Exploring the potential of Buzhong Yiqi decoction in the treatment of membranous nephropathy: a comprehensive study combining bioinformatics, network pharmacology, and experimental validation

Abstract

Background and Objective: Buzhong Yiqi Decoction (BZYQ), a classic Chinese herbal formula, demonstrates clinical efficacy in treating membranous nephropathy (MN), yet its molecular mechanisms remains undefined. This study aims to uncover the therapeutic mechanisms of BZYQ against MN at the molecular level. Methods: BZYQ components/targets were retrieved from TCMSP and UniProtKB, while MN-related targets were obtained from GeneCards and GEO. Core targets were identified via Protein-Protein Interaction (PPI) analysis and machine learning (ML). Immune cell infiltration was analyzed using CIBERSORT, and diagnostic values were evaluated via Receiver Operating Characteristic (ROC) curves. Molecular docking validated interactions between core targets and BZYQ’s active ingredients, with efficacy and mechanisms further verified in a cationic bovine serum albumin-induced mouse MN model. Results: Screening identified MYC and NFKBIA as core targets, primarily enriched in B-cell receptor signaling. Immune infiltration analysis showed their correlation with follicular helper T cell abundance. ROC curves confirmed area under the curve > 0.7 for both targets. Molecular docking revealed strong binding (average energy: -7.13 kcal/mol) between targets and BZYQ components. In vivo, BZYQ significantly reduced 24-h urinary protein (P < 0.001), increased serum albumin (P < 0.05), decreased serum creatinine (P < 0.05), and alleviated renal pathology/immune complex deposition. Mechanistically, immunohistochemistry/Western blotting showed BZYQ downregulated MYC (P < 0.05) and upregulated IκBα (P < 0.05) versus the model group. Conclusion: BZYQ emerges as a promising therapeutic for MN. Elucidating its anti-MN mechanisms provides a theoretical basis for clinical translation.

Introduction

Membranous nephropathy (MN) is one of the major causes of nephrotic syndrome in adults, with pathological features characterized by the deposition of immune complexes in the subepithelial space of the glomerular basement membrane (GBM), accompanied by diffuse thickening of the basement membrane [1]. These immune complexes are mainly composed of immunoglobulin G (IgG) and target antigens [2,3]. Currently, in addition to basic treatments such as diuresis, antihypertensive therapy, lipid-lowering therapy, and anticoagulant therapy, the treatment of MN mainly relies on glucocorticoids combined with immunosuppressive agents and biologicals. However, long-term treatment with glucocorticoid hormones and immunosuppressants is often associated with varying degrees of side effects, as well as issues like poor patient compliance and heavy economic burden. Thus, there remains a pressing need to continuously explore novel therapeutic approaches.

Buzhong Yiqi Decoction (BZYQ) consists of eight Chinese herbs: Astragalus (Huangqi), Ginseng (Renshen), Atractylodes (Baizhu), Cimicifuga (Shengma), honey-fried Licorice (Zhigancao), Angelica sinensis (Danggui), Pericarpium Citri Reticulatae (Chenpi), and Bupleurum chinense (Chaihu). It is recognized for its efficacy in regulating spleen-stomach function [4], improving fatigue, enhancing immunity [5], and treating stress urinary incontinence [6]. Similarly, BZYQ has also demonstrated significant efficacy in the clinical management of MN [7], which provides a basis for BZYQ as a promising candidate drug for the treatment of MN. Nevertheless, the exact mechanism underlying the efficacy of BZYQ in MN has not been fully elucidated, warranting further analytical methods and multi-dimensional studies.

Network pharmacology constructs biological networks to integrate drug-target relationships and reveal multi-target mechanisms of action [8]. Machine Learning (ML) accurately identifies disease-related genes, signaling pathways, and functional proteins to facilitate new target discovery and drug efficacy prediction [9]. Molecular docking predicts molecular binding modes and affinities for drug discovery [10]. This study integrated the above methods and performed the following explorations (Figure 1).

Figure 1.

A flowchart summarizing the study design.

(1) The active ingredients and corresponding targets of BZYQ were screened via the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) and Universal Protein Resource (UniProt), while the targets associated with MN were retrieved from the GeneCards and Gene Expression Omnibus (GEO).

(2) Protein-protein interaction (PPI) data among the targets were obtained from the Search Tool for the Retrieval of Interacting Genes (STRING) database, after which we constructed a PPI network to identify hub genes.

(3) Three algorithms of ML, namely Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest (RF), and Support Vector Machine-Recursive Feature Elimination (SVM-RFE), were employed for further optimization to screen core genes.

(4) Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis, immune infiltration analysis, and evaluation of diagnostic performance were performed on the core genes. Additionally, molecular docking was utilized to assess the binding affinity between core genes and active compounds, thereby validating their binding specificity.

(5) An animal model of MN was established to validate the therapeutic efficacy of BZYQ and the expression levels of core genes.

In summary, this study is designed to systematically explore and experimentally validate the therapeutic mechanism of BZYQ for treating MN, with the ultimate goal of providing novel therapeutic strategies and scientific theoretical support for clinical MN management.

Materials and methods

Screening for active ingredients and targets of BZYQ

The BZYQ active ingredients and their corresponding target proteins were obtained from the TCMSP (http://lsp.nwu.edu.cn/tcmsp.php). Considering that BZYQ tablets are administered orally, the screening criteria of Oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥ 0.18 were adopted. The corresponding target proteins were subsequently mapped to their standard gene names through the UniProt (https://doi.org/10.1093/nar/gkaa1050).

Screening for MN target

MN-related targets were obtained from the GeneCards (http://www.genecas.org/) and the GEO (GSE73953) (http://www.ncbi.nlm.nih.gov/geo/). Differential analysis of the GSE73953 dataset was performed using the R package Limma, with adjusted P-values < 0.05 and |log2FC| > 0.58 set as screening criteria to identify upstream and downstream genes influencing MN progression.

Prediction of BZYQ targets for MN treatment

Based on the above information, drug-disease target intersections were plotted on the Venny 2.1 (http://bioinfogp.cnb.csic.es/tools/venny/), and component-target network diagrams were plotted using Cytoscape 3.7.1 software.

Construction of the PPI network

The intersection targets were imported into the STRING database (https://string-db.org/), and the filtering criteria of “Homo sapiens” species and “minimum interaction score” ≥0.9 were applied to remove isolated nodes. Subsequently, PPI data were exported in TSV format. The data were then imported into Cytoscape 3.7.1 software to generate a PPI network diagram, and the CytoHubba plugin was used to analyze the interactions among the targets to identify central targets (hubs).

GO and KEGG enrichment analysis

The intersection targets of BZYQ treatment for MN were imported into the Database for Annotation, Visualization and Integrated Discovery (DAVID, https://david.ncifcrf.gov/) for GO (biological process, BP; cellular component, CC; molecular function, MF) and KEGG enrichment analysis, and the results were visualized using Weishengxin online platform (http://www.bioinformatics.com.cn/).

Machine learning

To identify the core targets, we applied three algorithms - LASSO, RF, and SVM-RFE - to refine the key targets from the previously identified major drug-disease intersection targets. First, stringent 10-fold cross-validation was implemented using the glmnet package in R to enhance the model’s robustness and reliability. Next, an RF classification model was built with the randomForest package, where key genes were prioritized according to the Gini index. Furthermore, the SVM-RFE algorithm was utilized via the e1071 and caret packages (with the “svmRadial” kernel) to further optimize feature selection, which effectively identified target genes with distinct expression profiles. Ultimately, by integrating the hub genes from PPI screening, we identified the potential core genes mediating the therapeutic effects of BZYQ in MN.

Gene set enrichment analysis (GSEA)

To assess the enrichment of core genes and identify their functional relevance, GSEA was conducted using the gseGO and gseKEGG functions from the clusterProfiler R package. This analysis evaluated the non-random enrichment of core genes within GO terms and KEGG pathways, ranking gene sets by their enrichment significance.

Immune-infiltration analysis

In this study, the GSE73953 dataset was analyzed using the CIBERSORT algorithm via its online platform to generate results. The relative proportions of 22 immune cell types in MN were then quantified using the Limma and dplyr packages. Subsequently, the ggpubr and ggExtra packages were employed to explore associations between the identified core targets and these immune-infiltrating cells, and the results were visualized to facilitate further analysis.

Receiver operating characteristic (ROC) curve

In this study, gene expression data from the GEO database GSE73953 were utilized to develop a training model. The model’s performance was assessed using the pROC package in R, with ROC curves generated to visualize performance. The area under the curve (AUC) was calculated to measure the model’s accuracy in differentiating MN patients from healthy controls.

Molecular docking

Molecular docking simulations were conducted to investigate the interactions between the active ingredients of BZYQ and core targets. It is worth noting that the presence of these compounds in the BZYQ formulation was experimentally validated via liquid chromatography-mass spectrometry (LC-MS) [11]. The 3D structures of target proteins were retrieved from the Protein Data Bank (PDB, https://www.wwpdb.org/) with a resolution threshold of ≤ 3.0 Å, with core active ingredients from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). These protein structures were preprocessed using PyMOL v3.0.4 to remove solvents and ligands, ensuring the proteins were in a suitable state for docking. Subsequently, molecular docking was performed using AutoDock 1.5.7 to simulate binding interactions, and the docking results were visualized using PyMOL3.0.

Animals

Thirty healthy female BALB/c mice (6-8 weeks old, 15-20 g) were obtained from Chengdu Pharmachem Biotechnology Co., Ltd. (Certificate of Conformity No. SCXK [Chuan] 2020-034). The animal experiment design and operating procedures of this study comply with the National Guidelines for the Care and Use of Laboratory Animals and have been approved by the Animal Ethics Committee of Southwest Medical University (Approval No.: 20221026-013). At the conclusion of animal experimentation, euthanasia was performed via intraperitoneal injection of a lethal dose of pentobarbital sodium (150 mg/kg) to ensure minimal suffering.

Modeling, grouping and drug administration

After one week of adaptive feeding, the mice were randomly divided into a normal group of 6 and a model group of 24. The administration method was as follows [12]: Pre-immunization for 2 weeks: 0.2 mg C-BSA (2 g/L) was fully emulsified with an equal volume of complete Freund’s adjuvant (CFA) and the mixture was injected into mice in the model group via multi-point subcutaneous injection; mice in the normal group were given an equal amount of CFA. Formal immunization for 6 weeks: C-BSA (6.5 mg/kg) was administered to the model group mice via tail vein injection, and an equal amount of physiological saline was injected into the normal group. Two mice died during the modeling process. Subsequently, 24-hour urine protein (24-UP) levels of each mouse in the model group were measured, and all levels were greater than 60 μg/day, indicating abnormal proteinuria [13]. Two mice were randomly selected for renal tissue immunofluorescence assay, and both exhibited IgG immune complex deposition, confirming the validity of the model. The remaining 20 model mice were randomly divided into four groups: a model group, a BZYQ high-dose group (7.42 g/kg), a BZYQ low-dose group (3.71 g/kg), and a Captopril group (1.3 mg/kg), with 5 mice in each group. The corresponding dose of the drug was administered orally once daily for 1 month.

Determination of biochemical indexes in mice

Prior to sampling, each mouse was individually placed in a metabolic cage for 24-hour urine collection, with group information recorded accordingly. Mouse urine samples were centrifuged at 4,000 rpm for 10 minutes (centrifuge radius: 10 cm), and 24-UP was measured following the instructions provided with the urine protein assay kit. After sampling, serum was collected from each group of mice, allowed to stand overnight, and centrifuged at 4,000 rpm for 10 minutes (centrifuge radius: 10 cm). We took the supernatant and measured the levels of albumin (Alb), serum creatinine (SCr), and blood urea nitrogen (BUN) according to the instructions of the corresponding reagent kit.

Histopathological analysis of the kidney

The tissues were fixed in 4% paraformaldehyde, then dehydrated, embedded in paraffin, sectioned, and stained with HE, Masson, and PAS stains, and the morphological changes of renal tissues were observed under a light microscope.

Detection of IgG deposition in the mouse kidney

Fresh kidney tissues were embedded in OCT embedding medium and immediately placed into a -20°C cryostat for sectioning. Sections were cut to a thickness of 4 μm, fixed in acetone for 5 minutes, and then washed three times with phosphate-buffered saline (PBS).The sections were soaked in 5% bovine serum albumin (BSA) blocking buffer before incubation with FITC-labeled mouse IgG antibody in the dark overnight. On the following day, the sections were brought to room temperature, washed again with PBS, mounted with glycerol, and air-dried. The sections were then examined under a fluorescence microscope and imaged.

Immunohistochemistry

Following deparaffinization, hydration, and antigen retrieval, kidney sections were blocked with 5% BSA and subsequently incubated overnight with primary antibodies against MYC and IκBα. The following day, the sections were washed with PBS, incubated with horseradish peroxidase (HRP)-conjugated corresponding secondary antibodies at room temperature for 1 h, stained with 3,3’-diaminobenzidine (DAB) and counterstained with hematoxylin.

Western blotting

Kidney tissues were homogenized in 400 μL of RIPA lysis buffer supplemented with 100 mmol/L PMSF. The protein concentration was determined and adjusted to 10 μg/μL using a microplate reader. Subsequently, protein samples were denatured in SDS loading buffer at 100°C for 10 min. 10 μL of each protein sample was subjected to SDS-PAGE on a 10% polyacrylamide gel and then electrotransferred onto a PVDF membrane. The membrane was blocked with 5% BSA in TBST at room temperature for 1 h, followed by incubation with primary antibodies (GAPDH 1:10,000, MYC 1:1,000, IκBα 1:1,000) at 4°C overnight. After washing with TBST, the membrane was incubated with corresponding horseradish peroxidase-conjugated IgG secondary antibody at room temperature for 1 h and washed again with TBST. Chemiluminescence was detected using ECL chemiluminescence substrate, and the membrane was imaged using the Qinxiang imaging system. The relative gray value ratio of the target protein to the internal control was quantified using ImageJ software.

Statistical analysis

Statistical data analysis was performed using GraphPad Prism 9.0 statistical software. Quantitative data were described, and intergroup comparisons were performed using one-way ANOVA. Unless otherwise specified, the significance level α was set at 0.05.

Results

Main target screening and component-target network diagram establishment

A total of 131 active ingredients of BZYQ were screened by TCMSP, and 232 active targets were obtained by querying the action targets of the above active ingredients and removing the duplicates; 1,840 MN-related target genes were screened by GEO-DEGs, of which 1,174 were up-regulated and 666 were down-regulated (Figure 2A); 4,478 MN targets were screened by GeneCards; 16 major targets of BZYQ for MN treatment were identified by taking the intersection of the three sets of data (Figure 2B). Finally, a total of 68 components corresponding to 16 major targets were found among the 131 active ingredients of BZYQ; a component-target gene network diagram (Figure 2C) was constructed, and the core components of BZYQ were identified based on the number of interactions: quercetin, kaempferol, stigmasterol, and licochalcone a (Table 1).

Figure 2.

Expression profile of the GSE73953 dataset. A. Volcano map of detected genes. Each dot represents a gene. Blue dots represent downregulated genes, and pink dots represent upregulated genes. The screening criteria for significant genes were |log2FC| > 1 and P < 0.05. B. Heatmap display of the clustering of the differentially expressed genes (DEGs). Blue represents low expression, pink represents high expression. C. BZYQ-component-target network. The Chinese medicines are depicted as red rectangle nodes, their compounds as yellow ellipse nodes, and the disease genes as orange polygon nodes.

Table 1.

Information on the core components of BZYQ

Mol ID	Molecule name	OB%	DL	Drug
MOL000422	kaempferol	41.88	0.24	HQ TSZ YMC
MOL000098	quercetin	46.43	0.28	HQ TSZ GQZ HL
MOL000449	Stigmasterol	48.03	0.76	SD GQZ
MOL000497	licochalcone a	40.79	0.29	GC JXT
MOL004328	naringenin	59.29	0.21	BH CP GC

PPI network analysis

The 16 targets common to diseases and drugs were entered into the STRING database to obtain a preliminary PPI network (Figure 3A). This work was then imported into Cytoscape 3.7.1 software, and the topological attributes of the PPI network were analyzed using the “NetworkAnalyzer” module. The color shades of the nodes were positively correlated with the size of the degree value (Figure 3B). The color depth of nodes was positively correlated with the degree value, the larger the degree value, the darker the color. As a result, the core targets in the network were identified as BCL2, MYC, NFKBIA (The NFKBIA gene encodes IκBα, an inhibitor of nuclear factor kappa B alpha), CCL2, and GSK3β (Table 2).

Figure 3.

A. Protein-Protein Interaction (PPI) of BZYQ targets for treating MN from STRING. B. PPI network processed using Cytoscape. C. Signaling pathways identified from Kyotoencyclopedia of genes and genomes (KEGG) analysis. D, E. Results of Gene Ontology (GO) Enrichment Analysis.

Table 2.

Core targets information

Number	Rank	Gene name	Protein name	Degree
1	1	BCL2	B-cell lymphoma 2 protein	13
2	1	MYC	Myc proto-oncogene protein	13
3	3	CCL2	Chemokine (C-C motif) ligand 2	10
4	4	GSK3β	Glycogen Synthase Kinase 3 Beta	9
5	5	NFKBIA	NF-kappa-B inhibitor alpha	8

Enrichment analysis

GO and KEGG analysis was performed on 16 targets using the DAVID platform, and the results with the lowest P-values (i.e., the most significant results) were selected for visualization. The results showed that the highly relevant signaling pathways include: Lipid and atherosclerosis pathway, B-cell receptor signaling pathway, IL-17 signaling pathway, PI3K-Akt signaling pathway, and MAPK signaling pathway (Figure 3C). Additionally, the results showed that BPs were mainly involved in cellular responses to peptides, responses to peptide hormone, positive regulation of cellular activation and responses to exogenous stimuli; CCs were mainly involved in macromolecular complexes such as Bcl-2 family protein complex, CD40 receptor complex, peptidase inhibitor complex, RNA polymerase II transcriptional repressor complex and macromolecular complexes; MFs were mainly involved in DNA-binding transcription factor binding, NF-κB binding and binding to various enzyme (Figure 3D and 3E).

Machine learning

The LASSO algorithm narrowed down the initial 16 genes to 8 key genes (Figure 4A). Meanwhile, the RF algorithm assigned an importance score to each gene; and a threshold of 0.1 was set, resulting in the identification of 12 key genes (Figure 4B). In addition, the SVM-RFE algorithm identified 9 genes, achieving a peak classification accuracy of 1 and a minimum RMSE of 0 (Figure 4C). By taking the intersection of the above three gene sets with the five core targets screened from PPI, two key genes were obtained: NFKBIA and MYC (Figure 4D).

Figure 4.

A. Screening for core genes using the Least absolute shrinkage and selection operator (LASSO). B. Screening for core genes using the random forest algorithm (RF). C. Screening for core genes using the Support Vector Machine-Recursive Feature Elimination (SVM-RFE). D. Core target genes identified via integration of three machine learning (ML) methods and PPI network analysis.

GSEA

Through individual GSEA analysis performed on the two core targets, their corresponding functional attributes and associated pathways were obtained (Figure 5A and 5B). It was found that both targets were enriched in the B-cell receptor (BCR) signaling pathway, and the mechanism of the B-cell receptor pathway is illustrated in Figure 5C.

Figure 5.

A. GSEA-GO/KEGG of MYC. B. GSEA-GO/KEGG of NFKBIA. C. B-cell receptor pathway mechanisms.

Analysis of immune infiltration

The level of each immune cell infiltration in MN was assessed. It was found that naive B cells and M2 macrophages had higher proportions in MN compared with those in the control group, and their activations might be involved in the pathogenesis of MN (Figure 6A). Meanwhile, all of these immune cells were correlated with core targets to varying degrees, as presented in the corresponding heatmap (Figure 6B and 6C). MYC is strongly associated with naive B cells and mast cell activation, whereas NFKBIA exhibits a strong association with neutrophils and regulatory T cells. Additionally, both MYC and NFKBIA are strongly associated with follicular helper T cells (Tfhs).

Figure 6.

A. Immune cell infiltration analysis evaluated the abundance of 22 immune cells in control and MN groups using ssGESA in the GSE73953 dataset. B, C. Relevance of immune cells to MYC and NFKBIA. D. The performance of the MN diagnostic model. E, F. ROC analysis of MYC and NFKBIA.

ROC curve analysis

The ROC curve was plotted to demonstrate the diagnostic performance of the same model under different conditions and the diagnostic performance of the two core genes. Analysis revealed that the AUC values of the model under non-interference and interference conditions were 0.758 and 0.641, respectively (Figure 6C and 6D), indicating that the data as a whole had good diagnostic capability. The AUC values of MYC and NFKBIA were greater than 0.7 (Figure 6E and 6F), indicating that the two genes exhibit good sensitivity and specificity.

Molecular docking

The core components of BZYQ were molecularly docked with MYC and IκBα (encoded by NFKBIA), and it was found that each component had a good binding affinity with these proteins, with an average binding energy of -7.13 kcal/mol. The results were visualized (Figure 7A-F), and the heatmap is presented in Figure 7G.

Figure 7.

Visualization of molecular docking between core targets and components. A. MYC and Stigmasterol. B. IκBα and Kaempferol. C. MYC and Naringenin. D. IκBα and Licochalcone a. E. MYC and Quercetin. F. IκBα and Naringenin. G. Docking thermal diagram of the main components of BZYQ with MYC and IκBα.

Protective effect of BZYQ on MN in mice

Compared with the normal group, mice in the model group exhibited significantly increased 24h-UP (P < 0.001) and SCr (P < 0.01) levels, along with significantly decreased Alb (P < 0.001) levels, all with statistically significant differences. Although BUN levels were increased, the difference was not statistically significant (P > 0.05). In contrast, compared with the model group, mice treated with BZYQ or captopril showed significantly decreased 24h-UP (P < 0.001) and SCr (P < 0.05) levels, as well as significantly increased, Alb levels (P < 0.05), with all these changes reaching statistical significance. BUN levels were decreased, but the difference was not statistically significant (P > 0.05) (Figure 8A-D).

Figure 8.

Changes in biochemical indicators and pathological alterations in mice of each group. A. The 24-hour urine protein quantification (24-UP). B. Serum albumin (Alb). C. Creatinine (Scr). D. Blood Urea Nitrogen (BUN). ***P < 0.001, **P < 0.01 compared to the Control; #P < 0.05, ###P < 0.001 compared to the Model. E. HE, Masson, Pas staining results of mouse kidney tissue (original magnification ×400).

BZYQ attenuates kidney injury in mice

In addition, the effect of BZYQ on renal histopathology in MN mice was assessed by HE staining, Masson staining, and PAS staining. It was found that the normal group of mice had an intact structure and a normal size of the tubular lumen, while the model group showed increased glomerular volume, mesangial cell proliferation, and basement membrane thickening. The degree of pathological damage was significantly reduced in the BZYQ-treated groups (Figure 8E).

IgG deposition in mouse kidney tissue

Under identical imaging conditions, no immune complex deposition was observed in the normal group. In contrast, the model group exhibited a substantial amount of IgG deposited diffusely along the capillary loops, which displayed intense fluorescence. In the BZYQ-treated groups, the IgG deposition was diminished, and the fluorescence intensity was attenuated (Figure 9A).

Figure 9.

A. IgG deposition in mouse kidney tissue (original magnification ×400). B. Immunohistochemistry staining of MYC and IκBα (original magnification ×400). C. Protein expression levels of MYC and IκBα in kidney tissue were determined using Western blotting. Relative quantitative analysis of MYC and IκBα expression.

Effect of BZYQ on MYC and IκBα protein expression

A significant difference was observed in the expression of core proteins MYC and IκBα in normal and model tissues by immunohistochemistry. Compared with the normal group, the MYC level was increased and the IκBα level was decreased in the model group; BZYQ and Benazepril HCl significantly decreased MYC expression and increased IκBα expression compared with the model group (Figure 9B).

Western blot further validated that MYC expression levels were significantly increased (P < 0.05) in the model group compared with the control group while that of IκBα was significantly decreased (P < 0.05). However, BZYQ reversed these abnormal changes (Figure 9C).

Discussion

Currently, the clinical management of MN primarily relies on immunosuppressive therapies, including cyclophosphamide combined with corticosteroids, calcineurin inhibitors, and CD20-targeted B-cell depletion strategies. However, these interventions are frequently constrained by adverse reactions and high recurrence rates [14,15]. Notably, patients refractory to immunosuppressive agents may experience accelerated progression to end-stage renal disease (ESRD) [16]. In recent years, the efficacy of traditional Chinese medicine in improving renal function and treating MN has gained increasing recognition [17,18]. Against this backdrop, the present study aims to integrate bioinformatics, network pharmacology, and ML approaches to investigate the potential therapeutic mechanisms of the classic traditional formula BZYQ, thereby providing additional options for clinical practice.

The core components of BZYQ identified in our study include kaempferol, quercetin, stigmasterol, Licochalcone A, and naringin, among others. Experimental evidence has demonstrated that these components not only exert inhibitory effects on inflammatory responses [19-22] but also possess significant anti-fibrotic activity in various disease settings [23-25]. Such properties underscore their potential therapeutic significance in chronic kidney disease, prompting the hypothesis that these components may play a pivotal role in the therapeutic effects of BZYQ against MN.

Ultimately, MYC and NFKBIA were identified as the core targets of BZYQ in the treatment of MN. As a key nuclear transcription factor, MYC has been demonstrated to regulate the expression of numerous human genes and is extensively involved in various physiological processes, including cell growth, differentiation, and apoptosis [26]. Research has indicated that MYC overexpression significantly mediates inflammatory responses in glomerular endothelial cells in both diabetic nephropathy patients and rat models [27]. The NF-κB pathway is widely implicated in immune and inflammatory responses, and it can mediate renal inflammation in various cell types [28]. As the core negative regulator in the NF-κB signaling pathway, IκBα is capable of binding to NF-κB in the resting state and preventing its translocation to the nucleus - thereby inhibiting the transcriptional activity of NF-κB [29] and playing a key role in alleviating renal inflammation [30]. The close relationship between core targets and inflammation regulation coincides with the pathogenesis of MN [31,32]. In addition, activation of podocyte autophagy has been shown to mitigate podocyte injury and exert renal protective effects [33]. Prior studies have clarified that MYC expression regulates autophagy [34]. Moreover, MYC is highly expressed in multiple kidney diseases, and targeted reduction of MYC can exert dual effects of anti-inflammation and autophagy promotion [26,35,36]. These findings further underscore its potential as a key therapeutic target for MN.

Next, we performed single-cell enrichment analysis on two core targets, which revealed that these targets are closely associated with the BCR signaling pathway. Abnormal BCR signaling or elevated BAFF levels are known to induce B cell activation and differentiation into plasma cells, leading to the secretion of large amounts of autoantibodies and thereby promoting the progression of MN [37]. Notably, MYC expression is often accompanied by the activation of BCR signaling pathway-related proteins, including CD79α, Btk, Plcγ2, and Erk1/2 [38]. Based on these findings, we hypothesize that MYC may promote B cell survival and proliferation by enhancing the activity of the BCR signaling pathway, thereby participating in the pathogenesis of MN. Subsequent immune infiltration analysis revealed a significant association between the two core targets and Tfhs. Tfhs is involved in immune dysfunction in MN, and its subpopulation imbalance induces B cell differentiation and plays a key role in the pathological process of MN [39,40]. Additionally, ROC curve analysis demonstrated that MYC and NFKBIA exhibit favorable diagnostic performance, while molecular docking results indicated that MYC and IκBα form favorable binding interactions with the key active components of BZYQ. All these results suggest that these two core targets are promising candidates for further investigation.

Based on the above screening results, we selected female mice, which are more likely to produce typical pathological changes and clinical manifestations, to establish a model of MN [41]. Through biochemical and histopathological analysis, we found that compared with the control group, BZYQ administration significantly reduced the 24-hour urine protein level and increased the serum albumin level in mice, alleviated pathological damage, and reduced the expression of IgG, the main immune deposit antibody in MN [42]. Additionally, both IHC and WB analyses showed significant differences in the expression of two core proteins, MYC and IκBα, between the control and model groups, with positive changes observed after BZYQ administration. This suggests that BZYQ may target MYC and IκBα in the treatment of MN.

A key advantage of this study is the integration of traditional network pharmacology with ML. By analyzing data via algorithms, this integration partially addresses the limitations encountered in identifying new drug targets in network pharmacology due to technical and knowledge limitations, as well as poor-quality databases [43]. Additionally, various bioinformatics methods were employed to further validate the associations between core genes, drugs, and diseases. A limitation of this study is that the protein expression of core genes was only validated via in vivo experiments. Future research should further optimize in vitro experiments and pathway mechanism validation.

Conclusion

In summary, BZYQ exerts therapeutic effects on renal injury in MN by modulating the expression of MYC and IκBα. These findings offer valuable insights into the clinical application of BZYQ. Future studies should focus on validating these findings through experimental modeling. Additionally, an in-depth investigation of its long-term effects and potential interactions with other therapeutic agents will provide a crucial foundation for a comprehensive assessment of its clinical application potential.

Disclosure of conflict of interest

None.

Abbreviations

BZYQ

Buzhong Yiqi decoction

MN

membranous nephropathy

GEO

Gene Expression Omnibus

DEG

Differential expression gene

OB

Oral bioavailability

DL

drug-likeness

PPI

Protein-Protein Interaction

GO

Gene ontology

KEGG

Kyotoencyclopedia of genes and genomes

GSEA

Gene set enrichment analysis

ML

Machine learning

LASSO

Least absolute shrinkage and selection operator

RF

Random forest

SVM-RFE

Support Vector Machine-Recursive Feature Elimination

ROC

Receiver operating characteristic curve

AUC

Area under curve

Tfhs

follicular helper T cells