Identification of CTSK as a TLR-related critical biomarker in liver cirrhosis via integrative bioinformatics and pathological characterization

Abstract

Liver cirrhosis (LC) is a common chronic disease worldwide with a poor prognosis, and its pathogenesis has not been fully elucidated. Toll-like receptors (TLRs) are crucial in LC progression. Here, we identified TLR-related genes, providing novel insights related to LC diagnosis, pathogenesis, and treatment. Data from public databases were analyzed using “limma” and WGCNA to screen candidate genes, and four hub genes (CXCL9, CXCL10, SPP1, CTSK) were selected through machine learning. These hub genes were validated through bioinformatics, quantitative real-time PCR (qRT-PCR), and immunohistochemistry (IHC). Both the hub genes and risk models demonstrated strong diagnostic potential for LC. The hub genes were enriched in various pathways and strongly correlated with immune infiltration. Subtypes characterized by different TLR signaling activity exhibited distinct immune responses. scRNA-seq analysis revealed significant differences in hub gene expression and TLR signaling activity across different cell types. ceRNA network analysis revealed interactions involving miRNAs, lncRNAs, and hub genes. Molecular docking supported the potential value of the hub genes as drug targets. In conclusion, TLR-related hub genes exhibit excellent diagnostic and therapeutic value in LC, and their dysregulation contributes to immune disorders and activation of pathogenic signaling pathways. CTSK may stimulate the TLR4-MyD88-NF-κB axis to facilitate LC progression.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-11606-6.

Introduction

Liver cirrhosis (LC), characterized by the replacement of normal liver tissue with regenerating nodules, typically occurs because of chronic inflammation that gradually progresses; it presents as the end stage of liver fibrosis and ultimately results in liver failure¹. Patients with asymptomatic cirrhosis, also known as compensated cirrhosis, often miss early diagnosis. This stage can persist for months to years before transitioning into the decompensated phase. Decompensated cirrhosis presents with various complications, frequently leading to hospitalization and thereby imposing a significant burden on both patients and caregivers. Mortality rates in this phase are high without liver transplantation¹. No effective drug has yet been approved². Therefore, a comprehensive understanding of the pathogenesis of LC is urgently needed to identify new targets for preventing the progression of cirrhosis.

Activated hepatic stellate cells (HSCs) differentiate into myofibroblasts and secrete type I collagen, serving as pivotal contributors to liver fibrosis and cirrhosis³. These cells are primarily stimulated by cytokines such as TGF-β, the most potent profibrogenic factor. TGF-β is secreted by Kupffer cells, sinusoidal endothelial cells, and hepatocytes, subsequently inducing HSC activation through Smad2/3 signaling pathways⁴. Inhibiting TGF-β/Smad signaling has been shown to suppress HSC activation^4,5. Although the pathogenesis of LC is known to involve complex signaling pathways, the regulatory factors involved remain incompletely understood.

Toll-like receptors (TLRs) are pattern recognition receptors widely expressed in both parenchymal and nonparenchymal liver cells and play critical roles in hepatic inflammation, injury, and fibrosis⁶. Recent research has underscored the importance of TLR-mediated signaling in the pathogenesis of diverse liver conditions, such as alcoholic⁷ and nonalcoholic liver disease⁸fibrosis⁹ and hepatocellular carcinoma¹⁰. TLR4 plays a central role in advancing liver fibrosis and cirrhosis. In quiescent HSCs, TLR4-MyD88-NF-κB signaling enhances chemokine secretion, facilitates Kupffer cell chemotaxis, and downregulates the expression of the TGF-β pseudoreceptor Bambi, thus sensitizing HSCs to TGF-β and promoting their activation¹¹. Other TLRs, such as TLR9¹², TLR3¹³, and TLR2⁹, also play important regulatory roles in this disease. In light of aberrant TLR signaling pathway activity in LC, exploring novel TLR-related genes critical for LC could offer a more profound understanding of its pathogenesis and advance molecular diagnostic and therapeutic strategies.

In our research, 4 hub genes (CXCL9, CXCL10, SPP1, CTSK) for LC diagnosis were identified, and risk models for LC were subsequently constructed on the basis of the hub genes. Comprehensive analyses were conducted to elucidate the involvement of these hub genes in LC progression. Finally, molecular docking revealed these genes as promising therapeutic targets for LC. The technological workflow is shown in Fig. 1.

Fig. 1.

The technological workflow for this study.

Methods

Data collection and preprocessing

Five datasets, namely, the microarray datasets GSE14323, GSE36411, GSE89377, and GSE6764 and the RNA sequencing dataset GSE136103¹⁴, were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). One outlier sample was deleted from the GSE36411 dataset. In the GSE136103 dataset, single-cell transcriptomic data of CD45 + liver leukocytes were selected from the liver tissues of 5 healthy control (HC) individuals and 5 LC patients. Table 1 provides detailed information on these datasets. To address batch effects, the “removeBatchEffect” function in the “limma” package in R was applied, combining the GSE14323 and GSE36411 datasets into a unified expression matrix for subsequent analysis. The GSE89377 and GSE6764 datasets were combined into a validation set using the same methodology.

Table 1.

The information of all the datasets in the study.

Dataset	Platform	HC	LC
GSE14323	GPL571	19	41
GSE36411	GPL10558	20	21
GSE89377	GPL66947	13	12
GSE6764	GPL570	10	13
GSE136103	GPL2301	5	5

Additionally, 102 TLR signaling pathway-related genes were acquired from KEGG_TOLL_LIKE_RECEPTOR_SIGNALING_PATHWAY in the MsigDB database (https://www.gsea-msigdb.org/gsea/msigdb/) and defined as TLR-related genes.

Weighted gene coexpression network analysis (WGCNA)

Using the “WGCNA” package in R, WGCNA was performed to develop scale-free coexpression networks correlated with clinical traits. An optimal soft power β was chosen for building a weighted adjacency matrix, which was subsequently converted into a topological overlap matrix (TOM). Modules were assigned colors, and module eigengenes (MEs) were calculated.

We selected key modules using r > 0.50 and P < 0.05 as thresholds. Disease-related genes from the key modules were extracted according to the following criteria: gene significance (GS) > 0.4 and module membership (MM) > 0.5.

Identification of differentially expressed genes (DEGs) and functional enrichment analysis

Using the “limma” package in R, DEGs were assessed between two groups. The criteria for statistical significance were |log2FoldChange| > 0.5 and p-value < 0.05.

Gene Ontology (GO) enrichment, including biological process (BP), molecular function (MF), and cellular component (CC) enrichment, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis^15–17 were performed using the “ClusterProfiler” and “org.Hs.eg.db” packages in R.

Identification of candidate genes

Candidate genes were identified by intersecting TLR-related genes, DEGs, and disease-related genes using the “VennDiagram” package in R. A protein-protein interaction (PPI) network was generated using the STRING database (version 12.0). Disconnected nodes were excluded, and the remaining network was visualized with Cytoscape software (version 3.9.1). Correlations between candidate genes were analyzed using the “corrplot” package in R, and univariate logistic regression with the “rms” package in R was employed to determine odds ratios (ORs).

Hub gene selection and model construction

Three machine learning methods, i.e., least absolute shrinkage and selection operator (LASSO), support vector machine (SVM), and random forest (RF), were applied to identify hub genes; they were run with the “glmnet”, “e1071”, and “randomForest” packages, respectively, in R. The “pROC” package in R was applied to conduct receiver operating characteristic (ROC) curve analysis, which was used to evaluate the predictive performance of the hub genes and diagnostic models. A nomogram was developed with the “rms” package in R, and calibration curves and decision curve analysis (DCA) were subsequently used to test the reliability of the nomogram.

Tissue collection and preprocessing

Clinical liver tissue samples, including three control samples and three LC samples, were obtained from the Second Affiliated Hospital of Harbin Medical University (ethics number: KY2024-114). The liver tissues were fixed in 4% paraformaldehyde before being dehydrated and embedded in paraffin. Tissue Sect. (5 μm thick) were subsequently prepared.

Histological analysis

Histological examination was performed via hematoxylin-eosin (HE) staining, and fibrosis severity was evaluated by Masson’s trichrome staining.

Quantitative real-time PCR (qRT-PCR)

Total RNA in liver tissue was extracted with TRIzol (Invitrogen, USA), reverse transcribed into cDNA with a reverse transcription kit from Roche, and analyzed via qRT-PCR using a SYBR Green kit (Roche, Switzerland). β-actin was used as the internal control, and the 2^−△△CT method was used to quantify relative gene expression. The primer sequences are provided in Table 2.

Table 2.

Sequences of hub gene-specific primers used for qRT-PCR.

Hub genes	Forward primer (5’−3’)	Reverse primer (5’−3’)
CXCL10	GGTGAGAAGAGATGTCTGAATCC	GTCCATCCTTGGAAGCACTGCA
CTSK	GAGGCTTCTCTTGGTGTCCATAC	TTACTGCGGGAATGAGACAGGG
CXCL9	CTGTTCCTGCATCAGCACCAAC	TGAACTCCATTCTTCAGTGTAGCA
SPP1	CGAGGTGATAGTGTGGTTTATGG	TGAACTCCATTCTTCAGTGTAGCA
β-actin	GGGAAATCGTGCGTGACATT	GGAACCGCTCATTGCCAAT

Immunohistochemistry (IHC)

To prepare tissue sections for immunohistochemical analysis, antigen retrieval with EDTA was performed, and peroxidase activity was blocked with 3% hydrogen peroxide. The sections were then incubated with anti-CTSK antibodies overnight at 4 °C. The following day, the sections were incubated with a secondary antibody at room temperature for 60 min, counterstained with hematoxylin, dehydrated, and sealed, after which CTSK expression was evaluated.

Gene set enrichment analysis (GSEA)

Utilizing the “clusterProfiler” and “org.Hs.eg.db” packages in R, GSEA was performed to explore the possible functions of the hub genes. Predefined gene sets were derived from the GO and KEGG collections in the MsigDB database.

Immune infiltration analysis

Immune profiling was conducted using three computational tools. CIBERSORT was used to estimate the relative proportions of immune cells via the “CIBERSORT” package in R¹⁸. The MCP-counter method was used to calculate the absolute abundances of 10 cell types using the “MCPcounter” package in R¹⁹. Immune function activities were assessed with the ssGSEA algorithm implemented in the “GSVA” package in R²⁰.

Consensus cluster analysis

With the “ConsensusClusterPlus” package in R, consensus clustering was conducted on LC samples according to the expression profiles of the hub genes. The ideal number of clusters was identified by analyzing the consensus matrix and the cumulative distribution function (CDF) curve. To evaluate the geometrical distance and expression patterns within the identified clusters, principal component analysis (PCA) was applied.

Distinctions in function and immune infiltration between clusters

To compare the enrichment of the TLR signaling pathway between clusters, the R package “GSVA” was utilized. The cluster-associated DEGs were subjected to GSEA with the “clusterProfiler” package in R. The immune infiltration levels between clusters were determined using the “CIBERSORT” package in R.

scRNA-seq analysis

The “Seurat” package in R was used for data processing. Cells expressing between 200 and 4,500 genes were selected, with each gene detected in at least 3 cells and a mitochondrial content less than 10% per cell. Batch effects were corrected using the “harmony” package in R. Cell types were identified on the basis of characteristic markers obtained from the literature^14,21,22. Uniform manifold approximation and projection (UMAP) was utilized for nonlinear dimensionality reduction and visualization. Additionally, the “AUCell” package in R was used to calculate signaling pathway scores across different cell types.

Construction of the ceRNA network

The miRWalk database was utilized to identify mRNA-miRNA interaction pairs. By utilizing the starBase database, lncRNAs associated with these miRNAs were identified. The ceRNA network was constructed with Cytoscape software. Within this network, the hub genes are displayed as red nodes, the miRNAs are displayed as orange nodes, and the lncRNAs are displayed as purple nodes.

Protein-ligand interaction analysis

The protein structures were chosen from the UniProt database and filtered to include only human proteins. The structure file for the small molecule was obtained from the PubChem database. Key targets were molecularly docked with small-molecule compounds using AutoDock Tools (version 1.5.7). The binding activities were analyzed on the basis of docking energy values, and the docking outcomes were visualized with PyMOL software.

Statistical analysis

R software and GraphPad Prism were used for statistical analysis and graphing. The Pearson test was utilized to evaluate correlations between two variables, and the Wilcoxon test was utilized to compare intergroup differences. Statistical significance was defined as * p < 0.05, ** p < 0.01, and *** p < 0.001.

Ethics statement

All experimental procedures have been reviewed and approved by the Medical Ethics Committee of the Second Affiliated Hospital of Harbin Medical University (ethics number: KY2024-114). All experiments were performed in accordance with relevant guidelines and regulations. Informed consent was obtained from all participants.

Results

WGCNA and differential expression analysis were used for screening genes

To identify the crucial genes linked to disease characteristics, WGCNA was conducted. We assessed the fitting index for scale-free networks and the average connectivity with a soft-thresholding power ranging from 1 to 20. A soft-thresholding power of 4 was identified as optimal, leading to the identification of 21 distinct modules (Fig. 2A, B). A correlation heatmap was generated, highlighting the red, turquoise, and lightcyan modules as key modules (Fig. 2C). The results revealed a strong positive correlation between GS and MM in the lightcyan, red, and turquoise modules (Fig. 2D-F). Ultimately, on the basis of thresholds of GS > 0.4 and MM > 0.5, a total of 943 disease-related genes were screened from key modules for further investigation.

Fig. 2.

Identification of disease-related genes and DEGs. (A) Scale-free fit index analysis for soft-threshold powers and mean connectivity for various soft-threshold powers. (B) The module clustering dendrogram for the coexpression network from WGCNA. (C) Heatmap of the correlations between gene modules and traits. Scatter plot of MM vs. GS in the (D) red module, (E) turquoise module and (F) lightcyan module. (G) Volcano map of DEGs between the LC group and the HC group. (H) Heatmap of DEGs between the LC group and the HC group.

Differential expression analysis between the LC and HC groups revealed 1,113 DEGs, with 771 upregulated and 342 downregulated (Fig. 2G). A heatmap was generated to visualize the DEG expression patterns and demonstrate intragroup consistency (Fig. 2H). This comprehensive analysis provided a robust foundation for subsequent analyses.

Enrichment analysis of DEGs

To elucidate the latent mechanisms driving LC progression, GO and KEGG pathway analyses were performed using both the upregulated and downregulated genes. GO enrichment analysis revealed the terms in the BP, CC, and MF categories that were enriched in the DEGs; these included chemotaxis, the collagen-containing extracellular matrix, intracellular zinc ion homeostasis, blood microparticle, and extracellular matrix structural constituent (Fig. 3A, C; Supplementary Data S1, S2). The KEGG enrichment analysis results revealed alterations in multiple immune response-related pathways, such as antigen processing and presentation; complement and coagulation cascades, and the chemokine signaling pathway (Fig. 3B, D; Supplementary Data S3, S4). These observations emphasize the critical influence of immune-related mechanisms on the progression of LC. The TLR signaling pathway was significantly enriched, which prompted further investigation into the potential role of TLRs in LC development.

Fig. 3.

Functional enrichment analyses of DEGs were conducted, and candidate genes were identified. (A) GO and (B) KEGG enrichment analyses of the upregulated DEGs. (C) GO and (D) KEGG enrichment analyses of the downregulated DEGs. (E) Venn diagram showing 8 overlapping candidate genes common to DEGs, TLR-related genes and disease-related genes. (F) The PPI network of candidate genes. (G) Correlations of candidate genes. (H) Univariate logistic regression analysis of candidate genes.

Identification of candidate genes

Eight overlapping candidate genes common to TLR-related genes, DEGs, and disease-related genes were identified (Fig. 3E). The interactions between these genes were subsequently explored within the PPI network and visualized with Cytoscape. Notably, PIK3R3 did not directly interact with the other candidate genes (Fig. 3F). Correlation analysis revealed significant synergistic relationships between the candidate genes (Fig. 3G). Univariate logistic regression analysis was then employed to calculate the ORs, and the findings suggested that these eight genes exhibited a risk effect and might have implications in promoting LC (Fig. 3H).

Exploration of hub biomarkers and construction of diagnostic models

To further refine candidate genes, LASSO regression was applied, narrowing the candidate genes down to four (Fig. 4A, B). Similarly, the SVM algorithm prioritized six genes as critical variables capable of distinguishing between LC and HC samples (Fig. 4C, D). In addition, the top six genes for mean decrease accuracy and mean decrease Gini were selected from the RF algorithm analysis (Fig. 4E, F, G). Utilizing a Venn diagram to intersect the gene sets derived from the aforementioned methods, four hub genes were identified: CXCL9, CXCL10, SPP1, and CTSK (Fig. 4H). Expression analysis revealed significantly higher levels of these four hub genes in LC samples than in HC samples in both the training and testing sets (Fig. 4I; Supplementary Fig. S1A). Furthermore, to assess the diagnostic potential of the hub genes, ROC curves were generated, and area under the curve (AUC) values served as indicators of predictive accuracy. The results (Fig. 4J) indicated that all four genes had AUC values above 0.90, with CXCL10 (AUC: 0.953), CTSK (AUC: 0.943), SPP1 (AUC: 0.924), and CXCL9 (AUC: 0.947), having excellent predictive capabilities for LC.

Fig. 4.

Screening hub genes and constructing disease prediction models. (A) Four genes were screened via LASSO regression. (B) Ten-fold cross-validation for tuning parameter selection in LASSO regression. (C) Error and (D) accuracy of five-fold cross-validation in SVM algorithms to extract 6 genes. (E) Model error and the number of RF trees in a correlation plot. (F), (G) The RF algorithm selected 6 genes. (H) Venn diagram to intersect 4 gene subsets. (I) Expression of the 4 hub genes between the LC and HC groups in the training set. (J) ROC curve for the hub genes for LC diagnosis in the training set. ROC curves for the diagnostic risk models in the (K) training and (L) testing sets. (M) Individual risk scores based on LASSO regression. (N) A nomogram was used to predict the occurrence of LC. Calibration curves for the (O) training and (Q) testing sets. DCA for the (P) training and (R) testing sets.

To further determine their predictive performance, we integrated the four hub genes into the LASSO, SVM, and RF algorithms to construct disease prediction models. ROC curve analysis of both the training and testing sets demonstrated AUC values exceeding 0.90 for all three models (Fig. 4K, L). Notably, the LASSO regression and SVM models achieved identical AUC values for the testing set, both outperforming the RF model. Consequently, the LASSO regression model was selected to calculate individual risk scores on the basis of hub gene expression. As shown in Fig. 4M and Supplementary Fig. S1B, individuals with elevated risk scores were predisposed to a higher incidence of LC. Additionally, a nomogram incorporating the four hub genes was constructed to estimate the incidence of LC (Fig. 4N). The calibration curves for the training and testing sets demonstrated close alignment with the ideal curve, indicating high model reliability (Fig. 4O, Q). DCA further highlighted the practical value of the model in clinical settings (Fig. 4P, R). In summary, through extensive analyses of public datasets, we confirmed the robust predictive ability of the hub genes and their associated diagnostic models for LC.

Experimental validation of the hub genes

To validate the bioinformatics findings, partial liver tissue samples were collected from clinical cases. HE and Masson staining results (Fig. 5A, B) revealed a normal liver structure in the HC group, whereas the LC group exhibited pseudolobule formation and extensive collagen fiber deposition. qRT-PCR analysis confirmed that the mRNA levels of the four hub genes were significantly higher in the LC group than in the HC group (Fig. 5C-F). IHC staining further revealed increased CTSK expression in the LC group (Fig. 5G, H). All these results align closely with the bioinformatics predictions.

Fig. 5.

Experimental validation of the hub genes and GSEA. (A) HE staining and (B) Masson staining of liver tissue. mRNA expression of (C) CXCL10, (D) CXCL9, (E) SPP1, and (F) CTSK was verified via qRT‒PCR. Expression of CTSK in (G) HC tissue and (H) LC tissue by IHC staining. GSEA of KEGG (top 4 and bottom 4) based on the enrichment scores for (I) CTSK, (J) SPP1, (K) CXCL10, and (L) CXCL9.

Biological significance of hub genes

The functions of the four hub genes were investigated by GSEA. The results were ranked by enrichment score, and the top 4 and bottom 4 results were visualized (Fig. 5I-L; Supplementary Fig. S1C-F). The overexpression of CXCL9, CXCL10, CTSK, and SPP1 was involved in many biological processes, cellular components, and molecular functions (Supplementary Data S5-S8). And these genes were significantly enriched in immune-related pathways, including antigen processing and presentation, NOD-like receptor signaling pathway, Th1 and Th2 cell differentiation, and Toll-like receptor signaling pathway (Supplementary Data S9-S12), suggesting that the hub genes may contribute to the occurrence and development of LC by regulating immune-related pathways.

Analysis of differences in the immune microenvironment

Given the critical involvement of immune-related pathways in LC highlighted by the enrichment analysis, we investigated the relative abundance of infiltrating immune cells across samples using the CIBERSORT algorithm. Naive CD4 T cells were absent in all the samples; however, 21 other infiltrating immune cell types were identified in liver tissues. A bar plot was used to illustrate the overall landscape of immune cell distribution (Fig. 6A). The Wilcoxon test findings revealed noticeably more plasma cells, resting memory CD4 T cells, gamma delta T cells, M1 macrophages, and M0 macrophages in the LC samples than in the HC samples, whereas there were more naive B cells, CD8 T cells, follicular helper T cells, Tregs, resting NK cells, monocytes, M2 macrophages, activated dendritic cells, and neutrophils in the HC samples (Fig. 6B). Correlation analysis between the four hub genes and significantly different immune cells in LC tissues revealed both positive and negative relationships, implying that these genes may contribute to immune dysregulation in LC (Fig. 6C). Further exploration revealed a close interplay among immune cells (Fig. 6D). The immune cell composition in different samples was analyzed using the MCPcounter algorithm, and the Wilcoxon test revealed noticeable differences in the proportions of cells other than CD8 + T cells in the LC samples compared with the HC samples (Fig. 6E). The ssGSEA algorithm revealed that the immune pathways were obviously activated in LC tissues (Fig. 6F). These results highlight the importance of immune dysregulation in the pathogenesis of LC.

Fig. 6.

Immune infiltration analysis and connection analysis between hub genes and immune cells. (A) Using CIBERSORT, the bar plot provides an overview of the distribution of immune cells. (B) The percentages of immune cells in the LC and HC samples were determined using CIBERSORT. (C) Heatmap of the correlations between immune cells and hub genes. (D) Heatmap of the correlations between immune cell types. (E) The number of immune cells in the LC and HC samples was determined with an MCP-counter. (F) ssGSEA of pathways in LC and HC samples.

Identification of TLR-related subtypes in LC

To characterize the TLR-related patterns in LC, we conducted unsupervised clustering using 62 LC samples from the training dataset on the basis of the expression levels of the four hub genes. The optimal number of clusters was determined to be k = 2, as indicated by stable isoform numbers (Fig. 7A) and changes in the CDF curve area from k = 2 to k = 5 (Fig. 7B). PCA further confirmed the distinct separation, dividing the samples into two subtypes: cluster A (n = 28) and cluster B (n = 34) (Fig. 7C).

Fig. 7.

Identification of TLR-related subtypes. (A) Consensus clustering matrix when k = 2. (B) Consensus CDF curves for k values ranging from 2 to 5. (C) PCA diagram separating the subtype A and subtype B samples. (D) Heatmap and (E) boxplot of the expression of the 4 hub genes between subtypes. (F) Different GSVA scores for the TLR signaling pathway between subtypes. GSEA of subtype-related DEGs in the (G) GO and (H) KEGG analyses. (I) Heatmap delineating the abundance of various infiltrated immune cell types between subtypes. (J) Boxplot showing the differences in infiltrating immune cells between subtypes.

All the hub genes presented significantly greater expression in cluster B than in cluster A (Fig. 7D, E). We then evaluated TLR signaling pathway activity using GSVA, and the result revealed that for this pathway, the GSVA scores were significantly higher for cluster B than for cluster A (Fig. 7F).

GSEA of DEGs between the two subtypes provided further insight into functional and pathway differences. In cluster B, GO enrichment highlighted activated biological processes, such as negative regulation of viral genome replication; molecular functions, such as chemokine receptor binding; and cellular components, including the external side of the plasma membrane (Fig. 7G; Supplementary Data S13). KEGG pathway analysis revealed the enrichment of immune-related pathways, including antigen processing and presentation and the NOD-like receptor signaling pathway (Fig. 7H; Supplementary Data S14).

Immune cell infiltration also differed between the clusters. Cluster A included more follicular helper T cells, regulatory T cells, resting NK cells, and M2 macrophages. Cluster B showed increased infiltration of gamma delta T cells and M1 macrophages (Fig. 7I, J).

Single-cell analysis

To elucidate the immune landscape at the single-cell level, a scRNA-seq dataset was utilized in this study. After quality control, the cells were categorized into nine primary cell types on the basis of marker genes: T cells, ILCs, MPs, B cells, endothelial cells, hepatocytes, plasma cells, fibroblasts, and pDCs (Fig. 8A; Supplementary Fig. S2A). Moreover, the distribution of hub gene expression was mapped across these cell types. We found that CXCL10 and CXCL9 were strongly distributed in MPs, CTSK was highly localized in fibroblasts, and SPP1 was mainly expressed in hepatocytes (Fig. 8B, C). In addition, we analyzed the activation of the TLR signaling pathway in different cell clusters. The results demonstrated that MPs presented the highest AUCell score, followed by ILCs and plasma cells, which also presented elevated scores (Fig. 8D, E). As shown in Fig. 8F, the AUCell scores for most cell types were significantly different between the two groups, highlighting substantial differences in TLR pathway activity.

Fig. 8.

scRNA-seq analysis and molecular docking of drug and hub genes. (A) UMAP plot of the cell clusters in the LC and HC groups. (B) UMAP plot of the hub genes. (C) Hub gene enrichment in cell clusters. (D) AUCell scores for the TLR signaling pathway in different cell clusters. (E) UMAP plot of the TLR signaling pathway. (F) Boxplot showing the differences in AUCell scores for the TLR signaling pathway in different cell clusters between the two groups. (G) Molecular docking between CTSK and baicalin. (H) Molecular docking between SPP1 and baicalin. (I) Molecular docking between CXCL9 and baicalin. (J) Molecular docking between CXCL10 and baicalin.

Construction of the ceRNA network and the molecular docking landscape of LC drugs against hub genes

Twelve miRNAs and 42 lncRNAs were identified, after which a ceRNA network was constructed and displayed (Supplementary Fig. S2B). Such a network is essential for elucidating the complexities of genetic regulation, illustrating how gene expression is modulated, how miRNAs and lncRNAs fine-tune this process, and how their dysregulation leads to disease.

Molecular docking was employed to evaluate the binding affinity between molecules and their targets on the basis of binding energy. The docking scores for the four genes were all less than − 5 kcal/mol, suggesting favorable affinities between these genes and baicalin (Fig. 8G-J).

Discussion

LC, the major cause of ascites, upper gastrointestinal bleeding, and even liver failure worldwide, poses significant challenges to human society because of its progressive nature and increasing prevalence¹. The objective of this study was to identify reliable diagnostic biomarkers, enhance the understanding of the underlying mechanisms of LC, and formulate appropriate therapeutic strategies, leveraging bioinformatics, machine learning, and experimental validation.

In this study, enrichment analyses of DEGs revealed TLR signaling pathway activation in the LC group. Thus, we subsequently investigated the potential value of TLR-related genes in the occurrence and progression of LC. On the basis of public databases and experimental verification, we identified four TLR-related hub genes (CXCL9, CXCL10, SPP1, and CTSK) that exhibit risk effects and might promote LC. Moreover, these genes, along with their risk prediction models, demonstrated powerful diagnostic potential. Thus, these four genes may offer novel insights for the future biological diagnosis of LC.

GO and KEGG enrichment analyses revealed that the DEGs were enriched not only in the TLR signaling pathway but also in pathways such as the NOD-like receptor signaling pathway and the PI3K-Akt signaling pathway. NLRP3 is pivotal for the NOD-like receptor signaling pathway, which promotes liver inflammation and fibrosis through the formation of the NLRP3 inflammasome²³. In addition, the PI3K-Akt signaling pathway can facilitate fibrosis progression by regulating DC function, the inflammatory response, and HSC activation²⁴. The GSEA results for the hub genes align with the above results, highlighting enrichment in pathways linked to the immune system (such as NOD-like receptor signaling) and signal transduction (such as PI3K-Akt signaling).

The interplay among immune cells contributes to liver fibrosis and cirrhosis development²⁵ which is consistent with our findings from immune infiltration analyses using CIBERSORT, MCP-counter, and ssGSEA methods. These analyses revealed significant variations in both the relative abundance and absolute quantity of immune cell types between the studied groups. Strong correlations were identified between multiple immune cell types and hub genes. Notably, the proportions of NK cells and M1 and M2 macrophages exhibited robust correlations with hub gene levels, implying the potential regulation of these cell types by these genes and the key functions of these cells in fibrosis progression. Previous studies indicate that TLR4 suppresses NK cell activity²⁶. During liver fibrosis, NK cells exhibit reduced abundance and functional impairment, attenuating their cytotoxic effects on HSCs and thereby exacerbating fibrosis^27–29. Furthermore, M1 macrophage polarization exacerbates liver fibrosis by amplifying inflammatory responses³⁰ TLR-related MyD88 and CXCL10 can be involved in this process³¹. Promoting the shift from pro-inflammatory M1 to anti-inflammatory M2 macrophages may inhibit fibrotic progression³². These findings align with our observed correlations between hub gene levels and infiltrating immune cell levels. Collectively, the results suggest that regulation of hub genes may affect the enrichment of immune cell subpopulations and thereby restore immune homeostasis in the context of liver cirrhosis. Recent studies have highlighted the prominent role of myeloid-derived suppressor cells (MDSCs) in liver, for example, TLR4-dependent mMDSCs suppress CD8 + T cells, thereby promoting fibrogenesis and shaping the hepatocellular carcinoma (HCC) microenvironment³³. CD8 + T cell suppression in our research is consistent with this view. Further investigation into TLR4-dependent mMDSCs in the future may uncover novel mechanisms of promoting liver fibrosis. scRNA-seq analysis further revealed differences in cell cluster distributions between the groups, with hub gene expression varying in distinct cell clusters. These findings collectively illustrate the complexity of the immune response in LC pathogenesis and underscore the importance of hub genes in driving immune cell changes and immune function alterations.

Consensus clustering revealed that subtypes with different TLR signaling activities were clearly different in terms of immune and inflammatory responses. scRNA-seq analysis further confirmed significant differences in TLR signaling pathway activity between the LC and HC groups across cell clusters. These results underscore the pivotal role of TLRs in immune dysregulation in LC.

CXCL9 and CXCL10 expression was higher in LC tissues than in HC tissues in our study, a finding that aligns with the results of prior studies³⁴. Previous studies have established that CXCL10 is instrumental in the pathogenesis of LC³¹. It is well-established that CCL2, a key chemokine secreted upon TLR4/NF-κB activation, drives hepatic fibrosis progression primarily through the CCL2/CCR2 axis, which facilitates the recruitment and polarization of macrophages toward a pro-inflammatory phenotype^35,36. Intriguingly, CXCL10 has also been implicated in M1 macrophage polarization through the activation of JAK-STAT1 signaling, further exacerbating fibrosis³¹. In this study, a strong positive association between CXCL10 and STAT1 expression was observed, further corroborating this mechanism.These findings suggest that CCL2 and CXCL10 exhibit synergistic yet distinct roles in fibrogenesis, and dual targeting of these pathways may offer a more effective strategy to disrupt fibrotic progression at different stages. Conversely, CXCL9 has been shown to attenuate fibrosis-associated angiogenesis, exerting anti-fibrotic effects^37,38. However, emerging evidence indicates that CXCL9 recruits pro-inflammatory macrophages, amplifying inflammation and fibrosis³⁹. This dual functionality underscores the pleiotropic role of CXCL9 in hepatic fibrogenesis.

The SPP1 gene, encoding osteopontin (OPN), is implicated in cell migration, inflammation, metabolic dysregulation, and tumorigenesis and is known to serve as a critical mediator in the progression of chronic liver diseases⁴⁰. Recent research has demonstrated that the transcription factor E4BP4 promotes the expression and secretion of OPN by stabilizing YAP, a known activator of OPN, thereby facilitating the progression of liver fibrosis⁴¹. Previous studies have demonstrated that CD44 is a well-established pro-fibrotic gene^42–44 and is highly expressed in activated HSCs⁴⁵. OPN binds to CD44 through the C-terminal domain and plays a synergistic role in liver fibrosis, collectively promoting HSC activation and fibrosis progression^40,46. Furthermore, OPN plays a distinctive profibrogenic role. For instance, it drives type I collagen upregulation and liver fibrogenesis via integrin αvβ3-dependent PI3K-pAkt-NFκB signaling, independent of CD44 binding⁴⁷. The similarities and differences in their mechanisms provide valuable clues for future research into the pathogenesis of hepatic fibrosis.

CTSK has been studied primarily for its role in bone diseases⁴⁸. Small-molecule drugs targeting CTSK can alleviate osteoporosis⁴⁹. However, the specific role of CTSK in LC remains largely unexplored. This research aimed to address the aforementioned gap. Our study revealed higher CTSK expression in LC tissues than in HC tissues, a finding that was supported by bioinformatics and qRT-qPCR analyses and attracted our attention. Therefore, we selected CTSK as the gene of interest, further confirming its overexpression through immunohistochemistry. GSEA revealed that CTSK was significantly enriched in the Toll-like receptor signaling pathway, MyD88-dependent toll-like receptor signaling pathway, and NF-kappa B signaling pathway. Others have previously demonstrated that the TLR4-MyD88-NF-κB axis can regulate TGF-β signaling, thereby facilitating liver fibrosis progression¹¹. Previous studies indicate that there is a synergistic relationship between CTSK and MyD88⁵⁰. The upregulation of CTSK expression is likely to induce activation of the NF-κB pathway⁵¹. Studies have demonstrated that TLR4 is significantly upregulated in the CTSK overexpression model, CTSK deficiency or pharmacological inhibition leads to a reduction in TLR4 expression and concurrently mitigates the inflammatory response^52–55. In CTSK-knockout rheumatoid arthritis models, both the number of immune cells and the levels of TLRs, including TLR4, TLR5, and TLR9, were significantly reduced⁵⁶. Moreover, the classical TLR4/MyD88 pathway is activated under CTSK stimulation, and CTSK inhibition results in a marked decrease in TLR4 expression⁵⁷. These results indicate a close interaction between CTSK and TLR4-MyD88-NF-κB axis, supporting the hypothesis that CTSK interacts with TLR4 to activate TLR4-mediated pathophysiological changes in LC (Fig. 9). Additionally, scRNA-seq data analysis revealed that CTSK is closely related to fibroblasts, suggesting that CTSK might interact with fibroblasts during the pathogenesis of liver cirrhosis.

Fig. 9.

Mechanism’s diagram.

To date, the roles of the 12 miRNAs in the ceRNA network have not been directly investigated in LC. Our findings imply that these miRNAs may function as pivotal modulators in the progression of LC. However, this hypothesis and the underlying mechanisms require further research.

As mentioned above, the four hub genes are implicated in LC pathogenesis; therefore, they may serve as potential therapeutic targets. Some studies have shown that baicalin has a positive effect on the treatment of LC^58,59. The molecular docking results suggest that the hub genes are potential targets for baicalin in the treatment of LC. Shortly, therapeutic strategies targeting hub genes could emerge as novel and efficacious approaches for treating LC.

Despite the novel findings presented in this study, several limitations must be acknowledged. First, reliance on publicly available datasets restricts the generalizability of our diagnostic model validations; and among these datasets, only two (GSE14323 and GSE6764) contain well-annotated etiological information, thereby limiting cross-etiological comparisons. Future work will enroll LC patients with diverse etiologies from clinical practice. This multi-etiology cohort will enable both model validation and investigation of potential etiology heterogeneity of TLR signaling dynamics. Second, while CTSK is associated with cirrhosis, its specific role in LC progression requires more extensive research. Subsequent studies will utilize in vitro approaches (e.g., CTSK knockdown/overexpression in hepatic stellate cells) and in vivo models (e.g., CTSK-deficient mice) to establish causal relationships. Third, although baicalin has been proposed as a potential therapeutic agent targeting the hub genes, experimental studies are necessary to confirm its mechanisms of action. In subsequent studies, in vitro enzymatic assays and in vivo animal models will be employed to evaluate its therapeutic potential in LC.

Conclusions

Our research demonstrated that four TLR-related hub genes play crucial roles in LC pathogenesis by promoting immune disorders and activating pathogenic signaling pathways; additionally, the four hub genes showed excellent value in the diagnosis and treatment of LC. Moreover, CTSK may activate the TLR4-MyD88-NF-κB axis to promote LC progression.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Jx.W., N.L., Xr.L. and Sz.J. designed research; Jx.W., Ry.G., and Hl.C. analyzed data; Xy.C. and Xy.J. interpretated results; Jx.W., Ly.X.,Fc.L., and Jc.H. wrote the manuscript; Sz.J., R.P.C, Xy.G., and Jh.Q. revised manuscript. All authors read and approved the final manuscript.

Funding

This study was supported by Natural Science Foundation of Heilongjiang Province (LH2023H036).

Data availability

The datasets analysed during the current study are available in the GEO database (https://www.ncbi.nlm.nih.gov/geo/). The datasets analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.