A network pharmacology approach to reveal the protective mechanism of Salvia miltiorrhiza-Dalbergia odorifera coupled-herbs on coronary heart disease

Salvia miltiorrhiza-Dalbergia odorifera coupled-herbs (SMDOCH) has been used to treat coronary heart disease (CHD) for thousands of years, but its unclear bioactive components and mechanisms greatly limit its clinical application. In this study, for the first time, we used network pharmacology to elucidate the mechanisms of action of SMDOCH on CHD. We collected 270 SMDOCH-related targets from 74 bioactive components and 375 CHD-related targets, with 58 overlapping common targets. Next, we performed enrichment analysis for common-target network and protein-protein interaction (PPI) network. The results showed that SMDOCH affected CHD mainly through 10 significant signaling pathways in three biological processes: ‘vascular endothelial function regulation’, ‘inflammatory response’, and ‘lipid metabolism’. Six pathways belonged to the ‘vascular endothelial function regulation’ model, which primarily regulated hormone (renin, angiotensin, oestrogen) activity, and included three key upstream pathways that influence vascular endothelial function, namely KEGG:04933, KEGG:05418, and KEGG:04066. Three pathways, namely KEGG:04668, KEGG:04064, and KEGG:04620, belonged to the ‘inflammatory response’ model. One pathway (KEGG:04920) belonged to the ‘lipid metabolism’ model. To some extent, this study revealed the potential bioactive components and pharmacological mechanisms of SMDOCH on CHD, and provided a new direction for the development of new drugs for the treatment of CHD.

TCM formulas exert their effect on multiple biological processes to treat diseases via its diverse bioactive components, which act on multiple targets. However, most studies still use the term 'single-component' , 'single-target' , or 'single-pathway' to investigate the mechanism of TCM. With the development of bioinformatics, network pharmacology has become a highly effective method for studying TCM, as it can reveal the relationship between the bioactive components of a TCM and their potential mechanism of action in a systematic and comprehensive manner 4 .
To comprehensively elucidate the mechanism of SMDOCH on CHD using network pharmacology, we first screened the targets of SMDOCH bioactive components and CHD-related targets. Next, we constructed a SMDOCH-CHD common-target network and a core-target protein-protein interaction (PPI) network, and then performed a cluster analysis of the core-target PPI network. Finally, we screened the key targets and signaling pathways by GO and KEGG pathway enrichment analysis to further explore the protective mechanism of SMDOCH on CHD.

SMDOCH component-target network.
A total of 298 components of SMDOCH were collected from two natural product databases: TCMSP and TCM@Taiwan, 202 of which were from S. miltiorrhiza (SM), 98 from D. odorifera (DO), and 2 (palmitic acid, alpha-Farnesene) from SM and DO. The components were screened with the criteria of OB ≥ 30% and DL ≥ 0.18. In total, 104 bioactive components of SMDOCH were included, 62.5% (65/104) and 37.5% (39/104) of which were from SM and DO, respectively (Fig. 1A,B). Subsequently, we obtained the structural information of the bioactive components, including molecular structures, canonical smiles, and their 'sdf ' files from the product databases of PubChem and ZINC.
Next, we used a similarity-based method to identify the targets of the 104 bioactive components with the public webservers of Swiss Target Prediction and STITCH. Only 74 of 104 bioactive compounds had effective targets, 41 of which were from SM and the remaining from DO. From the 74 bioactive components, 270 potential targets were explored after removing duplicates. We subsequently constructed a visual SMDOCH component-target network containing 344 nodes and 691 edges by using Cytoscape (Fig. 1C). The number of these targets in SM and DO was 216 and 118 respectively, and 64 targets overlapped in the two herbs, which indicated that SM and DO showed the propensity to interact with each other by acting on the same or similar targets. We then conducted further studies to explore the interaction between SM and DO.
common-target network. The occurrence and development of CHD involves the co-regulation of multiple genes. Investigation of gene and gene-environment interaction is beneficial to elucidate the pathogenesis of CHD. In this study, 375 CHD-related targets were collected from human genomic databases. The number of these targets in OMIM, TTD, NCBI Gene, PharmGkb, Drugbank, CTD, and GeneCards was 248, 30, 73, 27, 31, 48, and 17, respectively. In addition, 58 targets were common to SMDOCH and CHD ( Fig. 2A,B), and these targets were related to 20 and 24 bioactive components from SM and DO, respectively. We used degree, a network pharmacological parameter based on topological analysis method that reflects the importance of nodes through the number of connections to other nodes, to screen 18 candidate components with a criteria of degree > 2 in the common-target network. Some of the candidate components, such as tanshinone IIA from SM and formononetin and butin from DO, were highly abundant in SMDOCH and their pharmacological activities will be investigated in future studies.
CHD development is closely related to several biological processes, including vascular tone/nitric oxide signaling, cellular proliferation/vascular remodelling, and inflammation/cell adhesion/transendothelial migration, as well as molecules such as Low Density Lipoprotein (LDL) cholesterol and triglyceride-rich lipoproteins 5 . These pathogenesis-related biological processes and molecules mainly involve three functional modules, including vascular endothelial function regulation (vascular tone/nitric oxide signaling, cellular proliferation/vascular remodelling), inflammatory response (inflammation/cell adhesion/transendothelial migration), and lipid metabolism (LDL cholesterol and triglyceride-rich lipoproteins). To elucidate the mechanism of SMDOCH in the treatment of CHD, the 58 common targets were input into DAVID and ClueGO for GO and KEGG pathway analysis, producing 35 biological processes (BPs), 7 cell components (CCs), 13 molecular functions (MFs), and 93 KEGG pathways.
According to the aetiologies of CHD, these biological processes can be divided into three functional modules, including vascular endothelial function regulation ( Table 1).

SMDOCH-CHD PPI network.
The PPI networks of SMDOCH-and CHD-related targets were constructed by using Bisogenet (Fig. 3A,B). We selected the same nodes and edges from the two PPI networks to obtain an intersection (Fig. 3C). Subsequently, we used CytoNCA to perform a central network assessment of the intersection of the PPI networks through topological analysis, and determined the significant targets for SMDOCH on CHD PPI network by a screening criteria of 'DC ≥ 62' (Fig. 3D). Six screening criteria were used to further screen the network, and a core-target PPI network containing 297 candidate targets for SMDOCH on CHD was finally obtained (Fig. 3E).
To clarify the mechanism of the 297 candidate targets for SMDOCH on CHD, we analysed the core-target PPI network using cluster analysis by MCODE, generating eight clusters (Fig. 3F) that were subsequently analysed. The cluster analysis produced 40 input targets from SMDOCH-or CHD-related targets, 8 of which belong to the (2) Three of them were related to inflammatory response, suggesting that SMDOCH may inhibit inflammatory response by affecting KEGG:04668, KEGG:04064, and KEGG:04620. These KEGG pathways were primarily associated with the common targets TNF, IL1B, TLR4, FOS, CCL2, and PTGS2. (3) One of them was related to lipid metabolism, indicating that SMDOCH may regulate lipid metabolism by affecting KEGG:04920. This KEGG pathway was primarily associated with the common targets ADIPOQ, MTOR, PPARA, and IRS1 (Fig. 4).
A PPI network was established to elucidate the link between SMDOCH-related targets, CHD-related targets, and other human proteins in the action mechanism of SMDOCH on CHD. As an extension of the common-target networks, we discovered new biological processes, which suggested that SMDOCH may also regulate vascular endothelial function through (KEGG:04014) ras signaling pathway, (KEGG:04110) cell cycle, and (KEGG:04115) p53 signaling pathway, which were related to cell proliferation, survival, growth, migration, differentiation, or cytoskeletal dynamism. SMDOCH may also regulate inflammatory response through (KEGG:04062) chemokine signaling pathway, which can provide directional cues for the trafficking and recruitment of leukocytes to the site of inflammation upon foreign insult. In addition, SMDOCH may also regulate lipid metabolism through the (KEGG:04910) insulin signaling pathway, which maintains lipid and glucose homeostasis. From data of GO terms, we found that SMDOCH may treat CHD by regulating gene repair and expression as well as macromolecular metabolism at the gene level with multiple synergies, such as (GO:0000398)mRNA splicing via spliceosome, (GO:0010467)gene expression, (GO:0010604)positive regulation of macromolecule metabolic process, and (GO:0010605)negative regulation of macromolecule metabolic process.

Discussion
As a complex cardiovascular disease, CHD is a serious threat to the human health. At present, the commonly used chemical drugs mainly control the corresponding symptoms of CHD. TCM has characteristics of multi-component and multi-target, which can affect different biological processes to control symptoms and solve the fundamental problems. SMDOCH has been used for more than a thousand years to treat CHD 6-8 , but its unclear bioactive components and mechanisms greatly limit its clinical application. However, the relationships between bioactive components and TCM mechanisms have been explored by high-efficiency strategy of network pharmacology. Thus, in this study, we investigated the action mechanism of SMDOCH on CHD through network pharmacology using three aspects (Fig. 5).
Through GO and KEGG pathway enrichment analysis of common-target and PPI networks, 10 significant signaling pathways were screened as key action mechanisms of SMDOCH on CHD, and these pathways can be divided into three functional modules: 'vascular endothelial function regulation' , 'inflammatory response' , and 'lipid metabolism' . Next, we used the KEGG Mapper to analyse the connections between upstream and downstream genes in the key signaling pathways.
The 'Vascular endothelial function regulation' model showed that SMDOCH may influence vascular endothelial function by regulating hormone (renin, angiotensin, oestrogen) activity and three key upstream pathways: KEGG:04933, KEGG:05418, and KEGG:04066. CHD is a heart disease caused by coronary artery atherosclerosis or organic stenosis and obstruction, resulting in myocardial ischemia, hypoxia, or necrosis 9 . Coronary artery stenosis disturbs blood flow with associated reciprocation, resulting in low-shear stress in the artery 10 . Myocardial ischemia-reperfusion will also lead to production of advanced glycation end products (AGEs) 11 . Through the www.nature.com/scientificreports www.nature.com/scientificreports/ KEGG:04933 and KEGG:05418 pathways, AGEs and low-shear stress activate intracellular signaling pathways that regulate five functions: firstly, AGEs and low-shear stress mediate the activation of PI3K-Akt and MAPK [12][13][14] , increasing the expression of activator protein 1 (AP-1) and NF-κB 15 , which activate pro-inflammatory cytokines, such as IL1B and TNF 16 , and various atherosclerosis-related genes, including CCL2 17 , MMP2, MMP9, VEGFA 18 , and END1 19 . Second, AGEs mediate the activation of the JAK-STAT pathway, leading to a cardioprotective response against ischemia, modulation of the cardiac microenvironment, and communication with cardiac fibroblasts 20 . Third, AGEs and low-shear stress mediate the activation of KEGG:04210, which eliminates damaged or redundant cells by activating CASP3 and CASP7, which are activated by the TNF and P53 signaling pathways 21 . Fourth, AGEs mediate the activation of AGT, which is closely related to extracellular fluid volume and blood pressure balance in the body 22 . Fifth, low-shear stress mediate the activation of KEGG:04370, which results in the expression of genes related to endothelial cell proliferation and migration, thereby promoting endothelial cell survival and improving vascular permeability 23 .
The 'Lipid metabolism' model showed that SMDOCH may influence lipid metabolism by regulating KEGG:04920. Abnormal glucose and lipid metabolisms are the main risk factors of CHD 31 . Adipocytokines, such as ADIPOQ and TNF, play an important role in regulating energy balance, metabolic homeostasis, and inflammation 32 . ADIPOQ not only increases the volume and number of adipocytes but also decreases plasma glucose and free fatty acids (FFAs) 33 . Through the KEGG:04920 pathway, ADIPOQ mediates the activation of the AMPK signaling pathway (KEGG:04152) and PPAR signaling pathway (KEGG:03320), which in turn regulates energy-consuming pathways, such as protein, glycogen, and fatty acid synthesis, as well as energy-producing pathways, such as glycolysis and fatty acid oxidation 34,35 . Some studies revealed that the proinflammatory factor TNF inhibits tyrosine phosphorylation of IRS1 by promoting the binding of IRS1 to serine, leading to insulin resistance 36 .
SMDOCH may regulate these key signaling pathways and their genes directly to treat CDH. Moreover, SMDOCH may indirectly treat CHD through signaling pathways that do not overlap with the common-target network, such as KEGG:04014, KEGG:04110, KEGG:04115, KEGG:04062, and KEGG:04910. Further research is needed to verify or modify this finding.
However, there were some shortcomings in this study. First, the reliability of our investigation of potential targets of bioactive compounds using a similarity-based method was limited by the quality of the existing databases, and the target prediction methods cover only a few hundreds to thousands of targets that may introduce biases to the enrichment analysis. Second, network pharmacology is a virtual screening of potential bioactive components, targets, and pathways in vitro, which cannot reflect the real dynamic situation of drugs in the body. Third, the results can only be taken as hypotheses rather than solid conclusions proved by experiments. In the future, we strive to elucidate the potential bioactive components and pharmacological action mechanisms of SMDOCH on CHD in vivo.
In conclusion, the cardioprotective effect of the bioactive components of SMDOCH, such as tanshinone IIA from SM as well as formononetin and butin from DO, can be explained, at least in part, by the biological processes 'vascular endothelial function regulation' , 'inflammatory response' , and 'lipid metabolism' through the core targets TP53, ESR1, AKT1, STAT3, and MAPK1. In addition, 10 significant signaling pathways were screened to clarify the protective mechanism of SMDOCH on CHD. Six pathways belonged to the 'vascular endothelial function regulation' model, which primarily regulated hormone (renin, angiotensin, oestrogen) activity, and included three key upstream pathways that influence vascular endothelial function, namely KEGG:04933, KEGG:05418, and KEGG:04066. Three pathways, namely KEGG:04668, KEGG:04064, and KEGG:04620, belonged to the 'inflammatory response' model. One pathway (KEGG:04920) belonged to the 'lipid metabolism' model. To some extent, this study revealed the potential bioactive components and pharmacological mechanisms of SMDOCH on CHD, and provided a new direction for the development of new drugs for the treatment of CHD.
Screening of bioactive components. In clinical treatment, a TCM often used by oral administration. Oral bioavailability (OB) 42 and drug-likeness (DL) 43 , two ADME-related models, are the main variables affecting the absorption of drugs by the gastrointestinal tract. Therefore, we screened bioactive components under the conditions of OB ≥ 30% and DL ≥ 0.18 44 (Supplementary Table S2). www.nature.com/scientificreports www.nature.com/scientificreports/ Target prediction of bioactive components. At present, many methods are available for target prediction, and these methods can be divided into four categories based on the prediction principles: 1) molecular docking-based methods, which are based on the three-dimensional (3D) structures of targets; 2) pharmacophore-based methods, which are based on structure and ligand pharmacophore mapping; 3) machine learning-based methods, which are based on databases; and 4) similarity-based methods, which are based on a prediction principle that similar drugs act on similar targets 45 . In this study, we selected two similarity-based webservers, Swiss Target Prediction (http://www.swisstargetprediction.ch/) 46 and STITCH (http://stitch.embl.de/) 47 , to predict the target of TCM components. Considering that the chemical-protein associations integrated in these databases are from pathway and experimental databases as well as from the literature, each proposed interaction can be traced back to the original data sources.
First, we obtained information on the structure of the bioactive components, including molecular structures, canonical smiles, and their 'sdf ' files from the product databases of PubChem (https://pubchem.ncbi.nlm.nih. enrichment analysis. We used the Database for Annotation Visualization and Integrated Discovery (DAVID, https://david.nicifcrf.gov/, version 6.8) for gene ontology (GO) enrichment analysis 65,66 with the screening criteria of P ≤ 0.05 using the Bonferroni correction 67 . The Bonferroni correction compensates for multiple comparisons by dividing the level of significance by the number of comparisons 68 . To uncover functionally grouped gene pathway annotation networks, we used ClueGO, a plug-in of Cytoscape, to apply Kyoto Encyclopaedia of Genes and Genomes (KEGG) pathways enrichment analysis with P ≤ 0.05 and Kappa Score ≥0.4 as screening criteria 69 . Moreover, we used the KEGG Mapper (https://www.genome.jp/kegg/mapper.html), a collection of tools for 'Search Pathway' and 'Color Pathway' , to analyse the connections between upstream and downstream genes in key signaling pathways [70][71][72] . Next, we performed enrichment analysis for common-target networks and PPI networks to elucidate the action mechanism of SMDOCH in the treatment of CHD (Supplementary Tables S8-10).