The therapeutic targets for GC were obtained from the Malacard (https://www.malacards.org/) (Rappaport et al., 2013) module in GeneCards (Rappaport et al., 2017) (https://www.genecards.org/) using “gastric cancer” as the keyword for the search (Retrieval deadline: 2020.10).

The active ingredients of HJT were searched in the Traditional Chinese Medicine Integrated Database (TCMID, http://www.megabionet.org/tcmid/) (Huang et al., 2018), Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP, https://old.tcmsp-e.com/tcmsp.php) (Ru et al., 2014), and Herb Ingredients’ Targets (HIT, http://hit2.badd-cao.net/) (Ye et al., 2011) databases. A total of 331 active compounds were obtained following the exclusion of compounds that lacked target information. Additionally, the Search Tool for Interacting Chemicals (STITCH, http://stitch.embl.de) (Szklarczyk et al., 2016) database was used to retrieve the compound targets. A total of 8,322 targets were obtained by selecting targets with compound-target association scores greater than 400 in the STITCH database and normalizing the target information using the Gene module in the NCBI database.

Drug-likeness evaluation of chemical components was mainly used to evaluate the absorption, distribution, metabolism, and excretion properties of compounds. An effective quantitative estimate of the drug-likeness evaluation index described previously (Bickerton et al., 2012; Liang et al., 2014; Yang et al., 2018) was used to identify pharmaceutically active compounds in HJT. By referring to DrugBank drugs with a quantitative estimate of drug-likeness value greater than 0.2, 292 compounds satisfying the drug-likeness evaluation were obtained.

Finally, by using a binomial statistical model (Liang et al., 2014; Yang et al., 2018) to screen the core targets of HJT, 738 major targets and 257 major compounds were identified.

HJT-related and GC-related targets were subjected to Venny analysis to identify the cross-targets, and ultimately reveal the corresponding active compounds. The cross-targets were then uploaded to the Search Tool for the Retrieval of Interacting Genes (STRING) Database (https://string-db.org). The PPI network was analyzed using CytoHubba, a plug-in of Cytoscape (v3.8.0), and screened with a median value of degrees (Sun et al., 2019).

In the network, the size, form, and color of a node represent the value of the degree. Of note, the value of the degree increases as the node becomes more important.

The RGUI and ClusterProfiler packages in R (4.0) (http://bioconductor.org/packages/release/bioc/html/clusterProfiler.html) (Yu et al., 2012) were used to perform the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses, with adjusted p-values of <0.01 using the Bonferroni algorithm (Retrieval deadline: 2020.10).

The chemical structure of the active compound was obtained from the Zinc [23] database (http://zinc15.docking.org) and then imported into AutoDockTools-1.5.6 software to add polar hydrogen, distribute the charge, and set the rotatable bond. The resulting structure was saved in “pdbqt” format. The 3D structure of the protein was retrieved from the Protein Data Bank (PDB, http://www.rcsb.org/) and then inputted into PyMOL (v2.3.0) to remove water molecules, co-crystallized ligands, and ions. The AutoDockTools-1.5.6 software was used to add polar hydrogen and distribute the charge; the resulting structure was saved in “pdbqt” format. Molecular docking analysis was performed using AutoDock Vinna (v1.1.2) software. Affinity reflects the score for molecular docking; a lower score indicates stronger binding affinity. In this study, an affinity of less than −7 kcal/mol was considered to indicate strong binding activity.

HJT herbal drink (448 g, Tongrentang, Beijing, China) was soaked in 1,000 ml of distilled water for 30 min and condensed and refluxed for 60 min. The filtrate was then collected and concentrated to 170 ml under reduced pressure in a rotary evaporator at 45°C to obtain a final HJT concentration of 2.64 g/ml. Finally, the HJT solution was diluted with water to concentrations of 0.66 and 1.32 g/ml.

HPLC analysis were performed using the Agilent 1260 HPLC system (Agilent Technologies, Santa Clara, CA, United States) equipped with a Zorbax C₁₈ column (250 mm ⅹ 4.6 mm, 5 μm). The following gradient elution of methanol-water was performed: 0–10 min, 30% methanol; 10–20 min, 40% methanol; 20–30 min, 50% methanol; 30–40 min, 65% methanol; 40–50 min, 80% methanol; 50–60 min, 95% methanol. The column temperature was maintained at 25°C, the detection wavelength was 250 nm, the flow rate was 1.0 ml/min, and the injection volume was 10 μL.

Human GC cells (MGC-803) were purchased from the Chinese Academy of Sciences Shanghai Cell Bank (Shanghai, China) and cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (Gibco, Grand Island, NY, United States) at 37°C with 5% CO₂.

Male BALB/c nude mice (6-week-old) were housed in a controlled environment under a 12-h light/dark cycle with free access to water and food. MGC-803 cells (2 × 10⁶) were inoculated subcutaneously into the right side of each mouse, and tumor size was observed daily. When the tumor reached approximately 3 mm × 3 mm, mice were randomly divided into four groups (n = 5 in each group). Mice in the HJT group were administered 0.66 g/ml (low-dose) and 1.32 g/ml (high-dose) HJT (0.5 ml/20 g) via gavage each day; mice in the model group were administered equal amounts of saline via gavage; and mice in the positive control group were administered an intraperitoneal injection of 5-fluorouracil (5-FU, 20 mg/kg, Sigma-Aldrich, St. Louis, MO, United States) every 2 days. The tumor volume was calculated every 3 days (formula: A × B²/2, where A is the long diameter and B is the short diameter); only one measurement was performed. After 15 days, mice were killed, and the tumors were isolated and weighed. All animal experiments were approved by the Experimental Animal Management and Ethics Committee of Xiamen University.

To detect cellular apoptosis, the tumor tissue sections were deparaffinized and rehydrated. The sections were then incubated with 20 μg/ml proteinase K for 30 min at 37°C and TdT reaction mix for 60 min at 37°C. After three rounds of washing with phosphate-buffered saline, the sections were stained with 4′,6-diamidino-2-phenylindole staining solution for 10 min at room temperature in the dark. Finally, the sections were observed using a fluorescence microscope (Olympus, Tokyo, Japan).

Total RNA was extracted from the tumor tissues using TRIzol reagent (Invitrogen, Carlsbad, CA, United States) and reverse-transcribed into cDNA using cDNA reverse transcription kits (TransGen Biotech Co., Ltd., Beijing, China). SYBR Green PCR Master Mix (Takara, Shiga, Japan) was used to amplify the samples on the MX3000P Real-Time QPCR System (Agilent Technologies, Santa Clara, CA, United States). The following cycling conditions were employed for PCR: 95°C for 3 min, 40 cycles at 95°C for 12 s, and 60°C for 40 s. Relative mRNA expression was detected using the 2^−ΔΔCt method, with Gapdh as the reference gene. The primers used in the qPCR assays are listed in Supplementary Table S1.

Tumor tissue homogenates were prepared in lysis buffer (Beyotime, Shanghai, China). The proteins in the homogenates were separated via sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto polyvinylidene fluoride membranes. After blocking with 5% skim milk for 90 min, the membranes were incubated with anti-Mapk3 (ab32537), anti-Akt1 (ab227385), anti-Vegfa (ab214424), and anti-Gapdh (ab9485) (1:1000, Abcam, Cambridge, United Kingdom) overnight at 4°C, followed by the corresponding secondary antibody (1:1000, ab6721, Abcam) for 1 h at room temperature. The blots were detected using an enhanced chemiluminescence kit and GeneGnome XRQ (Gene Company, China). The gray values of the bands were quantified using ImageJ software (NIH, Bethesda, MD, United States).

GO and KEGG pathway enrichment analyses were performed with adjusted p-values of <0.01 based on the Bonferroni algorithm.

All data are expressed as mean ± SD and were analyzed using GraphPad Prism 7.0 software (GraphPad, Inc., La Jolla, CA, United States). One-way analysis of variance followed by Tukey’s test was used for multiple comparisons among groups. A p value <0.05 indicated statistical significance.

A total of 538 GC targets were obtained from the GeneCards database. Further, 72 cross-targets were obtained between HJT and GC using Venny (Figure 1A; Table 1). Cytoscape (v3.8.0) was used to construct the PPI network, which contained 72 nodes and 1,194 edges with an average degree value of 33.2 (Figure 1B). Ten hub genes (FOS, EGF, MAPK1, CASP3, MAPK3, JUN, AKT1, VEGFA, MYC, and TP53) were selected using CytoHubba software with high degrees. Thereafter, a hub gene network of 10 nodes and 45 edges was constructed using the genes listed above (Figure 1C). A total of 89 bioactive compounds that met the criteria for drug-likeness screening and had GC treatment targets were obtained using the 72 cross-targets. A total of 63 bioactive compounds were selected from 89 bioactive compounds based on a frequency ≥10; further details are provided in Table 2. Cytoscape was used to construct a visual HJT herb-component-target network, which consisted of 139 nodes and 942 edges. Quercetin, folic acid, oxaliplatin, choline, IFP, and berberine had a high degree in this study, suggesting that they play important roles in the effects of HJT on GC (Figure 2).

To further investigate the 72 cross-targets of the HJT bioactive compounds, GO enrichment analysis was performed to elucidate the biological process (BP), molecular function (MF), and cellular component (CC) terms. A total of 41 MFs, 1,337 BPs, and 28 CCs were found to be enriched (p < 0.01). The GO analysis results for the top 15 markedly enriched MF, BP, and CC terms are shown in Figures 3A–C. KEGG pathway enrichment analysis revealed that the 72 cross-targets of the HJT bioactive compounds were enriched in 140 pathways (p < 0.01). The top 15 enriched KEGG pathways for the HJT bioactive compounds are shown in Figure 3D. Furthermore, to illustrate the correlation between cross-targets and the top 15 BP-related terms and top 15 KEGG pathways, a target-BP-pathway network diagram was constructed (Figure 4).

The main active ingredients in HJT were analyzed by HPLC. By comparing HJT to the standard, formononetin was identified as the dominant compound, followed by cinnamaldehyde, gingerol, ursolic acid, anethole, and berberine (Supplementary Figure S1).

To validate the candidate GC targets in HJT, molecular docking was used to identify the binding ability between the bioactive components (berberine, formononetin, ursolic acid, gingerol, anethole, and cinnamaldehyde). The corresponding 2D-chemical structures of HJT and the hub targets (MYC, VEGFA, AKT1, JUN, MAPK3, CASP3, MAPK1, EGF, and TP53), generated using Zinc software, are presented in Table 3. As shown in Table 4, 41 pairs of docking results were obtained, and berberine had a strong binding activity to MAPK3 (affinity = -9.2 kcal/mol). The top 10 affinities of the combination of bioactive components and hub targets are shown in Figure 5.

Nude mice xenografted subcutaneously with MGC-803 cells were used to investigate the potential therapeutic effects of HJT. Compared to treatment with the model compound, HJT and 5-FU significantly decreased the tumor size and weight (p < 0.01). Moreover, compared with low-dose HJT and 5-FU, high-dose HJT most effectively inhibited tumor growth (p < 0.01) (Figure 6A).

The apoptotic effects of HJT on tumor tissues were determined using a TUNEL assay. As shown in Figure 6B, treatment with high-dose HJT led to a higher apoptosis rate of tumor tissues.

To further study the anti-GC effects of HJT, we selected the hub genes identified in the network pharmacology analysis as potential targets of HJT for qRT-PCR and western blot analysis. Based on the qRT-PCR results, the expression levels of Akt1, Casp3, Egf, Jun, Mapk1, Myc, Tp53, Mapk3, and Vegfa in the HJT and 5-FU treatment groups were notably downregulated compared to those in the model group, (p < 0.05, p < 0.01). Further, based on western blotting, the protein expression levels of Mapk3, Akt1, and Vegfa in the treatment group were significantly decreased compared to those in the model group (p < 0.05, p < 0.01) (Figure 7).