Network Pharmacology, Machine Learning, and In Vivo Validation of Danzhi Jiangtang Capsule in Diabetic Nephropathy

Summary

Danzhi Jiangtang Capsule (DJC) may attenuate diabetic nephropathy (DN), as evidenced by changes in chemokine signaling and pyroptosis markers, supported by bioinformatics and in vivo validation.

Abstract

Although Danzhi Jiangtang Capsule (DJC) is a traditional Chinese herbal preparation used clinically for diabetes, how it may protect the kidney during diabetic nephropathy (DN) has not been fully clarified. This investigation was designed to explore potential mechanisms by which DJC affects DN, with a focus on the NLR family pyrin domain-containing 3 (NLRP3)/Caspase-1/Gasdermin D (GSDMD) pyroptosis-related signaling cascade.

An integrated strategy combining network pharmacology and machine learning was employed. The Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) and the Bioinformatics Analysis Tool for Molecular Mechanism of Traditional Chinese Medicine (BATMAN-TCM) were used to screen bioactive ingredients and their corresponding protein targets of DJC. DN-associated genes were retrieved from Gene Expression Omnibus (GEO), GeneCards, and Online Mendelian Inheritance in Man (OMIM). Key candidate targets were screened and ranked using multiple machine learning algorithms. The binding affinity between DJC’s active ingredients and core targets was assessed via molecular docking. Finally, the therapeutic efficacy and predicted mechanisms were evaluated in db/db diabetic mice.

Network pharmacology analysis identified 599 DJC targets and 68 overlapping genes shared with DN. Using machine learning algorithms, C-C motif chemokine ligand 2 (CCL2) and CASP1 were identified as prioritized candidate targets. Molecular docking predicted possible strong binding affinities between DJC active ingredients and these core proteins. Functional enrichment analyses (GO/KEGG) suggested that DJC modulation is associated with inflammatory responses and the MAPK pathway. In vivo validation showed that DJC treatment attenuated renal injury and fibrosis markers. These findings suggest that DJC may attenuate DN in part through CCL2/C-C motif chemokine receptor 2 (CCR2)-related pyroptosis signaling changes.

Introduction

The global burden of diabetes mellitus (DM) continues to escalate, presenting a formidable public health challenge. An estimated 5.2 million deaths annually stem from DM and its related complications¹. DN counts as a severe microvascular complication and a major contributor to end-stage renal disease (ESRD) development, affecting approximately 30-40% of the diabetic population^2,3. Multiple intertwined factors contribute to DN pathological progression, including metabolic dysfunction, hemodynamic remodeling, and persistent inflammatory responses. Pathological insults, including high glucose and hyperlipidemia, activate the NLRP3/caspase-1/GSDMD-mediated pyroptosis pathway, inducing inflammatory programmed death in renal mesangial and tubular epithelial cells, followed by progressive tubulointerstitial lesions⁴. Pyroptosis is characterized by plasma membrane rupture, which drives the release of IL-1β and IL-18⁵. Such proinflammatory factors increase C-C motif chemokine ligand 2 (CCL2) transcription, promote macrophage infiltration, and further potentiate pyroptotic activation, thereby progressively exacerbating renal injury⁶. In clinical practice, this condition manifests with progressive albuminuria and gradual reduction of glomerular filtration rate, which can culminate in irreversible kidney damage and a marked deterioration in the quality of life for affected individuals^7,8. Despite advancements in medical care, current therapeutic strategies, which mainly focus on strict glycemic and blood pressure control, often fail to completely arrest disease progression. While emerging pharmacotherapies, including SGLT2 blocking agents, GLP-1 agonists, and mineralocorticoid receptor blockers, as well as stem cell therapies and dietary interventions, have garnered significant attention and are becoming more prevalent in practice⁹, the residual risk of DN remains high. Hence, developing innovative therapeutic substances remains an urgent research priority, particularly those from traditional medicine that may target multiple pathological pathways simultaneously.

Danzhi Jiangtang Capsule (DJC), an in-hospital-exclusive formulation developed at the First Affiliated Hospital of Anhui University of Chinese Medicine (Anhui Provincial Medical Institution Preparation Approval No. Z20090006; Patent No. ZL200310112845.1; Batch No. 20220427), is a promising therapeutic candidate. It must be sealed, stored in a dry, ventilated environment, and protected from light. The formula is composed of six traditional Chinese herbs: Pseudostellaria heterophylla (Radix Pseudostellariae), Whitmania pigra (Hirudo), Alisma orientale (Alismatis Rhizoma), Cuscuta chinensis (Cuscutae Semen), Paeonia suffruticosa (Moutan Cortex), and Rehmannia glutinosa (Rehmanniae Radix Praeparata) (5:4:4:3:2:5)¹⁰. In clinical practice, DJC has demonstrated significant hypoglycemic efficacy and the ability to reduce renal injury markers in diabetic patients¹¹. Previous research suggests that DJC may exert renoprotective effects by regulating blood glucose levels and lipid metabolism^12,13,14. Previous studies of DJC have mainly focused on single-pathway validation with unsystematic target selection. Therefore, we integrated network pharmacology and machine learning to screen active components and identify core hub genes, followed by preliminary in vivo validation, to strengthen the analysis.

Network pharmacology offers an effective strategy for dissecting complicated herbal prescriptions. This framework integrates bioinformatic and systematic biological approaches, computational chemistry, and pharmacology to decipher the "multi-component, multi-target, and multi-pathway" mechanisms of drug actions^15,16,17. This holistic approach facilitates a structured examination of the intricate interactions between herbal compounds and disease-specific networks, uncovering the synergistic effects inherent in TCM¹⁸. However, traditional network pharmacology can sometimes generate excessive noise in the data. To overcome this, integrating machine learning algorithms enables precise screening of core characteristic genes from large-scale datasets. Furthermore, molecular docking provides an intuitive, dynamic perspective for predicting potential drug-biomolecule binding affinities¹⁹.

Protocol

Screening of bioactive compounds and potential targets of DJC

Bioactive compounds of the six herbs in DJC, Pseudostellaria heterophylla (Radix Pseudostellariae), Paeonia suffruticosa (Moutan Cortex), Cuscuta chinensis (Cuscutae Semen), Alisma orientale (Alismatis Rhizoma), Rehmannia glutinosa (Rehmanniae Radix Praeparata), and Whitmania pigra (Hirudo) were retrieved from the TCMSP database. Screening was performed based on pharmacokinetic parameters: oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥ 0.18. For Whitmania pigra, which lacks data in TCMSP, its active compounds and targets were obtained from the BATMAN-TCM, using a score ≥ 20 and a P-value < 0.05 as selection criteria. UniProt was applied to unify all collected targets into a standard gene nomenclature.

Acquisition of DN-related genes

DN-related genes were collected from three sources. First, GeneCards and OMIM were searched with "diabetic nephropathy" as the search term. Genes from GeneCards with a relevance score > 1 were retained. Second, three microarray profiles (GSE30529, GSE104948, GSE96804) were obtained from the GEO repository. The samples in GSE30529 were derived from renal tubules, whereas those in GSE104948 and GSE96804 originated from renal glomeruli. All database searches were conducted on 15 December 2024. Differentially expressed genes (DEGs) between DN patients and healthy controls in the GSE30529 dataset were identified using the limma package in R software. The screening criteria were set at an adjusted p-value < 0.05 and |log2(Fold Change)| > 0.5. Volcano graphs and heatmaps displayed differentially expressed genes.

Identification of putative therapeutic targets of DJC for DN

Venn diagram analysis identified shared loci between the DJC target pool and DN-associated genes from GeneCards, OMIM, and GSE30529 DEGs. These shared loci served as candidate therapeutic genes for all downstream research procedures.

Establishment and profiling of protein interaction (PPI) networks

Putative therapeutic loci were uploaded to STRING to construct a protein-protein interaction network, with an interaction cutoff set above a 0.4 confidence value. Cytoscape processed the network datasets for visual presentation. Topological indices such as Degree were computed via the CytoNCA plugin to identify core nodes throughout the network.

Enrichment analysis

To delineate the biological roles and signaling cascades linked to candidate targets, functional enrichment was analyzed in R with the clusterProfiler package, using org.Hs.eg.db for annotation and ggplot2 for visualization. Gene Ontology (GO) terms were evaluated across biological process (BP), cellular component (CC), and molecular function (MF) categories. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were retained when the enrichment p-value was < 0.05.

Application of the CytoHubba algorithm

The CytoHubba plugin in Cytoscape was used to calculate four topological centrality indices for each network node. Degree, the count of directly connected neighbors, reflects the breadth of each protein's interactions. Maximum neighborhood component (MNC), a metric that quantifies the size of the largest connected component in the node’s neighborhood subgraph, measures local regulatory influence. The edge percolated component (EPC) assesses global structural importance via simulated random edge removal and quantification of the average connected component size. Genes ranking in the top 20 for each index were intersected to yield the final hub genes.

Machine learning-based core gene screening

Three machine learning approaches were applied to expression profiles from the GSE30529 dataset to identify reliable hub genes among candidate targets.

(1) LASSO (Least Absolute Shrinkage and Selection Operator): The regularization parameter λ was tuned via five-fold cross-validation, with the optimal value determined by average performance across folds. Input gene expression features underwent automatic Z-score standardization to eliminate effects of expression magnitude. With the penalty parameter α fixed at 1 for pure LASSO regularization, a predefined panel of 100 λ values was used for screening. The L1 penalty shrank the coefficients of low-relevance features to zero, enabling simultaneous feature selection and complexity control, while the cross-validation-tuned regularization strength mitigated the risk of overfitting.

(2) SVM-RFE (Support Vector Machine-Recursive Feature Elimination): A two-layer 5-fold cross-validation framework was employed to ensure robust feature screening. Five non-overlapping subsets were generated by stochastically dividing the entire dataset, with four used for training and one for testing across five iterations; fold-wise averaging mitigated bias from single-sample splitting. Recursive feature elimination was applied to reduce feature dimensionality and constrain model complexity, alleviating overfitting. Classification error rate and accuracy served as performance metrics to determine the optimal number of features.

(3) Random Forest (RF): An initial 500-tree random forest was constructed, with the optimal tree number determined by the minimum out-of-bag classification error rate to retrain the final model. All remaining hyperparameters, including the random feature subset size at each node split, were kept at default package settings. Gene importance scores from the final model were ranked in descending order to generate the disease signature gene set. The final set of core genes was determined by intersecting the results from these three algorithms.

Independent dataset-based verification of hub gene expression

Two independent external datasets, GSE104948 and GSE96804, were used to validate transcript levels of the hub genes. Gene expression datasets underwent normalization; expression differences between DN cases and healthy subjects were plotted using R’s ggplot2 and ggsignif toolkits.

Molecular docking

The 3D conformations of the hub target proteins (CASP1, PDB ID: 3E4C; CCL2, PDB ID: 7S00) were obtained from the Protein Data Bank. Bioactive compound 3D conformations were fetched via PubChem. Ursolic acid, 3β-hydroxyurs-12-en-28-oic acid, and quercetin were selected for docking because network pharmacology predicted interactions between these active compounds and the core targets CCL2 or CASP1. Protein structures were pretreated by stripping native ligands and water molecules, introducing polar hydrogens, and subsequently subjected to molecular docking using AutoDock Vina. The binding affinity, represented by the docking score (in kcal/mol), served to assess the strength of interactions. A binding energy threshold of −5.0 kcal/mol was set to define stable intermolecular interactions. The docked conformations were visualized using PyMOL.

Animals and experimental design

The present study used 30 8-week-old male db/db mice of specific pathogen-free (SPF) grade, along with 10 age-matched male SPF db/m mice. All experimental mice were reared in an SPF-grade facility under a 12-h light–dark photoperiod and maintained under stable conditions: 22 ± 2 °C, 50–70% relative humidity, and unrestricted access to regular chow and drinking water. Ethical approval for all in vivo experimental protocols was granted by the Anhui University of Chinese Medicine Animal Ethics Committee (Approval No. AHUCM-mouse-2023059).

Rodents underwent a 7-day adaptation phase before random grouping (6 animals each) and 8 weeks of daily intervention. Fasting blood glucose was measured in all experimental animals; animals were allocated using a random-number table, and histology, IHC, and western blot quantification were performed under blinded assessment.

Normal control (CTL) group: db/m mice receiving saline by oral gavage.

Model (M) group: db/db mice receiving saline by oral gavage.

DJC-Medium Dose (DJC-M) group: db/db mice receiving DJC at 0.78 g/kg (equivalent to the clinical daily dose for adults).

DJC-High Dose (DJC-H) group: db/db mice receiving DJC at 1.56 g/kg (twice the clinical daily dose for adults).

Dosing preparation

Human-to-murine dose scaling was derived using the body surface area normalization approach, as outlined in Methodology of Pharmacological Experiments, with a conversion coefficient of 9.1. The clinical daily dose of DJC was 6 g for a 70-kg adult (0.0857 g/kg), yielding mouse equivalent doses of 0.78 g/kg (1-fold clinical dose) and 1.56 g/kg (2-fold clinical dose). All mice received intragastric administration at a constant volume of 10 mL/kg. DJC powder was precisely weighed and suspended in 0.5% CMC-Na solution to form uniform suspensions at 78 mg/mL (medium dose) and 156 mg/mL (high dose).

After 8 weeks of daily administration, anesthesia was induced in all experimental mice using 1% sodium pentobarbital. Blood specimens were collected via the abdominal aorta and centrifuged to isolate the supernatants for kit-based biochemical assays. Metabolic cages were used to collect murine urine samples. Partial renal tissues were fixed in 4% paraformaldehyde for hematoxylin-eosin (HE) and Masson staining. Snap-frozen in liquid nitrogen, residual renal tissues were preserved at −80 °C for downstream analytical assays.

Histological analysis

Following fixation in 4% paraformaldehyde, renal tissues underwent graded ethanol dehydration, clearing, paraffin embedding, and 4-µm serial sectioning for subsequent HE and Masson’s trichrome staining. HE staining was performed following standard procedures (deparaffinization, rehydration, hematoxylin nuclear staining, acidified ethanol differentiation and blueing, eosin counterstaining, dehydration, clearing, and neutral balsam mounting) to observe renal histopathological changes. Sections were immersed in hematoxylin stain for 3 min, followed by eosin counterstaining for 15 s. For Masson’s trichrome staining, sections underwent deparaffinization, rehydration, sequential staining, differentiation, and acetic acid color separation, followed by dehydration, clearing, and mounting to assess renal collagen accumulation and fibrotic changes. Bright-field microscopy was employed to inspect and photograph all stained tissue sections at 200× magnification. HE-stained sections were scored using the Glomerular Sclerosis Index (GSI). For Masson’s trichrome staining, the Masson Trichrome preset was applied to isolate the blue collagen channel, and the ratio of collagen area to red-stained tissue area was calculated.

Immunohistochemistry (IHC)

Following 24-h fixation with 4% paraformaldehyde, tissue specimens were dehydrated through serial ethanol gradients, and then paraffin-embedded and cut into sequential 5 µm sections. After deparaffinization, sections underwent antigen retrieval and endogenous peroxidase blocking. Subsequent to blocking, tissue slides were incubated with diluted primary antibodies at 4 °C overnight in a moist chamber. Slides received a secondary antibody incubation at room temperature after PBS washes, followed by a DAB chromogenic reaction until optimal staining intensity was achieved. The sections then underwent sequential hematoxylin counterstaining, differentiation, bluing, dehydration, clearing, coverslip mounting, and bright-field microscopic imaging. Antigen retrieval was performed via autoclaving for 10 min, followed by natural cooling to room temperature. After blocking with 5% goat serum for 60 min at room temperature, sections were incubated with primary antibody (1:200 dilution) and then HRP-conjugated goat anti-rabbit IgG secondary antibody (1:300 dilution) for 60 min each at room temperature. The DAB chromogenic reaction was conducted for 3–5 min at room temperature, monitored microscopically, and terminated by rinsing with distilled water. Images were acquired at ×200 magnification, and the percentage of positive staining area was quantified using ImageJ.

ELISA detection

Assay steps complied with the official protocols supplied by kit vendors. Washing buffer and serially diluted standards were prepared in advance. Blank, standard, and sample wells were loaded with reagents and incubated in the dark. Plates were washed thoroughly after incubation, then incubated with the enzyme conjugate in the dark. After washing, a chromogenic substrate was added for a light-protected color reaction, which was terminated with stop buffer once distinct blue gradients emerged in the standard wells. The absorbance (OD) of each well was measured at the target wavelength using a microplate reader. Serum IL-1β and IL-18 concentrations were quantified using calibration curves generated from OD values of standard concentrations (undiluted serum samples; absorbance measured at 450 nm; eight technical replicate wells per sample). Urinary creatinine and 24 h urinary protein were measured in undiluted urine samples collected using 24 h metabolic cages (absorbance read at 546 nm; eight technical replicates per specimen).

Real-time quantitative PCR (RT-qPCR)

Total RNA was isolated from renal tissues, and its purity and concentration were determined by spectrophotometry. Purified RNA was reverse-transcribed to generate cDNA templates. RT-qPCR was performed on a real-time PCR instrument with the following thermal cycling program: 5 min pre-denaturation (95 °C), followed by 40 amplification rounds: 15 s at 95 °C, 1 min at 60 °C. RT-qPCR was performed in a 10 µL reaction containing 5 µL of SYBR Green Master Mix, 0.2 µL of forward primer, 0.2 µL of reverse primer, 1 µL of cDNA template, and 3.6 µL of RNase-free water. Reverse transcription for cDNA synthesis was performed in a total reaction volume of 20 µL. Melting-curve analysis was performed to confirm amplification specificity, and each sample was assayed in three technical replicates. Transcript levels were assessed by 2⁻^ΔΔCq normalization to β-actin, with all primer sequences listed in Table 1.

Western blot analysis

Roughly 40 mg renal tissue samples were harvested from each mouse cohort, minced on ice, and fully homogenized in protein lysis buffer. After a 15 min spin at 12,000 rpm at 4 °C, the total protein-containing supernatant from tissue homogenates was collected, aliquoted, and cryopreserved at −80 °C. The protein supernatant was mixed with loading buffer and denatured by boiling in a water bath. Approximately 10 µL of normalized protein sample was loaded per lane and separated on 10% SDS-polyacrylamide gels. Target proteins were transferred to nitrocellulose filters through wet electroblotting. The transfer was performed at a fixed current of 400 mA in an ice bath for 30–45 min. Post-transfer, blots were incubated in cold non-fat milk for 2 h, then incubated overnight at 4 °C with 1:1,000 primary antibodies. Washed blots were covered with matching secondary antibodies (1:10,000) at 4 °C for 2 h, and luminescent band signals were recorded and quantified from blot photos using ImageJ software. For GSDMD-N, Tubulin was used as the loading control because its expected molecular weight (~35 kDa) overlaps with that of GAPDH (~36 kDa). All western blot experiments were performed with three independent biological replicates.

Statistical analysis

Quantitative data are presented as the mean ± standard deviation (SD). Differences between two groups were analyzed with an unpaired Student’s t-test, whereas comparisons among multiple groups were evaluated by one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test. A p-value < 0.05 was considered statistically significant.

Results

DJC bioactive component and target identification

A total of 66 bioactive compounds that met the screening criteria (OB ≥ 30% and DL ≥ 0.18) were identified from the six herbs comprising DJC. These included 26 compounds from Pseudostellaria heterophylla, 6 from Paeonia suffruticosa, 8 from Cuscuta chinensis, 6 from Alisma orientale, 10 from Rehmannia glutinosa, and 10 from Whitmania pigra. Standardization of tar...

Discussion

DN remains a leading cause of ESRD, characterized by a complex and multifactorial pathogenesis that is not yet fully understood^20,21. Although DJC has shown promising clinical efficacy in ameliorating DN, the underlying molecular mechanisms have remained largely elusive²². In this study, we used a multi-omics analytical scheme integrating network pharmacology, machine learning tools, binding simulation assays, and in vivo verifica...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Creatinine (CRE) Assay Kit	Nanjing Jiancheng Bioengineering Institute	C011-2-1
ECL	Biosharp	BL520B
Eosin Staining Kit	Servicebio	G1002
Goat anti-rabbit IgG-HRP Conjugate	Abbkine	A21020
Hematoxylin Staining Kit	Servicebio	G1004
Masson's Trichrome Stain Kit	Servicebio	G1006
Mouse IL-1β ELISA Kit	MeiMian	MM-0040M1
Mouse IL-18 ELISA Kit	MeiMian	MM-0169M1
Rabbit anti-mouse Caspase1 Antibody	Affinity	AF5418
Rabbit anti-mouse CCL2 Antibody	Abcam	ab315478
Rabbit anti-mouse cleaved-Caspase-1 Antibody	Affinity	AF4005
Rabbit anti-mouse GAPDH Antibody	Zenbio	R380626
Rabbit anti-mouse GSDMD-N Antibody	Abcam	ab219800
Rabbit anti-mouse NLRP3 Antibody	Abcam	ab263899
Rabbit anti-mouse Tubulin Antibody	Affinity	AF7011
Reverse Transcription Kit	Biosharp	BL696A
RIPA Lysis Buffer	Beyotime Biotechnology	P0013
RIPA Lysis Buffer	Leagene	PS0013
RNA isolation reagent	Servicebio	G3013-100ML
SYBR Green Master Mix	Servicebio	G3326-05
Urinary Protein Quantification Kit	Nanjing Jiancheng Bioengineering Institute	C035-2-1
software name	version
AutoDock Vina	1.2.3
Cytoscape software	3.10.0
GraphPad Prism	9.0
ImageJ	java 1.8.0_345
PyMOL	2.5
R software	4.2.0