Synergistic mechanisms of Sanghuang–Danshen phytochemicals on postprandial vascular dysfunction in healthy subjects: A network biology approach based on a clinical trial

With the increased risk of cardiovascular disease, the use of botanicals for vascular endothelial dysfunction has intensified. Here, we explored the synergistic mechanisms of Sanghuang–Danshen (SD) phytochemicals on the homeostatic protection against high-fat-induced vascular dysfunction in healthy subjects, using a network biology approach, based on a randomised crossover clinical trial. Seventeen differential markers identified in blood samples taken at 0, 3 and 6 h post-treatment, together with 12SD phytochemicals, were mapped onto the network platform, termed the context-oriented directed associations. The resulting vascular sub-networks illustrated associations between 10 phytochemicals with 32 targets implicated in 143 metabolic/signalling pathways. The three key events included adhesion molecule production (ellagic acid, fumaric acid and cryptotanshinone; VCAM-1, ICAM-1 and PLA2G2A; fatty acid metabolism), platelet activation (ellagic acid, protocatechuic acid and tanshinone IIA; VEGFA, APAF1 and ATF3; mTOR, p53, Rap1 and VEGF signalling pathways) and endothelial inflammation (all phytochemicals, except cryptotanshinone; 29 targets, including TP53 and CASP3; MAPK and PI3K-Akt signalling pathways, among others). Our collective findings demonstrate a potential of SD to protect unintended risks of vascular dysfunction in healthy subjects, providing a deeper understanding of the complicated synergistic mechanisms of signature phytochemicals in SD.

Identification of differential markers in the clinical setting. The effects of SD on biochemical markers in blood over 6 h are shown in Fig. 2A. A high-fat intake yielded significant variations in the triglyceride (TG) and insulin levels in plasma, and the collagen/epinephrine (Col/Epi)-induced closure time (CT) in whole blood. SD consumption suppressed the extent of the TG fluctuations, resulting in a decrease in the areas under the  Figure 1. CONSORT flow diagram of the study, including enrolment of the subjects through to data analysis, as well as the primary reasons for exclusion. All subjects who completed the study were analysed.
www.nature.com/scientificreports www.nature.com/scientificreports/ curves (AUCs) (P < 0.0001). A similar tendency was shown in the insulin level, but this did not reach statistical significance because of the large variations within the groups. Consumption of the high-dose SD also effectively suppressed high-fat-induced platelet defects, as evidenced by the increase in the AUCs of the Col/Epi-induced CT (P = 0.0207).
Finally, the metabolic profiles were determined in the placebo and high-dose groups at 6-h, using gas chromatography-time-of-flight-mass spectrometry (GC-TOF-MS). Both principal component analysis (PCA) and orthogonal partial least squares-discriminant analysis (OPLS-DA) exhibited a significant separation of the clusters between the two groups, suggesting a considerable modification of plasma metabolites by SD administration (Fig. 2C). The heat map demonstrated the top 13 metabolites were significantly different between the two groups (Fig. 2D). The 7 metabolites (oleamide, cholesterol, oleonitrile, stearic acid, pyrophosphate, tryptophan and proline) were significantly increased, and the 6 metabolites (aspartic acid, 9,12-octadecadienoic acid, glucose, glycine, arachidonic acid and 5-oxoproline) were significantly decreased in the SD group compared with the placebo group (Supplementary Table S1).
Construction of the target-specific metabolic/signalling pathways associated with the vascular endothelial function. To explore the underlying mechanisms, we further analysed the metabolic/signalling pathways related to vascular endothelial dysfunction, using the CODA network platform. A total of 143 metabolic/signalling pathways, involving 32 targets and 10 SD phytochemicals were identified (Fig. 4). The functional annotation and pathway enrichment analysis revealed that α-linolenic acid metabolism was implicated in adhesion molecule production with PLA2G2A while arachidonic acid metabolism and linoleic acid metabolism were www.nature.com/scientificreports www.nature.com/scientificreports/ involved either in adhesion molecule production with PLA2G2A or in endothelial inflammation with CYP2E1. The mTOR, p53, PI3K-Akt, Rap1 and VEGF signalling pathways were closely related either to platelet activation with VEGFA/APAF1 or endothelial inflammation with AKT1/BCL2/CASP3/NFKB1/RELA/TP53. Of the metabolic/signalling pathways speculated to be involved in endothelial inflammation, the MAPK signalling pathway had the most number of target proteins (AKT1/CASP3/CD14/JUN/NFkB1/NFkB2/RELA/TGFBR1/TGFBR2/  Table S4).

Discussion
To the best of our knowledge, this is the first attempt to apply emerging analytical technologies and computational network biology to the traditional clinical setting, to overcome the current limitations in understanding how botanical phytochemicals exert homeostatic control against unintended damages confronted in daily life. As a result, we were able to provide a detailed description of the synergistic actions of SD phytochemicals against postprandial-lipaemia-induced vascular endothelial dysfunction in healthy adults.
The first hurdle facing botanical research in the traditional clinical setting of healthy adults is that subtle and early changes cannot be captured by testing the statistical significance of individual biomarkers between groups. Therefore, we employed a high-fat challenge model, to magnify the responses to botanical intervention. This strategy was based on a growing body of evidence suggesting that an extended postprandial state in daily life may result in a temporary and reversible perturbation of platelet hyperactivity in the bloodstream, along with hyperlipaemia, hyperglycaemia and hyperinsulinaemia [13][14][15][16][17] . We then adopted the analysis of metabolites and gene expressions, for understanding the physiological processes potentially affected by SD consumption, using a subtle and holistic approach [18][19][20] . Owing to the inherent sensitivity of metabolomics and gene expression, we could detect subtle alterations in biological pathways that might be useful for mining the underlying mechanisms in the www.nature.com/scientificreports www.nature.com/scientificreports/ following network analysis 21 . For the gene expression experiment, we decided to use PBMCs because over 80% of the gene expression in PBMCs is shared with most genes expressed in different human tissues, thereby serving as genetic "footprints" in blood 22 . Our recent animal study 7 also confirmed that gene expression in PBMCs was remarkably compatible with that in the aorta, demonstrating that it can be applied to understand the events in the vascular endothelium.
The next hurdle encountered in a traditional clinical trial is that it is not designed to explain the synergistic actions of botanical phytochemicals. To date, preclinical studies have been carried out to investigate the synergistic effect of phytochemicals using more than two compounds individually and in combination 23 . More recently, high-throughput screening has been used in the identification of potential targets, bioactive components in botanicals and their synergistic interactions. However, these approaches are rarely successful in screening all possible cases, due to component diversity and target complexity 11 . The computational network approach is now available, providing increased opportunities to understand complex interactions between multiple phytochemicals in botanicals and multi-targets in the human body 24 . In our study, we constructed a vascular subnetwork to expand our knowledge on the synergistic effects of SD phytochemicals and fundamental biological mechanisms, by mapping the observed impact and omics data obtained from a comprehensive clinical trial to the CODA network platform.
The vascular subnetwork constructed in this study captured platelet activation, adhesion molecule production and endothelial inflammation, as key biological events related to the effects of SD against postprandial lipaemia-induced vascular dysfunction. The network demonstrated that fumaric acid in Sanghuang, together with cryptotanshinone in Danshen, were directly connected with VCAM-1 and ICAM-1, related to adhesion molecule production 25 . Meanwhile, ellagic acid (in Sanghuang) was also involved in adhesion molecule production, by regulating PLA2G2A, implicated in arachidonic acid, α-linolenic acid and linoleic acid metabolism. PLA2G2A, a member of the phospholipase A2 family, hydrolyses membrane phospholipids to fatty acids, which activate NF-kB activation and, ultimately, ICAM-1 expression 26,27 . In this way, fumaric acid, cryptotanshinone and ellagic acid would exert a synergistic influence on vascular health, by regulating adhesion molecule production, using common or separate mechanisms.
The protocatechuic acid in Sanghuang was recognised to bind to VEGFA, implicated in the mTOR, PI3K-Akt, Rap1 and VEGF signalling pathways. VEGFA is released during platelet activation, as a growth factor involved in microvascular development via its phosphorylation of downstream targets of Akt and mTOR in endothelial cells 28 . Also, VEGFA stimulation induces activation of Rap1, which regulates endothelial cell growth, migration, proliferation and tubule formation, by triggering Akt-eNOS signalling 29 . Concurrently, ellagic acid (in Sanghuang) and tanshinone IIA (in Danshen) were linked to APAF1, implicated in the p53 signalling pathway. www.nature.com/scientificreports www.nature.com/scientificreports/ In human platelets, APAF1, cytochrome c, and caspases-3 and -9 cooperate as an essential element of the mitochondrial death pathway 30,31 .
All phytochemicals, except cryptotanshinone in Danshen, were linked to 29 different target proteins related to endothelial inflammation. Among them, ellagic acid, in Sanghuang, was highly integrated with many different signalling pathways in endothelial inflammation, and thereby presented the broadest impact. This finding is compatible with previous studies that report protective effects of ellagic acid against oxidant-induced endothelial inflammation and atherosclerosis 32,33 . Conversely, the target interacting the greatest number of phytochemicals (ellagic acid, fumaric acid and protocatechuic acid in Sanghuang, and danshensu, salvianolic acid A, salvianolic acid B and tanshinone IIA in Danshen) was TP53, which is a transcriptional factor inducing cell cycle arrest, apoptosis or changes in metabolism in response to cellular stress 34 . Activation of TP53 is speculated to be influenced by several mechanisms, including MAPK, PI3K-Akt, sphingolipid, thyroid hormone, p53 and Wnt signalling pathways, suggesting its role as a hub. The next rank was CASP3. Ellagic acid, caffeic acid and protocatechuic acid in Sanghuang, and salvianolic acid A, salvianolic acid B and tanshinone IIA in Danshen, were connected to CASP3, mediated by the MAPK, TNF and p53 signalling pathways. CASP3 is involved in the sequential activation of caspases responsible for the execution of cell apoptosis 35 .
In summary, we demonstrated that the strategy of applying a high-fat challenge with metabolite and gene analyses to a traditional crossover clinical trial allowed detecting subtle and early effects of SD on maintaining vascular homeostasis in healthy subjects. Also, subsequent mapping of the outcomes observed in a clinical trial onto an in silico network model (CODA), enabled a deep understanding of the complicated synergistic mechanisms of SD phytochemicals. However, it is worth mentioning the limitations of our study. This study describes the acute outcomes after a single administration of SD. The results of another clinical trial has been recently published by our research group to report the vascular endothelial effects of 4-week SD consumption at 900 mg/ day in healthy chronic smokers 36 . In addition, the contribution of this study is limited to provide a qualitative description of the relationship between the phytochemicals and targets. An extensive statistical analysis is currently under way to quantify. Even with these limitations, our approach is a novel and useful tool to overcome the inherent limitations of traditional clinical trials, for evaluating the effectiveness of botanicals in healthy subjects, although further refinement is necessary. This in silico model may also be used reversely, to find botanicals containing bioactive phytochemicals. www.nature.com/scientificreports www.nature.com/scientificreports/ subjects and Methods Study products. The SD and a colour-matched placebo were provided by Pulmuone Foods Co., Ltd. (Seoul, Korea). Details on the preparation of SD are described in our previous publication 7 . The following 12 signature phytochemicals were quantified using a high-performance liquid chromatograph coupled with a diode array detector and MS/MS: caffeic acid, ellagic acid, fumaric acid, hispidin and protocatechuic acid from Sanghuang, and cryptotanshinone, danshensu, rosmarinic acid, salvianolic acid A, salvianolic acid B, tanshinone I and tanshinone IIA from Danshen ( Supplementary Fig. S1) 7 . For standardisation purposes, protocatechuic acid (0.45-1.5 μg/g) and tanshinone IIA (50 μg/g) were chosen as marker components. The gelatine capsules were used for a double-blind challenge.
Subjects and study design. Based on a previous study 37 , the estimated sample size of 56 subjects per group provides a statistical power of 80% to detect a difference in platelet aggregation, using a two-sided significance level of α = 0.05 and assuming a 20% attrition rate. The study followed a randomised, double-blinded, placebo-controlled crossover design. Subjects were recruited through poster advertisements. Eligible subjects were apparently healthy adults aged between 30 and 65 years old. Subjects were excluded if any of the following criteria were present: (1) body mass index >35 kg/m 2 ; (2) a history of body weight change ≥10% in the previous 8 weeks; (3) exercising >10 h/week; (4) cigarette smoking >1 pack/day; (5) alcohol consumption >140 g/week; (6) use of medication or dietary supplements in the previous 4 weeks; (7) a history of platelet dysfunction, hypertension, stroke, diabetes and thyroid disease; (8) a history of hypersensitivity in test material; and (9) pregnancy or breastfeeding. Written informed consents were received from all participants before their participation in this study.
Fifty-six eligible subjects were enrolled in the trial and randomly assigned to receive one of the test samples in four sequences (placebo, 300, 600 or 900 mg SD), using computer-generated block randomisation (block size of four), performed by an independent researcher. Each session began with the collection of venous blood at fasting (t = 0) and 3 and 6 h after consuming each a high-fat-loaded test sample (total 900 kcal; 58.9% energy from fat, 33.3% energy from carbohydrate, and 7.6% energy from protein). Treatment visits were scheduled 1 week apart. During the entire trial, the subjects were advised to maintain their regular diet and lifestyle, which were monitored through a mobile phone app. The study was conducted according to the Declaration of Helsinki and approved by the Institutional Review Board of Ewha Womans University Medical Centre (EUMC 2014-04-012-014) and Ewha Womans University (79-1). The study was also registered in the World Health Organisation (WHO) International Clinical Trials Registry Platform, under the following identification: KCT0001193 (05/08/2014).
Biochemical analysis of plasma. Plasma TG was measured by an automatic analyser (Hitachi 7600, Hitachi Co., Tokyo, Japan). Plasma insulin was determined by a human insulin enzyme-linked immunosorbent assay kit (Abcam, Cambridge, UK). CT was measured in whole blood collected in 3.8% trisodium citrate, by using a Col/Epi cartridge in a PFA-100 instrument (Siemens Healthcare Diagnostics, Marburg, Germany).

Target gene analysis in PBMCs.
Total RNA was extracted from PBMCs using TRIzol (Invitrogen, San Diego, CA, USA). The concentration and quality of the RNA were measured using a BioSpec-nano (Shimadzu Corp., Tokyo, Japan), and total RNA was reverse-transcribed using a high-capacity cDNA reverse transcription kit (Applied Biosystems, Foster City, CA, USA). The TaqMan method was used to quantify the expression of COX-1 (Ptgs1; Hs00377726_m1), COX-2 (Ptgs2; Hs00153133_m1), ICAM-1 (Icam; Hs00164932_m1), VCAM-1 (Vcam; Hs01003372_m1) and β-actin (Actb; Hs01060665_g1). The relative amounts of these mRNAs were normalised to the amount of β-actin, and the relative amounts of the RNAs were calculated using the comparative C T method.
Metabolomic analysis in plasma. Each plasma sample of the placebo and high-dose SD groups was treated with ice-cold MetOH (containing 2-chloro-l-phenylalanine as an internal standard), sonicated, left at 4 °C for 1 h and then centrifuged (13,500 g/4 °C/10 min). The supernatant was filtered, dried, oximated (30 °C/90 min) with methoxyamine hydrochloride in pyridine, and trimethylsilylated with N-methyl-N-trimethylsilyl-trifluoroacetamide (37 °C/30 min). The GC-TOF-MS analysis was performed by using an Agilent 7890 gas chromatograph system (Agilent Technologies, Palo Alto, CA, USA) coupled with an Agilent 7693 auto-sampler (Agilent Technologies) and equipped with a Pegasus ® HT TOF-MS (LECO, St. Joseph, MI, USA) system. An Rtx-5MS column (30 m × 0.25 mm, 0.25 μm particle size; Restek Corp., Bellefonte, PA, USA) was used with a constant flow (1.5 mL/min) of helium as carrier gas. One microliter of the sample was injected into the GC. The oven temperature was initially maintained at 75 °C for 2 min and then ramped at 15 °C/ min to 300 °C, and held for 3 min. The temperatures of the front inlet and transfer lines were 250 and 240 °C, respectively. The electron ionisation was carried out at −70 eV and full-scan data were acquired over a range of 50-1000 m/z. Statistical analysis. The postprandial time-series data for each subject were normalised to the respective t = 0 baseline value before the mean values were computed. The response of each marker was quantified as the AUC, using the trapezoidal method 38 . Data were analysed by one-way ANOVA for repeated measures, followed by Tuckey's multiple comparison test. Data were analysed using the SAS version 9.4 (SAS Institute, Cary, NC, USA), and a significance was assumed at a two-tailed P < 0.05. The quantitative GC-TOF-MS data were subjected to multivariate statistical analysis using SIMCA-P ver. 14.1 (Umetrics, Umea, Sweden). The overall effect was visualised by the PCA and OPLS-DA, followed by variable importance in the projection analysis, to identify significant and important variables. A heat map was generated based on differential metabolites, using MetaboAnalyst 3.0 (http://metaboanalyst.ca).