Exploring the Effects and Mechanisms of Neohesperidin Dihydrochalcone on Acute Lung Injury in Mice with Sepsis Using Network Pharmacology and Machine Learning

Abstract

Neohesperidin dihydrochalcone (NHDC) is a synthetic sweetener derived from neohesperidin and can improve pathological changes in sepsis-associated acute lung injury (SALI), but the mechanism by which NHDC inhibits SALI remains unclear. We evaluated the therapeutic effect of NHDC (100 mg/kg) and its potential mechanism using bioinformatics approaches with a Lipopolysaccharide (LPS)-induced SALI model (LPS: 10 mg/kg) in mice (n = 6). Bioinformatics analysis identified 176 shared targets between NHDC and SALI, which were enriched in the MAPK signaling pathway. Further screening yielded five key targets (MAPK14, MAPK8, KDR, CASP3, and RHOA) with significant clinical expression differences (p < 0.01). Molecular docking suggested that NHDC could bind to all five targets, with binding energies <−5.0 kJ/mol, and molecular dynamics indicated stable binding between NHDC and MAPK8 (total binding energy ΔG = −181.320 kJ/mol). In vivo, NHDC reversed oxidative stress markers (catalase, superoxide dismutase, glutathione, malondialdehyde, and reactive oxygen species), decreased TNF-α and IL-6 levels, and alleviated lung pathological injury (p < 0.05 vs. model group); it also significantly decreased phosphorylation of mitogen-activated protein kinases(MAPK) pathway proteins (p < 0.001 vs. model group). In summary, our research revealed that NHDC decreased the oxidative stress and inflammatory response of SALI; its specific mechanism is associated with the MAPK pathway. NHDC has a lot of potential as a medication for anti-SALI treatment.

1. Introduction

Sepsis is a serious condition that can result in deadly multi-organ failure and is brought on by an improper host response to infection. It continues to have a high incidence and fatality rate [1,2,3]. Acute lung injury (ALI) is frequently caused by sepsis, which is mainly characterized by extensive lung inflammation and edema. It might progress to acute respiratory distress syndrome (ARDS). Reports suggest that sepsis-associated acute lung injury (SALI) may occur in 20–50% of sepsis patients [4,5]. The pathogenesis of SALI is complex. It involves lung barrier disruption, inflammatory responses, oxidative stress, apoptosis, coagulation disorders, and related signaling pathways. Current management mainly depends on lung-protective mechanical ventilation, which can improve survival. Although drugs such as corticosteroids (e.g., dexamethasone) and statins (e.g., atorvastatin and simvastatin) may benefit some SALI patients, effective and targeted drug therapies are still lacking [6,7]. Therefore, finding new medications for SALI is critically important.

In recent years, many studies have suggested that natural products may have therapeutic value in SALI [8,9]. Traditional herbs, fruits, and vegetables are all rich sources of flavonoids, a broad class of chemicals [10]. Many flavonoids, such as norwogonin, Baicalein, and faroanol, show anti-inflammatory and antioxidant activities, and they may help treat lung injury [11,12,13].

Neohesperidin dihydrochalcone (NHDC) is a synthetic sweetener derived from neohesperidin (NH), a flavonoid mainly obtained from citrus fruits [14]. NHDC is widely used as a food additive because of its good safety profile, high sweetness, and low calorie content. It also shows several biological effects, including immune modulation, lipid-lowering activity, anti-inflammatory and antioxidant actions, and anti-apoptotic effects. Its potential has been explored in inflammatory skin diseases, food allergy, colitis, obesity, liver injury, and other conditions [15,16,17,18,19,20]. NHDC can regulate the production of inflammatory mediators, especially those induced by lipopolysaccharide (LPS). In LPS-treated RAW264.7 macrophages, NHDC showed anti-inflammatory and antioxidant effects. It reduced oxidative and inflammatory markers and raised the release of the anti-inflammatory cytokine IL-10, partly by preventing LPS from binding to toll-like receptor 4(TLR4) [16,19]. Recently, NHDC has also attracted attention in sepsis-related research. One study reported that NHDC decreased acute kidney damage brought on by sepsis by suppressing inflammation and apoptosis, and this may relate to the regulation of the mitogen-activated protein kinase(MAPK) signaling pathway [21]. The protective effects of NHDC on the liver, kidney, lung, and intestine were further investigated in a mouse sepsis model generated by LPS. The results suggested that NHDC mainly improved vascular endothelial dysfunction through its antioxidant action [22]. However, most sepsis studies on NHDC have focused on kidney injury and systemic responses. Its specific role in SALI and the underlying mechanism are still unclear.

Bioinformatics tools, including network pharmacology, machine learning, and in silico methods, have been increasingly used to improve the efficiency of identifying candidate drugs and targets for SALI. Based on systems biology, network pharmacology connects “drug–disease–target” relationships and predicts how drugs may act on disease networks. Machine learning can further support target prediction from large datasets. Common algorithms, such as random forest (RF), support vector machine recursive feature elimination (SVM-RFE), and least absolute shrinkage and selection operator (LASSO), are often used to screen key targets shared by drugs and diseases. These approaches have been applied in studies such as ginger-gastric cancer, Huperzine A-rosacea, resveratrol-diabetic foot ulcer, and others [23,24,25]. In addition, in silico studies like molecular docking (MD) and molecular dynamics simulation (MDS) are used to explore binding patterns between compounds and targets. For example, one study suggested that MMP9 is a key target of ethyl caffeate in SALI based on MD and MDS [26]. Celastrol was reported to reduce ALI by targeting TLR4 and nuclear factor kappa B(NF-κB) [27]. Other natural compounds, such as saikosaponin B1 and anemoside B4, have also been suggested as potential agents for ALI [28,29].

In this research, we want to shed light on how NHDC improves SALI by combining network pharmacology, machine learning, in silico methods, and animal experiments. A theoretical foundation for the next studies and possible therapeutic applications may be provided by this work.

2. Materials and Methods

2.1. Identifying Potential NHDC Targets

The 3D structure of NHDC (CID:30231) was downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/, accessed on 8 February 2025). The human protein targets of NHDC were identified by importing the structures into PharmMapper (http://www.lilab-ecust.cn/pharmmapper/, accessed on 8 February 2025) and using Norm Fit > 0.5 as the threshold parameter. The UniProt database (https://www.uniprot.org/, accessed on 10 February 2025) was used to standardize NHDC targets and assign official gene symbols. Finally, we identified 225 unique target proteins for NHDC.

2.2. Identification of SALI-Related Target Genes

GeneCards (https://www.genecards.org/, accessed on 14 February 2025) and OMIM (https://www.omim.org/, accessed on 14 February 2025) databases were screened using ‘septic acute lung injury’ and ‘acute lung injury secondary to sepsis’ as search terms to identify SALI-related genes. After combining targets from both these databases and eliminating duplicates, we identified 8285 potential disease targets linked to SALI.

2.3. Protein–Protein Interactions (PPIs)

A Venn diagram was constructed to compare the SALI and NHDC targets using the Weishengxin online platform (http://www.bioinformatics.com.cn/, accessed on 14 February 2025) to identify potential SALI-related targets of NHDC. These genes were imported into the STRING (http://string-db.org/, accessed on 14 February 2025) database, and a protein–protein interaction (PPI) network was constructed using interaction score > 0.7 (high confidence) as the threshold parameter and Homo sapiens as the species. The STRING TSV file was then loaded into the Cytoscape 3.10.2 software, and the top 20 core targets were found using CytoHubba’s MCC method.

2.4. GO and KEGG Pathway Enrichment Analysis

The gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of the disease targets in Homo sapiens was performed using the DAVID bioinformatics platform (https://davidbioinformatics.nih.gov/, accessed on 14 February 2025) and p < 0.05 as the threshold parameter. The results were presented utilizing the Weishengxin internet platform.

2.5. Identification of Core Targets Using Machine Learning Algorithms

The GSE236713 dataset was obtained from the NCBI gene expression database (GEO, https://www.ncbi.nlm.nih.gov/geo/, accessed on 20 February 2025). We collected blood samples from 30 healthy volunteers and 73 patients with pulmonary sepsis from this dataset.

To further filter the important targets, we used three well-known machine learning techniques: RF, SVM-RFE, and LASSO in R software (v 4.3.2). To guarantee that the algorithm’s output is repeatable, we all set the seeds to 123.

(1) LASSO is a linear regression technique that selects features using L1 regularization. The optimal lambda-min was determined through tenfold cross-validation after building the regression model with the R package “glmnet.” (2) The “MSVM-RFE” and “e1071” R packages are used to create the SVM-RFE model. Tenfold cross-validation is used to select the important features. (3) RF is a decision tree-based ensemble learning approach. This method constructs the model using the “randomForest” R library. We found that 500 decision trees are optimal and use them to rank genes based on relevance. Ultimately, the intersection of those algorithms yields the core genes.

2.6. Evaluation of the Expression and Diagnostic Efficacy of Key Target Genes

After identifying the important targets, we analyzed the expression levels of these key targets in the GSE236713 dataset, which includes whole-blood samples from 73 patients with septic pneumonia and 30 healthy volunteers. We then generated receiver operating characteristic (ROC) curves using the “pROC” package and evaluated diagnostic performance by calculating the area under the curve (AUC).

2.7. Molecular Docking (MD)

MD was performed to predict the binding affinity between NHDC and the target proteins. The workflow was as follows. First, OpenBabel (v3.1.1) was used to transform the 3D structure of NHDC to MOL2 format after it was downloaded from PubChem. The RCSB PDB database (http://www.rcsb.org/, accessed on 13 April 2025) was then used to pick appropriate human protein targets, and their crystal structures were acquired. Water molecules and co-crystallized ligands were eliminated from the proteins in PyMOL(v 3.1.0a0 open-source, Schrodinger LLC). The proteins and NHDC were then processed for hydrogen addition, charge assignment, and grid box creation in AutoDockTools (v1.5.7). AutoDock Vina was used for docking with the following configurations: exhaustiveness = 50, num_modes = 9, and energy_range = 3 kcal·mol⁻¹. Docking poses were visualized in PyMOL. An affinity value below −5 kcal/mol was considered as effective binding.

2.8. Molecular Dynamics Simulation (MDS)

Guided by the MD results, we chose the most representative target gene and further explored the time-dependent behavior of its protein–ligand complex in GROMACS (v2020.6). The protein was parameterized with the Amber99SB-ILDN force field, while the ligand was described using GAFF, and water was represented by the TIP3P model. The complex was centered in a dodecahedral simulation box, keeping a minimum distance of 1.0 nm from the box boundary. The system was then solvated with SPC216 water and neutralized by adding Na⁺ and Cl⁻ ions. Before production simulation, the system was relaxed by energy minimization, first with the steepest descent algorithm (10,000 steps) and then with the conjugate gradient algorithm (1000 steps). Next, equilibration was carried out in two stages: 100 ps under NVT conditions (V-rescale thermostat, 300 K, τT = 0.1 ps), followed by 100 ps under NPT conditions (Parrinello–Rahman barostat, 1 bar, τP = 2.0 ps). A 200 ns MD simulation was subsequently performed, and snapshots were recorded every 30 ps. After the run, we evaluated the stability of the protein–ligand association by analyzing solvent-accessible surface area (SASA), radius of gyration (Rg), hydrogen bond number (HB), root mean square deviation (RMSD), and root mean square fluctuation (RMSF). All trajectories and analyses were visualized in QtGrace (v0.2.6). In addition, RMSD and Rg values were exported and processed in Python(v3.9) to construct the free energy landscape (FEL), allowing us to identify the main collective-variable space and the stable conformational states of the simulated complex.

2.9. Binding Energy Calculation

GROMACS’s gmx_mmpbsa script was used to estimate binding free energy. The following are the pertinent calculation equations:

$$\mathsf{\Delta }\mathrm{G}\_\mathrm{b}\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{i}\mathrm{n}\mathrm{g}=\mathsf{\Delta }\mathrm{G}\_\mathrm{c}\mathrm{o}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{e}\mathrm{x}-(\mathsf{\Delta }\mathrm{G}\_\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{t}\mathrm{e}\mathrm{i}\mathrm{n}+\mathsf{\Delta }\mathrm{G}\_\mathrm{l}\mathrm{i}\mathrm{g}\mathrm{a}\mathrm{n}\mathrm{d})$$

$$\mathsf{\Delta }\mathrm{G}\_\mathrm{b}\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{i}\mathrm{n}\mathrm{g}=\mathsf{\Delta }\mathrm{E}\_\mathrm{M}\mathrm{M}+\mathsf{\Delta }\mathrm{G}\_\mathrm{P}\mathrm{B}+\mathsf{\Delta }\mathrm{G}\_\mathrm{S}\mathrm{A}-\mathrm{T}\mathsf{\Delta }\mathrm{S}$$

$$\mathsf{\Delta }\mathrm{E}\_\mathrm{M}\mathrm{M}=\mathsf{\Delta }\mathrm{E}\_\mathrm{C}\mathrm{o}\mathrm{u}\mathrm{l}\mathrm{o}\mathrm{m}\mathrm{b}+\mathsf{\Delta }\mathrm{E}\_\mathrm{V}\mathrm{D}\mathrm{W}$$

In these equations, ΔG_binding denotes the overall binding free energy. ΔG_complex, ΔG_protein, and ΔG ligand refer to the free energies of the protein–ligand complex, the protein, and the ligand, respectively. ΔE_MM represents the molecular mechanics energy and consists of the van der Waals term (ΔE_VDW) and the electrostatic term (ΔE_Coulomb). ΔG_SA and ΔG_PB indicate the non-polar and polar contributions to solvation free energy, respectively, while TΔS describes the entropic contribution.

2.10. Antibodies and Reagents

Antibodies against the following targets were sourced from ABclonal (Wuhan, China): rabbit IgG (AS014), GAPDH (AC001), extracellular signal-regulated kinase (ERK, A4782), c-Jun N-terminal kinase (JNK, A22376), p-P38 (AP1556), and p-JNK (AP0631). p-ERK (28733-1-AP) and P38 (14064-1-AP) antibodies were procured from Proteintech (Wuhan, China). The LY6G antibody (ab238132) was acquired from Abcam (Cambridge, UK). For protein quantification and detection, the BCA protein assay kit and ECL chemiluminescence substrate kit were purchased from Biosharp (Beijing, China). Kits for assessing oxidative stress markers (glutathione, catalase, superoxide dismutase, malondialdehyde, and reactive oxygen species), lipopolysaccharide, along with Wright-Giemsa and H&E staining kits, were supplied by Solarbio (Beijing, China). ELISA kits for IL-6 and TNF-α were obtained from Ruixin Biotechnology (Quanzhou, China).

2.11. Mouse Model of SALI

Six-to-eight-week-old male C57BL/6J mice weighing 20–25 g were purchased from Hunan SJA Laboratory Animal Co., Ltd.(Changsha, China). They were kept in a specified pathogen-free (SPF) facility under a 12 h light/dark cycle and allowed a week to adapt. They were provided ready access to food and water. Then, the 24 mice were randomly assigned to the following four groups (n = 6): control, NHDC, LPS, and NHDC + LPS. Then, they were treated according to the protocol published by Yang et al. [21,22]. Briefly, NHDC was dissolved in a 0.5% carboxymethyl cellulose sodium (CMC-Na) solution by ultrasonication for 10 min. The control and LPS groups of mice received intragastric administration of identical amounts of 0.5%CMC-Na solution, whereas the NHDC and NHDC + LPS groups of mice received intragastric administration of 100 mg/kg NHDC, once daily for five days. After the oral treatment, mice in the LPS and NHDC + LPS groups received intraperitoneal injections of 10 mg/kg LPS, whereas mice in the control and NHDC groups received intraperitoneal injections of the same volume of normal saline. The mice’s activity levels, fur appearance, and secretions and excretions were all noted during the trial. Finally, all mice were put to death by intraperitoneal injection of excess pentobarbital sodium within 12 h following LPS administration, according to the study by Li D et al. [30,31], and organ samples were taken for additional examination. The study was approved by the Chongqing Medical University Second Affiliated Hospital’s Animal Committee (IACUC-SAHCQMU-2024-00147) (Approval date: 23 November 2024).

2.12. Assessment of Bronchoalveolar Lavage Fluid (BALF)

After ligation of the right bronchus, the left lung was lavaged three times with pre-cooled PBS. A BALF recovery rate of >80% was considered indicative of a successful procedure. The collected BALF was centrifuged at 1500 rpm for 10 min at 4 °C to separate the cell pellet from the supernatant. Total protein in the supernatant was measured using the BCA assay. The cell pellet was resuspended in 200 μL PBS, and total cell numbers were counted using a cell counter. A portion of the cell suspension was then stained with the Wright–Giemsa Composite Staining Kit according to the manufacturer’s instructions. Stained cells were observed under a microscope, and BALF neutrophils were identified and counted based on their purple, segmented nuclei.

2.13. Lung Wet-to-Dry Ratio

We measured the ratio of lung wet weight to dry weight using the upper right two lobes of the mice’s lungs. The lung tissues were collected after the mice were killed. A precise balance was used to measure their moist weights as soon as the extra blood was absorbed by filter paper. After 48 h of drying in an oven set at 65 °C, its dry weight was determined.

2.14. Estimation of Catalase (CAT), Glutathione (GSH), Malondialdehyde (MDA), and Superoxide Dismutase (SOD)

The murine lung tissue was weighed. Then, a suitable quantity of the extraction buffer was added and homogenized by grinding. Then, the extract was centrifuged at 8000× g for 10 min at 4 °C, and the supernatant was reserved to estimate MDA, SOD, CAT, and GSH levels and activities using specific kits according to the manufacturer’s instructions (please see Antibodies and Reagents). The absorbances for CAT, GSH, MDA, and SOD were measured at 240 nm, 412 nm, 532/600 nm, and 560 nm, respectively.

2.15. Enzyme-Linked Immunosorbent Assay (ELISA)

Mice were used to obtain BALF supernatants, and the samples were processed in accordance with the ELISA kit’s instructions. A microplate reader was used to measure absorbance at 450 nm. Following the creation of the standard curves, each group’s TNF-α and IL-6 levels were determined.

2.16. Estimation of Reactive Oxygen Species (ROS)

The pieces of frozen lung tissue were prepared for staining by reheating, cleaning, and drying according to the instructions. They were then incubated with the DHE staining solution (1:800) at room temperature for an hour in the dark. Then, after washing off excess stain, the tissues were sealed with a sealing reagent containing DAPI and observed under a fluorescent microscope. Five histological sections were chosen at random for each group and analyzed using the ImageJ software (v 1.54g).

2.17. Hematoxylin and Eosin (H&E) Staining

Hematoxylin and eosin (H&E) staining was applied to the paraffin-embedded lung tissues, and the morphological features were inspected under a microscope. The Smith lung injury scoring criteria was used to evaluate the lung tissue sections. This scoring method quantifies the extent of several pathological features, including pulmonary edema, atelectasis, hyaline membrane development, alveolar and interstitial swelling, and alveolar and interfacial bleeding. Depending on the magnitude of the damaged area inside the visual field, each indicator’s score ranges from 0 (no harm) to 4 (most severe injury). Five histological sections were randomly chosen for analysis from each sample group. The total score was calculated by adding the scores for all indicators.

2.18. Immunohistochemistry

Paraffin-fixed lung tissue sections were hydrated and dewaxed. Then, the sections were treated with H₂O₂ to inhibit endogenous peroxidase activity. The sections were blocked with sheep serum. Then, after washing with PBS, the sections were incubated with the LY6G antibody overnight in the cold. After rewashing, they were incubated with the secondary antibody overnight. Finally, they were incubated with the DAB staining solution, followed by examination under an optical microscope. After washing under tap water to stop the cytoplasm from turning brown, they were counterstained with hematoxylin and sealed with neutral resin. The sections were evaluated under a light microscope and imaged. From each sample, five histological sections were chosen at random and analyzed with the ImageJ software (v 1.54g).

2.19. Western Blotting

The mouse lung tissue was weighed and homogenized with a tissue grinder. Then, the homogenized tissue was lysed with the RIPA buffer containing phosphatase and protease inhibitors. Total protein content was estimated using the BCA assay kit according to the manufacturer’s instructions. Then, an equal amount of protein samples (25 μg/sample) was separated on a 10% SDS-PAGE by electrophoresis and then transferred to an activated PVDF membrane. The membrane was blocked with a high-efficiency blocking solution, followed by overnight incubation at 4 °C with antibodies against p-P38 (1:30,000), p-ERK (1:2000), p-JNK (1:35,000), P38 (1:1200), ERK (1:3500), JNK) (1:10,000), and GAPDH (1:35,000). Then, after washing with TBST, the sections were incubated with the HRP-conjugated Goat anti-Rabbit IgG (H + L) (1:35,000) for one hour at room temperature. The protein bands were developed using ECL chemiluminescence and photographed. Specific protein quantitative analysis was performed using the ImageJ software (v 1.54g).

2.20. Statistical Analysis

The data of this study were analyzed using GraphPad Prism 9.5.0 software, and the results were reported in the form of mean ± standard deviation (SD) (n = 6). During the calculating process, the differences between the two groups were analyzed using an independent two-sided Student’s t-test; comparisons among several groups were undertaken using one-way analysis of variance and Tukey’s post hoc test. A p-value of less than 0.05 was considered statistically significant.

3. Results

3.1. Identification of SALI-Related Targets of NHDC by Network Pharmacology

Bioinformatics tools were used to predict the mechanistic role of NHDC in SALI. After obtaining the 3D structure of NHDC in the PubChem database (Figure 1A), we identified 225 potential targets of NHDC in the PharmMapper database. Furthermore, GeneCards and OMIM database searches led to the identification of 8285 potential targets related to SALI (n = 8285). Intersection analysis of the NHDC and SALI targets using the Venn diagram (Figure 1B) resulted in the identification of 176 common targets. PPI network (Figure 1C) analysis of the common targets using the STRING database showed 176 nodes and 380 edges. We used Cytoscape 3.10.2 to analyze the PPI network diagram (Figure 1D), and the top 20 core genes were identified using the CytoHubba plugin (Figure 1E), which included LCK, RHOA, SRC, ESR1, AKT1, AR, MAPK14, ESR2, TNK2, HSP90AA1, ITK, HCK, MAPK8, IGF1R, HSPA8, KDR, MET, PRKACA, EGFR, and CASP3.

GO analyses of 176 intersection targets (p < 0.05) using the DAVID database revealed that they were abundant in 270 important biological processes (BP) (Figure 1F). The top BP terms were signal transduction, negative regulation of the apoptosis process, innate immune response, angiogenesis, response to lipopolysaccharide, response to allobiotin stimulation, inflammatory response, and so on. Additionally, GO analysis identified 127 enriched molecular functions (MF) associated with the targets, such as vascular endothelial growth factor receptor activity, signal receptor binding, oxidoreductase activity, protein tyrosine kinase activity, and protein serine kinase activity. In addition, GO analysis identified 47 enriched cellular components (CC) associated with the targets, including cytoplasm, nuclear, plasma membrane, organelles, nucleus, mitochondria, and cell membrane receptor complexes like membrane rafts.

KEGG pathway analysis identified 100 enriched pathways, including MAPK, RAS, and PI3K-AKT pathways, which were linked with NHDC-related SALI therapy (p < 0.05) (Figure 1G). These pathways support the mechanistic correlation of NHDC in targeting the pathophysiology of SALI since they are intimately linked to the regulation of inflammation.

3.2. Identification of Core Targets Using Machine Learning

We used three machine learning methods—RF, SVM-RFE, and LASSO regression—to assess the 20 core target genes, aiming to select the most significant ones. Through LASSO regression, we filtered 10 candidate genes (Figure 2A,B): LCK, RHOA, AR, MAPK14, ESR2, TNK2, MAPK8, KDR, MET, and CASP3. RHOA, MAPK14, EGFR, HCK, PRKACA, ESR1, KDR, MAPK8, CASP3, and ITK were the ten most important targets chosen by the RF algorithm, which indicated that 500 trees was the ideal number (Figure 2C,F). SVM-RFE analysis suggested all 20 genes were relevant (Figure 2D,E). Finally, intersection analysis identified five core NHDC target genes: MAPK8, MAPK14, KDR, CASP3, and RHOA (Figure 2G).

The names of these five key genes were then transformed into their gene IDs using the David database and displayed on the Weishengxin online platform. Four of the five genes (MAPK8, MAPK14, KDR, and CASP3) are closely linked to the MAPK pathway (Figure 3). A previous study showed that NHDC alleviates septic acute kidney damage in mice by regulating the MAPK pathway [21]. Furthermore, the MAPK pathway was among the most enriched pathways according to earlier KEGG pathway study results. As a result, we concentrated on the MAPK signaling pathway’s role in SALI.

3.3. Validation of the Expression and Diagnostic Efficacy of the Key Target Genes in SALI

Next, we used the GSE236713 dataset from the GEO database to examine the expression of five key target genes in whole blood. The SALI group included 73 patients, and the control group included 30 healthy volunteers. As shown in Figure 4A, all five genes showed significant differences between the two groups (p < 0.01), suggesting that these genes are closely associated with SALI and may reflect disease-related changes in blood. We then performed ROC curve analysis to evaluate their potential clinical value. Among the five genes, MAPK8 achieved the best diagnostic performance, with an AUC of 0.951 (Figure 4B). Overall, those genes demonstrated a strong capacity to distinguish between healthy individuals and SALI patients.

3.4. MD and MDS Results of the Target Proteins

The interactions between NHDC and the five major NHDC target genes were evaluated by MD, and PyMOL was used to show the findings. MD results for all five primary targets showed low binding energies, thereby indicating significant drug-target affinity (Figure 5). The strongest interaction was observed between MAPK8 and NHDC, with a binding energy of −11.0 kJ/mol.

Among all 5 primary targets, MAPK8 showed the strongest binding with NHDC based on the molecular docking results. To better understand the dynamic behavior and stability of the NHDC-MAPK8 complex, we conducted a detailed analysis of molecular dynamics simulations from multiple angles, including molecular structure, stability, conformational changes, and free energy. According to the RMSD data, the NHDC-MAPK8 complex was highly stable, reaching a plateau at around 150 ns and oscillating at 0.32 nm (Figure 6A). The Rg value and SASA showed slight fluctuations throughout the simulation process, indicating conformational changes (Figure 6B,D). The low RMSF value and the number of hydrogen bonds (typically about 5 hydrogen bonds) indicate that the NHDC-MAPK8 complex has strong stability (Figure 6C,E). Based on the free energy landscape (FEL) data (Figure 6F), the conformation with an RMSD around 0.32 nm and a fluctuating Rg of 2.22 nm was more stable. To further characterize the binding state of the protein-ligand complex, we calculated the FEC between 190 and 200 ns. The results showed a strong interaction between NHDC and MAPK8, with a total binding energy of −181.320 kJ/mol, suggesting stable binding during this period (Figure 6H). Furthermore, the binding energies of 23 amino acid residues in the MAPK8–NHDC complex were below −1.0 kJ/mol, and the binding energy of P37Gln was the highest absolute value at −10.323 kJ/mol (Figure 6G). It suggested weak individual contributions, though the overall interaction remains strong.

3.5. NHDC Mitigates Pathological Alterations in the LPS-Induced SALI Model Mice

We first analyzed whether NHDC mitigated SALI using the LPS-induced SALI model mice. Based on previously published research, the NHDC dosage utilized in this investigation was established [21,22]. The pertinent model proved its efficacy and lack of negative side effects. We observed that the mice in the LPS group had dull, disheveled hair, walked slowly, and had visible fluids and excretions in their eyes and anus. The mice in the treatment group showed improvement when compared to the animals in the LPS group. Lung samples from several mouse groups were histopathologically analyzed using H&E staining. The control mice administered with NHDC demonstrated normal lung morphology (Figure 7A). The mice in the LPS group demonstrated a disordered alveolar structure, but the NHDC + LPS group mice (NHDC + LPS) demonstrated improved interstitial edema and significantly reduced lung damage (Figure 7B). The wet-to-dry lung tissue ratios for all groups of mice are shown in Figure 7C. The ratio was significantly increased in LPS-treated mice compared with the control group, whereas it was significantly decreased after NHDC treatment (NHDC + LPS group; p < 0.01). Consistently, BALF protein concentration was also reduced in the NHDC + LPS group relative to the LPS group (Figure 7D, p < 0.05). The findings showed that NHDC (100 mg/kg) did not induce lung toxicity in mice. It protected the pulmonary barrier and alleviated the lung pathological changes caused by LPS.

3.6. NHDC Decreases Oxidative Stress in Murine Lung Tissues Exposed to LPS

ROS is a key marker of SALI, so we used immunofluorescence to test whether NHDC can reduce LPS-induced oxidative injury in mouse lungs. ROS signals were clearly increased in the LPS group, while they were lower in the NHDC + LPS group (Figure 8A,B, p < 0.05). We also measured MDA and several antioxidant indicators (GSH, CAT, and SOD) to further evaluate oxidative stress. In the LPS group, MDA levels were significantly higher, whereas GSH, CAT, and SOD levels were significantly reduced. After NHDC treatment, these changes were largely reversed in the NHDC + LPS group (Figure 8C–F, p < 0.05). These results indicate that NHDC may alleviate oxidative stress in SALI.

3.7. NHDC Decreases Infiltration of Inflammatory Cells into the Lung Tissues of SALI Model Mice

LY6G immunohistochemistry was used to visualize neutrophil infiltration in the lung tissue specimens. NHDC treatment reduced the amount of LY6G in the lungs of the NHDC + LPS mice compared to the LPS group mice (Figure 9A,B, p < 0.05). We also measured inflammatory responses in BALF. TNF-α and IL-6 levels, as well as total cell counts and neutrophil counts, were markedly increased in LPS-treated mice relative to control mice. These increases were significantly attenuated in the NHDC + LPS group (Figure 9C–F, p < 0.05). Overall, these results suggest that NHDC exerts anti-inflammatory effects in SALI model mice.

3.8. NHDC Protects Against Lung Injury in the SALI Model Mice by Regulating the MAPK Signaling Pathway

Our machine learning and network pharmacology analyses suggested that the MAPK pathway is important for the therapeutic action of NHDC in SALI. To verify this, we performed Western Blot in lung tissues. The ratios of p-P38/P38, p-JNK/JNK, and p-ERK/ERK were significantly increased in the Model group, indicating activation of MAPK signaling (Figure 10, p < 0.001). In contrast, these phosphorylation ratios were significantly lower in the NHDC + LPS group. This result indicates that the pre-treatment of NHDC affects the activation of MAPK induced by LPS. Overall, the Western Blot data support our bioinformatics findings and indicate that MAPK signaling is a key pathway involved in the protective effects of NHDC against SALI.

4. Discussion

NHDC is a widely used food additive. After oral consumption, NHDC is absorbed into the blood and distributed to various organs, including the lungs, and is known for its high safety profile [32]. Several studies have evaluated NHDC’s teratogenicity, embryotoxicity, and sub-chronic oral toxicity in rat models, with no negative effects reported [33,34]. NHDC has demonstrated anti-inflammatory, immunomodulatory, lipid-lowering, and antioxidant properties. While NHDC’s role in sepsis has been studied, particularly regarding kidney damage and systemic reactions, its effect on SALI and the mechanisms underlying it remain unknown.

The lungs are the most vulnerable organ when sepsis leads to multi-organ failure. Current treatments for SALI include fluid management, respiratory support, infection control, immune modulation, and so on [4]. However, aside from lung-protective ventilation, the efficacy of other approaches is limited due to the complex nature of SALI. Therefore, identifying new, effective treatments is crucial. NHDC has shown promise in sepsis models due to its immune-regulatory and anti-inflammatory properties. In our study, we created a mouse model of acute lung injury induced by sepsis and assessed the impact and mechanism of NHDC on SALI. Our results suggest that NHDC may lower protein levels in BALF and lessen lung pathology brought on by LPS, which is consistent with earlier studies. We observed that LPS-induced SALI in mice resulted in higher levels of LY6G, TNF-α, IL-6, ROS, and MDA, and decreased antioxidant levels (GSH, CAT, SOD), contributing to lung injury. However, NHDC treatment helped restore some of these oxidative stress and inflammatory markers, suggesting that its anti-inflammatory and antioxidant properties may help reduce lung damage.

Network pharmacology has become an effective tool for discovering new drugs and understanding the mechanisms of diseases like ALI. Moreover, combining machine learning techniques—such as SVM-RFE, RF, and LASSO—has proven crucial for screening key target genes in biomedical research. Using network pharmacology, we identified 176 shared targets between SALI and NHDC, with the MAPK pathway showing significant enrichment of NHDC targets, as indicated by KEGG pathway analysis. Further, through machine learning, we identified five primary targets—MAPK8, MAPK14, KDR, CASP3, and RHOA—associated with the MAPK signaling pathway. These targets have been linked to the regulation of ALI in multiple studies [35,36,37,38]. Our findings aligned with clinical expression differences confirmed using the GSE236713 dataset. MD and MDS were used to investigate the binding affinity of NHDC to these key targets. The results showed that NHDC successfully bound to all five targets, with strong binding affinity (binding energy < −5 kJ/mol), especially for MAPK8, which exhibited the strongest binding energy. The stability and binding capacity of the NHDC-MAPK8 complex were confirmed by molecular dynamics simulations, with van der Waals forces contributing significantly to the overall binding energy (−314.361 kJ/mol), resulting in a high total binding energy of −181.320 kJ/mol. These results demonstrate how important the MAPK pathway is to NHDC’s ability to protect against SALI.

The MAPK signaling pathway includes three main subfamilies: JNK, P38, and ERK. ERK regulates cell proliferation and differentiation, P38 is involved in inflammatory and stress responses, and JNK controls apoptosis and stress responses [39]. In our study, we found that LPS induced phosphorylation of P38, JNK, and ERK in the SALI model by activating the MAPK pathway. NHDC treatment effectively blocked these changes. Wang B et al. also reported that kaempferol reduces SALI by inhibiting the MAPK signaling pathway, which supports our findings [40]. Other studies have shown that isostrictiniin improves SALI by acting as an antioxidant and modulating inflammation through the MAPK/NF-κB and Keap1-Nrf2/HO-1 pathways [31]. Additionally, NHDC has been shown to influence other inflammatory pathways in various diseases. For example, NHDC reduces NF-κB expression in experimental colitis and modulates TLR4/Nrf2/NF-κB signaling in liver injury [15,16,41]. We hypothesize that NHDC may also mitigate SALI through regulation of the TLR4/Nrf2/NF-κB pathway, given the complex interactions between these pathways. Future studies should further explore this mechanism.

This study integrated in vivo mouse models with bioinformatics approaches to elucidate how NHDC reduces SALI. However, some limitations in the study remain, which could be addressed in future work: (1) Machine learning prediction limitations: While machine learning has significantly improved target prediction, factors such as inherent model performance differences and limited training data can still affect prediction accuracy. (2) Enhancing in vivo assessment: The preventive effects of NHDC on SALI have not been fully explored. For more comprehensive validation, additional physiological markers, such as arterial blood gas measurements and vital sign monitoring, should be included. (3) Limited exploration of signaling pathway interactions: Several inflammatory signaling pathways interact intricately during the pathophysiology of SALI. This study has not fully examined the precise mechanisms of NHDC’s intervention in these pathways (e.g., TLR4/Nrf2/NF-κB interactions). Future research should investigate these interactions in greater detail.

5. Conclusions

This study employed bioinformatics and in vivo experiments to explore how NHDC (Figure 11) reduces SALI. Key targets associated with SALI, including MAPK8, MAPK14, KDR, CASP3, and RHOA, which are strongly linked to the MAPK signaling pathway, were found to be influenced by NHDC. In vivo tests revealed that NHDC significantly reduces oxidative stress and inflammatory responses (p < 0.05). Moreover, NHDC regulates inflammation by inhibiting the phosphorylation of p-ERK, p-JNK, and p-P38 (p < 0.001). In conclusion, our study has demonstrated that NHDC is a potentially successful treatment for SALI and that it works via controlling the MAPK signaling pathway.