Pancreatic cancer peritoneum effusion samples were obtained from two patients at Huadong Hospital Affiliated to Fudan University (Shanghai, China) for testing the cell viability. Fresh samples were collected from two patients upon obtaining written informed consent in Oct. 2018. The detailed information concerning these two patients are summarized in Table I. Cells were acquired from peritoneum effusion by centrifugation at 1,500 × g for 5 min. This study was approved by the Internal Review and Ethics Boards of Huadong Hospital. The pancreatic cancer cell lines PANC-1 and BxPC3 were purchased from the American Type Culture Collection (ATCC; Manassas, VA, USA). Cells were cultured in RPMI-1640 medium, supplemented with 10% fetal bovine serum at 37°C with 5% CO₂.

The human genes associated with β-elemene were acquired from the BATMAN-TCM (http://bionet.ncpsb.org/batman-tcm/) database. BATMAN-TCM is an online bioinformatics analysis tool specially designed for studying the molecular mechanisms of traditional Chinese medicine, and is based on traditional Chinese medicine ingredients' target prediction (25). The keyword ‘6918391’, which is the PubChem CID (https://pubchem.ncbi.nlm.nih.gov/compound/6918391) of β-elemene, was input into the BATMAN-TCM database. To make the results more credible, the cutoff score was set at 5.

The human genes associated with pancreatic cancer were acquired from the DigSee (http://210.107.182.61/geneSearch/) database. DigSee is a text mining search engine that provides evidence sentences describing which ‘genes’ are involved in the development of ‘disease’ through ‘biological events’ (26,27). The keyword ‘pancreatic neoplasms’ was input into the DigSee database. To make the results more credible, the required number of academic papers was set to 5.

The Search Tool for the Retrieval of Interacting Genes (STRING, https://string-db.org/, version 10.5) database, which provides information regarding the predicted and experimental interactions of proteins, was used to obtain protein-protein interaction (PPI) data (28). Networks of β-elemene-target genes and pancreatic cancer-target genes were constructed using PPI analysis based on β-elemene-target genes and pancreatic cancer-target genes, respectively. The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of target genes in common between β-elemene and pancreatic cancer were performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID, http://david.abcc.ncifcrf.gov/, version 6.8) database (29,30). A network of the common target genes of β-elemene and pancreatic cancer was constructed by linking PPI analysis and KEGG pathway analysis based on the common target genes. All networks were visualized utilizing Cytoscape software (31) (https://cytoscape.org/, version 3.4.0). In addition, the Cytoscape plugin Molecular Complex Detection (MCODE) was applied to analyze clustering modules in the network of target genes in common between β-elemene and pancreatic cancer.

The molecular docking simulation was conducted with AutoDockTools (32) (http://mgltools.scripps.edu/, version 1.5.6), which is an automated docking software designed to predict how small molecules bind to a receptor of known three dimensional (3D) structure, to find the preferred binding conformation of HIF1A and β-elemene. The human HIF1A structure was retrieved from the PDB database (PDB code: 1H2M). PubChem was referenced for the 3D structures of β-elemene. The preparation work included affixing hydrogen atoms and removing co-crystallized ligands as well as water molecules from the 1H2M. The detailed docking operation process strictly followed the software instruction manual. To cover all residues with grid map, each grid point in the x, y, and z axes was 126×126×126 with X=21.796, Y=28.466, Z=30.538 co-ordinates. In every docking test, 10 runs were executed with the population size set at 150 and a maximum number of evaluation of 2,500,000. The remaining parameters were set to default. To visualize the 3D structure of the docking result, PyMOL software (33) (https://pymol.org/2/, version 2.2) was used.

The Cancer Genome Atlas (TCGA) is a collaboration between the National Cancer Institute and the National Human Genome Research Institute that has generated comprehensive, multi-dimensional maps of the key genomic changes in 33 types of cancer (34). The dataset of pancreatic adenocarcinoma (N=177, tumor samples with mRNA data) from TCGA was selected and a z-score threshold was set at ±2.0. Genetic alteration and co-expression of HIF1A and VEGFA were analyzed using cBioPortal (35,36) (http://www.cbioportal.org/). Analysis of overall survival (OS) was performed using Kaplan-Meier Plotter (37) (http://kmplot.com/analysis/).

PANC-1, BxPC3, and cells from two peritoneum effusion samples were seeded in 96-well plates at 5,000 cells/well in RPMI-1640 medium, supplemented with 10% fetal bovine serum at 37°C with 5% CO₂ and cultured overnight. Serial concentrations of β-elemene [CSPC Pharmaceutical Group Company Ltd. (YUANDA), Dalian, China] were added to the cells: 0, 0.5, 1, 2, 4, 8 and 16 µM. The stocking solution used here was prepared with an evenly-distributed aqueous solution of β-elemene, not the conventional emulsion dosage form of β-elemene. After 72 h of culture, the cells were incubated with 100 µl of 0.5 mg/ml sterile MTT (3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide; Sigma-Aldrich; Merck KGaA) for 4 h at 37°C. Then, the culture medium was removed and 150 µl of DMSO (Sigma-Aldrich; Merck KGaA) was added. The absorbance values were measured at 490 nm. The assay was performed in three replicates, and three parallel experiments were performed for each sample.

Total protein was extracted from PANC-1 and BxPC3 cells with radioimmunoprecipitation assay lysis buffer (Beijing Solarbio Science & Technology Co., Ltd., Beijing, China) and quantified using a Bicinchoninic Acid Protein Assay Kit (Pierce; Thermo Fisher Scientific, Inc.). A total of 10 µg protein was denatured and separated via sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). The percentage of stacking gel and separation gel used in this study was 5 and 10%, respectively. The proteins were transferred to polyvinylidene difluoride membranes (EMD Millipore), followed by blocking with 5% bovine serum albumin and incubation with primary antibodies (dilution 1:1,000) against GAPDH (cat. no. 2118; Cell Signaling Technology, Inc.), HIF1A (cat. no. 14179; Cell Signaling Technology, Inc.), and VEGFA (cat. no. ab52917; Abcam) overnight at 4°C. The membranes were then incubated with secondary antibody (dilution 1:2,000) (cat. no. 7074; Cell Signaling Technology, Inc.) for 1 h at 37°C. Finally, the membranes were imaged using the ChemiDOC™ XRS system (Bio-Rad Laboratories, Inc.) following detection with an enhanced chemiluminescence kit (Beijing Solarbio Science & Technology Co., Ltd.). The protein expression levels were calculated based on the greyscale values of the blots in ImageJ (version 1.52; NIH) and normalized to that of GAPDH.

Total RNA was extracted using the RNeasy mini kit (Qiagen) according to the manual without further modifications. RNA purity and quantity were assessed with NanoDrop 100 spectrophotometer (Thermo Scientific, Inc.). cDNA was synthesized with the Verso cDNA Synthesis kit (Thermo Scientific, Inc.) following the protocol described by the manufacturer. For qPCR, GoTap^® qPCR Master Mix (Promega) was utilized to set up the reactions as suggested by the manufacturer. The sequences of the primers are: GAPDH (upstream GCACCGTCAAGGCTGAGAAC and downstream TGGTGAAGACGCCAGTGGA); HIF1A (upstream TCAAAGTCGGACAGCCTCAC and downstream ATCCATTGATTGCCCCAGCA); VEGFA (upstream TTGCAGATGTGACAAGCCGA and downstream GGCCGCGGTGTGTCTA).

SPSS statistical software (version 22.0; IBM Corp.) was used for statistical analysis. One-way analysis of variance followed by Tukey's multiple comparison post-hoc test was performed and a Student's t-test was used when only two groups were compared. Data are presented as the mean ± standard deviation from three independent biological replicates. P<0.05 was considered to be statistically significant.

To acquire a list of human genes associated with β-elemene, the BATMAN-TCM database was employed. The PubChem CID of β-elemene is 6918391. After entering the PubChem CID of β-elemene (6918391) as the keyword into the BATMAN-TCM database with a cutoff score ≥5, 522 β-elemene-target genes were obtained. A protein-protein interaction (PPI) network was built using the String database and Cytoscape software to uncover the relationships between these β-elemene-target genes. In total, 519 nodes and 3,672 edges were present in the network of β-elemene-target genes (Fig. 1).

The DigSee database was used to search for human genes relevant to pancreatic cancer. The required number of academic papers was set to be >5 on the website and 319 pancreatic cancer-target genes were obtained. Similarly, a PPI network was constructed to establish the relationship between these pancreatic cancer-target genes. A total of 318 nodes and 8,733 edges made up the network of the pancreatic cancer-target genes (Fig. 2).

To further uncover the potential pharmacological mechanisms of β-elemene activity against pancreatic cancer, target genes common to both β-elemene and pancreatic cancer were selected in both networks. A total of 33 genes belonging to both the β-elemene-target gene and pancreatic cancer-target gene networks were screened via Venn analysis (Fig. 3A). Additionally, KEGG pathway analysis based on the DAVID database was performed to determine the potential biological roles of these 33 genes. Subsequently, a network of target genes for β-elemene against pancreatic cancer was built using the String database and Cytoscape software (Fig. 3B). Two modules were identified by MCODE. Module 1 consisted of HIF1A, PTEN, PPARG, HGF, IL10, IL1B, JAK2, PTPN11, IGF1R, IGF1, MAPL1, MAP2K1 and CXCL12, and was highly relevant to proteoglycans in cancer, focal adhesions, and the HIF-1, PI3K-Akt, FoxO, Rap1, and Ras signaling pathways, among others. Module 2, containing CDK2 and CCND1, was of great importance in the PI3K-Akt, FoxO and p53 signaling pathways.

Molecular docking mimics the binding between molecules and their corresponding receptors. To provide deeper insight into the binding interactions between β-elemene and HIF1A, a molecular docking simulation was performed using AutoDockTools (http://mgltools.scripps.edu/, version 1.5.6) yielding a binding energy of −5.3 kcal/mol, a ligand efficiency of −0.35 and an inhibitor constant of 129.58 µM. Collectively, these results suggest the potential of direct binding between β-elemene and HIF1A (Fig. 4).

The expression profiles of HIF1A and VEGFA were not separate from each other. The dataset of pancreatic adenocarcinoma (N=177, tumor samples with mRNA data) from TCGA was selected and analyzed with cBioPortal. HIF1A and VEGFA were both upregulated in pancreatic cancer (Fig. 5A). Furthermore, there was a trend of co-expression between HIF1A and VEGFA, verified by bioinformatic analysis. (Fig. 5B). Additionally, elevated expression of HIF1A and VEGFA corelated with poor overall survival, as evaluated using Kaplan-Meier Plotter (Fig. 5C and D).

To investigate the antitumor effect of β-elemene in pancreatic cancer, an MTT assay was performed using PANC-1, BxPC3 cells and cells from the peritoneum effusion of two pancreatic cancer patients. The cells were treated with serial concentrations of β-elemene. Analysis of the PANC-1 and BxPC3 cells revealed that treatment with β-elemene reduced cell viability (Fig. 6A and B). In addition, cells from the peritoneum effusion of two pancreatic cancer patients demonstrated decreased cell viability when treated with β-elemene (Fig. 6C and D). The IC₅₀ values of β-elemene were 6.94±0.86, 17.36±1.25, 15.80±0.63, 14.86±0.69 µM in the PANC-1 and BxPC3 cell lines, and in the two peritoneum effusion samples, respectively. Collectively, β-elemene showed dose-dependent antitumor activity in pancreatic cancer.

Based on our network pharmacology analysis, β-elemene is expected to ameliorate malignant peritoneum effusion in pancreatic cancer by targeting the HIF1A/VEGFA pathway. Thus, an in vitro experiment was conducted to further explore the molecular mechanism of β-elemene against peritoneum effusion in pancreatic cancer. The HIF1A protein levels and those of its downstream target, VEGFA, were evaluated via western blotting assay in both PANC-1 and BxPC3. Cells were treated with β-elemene at concentrations of 0, 2, 4, and 8 µM. As the concentration of β-elemene increased, the expression of both HIF1A and VEGFA decreased, especially at the highest dosage (Fig. 7A and B), implying that β-elemene attenuates the expression of HIF1A, thereby inhibiting the expression of its downstream target, VEGFA. Apart from the evaluation of the protein expression, RT-qPCR was also conducted to assess expression of these genes at the RNA level. Similarly, the mRNA levels of HIF1A and VEGFA were decreased as the concentration of β-elemene increased in the PANC-1 and BxPC3 cell lines (Fig. 8), which was in concordance with the western blot assay. Taken together, all these results demonstrate that β-elemene functions to inhibit the HIF1A/VEGFA pathway, thereby conferring suppression of malignant peritoneum effusion in pancreatic cancer.