In Silico Docking Analysis of A-type Proanthocyanidins to α-Glucosidase Constructed by Correlation with in Vitro Bioassay

The type A proanthocyanidins (2−8) with (2β→O→7, 4β→8) interflavane linkage, isolated from Machilus philippinensis, have been found to possess inhibitory activity against α-glucosidase (EC 3.2.1.20 from Bacillus stearothermophilus). To rationalize such activity, computer assisted docking of these compounds and the positive control, acarbose, on the conformation model of α-glucosidase (AG), built by using human intestinal maltase glucoamylase as a template, was undertaken in this study. The result showed good correlation between IC50 values and docking scores, expressed as binding energy (∆G), obtained from London (trimatch)-refinement (forcefield Affinity∆G) mode. Among these isolates, the most potent aesculitannin B (2) (IC50 3.5 μM) showed the best docking score (∆G -21.48 kcal/mol). Being interested in clarification of structure and activity relationship, virtual screening on the related compounds, including the de-unit III homologs of 2−8 (i.e., norseries) and additional 13 stereoisomers of the trimeric 2 at the C-2 and C-3 positions of units II and III, was further carried out. This docking study indicated the de-unit III homologs of 2−8 did not have better binding energies than 2. As for the trimers, 3-entC, 3C-entE, 3ent-C, 3C, and 3ent, showed comparable docking score to 2. The verification of this virtual screening was partially done by evaluating the inhibitory activity of the dimeric 2-nor-ent, 3-nor, 3-nor-ent, and iso-2-nor-ent, isolated from peanut skins, against α-glucosidase. Of these, iso-2-nor-ent, the only proanthocyanidin with (2β→O→7, 4β→6) interflavane linkage, showed the best activity (IC50 9.72 μM). Their simulation profiles of docking score also displayed a reasonable qualitative consistency with the overall trend of the bioassay results. This study demonstrates that virtual screening using this built model to search α-glucosidase inhibitors is facile and feasible and peanut skin might be used as a hypoglycemic food.


Construction of α-Glucosidase Conformation as
Docking Model The conformation model of α-glucosidase (AG; EC 3.2.1.20 from Bacillus stearothermophilus) using human intestinal maltase glucoamylase (PDB ID: 2QLY) [10] as a template had been constructed as described in our recent publication [7]. All the computational and structural studies were carried out using MIFit (a cross-platform interactive graphics application for molecular modelling) and Molecular Operating Environment (MOE, 2010.10; Chemical Computing Group Inc. software) [11]. To model the binding sites of α-glucosidase, the SiteFinder function of MOE based on the constructed three-dimensional α-glucosidase was used.

Ligands for Docking
Seven A-type proanthocyanidins (2−8) together with epicatechin (1), isolated from the EtOH extract of the leaves of Machilus philippinensis [8] and four dimeric isomers (2-nor-ent, iso-2-nor-ent, 3-nor, and 3-nor-ent), isolated from peanut skins [9], were chosen for this in silico docking analysis ( Figure 1). Next, the docking of the truncated 2−8 (2-nor−8-nor), obtained by deleting unit III, and all possible trimeric stereoisomers at the C-2 and C-3 positions of units II and III of 2−4 ( Figure 2) to AG were carried out. The three dimension structures of these compounds and acarbose were constructed using ChemBio3D, then imported them to MOE for molecular docking simulation.

Molecular Docking and Post-dynamic Analysis
Molecular docking study was done to reveal the ligand-receptor interactions and to compare affinities of various compounds to the target AG model. The AG structure was protonated in the MOE [15]. The triangle matcher algorithm of the MOE software packages was selected to dock the identified hit compounds into the chosen protein active site. Docking calculations were carried out using standard default variables for the MOE. Binding affinity was evaluated by the binding energies (S-score, kcal/mol), hydrogen bonds, and root-mean-square deviation (RMSD) values. The binding energy was calculated between the protein and the ligand, intramolecular hydrogen bonds and strains of the ligand. The RMSD was computed in terms of all the atoms in a protein backbone and the value was less than 0.6 Å which was indicative of considerable structural similarity. The compounds were docked into same groove of the binding site. Then, the initial model was loaded into MOE working environment ignoring water molecules and heteroatoms. The structure with all the atoms shown was put in generalized born implicit solvated environment at a temperature of 300K, pH of 7.0, and a salt concentration of 0.1 mol/L. Electrostatic potential was applied to a cut-off value of 1.5 Å at a dielectric value of 1.0. A non-bonded cut-off value of 5.0 Å was applied to the Leonard-Jones terms. The dock scoring in MOE software was done using London ∆G scoring function to estimate the ligand-protein binding free energy and enhanced by the Forcefield refinement method (Affinity ∆G or London ∆G) to relax the poses and then the refinement scores to rank the poses output to AG. Poses had been updated to ensure that refined poses satisfy the specified conformations. The rotatable bonds were allowed and then the best 20 poses were retained to analyze their binding scores. Energy minimization was conducted through Force-field MMFF94x optimization using a gradient cut-off value of 0.05 Kcal/mol/Å for determining low energy conformations with the lowest energy geometry [16]. From the final list of these 20 docked conformations, the pose with least docking score ligand was then chosen for further analysis.

Isolation of Proanthocyanidins from Peanut Skin
The dry peanut skins (7.00 kg) obtained from the baked peanut were stirred with 95% EtOH (1 × 45 L, 3 × 24 L) at 50 o C. The EtOH solutions were concentrated under reduced pressure at 50°C to give the EtOH extract (1.62 kg). The suspension of the EtOH extract (250.20 g) in H 2 O (2.5 L) was partitioned in sequence against CH 2 Cl 2 , EtOAc, and n-BuOH (saturated with H 2 O), each 3 × 2.5 L, to give fractions soluble in CH 2 Cl 2 (78.4 g), EtOAc (42.6 g), n-BuOH (78.8 g), and H 2 O (30.3 g) after evaporation of each fraction under reduced pressure at 50°C.

Docking Calculation
An inhibitory compound with a high binding energy (i.e. bigger negative numbers) should have a low IC 50 . Molecular docking of compounds 1−8 and the de-unit III homologs of 2−8 on α-glucosidase was undertaken initially via London (trimatch)−refinement (Forcefield-Affinity ∆G and Forcefield-London) mode and the results were listed in Tables 1 and 2. The correlation between docking score and bioassay result (IC 50 ) is roughly consistent. However, the binding scores yielded from Forcefield-Affinity ∆G mode agreed better with bioassay results (IC 50 ). Further docking and bioassay study on four proanthocyanidins (2-nor-ent, 3-nor, 3-nor-ent, and iso-2-nor-ent), isolated from peanut skins [9], also confirmed this observation. As such, molecular docking of 13 isomeric trimeric proanthocyanidin isomers at the C-2 and C-3 positions in units II and III on α-glucosidase was undertaken via London (trimatch)−refinement (Forcefield-Affinity ∆G) mode. The result was shown in Table 3.

Interaction of AG with Proanthocyanidins from
Machilus Philippinesis Our investigation was primarily based on the interplay of conformational change and spatial orientation associated with the binding affinity. Acarbose showed six H-bonds and five ionic interactions with the constructed docking model of α-glucosidase (EC 3.2.1.20 from Bacillus stearothermophilus) [7]. As listed in Table 1, the calculated docking energies were in good accordance with the IC 50 values.
As shown, while the binding affinity increased, the IC50 value decreased. Among these, the trimeric aesculitannin B (2) fitted into the hydrophobic pocket of AG with hydrophobic interactions, which strongly associated with F649, and hydrogen-bonding interactions with the side chain of D443 ( Figure 3A; Figure A1, Supplementary data). It was noted that the 7-OH in unit I had preferred orientation towards the binding pocket interior. These interactions made 2 the most potent against AG (IC 50 3.5 µM) among these proanthocyanidins. As for the other two trimeric isomers (3 and 4), they showed much weaker anti-AG activity, exemplified by their low docking affinity to AG (Table 1). For the tetrameric isomers, pavetannin C-1 (5), having an additional epicatechin residue (6→4β)-linked to unit I of cinnamtannin B-1 (3), is more potent than parameritannin A-1 (6), having an additional epicatechin residue (6→4β)-linked to unit II of 3. The better docking affinity of 5 than 6 to the active site (∆G -20.46 vs. -8.01) could be explained by possessing two more H-bonds (unit I′ 7-OH and 4′-OH to D443 and D645, respectively) ( Figure 4A; Figure  A2, Supplementary data). The additional epicatechin moiety in 6 was observed to hamper the flexibility of unit III, leading to a poor docking score to AG ( Figure 4A; Figure A2A, Supplementary data). Machiphilitannin A (7), having an additional epicatechin residue (8→4β)-linked to unit II′ of 6, had better structure flexibility than 6, arisen from the additional moiety. This allowed more n-π interactions to AG and made 7 have better anti-AG activity than 6 (IC 50 31.3 µM vs. > 100 µM). The structure of machiphilitannin B (8), having an additional epicatechin residue (8→4β)-linked to II″ of 7, was even more flexible than that of 7, allowing 8 to have one H-bond more than 7, thus had better anti-AG activity (IC 50 18.4 µM vs. 31.3 µM) ( Table 1).

Interaction of AG with de-unit III Homologues of 2−8
To examine the contribution of unit III to the anti-AG activity, this unit in 2−8 was truncated to give nor 2−8. Their virtual binding affinity to AG was shown in Table 2.
According to the binding affinity and the orientations of the dimeric isomers (2-nor-ent and 3-nor) to AG, the unit II 4′-OH, potentially responsible for hydrogen bonding to AG, was oriented toward the binding pocket interior. The higher binding affinity of 3-nor than that of 3 was due to the smaller molecular size, making it readily accessible into binding pocket. Since 2-nor and 2-nor-ent constituted more compact conformations than 2, they cannot form an ionic interaction with AG as that of 2, leading to much weaker binding affinity.
To validate our docking calculations, the same docking approach was carried out for epicatechin-(2β→O→7, 4β→6)-catechin (iso-2-nor-ent) [9]. Its docking score value was well correlated with IC 50 value ( Table 2). While comparing to the isomer 2-nor-ent, iso-2-nor-ent showed much better docking score. Such difference could be rationalized as follows. The steric hindrance effect derived from the 4β→8 linkage in 2-nor-ent, which influenced the spatial orientation and shifted its unit II residue toward the binding pocket interior of AG, forming only one H-bond (3′-OH to H674) ( Figure 5). As for iso-2-nor-ent, its unit I residue migrated toward the binding pocket interior to form three H-bonds (5-OH to W481, 7-OH to D404, and 3′-OH to D645), one π-π and one n-π interactions with AG ( Figure 5).

Correlation of IC 50 and Docking Score to AG of
De-unit III Homologues of 5−8 As indicated above, the steric hindrance between unit III and unit II′ will reduce the flexibility and hence decrease binding affinity. Thus truncation of this unit from 6−8 might increase the binding affinity. This is the case since the nor-analogues of 6−8 showed better docking scores than the corresponding parent compounds (Tables 1 and 2).
Removal of unit III from 5 (5-nor), however, decreased the binding affinity ascribable to the orientation change while docking as shown in Figure 4B and Figure A2B (Supplementary data). Thus, the spatial orientations of the flavan-3-ol residue should play an important role in designing AG inhibitors. It was noted that the phenolic hydroxyl groups not only served as the source of hydrogen bonding but also contributed lone-pair electrons to enhance n-π interactions to AG, providing additional stability to these complexes.

Interaction of AG with 3-entC, 3C-entE, 3ent-C, and 2
As indicated, those with unit II identical to unit III, e.g. 3 (both units being epicatechin), showed almost the worst docking affinity. While 3-entC, 3C-entE, 3ent-C, and 2 possessing the same chirality at C-3 in both units but different at C-2, e.g. 3-entC having 3-R in both units, 2-S in unit II but 2-R in unit III, showed good docking score. As shown, the structure conformations of 2 ( Figure 3A; Figure  A1, Supplementary data) and 3-entC ( Figure A3, Supplementary data) were folded into an extended and flexible shape, allowing them to have better docking probability.
As those of 2 and 3-entC, the structures of 3C, 3ent, and 2-C were also folded into an extended and flexible shape ( Figure A4, Supplementary data) and thus have better docking. While compounds 4 (3-C), 3-entE, and 3C-entC with the corresponding exchanged units II and III moieties as those three latter mentioned compounds (3C, 3ent, and 2-C) folded into compact and fairly rigid structures due to intramolecular hydrogen bonding or/ and π-π interaction (for 4: Figure A5, Supplementary data), leading to much weaker docking capacity.
The visual comparison of the MD simulation of these trimeric ligands to AG confirmed that the binding site of AG preferred to adopt a rather extended conformation instead of a compact one.
With the observation of various complex orientations, the MD simulation showed that residues D404, D443. D518, D524, and D616 appeared to be responsible for potentially favorable hydrogen-bonding interactions between the oligomeric 5,7,3′,4′-tetrahydroxyflavan-3-ols and the AG active site, similar to those reported for the active-site pocket (D404, I441, W481, W516, D518, M519, R600, D616, F649, and H674) [19] and those responsible for key catalytic activity (D518 and D616) [20−22]. As shown in Table 3, π−π interactions, mainly arisen from ring B of unit I in the ligands with the phenylalanine residue in AG, are also responsible for some bonding strength. The present study also indicated that for the good docking compounds such as 2 in Tables 1 and 3, the oxygen atom of the 3-OH in unit I formed an ionic bond with the guanidinium nitrogen of A600 in AG ( Figure 3A), greatly increasing the docking score.

Conclusion
In this work we have demonstrated a good correlation between the IC 50 values against α-glucosidase and in silico molecular docking scores using our built α-glucosidase model on seven proanthocyanidins, isolated from Machilus philippinensis (2−8). Extension of this study was undertaken on eight de-unit III homologs of 2−8, and 13 trimeric proanthocyanidin, belonging to stereoisomers of 2 at C-2 and C-3 of units II and III. The simulation profiles of binding energy displayed a reasonable qualitative agreement with the IC 50 on the dimeric 2-nor-ent, 3-nor, 3-nor-ent, and iso-2-nor-ent, isolated from peanut skins, partially verifying this MD study. In addition, five trimeric isomers (3-entC, 3C-entE, 3ent-C, 3C, and 3ent) were demonstrated to have docking score comparable to aesculitannin B (2, IC 50 3.5 µM) and they should be of value for further exploration as α-glucosidase inhibitors.  First and foremost, the docking analysis is very useful in virtual screening. While examining the docking data, not only the scores are concerned but also the conformation of ligand fitted to the target (AG in this study) since some better scores might come from unlikely twisted conformation. Our study has provided encouraging idea for screening in selecting compounds for further investigation.