3.1 High Frequency Herbal Medicines and Association Rule Analysis for Herb Combinations
Of the 216 herbal medicines included in the 114 prescriptions identified by literature screening, the total frequency of drug use was 1613 times (Supplementary Table 1). Atractylodis macrocephalae Rhizoma (Bai Zhu), Poria (Fu Ling), Hedyotis diffusa Willd. (Bai Hua She She Cao), Astragali Radix (Huang Qi), and Glycyrrhizae Radix Et Rhizoma (Gan Cao) were the herbal medicines most frequently used in the clinic (Fig. 1). Twenty-four Chinese medicines used more than 15 times were identified (Fig. 2A). These results suggest that those herbs were preferred for treatment of PC.
Chinese herbal medicine compatibility refers to the purposeful combination of two or more herbs, according to clinical requirements and pharmacodynamic effects, and is the main method used for clinical drug application and the basis for the composition of Chinese herbal medicine prescriptions.
Association rule analysis was carried out for 24 medicines used at high frequency using an apriori algorithm. We focused on two parameters: support and confidence level, where support was set as ≥ 20% and confidence level as ≥ 80% to obtain the top 10 herb pairs and suitable association rules (Table 1). The association between Atractylodis macrocephalae rhizoma (Bai Zhu) and Poria (Fu Ling) had the highest degree of support (52.63%), while those of Glycyrrhizae radix et rhizoma (Gan Cao) and Atractylodis macrocephalae rhizoma (Bai Zhu) with Poria (Fu Ling) had the highest confidence level (92%). The resulting association network diagram is presented in Fig. 2B.
Table 1
Apriori algorithm-based association rules for herbs used to treat pancreatic cancer
No.
|
Association rules
|
Support(%)
|
Confidence
(%)
|
1
|
Atractylodis macrocephalae rhizoma (Bai Zhu) => Poria (Fu Ling)
|
52.63
|
81.67
|
2
|
Poria (Fu Ling) = > Atractylodis macrocephalae rhizoma (Bai Zhu)
|
51.75
|
83.05
|
3
|
Scutellariae barbatae herba (Ban Zhi Lian) = > Hedyotisdiffusa Willd (Bai Hua She She Cao)
|
28.07
|
81.25
|
4
|
Citri reticulatae pericarpium (Chen Pi) = > Poria (Fu Ling)
|
26.32
|
80.00
|
5
|
Astragali radix (Huang Qi) and atractylodis macrocephalae rhizoma (Bai Zhu) => Poria (Fu Ling)
|
26.32
|
80.00
|
6
|
Astragali radix (Huang Qi) and Poria (Fu Ling) = > Atractylodis macrocephalae rhizome (Bai Zhu)
|
25.44
|
82.76
|
7
|
Codonopsis radix (Dang Shen) and atractylodis macrocephalae rhizoma (Bai Zhu) => Poria (Fu Ling)
|
22.81
|
88.46
|
8
|
Glycyrrhizae radix et rhizoma (Gan Cao) and atractylodis macrocephalae rhizoma (Bai Zhu) => Poria (Fu Ling)
|
21.93
|
92.00
|
9
|
Coicis semen (Yi Yi Ren) and atractylodis macrocephalae rhizoma (Bai Zhu) => Poria (Fu Ling)
|
21.93
|
82.00
|
10
|
Hedyotisdiffusa Willd (Bai Hua She She Cao) and Poria (Fu Ling) = > Atractylodis macrocephalae rhizoma (Bai Zhu)
|
21.05
|
91.67
|
3.2 Rules for Combination of Herbal Medicines Based on Cluster Analysis
Clustering classification is widely used to determine the compatibility of herbs and the rules for combination of different Chinese medicines. Here, we applied hierarchical cluster analysis to identify the core herbs that are used in combination for treatment of PC. The 24 drugs mentioned in the literature at the highest frequency were classified into five categories according to traditional Chinese medicine theory (Fig.3). Based on compatibility rules and clinical experience, the core prescription used for PC treatment included four herbs:Glycyrrhizae Radix et Rhizome (Gan Cao), Codonopsis Radix (Dang Shen), Citri Reticulatae Pericarpium (Chen Pi), andPinelliae Rhizoma (Ban Xia). Our results showed that these fourherbs, which are frequently used in the clinic, areoften used in combination to treat PC.
3.3 Identification of Potential Targets of Core Prescription for PC Treatment
To investigate the possible mechanism underlying the core prescriptions used to treat PC, the targets of the four selected herbs were obtained from the TCMSP database. In total, 295, 256, 141, and 365 potential targets were identified for Glycyrrhizae Radix et Rhizome (Gan Cao), Codonopsis Radix (Dang Shen), Citri Reticulatae Pericarpium (Chen Pi), and Pinelliae Rhizoma (Ban Xia), respectively. Importantly, 84 targets were shared by the four herbs, and these were defined as the core prescription targets. Target proteins were associated with tumors and apoptosis (e.g., TP53, TNF, BAX, BCL2, CASP3, and CASP9, among others).
In addition, 2940 PC-related proteins were identified from the DisGeNET database. Among the 84 core prescription targets, 44 overlapped with proteins in the 2940 PC-related group (hypergeometric p-value < 9.04e-23; Fig. 4A). The 44 common proteins identified as both targets of the herbs and related to PC were considered to represent likely targets of the herbal medicines during PC treatment.
These 44 shared proteins were imported into STRING, and an Herb–PC target PPI network constructed using Cytoscape (Fig.4B).From this PPI network, several nodes (TNF, AKT1, TP53, HSP90AA1, MMP9, JUN, CASP3, and IL6) had high degree values. Fig.4. Targets of core prescriptions used for PC treatment. (A) Molecules in common between herb targets and PC-associated proteins. (B) PPI network diagram of the common targets of the four core herbs used to treat pancreatic cancer. The PPI network contains 44 nodes and 297 edges. Circles represent protein targets; orange circles indicate higher degree values. The node size of gene targets is proportional to the number of degrees.
3.4 GO and KEGG Enrichment Analysis
To elucidate the potential molecular mechanisms by which core prescriptions act on PC, GO biological process and KEGG pathway enrichment analyses were performed using the 44 identified core proteins.
The top 20 enriched GO biological process terms were determined (Fig. 5A), and analysis showed that the targets were closely related to processes involved in responses to steroid hormones and apoptotic signaling pathways. The most significantly enriched KEGG pathways included those involved in cancer, hepatitis B, apoptosis, p53 signaling, and PI3K/Akt signaling (Fig. 5B).
Target–GO term andTarget–KEGG pathway networks were then constructed, basedon the targets involved in each GO term or KEGG pathway. The Target–GO network comprised 46 nodes and 138 edges (Fig.5C). The majority oftargets were primarily implicated in responses to steroid hormones and apoptotic signaling pathways. In addition, the targets participating in the largest number of terms were PTGS2 (also known as COX-2), AKT1, and TNF, which were involved in 10, 9, and 8 GO terms,respectively. The Target–KEGG network included 36 nodes and 126 edges (Fig.5D).The results of both GO and KEGG analysis suggest that the mechanism of action of core prescriptions for treatment of PC involves stimulation of responses to steroid hormones and apoptotic signaling pathways.Fig.5. Functional analysis of common targets. (A) GO enrichment analysis of putative targets. (B) Target–GO network terms. (C) KEGG pathway enrichment analysis of putative targets. (D) Target–Signaling pathway network. Pink diamond nodes represent main signaling pathways and blue circle nodes refer to putative common targets of the four core herbs used for treatment of pancreatic cancer. Node size is proportional to the number of degrees.
3.5 Molecular Docking
To evaluate whether active compounds from core prescription components that possess good pharmacokinetic properties could bind directly to proteins involved in responses to steroid hormone, we applied molecular docking analysis to explore potential binding modes. The top 10 compounds with highest oral bioavailability and drug-likeness values for each herb were identified as active compounds, and included flavonoids, alkaloids, amino acids, steroids, and volatile oils, among other substances (Supplementary Table 2).
As shown in Fig. 6A, stigmasterol could bind to PTGS2 with the lowest binding energy (− 10.2 kcal/mol). The binding site of stigmasterol in PTGS2 was GLY-225. Further, the binding sites for phaseol in PTGS2 were TYR-130 and VAL-47 and the binding energy for phaseol with PTGS2 was − 10.1 kcal/mol (Fig. 6B). These results suggest that stigmasterol and phaseol could directly bind to PTGS2.
Furthermore, perlolyrine could bind to ESR1 with a binding energy of − 8.8 kcal/mol and DIOP bind to ESR2 with the same binding energy (Fig. 6C,D). Notably, phaseol and AR were able to bind with a free binding energy of − 8.6 kcal/mol, while the free binding energy of licopyranocoumarin with PGR was − 9.7 kcal/mol (Fig. 6E,F).
These results indicate that several active compounds from the four identified medicines could bind to proteins that function in responses to steroid hormones.Fig.6. Schematic 3D representation of molecular docking models, active sites, and binding distances. Binding modes of: stigmasterol to PTGS2 (A), phaseol to PTGS2 (B), perlolyrine to ESR1 (C), DIOP to ESR2 (D), phaseol to AR (E), and licopyranocoumarin to PGR (F).