Unraveling the Gut Microbiota-mediated Anti-tumor Mechanisms of ShenXia KuanZhong Decoction in Gastric Cancer: a Systems Biology and Dose-weighted Network Pharmacology Approach.

Introduction
Gastric cancer (GC) ranks among the leading causes of cancer-related mortality worldwide, with a high incidence in East Asia and a median survival of approximately one year in advanced stages [1, 2]. Despite advances in chemotherapy (e.g., fluorouracil, oxaliplatin) and targeted therapies (e.g., trastuzumab for HER2-positive patients), complete cures remain elusive, and patients often face severe chemotherapy-related side effects and poor quality of life (QoL) [3, 4]. Immune checkpoint inhibitors like nivolumab have shown efficacy in specific GC subtypes, but overall response rates are modest [5]. Despite these therapeutic advances, patients frequently experience a substantial symptom burden and severe chemotherapy-related adverse effects, both of which compromise the QoL [6, 7]. These challenges underscore the urgent need for innovative, integrative therapeutic strategies.
Traditional Chinese medicine (TCM) has garnered increasing interest as a potential adjuvant therapy [8]. In ancient Chinese medical texts, Shengyang Yiwei Decoction (SYYWD) was widely used to treat gastrointestinal diseases such as chronic gastritis, dyspepsia, and ulcerative colitis [9]. Research shows that when combined with chemotherapy, SYYWD has demonstrated the potential to alleviate GC symptoms and decelerate disease progression [10]. Building on SYYWD, ShenXia KuanZhong Decoction (SXKZD) has been adapted as a specialized GC treatment formula [11]. It has exhibited notable efficacy in the long-term clinical management of GC, such as reducing postoperative complications, accelerating gastrointestinal recovery, alleviating chemotherapy-related side effects, and improving QoL. However, its underlying mechanisms require further elucidation.
The gut microbiota, a dynamic ecosystem, plays a pivotal role in cancer progression by modulating inflammation, immunity, and metabolism [12]. Dysbiosis in GC patients can exacerbate tumorigenesis, while microbial metabolites like short-chain fatty acids (SCFAs) and indole derivatives exhibit anti-tumor effects [13]. Research shows that TCM prescriptions can regulate the composition and structure of the gut microbiota in patients with GC, enhance the intestinal barrier, regulate immune responses, and maintain metabolic balance [14]. These changes support the significant anti-tumor effects of TCM. In addition, the gut microbiota can metabolize the components of TCM, further enhancing its anti-tumor potential [15]. Consequently, investigating the interaction between TCM and the gut microbiota provides a fresh lens to elucidate TCM’s anti-tumor mechanisms. This highlights the need to explore whether SXKZD exerts therapeutic benefits in GC via gut microbiota regulation.
Network pharmacology, with its complex models, elucidates the molecular mechanisms of TCM and has gained widespread recognition [16]. Nevertheless, it often ignores dosage, assuming uniform active ingredient levels across doses, which may skew outcome predictions. Systems biology and network pharmacology offer powerful tools to dissect the complex, multi-target actions of TCM [17, 18].
Based on the above theoretical foundation, we propose the preliminary research hypothesis: SXKZD exerts anti-gastric cancer effects via multi-target, multi-pathway mechanisms, possibly (1) containing a certain amount of antitumor active components that directly inhibit gastric cancer cell proliferation; (2) indirectly producing antitumor metabolites by modulating the gut microbiota; the two mechanisms act together through interactions with core targets to exert a combined antitumor effect, becoming an optional adjuvant therapeutic strategy.
This study integrates dose-weighted network pharmacology, molecular docking, molecular dynamics (MD) simulation, and pharmacokinetic analysis, aiming to: (1) construct a gut microbiota-substrate-metabolite (GM-S-M) network to identify key gut microbes and metabolites [19]; (2) identify core targets of SXKZD via dose-weighted network pharmacology; (3) validate interactions between key SXKZD components, gut microbes, metabolites, and core targets through docking and MD simulation; (4) evaluate the pharmacokinetic properties of key bioactive components; (5) validate related conclusions through in vitro experiments.
The research workflow is shown in Fig. 1.

Materials and Methods

Data Collection
SXKZD is a combination of six herbal compounds: Ginseng Radix et Rhizoma (9 g), Astragali Radix (30 g), Citri Reticulatae Pericarpium (9 g), Pinelliae Rhizoma (9 g), Curcumae Rhizoma (10 g), and Hedyotis Diffusae Herba (15 g). The detailed composition and dosage of SXKZD are presented in Table 1. Active ingredients of SXKZD were retrieved and screened from the TCMSP [20] using oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥ 0.18 as screening conditions. In addition, target prediction was conducted for the active ingredients that met the screening criteria, and all targets were corrected using Uniprot [21] after removing duplicates (Table S2). GC-related targets were collected from seven databases: GeneCards [22], OMIM [23], DisGeNET [24], PharmGKB [25], DrugBank [26], Comparative Toxicogenomics Database (CTD) [27], and Therapeutic Target Database (TTD) [28]. The search terms “gastric cancer” and “stomach adenocarcinoma” were used. For GeneCards and CTD, targets with correlation scores greater than or equal to twice the median were included (Table S3). The union of targets from all databases was taken, duplicates were removed, and target names were standardized using UniProt. Targets related to human gut microbiota and their metabolites were obtained from the gutMGene database [29]. Duplicate entries were removed, and the data were used to construct networks linking gut microbiota, substrates, metabolites, and associated targets (Table S4, Table S5). All database URLs and software versions are provided in Table S1.

Network Construction and Topological Analysis
The common targets obtained by taking the intersection of active ingredients, GC, and gut microbiota-related targets were imported into the STRING platform, selecting “Homo sapiens” as the species for protein-protein interaction (PPI) analysis, with an interaction score threshold of ≥ 0.7 to construct the network. Then, the core targets were imported into Cytoscape software to analyze the degree centrality (DC), betweenness centrality (BC), and closeness centrality (CC) of the network nodes using the CytoNCA plugin, where a higher DC, BC, and CC indicate a more critical role of the active ingredients and targets in the potential mechanism of SXKZD for treating GC. Additionally, we further explore the key node modules in the network through the Cytohubba [30] and MCODE [31] algorithms to demonstrate the core process. The top enriched terms were ranked by their -log10(P-value) and visualized as bar plots using the R package “ggplot2”.

Functional Enrichment Analysis
To investigate the molecular mechanisms and pathways of SXKZD regulating gut microbiota and their metabolites in treating GC, the “clusterprofiler” package was used to perform gene function enrichment analysis and visualization. Gene ontology (GO) analysis included biological process (BP), cellular component (CC), and molecular function (MF) [32], as well as Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis [33]. Significant statistical difference was defined as FDR-adjusted p < 0.05.

Dose-weighted Network Pharmacology
Due to the complexity of the dosage of the internal medicines in TCM formulas, it is not sufficient to explore the importance of targets using only conventional network pharmacology. To overcome the limitations, the VIKOR (VlseKriterijumska Optimizacija I Kompromisno Resenje) method, a multi-criteria decision-making framework, was employed to rank target importance [34, 35]. Through the VIKOR method, the weights of the nodes can be defined objectively or subjectively [36]. The process involved: (1) constructing a PPI network matrix; (2) calculating DC, BC, and CC as evaluation criteria; (3) normalizing the matrix and assigning equal weights (1/3) to each criterion; (4) determining the ideal and negative ideal solutions, utility measure (Si), regret measure (Ri), and VIKOR index (Qi); (5) converting Qi to Qi*=1-Qi for cost-type indexing; (6) assigning bilateral weights, where the drug-side weight (wi, D) was calculated by normalizing the frequency of active ingredient production based on dosage (assuming 1 g of herb corresponds to one production run), and the disease-side weight (wi, L) was set to 1; and (7) calculating the final drug score as DrugVIKORi=wi, D×Qi, D*.

Molecular Docking
The 3D structure of the protein is obtained from the RCSB PDB database [37]. The 3D structure of the ligand is obtained from the PubChem database [38]. PyMOL removes the original ligands and water molecules from the protein, adds hydrogen atoms to the receptor protein [39], and saves it in PDBQT format using AutoDock Tools [40]. The Gasteiger method is used to process atomic charges. The docking position is where the original ligand is located. The grid box is centered on the active binding site, and its size is adjusted to the size of the protein. Docking is performed with the ligand being flexible and the receptor being rigid. The number of Runs is 15, and other parameters are set to default values. The binding free energy (kcal/mol) is calculated based on the scoring function, and the pose with the lowest binding energy is selected for further analysis. The URL of the database and the version of the software are shown in Table S1.

Molecular Dynamics (MD) Simulation
MD simulations were conducted using GROMACS to assess the stability of the complex [41]. The ligand topology was generated using the Sobtop script (Tian, 2024), and the protein was simulated with the Amber99SB-ILDN force field. Each system was solved in a cubic box with TIP3P water molecules, neutralized with sodium (Na⁺) and chloride (Cl⁻) ions, and energy-minimized using the steepest descent algorithm. Equilibration was performed under isothermal-isochoric (NVT) and isothermal-isobaric (NPT) ensembles at 300 K and 1 bar, followed by a 100-ns production Run with a 2-fs time step. Trajectories were analyzed for root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), hydrogen bonds, and Gibbs energy landscapes.
Binding free energies were calculated using the Molecular Mechanics Poisson-Boltzmann Surface Area (MMPBSA) method with the “gmx_mmpbsa.bsh” script [43]. Snapshots from the last 10 ns of the MD trajectories were used to compute the total binding free energy (∆GTOTAL) as the sum of gas phase molecular mechanics energy (∆GGAS, including van der Waals [∆GVDW] and electrostatic [∆GELE] energies), solvation energy (∆GSOLV, including polar [∆GPOLAR] and non-polar [∆GNONPOLAR] contributions), and entropy contribution (− TΔS). The final binding free energy (dG) was used to calculate the dissociation constant (Ki) using the equation Ki = exp(dG/RT), where R is the gas constant and T is 300 K. Results were reported as mean ± standard error of the mean (SEM) in kJ/mol. The version of the software is shown in Table S1.

Physicochemical and Pharmacokinetic Property Analysis
The physicochemical properties were evaluated using the SwissADME platform [44]. Canonical SMILES strings were input to calculate parameters, including molecular weight (MW), number of heavy atoms, fraction of sp3 carbons (Fraction Csp3), rotatable bonds, H-bond acceptors and donors, molar refractivity (MR), topological polar surface area (TPSA), and various Log P values (iLOGP, XLOGP3, WLOGP, MLOGP, Silicos-IT Log P, Consensus Log P). Solubility was assessed using ESOL, Ali, and Silicos-IT models, with values reported in mg/ml and mol/l, and classified as soluble, moderately soluble, or poorly soluble. Drug-likeness was evaluated based on Lipinski, Ghose, Veber, Egan, and Muegge rules, alongside bioavailability scores, PAINS alerts, Brenk alerts, lead-likeness violations, and synthetic accessibility scores.
Pharmacokinetic properties were predicted using the pkCSM platform [45], assessing absorption (water solubility, Caco2 permeability, intestinal absorption, skin permeability, P-glycoprotein substrate/inhibitor status), distribution (volume of distribution, fraction unbound, blood-brain barrier (BBB) permeability, central nervous system (CNS) permeability), metabolism (CYP substrate/inhibitor status for CYP1A2, CYP2C19, CYP2C9, CYP2D6, CYP3A4), excretion (total clearance, renal OCT2 substrate status), and toxicity (AMES toxicity, maximum tolerated dose, hERG inhibition, oral rat acute and chronic toxicity, hepatotoxicity, skin sensitization, T. pyriformis toxicity, minnow toxicity). The URL of the platform is shown in Table S1.

Results

Construction of the Gut Microbiota-substrate-metabolite (GM-S-M) Network
To investigate the metabolic interactions within the gut microbiota, a tertiary regulatory network of gut microbiota-substrate-metabolite (GM-S-M) was constructed using data from the gutMGene database and visualized with Cytoscape. The network comprises 446 gut microbial nodes (diamonds), 86 substrate nodes (rounded rectangles), and 614 metabolite nodes (triangles), with 2418 edges (average degree: 4.22) (Fig. 2). Bacteroides exhibited the highest degree centrality (DC: 75) among gut microbiota, followed by Parabacteroides (DC: 51) and Akkermansia (DC: 46). Among substrates, D-Glucose (DC: 54) and Cianidanol (DC: 44) were the most connected, while butyrate (DC: 86), propionate (DC: 72), and acetate (DC: 66) were the top metabolites (Table S6).
Parts of a network with highly connected areas (sub-networks) have a higher probability of participating in biological regulation. At the same time, those with low connectivity do not play a key role in the entire network. Further analysis using cytoHubba revealed key players in this complex network, including bacteria such as Bacteroides, Parabacteroides, Lactobacillus, Lachnospiraceae, Akkermansia, and Oscillospiraceae; metabolic precursors Like D-Glucose and Cianidanol; and metabolites such as Acetate, Butyrate, and Propionate. These elements form the critical modules within the gut microbiota network. Additionally, MCODE analysis identified 12 distinct sub-networks. For instance, Cluster 1 includes bacteria such as Bacteroides, Lactobacillus, Bacteroides thetaiotaomicron, Lachnospirales, Clostridium sporogenes, and Clostridium sp. TM-40, precursor Daidzein, and metabolites Like 3-Indolepropionic acid, Indole, Propionate, Succinate, Phenylactic acid, Dihydrodaidzein, and Daidzein. These 12 subnetworks likely represent pivotal components of the gut microbiota, serving as typical examples of its potential functional roles.

Identification of Active Ingredients and Targets
Following screening with OB ≥ 30% and DL ≥ 0.18, 55 unique active ingredients involved in SXKZD (Fig. 3A) were identified: 17 from Ginseng Radix et Rhizoma (86 targets), 16 from Astragali Radix (164 targets), 5 from Citri Reticulatae Pericarpium (194 targets), 11 from Pinelliae Rhizoma (67 targets), 1 from Curcumae Rhizoma (14 targets), and 5 from Hedyotis Diffusae Herba (150 targets), yielding a total of 212 unique targets for SXKZD. Multiple databases retrieved GC-related targets, resulting in 14,926 unique targets after deduplication. The gutMGene database identified 55 gut microbes and 46 microbial metabolites, corresponding to 238 unique targets. A Venn diagram revealed 48 overlapping targets between SXKZD ingredients, GC, and gut microbiota, suggesting a mechanistic link through shared molecular pathways (Fig. 3B).

GO and KEGG Enrichment Analysis
Functional enrichment analyses were performed to elucidate the potential mechanisms of SXKZD in GC treatment through its interactions with gut microbiota and microbial metabolites. The obtained targets were subjected to GO analysis, yielding 2,305 BP terms, 89 CC terms, and 42 MF terms. The BP terms included processes such as response to nutrient levels, response to lipopolysaccharide, and response to molecules of bacterial origin. The CC terms encompassed components like membrane raft, membrane microdomain, and caveola, while the MF terms involved functions such as cytokine receptor binding, cytokine activity, and phosphatase binding (Fig. 3C, Table S7). These GO terms suggest that SXKZD influences immune regulation, inflammatory responses, and cellular signaling, potentially through gut microbiota-derived signals. KEGG enrichment analysis was also performed, resulting in 172 enriched pathways. These pathways included signaling cascades such as Gastric cancer, IL-17, and TNF signaling pathways (Fig. 3D, Table S8). These pathways collectively indicate that SXKZD exerts its anti-GC effects through modulation of immune responses, inflammation, apoptosis, and tumor microenvironmental factors, potentially mediated by gut microbiota interactions.

Network Analysis of SXKZD in the Context of GC and Gut Microbiota
Around 48 common targets, we constructed the “SXKZD-Gut-GC” axis to visually demonstrate the multi-layered and multi-dimensional regulatory mechanism of the SXKZD. This network aligns with the characteristics of TCM formulas with multiple ingredients and targets, providing a theoretical basis for further elucidating its action mode (Fig. 4A). Further, a PPI network of these common targets was constructed using the STRING platform (interaction score ≥ 0.7) and visualized in Cytoscape, comprising 44 nodes and 308 edges (average degree: 12.8) (Fig. 4B-C). Topological analysis using DC, BC, and CC identified TNF (DC: 28, CC: 0.62), IL6 (DC: 26, CC: 0.59), TP53 (DC: 24, CC: 0.57), AKT1 (DC: 22, CC: 0.55), BCL2 (DC: 20, CC: 0.53), and PTGS2 (DC: 18, CC: 0.51) as primary targets (Fig. 4D). These targets are associated with inflammation (TNF, IL6), cell cycle regulation (TP53), cell survival (AKT1, BCL2), and prostaglandin synthesis (PTGS2), indicating their central roles in the anti-GC effects of SXKZD.

Dose-weighted Network Pharmacology Analysis
We adopt the dose-weighted network pharmacology approach to incorporate the dose dimension into mapping active ingredient targets. The Sankey diagram was generated using the R package “networkD3” to visualize the relationships between herbs, ingredients, and targets (Fig. 5A). The diagram highlights the complex, multi-component, and multi-target nature of SXKZD, with each herb contributing multiple active ingredients that collectively interact with a wide range of targets. For instance, Huangqi (30 g) and Baihuasheshecao (30 g) showed the most extensive connections, reflecting their higher dosage and likely more significant contribution to the formula’s therapeutic effects. The VIKOR method was applied to quantify and rank the importance of targets based on their Weighted interactions, considering the dosage of each herb. The VIKOR scores were calculated using a multi-criteria decision-making framework, integrating DC, BC, CC, and other topological metrics. The top 10 core targets of SXKZD with VIKOR scores from high to low were PTGS2, BCL2, TNF, AKT1, CASP3, TP53, PPARG, RELA, HMOX1, and IL1B (Fig. 5B, Table S9). These metrics indicate that PTGS2, BCL2, and TNF are central hubs in the network, likely playing critical roles in mediating the therapeutic effects of SXKZD.

Integrated Network Analysis of Gut Microbiota, Metabolites, and Ingredients
To investigate the potential mechanisms of SXKZD, we further constructed and analyzed three sub-networks: the Gut Microbiota-Target (GM-T) network, the Metabolite-Target (M-T) network, and the Ingredient-Target (I-T) network. These sub-networks were developed to explore the relationships between gut microbiota, metabolites, active ingredients, and their corresponding therapeutic targets. Venn diagrams (Fig. 6A) revealed significant overlaps, with the GM-T network identifying 14,816 disease-related targets, of which 109 overlapped with gut microbiota targets, the M-T network showing 14,783 targets with 142 overlapping with metabolite targets, and the I-T network indicating 14,722 targets with 203 overlapping with ingredient targets. GM-T network mapped the interactions between gut microbiota and their associated targets, highlighting the role of microbial communities in mediating prescription effects. M-T network focused on the interactions between microbial metabolites and their targets, emphasizing the role of gut microbiota-derived metabolites in the therapeutic activity of the formula. I-T network illustrated the interactions between active ingredients and their molecular targets, reflecting the direct pharmacological effects of the formula.
Topological analysis of the GM-T network identified Lactobacillus plantarum as the most central microbiota (DC: 243.80, BC: 0.43, CC: 0.46), alongside Akkermansia muciniphila (DC: 808.12, BC: 0.36, CC: 0.40) and Escherichia coli (DC: 358.12, BC: 0.39, CC: 0.38), with key targets including IL10 (DC: 15), CXCL8 (DC: 15), and TNF (DC: 14) (Fig. 6B, Table S10). In the M-T network, urocanic acid emerged as the most connected metabolite (DC: 721.34, BC: 0.41, CC: 0.46), followed by Quercetin (DC: 632.38, BC: 0.41, CC: 0.46) and Butyrate (DC: 364.01, BC: 0.40, CC: 0.38), with top targets such as CXCL8 (DC: 6), CYPIA1 (DC: 6), and TNF (DC: 5) (Fig. 6B, Table S11). The I-T network highlighted Quercetin as the most prominent ingredient (DC: 186.42, BC: 0.52, CC: 0.51), alongside Kaempferol (DC: 162.31, BC: 0.38, CC: 0.46) and Naringenin (DC: 78.86, BC: 0.38, CC: 0.37), with key targets including PTGS2 (DC: 33), PTGS1 (DC: 28), and NCOA2 (DC: 24) (Fig. 6B, Table S12).
The intersection targets of the GM-T, M-T, and I-T sub-networks were identified as adjunct targets, including IL10, IL1B, CCL2, CXCL8, CLDN4, MAPK1, TNF, MAPK14, CDKN1A, IL6, PPARG, and GOT1. These targets contribute to complementary pathways, including immune regulation, cell adhesion, and metabolic processes. These results indicate that SXKZD exerts its therapeutic effects through a multifaceted, multi-pathway mechanism, involving the modulation of gut microbiota, the production of microbial metabolites, and the direct interaction of active ingredients with key molecular targets.
To focus on research targets, we used Venn diagrams to integrate primary and auxiliary targets, with TNF and IL6 overlapping in both groups. In addition, based on VIKOR scores, TNF was identified as a hub target worthy of further exploration (Fig. 6C).

Molecular Docking
First, a Gut Microbiota-Metabolite-Target-Ingredient (GM-M-T-I) network centered on TNF and IL-6 was constructed to illustrate the relationships among gut microbiota, metabolites, and ingredients (Fig. 7A).
The heatmap quantified the binding affinities of key ingredients and metabolites to primary targets. Notably, MOL005344 (Ginsenoside Rh2) exhibited strong binding to TNF (−8.7 kcal/mol) and IL6 (−8.0 kcal/mol). In comparison, 3-Indolepropionic acid showed favorable binding to TNF (−8.0 kcal/mol), indicating its potential as a key bioactive molecule in the mechanism of action (Fig. 7B). Molecular docking further validated the interactions of Ginsenoside Rh2 (MOL005344) and 3-Indolepropionic acid with TNF. The 2D and 3D diagrams of TNF-Ginsenoside Rh2 revealed key interactions, including hydrogen bonds with ASP-86 (2.3 Å), PRO-84 (2.1 Å), and ALA-322 (2.8 Å), and hydrophobic interactions with LEU-328, suggesting stable binding within the active site. Similarly, the TNF-3-Indolepropionic acid complex showed hydrogen bonds with SER-139 (2.8 Å) and TYR-305 (2.1 Å), alongside hydrophobic interactions with LEU-328, ALA-323, and TYR-319, confirming its strong binding affinity. The chemical structures of Ginsenoside Rh2 and 3-Indolepropionic acid provided insights into their molecular features, with Ginsenoside Rh2 featuring a steroidal backbone with multiple hydroxyl groups, and 3-Indolepropionic acid possessing an indole ring and a propionic acid group, both conducive to forming stable interactions with TNF. These findings demonstrate that therapeutic effects are mediated by the synergistic actions of its ingredients and microbial metabolites (Fig. 7C-D).

MD Simulation
The MD simulations for the TNF-Ginsenoside Rh2 complex revealed stable dynamics over 100 ns, as evidenced by the RMSD of TNF, Ginsenoside Rh2, and the complex, which fluctuated between 0.1 and 0.4 nm, indicating structural stability. The RMSF showed minimal atomic fluctuations (0.2–0.6 nm), with the complex exhibiting slightly higher flexibility. The Rg remained consistent at approximately 1.8–2.4 nm, suggesting that the complex maintained a compact conformation. Hydrogen bond analysis indicated the presence of 2–6 hydrogen bonds throughout the simulation, with occasional pairs within 0.35 nm, supporting the stability of the interaction. Gibbs energy landscapes further confirmed the stability of the TNF-Ginsenoside Rh2 complex, with a low-energy basin at 7.5 kJ/mol, as visualized in both 2D and 3D plots (Fig. 8A). Similarly, the MD simulation for the TNF-3-Indolepropionic acid complex (Panel B) showed stable RMSD values (0.1–0.3 nm), RMSF values (0.2–0.6 nm), and Rg values (1.6–2.2 nm), indicating a stable and compact structure. Hydrogen bond analysis revealed 2–4 hydrogen bonds, with pairs within 0.35 nm, and the Gibbs energy landscape displayed a low-energy state at 5.0 kJ/mol, confirming the stability of the complex (Fig. 8B). These MD simulation results demonstrate that Ginsenoside Rh2 and 3-Indolepropionic acid form stable complexes with TNF.

The MMPBSA analysis provided detailed insights into the binding free energies of Ginsenoside Rh2 and 3-Indolepropionic acid with TNF. For 3-Indolepropionic acid, the total relative binding energy (∆GTOTAL) was calculated as −109.64 kJ/mol, with contributions from van der Waals energy (∆GVDW: −133.85 ± 9.53 kJ/mol), electrostatic energy (∆GELE: −117.08 ± 23.78 kJ/mol), polar solvation energy (∆GPOLAR: 155.77 ± 27.24 kJ/mol), and non-polar solvation energy (∆GNONPOLAR: −14.47 ± 0.28 kJ/mol). The total gas phase energy (∆GGAS) was − 250.94 ± 16.75 kJ/mol, and the total solvation energy (∆GSOLV) was 141.30 ± 27.10 kJ/mol. The entropy contribution (-TΔS) was 11.01 kJ/mol, resulting in a final binding free energy (dG) of −98.63 kJ/mol, corresponding to a dissociation constant (Ki) of 5.26 × 10⁻¹² µM, indicating powerful affinity. For Ginsenoside Rh2, the ∆G_TOTAL was − 188.71 kJ/mol, with contributions from ∆GVDW (−295.89 ± 14.63 kJ/mol), ∆GELE (−5.66 ± 13.14 kJ/mol), ∆GPOLAR (145.96 ± 28.25 kJ/mol), and ∆GNONPOLAR (−33.12 ± 1.37 kJ/mol). The ∆GGAS was − 301.55 ± 20.91 kJ/mol, and ∆GSOLV was 112.84 ± 27.26 kJ/mol. The entropy contribution (-TΔS) was 41.08 kJ/mol, yielding a dG of −147.63 kJ/mol and a Ki of 1.37 × 10⁻²⁰ µM, indicating robust affinity. These results highlight that Ginsenoside Rh2 exhibits a stronger binding affinity to TNF than 3-Indolepropionic acid, driven primarily by favorable van der Waals and non-polar interactions. The MD simulations and MMPBSA analysis confirm these molecules’ stable and strong binding to TNF (Tables 2 and 3, Table S13).

Experimental Validation
We evaluated the anti-GC activity of Ginsenoside Rh2 (GRh2) and 3-Indolepropionic acid (IPA) to validate the computational predictions further using the CCK-8 assay on AGS cells. Both compounds exhibited concentration- and time-dependent inhibition, with IC50 values of 68.74 ± 1.27 µg/ml for GRh2 and 780.60 ± 24.40 µg/ml for IPA at 24 h (Fig. 9A-B). Western blot analysis confirmed that exogenous supplementation of GRh2 and IPA significantly suppressed TNFα expression in AGS cells, with GRh2 showing a stronger inhibitory effect (Fig. 9C-D). Additionally, a cellular thermal shift assay (CETSA) demonstrated that GRh2 and IPA increased the thermal stability of TNFα compared to DMSO, with IPA producing greater stabilization (Fig. 9E). These preliminary experimental results support the computational findings for TNF as a hub target, with GRh2 demonstrating superior inhibitory activity and IPA exhibiting stronger binding stability. However, these assays are limited to in vitro conditions and TNFα, and further validation for other targets and in vivo studies is needed to confirm their therapeutic relevance.

Physicochemical and Pharmacokinetic Properties Analysis
The physicochemical properties of Ginsenoside Rh2 and 3-Indolepropionic acid were analyzed to assess their drug-likeness and solubility profiles. Ginsenoside Rh2 has a molecular formula of C36H62O8 and a MW of 622.87 g/mol. It contains 44 heavy atoms, no aromatic heavy atoms, a fraction of sp3 carbons (Fraction Csp3) of 0.94, 7 rotatable bonds, 8 H-bond acceptors, and 6 H-bond donors. Its MR is 172.26, and TPSA is 139.84 Å2. It violated several drug-likeness Rules, including 2 Lipinski violations, 3 Ghose violations, 1 Egan violation, and 3 Muegge violations, with a bioavailability score of 0.17, 2 Brenk alerts, and a synthetic accessibility score of 8.04, indicating challenges in drug development. In contrast, 3-Indolepropionic acid has a molecular formula of C11H11NO2 and an MW of 189.21 g/mol. It contains 14 heavy atoms, 9 aromatic heavy atoms, a Fraction Csp3 of 0.18, 3 rotatable bonds, 2 H-bond acceptors, and 2 H-bond donors. Its MR is 54.65, and TPSA is 53.09 Å2. 3-Indolepropionic acid exhibited no violations of Lipinski, Ghose, Veber, or Egan Rules, a bioavailability score of 0.85, and a synthetic accessibility score of 1.51, suggesting favorable drug-like properties (Table 4).

The Pharmacokinetic Properties of Ginsenoside Rh2 and 3-Indolepropionic Acid Were Evaluated Using the PkCSM Platform (Table 5). Ginsenoside Rh2 showed a water solubility of −4.522 log mol/L, a Caco2 permeability of 0.196 log Papp (10⁻⁶ cm/s), and an intestinal absorption of 57.315% in humans, with a skin permeability (log Kp) of −2.762. It was a Pgp substrate and a Pgp I inhibitor, but not a Pgp II inhibitor. Its distribution profile included a volume of distribution of −1.046 log L/kg, a fraction unbound of 0.189, a BBB permeability (log BB) of −1.08, and a CNS permeability (log PS) of −3.358. In metabolism, it was a CYP3A4 substrate but did not inhibit CYP1A2, CYP2C19, CYP2C9, CYP2D6, or CYP3A4. Excretion analysis showed a total clearance of 0.44 log ml/min/kg; it was not a renal OCT2 substrate. Toxicity profiling indicated no AMES toxicity, a maximum tolerated dose (MTD) of −1.192 log mg/kg/day, no hERG I inhibition but hERG II inhibition, an oral rat acute toxicity (LD50) of 2.628 mol/kg, chronic toxicity (LOAEL) of 2.826 log mg/kg_bw/day, no hepatotoxicity or skin sensitization, a T. pyriformis toxicity of 0.285 log ug/L, and minnow toxicity of 1.883 log mM. In contrast, 3-Indolepropionic acid exhibited a water solubility of −3.217 log mol/L, a higher Caco2 permeability of 1.203 log Papp (10−6 cm/s), and an intestinal absorption of 93.19%, with a skin permeability of −2.734 log Kp. It was a Pgp substrate but not a Pgp I or II inhibitor. Its distribution profile included a VDss of −1.205 log L/kg, an Fu of 0.325, a log BB of 0.314, and a log PS of −2.249. It was not a substrate or inhibitor of CYP2D6, CYP3A4, CYP1A2, CYP2C19, CYP2C9, or CYP3A4. Excretion showed a total clearance of 0.536 log ml/min/kg, with no renal OCT2 substrate activity. Toxicity profiling revealed no AMES toxicity, an MTD of 1.363 log mg/kg/day, no hERG I or II inhibition, an LD50 of 3.046 mol/kg, a LOAEL of 1.824 log mg/kg_bw/day, no hepatotoxicity or skin sensitization, a T. pyriformis toxicity of 0.287 log ug/L, and minnow toxicity of 1.006 log mM. These results suggest that 3-Indolepropionic acid has more favorable pharmacokinetic properties for absorption and distribution than Ginsenoside Rh2, supporting its potential as a therapeutic candidate for GC treatment.

Discussion
This study employed an integrative systems biology approach to investigate the therapeutic mechanisms of SXKZD in the treatment of GC, focusing on its modulation of gut microbiota, microbial metabolites, and bioactive ingredients. By leveraging dose-weighted network pharmacology, molecular docking, MD simulations, and pharmacokinetic analyses, we demonstrated that SXKZD exerts its anti-GC effects through a multi-target, multi-pathway mechanism. Central to these findings are the hub targets TNF, which consistently emerged as critical nodes across the GM-T, M-T, I-T, and dose-weighted network analyses. These targets are implicated in key biological processes. GO and KEGG enrichment analyses highlighted pathways such as IL-17 signaling, TNF signaling, and apoptosis. These pathways are hallmarks of GC progression, where chronic inflammation drives tumorigenesis, and dysregulated apoptosis facilitates tumor survival and metastasis. The multifaceted action of SXKZD suggests it may counteract these pathological processes holistically, offering a complementary strategy to conventional GC therapies.
A cornerstone of this study is the pivotal role of the gut microbiota in mediating the therapeutic effects of SXKZD. The tertiary regulatory network GM-S-M constructed using the gutMGene database revealed a complex interplay involving key microbial taxa such as Bacteroides, Lactobacillus plantarum, and Akkermansia muciniphila, alongside metabolites Like 3-Indolepropionic acid, butyrate, and succinate. The gut microbiota, comprising trillions of microorganisms, is a dynamic ecosystem critical to host health, maintaining intestinal homeostasis, aiding digestion, and modulating systemic immunity [46]. Microbial dysbiosis can exacerbate tumor progression by promoting inflammation or producing genotoxic metabolites [47]. Our topological analyses identified Lactobacillus plantarum and Akkermansia muciniphila as central players in the GM-T network, with high DC values reflecting their extensive interactions with therapeutic targets. These taxa are known for their anti-inflammatory properties and ability to reinforce the gut barrier, potentially counteracting the pro-tumorigenic microenvironment in GC. Similarly, metabolites in the M-T network are microbial products with documented anti-inflammatory and anti-tumor effects.
Among these metabolites, 3-Indolepropionic acid, a tryptophan-derived product of gut microbial metabolism, exhibits notable therapeutic potential [48]. Molecular docking and MD simulations revealed stable binding interactions between 3-Indolepropionic acid and TNFα, with favorable binding free energies, underscoring its role as a key effector molecule in SXKZD’s anti-GC mechanism. Furthermore, CETSA confirmed that 3-Indolepropionic acid enhanced TNFα thermal stability compared to DMSO. Recent studies have also demonstrated that microbial metabolites like indole-3-acetic acid can elevate CXCL9 and IFN-γ levels, enhancing tumor-infiltrating T-cell abundance and activation. Similarly, supplementation with Lactobacillus johnsonii or 3-Indolepropionic acid has been shown to bolster CD8+ T-cell-mediated αPD-1 immunotherapy efficacy [49], suggesting a broader immunomodulatory role for microbial metabolites in cancer therapy.
Ginsenosides, the primary bioactive compounds in ginseng, are typically administered orally as dietary supplements or prescription drugs [50]. However, their pharmacological efficacy is hindered by poor intestinal absorption due to high polar surface area, extensive hydrogen bonding, and the molecular flexibility imparted by sugar moieties. Fortunately, gut microbiota facilitates the stepwise cleavage of glycosidic or glucuronide groups, yielding secondary glycosides or aglycones (e.g., Ginsenoside Rh2, Ginsenoside Rg3, Compound K, 20(S)-protopanaxatriol, and 20(S)-protopanaxadiol) with improved bioavailability [51, 52]. Ginsenoside Rh2, a rare ginsenoside with potent anti-cancer properties, exerts its effects by promoting apoptosis, inhibiting proliferation, invasion, and metastasis, and inducing cell cycle arrest in various cancers [53]. Within the gut, Bacteroides and Eubacterium further metabolize Ginsenoside Rh2 into 20(S)-protopanaxadiol (PPD), enhancing its systemic availability [54].
Pharmacokinetic profiling further revealed that 3-Indolepropionic acid exhibits superior drug-like properties to Ginsenoside Rh2, including enhanced intestinal absorption, higher solubility, and fewer violations of drug-likeness criteria. These attributes position 3-Indolepropionic acid as a more bioavailable candidate for anti-tumor activity.
Given that TNF emerged as a shared target for both Ginsenoside Rh2 and 3-Indolepropionic acid in network analysis, molecular docking, MD simulations, and pharmacokinetic studies, an intriguing hypothesis arises: could Ginsenoside Rh2 intake indirectly contribute to 3-Indolepropionic acid production and its therapeutic effects? Structurally, Ginsenoside Rh2 (C36H62O8), a triterpenoid saponin with a dammarane backbone, differs significantly from 3-Indolepropionic acid (C11H11NO2), which features an indole ring derived from tryptophan. Direct biotransformation of Ginsenoside Rh2 into 3-Indolepropionic acid is improbable due to the absence of an indole precursor in its structure. However, an indirect mechanism is plausible: Ginsenoside Rh2 modulates the gut microbiota to enhance 3-Indolepropionic acid synthesis from dietary tryptophan. For instance, Lactobacillus plantarum and Clostridium species, known tryptophan metabolizers, may be enriched by Ginsenoside Rh2 or its metabolite PPD, thereby increasing 3-Indolepropionic acid levels. This hypothesis aligns with the observed overlap in the GM-T, M-T, and I-T networks. Moreover, prior studies indicate that Ginsenoside Rh2 alters microbial composition in liver fibrosis models, increasing Bacteroidetes and Tenericutes while reducing Firmicutes, Proteobacteria, and Actinobacteria [55], further supporting its microbiota-modulating potential. This hypothesis is speculative and requires validation through targeted experiments, such as in vitro microbial culturing with Ginsenoside Rh2 and liquid chromatography-mass spectrometry (LC-MS) to quantify 3-Indolepropionic acid levels.
Despite its potent anti-cancer properties, Ginsenoside Rh2’s poor oral bioavailability limits its direct clinical applicability due to its high molecular weight, extensive hydrogen bonding, and multiple drug-likeness violations. Gut microbiota-mediated conversion to PPD enhances its systemic availability [56]. However, formulation strategies, such as nanoparticle encapsulation or structural modification to reduce polar surface area, are needed to improve its therapeutic potential. Future studies will explore these approaches to optimize Ginsenoside Rh2’s pharmacokinetic profile for clinical use. In contrast, 3-Indolepropionic acid demonstrates a favorable pharmacokinetic profile, with high bioavailability, low TPSA, and compliance with Lipinski’s Rule, positioning it as a promising orally viable therapeutic agent. Unlike Ginsenoside Rh2, which may require non-oral delivery or formulation optimization, 3-Indolepropionic acid’s properties suggest potential as a standalone agent or adjuvant, particularly given its immunomodulatory effects that may synergize with immunotherapies.
However, several limitations warrant consideration. The proposed indirect biotransformation pathway from Ginsenoside Rh2 to 3-Indolepropionic acid lacks direct experimental evidence, and the specific microbial enzymes and intermediates involved remain unidentified. Inter-individual variability in gut microbiota composition, influenced by diet, genetics, and lifestyle, may also affect the reproducibility of SXKZD in clinical settings. Additionally, the gut microbiota data presented in this manuscript were derived from fecal samples (gutMGene database). Compared to biopsy samples, microbial composition from fecal samples varies between individuals. It may not fully reflect the true state of the gut microbiota, particularly in the mucosal environment where interactions with the host are most direct. The data used in this study for network pharmacology predictions require further validation using additional methods to confirm the accuracy of the gut microbiota profiles.
To address these gaps, future research should prioritize: (1) microbial profiling via 16 S rRNA sequencing or metagenomics to assess SXKZD-induced changes in GC models, focusing on 3-Indolepropionic acid-producing taxa; (2) metabolomics using LC-MS/MS to quantify 3-Indolepropionic acid levels in gut and plasma following SXKZD administration, with and without tryptophan supplementation; (3) in vitro co-culture experiments combining Ginsenoside Rh2 with tryptophan-metabolizing bacteria to explore co-metabolic pathways; and (4) enzyme activity assays to identify microbial glycosidases and tryptophanases involved in these processes. Longitudinal studies in preclinical models could further evaluate the durability of microbial and therapeutic effects, while clinical trials could assess the efficacy and safety of SXKZD in GC patients.

Conclusion
In conclusion, this study elucidates the gut microbiota-mediated mechanisms of SXKZD in GC treatment, highlighting the synergistic roles of Ginsenoside Rh2 and 3-Indolepropionic acid in targeting TNF and modulating inflammation and immunity. By integrating systems biology and dose-weighted network pharmacology, we provide a comprehensive framework for understanding multi-target action, offering novel insights into the therapeutic potential of SXKZD and paving the way for microbiota-targeted strategies in cancer management.

Supplementary Information