QSAR, structure-based pharmacophore modelling and biological evaluation of novel platelet ADP receptor (P2Y₁₂) antagonist

Abstract

P2Y₁₂ has a key role in platelet aggregation and thrombus formation via an ADP-induced platelet activation mechanism. Recently, P2Y₁₂ antagonists have become of great interest in the clinical management of antithrombotic therapy. In light of this, we explored the pharmacophoric space of P2Y₁₂ using structure-based pharmacophore modelling. Subsequently, genetic algorithm and multiple linear regression analyses were conducted to select the best combination of physicochemical descriptors and pharmacophoric models to create useful predictive quantitative structure–activity relationship (QSAR) equation (r² = 0.9135, r_(adj)² = 0.9147, r_(PRESS)² = 0.9129, LOF = 0.3553). One pharmacophoric model emerged in the QSAR equation and was validated by analysing receiver operating characteristic (ROC) curves. The model was then used to screen 200 000 compounds from the National Cancer Institute (NCI) database. The top-ranked hits were in vitro tested, where their IC₅₀'s range between 4.20 to 35.00 μM when measured via the electrode aggregometry assay. Whilst, the VASP phosphorylation assay showed 29.70% platelet reactivity index for NSC618159, which is superior to that of ticagrelor.

In silico and in vitro discovery of P2Y₁₂ antagonists utilizing structure-based pharmacophore modelling directed by quantitative structure-activity relationship (QSAR) analysis.

1. Introduction

The P2Y₁₂ protein is a G_i-coupled metabotropic plasma membrane receptor.^1,2 P2Y₁₂ is embedded in the platelet's bilayer membrane and consists of 342 amino acids and two putative N-linked glycosylation sites at its N-terminus. Thus, it has seven hydrophobic transmembrane regions (TM1–7) connected by three extracellular and three intracellular loops (EL1–3 and IL1–3, respectively).^2,3 The structure of P2Y₁₂ contains four cysteine (Cys) residues at 17, 97, 175, and 270 positions, with the disulphide bridge between Cys97 and Cys175 being vital for receptor expression. Meanwhile, thiol-containing ligands and thienopyridine active metabolites interact with Cys17 and Cys270.^4,5 P2Y₁₂-mediated inhibition of adenylyl cyclase is stimulated by diphosphate analogues like adenosine diphosphate (ADP). In contrast, triphosphate analogues, such as adenosine triphosphate (ATP), antagonise P2Y₁₂ activity.⁶

P2Y₁ receptors are activated during coagulation via binding to ADP. This activation mediates platelets' shape-change and transitory reversible aggregation. The role of P2Y₁₂ is to maintain and amplify these changes and effects.⁷ Although ADP cannot release the contents of platelets' granules on its own, its interaction with P2Y₁₂ enhances their release considerably. This mechanism is triggered by thromboxane A₂ and the thrombin receptor-activating peptide.^8,9 Platelet aggregation and thrombus formation are stabilised by ADP-induced activation of P2Y₁₂.⁶ The activation of G_αi2 and the inhibition of adenylyl cyclase were shown to have a role in P2Y₁₂ mediated ADP-induced platelet activation.^10,11 Moreover, platelet aggregation is regulated by P2Y₁₂ activation in response to thromboxane A₂ or thrombin is also mediated by phosphoinositide 3-kinase (PI3-K).⁴ As a result of the ADP-induced platelet activation, P2Y₁₂ plays a vital role in platelet aggregation and thrombus formation. Currently, there are five P2Y₁₂ antagonists available commercially as antiplatelets. These antiplatelets are classified into nucleoside–nucleotide derivatives and thienopyridines. The nucleoside–nucleotide reversible antagonists do not require metabolism for their activity and include ticagrelor and cangrelor. In contrast, thienopyridines, including clopidogrel, ticlopidine, and prasugrel, are prodrugs metabolized by the hepatic cytochrome P450. Upon metabolism, active metabolites bind irreversibly to the Cys residues of P2Y₁₂ and antagonise its activity.^12–14 A thorough review of the approved drugs, potential naturally isolated and synthesised compounds, and related in silico studies has been made recently by our team.⁵

Designing pharmacophore models regarding the structural features of the target protein is commonly called structure-based pharmacophore modelling. The potential interactions of the co-crystallized ligand in proteins will be investigated using this approach. Various energy- or geometry-based approaches will then be translated into pharmacophore models. Structure-based pharmacophore modelling is used to represent better the interactions that might occur when target proteins and ligands fit each other. In 2014, a breakthrough was attained by resolving the 3D crystal structure of the human P2Y₁₂ receptor.^15,16 Two P2Y₁₂ crystal structures were deposited at the Protein Data Bank (PDB) co-crystallised with 2MeSATP, a partial agonist – PDB ID: 4PY0, and 2MeSADP, a full agonist – PDB ID: 4PXZ.¹⁵ Another crystal structure in complex with an antagonist, AZD1283 (ethyl-6-(4-((benzylsulfonyl)carbamoyl)piperidin-1-yl)-5-cyano-2-methylnicotinate), is also available in the PDB (PDB ID: 4NTJ).¹⁶

Herein, our group reports the discovery of a novel antagonist of the P2Y₁₂ receptor utilizing a structure-based pharmacophore modelling strategy directed by quantitative structure–activity relationship (QSAR) analysis. Finally, the results' significance prompted us to optimize the hits further using the prepared 3D-QSAR equations.

2. Results and discussion

2.1. Pharmacophore models and validation

Running the receptor–ligand pharmacophore generation protocol embedded in Discovery Studio has created 80 pharmacophore models upon varying the maximum features, minimum inter-feature distance, and maximum exclusion volume distance (Table S1, ESI†). The capability of the pharmacophoric model to selectively categorize a set of compounds as actives and inactives is determined by the area under the ROC curve. Table 1 shows the results of the ROC analysis of the selected pharmacophore to be used in QSAR analysis, along with their calculated statistical factors. The pharmacophore models mainly contain hydrogen bond acceptor (HBA), aromatic ring (RingArom), and hydrophobic (H_bic) features. Furthermore, additional excluded volumes were allowed to be added in several pharmacophore models.

Selected pharmacophore models for QSAR analysis and their statistical factors.

Pharmacophore model	Featuresa	ROCb – AUCc	TPRd	TNRe	ACCf	EFg	GHh
T1D3P6	AAHHRE	0.909	1	0.306	0.523	1.724	0.646
T1D3P8	AAHHRE	0.901	1	0.278	0.504	1.709	0.635
T2D3P6	AAHHRE	0.909	1	0.306	0.523	1.724	0.646
T2D3P8	AAHHRE	0.901	1	0.278	0.504	1.709	0.635
T3D3P6	AAHHRE	0.909	1	0.306	0.523	1.724	0.646
T3D3P8	AAHHRE	0.901	1	0.278	0.504	1.709	0.635
T4D1P4	AAAHHHR	0.928	1	0.556	0.695	1.869	0.758
T4D1P5	AAAHHHR	0.921	1	0.556	0.695	1.869	0.758
T4D3P6i	AAHHR	0.915	1	0.278	0.504	1.709	0.635
T4D3P9	AAAHH	0.916	1	0.361	0.561	1.754	0.670

^aA: hydrogen bond acceptor; H: hydrophobic; R: ring aromatic; E: excluded volume.

^bReceiver operating characteristic curve.

^cArea under the curve

^dSensitivity.

^eSpecificity.

^fAccuracy.

^gEnrichment factor.

^hGoodness-of-hit score

ⁱPharmacophore hypothesis appeared in the best QSAR equation.

As illustrated in Table 1, both models T4D1P4 and T4D1P5 showed equally the highest TNR, ACC, and GH score values of 0.556, 1.869, and 0.758, respectively. Still, T4D3P6 exhibited an EF value higher than 1 (1.709), and a GH score of 0.635, which is also considered as a valid pharmacophore model. Thus, classical QSAR analysis, another well-documented method for validating the selected models, was implemented to select the best pharmacophore based on its fit values equation.

2.2. QSAR analysis

The pharmacophoric models explain how ligands and receptors interact and may be utilized to find novel pharmacologically active compounds. We employed classical QSAR analysis in this work to identify the optimal combination of pharmacophores and other 2D and 3D descriptors capable of justifying the biological activities of ligands. To get the best QSAR equation, multiple linear regression and GFA were used. The fit values obtained from the dataset's pharmacophoric mapping (118 ligands) were uploaded with various physicochemical descriptors. Eqn (1) and Fig. 1 represent the details of the best QSAR model, as well as the corresponding scatter plots of experimental vs. predicted bioactivities for the training antagonists, respectively.

1

wherein, IC₅₀ is the half-maximal inhibitory concentration; Mwt is the molecular weight; HD is the number of hydrogen bond donors; r² is the correlation coefficient against 118 compounds, equals to 0.9135; r_(adj)² is r² adjusted for the number of terms in the model, equals to 0.9147; r_(PRESS)² is the prediction r², equivalent to q² from a leave-1-out cross-validation, equals to 0.9129; LOF is the Friedman lack-of-fit score, equals to 0.3553.^17–19

Fig. 1. Correlation between experimental versus predicted log(1/IC₅₀) values of test set (r² = 0.9135, r_(adj)² = 0.9147, r_(PRESS)² = 0.9129, LOF = 0.3553) calculated from the best QSAR model eqn (1). The dotted line is the regression line for the fitted and predicted bioactivities, while the dashed lines indicate 1.0 log point error margins.

T4D3P6 in the QSAR equation (eqn (1)) represents the fit values of the ligands against the sixth pharmacophoric hypothesis, using the interfacial distance of 3.0 Å from the fourth trial. The emergence of the pharmacophoric model in the equation means that one of the pharmacophores can optimally explain the bioactivities. Fig. 2 shows the pharmacophoric model, which is composed of HBA, RingArom, and H_bic features mapped to the co-crystallised ligand. These features suggest the importance of such interaction between the ligand and the amino acids in the P2Y₁₂ binding site. For further insight into the influence of the model on the binding site, Fig. 3 shows the intermolecular interactions between the co-crystallized ligand and the residues in the binding site explained by the pharmacophoric features. As shown in the figure (Fig. 3), the ligand performs hydrogen bond interactions with Asn159 and Lys280, aromatic interaction with Tyr105, and hydrophobic interactions with Leu155, Phe252, and Ala255. Consequently, based on these findings, in addition to the acceptable EF and GH score values of T4D3P6, this pharmacophore was selected as a rational model for virtual screening.

Fig. 2. 3D and 2D representations of the chemical structure of the co-crystallised ligand (AZD1283) obtained from crystal structure 4NTJ mapped to the T4D3P6 pharmacophore model. Hydrogen bond acceptor (HBA) features are illustrated as green vectored spheres, hydrophobic (H_bic) as light blue spheres, and aromatic ring (RingArom) as vectored orange spheres. Figures were generated via ChemDraw 16.0 and Discovery Studio.

Fig. 3. Stick representation of the interacting residues with the co-crystallised ligand (AZD1283) mapping with pharmacophore model T4D3P6. Hydrogen bond acceptor (HBA) features are illustrated as green vectored spheres, hydrophobic (H_bic) as light blue spheres, and aromatic ring (RingArom) as vectored orange spheres. The figure was generated via Discovery Studio.

2.3. In silico screening and subsequent in vitro evaluation

T4D3P6 was utilized as a 3D search query against 284 176 compounds from National Cancer Institute (https://wiki.nci.nih.gov/display/NCIDTPdata/Chemical+Data) using the “Best Flexible Database Search” option within Discovery Studio. Compounds that spatially map with features in T4D3P6 were considered as hits. Then, the fitting values and other correlated 2D and 3D molecular descriptors were substituted in the QSAR eqn (1) to predict their IC₅₀ values. The highest-ranking available compounds were in vitro evaluated for their potential P2Y₁₂ antagonist activity. Hits were first screened at 100 μM concentrations, and compounds that substantially inhibited P2Y₁₂ were then evaluated further to establish their percentage of inhibition according to eqn (7) (see Experimental). Among the screened compounds, NSC380323, NSC617595, and NSC618159 were selected for further investigations. The compounds mapped features in the T4D3P6 pharmacophore model with fit values of 4.8, 4.2, and 4.3 for NSC380323, NSC617595, and NSC618159, respectively, as shown in Table 2 and Fig. 4. Experimentally, the compounds inhibited platelet aggregation significantly (Fig. S1, ESI†). The percentage of inhibition of platelet aggregation for NSC380323 was 57.78%, 33.65%, and 22.26%, for 10, 1, and 0.1 μM, respectively. NSC617595 showed 61.84%, 38.16%, and 36.96% inhibition, at 10, 1, and 0.1 μM, respectively. On the other hand, NSC618159 showed 41.12%, 22.77, and 4.75% inhibition with the same series of concentrations (Table 2).

Predicted and experimental bioactivities of selected compounds.

Compounda	T4D3P6 fit value	Predicted IC50b (nM)	%Inhibition at 10, 1, and 0.1 μM (±StErr)	Experimental IC50c (±StErr) (μM)	%PRId at 10 μM (±StErr)
NSC380323	4.80	18.20	57.78 (±14.08)	6.49 (±3.42)	95.00 (±0.67)
			33.65 (±6.46)
			22.26 (±13.75)
NSC617595	4.20	38.50	61.84 (±4.21)	4.20 (±2.61)	92.80 (±0.85)
			38.16 (±9.25)
			36.96 (±3.63)
NSC618159	4.30	33.10	41.12 (±0.98)	35.00 (±6.84)	29.70 (±1.59)
			22.77 (±4.57)
			4.75 (±0.15)

^aChemical structures are shown in Fig. 4.

^bPredicted IC₅₀ nM according to QSAR eqn (1).

^cExperimental inactivation was determined in triplicates via electrode aggregometry assay.

^dPlatelet reactivity index measured via VASP phosphorylation assay for each compound at 10 μM.

Fig. 4. Mapping compounds (a) NSC380323, (b) NSC618159, and (c) NSC617595 against T4D3P6 pharmacophoric model. Hydrogen bond acceptor (HBA) features are illustrated as green vectored spheres, hydrophobic (H_bic) as light blue spheres, and aromatic ring (RingArom) as vectored orange spheres. Figures were generated via ChemDraw 16.0 and Discovery Studio.

Platelets aggregation results and the corresponding dose–response curves can be found in Table 2, and Fig. S1 (ESI†). The IC₅₀'s of NSC380323, NSC617595, and NSC618159 were calculated as 6.49, 4.20, and 35.00 μM, respectively.

ADP-induced platelet aggregation is mediated by a dual receptor system involving the activation of P2Y₁ and P2Y₁₂ receptors.²⁰ Inhibition of platelet aggregation by the 3 tested compound using multiplate platelet aggregometer may indicate that the tested compound is P2Y₁ antagonist, P2Y₁₂ antagonist or both. To check if the activity of the tested compound was specific for P2Y₁₂, VASP phosphorylation assay (CY-QUANT VASP/P2Y₁₂) was performed. Compound NSC618159 (10 μM) was found to have platelet reactivity index (%PRI) of 29.70% compared to 45.02% activity exerted by ticagrelor (Table 2 and Fig. 5). This indicates a better inhibitory activity than ticagrelor towards P2Y₁₂. On the other hand, NSC380323 and NSC617595 had no significant %PRI in VASP phosphorylation assay (>92.0%) compared to the vehicle (0.1% DMSO; %PRI = 97.0%). This may indicate that both compounds (NSC380323 and NSC617595) may function as P2Y₁ antagonists rather than P2Y₁₂ antagonists. Still the activity of the 3 tested compounds against P2Y₁ need to be investigation.

Fig. 5. Histogram representation of the platelet VASP phosphorylation assay for testing the antagonistic activity of compounds against P2Y₁₂ receptor. ****p ≤ 0.0001.

Fig. 6 shows the chemical structures of the tested compounds and compares its binding in the protein crystal structure (4NTJ) with how it maps against the T4D3P6 model without conformational adjustments. Fitting the two carbonyl groups in the acridone moiety of compounds NSC380323 and NSC618159 with an HBA feature corresponds to hydrogen-bonding interactions with Asn159 and Lys280. Similarly, mapping the acridone moieties against the H_bic feature is associated with fitting these moieties into a hydrophobic pocket composed of the hydrophobic side chain residues Leu155 and Ala255. Finally, mapping the aromatic ring against the RingArom feature correlates with stacking against the aromatic amino acid Tyr105. Compound NSC617595 showed similar mapping with the amino acids Asn159, Lys280, Leu155, Ala255, and Tyr105 in the binding site.

Fig. 6. Stick representations of mapped compounds (a) NSC380323 (magenta), (b) NSC618159 (orange), (c) NSC617595 (blue), and (d) the co-crystallized ligand AZD1283 (green) within the 4NTJ crystal structure binding pocket (amino acids in grey colour). Hydrogen bond acceptor (HBA) features are illustrated as green vectored spheres, hydrophobic (H_bic) as light blue spheres, and aromatic ring (RingArom) as vectored orange spheres. Figures were generated via Discovery Studio.

3. Conclusions and future work

P2Y₁₂ antagonists have attracted great attention recently for the clinical management of thrombosis. The pharmacophoric space of P2Y₁₂ antagonists was explored via the receptor–ligand pharmacophore generation protocol of Discovery Studio to identify a high-quality binding model. Subsequently, genetic algorithm and multiple linear regression analyses were utilized to obtain an optimal QSAR model capable of explaining activity variation across 118 compounds. One orthogonal pharmacophoric model that appeared in the QSAR equation indicates the importance of such a model in explaining the activity. The QSAR equation and the associated pharmacophoric model were experimentally confirmed by discovering a micromolar P2Y₁₂ antagonist retrieved via in silico screening. The significance of the results prompted us to optimize the hit further using the prepared 3D-QSAR equation. Finally, further in vitro investigations are required for compounds NSC380323 and NSC617595 on P2Y₁, in addition to in vivo studies using NSC618159 to confirm its activity and study its metabolites' antithrombotic activities.

4. Experimental

4.1. Structure-based pharmacophore modelling and validation

X-ray crystal complexes were reviewed previously to construct structure-based pharmacophore models.⁵ The selected receptor, 4NTJ, co-crystallised with the antagonist AZD1283,¹⁶ was selected to be uploaded in the receptor–ligand pharmacophore generation protocol in Discovery Studio 3.1.²¹ The Interaction Generation protocol was then used to produce pharmacophoric features from the active site. Through the analysis of the active site residues, this approach may identify hydrogen bond donors (HD), hydrogen bond acceptors (HA), and hydrophobic pockets (HY). The Edit and Cluster tool was used to cluster and eliminate duplicate features or features with no catalytic impact as the last step in optimizing the structure-based pharmacophore.

The pharmacophoric models were then evaluated against a dataset of synthetic or organic substances studied for P2Y₁₂ antagonist activity using a similarly biological testing procedure. The testing list was downloaded from the ChEMBL database (https://www.ebi.ac.uk/chembl/). These ligands were divided into active and inactive molecules based on their agonistic activity.^20,22–27 Validation using the ligand pharmacophore mapping approach will determine if the created pharmacophores can tell the difference between the active and inactive compounds. The statistical factors of the generated pharmacophore models, including the sensitivity (TPR), specificity (TNR), goodness-of-hit score (GH), enrichment factor (EF), and accuracy (ACC), were calculated using eqn (2) through (6), respectively. The value of the GH score ranges between 0, which indicates an invalid (null) model, and 1, which indicates an ideal model, with a cut-off value of 0.6.²⁸ In an ideal model, the ROC curve should have a steep slope as the model selects all active hits and omits all the inactive ones. Plus, it should have high values for AUC, EF, and the greatest possible values of sensitivity (TPR) and specificity (TNR).²⁹ Finally, an internal validation was conducted by providing a set of active and inactive ligands where the results were plotted as a receiver operating characteristic (ROC) curve.

2

3

4

5

6

wherein, TPR, sensitivity; TP, true positives; A, true positives in the database; TNR, specificity; TN, true negatives; D, true decoys in the database; GH, goodness-of-hit score; Ha, active hits; Ht, total hits; EF, enrichment factor; ACC, accuracy.

4.2. QSAR modelling

Since the statistical factors are not conclusive in validating the best pharmacophore model, QSAR is another approach which can be used to validate the chosen models. Herein, a group of 118 antagonists' test set with their activities was imported into Discovery Studio, and then different 2D and 3D descriptors were calculated. The chosen pharmacophores were utilized in a “ligand pharmacophore fitting” methodology for the ligands, with the best fit option in catalyst. The fit values were uploaded with additional QSAR descriptors. The QSAR equation was established utilizing genetic function approximation (GFA). The optimal GFA settings are as follows: investigate linear, quadratic, and spline equations with 50% mating and mutation probabilities, a population size of 500, 30 000 genetic iterations, and a lack-of-fit (LOF) smoothness parameter of 1.0.^17,19

4.3. Database screening

The generated pharmacophore model and QSAR equation were used as 3D queries to extract chemical compounds from commercially available databases. Virtual screening was performed on about 200 000 chemicals obtained from the NCI database (https://dtp.cancer.gov). The number of conformations was set to 250, while the conformation method was set to BEST, which provided a complete and improved coverage of the conformational space.³⁰ A chosen molecule from the screening process should map all the pharmacophoric features to be considered a hit. The fit values and the related molecular descriptors of each hit were uploaded in the early prepared QSAR equation. The top-ranking compounds based on QSAR predictions were obtained to be in vitro evaluated.

4.4. In vitro evaluation: measurement of platelet aggregation by multiple electrode aggregometry

This study was approved by the Research Ethical Committee at Al-Ahliyya Amman University (ethical approval number IRB: AAU/1/1/2021–2022). Blood samples (6 mL) were collected from healthy volunteers using a 19 gauge needle and a plastic syringe after signing an informed consent form. Blood was emptied carefully into hirudin tubes (Roche, Austria) containing 1/10 volume of recombinant hirudin. All volunteers refrained from taking any medication for the last seven days.

Freshly collected whole blood samples were used to test for in vitro aggregation. All measurements were performed within two hours after venepuncture. Multiplate® platelet function analyzer (Dynabyte GmbH, Munich, Germany) was used to measure plate aggregation. To perform the tests in parallel, all five channels of the device were used, as well as single-use test cells with duplicate impedance sensors. Each sensor's measured impedance change was recorded separately. Discardable polytetrafluoroethylene-coated magnetic stirrers were used to stir the sample-reagent mixtures. Three hundred microliter of hirudin blood was mixed with 300 μL pre-warmed isotonic saline solution containing vehicle (1% DMSO final concentration) or the tested compound (0, 0.1, 1, 10, or 100 μM final concentrations dissolved in 1% DMSO (Fisher BioReagents, USA, >99.7%)). Ticagrelor (Hikma Pharmaceuticals LLC, Jordan) was used as a positive control. Blood samples were incubated with the vehicle or test compound at 37 °C for five minutes before adding 20 μL of 0.2 mM ADP (HART, UK) and measuring platelet aggregation. The Multiplate® analyzer allows duplicate measurements of the sample with dual-electrode probes. The impedance change by the adhesion and aggregation of platelets on the electrode wires was continuously detected, and the mean values and area under the curve (AUC) were calculated. Each test was repeated in triplicate using blood from three to five different subjects. The percentage of inhibition was calculated as in eqn (7), and the IC₅₀'s were calculated accordingly.

7

Moreover, CY-QUANT VASP/P2Y₁₂ (BioCytex, France), an enzyme-linked immunosorbent assay, was used for the determination of serine 239-phosphorylated VASP (VASP-P) in platelets from fresh human whole blood. This test is used for the measurement of specific platelet ADP receptor (P2Y₁₂) antagonists. The test was performed according to manufacturer's directions. Platelet reactivity index (%PRI) was calculated using eqn (8).

8

wherein, %PRI, platelet reactivity index; OD, optical density; PGE1, prostaglandin E1; ADP, adenosine diphosphate.

Ethical approval statement

All experiments were performed in accordance with the Guidelines of Helsinki Declaration and were approved by the ethics committee at Al-Ahliyya Amman University (ethical approval number IRB: AAU/1/1/2021–2022). Informed consents were obtained from human participants of this study.

Author contributions

Conceptualization: Belal O. Al-Najjar; in vitro assessment: Manal A. Abbas, Obada A. Sibai and Aya Y. Al-Kabariti; resources: Belal O. Al-Najjar and Fadi G. Saqallah; writing-original draft preparation: Belal O. Al-Najjar, Manal A. Abbas and Fadi G. Saqallah; writing-review and editing: Fadi G. Saqallah; illustrations: Fadi G. Saqallah and Belal O. Al-Najjar; supervision: Belal O. Al-Najjar.

Conflicts of interest

The authors have no conflicts of interest to declare.