Journal of Chemical Information and Modeling Article (b) 22● Figure 3. a, Superimposition of the five major S-HTIAR conformations obtained from 100 ns MD simulation. Cluster-I representative structure is shown in white surface. Conserved residues in the binding pocket of Cluster -L, -Il,Ill,IV, and -V representative conformations are eparately in white, blue, green, yellow, and pink. b, Top view of the pharmacophore model mapped into the active pocket of s-HTIAR(only I representative structure was shown for clarity). Two hydrophobic elements(Hydro-l, -2), a positive ionizable element(POS), a hydrog acceptor element, and three excluded volumes(Excl.1,2,3)are represented in cyan, red, green, and black spheres, respectively. pocket(enclosed by ECL2 and the extracellular segments of Validation of a Pharmacophore Model. The Guner TM-3,-7)which is also typical among dopamine receptors. MD Henry (GH) scoring method"was used to assess the ability of simulation was started with the obtained R-8-OH-DPAT-5- our pharmacophore model to discriminate a small number of HTIAR complex embedded in a hydrated POPC lipid bilayer known active molecules against a greater number of decoy (Figure 2, left). During the simulation 8-OH-DPAT was molecules in the database. The decoy set contains a total of constantly trapped in the binding pocket of 5-HTIAR Taken 1000 compounds, including 25 highly active 5-HTIAR agonists together, our pi predicted 5-HT1AR model is credible, and taken from ChEMBL database, and 975 matched decoys conformations generated from MD simulation shall stand for generated from online DUD-E(Directory of Useful Decoys active states of the recept Enhanced)tool. The pharmacophore model was used to The Dynamic Pharmacophore Model Generation. Five screen the decoy set employing the BEST flexible searching representative structures obtained from clustering were used to method. A set of statistical parameters were set to analyze the represent the flexibility of s-HTIAR and to build the dynamic result (Table 1). 33 compounds were retrieved as hits with a hit pharmacophore model. GRID Probed the active site with five rate of 63.6%. In addition, the pharmacophore model showed types of functional groups(COO-, NI+, O, Nl, and DRY) to an enrichment factor of 25.45 and a GH score of 0.68, find the most favorable regions for them to interact with the indicating the good quality of our model protein, generating a total of 25 interaction maps(5 for each Database Searching and Bioassay Results. The protein conformation), and they were overlaid based on the validated pharmacophore model was then used as a 3D query superimposition of the five major S-HTIAR conformations to screen Maybridge and Specs databases, of which nondruglike (Figure 3a). Backbone atoms RMSD of representative compounds were rejected. A total of 18, 976 compounds respectively. On the basis of site-directed mutagenesis data and with GoldScore >45 were extracted for visual inspection. omputational studies available in the literature, #+344ie Compounds that form favorable interactions with key residues hydrogen bond contacts with D3. 32, Ser5. 42, T5.43, N7.39, in the active site such as D3. 32, SS. 42, F6.51, and F6.52 Y7.43 and T-t stacking interactions with Y2.64, w6.48, F6.51, chosen, and those with unreasonable binding modes F6.52, W7.40, Y7. 43 are crucial for S-HTIAR agonist binding discarded. At last, a total of 45 compounds, 16 from Specs only the features complementary to these key residues were 29 from Maybridge, were selected to purchase from onsidered. Generally, a positive ionizable feature(POS)was suppliers. The flowchart of hybrid searching approach was chosen to target the negatively charged carboxylic side chain of shown in Figure 4 D3. 32, the hydrogen-bond acceptor(HBA)feature located ge clusters found in Specs, May Bridge hydrophobic regions, which are surrounded by TMS/TM6 and 228401 Lipinski's Ro5& Veber's Rules TM2/TM7/ECL2, were combined into two hydrophobic 210,483 features(HYDRO-1,-2), respectively. Whereas, clusters of COO- probes were discarded due to the electrostatic repulsion between COO-and negatively charged residues in the binding D8-VS site of S-HTIAR(Figure 7), and the hydrogen-bond donor Visual inspection to SS.42/T5.43 was also omitted for it partially Bioassay overlapped with HYDRO-1. Thus, a four-feature pharmaco- phore model was produced( Figure 3b). Three exclusion volumes(Excl-1,-2, -3)were in situ generated from the side Figure 4. Workflow of dynamic pharmacophore-based virtual chains of the conserved residues D3. 32, F6.51, and Y7.43. searching approach. dx dolor/10.1021/c400481plJ Chem Inf Model. 2013, 53, 3202-3211pocket (enclosed by ECL2 and the extracellular segments of TM-3, -7) which is also typical among dopamine receptors. MD simulation was started with the obtained R-8-OH-DPAT-5- HT1AR complex embedded in a hydrated POPC lipid bilayer (Figure 2, left). During the simulation 8-OH-DPAT was constantly trapped in the binding pocket of 5-HT1AR. Taken together, our predicted 5-HT1AR model is credible, and conformations generated from MD simulation shall stand for active states of the receptor. The Dynamic Pharmacophore Model Generation. Five representative structures obtained from clustering were used to represent the flexibility of 5-HT1AR and to build the dynamic pharmacophore model. GRID probed the active site with five types of functional groups (COO-, N1+, O, N1, and DRY) to find the most favorable regions for them to interact with the protein, generating a total of 25 interaction maps (5 for each protein conformation), and they were overlaid based on the superimposition of the five major 5-HT1AR conformations (Figure 3a). Backbone atoms RMSD of representative structures in each Cluster (II, III, IV, V) to the major conformer (Cluster I) is 1.48 Å, 1.38 Å, 1.74 Å, 1.39 Å, respectively. On the basis of site-directed mutagenesis data and computational studies available in the literature,41,43,44 i.e. hydrogen bond contacts with D3.32, Ser5.42, T5.43, N7.39, Y7.43 and π−π stacking interactions with Y2.64, W6.48, F6.51, F6.52, W7.40, Y7.43 are crucial for 5-HT1AR agonist binding, only the features complementary to these key residues were considered. Generally, a positive ionizable feature (POS) was chosen to target the negatively charged carboxylic side chain of D3.32, the hydrogen-bond acceptor (HBA) feature located close to Y7.43 was retained, and large clusters found in two hydrophobic regions, which are surrounded by TM5/TM6 and TM2/TM7/ECL2, were combined into two hydrophobic features (HYDRO-1, -2), respectively. Whereas, clusters of COO- probes were discarded due to the electrostatic repulsion between COO- and negatively charged residues in the binding site of 5-HT1AR (Figure 7), and the hydrogen-bond donor feature adjacent to S5.42/T5.43 was also omitted for it partially overlapped with HYDRO-1. Thus, a four-feature pharmaco￾phore model was produced (Figure 3b). Three exclusion volumes (Excl-1, -2, -3) were in situ generated from the side chains of the conserved residues D3.32, F6.51, and Y7.43. Validation of a Pharmacophore Model. The Gü ner￾Henry (GH) scoring method45 was used to assess the ability of our pharmacophore model to discriminate a small number of known active molecules against a greater number of decoy molecules in the database. The decoy set contains a total of 1000 compounds, including 25 highly active 5-HT1AR agonists taken from ChEMBL database,46 and 975 matched decoys generated from online DUD-E (Directory of Useful Decoys, Enhanced) tool.47 The pharmacophore model was used to screen the decoy set employing the BEST flexible searching method. A set of statistical parameters were set to analyze the result (Table 1). 33 compounds were retrieved as hits with a hit rate of 63.6%. In addition, the pharmacophore model showed an enrichment factor of 25.45 and a GH score of 0.68, indicating the good quality of our model. Database Searching and Bioassay Results. The validated pharmacophore model was then used as a 3D query to screen Maybridge and Specs databases, of which nondruglike compounds were rejected. A total of 18,976 compounds obtained from PB-VS were docked into the representative structure of 5-HT1AR in the biggest cluster. 1500 compounds with GoldScore >45 were extracted for visual inspection. Compounds that form favorable interactions with key residues in the active site such as D3.32, S5.42, F6.51, and F6.52 were chosen, and those with unreasonable binding modes were discarded. At last, a total of 45 compounds, 16 from Specs and 29 from Maybridge, were selected to purchase from their suppliers. The flowchart of hybrid searching approach was shown in Figure 4. Figure 3. a, Superimposition of the five major 5-HT1AR conformations obtained from 100 ns MD simulation. Cluster-I representative structure is shown in white surface. Conserved residues in the binding pocket of Cluster -I, -II, -III, -IV, and -V representative conformations are colored separately in white, blue, green, yellow, and pink. b, Top view of the pharmacophore model mapped into the active pocket of 5-HT1AR (only Cluster￾I representative structure was shown for clarity). Two hydrophobic elements (Hydro-1, -2), a positive ionizable element (POS), a hydrogen bond acceptor element, and three excluded volumes (Excl-1, -2, -3) are represented in cyan, red, green, and black spheres, respectively. Figure 4. Workflow of dynamic pharmacophore-based virtual searching approach. Journal of Chemical Information and Modeling Article 3206 dx.doi.org/10.1021/ci400481p | J. Chem. Inf. Model. 2013, 53, 3202−3211