Homology Modeling of Dopamine D2 and D3 Receptors: Molecular Dynamics Refinement and Docking Evaluation

Dopamine (DA) receptors, a class of G-protein coupled receptors (GPCRs), have been targeted for drug development for the treatment of neurological, psychiatric and ocular disorders. The lack of structural information about GPCRs and their ligand complexes has prompted the development of homology models of these proteins aimed at structure-based drug design. Crystal structure of human dopamine D3 (hD3) receptor has been recently solved. Based on the hD3 receptor crystal structure we generated dopamine D2 and D3 receptor models and refined them with molecular dynamics (MD) protocol. Refined structures, obtained from the MD simulations in membrane environment, were subsequently used in molecular docking studies in order to investigate potential sites of interaction. The structure of hD3 and hD2L receptors was differentiated by means of MD simulations and D3 selective ligands were discriminated, in terms of binding energy, by docking calculation. Robust correlation of computed and experimental Ki was obtained for hD3 and hD2L receptor ligands. In conclusion, the present computational approach seems suitable to build and refine structure models of homologous dopamine receptors that may be of value for structure-based drug discovery of selective dopaminergic ligands.


Introduction
The dopaminergic systems in the central nervous system (CNS) have been extensively studied over the past 50 years [1]. Dopamine exerts its action through five distinct G-protein coupled receptors (D 1-5 receptors), grouped in two classes, D 1 -like and D 2like receptors, that differ in their signal transduction, binding profile and physiological effects [1]. D 1 -like receptors (D 1 and D 5 ) are principally coupled to stimulatory G s -proteins and enhance the activity of adenylyl cyclase (AC), whereas D 2 -like receptors (D 2 , D 3 , and D 4 ) are primarily coupled to inhibitory G i -proteins and suppress the activity of AC [1].
Alternative splicing of D 2 receptor mRNA leads to generation of two isoforms: D 2 short (D 2S ) and D 2 long (D 2L ), which have been associated (though not exclusively) with presynaptic and postsynaptic populations of D 2 receptors, respectively [2]. The difference between these two splicing isoforms is represented by 29 amino acid residues in the III intracellular loop (3ICL), involved in the G protein coupling. The D 2S is mainly considered as a presynaptic receptor, whereas, the D 2L as a postsynaptic receptor [2], like the D 3 [3]. However, it has been suggested that D 3 , in addition to the classical postsynaptic location, is also localized in the presynapse, where it modulates dopamine release and synthesis [4,5]. D 2 and D 3 receptors display a high degree of sequence homology and share the putative binding site for dopamine and synthetic ligands at the interface of transmembrane helices [6]. D 2 and D 3 receptors also share the signal-transduction mechanism, though under certain conditions the latter may exert a weaker stimulation of effectors like AC [7,8]. Several patholog-ical conditions such as schizophrenia, Parkinson's disease, Tourette's syndrome, and hyperprolactinemia have been linked to a dysregulation of dopaminergic transmission [1]. Furthermore, D 2 and D 3 receptor have been implicated as potential target for drug development in ocular diseases such as glaucoma [9,10,11,12,13,14,15]. D 2 -like receptors represent the most relevant class in the pathophysiology of neurological and psychiatric disorders. However, while D 2 receptor is considered the principal target to control the positive symptoms of schizophrenia, none of antipsychotics approved so far discriminates D 2 from D 3 receptors; on the other hand, the functional significance of D 4 receptor largely remains to be defined.
The crystal structure of hD 3 has been solved [31] and identified as a powerful tool for structure-based drug discovery of selective dopaminergic D 2 -like ligands [32]. This crystallized receptor is a hD 3 -lysozyme chimera, where the 3ICL is replaced by the lysozime protein; moreover, the receptor bears the mutation Leu119Trp in order to increase the thermal stability of the system. Recently, the determination of the crystal structure of hD 3 receptor and subsequent efforts in molecular modeling led to successful prediction of the pose of eticlopride in complex with a refined homology model of D 3 receptor [33]. Kortagere et al [34] analyzed in 2011 the binding mode of preferential D 3 ligands by means of site-directed mutagenesis and homology modeling studies (template structure 2RH1); these authors identified Ser 192 of V helix as an important site of interaction for the activation of D 3 receptor. Ser 192 belongs to a cluster of three serine residues (Ser 192,Ser 193,Ser 196); thus we have carefully looked at these residues, and their homologous (Ser 193,Ser 194,Ser 197) in hD 2L subtype, in our docking protocol. The subtype selectivity of D 2 -like ligands had been also studied before, by Wang et al [35], in the absence of structural information on D 3 and D 2 receptors, by a mixed structure-based (homology modeling using b 2 -adrenergic and rhodopsin receptors, molecular dynamics of haloperidolreceptor complexes) and ligand-based approach (3D-QSAR). These authors, however, did not carry out docking calculations. No study published so far has used a total structure-based approach for modeling ligand interactions with the hD 3 and hD 2L . In the present study we aimed at building and validating homology models of hD 3 and hD 2L receptors using the hD 3 receptor structure (3PBL) as template. Furthermore, in order to better discriminate their structural difference as well as selective ligands, we have carried out a structural optimization by molecular dynamics (MD) simulations of these two receptors for 3 ns in an explicit palmitoyl-oleoyl-phosphatidyl-choline (POPC) bilayer, that mimics the plasma membrane lipid environment, reaching a structural differentiation of these homologous receptors. The short-term MD simulations were adequate to obtain optimized structures of hD 3 and hD 2L receptors, because of the high homology and sequence identity between target and template proteins. We have validated these optimized structures using a total structure-based approach by molecular docking calculations that are extremely influenced by the reliability of receptor structure. The validation of optimized structure models was successful, giving good correlation between experimental and predicted K i of agonists.

Homology Modeling
The retrieved (Swiss-Prot) protein sequences of hD 3 and hD 2L receptors are respectively: P35462.2 and P14416.2. Homology models of hD 3 and hD 2L receptors were obtained by the Automated Modeling tool of Swiss Model web service http:// swissmodel.expasy.org/ [36,37] using the crystal structure of the human D 3 dopaminergic receptor-lysozyme chimera (Protein Data Bank-code 3PBL) in complex with the antagonist eticlopride as template. N-terminals of receptors were not modeled, because we focused on the binding pocket. Moreover the structure of Nterminal of hD3 was not solved by Chien et al [26]. The terminal residues Tyr 32 in hD3 and Tyr 37 in hD 2L were blocked in the homology models by acetylation. The hD 3 model was validated by docking eticlopride in the binding pocket. The model validation was carried out using two different molecular docking software (the docking protocol is reported in the Docking section): Autodock Vina (Vina) and Autodock 4.2 (AD4.2).

Molecular Dynamics
Homology models of dopaminergic receptors were embedded in a pre-equilibrated POPC bilayer. Then, the systems were hydrated with TIP3P water molecules, and neutralized adding NaCl up to 150 mM. CHARMM 27 parameters were assigned to all molecules. Disulfide bridges of hD 3 were parameterized by involving the following residues: Cys 103-Cys 181 connecting the III helix with the II extracellular loop (2ECL) and Cys 355-Cys 358 in the 3ECL. In the hD 2L model we parameterized the conserved disulfide bridge between the III helix and 2ECL involving the Cys 109-Cys 187 residues. The system preparation processes (building of bilayer, embedding of the proteins into the membrane, hydration and neutralization) were done using VMD v1.8.7 [38]. Before MD simulations the systems were equilibrated as follows: i) MD of lipid tails for 50 ps (time-step = 1 fs) while protein, water, ions and lipid head groups were kept fixed; ii) equilibration for 100 ps (time-step = 1fs) of water-ions-lipids, while proteins were kept fixed by applying harmonic constraints; iii) 500 ps (time step = 1 fs) of system equilibration, with no constraints applied to molecules. After the described steps of equilibration, 3 ns of MD simulation were carried out with time-step of 2 fs, collecting trajectory data every 10 ps. The SHAKE algorithm, which constraints the hydrogen-heavy atom bonds.was applied. Equilibration steps and simulations were carried out using NAMD v2.7 [39]. Langevin dynamics and piston were used to maintain constant temperature (300 K) and pressure (1 atm) during simulation. The area per lipid was maintained constant, after the equilibration steps (NPAT ensemble). The particle number of systems was 83242 for hD 3 -lipids-water-ions and 83429 for hD 2L in membrane. Periodic Boundary Conditions (PBC) and Particle Mesh Ewalds (PME) method [40] were used to treat long-term electrostatics (time-step of 4 fs). The cut-off at 10 Å was applied to Van der Waals and coulombic interactions and switching functions started at 9 Å . First stage minimization was performed using the steepest descent algorithm whereas the conjugate gradient was used during production runs.

Docking and Virtual Screening
We carried out two different molecular docking studies using Vina and AD4.2 software. Vina [41] is an accurate algorithm faster than AD4.2; for this reason it was used for docking calculation of a large group of D 2 -like ligands and for virtual screening study. AD 4.2 [42] provided the best prediction of pose of eticlopride in the hD 3 homology model, thus we have chosen it for accurate docking calculation such as prediction of K i of wellknown D 2 -like agonists docked into the refined homology models of hD 3 and hD 2L receptors. File preparation for AD4.2 docking calculations was carried out using the AutodockTool (ADT), a free graphics user interface (GUI) of MGL-tools.
The search space for all docking calculations included the orthosteric binding pocket individuated by eticlopride in 3PBL, the allosteric binding pocket reported by Chien et al [31] and the extracellular domain of receptors. An high exhaustiveness, 32, was used in Vina calculation because the search space applied to hD 3 and hD 2L receptor is relatively wide. In calculations carried out with AD4.2 we chose, as search algorithm, the time-consuming Lamarkian genetic algorithm (GA), that generated the best docking results for eticlopride in hD 3 homology model. Hundred iterations of GA with 2,500,000 energy evaluations per run were carried out. Population size was set to 150 and a maximum of 27,000 generations per run was carried out, followed by automatic clusterization of poses. Top scored (lowest energy) and more populated poses with orthosteric binding, as reported by Kortagere et al [34],were selected for analysis of ligand-protein interactions using the GUI ADT. AD 4.2 uses a semi-empirical free energy function and a charge-based method for desolvation contributes; the free energy function was calibrated using a set of 188 structurally known ligand-complexes with experimentally determined binding constants [43]. The binding energy of ligand poses (Kcal/mol) is the sum of intermolecular energy, internal energy of the ligand and torsional free energy minus the unboundsystem energy (see in Supporting Information S1 about the calculation of K i from AD4.2 binding energy values and Supporting Information S2 for ligand poses and optimized structure of receptors).

Homology Modeling
We built the homology models of hD 3 and hD 2L receptors. Two disulfide bridges were modeled in hD 3 receptor according to the crystal structure 3PBL [31], the canonical one that connect the 2ECL with the III helix and the disulfide bridge in the 3ECL involving residues Cys 355 and Cys 358. In hD 2L receptor only the conserved disulfide bridge was modeled, because we considered that a single residue of distance between the two conserved cysteine residues (Cys 399 and Cys 401) may lead to unstable disulfide bond. Validation for the hD 3 model, by docking eticlopride with Vina and AD4.2 was performed. Both software were able to reproduce the eticlopride conformation in the binding pocket; AD4.2 gave the lowest root mean square deviation (RMSD, 0.4 Å ) and better reproduced the internal H-bonds (Figure 2A), compared to VINA ( Figure 2B), that gave 0.6 Å RMSD for re-docked eticlopride. We have evaluated the similarity of hD 3 and hD 2L homology models by means of structural alignment. The tridimensional alignment revealed that the two homology models did not differ in transmembrane core structure ( Figure 3A), as expected from their high sequence identity; furthermore, RMSD between the two aligned GPCRs was very low (0.033 Å ). We have, further, analyzed the structural similarity and capacity of discrimination of active D 2 -like ligands by fast docking calculations, with the Vina docking software. The structure similarity was reflected by the high correlation (R 2 = 0.91, Figure 3C) of predicted binding energy of D 2 -like ligands docked into the homology models of hD 3 and hD 2L . Thus, these two homology models do not seem useful, without a structural refinement, for virtual screening directed at the recognition of selective ligands.

Molecular Dynamics
We have simulated for 3 ns the hD 3 and hD 2L homology models in a water-membrane environment that reproduces the biological milieu where these two GPCRs are located, to further discriminate their structural difference. By reporting the RMSD of protein structure from the starting homology model, both receptors differentiate in structure and reach a relative stable conformational minimum roughly after 1.25 ns (Figure 4). Total energy (E tot ) and potential energy (E p ) of systems are constant during the MD simulation (Supporting Information S1) and energy values of D 3 receptor are slightly lower compared to the energy of D 2L subtype. We stopped simulations at 3 ns because we reached stable local minima and distinct conformations for hD 3 and hD 2L receptors. Longer simulations (over 30 ns) might reveal other local minima and further characterize the conformational space of these receptors; this goal, however, is beyond the aim of our study. GPCRs are in equilibrium between active and inactive conformation, and, as far as the inactive conformation is concerned, a structural marker, the ''ionic lock'' was described in several studies [48,49,50,51] and was also revealed in the crystal structure of eticlopride-hD 3 complex (3PBL) [31]. This ionic lock involves, four conserved residues, Arg128-Asp127-Glu324-Tyr138 in hD 3 ( Figure 5A), and Arg132-Asp131-Glu368-Tyr142 in hD 2L receptor ( Figure 5B), respectively. The salt-bridges that constitute the ionic lock are retained during the 3 ns of simulation. We can assume that the conformation of receptors, that reached the relative minimum, describes the inactive state. The superimposition of the simulated hD 3 and hD 2L receptors confirmed the structural deviation of receptors in membrane, as the RMSD was 1.63 Å ( Figure 3B). The differentiation of the two homologous receptors was further strengthen by the lower correlation (R 2 = 0.74) of binding energies of D 2 -like ligands docked, with VINA, into hD 3 and hD 2L optimized structures ( Figure 3D). We have measured the Ca deviation of residues belonging to the orthosteric binding pocket of receptors in order to further characterize the structural modification of hD 3 and hD 2L induced by the membrane environment. The deviations of these residues, comparing the initial homology models with the refined structures are reported in Table 1. The residues of binding pocket of hD 2L receptor deviated from starting model more than residues of hD 3 subtype ( Table 1). The V helix of hD 2L receptor had the greater deviation than other helices after the simulation (Supporting Information S1), involving the extracellular and intracellular side (transversal to the plane of the membrane). The VI and VII helices deviated mostly in the extracellular side and the greater deviation is shown for the VII helix (Supporting Information S1). Within the seven helices of hD 2L receptor, only IV helix had a major  transversal deviation and a sensible deviation along the z-axis of membrane (Supporting Information S1). Furthermore, the binding pocket of hD 3 receptor was also remodeled in membrane, because there were major structural deviations involving the residues of V helix (Ser 192,Ser 193,Ser 196), VI helix (His 349) and VII helix (Tyr 375) (Tables 1 and Supporting Information S1). We further characterized the binding pocket of hD 3 and hD 2L , before and after refining with MD simulations, by using the web service fpocket http://fpocket.sourceforge.net/ [52]. Fpocket generates clusters of spheres to describe each pocket of a given protein; in Figure 6 we have assigned different colors to pockets of hD 3 and hD 2L receptors, before and after optimization. Before simulation in membrane, the binding pockets of the two receptors were very similar in shape and dimension. After simulation, the pocket of hD 3 became smaller than that of hD 2L and divided in three pockets ( Figure 6C); the one in blue includes the orthosteric and the allosteric pockets, the one in magenta is surrounded by the extracellular loops, and the deepest and smallest pocket is colored in red. In docking calculations, we did not find poses in the red pocket, that was occupied by water molecules during MD simulation (data not shown). The pocket of hD 2L after simulation became bigger than that of D 3 subtype ( Figure 6B and 6D). The hD 2L receptor after simulation shows a big pocket (orange spheres) and a smaller pocket (magenta) located along the big one, between the III and IV helices. After simulation the red pocket of hD 2L appears included within the orange one ( Figure 6B and 6D). The optimized structures of hD 3 and hD 2L used for analysis and docking calculations were extracted randomly from one of the last frames of simulations that characterize the relative conformational equilibrium, by considering as equivalent frames belonging to the same local minimum. To confirm this assumption we randomly selected one additional frame from each local minimum of the hD 3 and hD 2L MD simulations. These two additional frames resulted equivalent to the previous, because, when carrying out docking of pramipexole superimposable results were obtained both in terms of binding energy ( Table 2, values in brackets) and poses (data not shown). We did not carried out a clusterization of trajectories because we have reached one local minimum in each simulation. Furthermore, as reported by Yap et al [53] clusterization of GPCR trajectories, is not useful for selecting the representative structure to be used in docking calculation.

Docking
We validated the optimized structures of hD 3 and hD 2L receptors by docking D 3 -preferring receptor agonists into receptor binding pockets using AD 4.2 docking software, which provided the best result of eticlopride pose prediction in the hD 3 homology model. Binding energy of agonists docked in hD 3 and hD 2L receptors correlates with their higher affinity for the D 3 subtype (Table 2), consistent with more polar contacts of ligands docked into D 3 receptor compared to ligands docked into the D 2L subtype ( Table 3). The experimental pK i values (retrieved from http://  (Figure 7, see also Supporting Information S1) obtaining a good correlation as indicated by Pearson coefficients relative to hD 3 and hD 2L receptors equal to 0.88 and 0.83 respectively (p,0.005). Linear regression coefficients however were low (Figure 7), due to the limitations of AD4.2 in predicting absolute values of K i , as reported by Lape et al [54] and by Yap et al [55]. Another explanation to the mentioned issue might be related to the heterogeneity in K i determination assays. Quinpirole was not included in the regression analysis because it was an outlier, even though its predicted binding energies for hD 3 and hD 2L correlate with the higher affinity toward the D 3 subtype. Quinpirole is a bioisoster of DPAT, among other ligands included in the regression model (Figure 1), with a tricyclic structure where the hydroxyphenyl group is substituted with a pyrazolic group. On the contrary, PD-128907, a tricyclic compound with the hydroxyphenyl group, fits in the regression model of pK i for hD 3 and hD 2L receptor. Another tricyclic compound included in the regression model is cis-8-OH-PBZI (PBZI), which retains the position of hydroxyl and amine groups of 7-OH-DPAT. The affinity of PBZI was determined for D 2S , D 3 and D 4 receptors but not for D 2L receptor, therefore we did not include it in the regression model for hD 2L receptor. Recently, PBZI was found to not induce tolerance and slow response termination, in comparison to known agonists such as 7-OH-DPAT and pramipexole [56]. Comparing the tricyclic structures of PD-128907, PBZI and quinpirole, this latter might behave as an outlier in the chemical space, due to the substitution of the hydroxyphenyl moiety with the pyrazol condensed group.

Virtual Screening
Pramipexole is a selective D 3 agonist (D 2 /D 3 = 75.5) indicated in the treatment of early-stage Parkinson disease. This agonist was chosen as reference for building a small ligands database (89 molecules), where drug-like compounds are 70% similar to pramipexole. We carried out a virtual screening by docking these ligands into the refined hD 3 and hD 2L models. The top scored compound is a novel selective D 2 -like agonist synthesized by Ghosh et al [57] (-)-(S)-N6-Propyl-N6-(2-(4-(4-(pyridin-4-yl)phe-   (Table 4). ZINC45254546 (Figure 1) is an hybrid compound bearing a pramipexole moiety and a piperazin(4-phenyl(4pyridyl)) antioxidant group. This compound was re-docked with AD4.2, into hD 3 and hD 2L receptors. As shown in Figure 8, polar contacts involved aspartate and threonine residues in III helix and the cluster of serine residues in V helix that interact with the pramipexole group. The analysis of pose of ZINC45254546 did not show the H-bond with Asp114 in hD 2L , which may explain its lower affinity toward the D 2L subtype. The piperazin(4-phenyl(4pyridyl)) group interacted with part of the 2ECL in hD 3 subtype and with residues of II and VII helices in hD 2L receptor, that characterize the allosteric pocket. The top 30 compounds (ZINC-db code), docked into hD 3 and hD 2L receptors, are reported in Supporting Information S1.

Discussion
In the present study we have successfully modeled and optimized the structure of two high homologous GPCRs, the hD 3 and hD 2L receptors. The homology modeling is a powerful tool in the prediction of protein structure. The strength of this methodology is related to the sequence identity shared between the target and the template protein: the highest sequence identity determines the best structure model. We built and validated the homology models of hD 3 and hD 2L receptor using the x-ray structure of hD 3 receptor, a lysozyme-chimera protein. The high sequence identity shared by these two receptors did not allow us to differentiate their homology models that were therefore unsuitable for prediction of binding energies and subtype selectivity of D 2 -like ligands. The high structure similarity of hD 3 and hD 2L arises from the energy minimization process, and represents a weakness in the homology modeling approach. Usually, in homology modeling, the energy optimization of the modeled protein structure is performed by energy minimization in vacuo, with some exceptions such as the GPCRRD server http:/zhanglab.ccmb.umich.edu/  and physics-based atomic potentials (AMBER99 forcefield) [58,59]. So far protein-lipid and protein-water explicit interactions, based on empirical physics-based atomic potentials, are not taken into account by homology modeling software. Thus, we attempted to optimize the structure of the hD 3 and hD 2L models by MD in an explicit water-membrane environment, reaching a local conformational minimum within 3 ns. The MD simulations led to structural adaptation and differentiation of the two receptors in membrane, enabling the prediction of trends of pK i values and the modeling of ligand-protein interactions of D 3 -preferring receptor agonists. Moreover, the refined models were useful in the identification, by a virtual screening approach, of an agonist (ZINC45254546) referred to be selective for D 3 over D 2 [57]. Our results are consistent with the findings of Chien et al [26]; the hD 3 homology model we built was validated by docking eticlopride and by obtaining with AD 4.2 a pose highly similar to the one in the xray structure 3PBL. Because the ionic lock, a marker of inactive state described in 3PBL, was retained during MD simulations in both hD 3 and hD 2L receptors, we can assume that refined models represent an inactive state of the receptors. Moreover, we modeled both disulfide bridges solved in 3PBL in hD 3 model and we modeled just one disulfide bridge, the canonical one, in hD 2L . We made this choice because the conserved cysteine residues in the 3ECL, Cys 399 and Cys 401, are separated just by one residue Asp 400, leading to a high constrained loop in the case a disulfide bridge is formed. The lack of the accessory disulfide bridge in the 3ECL might have influenced the dynamics of hD 2L receptor, leading to the swelling of its binding pocket, in comparison to the hD 3 which is restrained by two disulfide bridges. Wang et al [60] have predicted the structural differences of hD 3 and hD 2 receptors. The homology models of these GPCRs were built in complex with haloperidol (previously aligned to the b 2 -adrenergic inverse agonist s-carazolol), using the crystal structure of b 2adrenergic receptor (2RH1); the complexes were subsequently simulated in a POPC bilayer for 1.5 ns. Haloperidol in complex with simulated D 3 and D 2 receptors was also used to carry out 3D-QSAR studies using 163 compounds. These authors [35] concluded that the higher affinity of bigger ligands for D 3 receptor over D 2 subtype is related to the shape of binding pocket, which is shallower in D 2 receptor. We found that the binding pocket of hD 3 receptor, after adapting in the membrane environment, significantly deviates from the initial homology model, becoming smaller and partitioned. The binding pocket of hD 3 in membrane environment is also smaller than the one of hD 2L receptor. We carried out docking calculations rather than 3D-QSAR (ligand-based method) because we considered our refined models highly predictive due to the crystal structure of hD 3 receptor, used as template for homology modeling. Docking calculations (structurebased method) are strictly related to the reliability of the receptor structure, and we obtained a good correlation of experimental and computed K i values for agonists docked into hD 3 and hD 2L binding sites. Although the prediction of absolute K i values is a difficult task, AD 4.2 was a powerful tool in order to validate homology model of hD 3 receptor (eticlopride re-docking) as well as to validate the refined models by MD simulations. In fact, the predicted trend of K i values is well correlated (high Pearson coefficients) with the experimental trend. This correlation was carried out with aminotetraline derivatives, a congeneric chemical class that does not include quinpirole. This latter is a preferential D 3 agonist, but behaved as an outlier in the chemical space of docked ligands, due to the tricyclic structure and the pyrazole moiety. Neverthless, our optimized models were able to predict the affinity of quinpirole higher for D 3 than for D 2L receptor. In conclusion, the computational approach, totally structure-based, adopted in the present study is able to build and refine structure models of homologous dopamine receptors that may be of interest for structure-based drug discovery of selective dopaminergic ligands, potentially useful to treat neurological, psychiatric and ocular disorders.

Supporting Information
Supporting Information S1 Figure S1: Energy plots of systems. Potential energy (E pot ) and total energy (E tot ), of hD 2L and hD 3 receptors. Table S1: Ca deviations of transmembrane helices (TM) of D 3 and D 2L simulated receptors from the starting models. Ca deviation values were determined by structural alignment of each helix of the model and of the optimized structure. Figure S2: Deviation of helices of optimized hD 2L receptor (cyan cartoon) respect the starting model (yellow cartoon). The upper side of the figure corresponds to the extracellular side. Table S2: Computed pK i for ligands docked into hD3 and hD2L receptors. Values are reported for ligands inserted in the regressions represented in Figure 7. Figure S3: Superimposition of template (3PBL)-homology model-optimized model of hD 3 receptor and hD 2L receptor. The template structure (green cartoon) is the A chain of hD 3 receptor crystal structure (3BPL). The cyan cartoon corresponds to the homology model of hD 3 receptor, the yellow cartoon corresponds to the homology model of hD 2L receptor. The optimized models of hD3 and hD2L receptor are respectively the magenta and orange cartoons.
Supporting Information S2 Supplemental files (.pdb files) contained in the compressed directory File S2 include poses of ligands, shown in Figure 1, docked into hD 3 and hD 2L optimized receptors, whose.pdb files are also included in File S2. All.pdb files can be visualized with Open Pymol. Files named ligand_D2.pdb correspond to poses of ligand docked into hD 2L receptor, whereas files named ligand_D3.pdb correspond to poses into hD 3 receptor. The optimized structure of hD3 and hD2L receptor are named respectively opt_D3_receptor.pdb and opt_D2L_receptor. (ZIP)