Introduction

Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection and remains a major global health challenge.1,2 Among its pulmonary complications, sepsis-induced acute lung injury (SALI), which may progress to acute respiratory distress syndrome (ARDS), is characterized by diffuse inflammatory lung injury, increased pulmonary endothelial and epithelial permeability, pulmonary edema, hypoxemia, and a loss of aerated lung tissue.3 Sepsis-associated pulmonary inflammation involves complex interactions among alveolar and interstitial macrophages, neutrophils, pulmonary endothelial cells, and alveolar epithelial cells. In particular, activated macrophages can amplify inflammatory cytokine production, oxidative stress, endothelial dysfunction, and disruption of the alveolar–capillary barrier.4–6 However, the mechanisms governing macrophage activation and regulated cell-death pathways in SALI remain incompletely understood, thereby limiting the development of effective targeted therapies.

Accumulating evidence indicates that ferroptosis contributes functionally to the pathogenesis of SALI rather than merely serving as a secondary marker of oxidative injury. Ferroptosis is a regulated form of cell death characterized by iron-dependent phospholipid peroxidation resulting from disrupted iron homeostasis and impaired antioxidant defenses.7–9 In SALI, dysregulation of GPX4-dependent lipid peroxide detoxification, SLC7A11-mediated cystine uptake, ACSL4-dependent phospholipid remodeling, and mitochondrial redox metabolism may influence the severity of pulmonary inflammation, vascular permeability, and tissue injury.10–12 Accordingly, interventions that preserve or restore anti-ferroptotic defenses have demonstrated protective effects in recent models of CLP- and LPS-induced lung injury.

Matrine is a bioactive quinolizidine alkaloid derived from Sophora flavescens, commonly known as Kushen (KS), and possesses anti-inflammatory, antioxidant, and immunomodulatory properties.13,14 Previous studies have demonstrated the protective effects of matrine in liver fibrosis, cardiovascular injury, and neuroinflammation.15–17 Emerging evidence further suggests that matrine can modulate ferroptosis-related processes. In a model of severe acute pancreatitis-induced acute lung injury, matrine was reported to activate the mitochondrial UCP2/SIRT3/PGC-1α signaling pathway, reduce mitochondrial reactive oxygen species (ROS) production, restore GPX4 expression and glutathione levels, and suppress lipid peroxidation, thereby attenuating ferroptosis-associated injury and improving lung histopathology.18 These findings suggest that upstream regulators of redox homeostasis may contribute to the anti-ferroptotic effects of matrine.

To systematically investigate the molecular mechanisms underlying the protective effects of matrine against SALI, we integrated single-cell RNA sequencing, network pharmacology, transcriptomic bioinformatics, molecular docking, molecular dynamics simulations, surface plasmon resonance (SPR), the cellular thermal shift assay (CETSA), in vivo and in vitro experiments, and pharmacological inhibition of NQO1. The therapeutic effects of matrine were evaluated in mice with CLP-induced SALI and in LPS-stimulated MH-S cells, whereas its binding to NQO1 and intracellular target engagement were assessed using complementary biophysical and cellular approaches. The NQO1 inhibitor ES936 was subsequently used to examine the functional contribution of NQO1 to the protective effects of matrine. This multilevel strategy provided evidence that matrine reinforces NQO1/SLC7A11/GPX4-associated anti-ferroptotic defenses and attenuates NF-κB-driven inflammation, while identifying NQO1 as a functionally relevant candidate target. The overall study workflow is presented in Figure 1.

Figure 1 Outline Diagram.

Materials and Methods Animals

Male C57BL/6 mice aged 6–8 weeks and weighing 20–25 g were obtained from the Hubei Provincial Center for Animal Experimentation. The mice were housed in groups under controlled environmental conditions, including a 12 h light/dark cycle, an ambient temperature of 22 ± 2°C, and a relative humidity of 55 ± 5%. Standard rodent chow and sterilized water were available ad libitum. All animal procedures were reviewed and approved by the Institutional Animal Care and Use Committee of Wuhan University (approval No. WDRY2022-K046) and were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. For surgical procedures, mice were anesthetized with sodium pentobarbital (50 mg/kg, intraperitoneally; Sigma, USA), and an adequate depth of anesthesia was confirmed by the absence of the pedal withdrawal reflex. For terminal sample collection, the mice were deeply anesthetized with sodium pentobarbital using the same dose and route of administration. After the absence of the pedal withdrawal reflex had been confirmed, the mice were euthanized by cervical dislocation under deep anesthesia, and the required tissues were immediately collected.

CLP-Induced Sepsis, Drug Administration, and Dose Selection

Sepsis was induced using the cecal ligation and puncture (CLP) model under anesthesia with sodium pentobarbital (50 mg/kg, intraperitoneally; Sigma, USA). Before surgery, an adequate depth of anesthesia was confirmed by the absence of the pedal withdrawal reflex. A midline laparotomy was performed to expose the cecum, and the distal portion of the cecum was ligated with a 4–0 silk suture. The cecum was then punctured once through both walls using a 21-gauge needle, and a small amount of fecal material was gently extruded through the puncture site to ensure patency. The cecum was subsequently returned to the abdominal cavity, and the abdominal wall was closed in two layers. Fluid resuscitation was provided by the subcutaneous administration of prewarmed sterile saline. Sham-operated mice underwent the same laparotomy and cecal manipulation without ligation or puncture.

To evaluate the therapeutic effects of matrine and select an appropriate dose for subsequent experiments, mice subjected to CLP received a single dose of matrine (MAT; C15H24N2O; molecular weight, 248.36 g/mol; purity ≥98.0%; MedChemExpress, HY-N0164) at 50 or 100 mg/kg by oral gavage 2 h after CLP.19 These groups were designated the low-dose matrine group (MAT-L) and high-dose matrine group (MAT-H), respectively. Dexamethasone (DEX; 1 mg/kg) was included as a positive control, and the corresponding vehicle-control mice received an equivalent volume of phosphate-buffered saline (PBS). Dexamethasone was selected as a reference anti-inflammatory and lung-protective agent because of its established use in experimental models of sepsis and acute lung injury.20 At 24 h after CLP, mice were assigned to separate cohorts for BALF collection, lung wet-to-dry weight determination, or lung tissue assessment. BALF was obtained under deep anesthesia before euthanasia, whereas non-lavaged mice were euthanized under deep anesthesia and their lungs were immediately harvested for gravimetric, histopathological, or molecular evaluation. Matrine administration 2 h after CLP was selected to model an early therapeutic intervention,21 whereas the 24 h sampling time point was chosen to evaluate acute SALI-related pathological and molecular alterations,22,23 consistent with previous studies using CLP-induced acute lung injury models. Among the doses tested, 100 mg/kg produced the most consistent improvements in BALF protein concentration, pulmonary edema, and histopathological lung injury without overt signs of toxicity and was therefore selected for subsequent in vivo experiments.

Cell Culture and Drug Treatment

The murine alveolar macrophage cell line MH-S was obtained from the American Type Culture Collection (ATCC; CRL-2019). For biochemical and immunoblotting analyses, MH-S cells were seeded in 6-well plates at a density of 3×105 cells per well and cultured for 24 h until they reached approximately 80–85% confluence. Before LPS stimulation, the cells were pretreated with matrine (400 μM; MedChemExpress, China) dissolved in sterile PBS for 2 h.24 LPS (1 μg/mL; Sigma-Aldrich, USA) was subsequently added, and the cells were incubated for an additional 24 h.25 Vehicle-control cells received an equivalent volume of PBS under the same conditions.

Cell Viability Assay

MH-S cells were seeded in 96-well plates at a density of 1×104 cells per well and allowed to adhere for 24 h. After the indicated treatments, 10 μL of Cell Counting Kit-8 (CCK-8) reagent (Solarbio, CA1210, China) was added to each well, and the plates were incubated at 37°C for 2 h. Absorbance was measured at 450 nm using a microplate reader. Cell viability was calculated relative to that of the corresponding vehicle-treated control and expressed as a percentage.

Single-Cell RNA-Sequencing Analysis

The publicly available single-cell RNA-sequencing (scRNA-seq) dataset GSE273924 was analyzed to characterize immune-cell heterogeneity in murine lung tissue. According to the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) record and the associated publication, the dataset was generated from CD45-enriched pulmonary immune cells isolated from sham mice and mice with intratracheal Escherichia coli-induced pneumonia.26 Therefore, all analyses of cell-type composition were restricted to the CD45-enriched pulmonary immune-cell compartment. Count matrices were imported into Seurat version 5 for preprocessing and downstream analysis. Cells that did not meet the predefined quality-control criteria were excluded. The retained data were subsequently normalized, and highly variable genes were identified. Principal component analysis (PCA) was performed for dimensionality reduction, and batch effects were mitigated using Harmony. After clustering, uniform manifold approximation and projection (UMAP) was used to visualize the cellular landscape. The expression patterns of Slc7a11, Gpx4, Acsl4, and Nqo1 across the identified immune-cell populations were visualized using feature plots based on the same UMAP embedding. To assess transcriptional differences between experimental groups within defined immune-cell subsets, gene counts were aggregated at the sample level using a pseudobulk approach, followed by differential expression analysis with edgeR. Differentially expressed genes (DEGs) were visualized using volcano plots and heatmaps.

Identification of Bioactive Constituents and Potential Targets of Kushen

A preliminary search of the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP; https://old.tcmsp-e.com/tcmsp.php) identified 45 chemical constituents of Kushen (KS) that satisfied the prespecified thresholds of oral bioavailability (OB) ≥30% and drug-likeness (DL) ≥0.18. The pharmacokinetic and drug-likeness properties of these compounds were then reassessed using SwissADME (http://www.swissadme.ch/). Compounds predicted to exhibit high gastrointestinal absorption and to pass at least two drug-likeness filters were included in the subsequent analyses, resulting in the retention of 25 candidate bioactive constituents. Potential targets associated with these compounds were further predicted using SwissTargetPrediction (http://www.swisstargetprediction.ch/) and PharmMapper (https://www.lilab-ecust.cn/pharmmapper/).

Identification of Sepsis- and SALI-Related Targets

Genes associated with sepsis and SALI were systematically retrieved from four databases: the Therapeutic Target Database (TTD; https://db.idrblab.net/ttd/), Online Mendelian Inheritance in Man (OMIM; https://www.omim.org/), GeneCards (https://www.genecards.org/), and DisGeNET (https://www.disgenet.org/). The target lists obtained from these databases were merged, and duplicate entries were removed to generate a nonredundant set of sepsis- and SALI-related targets for subsequent analyses.

Identification of Differentially Expressed Genes

The transcriptomic dataset GSE245013 was obtained from the GEO. Of the 20 murine samples included in the dataset, five sham-operated samples and five samples from mice with CLP-induced sepsis were included in the comparative analysis. Differential expression analysis was performed using the limma package through the Sangerbox platform (https://sangerbox.com/home.html). Genes with a false discovery rate (FDR) <0.05 and an absolute log2 fold change (|log2FC|) >1 were considered differentially expressed.

Weighted Gene Co-Expression Network Analysis

Weighted gene co-expression network analysis was performed using the WGCNA package (version 1.74). After sample and gene quality assessment using goodSamplesGenes, the normalized expression matrix containing 17,980 genes from 10 samples was used for network construction. Candidate soft-thresholding powers from 1 to 30 were evaluated using pickSoftThreshold. A signed scale-free topology fit cutoff of R2 ≥ 0.85 was predefined. A power of β = 10 was retained because it yielded a signed scale-free topology fit index of 0.884 while maintaining a mean connectivity of 1,129.79. An unsigned adjacency network and unsigned topological overlap matrix were constructed using Pearson correlation. Modules were detected using blockwiseModules with a maximum block size of 5,000 genes, minimum module size of 30, deepSplit = 2, and mergeCutHeight = 0.25. Module eigengenes were subsequently correlated with the CLP and sham traits.

GO and KEGG Enrichment Analyses

Functional enrichment analysis of the overlapping targets was performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID; https://david.ncifcrf.gov/). Gene Ontology (GO) enrichment analysis included biological process (BP), cellular component (CC), and molecular function (MF) categories, whereas Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was used to identify enriched signaling pathways. The enrichment results were visualized using the Bioinformatics Analysis Platform (https://www.bioinformatics.com.cn/) to characterize the biological functions and pathways potentially associated with the effects of KS on SALI.

Identification of Core Targets

The overlapping targets were uploaded to the STRING database (version 12.0; https://cn.string-db.org/) for protein–protein interaction (PPI) analysis. The resulting PPI network was imported into Cytoscape (version 3.10.1) for visualization and topological analysis. Hub targets were identified using the Molecular Complex Detection (MCODE) plugin and subsequently intersected with ferroptosis-associated genes retrieved from FerrDb, yielding four final core targets.

Molecular Docking

The three-dimensional structures of the four core targets—NQO1 (PDB ID: 1D4A), NOX4 (AlphaFold DB ID: AF-Q9NPH5-F1), NOX1 (AlphaFold DB ID: AF-Q9Y5S8-F1), and LCN2 (PDB ID: 4MVK)—were obtained from the RCSB Protein Data Bank27 and the AlphaFold Protein Structure Database.28 The protein structures were prepared using PyMOL (version 2.3.0). The three-dimensional structure of matrine, the principal bioactive compound selected for validation, was retrieved from PubChem (CID: 91466) and subjected to energy minimization using the Merck Molecular Force Field 94 (MMFF94) implemented in Open Babel (version 3.1.1). Receptor and ligand files were prepared using AutoDockTools (version 1.5.6), and semi-flexible molecular docking was performed using AutoDock Vina (version 1.2.5). Five independent docking runs were conducted for each matrine–target pair. The predicted binding energies were expressed as the mean ± standard deviation (SD). The estimated inhibition constant (Ki) and negative logarithm of the inhibition constant (pKi) were derived from the predicted binding energy using the equations described below.29

(1)

(2)

(3)

Molecular Dynamics Simulations

Molecular dynamics (MD) simulations were performed for the matrine–NQO1 complex and the apo-NQO1 system using GROMACS version 2024.4. The protein and ligand were parameterized using the AMBER14SB and General Amber Force Field 2 (GAFF2) force fields, respectively. Each system was solvated in a periodic box containing TIP4P water molecules and neutralized by adding Na+ and Cl− ions. Following energy minimization and sequential equilibration, a 100 ns production simulation was conducted with a 2 fs integration time step under periodic boundary conditions.

Trajectory analyses included the root-mean-square deviation (RMSD), root-mean-square fluctuation (RMSF), radius of gyration (Rg), solvent-accessible surface area (SASA), hydrogen-bond occupancy, and binding free-energy estimation. Representative structures were extracted at 0, 25, 50, 75, and 100 ns to evaluate the temporal evolution and conformational stability of the matrine–NQO1 complex.

Surface Plasmon Resonance

Surface plasmon resonance (SPR) analysis was performed to determine the binding affinity between matrine and NQO1. An activation solution was freshly prepared by combining 400 mM 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (EDC) and 100 mM N-hydroxysuccinimide (NHS) immediately before use. A CM5 sensor chip was activated with the EDC/NHS solution for 420 s at a flow rate of 10 μL/min. NQO1 was diluted to 20 μg/mL in immobilization buffer and injected over the active flow cell (Fc2) at 10 μL/min, resulting in an immobilization level of approximately 12,600 response units (RU). The reference flow cell (Fc1) underwent the same activation and deactivation procedures without NQO1 immobilization. The remaining activated carboxyl groups were blocked with 1 M ethanolamine hydrochloride for 420 s at 10 μL/min.

Matrine was prepared in analyte buffer at seven concentrations ranging from 0.16 to 10 μM and injected over both Fc1 and Fc2 at a flow rate of 20 μL/min. Each injection consisted of a 100 s association phase followed by a 180 s dissociation phase in analyte buffer. Seven injection cycles were performed in ascending order of matrine concentration, and the sensor surface was regenerated after each cycle.

Cellular Thermal Shift Assay

A cellular thermal shift assay (CETSA) was performed to evaluate the intracellular target engagement of NQO1 by matrine. MH-S cells were treated with matrine (400 μM) or an equivalent volume of PBS (vehicle control) for 2 h and then divided equally into eight aliquots. The aliquots were heated at temperatures ranging from 37°C to 72°C in 5°C increments, with each temperature maintained for 3 min. After heating, the samples were cooled, lysed, and centrifuged to remove thermally precipitated proteins. NQO1 abundance in the soluble protein fractions was determined by immunoblotting. The experiment was performed using five independent biological replicates.

Histological Analysis

Paraffin-embedded lung sections were mounted on poly-L-lysine-coated slides, deparaffinized in xylene, and rehydrated through a graded ethanol series. For hematoxylin and eosin (H&E) staining, the sections were stained with Mayer’s hematoxylin, differentiated in 0.3% acid alcohol, blued in Scott’s tap water substitute, and counterstained with eosin Y. For Masson’s trichrome staining, the sections were sequentially stained with Weigert’s iron hematoxylin and Biebrich scarlet–acid fuchsin, differentiated in phosphomolybdic–phosphotungstic acid, and counterstained with aniline blue. After dehydration and clearing, the stained sections were mounted with Permount. Histopathological lung injury was evaluated using a standardized scoring system.

Elisa

The concentrations of IL-1β, IL-6, and TNF-α were measured using commercially available enzyme-linked immunosorbent assay (ELISA) kits (Jianglai Bio, China) according to the manufacturer’s instructions.

Reverse Transcription Quantitative PCR

Total RNA was extracted from lung tissues using TRIzol reagent (Invitrogen, USA), and complementary DNA (cDNA) was synthesized using a reverse transcription kit (Vazyme, China). Quantitative PCR was performed using a CFX384 Touch Real-Time PCR Detection System (Bio-Rad, USA). Relative mRNA expression was calculated using the 2^−ΔΔCt method, with β-actin as the reference gene. The primer sequences are listed in Table 1.

Table 1 Quantitative PCR Primer Sequences

Western Blotting

Total protein was extracted from lung tissue homogenates and MH-S cell lysates, and protein concentrations in the clarified supernatants were determined using a BCA assay kit (Beyotime, China). Proteins were separated on 10% SDS–PAGE gels (Epizyme, China) and transferred to PVDF membranes. The membranes were blocked with rapid blocking buffer (Epizyme, China) and incubated overnight at 4°C with the indicated primary antibodies. After washing, the membranes were incubated with HRP-conjugated secondary antibodies (1:10,000; ABclonal or Proteintech). Protein bands were detected using an enhanced chemiluminescence substrate (Abbkine, China) and imaged with a ChemiDoc Imaging System (Bio-Rad, USA). Band intensities were quantified using ImageJ software. ACSL4, GPX4, SLC7A11, NQO1, and total NF-κB p65 levels were normalized to β-actin, whereas phosphorylated NF-κB p65 was normalized to total NF-κB p65.

The primary antibodies used were as follows: anti-ACSL4 (#22401-1-AP, Proteintech), anti-GPX4 (#67763-1-Ig, Proteintech), anti-SLC7A11 (#26864-1-AP, Proteintech), anti-NQO1 (#67240-1-Ig, Proteintech), anti-phospho-NF-κB p65 (p-NF-κB, Ser536, #TP56372F, Abmart), anti-NF-κB p65 (#T55034F, Abmart), anti-β-actin (#AC026, ABclonal).

Immunohistochemical Analysis

Paraffin-embedded lung sections were deparaffinized, rehydrated through a graded ethanol series, and subjected to antigen retrieval. The sections were then blocked with 5% bovine serum albumin (BSA) and incubated overnight at 4°C with the following primary antibodies: anti-F4/80 (#29414-1-AP, Proteintech), anti-NQO1 (#67240-1-Ig, Proteintech), and anti-phospho-NF-κB p65 (#PC0982, Abmart). After washing, the sections were incubated with the appropriate species-specific secondary antibodies (1:500; MXB Biotechnologies, China) at room temperature. Immunoreactivity was visualized using 3,3′-diaminobenzidine (DAB), and the nuclei were counterstained with Mayer’s hematoxylin. The stained sections were examined under a light microscope.

Immunofluorescence Staining

Lung sections were deparaffinized in xylene, rehydrated through a graded ethanol series, and subjected to heat-induced antigen retrieval. After permeabilization with 0.3% Triton X-100, nonspecific binding sites were blocked with 5% BSA at room temperature. The sections were then incubated overnight at 4°C with anti-NQO1 (#67240-1-Ig, Proteintech) and anti-phospho-NF-κB p65 (#PC0982, Abmart) primary antibodies. After washing with PBS, the sections were incubated with appropriate fluorophore-conjugated secondary antibodies and counterstained with DAPI to visualize the nuclei. The sections were mounted using an antifade mounting medium and examined under a fluorescence microscope.

Biochemical Assays

Lung tissue homogenates and MH-S cell lysates were analyzed using commercially available kits according to the respective manufacturers’ instructions. Reduced glutathione (GSH), malondialdehyde (MDA), and ferrous iron (Fe2+) levels were measured using a Reduced Glutathione Micro-Assay Kit (Nanjing Jiancheng Bioengineering Institute, China), a Lipid Peroxidation MDA Assay Kit (Beyotime, China), and a Total/Ionic Iron Quantification Kit (Beyotime, China), respectively.

Pharmacological Inhibition of NQO1

To examine the functional contribution of NQO1 to the protective effects of matrine, MH-S cells were treated with ES936, a mechanism-based NQO1 inhibitor (MedChemExpress, HY-100367). ES936 was dissolved in dimethyl sulfoxide (DMSO) and diluted to the indicated concentrations in complete culture medium. The final concentrations of PBS and DMSO were matched across all experimental groups.

To select a non-cytotoxic working concentration, cells were exposed to ES936 at 25, 50, 100, 200, or 500 nM for 24 h, with vehicle-treated cells serving as the control. Cell viability was then evaluated using the CCK-8 assay. Based on the absence of detectable cytotoxicity and previously reported conditions for cellular NQO1 inhibition, 100 nM ES936 was selected for subsequent experiments. This concentration and a 30 min pretreatment period were consistent with previous studies showing that 100 nM ES936 inhibited more than 95% of cellular NQO1 activity within 30 min.30–32

For mechanistic experiments, cells were pretreated with ES936 (100 nM) for 30 min, followed by the addition of matrine (400 μM) for 2 h. LPS (1 μg/mL) was then added without replacing the culture medium, and the cells were incubated for an additional 24 h. ES936 and matrine remained present throughout the period of LPS stimulation.

Statistical Analysis

Statistical analyses were performed using GraphPad Prism version 10.1.2 (GraphPad Software, USA). Data are presented as the mean ± SD, and n denotes independent biological replicates; technical replicates were not included. Normality and homogeneity of variance were assessed using the Shapiro–Wilk and Levene’s tests, respectively. Two-group comparisons were performed using an unpaired two-tailed Student’s t-test or Mann–Whitney U-test, as appropriate. Comparisons among three or more groups were performed using one-way ANOVA followed by Tukey’s test or the Kruskal–Wallis test followed by Dunn’s test. Survival curves were analyzed using the Kaplan–Meier method and Log rank test. All tests were two-sided, and P < 0.05 was considered statistically significant. Significance was defined as follows: ns, not significant; *P < 0.05, **P < 0.01 and ***P < 0.001.

Results Single-Cell RNA Sequencing Reveals Ferroptosis-Associated Programs and Inflammatory Activation in CD45-Enriched Pulmonary Immune Cells

To characterize immune-cell composition and transcriptional programs potentially relevant to SALI, we reanalyzed the publicly available scRNA-seq dataset GSE273924. Following stringent quality-control filtering, the distributions of detected gene numbers, unique molecular identifier (UMI) counts, and mitochondrial transcript proportions were within the predefined ranges, supporting the suitability of the retained cells for downstream analyses (Figure S1A). Unsupervised clustering and UMAP identified the major immune-cell populations within the CD45-enriched pulmonary compartment, including neutrophils, macrophages, CD8+ T cells, CD4+ T cells, natural killer cells, and B cells (Figure 2A). Cell-type annotations were further confirmed using the expression patterns of canonical marker genes (Figure 2B). To further examine the cell-type specificity of ferroptosis-associated transcriptional programs during pneumonia, we assessed the single-cell expression patterns of Slc7a11, Gpx4, Acsl4, and Nqo1 across the CD45-enriched pulmonary immune-cell landscape. UMAP feature plots revealed distinct cell-type-dependent and heterogeneous expression patterns for these genes (Figure 2C). Gpx4 was broadly expressed across multiple immune-cell populations, whereas Slc7a11 and Acsl4 showed relatively restricted expression patterns in specific immune-cell populations; Nqo1 expression was predominantly confined to a small subset of cells.

Figure 2 Single-cell RNA sequencing identifies ferroptosis-associated programs and inflammatory activation in CD45-enriched pulmonary immune cells from GSE273924. (A) UMAP visualization of the major CD45-enriched pulmonary immune-cell populations. (B) Marker genes used for cell-type annotation. (C) UMAP feature plots showing the single-cell expression patterns of Slc7a11, Gpx4, Acsl4, and Nqo1 across the major CD45-enriched pulmonary immune-cell populations. Color intensity indicates the normalized expression level of each gene. (D) Comparison of CD45-enriched pulmonary immune-cell distributions between sham samples (left) and samples from mice with intratracheal E. coli-induced pneumonia (right). (E) Volcano plot showing differential gene expression in macrophages between sham samples and samples from mice with intratracheal E. coli-induced pneumonia. (F) KEGG enrichment analysis of macrophage DEGs. (G) GO biological process enrichment analysis of macrophage pseudobulk data, showing significant enrichment of terms associated with macrophage activation, inflammatory signaling, and ferroptosis-associated processes. Dot size represents the gene count, and dot color represents −log10(P). (H) GSEA of macrophage pseudobulk data showing enrichment of the “response to oxidative stress” gene set.

Comparison of sham samples with samples from mice with intratracheal Escherichia coli-induced pneumonia revealed marked remodeling of the CD45-enriched pulmonary immune-cell landscape. This remodeling was characterized by increased representation of neutrophils and macrophages, together with redistribution among T-cell and natural killer-cell subsets (Figure 2D). Macrophage-specific pseudobulk differential expression analysis further revealed extensive transcriptional alterations, including the upregulation of numerous inflammation- and metabolism-related genes in the pneumonia-model samples (Figures 2E and S1B).

Pathway enrichment analysis identified NF-κB, TNF, Toll-like receptor, and MAPK signaling among the prominently enriched inflammatory pathways, together with significant enrichment of the ferroptosis pathway (Figure 2F). GO biological process analysis further highlighted terms associated with macrophage activation, NF-κB-related regulation, and ferroptosis-associated processes (Figures 2G and S1C). Consistently, gene set enrichment analysis indicated enrichment of gene sets related to oxidative stress responses and ferroptosis (Figures 2H and S1D).

Collectively, these findings demonstrate substantial remodeling of the CD45-enriched pulmonary immune-cell compartment in mice with intratracheal E. coli-induced pneumonia, with prominent inflammatory and ferroptosis-associated transcriptional changes, particularly in macrophages. These exploratory findings provided an immune-cell-specific molecular context for subsequent experimental validation in CLP-induced SALI.

Network Pharmacology Analysis of Potential Kushen Targets in SALI

A total of 45 Kushen (KS)-associated constituents were initially retrieved from the TCMSP database using the predefined criteria of oral bioavailability and drug-likeness. Further screening with SwissADME retained 25 candidate bioactive constituents (Table S1). Target prediction using SwissTargetPrediction and PharmMapper yielded 611 nonredundant protein targets after duplicate entries were removed.

In parallel, 3,235 targets associated with sepsis and SALI were compiled from the TTD, OMIM, GeneCards, and DisGeNET databases. Intersection analysis identified 313 common targets shared between the KS-associated and SALI-associated target sets (Figure S2A). These overlapping targets were imported into Cytoscape version 3.10.1 to construct a constituent–target–disease network (Figure S2B). In the network, KS, candidate constituents, SALI, and protein targets were represented by purple, blue, light-coral, and turquoise nodes, respectively.

To examine potential protein–protein interactions, the 313 common targets were submitted to the STRING database for network construction (Figure S2C). After four disconnected nodes were removed, the resulting protein–protein interaction network contained 309 nodes and 5,162 edges. Network topology was evaluated using the NetworkAnalyzer tool in Cytoscape. Node size and color intensity were scaled according to centrality values, with larger and more intensely colored nodes indicating potentially greater topological importance within the network (Figure S2D).

Functional enrichment analysis of the 313 common targets was performed using the DAVID platform and included GO enrichment across the biological process (BP), cellular component (CC), and molecular function (MF) categories, as well as KEGG pathway enrichment analysis (Figure S2E–F). The enriched GO terms included “inflammatory response,” “response to lipopolysaccharide,” and “protein phosphorylation,” whereas KEGG analysis identified ferroptosis and NF-κB signaling among the significantly enriched pathways. Collectively, these findings suggest that the potential effects of KS on SALI may involve multiple targets and biological processes associated with ferroptosis and inflammatory signaling.

Identification of Key Targets Through Integrated Differential Expression and WGCNA

Differential expression analysis of the GSE245013 dataset was performed using the limma package, with FDR < 0.05 and |log2FC| > 1 as the significance thresholds. Among the 22,404 genes analyzed, 2,127 were differentially expressed, including 1,037 upregulated and 1,090 downregulated genes. PCA was used to assess the overall variation and distribution of the samples (Figure S3A). The distribution and expression patterns of the differentially expressed genes (DEGs) were visualized using a volcano plot and a hierarchical clustering heatmap, respectively (Figure S3B–C).

To further characterize gene co-expression patterns, WGCNA was performed. A soft-thresholding power of β = 10 was selected to approximate scale-free network topology (Figure S3D). Genes were subsequently assigned to distinct co-expression modules (Figure S3E), and the relationships between module eigengenes and the SALI phenotype were evaluated (Figure S3F). Among the identified modules, the darkgrey module exhibited the strongest positive correlation with the SALI phenotype (r = 0.82), whereas the blue module showed the strongest negative correlation (r = −0.97).

Intersection of the 2,127 DEGs with the 313 targets shared between the KS-associated and SALI-associated target sets yielded 66 overlapping genes (Figure 3A). These genes were submitted to the STRING database to construct a PPI network, which was subsequently visualized using Cytoscape (Figure 3B and C). GO enrichment analysis identified terms associated with inflammatory and redox-related functions, including “response to lipopolysaccharide,” “inflammatory response,” “NADPH oxidase complex,” and “nuclear receptor activity” (Figure 3D). KEGG pathway analysis further highlighted ferroptosis and NF-κB signaling as key pathways implicated in SALI (Figure 3E).

Figure 3 Identification of candidate targets and prioritization of ferroptosis-associated targets. (A) Venn diagram showing the intersection between 2,127 DEGs from GSE245013 and 313 targets shared between the KS-associated and SALI-associated target sets, yielding 66 overlapping genes. (B) PPI network of the 66 overlapping genes constructed using the STRING database. (C) Cytoscape visualization of the PPI network. Node size and color intensity represent the degree of each node. (D) GO enrichment analysis of the 66 overlapping genes, classified into BP, CC, and MF categories. (E) KEGG pathway enrichment analysis of the 66 overlapping genes. The color gradient represents enrichment significance, and the red boxes highlight the ferroptosis and NF-κB signaling pathways. (F) MCODE clustering of the PPI network showing three densely connected clusters. The colored boxes and arrows indicate different MCODE clusters; the MCODE scores for Clusters 1, 2, and 3 were 10.235, 4, and 3, respectively. (G) Subnetwork of Cluster 1 highlighting highly connected hub targets. (H) Venn diagram showing the intersection between 18 hub targets from Cluster 1 and 484 ferroptosis-associated genes from FerrDb. The red box highlights the four final ferroptosis-associated core targets: NQO1, NOX4, NOX1, and LCN2.

Molecular Complex Detection (MCODE) analysis was applied to the PPI network generated from the 66 overlapping genes and identified three densely connected clusters (Figure 3F). Cluster 1 had the highest MCODE score of 10.235, indicating the greatest topological density among the identified clusters. The interaction network of Cluster 1 was further visualized to illustrate the relationships among its constituent targets (Figure 3G). Given the potential relevance of ferroptosis to SALI, the 18 genes represented in Cluster 1 were intersected with 484 ferroptosis-related genes retrieved from the FerrDb database. This analysis identified four shared genes—NQO1, NOX4, NOX1, and LCN2 (Figure 3H). These genes were subsequently prioritized as ferroptosis-associated candidate targets potentially relevant to the effects of KS in SALI.

Molecular Docking Analysis of KS Constituents and Ferroptosis-Associated Candidate Targets

Based on the preceding computational analyses, molecular docking was performed between the 25 candidate bioactive constituents of KS and four ferroptosis-associated targets: NQO1, NOX4, NOX1, and LCN2. The resulting docking-score matrix was visualized as a heatmap (Figure S4A). Within the applied docking framework, most constituents exhibited more favorable predicted docking scores for NQO1, NOX4, and NOX1 than for LCN2. The complete docking-score rankings of the 25 KS constituents against the four targets are provided in Table S2. Ranking based on the mean predicted docking score across the four targets indicated that matrine had the most favorable overall score among the 25 constituents evaluated. Matrine was therefore selected for further binding-pose analysis and experimental validation. Its predicted binding poses with NQO1, NOX4, NOX1, and LCN2 were visualized using PyMOL version 2.3.0 (Figure S4B–E). Within the predicted NQO1 binding pocket, matrine formed three hydrogen bonds with TYR-128, TYR-155, and HIS-161, together with a hydrophobic interaction involving TRP-105 (Figure S4B). In the predicted matrine–NOX4 complex, hydrophobic contacts were observed with PHE-354, VAL-374, ILE-433, and PHE-577 (Figure S4C). Matrine was predicted to interact with NOX1 primarily through hydrophobic contacts involving LEU-68, VAL-71, HIS-208, and PHE-211 (Figure S4D). In contrast, the predicted matrine–LCN2 complex showed fewer contacts, involving TYR-52, HIS-77, and LYS-79 (Figure S4E).

To compare the predicted binding strengths of the matrine–target complexes, the predicted docking energy (ΔG), Ki, and pKi were calculated (Table 2). Among the four targets, matrine exhibited the strongest binding affinity toward NQO1 (ΔG = −9.44 ± 0.10 kcal/mol; Ki = 0.12 ± 0.02 μM; pKi = 6.92 ± 0.07), followed by NOX4 and NOX1. In contrast, binding to LCN2 was substantially weaker (ΔG = –7.10 ± 0.64 kcal/mol; Ki = 8.99 ± 6.48 μM; pKi = 5.21 ± 0.47). Overall, these results implicate NQO1 as a potential target of matrine in SALI.

Table 2 Predicted Molecular Docking Parameters (ΔG, Ki, and pKi) for Matrine Against Four Core Targets (Mean ± SD, n = 5)

Molecular Dynamics Simulations Support Stable Matrine–NQO1 Binding

To evaluate the atomic-level stability and interaction energetics of matrine binding to NQO1 (Figure S5Ai), 100-ns molecular dynamics simulations were conducted for both the matrine–NQO1 complex and the apo-NQO1 system. The RMSD trajectories indicated that both systems remained structurally stable throughout the simulations, with fluctuations confined within 0.1 nm (Figure S5Aii). Compared with the apo protein, the matrine–NQO1 complex exhibited a modestly higher RMSD, consistent with minor ligand-associated conformational adjustments. The RMSF analysis revealed comparable residue-level mobility between the two systems (Figure S5Aiii), with no substantial differences, indicating that matrine binding did not significantly affect the local flexibility of NQO1 residues. The Rg remained stable at approximately 2.3 nm in both systems (Figure S5Aiv), supporting the preservation of protein compactness and overall architecture. Hydrogen-bond analysis showed that up to three hydrogen bonds were frequently formed between matrine and NQO1 during the trajectory, although transient loss of hydrogen bonding was also observed (Figure S5Av), indicating stable ligand–protein interactions. Similarly, the SASA profiles of the two systems were nearly identical and fluctuated around 240 nm2 (Figure S5Avi), further suggesting that ligand binding did not substantially alter the solvent exposure of NQO1.

Free-energy landscape (FEL) analysis showed a single, tightly clustered low-energy basin for the apo system (Figure S5B), indicating a stable and conformationally homogeneous state. In contrast, the matrine–NQO1 complex exhibited a moderately broader energy basin with a more diffuse distribution (Figure S5C), which may reflect a slight increase in conformational flexibility following ligand binding. Nevertheless, the predominant energy minimum corresponded to a stable conformational state. The binding free energy calculated using MM/GBSA was −43.43 kcal/mol (Figure S5D), indicating energetically favorable binding. Energy decomposition analysis highlighted GLN-104, TRP-105, TYR-155, and GLY-149 (chain B) as major contributors to matrine binding, with individual energy contributions ranging from −1.88 to −2.61 kcal/mol (Figure S5E), underscoring their roles in stabilizing the complex.

In summary, the simulations supported the stability of the matrine–NQO1 complex and provided a structural basis for the subsequent target-engagement and pharmacological loss-of-function experiments.

Matrine Engages NQO1 and Increases Its Thermal Stability

Building on molecular dynamics simulations showing a stable interaction between matrine and NQO1 (Figure S5), we next evaluated this interaction using biophysical and cellular assays (Figure 4). The chemical structure of matrine is shown in Figure 4A. Consistent with the predicted binding mode, molecular dynamics simulations suggested that matrine remained stably positioned within an NQO1 binding pocket (Figure 4B). We then quantified binding by SPR. The sensorgrams showed that matrine bound to immobilized NQO1 in a concentration-dependent manner (Figure 4C), and kinetic analysis yielded an estimated dissociation constant (KD) of about 1.8 μM (Figure 4D), supporting a moderate, detectable interaction in the micromolar range. To assess whether matrine affects NQO1 stability in cells, we performed CETSA. Immunoblotting showed higher levels of soluble NQO1 in matrine-treated samples than in PBS controls at elevated temperatures (Figure 4E). Quantification further demonstrated a rightward shift in the thermal denaturation profile of NQO1 in the matrine group (Figure 4F). Two-way repeated-measures ANOVA showed a significant interaction between treatment and temperature (P < 0.001), indicating that matrine alters the thermal stability profile of NQO1 in cells.

Figure 4 Matrine binds to NQO1 and increases its thermal stability. (A) Chemical structure of matrine. (B) Superimposed binding conformations of matrine within NQO1 extracted at 0, 25, 50, 75, and 100 ns, supporting the temporal stability of the matrine–NQO1 complex. (C and D) SPR analysis showing concentration-dependent binding of matrine to immobilized NQO1. (E and F) CETSA showing increased thermal stability of NQO1 in matrine-treated cells compared with PBS-treated control cells. Data are presented as the mean ± SD (n = 5).

Matrine Alleviates SALI and Attenuates Pulmonary Inflammation in CLP-Induced Septic Mice

In the CLP model of sepsis, matrine treatment at 50 and 100 mg/kg improved survival compared with that in the CLP group, with the 100 mg/kg dose producing the greatest benefit and an effect approaching that of dexamethasone (DEX) (Figure 5A). Matrine further attenuated pulmonary edema and protein leakage, as evidenced by reductions in the lung wet-to-dry weight ratio and total protein concentration in BALF. Compared with the 50 mg/kg dose, the 100 mg/kg dose produced more consistent improvements in these parameters (Figure 5B and C). Histopathological examination showed that matrine alleviated typical SALI-associated changes, including alveolar wall thickening, interstitial edema, hemorrhage, collagen deposition, and disruption of alveolar architecture (Figure 5F). Consistently, lung injury scores and collagen deposition were significantly lower in CLP mice treated with 50 or 100 mg/kg matrine than in untreated CLP mice, with the higher dose showing the greatest overall efficacy. Consistently, lung injury scores were significantly reduced by both 50 and 100 mg/kg matrine, whereas collagen deposition was significantly reduced by 100 mg/kg matrine but not by 50 mg/kg matrine (Figure 5D and E). On this basis, 100 mg/kg matrine (MAT-H) was selected for all subsequent in vivo experiments.

Figure 5 Matrine improves survival and alleviates lung injury in mice with CLP-induced sepsis. (A) Kaplan–Meier survival curves showing that matrine treatment improved the 7-day survival of septic mice (Log rank test; n = 12). (B) Total protein concentration in BALF (n = 5). (C) Lung wet-to-dry (W/D) weight ratio at 24 h after CLP (n = 5). (D) Lung injury scores based on histopathological evaluation (n = 5). (E) Quantification of collagen-positive areas in lung tissues following Masson’s trichrome staining using ImageJ version 1.54 (n = 5). (F) Representative lung histological sections: H&E (upper) and Masson’s trichrome (lower). Data are presented as the mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, and ***P < 0.001.

To further assess pulmonary macrophage accumulation, immunohistochemistry (IHC) for the macrophage marker F4/80 was performed. Matrine treatment substantially reduced pulmonary macrophage accumulation compared with that in the CLP group (Figure 6A). In parallel, matrine significantly reduced the BALF concentrations and pulmonary mRNA expression levels of TNF-α, IL-1β, and IL-6 (Figure 6B, C, E, F, H and I). CLP also markedly activated NF-κB signaling in the lung, as shown by increased phospho-NF-κB p65 immunostaining and protein levels. In contrast, matrine treatment significantly reduced NF-κB p65 phosphorylation, as consistently demonstrated by immunohistochemistry, immunoblotting, and immunofluorescence analyses (Figure 6D, G, J–L). Collectively, these findings indicate that matrine ameliorated SALI and attenuated macrophage-associated pulmonary inflammation.