Virtual Screening Description and Application

We used a slightly modified version of a VS workflow that was previously developed and experimentally validated [1] to identify DPP-IV inhibitors in a large in-house database of natural products (NPs) annotated with their natural source.

The VS workflow (see Figure 1) consisted of several sequential steps in which the output molecules of one step were the input molecules for the next step and so on. Central in this workflow is one structure-based common pharmacophore that captures the key intermolecular interactions needed for drugs to inhibit DPP-IV; this pharmacophore is formed by 2 mandatory sites (i.e., one positive/donor and one hydrophobic/aromatic ring) and 5 optional sties (i.e., two hydrogen-bond acceptors and three hydrophobic/aromatic ring sites). Both mandatory sites interact with crucial molecular anchors for DPP-IV inhibition (i.e., the hydrophobic/aromatic ring interacts with the hydrophobic S1 pocket at the DPP-IV binding site, and the positive/donor site interacts with the Glu205/Glu206 dyad [20]). Briefly, the VS workflow consists of (1) comparing ligand conformers to the common pharmacophore by allowing reorientation of the conformers to determine if they match the pharmacophore; (2) using ligands with at least one hit in the previous filter in a protein-ligand rigid-docking study and docking them onto the ligand binding site of the DPP-IV conformation present in the 3C45 PDB file; (3) comparing the resulting docking conformations to the structure-based common pharmacophore without reorienting the poses; and (4) submitting the poses that were hits in the previous filter to a shape and electrostatic-potential comparison with the experimental pose of the DPP-IV inhibitor in the PDB file 3C45 (the previously developed VS workflow [1] was altered for the current work to use slightly lower threshold values for the electrostatic and shape comparisons).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. The VS workflow used in the present work.

The number beside each VS step is the number of molecules that survived that step.

https://doi.org/10.1371/journal.pone.0044972.g001

This VS protocol was applied to an in-house database of 29,779 NPs with appropriate ADME/Tox properties. The first filter found that 10,883 molecules in our database have at least one conformer that after proper reorientation, matches the pharmacophore (see Figure 1). Only 332 out of these 10,883 molecules have docked conformations that without reorientation, are able to match the pharmacophore (see Figure 1). This reduction is useful because it discards those molecules that are predicted to bind in a non-productive way to the DPP-IV binding site. Finally, the later filter (i.e., electrostatic and shape similarity screening) aims to smooth differences in chemical structures and translate them into criteria important for their intermolecular interactions with the ligand-binding site. This filter has been reported to be a valuable VS tool for the discovery of novel scaffolds [21], and in our case, it was applied to rescore the 332 hits that survived the previous filter. Consequently, only those molecules that had at least one docked pose that met the following conditions in the comparisons made by EON were predicted to inhibit DPP-IV: (a) ET_pb ≥0.468 and (b) ST ≥0.237. Interestingly, the fact that DPP-IV inhibitors (a) have a significant positive electrostatic potential in the region that interacts with the Glu205/Glu206 dyad (see Figure 1) and (b) that this ligand area matches the mandatory positive/donor site justifies the dominance of the electrostatic contribution over the shape contribution in the selected thresholds. Finally, the VS workflow identified 84 molecules with potential DPP-IV inhibitory activity (see Figure 1).

Virtual Screening Hits in Natural Extracts with Known Antidiabetic Activity

According to the information available in our in-house NPs database, the 84 molecules that were predicted by the VS workflow as potential DPP-IV inhibitors have been isolated from 95 different natural sources. Interestingly, a systematic bibliographic search of PubMed (http://www.pubmed.org) revealed that the extracts of 12 out of these 95 natural sources have been reported to exhibit antidiabetic activity (see Table S1). Moreover, among these 12 sources we found 12 VS hits that may, through their role as DPP-IV inhibitors, contribute to the observed antidiabetic activity of their corresponding extracts (see Table S1). In fact, a search using SciFinder (http://www.cas.org/products/sfacad) revealed that 6 out of these 12 natural compounds correspond (or are similar in chemical structure) to molecules with known antidiabetic properties (see Table S1). This finding further validates our methodology (in addition to the experimental validation of the original VS workflow [1]) and suggests that the mode of action of these molecules could involve DPP-IV inhibition. The remaining 6 natural compounds not previously reported to have antidiabetic properties represent new molecules that may also exhibit this bioactivity. In the next paragraphs, the most significant compounds found in these 12 antidiabetic extracts are discussed:

• N-Nororientaline (CAS number 29079-44-5, see Table S1) has been isolated from five plant species from genus Erythrina (i.e., Erythrina variegata [22], Erythrina crystagalli, Erythrina indica, Erythrina poeppigiana [22], and Erythrina arborescens) and was identified as a DPP-IV inhibitor by our VS procedure. Extracts from Erythrina variegata, Erythrina senegalensis, Erythrina addisoniae, Erythrina abyssinica, and Erythrina mildbraedii are reported to enhance antidiabetic activity [23]–[27]. Isoprenylated flavonoids isolated from Erythrina mildbraedii and prenylflavonoids isolated from the Erythrina senegalensis roots have been described as inhibitors of two other proteins frequently targeted in T2DM treatment (i.e., protein tyrosine phosphatase-1B [25], [26] for the former class of molecules and acyl CoA:diacylglycerol acyltransferase [24] for the latter class). All of these flavonoid compounds are chemically related to our hit. Therefore, the antidiabetic action of extracts from these plants may be the result of more than one bioactive component and mode of action.
• Tecostamine, which is found in Tecoma stans, has hypoglycemic properties similar to those of tecomine [28]. Our results therefore suggest that the hypoglycemic properties of tecostamine could be mediated by the inhibition of DPP-IV. In addition, the Tecoma stans aqueous extract posses at least four antidiabetic-related activities (i.e., intestinal á-glucosidase inhibition, post-prandial antihyperglycemic, hypocholesterolemic and hypotriglyceridemic effects). Therefore, although it is possible that most of these activities are exerted by the phenolic compounds present in the Tecoma stans aqueous extract, bioguided studies are necessary to confirm this hypothesis [29].
• Epinephrine (also known as adrenaline), which is found in Scoparia dulcis, has been reported to improve hypoglycemia [30]. Scoparia dulcis has been described as a folk-medicinal plant and has been traditionally used as a remedy for diabetes mellitus in India and for hypertension in Taiwan [31]. From Indian Scoparia dulcis, an antidiabetic compound named amellin was also isolated and characterized by Nath [32].
• From the same family as epinephrine, two additional compounds were predicted by our VS to be potential DPP-IV inhibitors. These compounds are (+)-pseudoephedrine (CAS number 90-82-4) and (−)-ephedrine (CAS number 299-42-3), and both have also been reported to have hypoglycemic activity [33]. Both molecules are found in several Ephedra species (Ephedra alata [34], Ephedra distachya [35], Ephedra equisetina [36], Ephedra gerardiana [35], Ephedra shennungiana [37], Ephedra sinica, Ephedra vulgaris and Ephedra pflanze). However, Ephedra distachya and Ephedra alata are the only species reported to have antidiabetic properties [33], [38]; therefore, we proposed that the remaining Ephedra species may be new sources of antidiabetic extracts.
• The molecules ajmaline and isosandwichine, which are enantiomers with the same CAS number, 509-37-5, are found in several Rauwolfia species. Rauwolfia vomitoria has been investigated for the content of alkaloids, especially those alkaloids with hypotensive and anti-inflammatory properties [39] in addition to the antidiabetic properties. Rauwolfia serpentina is also used as antidiabetic extract [40]. The remaining species that contain these molecules, such as Rauwolfia canescens [41], Rauwolfia degeneri [42], Rauwolfia densiflora [43], Rauwolfia heterophylla [44], Rauwolfia indecora, Rauwolfia obscura [45] and Rauwolfia tetraphylla [42], are putative antidiabetic extracts.
• Serpinine (CAS number 509-38-6) is isolated from Vinca major and several Rauwolfia species (Rauwolfia obscura [45], Rauwolfia tetraphylla, Rauwolfia serpentina [46] and Rauwolfia sellowii [1]), belongs to the same cluster as ajmaline and isosandwichine (i.e., cluster 78; see Table S1). Therefore, they share similar chemical structures and natural sources. Moreover, Vinca major organic leaf extract strongly stimulates glucose utilization [47].
• One interesting hit predicted to be a DPP-IV inhibitor is an epicatechin derivate that is found in Vitis vinifera. An antihyperglycemic effect in streptozotocin-induced diabetic rats and insulinomimetic activity in insulin-sensitive cell lines have been described for grape seed procyanidin extracts (GSPE) [48]. In addition, it has been demonstrated that oligomeric procyanidins from GSPE activate the insulin receptor by interacting with and inducing the phosphorylation of the insulin receptor and that this interaction leads to increased glucose uptake [49]. Moreover, several epicatechin derivates have been reported to have antidiabetic properties (the most studied of which is epigallocatechin gallate [50]). Some findings demonstrate that epigallocatechin gallate may be a novel, plant-derived compound capable of reducing the risk of type 1 diabetes [51]. Therefore, the DPP-IV inhibition induced by this epicatechin derivate may contribute to the antihyperglycemic effect of GSPE [48].
• The remaining 3 molecules predicted to be DPP-IV inhibitors through our VS workflow and that are found in extracts with described antidiabetic properties are hydroxysmirnovine from Galega orientalis [52], (−)-halosaline (CAS number 26648-71-5) from Haloxylon salicornicum [53] and isochanoclavin-(I) (CAS number 1150-44-3) from Pennisetum typhoideum [54] (see Table S1). Our results suggest that these molecules could be DPP-IV inhibitors and that extracts containing these molecules could potentiate the glucose-induced insulin response by prolonging the half-lives of GLP-1 and GIP incretins, due to the inhibition of DPP-IV. This information is novel and relevant, as no mechanism that explains the antidiabetic properties of these extracts has been previously suggested.

Virtual Screening Hits in Natural Extracts with No Described Antidiabetic Activity

Taking into account that extracts from closely related species of the same genus may share a high number of components, we also determined if any of our VS hits that were isolated from plants with no described antidiabetic activity belong to the same genus as species with known antidiabetic properties. We identified 6 molecules isolated from 6 different plants, Aconitum japonicum, Ervatamia officinalis, Solanum nudum, Solanum sodomaeum, Stephania cepharantha and Tabernaemontana eglandulosa (see Table S2), that meet these criteria. The related species with described antidiabetic properties are Aconitum carmichaelii [33], Aconitum moschatum [2], Aconitum violaceum [2], Ervatamia microphylla [55], Solanum lycocarpum [56], Solanum nigrum [57], Solanum xanthocarpum [58], Stephania hernandifolia [59], Stephania glabra [60], Stephania tetrandra [61], and Tabernaemontana divaricata [55]. Therefore, it is plausible to hypothesize that these 6 plants could also have antidiabetic properties mediated, at least partially, by the inhibition of DPP-IV.

Finding New Scaffolds of Natural Origin for DPP-IV Inhibitors

The 18 molecules in Tables S1 and S2 that were predicted to be DPP-IV inhibitors are of interest. To quantify the number of new scaffolds for DPP-IV inhibitors that were identified in our study, we merged the 18 VS hits with 2,342 known DPP-IV inhibitors, and the resulting set was classified according to structural similarity into 99 clusters (results not shown). Interestingly, the 18 hits were classified in 13 clusters (see Tables S1 and S2) that do not contain known DPP-IV inhibitors (results not shown). Thus, these 18 predicted DPP-IV inhibitors correspond to 13 different chemical scaffolds that are unrelated to those present in known DPP-IV inhibitors, and, consequently, these new scaffolds could be used either in lead-hopping experiments to identify new DPP-IV inhibitors or in structure-activity studies to identify natural-product derivates with stringer DPP-IV inhibition activity than the original NPs from which they are derived.

Conclusions

In a previous study [1], we developed a VS workflow that was able to distinguish successfully molecules that inhibit DPP-IV and molecules that do not inhibit this enzyme. We experimentally demonstrated that our VS protocol was able to identify DPP-IV inhibitors that (a) were not structurally related to any known molecule that inhibits DPP-IV and (b) have never been reported to have antidiabetic activity. In the present work, we applied a slightly modified version of this VS workflow to an in-house database of 29,779 NPs annotated with their corresponding natural source(s). From this initial set of NPs, our VS procedure identified as potential DPP-IV inhibitors 84 hit molecules that have been isolated from 95 different natural sources. Interestingly, after an exhaustive bibliographic search, our results demonstrate that we are able to predict (a) 12 DPP-IV inhibitors that are present in 12 plant extracts with known antidiabetic activity and (b) 6 DPP-IV inhibitors that are present in 6 different plants species with no described antidiabetic activity but that share the same genus as plants with known antidiabetic properties (consequently, it could be suggested that these plants represent a potential new source of antidiabetic extracts). Moreover, none of these 18 hits exhibits chemical similarity with 2,342 known DPP-IV inhibitors, and, therefore, it is expected that a significant number of these hits could be lead-hopping candidates for the development of new DPP-IV inhibitors. At this point, it is also interesting to note that the analysis of the chemical structures of these 18 NP hits revealed that the majority of them are alkaloids containing basic nitrogen atoms (essential for proper interaction with the Glu205/Glu206 dyad). Moreover, our results provide a new hypothesis about the mechanisms by which, at least partially, these 12 extracts exert their antidiabetic effects (i.e., improving glucose homeostasis by prolonging the activity of GLP-1 and GIP through DPP-IV inhibition).

Lastly, we predicted that there are 77 other sources with no described antidiabetic activity that contain at least one VS hit. Consequently, our work opens the door to the discovery of new antidiabetic extracts of natural origin that could be of use, for example, in the design of functional foods aimed at preventing/treating T2DM. Therefore, the characterization of such extracts merits further attention, and such work is currently underway.

Initial Dataset of Natural Compounds Used

The database of NPs that was screened by the VS workflow contains 29,779 NPs from different origins (e.g., plants, and fungi) with appropriate ADME/Tox properties and no chiral ambiguities. An important characteristic of this database is that each molecule is annotated with (a) the natural sources from which it has been obtained and (b) the bibliographic references that describe how to extract the relevant molecule from each natural source. The 3D structures of the molecules in this NP database were processed with LigPrep v2.3 (Schrödinger LLC., Portland, USA; http://www.schrodinger.com) using the following parameters: (a) the force field used was OPLS 2005; (b) all possible ionization states at pH 7.0±2.0 were generated with Ionizer; (c) the desalt option was activated; (d) tautomers were generated for all ionization states at pH 7.0±2.0; (e) chiralities were determined from the 3D structure; and (f) one low-energy ring conformation per ligand was generated. Conformations and sites for the resulting ligand structures were determined during the generation of the corresponding Phase [62] databases with the Generate Phase Database graphic front-end. Default parameter values were used during this conformer generation process, with the exception of the maximum number of conformers per structure, which was increased from 100 (the default value) to 200. The conformer sites were generated with definitions made by adding to the default built-in Phase definitions the ability to consider aromatic rings as hydrophobic groups.

Virtual Screening Workflow

The VS workflow used in this work is the same as that described previously [1], except that the conditions of the last filter (i.e., the shape and electrostatic potential comparison) were slightly different, as outlined below.

The VS protocol used a structure-based pharmacophore that was built by (1) selecting from the PDB those reliable complexes of human DPP-IV and potent inhibitors of non-peptide nature (i.e., IC₅₀≤10 nM) that bind reversibly to the enzyme; (2) using their corresponding DPP-IV coordinates to guide the superposition of the remaining PDB files (the resulting re-oriented coordinates for these PDB files were also used in the pharmacophore-based searches, protein-ligand docking studies and shape and electrostatic-potential comparisons of the VS workflow); (3) using the resulting coordinates to derive the corresponding energetic structure-based pharmacophores; and (4) building the common structure-based pharmacophore for reversible DPP-IV inhibition by prioritizing energetically favorable features over energetically weaker ones. The resulting pharmacophore consists of two compulsory sites (one positive/donor and one hydrophobic/aromatic ring) and five optional sites (i.e., two acceptor sites and three hydrophobic/aromatic ring sites) and was completed with receptor-based excluded volumes that schematically represent the location of the DPP-IV residues that form the binding pocket in the PDB file 3C45.

The first step of the VS workflow uses the common structure-based pharmacophore to screen the conformer database with Phase v3.1 (Schrödinger LLC., Portland, USA; http://www.schrodinger.com) and allows the reorientation of the conformers to determine if they match the pharmacophore. Only those ligands with at least one conformer that matches the two compulsory sites of the common pharmacophore and at least one of the optional sites (without sterically colliding with the excluded volumes) survive this VS step. These ligands were docked onto the binding pocket of the PDB file 3C45 with eHiTS v2009 (SimBioSys Inc., Toronto, Canada; http://www.simbiosys.ca/ehits) [63] using default docking conditions, with the exception of the length of the sides of the cubic box encompassing the DPP-IV binding site, which was increased from 10 Å to 15 Å. Next, the resulting docked poses were again filtered with Phase through the common pharmacophore using the same filtering conditions as in the first Phase run (with the exception that no reorientation of the docked poses was allowed during the search). Thus, only docking poses compatible with the pharmacophore survived this filter. Finally, in the last step of the VS protocol, the poses that were hits in this second pharmacophore screen were submitted to a shape and electrostatic potential comparison with the DPP-IV inhibitor present in the 3C45 PDB file to rescore the hits. This comparison was completed with EON v2.0.1 (OpenEye Scientific Software, Inc., Santa Fe, New Mexico, USA; http://www.eyesopen.com) and used the Electrostatic Tanimoto combo (ET_combo) score as the similarity criterion. The ET_combo score is the sum of two calculations: (a) the Shape Tanimoto (ST) score, which is a quantitative measure of three-dimensional overlap (where 1 corresponds to a perfect overlap; i.e. the same shape), and (b) the Poisson-Boltzmann Electrostatic Tanimoto (ET_pb) score, which compares the electrostatic potential of two small molecules and ranges from 1 (identical potential) to a negative value that results from the overlap of positive and negative charges. In this work, we determined the EON threshold scores by comparing the inhibitor in 3C45 with the inhibitors found in a group of 24 PDB complexes of DPP-IV and reversible drugs that inhibit this protein (i.e., 1N1M, 20GZ, 2FJP, 2HHA, 2I78, 2IIT, 2IIV, 2OLE, 2ONC, 2OQI, 2OQV, 2QOE, 2QT9, 2QTB, 2RGU, 2RIP, 3C43, 3CCC, 3D4L, 3F8S, 3H0C, 3HAB, and 3HAC). This comparison of the experimental poses of DPP-IV inhibitors yielded threshold values for DPP-IV inhibition that were slightly lower than the ones used in the original VS workflow [1] (ET_pb ≥0.468 and ST ≥0.237 instead of ET_pb ≥0.623 and ST ≥0.244), probably because the original threshold values were obtained exclusively using potent DPP-IV inhibitors (i.e., IC₅₀≤10 nM); in this study, this condition was relaxed slightly. After rescoring with EON, only those NPs with at least one conformation with ET_pb ≥0.468 and ST ≥0.237 relative to 3C45’s inhibitor were considered to be DPP-IV inhibitor candidates.

Structural Similarity Analysis

The molecules that survived the electrostatics/shape similarity filter were merged with 2,342 known DPP-IV inhibitors obtained from the BindingDB database [64] and then clustered using Canvas v1.2 (Schrödinger LLC., Portland, USA; http://www.schrodinger.com). MOLPRINT2D fingerprints [65], using a fingerprint precision of 32 bits, were calculated for each molecule, and hierarchical clustering, based on Tanimoto similarities, was subsequently obtained. The number of clusters obtained was defined using the Kelley criterion [66], corresponding to a Tanimoto coefficient of 0.775 in this case.