Chromatogram-Bioactivity Correlation-Based Discovery and Identification of Three Bioactive Compounds Affecting Endothelial Function in Ginkgo Biloba Extract

Discovery and identification of three bioactive compounds affecting endothelial function in Ginkgo biloba Extract (GBE) based on chromatogram-bioactivity correlation analysis. Three portions were separated from GBE via D101 macroporous resin and then re-combined to prepare nine GBE samples. 21 compounds in GBE samples were identified through UFLC-DAD-Q-TOF-MS/MS. Correlation analysis between compounds differences and endothelin-1 (ET-1) in vivo in nine GBE samples was conducted. The analysis results indicated that three bioactive compounds had close relevance to ET-1: Kaempferol-3-O-α-l-glucoside, 3-O-{2-O-{6-O-[P-OH-trans-cinnamoyl]-β-d-glucosyl}-α-rhamnosyl} Quercetin isomers, and 3-O-{2-O-{6-O-[P-OH-trans-cinnamoyl]-β-d-glucosyl}-α-rhamnosyl} Kaempferide. The discovery of bioactive compounds could provide references for the quality control and novel pharmaceuticals development of GRE. The present work proposes a feasible chromatogram-bioactivity correlation based approach to discover the compounds and define their bioactivities for the complex multi-component systems.


Introduction
Ginkgo biloba Extract (GBE), extracted from Ginkgo biloba leaves, is mainly composed of terpene trilactones, flavonoid heterosides, ginkgolic acids, phenolic acids, proanthocyanidins, etc. [1,2]. GBE can significantly decrease serum ET-1 to reverse endothelial dysfunction [3][4][5]. Nowadays, chromatographic fingerprint plays a vital role in the quality control of GBE, including for authenticity determination and chemical information analyses. However, existing GBE studies with fingerprint tech mainly focus on the chemical characteristics, but do not elaborate the correlation between compounds and their bioactive effects. Based on the hypothesis that bioactive effects varied with differences between compounds, chromatographic fingerprint and bioactive tests of nine re-combined GBE samples were conducted, and their correlations were further analyzed ( Figure 1). Other than the

GBE HPLC Fingerprint and Identification of Components
With optimized HPLC conditions, the standard GBE HPLC fingerprint ( Figure 2) was established, and 21 compounds were identified or characterized through the HPLC-DAD-ELSD-MS/MS technique in our previous work [6] (Table 1). According to the retention time, UV spectra, and MS spectra of the reference standards, Protocatechuic acid (P4), Rutin (P12), Ginkgolide A (P24), Ginkgolide B (P25), and Bilobalide (P26) were identified unambiguously. The other compounds were characterized according to MS fragmentation pattern, UV spectra, and the reported literature.

GBE HPLC Fingerprint and Identification of Components
With optimized HPLC conditions, the standard GBE HPLC fingerprint ( Figure 2) was established, and 21 compounds were identified or characterized through the HPLC-DAD-ELSD-MS/MS technique in our previous work [6] (Table 1). According to the retention time, UV spectra, and MS spectra of the reference standards, Protocatechuic acid (P 4 ), Rutin (P 12 ), Ginkgolide A (P 24 ), Ginkgolide B (P 25 ), and Bilobalide (P 26 ) were identified unambiguously. The other compounds were characterized according to MS fragmentation pattern, UV spectra, and the reported literature.

Three Portions Separated from GBE and Nine Re-Combined GBE Samples
Portion A, portion B, and portion C were separated from GBE via D101 macroporous resin. They were re-combined with different compositions to get the nine GBE samples ( Figure 3). In accordance with the optimized HPLC conditions, the HPLC fingerprints of the nine GBE samples (S1-S9) were constructed ( Figure 4). 26 peak areas in nine GBE samples are shown in Table 2.

Three Portions Separated from GBE and Nine Re-Combined GBE Samples
Portion A, portion B, and portion C were separated from GBE via D101 macroporous resin. They were re-combined with different compositions to get the nine GBE samples ( Figure 3). In accordance with the optimized HPLC conditions, the HPLC fingerprints of the nine GBE samples (S 1 -S 9 ) were constructed ( Figure 4). 26 peak areas in nine GBE samples are shown in Table 2.

Three Portions Separated from GBE and Nine Re-Combined GBE Samples
Portion A, portion B, and portion C were separated from GBE via D101 macroporous resin. They were re-combined with different compositions to get the nine GBE samples ( Figure 3). In accordance with the optimized HPLC conditions, the HPLC fingerprints of the nine GBE samples (S1-S9) were constructed ( Figure 4). 26 peak areas in nine GBE samples are shown in Table 2.

Three Portions Separated from GBE and Nine Re-Combined GBE Samples
Portion A, portion B, and portion C were separated from GBE via D101 macroporous resin. They were re-combined with different compositions to get the nine GBE samples ( Figure 3). In accordance with the optimized HPLC conditions, the HPLC fingerprints of the nine GBE samples (S1-S9) were constructed ( Figure 4). 26 peak areas in nine GBE samples are shown in Table 2.

Cluster Analysis of Nine GBE Samples
Based on the data of the 26 peak areas, Cluster analysis was performed in SPSS 19.0. The clustering method was Nearest Neighbor. The distance calculation method was Euclidean Distance. The rescaled distance cluster combine was defined as 5. Nine GBE samples could be divided into seven categories ( Figure 5): S 2 and S 5 belonged to a class, S 3 and S 6 belonged to a class, and the remaining samples respectively represented a class each. Cluster analysis results indicated that the nine GBE samples had chemical differences in their compounds.

Cluster Analysis of Nine GBE Samples
Based on the data of the 26 peak areas, Cluster analysis was performed in SPSS 19.0. The clustering method was Nearest Neighbor. The distance calculation method was Euclidean Distance. The rescaled distance cluster combine was defined as 5. Nine GBE samples could be divided into seven categories ( Figure 5): S2 and S5 belonged to a class, S3 and S6 belonged to a class, and the remaining samples respectively represented a class each. Cluster analysis results indicated that the nine GBE samples had chemical differences in their compounds.

ET-1 Biotests of Nine GBE Samples
Plasma ET-1 in vivo was detected in the 11 treatment groups (Table 3). Compared with the normal group, plasma ET-1 content significantly increased in the model group. Compared with the model group, plasma ET-1 content significantly decreased in the S1, S2, S3, S4, S5, S6, S8, and S9 groups, but not for the S7 group. Biotest results indicated that nine GBE samples showed biological differences for ET-1.

ET-1 Biotests of Nine GBE Samples
Plasma ET-1 in vivo was detected in the 11 treatment groups (Table 3). Compared with the normal group, plasma ET-1 content significantly increased in the model group. Compared with the model group, plasma ET-1 content significantly decreased in the S1, S2, S3, S4, S5, S6, S8, and S9 groups, but not for the S7 group. Biotest results indicated that nine GBE samples showed biological differences for ET-1.

CA between Compound Differences and Biological Differences
Dimensionless data of the peak areas of 26 compounds and their ET-1 values are shown in Table S1. The Pearson correlation coefficients (PCC) are shown in Table 4. The results indicated that P 18 , P 22 , and P 23 had a significantly positive relation with ET-1, but that P 1 , P 2 , P 3 , P 4 , and P 6 were negatively correlated to ET-1. The scores of the extracted C1 and C2 were used as the new independent variables ( Table 5). The strict regression equation between C1, C2 and ET-1 was established as follows: ET-1 = 94.68 + 0.678 × C1 + 2.626 × C2 (R = 0.801, Sig. < 0.05). In accordance with the rotated component matrix (Table S3), C1 and C2 were replaced by the 26 original independent variables (P 1 -P 26 ). Regression coefficients (RC) of P 1 -P 26 are shown in Table 6. The results were in accordance with the PC analysis, indicating that P 18 , P 20 , P 22 , P 23 , and P 24 had a highly positive relation with ET-1, but that P 1 , P 2 , P 3 , P 4 , and P 6 showed a negative correlation.

Discussion
Current research methods for natural medicine mainly fall into two directions. The first is to separate single components or the active part and then assess the biological effects in vivo or in vitro; the second is to match up the compounds and bio-effects in the whole herb using computational modelling. It is understood that separating and assessing each compound one by one is almost impossible. Numerous existing studies of GBE focus on the chemical identification and the biological effects, separately, but not the correlation between them. ET-1 is a potent vasoconstrictor peptide released from endothelial cells [7]. Several studies have demonstrated that exposure to cold is associated with raised plasma ET-1 [8,9]. Thus, a rat model combined with subcutaneous injection of adrenaline and ice-bath was established, and similar data was observed in the present study.
GBE's main bioactive constituents include flavonoid glycosides and terpene trilactones. Flavonoid glycosides were detected by HPLC-UV [10][11][12]. Terpene trilactones were detected by Evaporative Light Scattering Detector (ELSD) due to their poor UV absorption property. Thus, GBE's chromatographic fingerprint was established by HPLC-UV-ELSD, in which 21 compounds were identified or characterized through the UFLC-DAD-Q-TOF-MS/MS technique. To prepare appropriate GBE samples with varying compounds, three portions were separated from GBE using D101 macroporous resin, and then re-combined to get nine GBE samples. The different ratios of the three portions were designed using a four-factor, nine-level Uniform Design (UD) method, which has been successfully applied to prepare different Chinese medicine samples [13,14]. To guarantee the differences of the GBE samples, cluster analysis was conducted that nine GBE samples could be divided into seven categories.
Correlation analysis was applied to discover and predict the compounds with bioactivities in our previous work [15,16]. The discovery of bioactive compounds was based on the hypothesis that the effect varies based on differences in the compounds. If a compound varies a little, while showing a big difference in the effect, the compound will be considered to have a close relevance; in the opposite case, the compound will be considered to have no effect contribution. In the cluster analysis, although S2 and S5, S3 and S6 belonged to a class, there were still relatively large differences among the discovered bioactive compounds, and this might be the reason behind the differences in effect among them. In this work, the Pearson Correlation and Multiple Linear Regression methods were used to evaluate the effect contribution of each compound, and the analysis results of the two methods were highly consistent. The connections between the identified compounds and ET-1 are presented dynamically in the electronic supplementary material (Compound-effect bubble chart). Kaempferol-3-O-α-L-glucoside (P 18 Figure 6). Numerous preclinical studies provide support for flavonoids exhibiting protective effects on endothelial dysfunction [17]. Quercetin, modified from quercetin flavonoid during metabolism, inhibits the overproduction and gene expression of ET-1 in vitro [18,19]. Kaempferol can improve the endothelial damage [20], but there is no direct evidence for either Kaempferol and Kaempferide on regulating ET-1. In GBE, not all the flavonoid glycosides have strong inhibitory activity on ET-1 release. As for terpene trilactones in GBE, Ginkgolide A and Ginkgolide B had a highly positive correlation, which also contributed to the effects. Moreover, P 1 , P 2 , P 3 , P 4 , and P 6 from portion A were negatively correlated with ET-1. Despite having no statistical meaning, the results suggested that water-soluble constituents might induce endothelial dysfunction, but this needs further experiments to confirm.

Animals and Materials
Sprague-Dawley male rats, Specific pathogen-free, 250-300 g, were purchased from Guangdong Medical Laboratory Animal Center (SCXK-(Yue) 2013-0002). Rats were fed on standard laboratory diet and water and kept in environmentally controlled quarters with temperature maintained at 25 °C and a 12 h dark-light cycle for a week before use. Experiments were approved by the Animal Care and Use Committee of Sun Yat-sen University (2015062529)

Preparation of GBE Samples
GBE (315 mg) was separated into three portions via D101 macroporous resin (20 g), with the eluent of 550 mL purified water (Portion A); 100 mL ethanol (40%, v/v, Portion B), and 100 mL absolute ethyl alcohol (Portion C). Each portion was evaporated with a rotary evaporator and dissolved in 1,2-propanediol (25%, g/mL) to 30 mL for HPLC analysis. According to a four-factor, nine-level UD (Table 7), three portions were re-combined to get nine GBE samples. GBE Samples were stored at 4 °C before use. Table 7. Volumes and percentage of three portions in nine GBE samples. Note: a % represents the nine levels (0, 25%, 50%, 75%, 100%, 125%, 150%, 175%, 200%) of each portion A, B, C, and the sequence was designed according to a four-factor, nine-level UD method.

Animals and Materials
Sprague-Dawley male rats, Specific pathogen-free, 250-300 g, were purchased from Guangdong Medical Laboratory Animal Center (SCXK-(Yue) 2013-0002). Rats were fed on standard laboratory diet and water and kept in environmentally controlled quarters with temperature maintained at 25 • C and a 12 h dark-light cycle for a week before use. Experiments were approved by the Animal Care and Use Committee of Sun Yat-sen University (2015062529)

Modelling and ET-1 Assay
Rats were randomly divided into eleven groups of normal (normal saline: NS, 7.2 mL/kg, n = 10) as blank, model (NS, 7.2 mL/kg, n = 10) as negative control, and nine GBE samples (7.2 mL/kg, n = 10), receiving intraperitoneal injection once daily for 7 consecutive days. After the 7th administration, the rats-except those in normal group-were subcutaneously injected with Adr (0.8 mg/kg). After 2 h, rats were kept in ice-water (0-2 • C) for 4 min, and 2 h later were subcutaneously re-injected with Adr (0.8 mg/kg). All the rats were fasted for 12 h. Blood was collected through abdominal aortic. Plasma ET-1 was detected by Elisa kit.

Correlation Analysis between Compound Difference and Bioactivity Difference
Pearson Correlation. 26 Peak areas were regarded as independent variables (P 1 -P 26 ). Average ET-1 value was regarded as a dependent variable. Every value of the peak areas and ET-1 in Table 2 was divided by the average of each column to get dimensionless data (Table S1). Pearson Correlation was used to analyze the correlation among P 1 -P 26 and ET-1. Multiple Linear Regression. 26 independent variables (P 1 -P 26 ) were recombined into two mutual independent principal components, which were regarded as new independent variables (C1 and C2, contributing to 96.388% of the total variance, Table S2). The regression equation between two components (C1 and C2) and ET-1 parameter was constructed by a stepwise regression analysis approach. Once a strict regression equation was established (p < 0.05), C1 and C2 would be replaced by the 26 original independent variables (P 1 -P 26 ) according to the rotated component matrix (Table S3). Then, the regression coefficients of P 1 -P 26 were used to evaluate the effect contribution.

Statistical Analysis
Experimental data were presented as mean ± standard deviation and analyzed by One-Way Analysis of Variance. p-values less than 0.05 or 0.01 were considered statistically significant.

Conclusions
Kaempferol Kaempferide were discovered to have the closest relevance to ET-1, which has not been reported so far and could provide further reference for the quality control and novel pharmaceutical development of GRE. Moreover, this work proposes a feasible approach for the discovery and prediction of compounds and their bioactivities in complex systems, especially for traditional Chinese medicine. The specific process is as follows: prepare the samples by the re-combination of different parts; establish the HPLC fingerprints; evaluate the bio-effects in vivo; regard the compound differences and effect differences as mathematical variables; analyze the relevance between the variables to find key bioactive compounds.