The issue of herb-drug interactions has been widely reported. Herbal ingredients can activate nuclear receptors and further induce the gene expression alteration of drug-metabolizing enzyme and/or transporter. Therefore, the herb-drug interaction will happen when the herbs and drugs are coadministered. This kind of interaction is called inductive herb-drug interactions. Pregnane X Receptor (PXR) and drug-metabolizing target genes are involved in most of inductive herb-drug interactions. To predict this kind of herb-drug interaction, the protocol could be simplified to only screen agonists of PXR from herbs because the relations of drugs with their metabolizing enzymes are well studied. Here, a combinational in silico strategy of pharmacophore modelling and docking-based rank aggregation (DRA) was employed to identify PXR’s agonists. Firstly, 305 ingredients were screened out from 820 ingredients as candidate agonists of PXR with our pharmacophore model. Secondly, DRA was used to rerank the result of pharmacophore filtering. To validate our prediction, a curated herb-drug interaction database was built, which recorded 380 herb-drug interactions. Finally, among the top 10 herb ingredients from the ranking list, 6 ingredients were reported to involve in herb-drug interactions. The accuracy of our method is higher than other traditional methods. The strategy could be extended to studies on other inductive herb-drug interactions.
In America, nearly forty percent of adults consume herbs or herbal products regularly every year and this number is still increasing [
Herb-drug interactions, as well as drug-drug interactions (DDIs), are generally divided into two categories: pharmacodynamics (PD) interactions and pharmacokinetic (PK) interactions [
The mode of inductive drug interactions.
Pregnane X Receptor (PXR), as a member of nuclear receptor families, is involved in most of inductive herb-drug interactions through regulating drug-metabolizing gene expression [
In the past years, several computational methods have been used for virtual screening PXR’s agonists, such as structure-based docking [
The complex crystal structure provides the binding information objectively, which is used for pharmacophore modelling and molecular docking. Three complex structures of PXR were obtained from the Protein Data Bank (PDB
266 compounds with EC50 values were obtained from the Binding Database (BindingDB
In order to evaluate performance of our method, a dataset of herb-drug interactions was needed. 421 herbs were checked in the PubMed database by text mining method. 90 herbs were found to interact with 230 drugs forming 380 herb-drug interactions. Besides, molecular structures of herbal ingredients should be provided for pharmacophore modelling and molecular docking. Among 421 herbs, 820 ingredients structures were obtained from the PubChem database (
As shown in Figure
The pharmacophore of PXR (F1: Hyd|Acc; F2: Acc|Acc2|Don2; F3: Hyd|Acc2; F4: Hyd|Acc; F5: ARO|Hyd; V1–V8: excluded volume).
The molecular structure of template by superposing three SRL12813 in three different conformations.
Docking-based rank aggregation (DRA) is a two-step process. Firstly, candidate ligands filtered out by the former pharmacophore model were docked into PXR with four different energy scoring functions. The possibility of candidate ligands was ranked according their energy scores. Secondly, four different ranks from four scoring functions were aggregated to obtain a final rank.
The complex crystal structure of PXR (PDB id: 1ILH) was used to define the active site and dock with other molecules. Molecular docking was performed in MOE-Dock 2008.10. The way to place ligand was alpha sphere triangle matching with 4 different scoring functions (ASE Scoring, Affinity dG Scoring, Alpha HB Scoring, and London dG Scoring), respectively. The molecular mechanics force field was used to minimize energy of the system. 0.0001 kcal/(mol·Å) was chosen as the cutoff of the root-mean-squared gradient and maximum iterations was 1000 with their defaulted parameters. Finally, four ranked lists (
Rank aggregation is a kind of multiview data analysis strategy aiming to fuse ranking results derived from individual views [
Some concepts and details which are used in the process of rank aggregation are introduced below. Spearman’s distance [
To evaluate ranking performance in comparison with the control rank, discounted cumulative gain (DCG), a usual method to measure effectiveness of a web search engine algorithm, is used for evaluating performance of ranking. Two assumptions are acknowledged along with the use of DCG. One was that highly relevant items are more important when having higher ranks. The other is that highly relevant items are more important than irrelevant items. For a particular rank, the discounted cumulative gain accumulated position
Due to the variety of lists in length relying on the query, the best rank would not be achieved if DCG is used along consistently. It was necessary for normalizing the cumulative gain of each rank. The normalized DCG (nDCG) was computed as
As a result of pharmacophore model, the true positive rate (sensitivity) is 53.52% (38/71) and the true negative rate (specificity) is 81.54% (159/195). The pharmacophore model of PXR is displayed in Figure
Two different views on how to superpose template molecules were used to construct the pharmacophore [
Firstly, a list including 107 ligands of PXR was sorted by EC50 value and was regarded as reference list, named
The value of nDCG to measure distance between ranks.
Rank | nDCG |
---|---|
EC50 | 1 |
ABD | 0.7149 |
AB | 0.5397 |
D (London dG) | 0.4599 |
ACD | 0.4023 |
B (Affinity dG) | 0.3972 |
BD | 0.3961 |
AD | 0.3947 |
BCD | 0.3743 |
CD | 0.3670 |
A (ASE) | 0.3650 |
ABCD | 0.3639 |
ABC | 0.3609 |
AC | 0.3423 |
C (Alpha HB) | 0.3416 |
BC | 0.3405 |
In order to find a ranking list of ligands, which was closer to the reference list, we aggregated ranking lists derived from docking results. The aggregated result showed that
Through our aggregated lists by docking result (shown in Table
The description of four scoring functions.
Index | Scoring function | Description |
---|---|---|
A | ASE Scoring | The distance between all ligand atom-receptor atom pairs and ligand atom-alpha sphere pairs. |
|
||
B | Affinity dG Scoring | The enthalpic contribution to the free energy of various interaction including interactions between hydrogen bond donor-acceptor pairs, ionic interactions, metal ligation, hydrophobic interactions, interactions between hydrophobic and polar atoms, and interactions between any two atoms. |
|
||
C | Alpha HB Scoring | Combination of two measurements between the geometric fit of the ligand to the binding site and hydrogen bonding effects. |
|
||
D | London dG Scoring | The free energy for binding of ligand including the gain/loss of rotational and translational entropy, the loss of flexibility of the ligand, geometric imperfections of hydrogen bonds and metal ligation, and the desolvation energy of atom. |
The inductive herb-drug interactions were predicted through screening agonist of PXR from herbal ingredients. Every ingredient is contained by one or more herbs. An ingredient will be considered to have the potential of inducing herb-drug interaction if a herb, containing the ingredient, is reported in our herb-drug interaction database. The ingredient is regarded as the positive sample. The detection rate is used to measure the performance of the computational method, which is the ratio of positive samples in listed rank of screened ingredients.
305 ingredients were picked out from 820 ingredients of 421 herbs by our pharmacophore model. Then, three ranking lists of these 305 ingredients were generated, respectively, by molecular docking from three individual scoring functions (ASE, Affinity dG, London dG). A final list is obtained by aggregating these three lists. In the top 10 percent of the ranking list, the detection rate reached 0.6 (18/30). The whole results of rank aggregation are shown in Supplementary Table S2.
As validity of methodology, the performance of our method was compared with traditional methods. We predict the inductive herb-drug interactions through screening agonist of PXR. Because candidate agonists screened by us are a ranking list, three methods for screening ligand of protein were chosen to compare, such as molecular docking, Partial Least Squares- (PLS-) based QSAR, Principal Component Regression- (PCR-) based QSAR. Likewise, 820 herbal ingredients are screened by different methods. As shown in Figure
The detection rate in different ranking lists obtained by four methods.
As a part of screened result, the top 10 ingredients in final ranking list are shown in Table
The top 10 of final rank for candidate agonist of PXR from herbal ingredients.
Rank | Ingredients | Herbs | Reference (Y/N) |
---|---|---|---|
1 | Sophoraflavoside IV | Sophorae flavescentis | Y |
2 | Hesperidin |
|
Y |
|
N | ||
3 | Sennoside C&D |
|
N |
4 | Ginsenosides Rgl |
|
Y |
5 | Chlorophy II |
|
N |
6 | Solanine | Fritillariae cirrhosae | Y |
7 | Senegenic acid |
|
N |
8 | Sophoraflavoside III | Sophorae flavescentis | Y |
9 | Phellanmurin |
|
Y |
10 | Torulosic acid |
|
N |
Sophoraflavoside III and Sophoraflavoside IV are isolated from the roots of
This interaction was also observed in Ueng’s study. Nifedipine is a dihydropyridine calcium channel blocker that primarily blocks L-type calcium channels [
A clinical case of a 61-year-old man indicated that fritillaria lessens anticoagulation of warfarin [
In our results, some ingredients were not reported to be associated with herb-drug interaction. Two potential interpretations are as follows:
In this study, a combinational in silico strategy was proposed to predict inductive herb-drug interactions. As a consequence, among 820 ingredients from 421 herbs, a ranking list of 305 ingredients was generated as candidate agonists of PXR. Among the top 10 herb ingredients from the ranking list, 6 ingredients were reported to involve herb-drug interactions. The strategy also could be extended to studies on other inductive herb-drug interactions. Besides, during the process of screening agonists for PXR, our pharmacophore model achieved a good performance across a broad dataset. What is more, the ranking result of traditional molecular docking was improved by rank aggregation. It is suggested that combining merits of scoring functions with less redundancies is a new orientation to optimize scoring functions.
The authors declare that there is no conflict of interests regarding the publication of this paper.
This work was supported by National Natural Science Foundation of China 30976611 (to RZ) and 31171272 (to WZ) and the Fundamental Research Funds for the Central Universities 2000219083 (to RZ). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the paper.