In Silico Investigation of the Pharmacological Mechanisms of Beneficial Effects of Ginkgo biloba L. on Alzheimer’s Disease

Based on compelling experimental and clinical evidence, Ginkgo biloba L. exerts a beneficial effect in ameliorating mild to moderate dementia in patients with Alzheimer’s disease (AD) and other neurological disorders, although the pharmacological mechanisms remain unknown. In the present study, compounds, their putative target proteins identified using an inverse docking approach, and clinically tested AD-related target proteins were systematically integrated together with applicable bioinformatics methods in silico. The results suggested that the beneficial effects of G. biloba on AD may be contributed by the regulation of hormone sensitivity, improvements in endocrine homeostasis, maintenance of endothelial microvascular integrity, and proteolysis of tau proteins, particularly prior to amyloid β-protein (Aβ) plaque formation. Moreover, we identified six putative protein targets that are significantly related to AD, but have not been researched or have had only preliminary studies conducted on the anti-AD effects of G. biloba. These mechanisms and protein targets are very significant for future scientific research. In addition, the existing mechanisms were also verified, such as the reduction of oxidative stress, anti-apoptotic effects, and protective effects against amyloidogenesis and Aβ aggregation. The discoveries summarized here may provide a macroscopic perspective that will improve our understanding of the molecular mechanism of medicinal plants or dietary supplements, as well as new clues for the future development of therapeutic strategies for AD.


Introduction
The number of patients with Alzheimer's disease (AD) is predicted to increase exponentially during the next few decades [1]. The current therapies for AD are based on five main strategies [2]: cholinergic treatment, antiglutamatergic treatment, vitamins and antioxidants, nonsteroidal anti-inflammatory drugs (NSAIDs), and pharmacological management of neuropsychiatric symptoms. However, single targeted therapies has often been unsuccessful, due to fact that the pathogenesis and etiology of AD have not yet been completely elucidated [3]. In contrast, Ginkgo biloba L. has long been thought to be "multivalent" [4] and it has a definite positive effect in ameliorating mild to moderate dementia in patients with AD and other neurological disorders associated with old age [5].
In the present study, the compounds were collected from The Traditional Chinese Medicine System Pharmacology Database and Analysis Platform [8] (TCMSP, http://ibts.hkbu.edu.hk/LSP/tcmsp.php). The TCMSP provides absorptions, distribution, metabolism, and excretion (ADME)-related pharmacokinetic properties, including bioavailability (OB), drug-likeness (DL), blood-brain barrier (BBB), etc. Values OB ≥ 30% and DL ≥ 0.18 were affirmed as ADME screening criteria for candidate compounds. Some studies confirmed that under pathological conditions such as AD, EGb761 is able to cross the BBB [4], but EGb761 has a limited ability to cross the BBB under normal physiological conditions. Thus, BBB penetration may be an important factor that alters the effects of EGb761 or G. biloba in vivo. Therefore, we removed the BBB cutoff value of ≥−0.3 from the ADME screening criteria.
Alzheimer's disease associated protein targets Information on AD-associated protein targets was identified from GeneCards [9] (http://www.genecards.org/) and the Comparative Toxicogenomics Database [10] (CTD, http://ctdbase.org/), which is a robust, publicly available database that provides comprehensive, user-friendly information on chemical-gene/protein interactions, and chemical-disease and gene-disease relationships. We also referred to the corresponding target protein's unique UniProtKB ID in the UniProt database (http://www.uniprot.org/), composing an AD-associated target protein database. It is noteworthy that these two databases provide an expert review ranking of protein targets based on scientific research and literature, labeled by Relevance score (GeneCard) and Inference Score (CTD). A higher score indicates a higher correlation with AD.

Inverse Docking Analysis
The 2-dimensional (2D) and 3D structures of candidate compounds were drawn using ChemBioOffice 2012 (PerkinElmer Inc., Cambridge, MA, USA). The mol2 files (mol2) of the 3D molecular structures of 25 candidate compounds were uploaded to the PharmMapper [7] (http://lilab.ecust.edu.cn/pharmmapper/) and the Human Protein Targets Only database was selected Nutrients 2018, 10, 589 3 of 17 for target prediction. Results include Protein Data Bank (PDB) database codes (PDBIDs), UniProtKB ID, target names, FitScores, and z'score. FitScores was adopted as the principal scoring to rank the proteins; among them, those with FitScores ≥ 4.5 were selected as the putative target proteins. The putative target proteins and AD-related target proteins were validated one by one, according to their unique UniProtKB ID, producing conditionally filtered results of AD-associated target proteins of candidate compounds.

Gene Ontology and KEGG Pathway Enrichment
The Database for Annotation, Visualization, and Integrated Discovery [11] (DAVID, v6.8, https://david.ncifcrf.gov/) was employed to conduct Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. The p-value was used to examine the significance of the GO/pathway term enrichment with a modified Fisher's exact test. The Benjamini value was used to globally correct the enrichment p-values of individual term members [11]. Those GO/pathway terms with a p-value of ≤0.05 and Benjamini value of ≤0.5 were regarded as significant and interesting.

Composite Network Integration
On the strength of above-mentioned target identification results, combined with Protein-Protein Interaction (PPI) data from STRING (https://string-db.org/) and pathway enrichment data from DAVID, Cytoscape 3.6.0 (Institute for Systems Biology, Seattle, WA, USA) [12] (http://www.cytoscape.org/) software was employed to construct Compound-Target (CT), Compound-Target-Disease and Compound-Group-Target-Pathway (CGTP) network models. Subsequently, engaging the NetworkAnalyzer plugin in Cytoscape, the parameters of the network topology were analyzed to get the Average Shortest Path Length (ASPL) and the Betweenness Centrality (BC), etc., and the more substantially contributing nodes were obtained. Besides this, R 3.4.3 software (R Core Team, Auckland, Tamaki-Makau-Rau, New Zealand) (https://www.r-project.org/) was employed to visualize the quantitative information. Figure 1 shows a schematic of the methodology and a summary of the results from each step. From the 307 native G. biloba compounds collected from the TCMSP database, 25 compounds were screened by ADME and prepared for further study as the candidate compounds as shown in Figure 2 and Supplementary Materials Table S1. The 25 compounds were divided into 6 categories: 12 flavonoids (quercetin, catechin, genkwanin, etc.), 5 terpene lactones (bilobalide, ginkgolide B, etc.), 3 sterols (beta-sitosterol, stigmasterol, and campest-5-en-3beta-ol), 3 fatty acid esters (mandenol, ethyl oleate, and linolenic acid ethyl ester), 1 polyprenol (flavoxanthin), and 1 lignan (sesamin). These compounds are the main components or active functional ingredients of G. biloba [4,13,14].

The Candidate Compounds and Putative Target Proteins
A total of 2500 target proteins were docked with the 25 candidate compounds. Among these target proteins, 97 were screened by ADME and named the putative target proteins. All results from the inverse docking calculation are presented in Supplementary Materials Table S2.
We compared the 97 putative target proteins for commonality and properties, and the results are shown in Figure 3. Panel (a) more intuitively shows that compounds from different categories mapped to different target proteins. Moreover, panel (b) depicts a Venn diagram that clearly shows the terpene lactone group associated with 26 exclusive protein targets-approximately one-third of the protein targets. Thus, we speculated that this finding may be consistent with the observation that terpene lactones are the predominant and unique primary bioactive substances in G. biloba. In other words, we inferred that the specificity of the inverse docking calculation was distinct. The gene entries related to AD were collected from the CTD and GeneCards databases. As a result, 21,249 and 7262 gene entries were collected from each database, respectively. The prioritized Inference Scores for the corresponding annotations are listed in descending order in Supplementary Materials Tables S3-1 and S4-1. The corresponding gene entries were converted into 108,145 and 58,432 UniProtKB IDs, respectively. The results are listed in Supplementary Materials Tables S3-2 and S4-2. This procedure was initiated by matching the unique UniProtKB IDs to determine the magnitude of the correlation between the putative target proteins and AD. We adopted the arithmetic average of the two scores from the CTD and GeneCards databases as integration scores, screening the top third of the putative target proteins for MOA (Molecular Mechanisms of Action) analysis. The one-third ratio was determined after several preliminary experiments. The results are listed in Supplementary Materials Table S5.  Tables S3-2 and S4-2. This procedure was initiated by matching the unique UniProtKB IDs to determine the magnitude of the correlation between the putative target proteins and AD. We adopted the arithmetic average of the two scores from the CTD and GeneCards databases as integration scores, screening the top third of the putative target proteins for MOA (Molecular Mechanisms of Action) analysis. The one-third ratio was determined after several preliminary experiments. The results are listed in Supplementary Materials Table S5.     Tables S3-2 and S4-2. This procedure was initiated by matching the unique UniProtKB IDs to determine the magnitude of the correlation between the putative target proteins and AD. We adopted the arithmetic average of the two scores from the CTD and GeneCards databases as integration scores, screening the top third of the putative target proteins for MOA (Molecular Mechanisms of Action) analysis. The one-third ratio was determined after several preliminary experiments. The results are listed in Supplementary Materials Table S5.

Exploration of the Molecular Mechanisms of Action
The top 30 putative target proteins were selected based on their integration score, and GO and KEGG pathway enrichment analyses were initiated. After filtering by a parameter p-value cutoff of ≤0.05, 84 GO terms and 30 KEGG pathway terms were returned, as shown in Figures

Exploration of the Molecular Mechanisms of Action
The top 30 putative target proteins were selected based on their integration score, and GO and KEGG pathway enrichment analyses were initiated. After filtering by a parameter p-value cutoff of ≤0.05, 84 GO terms and 30 KEGG pathway terms were returned, as shown in Figures      In order to show the result of the KEGG pathway enrichment in an intuitive and explicit way, a bubble diagram was employed. As shown in Figure 5, p-values are given the highest priority, in ascending order. Dual specificity mitogen-activated protein kinase kinase 1 (MAP2K1), GTPase HRas (HRAS), mitogen-activated protein kinase 14 (MAPK14), mitogen-activated protein kinase 10 (MAPK10), and proto-oncogene tyrosine-protein kinase src (SRC) were the most frequently occurring protein targets. According to the pathogenesis of AD, these KEGG pathway terms can be divided into 5 modules, as shown in Table 2.

An Integrated Network Model Analysis
The Compound-Target-Disease network model contained 123 nodes and 369 edges, as shown in Figure 6. The top 10 putative target proteins identified based on integration scores are highlighted, and the other putative target proteins are in gray, as shown in panel (b) of Figure 6. We identified 7 flavonoids, 3 terpene lactones, 3 sterols, 3 fatty acid esters, 1 polyprenol, and 1 lignan associated with these proteins. The results suggest that flavonoids and terpene lactones may primarily contribute to the anti-AD effects of G. biloba. The top 10 putative target proteins associated with AD, in turn, were
As shown in Figure 7, three subnetworks were integrated into the Compound-Group-Target-Pathway (CGTP) network, including the Protein-Protein Interaction (PPI) network, Compound-Group-Target (CGT) network, and Target-Pathway (TP) network. The PPI network is the premise and basis to obtain nodes with more substantial contributions. Nodes with a shorter ASPL and higher BC were considered as vital ones. In the PPI network, serum albumin (ALB), estrogen receptor (ESR1), and a proto-oncogene tyrosine-protein kinase (SRC) were the top three, consistent with their molecular functions of transport, connection, and signal communication. Clustering and topology approaches were utilized to identify individual variations and similarities among various protein targets. Of the top 30 putative protein targets, 3 well-organized clusters with 30 KEGG pathway terms were identified. The protein targets in cluster A were associated with 26 KEGG pathway terms, and targets in cluster C were associated with 4 KEGG pathway terms. and basis to obtain nodes with more substantial contributions. Nodes with a shorter ASPL and higher BC were considered as vital ones. In the PPI network, serum albumin (ALB), estrogen receptor (ESR1), and a proto-oncogene tyrosine-protein kinase (SRC) were the top three, consistent with their molecular functions of transport, connection, and signal communication. Clustering and topology approaches were utilized to identify individual variations and similarities among various protein targets. Of the top 30 putative protein targets, 3 well-organized clusters with 30 KEGG pathway terms were identified. The protein targets in cluster A were associated with 26 KEGG pathway terms, and targets in cluster C were associated with 4 KEGG pathway terms.

Discussion and Conclusions
Currently, an increasing number of people are consuming dietary supplements for health. Due to the amazing vitality induced by G. biloba, it has achieved therapeutic applications as a dietary supplement. Despite the fact that it has a definite positive effect in ameliorating mild to moderate dementia in patients with AD, its mechanism remains elusive. This prompted us to determine the pharmacological mechanism of G. biloba by inverse docking and system pharmacological approaches.
We collected 307 native G. biloba compounds from TCMSP, and 25 compounds were screened by ADME. Further calculations using PharmMapper identified 2500 target proteins, 97 of which were obtained by screening criteria. From the CTD and GeneCards databases, we collected 21,249 and 7262 AD-associated gene entries, respectively. The subsequent step was performed using UniProt and resulted in 108,145 and 58,432 UniprotKB IDs, respectively. Based on the integration score, the top 30 putative target proteins were selected to further explore the MOA. The functions of these putative target proteins include the following: antioxidant activity [15], protective effects on the mitochondria [16], anti-apoptotic [17], anti-inflammatory [18], protective effects on amyloidogenesis and amyloid β-protein (Aβ) aggregation [19,20], ion homeostasis [21], modulation of the phosphorylation of the tau protein [15], and induction of hormone synthesis [15]. These findings are consistent with existing experimental evidence [13,14,22]. Subsequently, 84 GO terms and 30 KEGG pathway terms were returned and classified. Finally, 3 networks were constructed and integrated.
In order to better understand the relationship between enriched KEGG pathways and putative target proteins, manual annotation was conducted based on KEGG pathway maps, as shown in Figure 8. The left panel shows the potential MOAs that are directly related to Aβ synthesis, transport, degradation, and clearance. The right panel shows MOAs that are indirectly linked or irrelevant to Aβ. Interestingly, the putative target proteins in the left panel are rarely enriched in the KEGG pathway terms, but the putative target proteins in the right panel are enriched in many signaling pathways. It seems that the putative target proteins in the left panel are more like "lone rangers", while the putative target proteins in the right panel play a physiological role by interacting with other proteins; this might also be related to most of them being protein kinases. The results in Figure 7 also confirm this viewpoint, and the putative target proteins in cluster B are mostly distributed in the left panel of Figure 8. This clustering may be a possible explanation for the findings from the DAVID enrichment algorithm, but it does not hinder the process of discovering their unique roles in AD pathology. These proteins, such as Endothelial Nitric Oxide Synthase (NOS3) [23][24][25][26][27], neprilysin (NEP) [28,29], Beta-secretase (BACE) [30][31][32][33][34], Monoamine oxidases (MAOs) [35][36][37], Prothrombin (F2) [38], Serum albumin (ALB) [39], Thyroid hormone (TTR) [32], and Matrix metalloproteinase 3 (MMP3) [40], are directly or indirectly involved in Aβ synthesis, processing, aggregation, degradation and transport; the formation of neurofibrillary tangles (NFTs); and tau proteolysis. In the right panel, the hormone-related signaling pathways were on top of the KEGG pathway enrichment results. Therefore, we hypothesize that hormone-related signaling pathways may play an important role in the anti-AD effects of G. biloba. These signaling pathways include the (hsa04917) prolactin signaling pathway, (hsa04915) estrogen signaling pathway, (hsa04912) GnRH signaling pathway, (hsa04910) insulin signaling pathway, (hsa04921) oxytocin signaling pathway, (hsa04919) thyroid hormone signaling pathway, (hsa04722) neurotrophin signaling pathway, and (hsa04071) sphingolipid signaling pathway. Based on accumulating evidence [41,42], non-Aβ-related pathways are also an important factor in AD etiology, particularly prior to Aβ plaque formation. Prolactin [43], estrogen [44,45], oxytocin [46,47], thyroid hormone [48], and insulin [49] potentially play substantial roles in non-Aβ-related mechanisms. As for the studies on G. biloba related to the above-mentioned hormones, estrogen-and insulin-related studies have been extensive, but there has been less research on prolactin, oxytocin, and thyroid hormone. Based on the results of present study, it is worth further research.
As illustrated in Table 3, we identified 6 putative protein targets that were significantly related to AD, but have not been researched or have had only preliminary studies conducted on the anti-AD effects of G. biloba. Neprilysin (NEP), estrogen receptor (ESR), Prothrombin (F2), Serum albumin (ALB), Thyroid hormone (TTR), and Matrix metalloproteinase 3 (MMP3) are significantly associated with the development of AD. These above-mentioned protein targets are highly matched with several representative compounds in G. biloba, but the specific actions and properties must be further verified and probed in future experiments.
Nutrients 2018, 10, x FOR PEER REVIEW 10 of 15 As illustrated in Table 3, we identified 6 putative protein targets that were significantly related to AD, but have not been researched or have had only preliminary studies conducted on the anti-AD effects of G. biloba. Neprilysin (NEP), estrogen receptor (ESR), Prothrombin (F2), Serum albumin (ALB), Thyroid hormone (TTR), and Matrix metalloproteinase 3 (MMP3) are significantly associated with the development of AD. These above-mentioned protein targets are highly matched with several representative compounds in G. biloba, but the specific actions and properties must be further verified and probed in future experiments.  No research (2) maintain blood-brain barrier (BBB) integrity (3) participate in neuroinflammation [50] Estrogen receptor (1) upregulated insulin-degrading enzyme (IDE) [51] (2) maintaining steroid homeostasis [52]   (1) Aβ degradation enzymes [28,29] (+)-Catechin Diosmetin Genkwanin No research (2) maintain blood-brain barrier (BBB) integrity (3) participate in neuroinflammation [50] Estrogen receptor (ESR) (1) upregulated insulin-degrading enzyme (IDE) [51] Genkwanin No research (2) maintaining steroid homeostasis [52] (3) altering synaptic plasticity [53,54] (4) participate in neurons oxidative stress-mediated injury [55] Prothrombin (F2) (1) coagulation cascade and endothelial cell integrity [38] Ginkgolide J No research (2) ideal molecular-biological indicator for AD [56] (3) proteolyzes the microtubule-associated protein tau [56] (4) inhibits phosphorylation of tau [56] Serum albumin (ALB) (1) bounded and transported Aβ, maintaining a constant concentration level in the brain [39] Ethyl oleate Flavoxanthin No research (2) Aβ excretion from the brain to the blood [57,58] Thyroid hormone (TTR) (3) a diagnostic biomarker for AD [61] In conclusion, the establishment of networks between AD-related protein targets and compounds in G. biloba may have important implications for elucidating the mechanisms underlying the beneficial effects of G. biloba on AD. Conceivably, a novel therapeutic strategy for AD may be developed from the protein targets and pathways identified in the present study. Hopefully, a novel paradigm presented in this study would help facilitate natural medicine development and the construction of a herbal compound library.

Conflicts of Interest:
The authors declare no conflict of interest.