Dopamine transporter comparative molecular ... - Wiley Online Library

Download PDF
2 downloads 0 Views 771KB Size Report
Comment
Sep 10, 2007 - Pieper U, Eswar N, Stuart AC, Ilyin VA, Sali A. MODBASE, a data- base of ... Petersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM,.
proteins STRUCTURE O FUNCTION O BIOINFORMATICS

Dopamine transporter comparative molecular modeling and binding site prediction using the LeuTAa leucine transporter as a template Martı´n Indarte,1* Jeffry D. Madura,2* and Christopher K. Surratt1* 1 Division of Pharmaceutical Sciences, Duquesne University, Pittsburgh, Pennsylvania 15282 2 Department of Chemistry and Biochemistry, Center for Computational Sciences, Duquesne University, Pittsburgh, Pennsylvania 15282

ABSTRACT Pharmacological and behavioral studies indicate that binding of cocaine and the amphetamines by the dopamine transporter (DAT) protein is principally responsible for initiating the euphoria and addiction associated with these drugs. The lack of an X-ray crystal structure for the DAT or any other member of the neurotransmitter:sodium symporter (NSS) family has hindered understanding of psychostimulant recognition at the atomic level; structural information has been obtained largely from mutagenesis and biophysical studies. The recent publication of a crystal structure for the bacterial leucine transporter LeuTAa , a distantly related NSS family homolog, provides for the first time a template for three-dimensional comparative modeling of NSS proteins. A novel computational modeling approach using the capabilities of the Molecular Operating Environment program MOE 2005.06 in conjunction with other comparative modeling servers generated the LeuTAa-directed DAT model. Probable dopamine and amphetamine binding sites were identified within the DAT model using multiple docking approaches. Binding sites for the substrate ligands (dopamine and amphetamine) overlapped substantially with the analogous region of the LeuTAa crystal structure for the substrate leucine. The docking predictions implicated DAT side chains known to be critical for high affinity ligand binding and suggest novel mutagenesis targets in elucidating discrete substrate and inhibitor binding sites. The DAT model may guide DAT ligand QSAR studies, and rational design of novel DAT-binding therapeutics. Proteins 2008; 70:1033–1046.

C 2007 Wiley-Liss, Inc. V

Key words: homology; comparative modeling; docking; drug; pharmacophore; medication; therapeutic; cocaine; psychostimulant; antagonist; addiction.

C 2007 WILEY-LISS, INC. V

INTRODUCTION Addiction to cocaine, methamphetamine, and related psychostimulants destroys millions of individuals, families, and careers, a societal scourge worldwide. Although, addiction to heroin, oxycodone, fentanyl, and other opiates can be effectively treated with buprenorphine and to some extent methadone, no such medications are available to combat psychostimulant addiction despite decades of research. Not coincidentally, opioid receptor structure and mechanism of action are much better understood than those of the brain receptors for psychostimulant drugs of abuse, the monoamine neurotransmitter transporter proteins. Pharmacologic and behavioral studies indicate that the dopamine transporter (DAT) protein is the principal binding site responsible for cocaine’s reward and reinforcement properties.1,2 The plasma membrane-bound DAT protein quenches dopamine-mediated neurotransmission by clearing the neurotransmitter from the synaptic cleft following Ca21-mediated exocytosis from presynaptic vesicles. Cocaine, a DAT inhibitor, blocks synaptic uptake of dopamine; the resultant accumulation of the neurotransmitter in the synapse leads to an increase in postsynaptic dopamine D2 and D3 receptor activation in the nucleus accumbens and other brain regions associated with addiction. Indeed, activation of these accumbal dopamine receptors has been linked with the reinforcing properties of the drug.3,4 Amphetamine also increases synaptic dopamine levels, but by mediating dopamine efflux from the presynaptic cell via the DAT.5,6 Logically, an agent that blocks cocaine and amphetamine binding at the DAT without substantially interfering with dopamine uptake should serve as an effective antiaddiction therapeutic. High resolution elucidation of the DAT structure, especially regarding its substrate and inhibitor recognition sites, is thus critical.

The Supplementary Material referred to in this article can be found online at http://www. interscience.wiley.com/jpages/0887-3585/suppmat. Grant sponsor: NIDA; Grant number: DA016604; Grant sponsor: Samuel and Emma Winters; Grant sponsor: DOE; Grant numbers: P116Z040100, P116Z050331. *Correspondence to: Dr. Christopher K. Surratt, Division of Pharmaceutical Sciences, Duquesne University, Mellon Hall, Room 453, 600 Forbes Avenue, Pittsburgh, PA 15282. E-mail: [email protected] or Martin Indarte, E-mail: [email protected] or Jeffry D. Madura, E-mail: [email protected] Received 11 September 2006; Revised 16 March 2007; Accepted 16 April 2007 Published online 10 September 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/prot.21598

PROTEINS

1033

M. Indarte et al.

The DAT and other plasma membrane monoamine transporters are members of the 12 transmembrane domain (TM) neurotransmitter:sodium symporter (NSS) family,7 in which electrogenic transport of a neurotransmitter substrate across the cell membrane is driven by a Na1/K1-ATPase-generated Na1 gradient. Cotransport of Cl2 is also required for the dopamine, norepinephrine, and serotonin transporter proteins (DAT, NET, and SERT, respectively); the SERT additionally transports K1, but in antiport fashion.8 Aligning the amino acid sequences of the NSS family members guided delineation of monoamine transporter TM domain borders and other aspects of transporter secondary structure.9 Such a sequence alignment can also yield clues as to which NSS residues probably contribute to the general protein infrastructure, which residues could play a role in substrate or ion recognition, and which residues are most likely to be responsible for a pharmacologic pattern unique to a given transporter. This sequence information alone spawned hundreds of NSS site-directed and chimeric mutants.10 The substituted cysteine accessibility mutagenesis (SCAM) methodology has especially contributed to defining monoamine transporter ligand binding cavities, substrate/ion pores, general TM domain infrastructure, and even detection of substrate- or inhibitorinduced conformational changes.11–13 Nevertheless, this approach only circumstantially implicates a given residue or protein region as a component of the binding pocket. Unequivocally identifying direct contacts between transporter protein and ligand has proven to be difficult; the lack of an X-ray crystal structure for any protein homologous to the NSS family has been the major impediment. Encouragingly, the recently published crystal structure of a bacterial leucine transporter (LeuTAa), a protein homologous with the NSS family,14 finally provides a suitable NSS template. Using LeuTAa as a template, the present study describes a novel modeling approach that employs comparative modeling to produce a feasible three-dimensional (3D) DAT structure. Three approaches may be employed in predicting a 3D macromolecular structure: ab initio prediction, ‘‘fold’’ recognition, and comparative (homology) modeling.15 These differ principally in the sequence and structural database information used. A true ab initio method bases structure prediction entirely on the physical and chemical information contained in the primary amino acid sequence. However, the term is also used when short experimental protein sequences and secondary structure prediction methods are incorporated.16–20 Fold recognition, or ‘‘threading,’’ relies heavily on the structural similarities between certain distantly related or unrelated proteins. Comparative modeling predicts the 3D structure of a target protein based primarily on its alignment with one or more template proteins of known structure.21 For proteins that share greater than 40% amino acid sequence identity, comparative modeling is straightfor-

1034

PROTEINS

ward and typically accurate.22 For proteins with less than 30% amino acid sequence identity (e.g., LeuTAa and the DAT), comparative modeling becomes more challenging. Still, the rhodopsin crystal structure has successfully guided the creation of useful comparative models for many other members of the G protein coupled receptor superfamily despite the absence of appreciable amino acid sequence identity.23 Upon obtaining a 3D protein model, the conformations and orientations (denoted as poses) of ligands that couple with the macromolecule are computationally determined (‘‘docking’’).24–27 To find the most energetically favorable ligand pose within a structurally determined receptor, the macromolecule is typically held rigid whereas the ligands are flexible and mobile.28 Here, a docking procedure similar to the earlier uses of DOCK29 was used to identify potential DAT binding sites. This approach should reveal DAT amino acid residues likely to participate in substrate and inhibitor recognition and thus define targets for mutagenesis and other structurefunction studies. In this way, it is hoped that a blueprint can be developed for rational design of DAT-binding therapeutics.

MATERIALS AND METHODS Comparative modeling Robetta server sequence alignment and model building

The comparative modeling module of the Robetta server aligns the target and the template using K*Sync, a more accurate method than PSI-BLAST or Pcons2.30 K*Sync estimates the most reliable alignment of target and template based on secondary structure information, residue information obtained by comparing statistical representations of protein families (‘‘profile–profile’’ comparisons), and information from multiple structural alignments of regions with high structural propensity to fold. The peptide backbone is constructed taking into account the geometry between template(s) and target via multiple independent simulations; the lowest energy models are selected. Side chains of these models are repacked and conformational space explored using 100 independent Monte Carlo simulations, with a backbone dependent side chain rotamer library and a full atom energy function to select the lowest energy conformation of the comparative model.31–33 The FASTA sequence of the rat DAT protein (SwissProt locus SC6A3_RAT; accession number P23977; NCBI accession number AAB21099)34 was utilized as the query for the hybrid template-based/de novo method of the Robetta server (http://robetta.bakerlab.org). The bacterial (Aquifex aeolicus) leucine transporter protein LeuTAa was employed as the template (PDB, www.rcsb.org, accession number 2A65; MMDB accession no. 34395). Five models DOI 10.1002/prot

Comparative Model of the Dopamine Transporter

were retrieved from the server and separately saved in a database using the Molecular Operating Environment (MOE) 2005.06 program (Chemical Computing Group, Montreal, Canada).35 The all-atom forcefield AMBER99 was used to add hydrogen atoms and assign partial charges to all models.36 Relaxation of the newly added hydrogen atoms via several cycles of energy minimization were performed using a conjugated gradient/truncated Newton optimization algorithm to convergence criteria of 0.05 kcal/mol and a dielectric constant (e) of 3. All nonhydrogen atoms were held fixed during the energy minimization. Pro_check (MOE version), a scientific vector language (SVL) code based on Ramachandran plots and custom-written by the Chemical Computing Group, was used to detect unfavorable van der Waals contacts and abnormal covalent bonds in the models. The few steric clashes found were relaxed by manually selecting backbone and side chain atoms of the implicated amino acids and by performing successive steps of energy minimization until the steric clash was removed. All steric clashes were far from the putative ligand binding sites. A final refinement of side chains was carried out utilizing AMBER99 (convergence criteria 5 0.1 kcal/mol, e 5 3). Backbone atoms were held fixed during the procedure to find local minima for the side chains of the DAT macromolecule. The final DAT model (herein referred to as Model 1) was selected using the following criteria: (1) Maximal spatial overlap of backbones between the DAT models (targets) and LeuTAa (template). (2) Similarity of Verify 3D scores between target and template models with respect to TM domains.37,38 (3) Optimal profile of atom contacts and fewest abnormal covalent bonds as reported by Pro_check (MOE version). (4) Lowest potential energy, as calculated using MOE 2005.06. 3D-JIGSAW server sequence alignment and model building

3D-JIGSAW employs PSI-BLAST39 to generate a position specific scoring matrix (PSSM) for the template and target sequence. This PSSM data is used by the PSI-Pred program40 to predict secondary structures for both sequences. The PSSM data and secondary structures are used in a dynamic programming algorithm to perform an initial alignment. A second dynamic programming algorithm refines the initial alignment via multiple alignment of template structures.41 Target protein side chains are positioned based on those in the template and are also added from a side chain rotamer library when needed. Finally, a mean-field calculation is performed to select the most probable, best packed side chain rotamers.41 The rDAT FASTA sequence was used as the query for the 3D-JIGSAW server (www.bmm.icnet.uk/ 3djigsaw/). Sequence alignment and DAT homology modeling relative to the LeuTAa template were derived using both ‘‘interactive’’ and ‘‘automatic’’ modes. The DOI 10.1002/prot

DAT atomic coordinates for the comparative model were obtained after model building and selection by the metaserver of the most energetically favorable structure. This single model was downloaded, read by MOE 2005.06 and saved in a molecular database. Using MOE pro_check, the few steric clashes found were resolved by selecting backbone and side chain atoms of the implicated amino acid residues and performing successive steps of AMBER99 energy minimization (convergence criteria 5 0.1 kcal/ mol, e 5 3). After resolving unfavorable contacts, the protocol described earlier for the Robetta models was applied to calculate partial charges and optimize hydrogen atoms and side chains, yielding Model 2. Yamashita et al. alignment and MOE model building

The rDAT FASTA sequence and crystal structure coordinates of LeuTAa were loaded into MOE 2005.06. The primary amino acid sequences of LeuTAa and DAT were manually aligned using the MSA proposed by Yamashita et al.14 Because the initial partial geometry between template and target was not specified, only the backbone coordinates of LeuTAa were used for the model creation. A series of 10 DAT models were independently constructed with MOE using a Boltzmann-weighted randomized procedure42 combined with specialized logic for the proper handling of sequence insertions and deletions.43 Each intermediate model was evaluated by a residue packing quality function sensitive to the degrees to which nonpolar side chain groups are buried within the lipid bilayer and hydrogen bonding opportunities are maximized. Before the final refinement of side chains, a coarse minimization of backbone atoms using AMBER99 and a conjugated gradient method (convergence criterion 5 1.0 kcal/mol, e 5 3) was performed to improve packing and intramolecular interactions. No steric clashes were observed. The same protocol described for the Robetta models was applied to calculate partial charges and optimize hydrogen atoms and side chains. The optimal MOE model (Model 3) was selected using the criteria listed above for Model 1, and by weighting the best scores for side chain packing according to MOE’s packing evaluation function. The sequence alignments underpinning Models 1–3 are shown in Figures 1–3, respectively. Modeling of Na1 binding sites

Two sodium atoms were placed in the DAT models using the corresponding LeuTAa crystal coordinates; their positions were manually refined in order to preserve coordination bonds established with adjacent residues. The side chains of such residues were adjusted to emulate the LeuTAa environment using the rotamer explorer module in MOE 2005.06. Side chains were relaxed (with the two Na1 atoms and backbone positions fixed) using AMBER99 (convergence criteria 5 1.0 kcal/mol, e 5 3). PROTEINS

1035

M. Indarte et al.

Figure 1 Sequence alignment used to build DAT Model 1, based on LeuTAa crystal structure information retrieved by the Robetta protein prediction server. The 12 LeuTAa TM domains are highlighted. Gray blocks indicate level of sequence similarity. Tallest blocks: Residue is identical at that position. Intermediate blocks: Residues are nonidentical but relatively conservative with respect to their properties. Small blocks: Residues share mild conservation with respect to structure or function. The absence of a block indicates no appreciable structure/function conservation. Gaps in one sequence relative to the other are indicated by dashes. The 12 DAT TM domains are highlighted and contrasted by varying the color. The UCSF Chimera Visualization System was used to generate this figure.44 [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

DAT model ligand docking Construction and geometry optimization of DAT substrates and inhibitors

Three-dimensional models of the DAT substrates dopamine and amphetamine were constructed using the molecule builder feature of MOE 2005.06 (structures pictured in Fig. 4). Partial charges and hydrogen atoms were added to protonated and unprotonated molecules using the Merck Molecular Force Field 94X (MMFF94X), suita-

1036

PROTEINS

ble for small drug-like molecules.46–48 All structures were energy minimized using the conjugated gradient/ truncated Newton optimization algorithm with convergence criterion 5 0.05 kcal/mol, e 5 1. Binding site selection and exploration

The ‘‘alpha site finder’’ module of MOE 2005.06 was used to identify possible DAT ligand binding pockets within the newly-generated DAT models. Hydrophobic or DOI 10.1002/prot

Comparative Model of the Dopamine Transporter

Figure 2 Sequence alignment used to build DAT Model 2, based on LeuTAa crystal structure information retrieved by the 3D-JIGSAW protein prediction server. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

hydrophilic alpha spheres served as ‘‘probes’’ denoting zones of tight atomic packing. All probe clusters of alpha spheres not situated in cytoplasmic or phospholipidfacing regions were used to identify potential binding sites that were used in the docking simulations. These alpha spheres were used as centroids for the creation of dummy atoms used to define potential binding sites during the docking process. MOE-Dock 2005.06

A binding region is identified by a cluster of hydrophobic and hydrophilic alpha spheres; hydrophobic spheres mark hydrophobic environments, and hydrophilic DOI 10.1002/prot

spheres mark hydrophilic environments. Ligand atoms are matched to corresponding alpha spheres during the docking process. The alpha spheres are used to calculate shape complementarity of small molecules fitting into macromolecules, as well as binding affinities of these conformers. Docking methods that employ alpha spheres may generate bound conformations that approach crystallographic resolution.49 The ligand explores the conformational space to locate the most favorable binding orientation and conformation (denoted as a ‘‘pose’’)24–27 by aligning and matching all triangles of the template points with compatible geometry and chemistry; the protein atoms remain fixed during the process. For each ligand, 100 poses were generated and scored in an effort PROTEINS

1037

M. Indarte et al.

Figure 3 Sequence alignment as proposed by Yamashita et al.19 used to build DAT Model 3, based on the multiple sequence alignment of NSSE and LeuTAa. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

to determine favorable binding modes. An affinity scoring function, G, was employed to rank candidate poses. This pairwise atomic contact scoring methodology estimates the enthalpic contribution to the free energy of binding using the following linear function: G 5 Chb fhb 1 Cion fion 1 Cmlig fmlig 1 Chh fhh 1 Chp fhp 1 Caa faa Figure 4

The fx terms represent the fractional atomic contacts for a specific interaction, x. The Cx terms are coefficients that weight the interaction contribution of x to the affin-

1038

PROTEINS

MOE 2005.06-generated 2D ligand representation used in DAT model docking.45 Protonated and unprotonated species were used in the docking simulations.

DOI 10.1002/prot

Comparative Model of the Dopamine Transporter

ity score. The individual terms are: hb, hydrogen bond donor-acceptor pair interactions (an optimistic view is taken; for example, two hydroxyl groups are assumed to interact in the most favorable way); ion, ionic interactions (a Coulomb-like term is used to evaluate the interactions between charged groups); mlig, metal ligation interactions (those involving nitrogen and sulfur atoms and transition metals are so classified); hh, hydrophobic interactions; hp, interactions between hydrophobic and polar atoms; aa, an interaction between any two atoms. Two different placement methodologies for docking DAT substrates and inhibitors were used. The alpha triangle placement method generates poses by superposition of ligand atom triplets and triplet points within the receptor site. The triangle matcher method generates poses in a systematic and more accurate way than the alpha triangle placement method by aligning ligand triplets of atoms with triplets of alpha spheres in cavities of tight atomic packing. The docking process accounted for the two protonation states of the amine group of ligands. Poses from molecular databases of each ligand were scored based on complementarity with binding pocket alpha spheres. ASEDock

Alpha Sphere Based Protein-Ligand Docking (Ryoka Institutes), or ASEDock, is a novel fast-docking program written in the SVL language (MOE platform) and based on the alpha shape method. Ligand atoms have alpha spheres within 1 A˚. On the basis of this, concave shape models can be created, and ligand atoms from a large number of conformations generated by superposition with these points can be evaluated and scored by maximum overlap with the alpha spheres and minimum overlap (repulsion) with receptor atoms. The initial ligand conformations were subjected to energy minimization using the MMFF94S force field46 and when converged, reproduced experimentally bioactive conformations.49 The scoring function used by ASEDock is based on protein–ligand interaction energies. The interaction energy of a given conformation is calculated using the following formula: Utotal 5 Uele 1 Uvdw 1 Uligand 1 Usolv Uele and Uvdw represent electrostatic and van der Waals interactions, respectively, between the protein macromolecule and the ligand. Uligand represents conformation energy. Usolv represents the energy because of solvation. The lowest Utotal of the multiple poses generated were considered optimal poses. All alpha spheres not situated in cytoplasmic or phospholipid-facing regions of the DAT model were used as centroids for the creation of dummy atoms used to dock DAT ligands. The docking process took into account the two protonation states of ligand amine groups. Poses from the molecular databases for each ligand were ranked based on Utotal. DOI 10.1002/prot

For each ligand, 500 conformations were generated using the default systematic search parameters in the ASEDock module. Five thousand poses per conformation were randomly placed onto the alpha spheres located within the TM domains. From the resulting 500,000 poses, 200 poses with the lowest Utotal values were selected, and these poses were further optimized with the MMFF94S force field. During this refinement step, the ligand was free to move within the rigid binding pocket (the transporter atoms were held fixed). MOE-DOCK 2004.03 GA

A Monte Carlo simulated annealing process is used, allowing a sampling of the conformational space for the ligand and an extensive screening of all possible binding sites in a particular region of the target macromolecule.50–52 Docking interaction energy (Utotal) of a given conformation is estimated from a set of energy grids centered in the macromolecule binding site using the formula given above for ASEDock. Macromolecule protein coordinates remain fixed during the process, while the flexible, mobile ligand moves along the grid to locate the most favorable binding orientation and conformation based on the interaction energy. A docking box of 45 3 45 3 45 grid points was employed with grid spacing of 0.375 A˚. The alpha spheres generated in the TM domains by the site finder module were used as the centroids for the docking box. Once the docking region was defined, the alpha spheres were deleted (and not used in any subsequent calculations). Minimized ligands were randomly placed inside the docking box, and the docking process initiated with an iteration limit of 10,000, cycle number of 50, and run number of 100. The two protonation states of the ligand amine group were taken into account in the docking process. The final molecular database contained 100 docked poses for each ligand as well as all energy terms discussed earlier. Validation of the DAT ligand docking process via LeuTAa-leucine docking

The three docking methods described above were used to assess the validity of the DAT-ligand docking predictions by calculating possible bound conformations of leucine-LeuT complexes. The crystal structure of LeuT was retrieved from the PDB and prepared for docking: partial charges and hydrogen atoms were added, calculated and relaxed within the protein structure as described earlier. No further minimizations of side chains were carried out. LeuTAa,-leucine docking poses were obtained using MOE-Dock 2005.06, ASEDock or MOEDOCK 2004.03 GA, and compared to the original crystal structure. The RMSDs of leucine bound in the crystal versus the predicted bound leucine conformations for the different methods were calculated using db_crystal_rmsd, PROTEINS

1039

M. Indarte et al.

three sequence alignments show that the rDAT main insertions and deletions relative to LeuTAa occur in intracellular loop (IL) 1 (deletion) and the beginning of TM 3 as well as EL 2 (insertion). No insertions or deletions are observed in regions related to substrate binding sites. Proline and glycine residues are highly conserved between the DAT and LeuTAa in the first eight TM domains, suggesting that the nature of a-helix disturbances is similar between the proteins. Regarding TM 1b, the DAT model polypeptide backbones do not completely overlap with the LeuTAa template backbone, leading to subtle differences in side chain rotamer orientation, and bound substrate conformations. Docking of DAT substrates and inhibitors

Figure 5 Panel A: Backbone superposition of the three comparative DAT models. Extensive spatial overlap is observed between Models 1 (pink), 2 (blue), and 3 (yellow). Regions of greatest divergence corresponded to sequences outside of the TM domains; note the hinge that connects TM 9 and TM 10 (blue arrow). Panel B: ‘‘Zoom’’ view of TM 9 and TM 10 superposition from a different angle. For clarity, not all TM domains are depicted. Panel C: ‘‘Zoom’’ view of the superposition of the models with respect to TM 1, indicating Model 2 backbone spatial differences (green arrow).

Three docking algorithms were employed: MOE-Dock 2005.06, ASEDock and MOE-DOCK 2004.03 GA. Using leucine/LeuTAa docking as a test system, ASEDock yielded bound conformations with the lowest RMSD scores (0.24 A˚), followed by MOE-DOCK 2004.03 GA (0.49 A˚) and MOE-Dock 2005.06 (0.7 A˚). Even though ASEDock appears to be the best in reproducing a physiologically relevant leucine-LeuTAa pose, all three methods were used to elucidate potential DAT binding sites; their

a SVL code custom-written by the Chemical Computing Group. RESULTS Comparative models

The three 3D DAT models obtained are overall in good agreement with respect to spatial overlap, especially in the TM domains [Fig. 5(A)]. The most prominent points of divergence between the 3 models occur in the extracellular TM loop (EL 5) connecting TMs 9 and 10 [Fig. 5(B), blue arrow], and within TM 1b [Fig. 5(C), green arrow] of the 3D-JIGSAW model. Indeed, the 10 plausible conformations for the Yamashita et al. based alignment (Model 3) diverge at the hinge region connecting TMs 9 and 10 (data not shown). Models 1 (Robetta-based) and 2 (3D-JIGSAW-based) possess a similar sequence alignment (Figs. 1–3) that creates a similar profile of possible residues involved in ligand binding. Contrasting with Model 3, Models 1 and 2 overlapped well at the TM 9/10 hinge (Fig. 5), as did the corresponding loop of LeuTAa (not shown). This suggests that differences in the sequence alignments obtained from the Robetta and 3D JIGSAW servers relative to the Yamashita et al. alignment contributed to the divergence at DAT loop structures. Loop positioning may be a critical feature in the extracellular substrate recognition process. An incorrectly oriented loop could occlude and remove from consideration a putative ligand binding pocket in docking experiments. The

1040

PROTEINS

Figure 6 Spatial similarity of LeuTAa and DAT Model 1 substrate binding sites. Leucine (yellow, line depiction) is superposed on DAT Model 1 using the 2A65 X-ray coordinates. Energetically optimal conformations for dopamine (white, ball-andstick) and amphetamine (pink, ball-and-stick) predicted by MOE-Dock 2005.06 are pictured. For a given ligand, the result of each docking simulation is represented by a single chemical structure. The hinged a–helices TM 1 (salmon) and TM 6 (orange), as well as TM 3 (green), TM 8 (white), TM 10 (cyan), and TM 11 (gray), are highlighted. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

DOI 10.1002/prot

Comparative Model of the Dopamine Transporter

Figure 7 A representative predicted binding scenario for dopamine and amphetamine obtained with different models and docking algorithms using MOE. The poses represent the top-ranked DAT-ligand associations based on total interaction energy using ASEDock with Model 1 for dopamine (white, ball-and-stick) and amphetamine (pink, balland-stick). Leucine exported from the crystal structure (yellow, line depiction) demonstrates the spatial similarity of binding pockets between rDAT and LeuTAa and the considerable overlap of leucine and the two docked structures. Models 1 and 3 and the three different docking methods pose charged substrate amino groups (nitrogen atoms in blue) close to D79, generating the corresponding interaction in the form of hydrogen bonds. The H-bond network (cyan) depicts and describes the strength of the bond interaction based on ideal angles and distances, represented as the percentage of possible interaction strength between a given residue and ligand. Model 2 displays a slightly different docking scenario (data not shown), possibly due to the tilted backbone and different side chain locations predicted by ASEDock. The two sodium atoms (green spheres) do not directly interact with the substrates. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

RMSD scores are similar yet could potentially yield different poses. An important and attractive feature of the ASEDock method is indicated by a correlation plot of RMSD value versus interaction energy (Utotal), revealing that the lowest RMSD values correlate to the lowest interaction energies (see online supplementary material). The three DAT models, refined as described in the Methods section, were initially employed in MOE-Dock 2005.06 docking simulations with the DAT substrates dopamine and d-amphetamine. Given that the two substrates are close structural analogs, it is not surprising that these ligands were found to dock essentially in the same primary binding site of the DAT (Fig. 6). Considering that DAT docking of these substrates employed an unbiased approach, entirely independent of LeuTAa docking of its leucine substrate, the substantial overlap between the dopamine/amphetamine DAT site and the leucine site of the analogous region of the LeuTAa crystal structure is remarkable (Fig. 6). The coincidence of substrate binding pockets within the DAT and LeuTAa, proteins largely dissimilar in sequence that recognize strucDOI 10.1002/prot

turally dissimilar substrates, in part validates the present DAT models. Like the LeuTAa substrate binding pocket, the primary DAT substrate pocket is at the approximate midpoint of the lipid bilayer and very close to the two Na1 binding sites. Regardless of the protonation state, each substrate optimally fits in to the substrate binding site; however, protonation introduces a pronounced drop in interaction energy. A close-up view of the protonated substrates in the binding site (Fig. 7) shows extensive spatial overlap of predicted best poses despite the variety of docking methods and homology models employed. The models and bound conformations suggest that the charged amine groups of dopamine and amphetamine can create a network of hydrogen bonds with the amide backbone carbonyl groups of A77 and V78 (TM 1) and S320 and L321 (TM 6), as well as direct interactions with the carboxylate side chain of D79 (TM 1). The DAT models do not imply direct interactions between the Na1 atoms and substrate, consistent with the finding that dopamine binds to the DAT in the absence of Na1.53–55 The PROTEINS

1041

M. Indarte et al.

Figure 8 Ligand interaction plot of the MOE-Dock 2005.06-generated DAT amphetamine (left panel) and dopamine (right panel) binding pockets. This plot depicts the 2D (‘‘flattened’’) spatial arrangement of ligand and DAT protein with respect to key interactions. The proximity contour (dashed lines) and solvent exposed areas (solid purple spheres) of the ligand atoms are indicated, as are the polar (pink), hydrophobic (green), and solvent-exposed (light blue shadow) binding pocket amino acids. Acidic and basic residues are highlighted with red and blue halos, respectively. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

m-hydroxyl group of dopamine can form hydrogen bonds with the amide backbone carbonyl group of S421 and A422 (TM 8). The substrate aromatic moiety can establish favorable hydrophobic interactions with V152 (TM 3) and V327 (TM 6). More importantly, p–p stacking of this substrate group with the phenyl rings of Y156 (TM 3), F319 (TM 6), and F325 (TM 6) are possible (Fig. 7). Recently developed MOE 2006.07 software was used to create ligand interaction plots for charged dopamine and amphetamine (Fig. 8), providing a more visually-digestible arrangement of putatively key intermolecular interactions that aids in interpreting the 3D juxtaposition of ligand and transporter protein. Docking calculations that take into account alpha sphere position also yielded a secondary substrate pocket that affords binding of dopamine and amphetamine with fairly low interaction energies (Fig. 9). This broad secondary binding region is located at the extracellular interface and defined by TMs 1, 6, 10, and 11 and ELs 3, 4a, and 4b. The ligand-docked DAT models identify discrete DAT amino acid residues as putative contributors to the substrate pocket. A constellation of DAT residues can be derived in this way for each ligand, providing targets for site-directed mutagenesis and subsequent pharmacology toward high-resolution determination of drug binding sites.

DISCUSSION Comparative models of membrane-spanning proteins with amino acid sequence identity to the template of less

1042

PROTEINS

than 25% can have TM Ca-RMSD values above 3.0 A˚ relative to the native protein.56 Models displaying such variation between native and predicted conformations

Figure 9 Primary and secondary substrate binding sites predicted from DAT ligand docking to DAT Model 1. The primary substrate pocket is occupied by the topranked ASEDock poses for dopamine (white, ball-and-stick) and amphetamine (pink, ball-and-stick). Leucine, exported from its crystal structure (yellow, line depiction) is included to aid in identification of the primary site. Docked poses with optimal interaction energies for dopamine (white, stick) and amphetamine (pink, stick) delineate a docking-derived broad secondary substrate site. The hinged a–helices TM 1 (salmon) and TM 6 (orange), as well as TM 3 (green), TM 8 (white), TM 10 (cyan), TM 11 (gray), EL3 (red), EL4a (yellow), and EL4b (yellow), are highlighted because of their contribution to the docking calculations. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

DOI 10.1002/prot

Comparative Model of the Dopamine Transporter

can nevertheless be useful in implicating residues key for ligand binding. The low sequence similarity (20%) between the DAT and LeuTAa proteins represents a challenge for comparative modeling. Accurate sequence alignments are difficult to obtain with conventional algorithms. Poor alignments will yield 3D models with abnormalities such as extratransmembranous loops in TM regions, or an unacceptable number of hydrophilic or charged amino acids in otherwise hydrophobic TM domains. To surmount these hindrances and create feasible molecular models with predictive power, two different meta-servers (3D-JIGSAW and Robetta) were used to create reliable comparative models for the DAT. These metaservers improve upon conventional alignment algorithms by incorporating secondary structure prediction data, producing alignments with an accuracy that challenges the predictive skills of experts.57 Modeling TM domain proteins with 3D-JIGSAW alignments resulted in models with statistically significant lower RMSD values than other alignments.58 The Robetta server has also displayed excellent structure prediction capabilities, enabling creation of a 3D model for an entire protein sequence in the absence of significant sequence identity to a template protein of known 3D structure.31,32,59 Accurate alignments of membrane protein sequences are achieved by using template and target profiles (statistical representations of protein families)57,58 that include secondary structure information in the scoring of alignments, similar to those used by the 3D-JIGSAW and Robetta servers. Classically, docking calculations are used to predict optimal ligand–protein conformations and to perform virtual screening of compound databases in discovering therapeutic lead compounds.24,28,29,60,61 Generally, the docking calculation is performed when the location of a binding pocket is known or suspected. Several methods have been developed to find probable ligand binding regions.62–72 None, however, takes full advantage of pharmacologically-characterized ligands in selecting the most feasible binding site. In the present study, a multiple docking approach was employed in which probable DAT binding sites are examined thoroughly by performing docking calculations using pharmacologically wellcharacterized DAT ligands as molecular probes. The recently published crystal structure of the bacterial leucine transporter LeuTAa,14 a protein distantly but clearly homologous to the DAT, provided surprising revelations concerning the putative 3D structure of the 12 TM NSS family of transporters. In the LeuTAa—leucine cocrystal, the center of the TM 1 and TM 6 a–helices is unwound, exposing main chain carbonyl oxygen and nitrogen atoms that H-bond to the leucine substrate and one of the two Na1 ions required for transport. These unwound regions also apparently serve as hinges involved in interconversion between outward- and inward-facing transporter conformations. The DAT comparative models DOI 10.1002/prot

obtained in the present work also largely adopt the LeuTAa 3D conformation, including centrally located disruptions in the TM 1 and 6 helices. Like LeuTAa, the DAT model suggests that TMs 3 and 8 combine with TMs 1 and 6 to form the substrate binding pocket. Using the modeling tool SCWRL3,73 the same four TM domains were found to form the substrate binding site of a 3D SERT model.74 Several NSS structure-function studies support direct contributions of TMs 1 and 3 to substrate recognition.10–12,75–83 TMs 6 and 8 had not been shown to directly contact the substrate prior to the LeuTAa crystallization, although TMs 3, 7, and 8 of the DAT were able to coordinately bind a Zn21 ion, indicating close proximity.13 Despite the different protein sequences and cognate substrate molecules, all ligand docking algorithms located an optimal binding pocket for dopamine (and its analog amphetamine) within the DAT model that was almost superimposable with that for leucine in the LeuTAa crystal (Figs. 6 and 7). DAT residues in TM 1 (e.g., F76 and D79) and TM 3 (e.g., V152) are within reach of the docked substrates (Figs. 7 and 8), and the pharmacology of engineered NSS mutants at these positions is consistent with the DAT model’s substrate binding site. The rDAT F76A mutation dramatically affected dopamine apparent affinity.84 Mutations of the identical position in the hSERT (Y95) and GAT-4 GABA (E61) transporters suggest a direct interaction with substrates.75,80 The DAT model indicates that the D79 residue, within the unwound region of TM 1, may directly interact with substrates, its carboxylate coordinating with the substrate amino groups of the best ranked poses regardless of the different docking methods. Cases have been made for85,86 and against81,82 a salt bridge forming between the DAT D79 or the analogous D98 SERT carboxylate and the substrate amine. In the present DAT model, the D79 carboxylate colocalizes with the carboxylic group of the leucine substrate in the LeuTAa crystal, arguing against an ionic bond with the substrate amino group. D79 is simultaneously able to establish a direct interaction with one of two Na1 sites in the model, the role played by the backbone carbonyl group of the analogous LeuTAa residue, G24.14 Accordingly, a D98 SERT mutant was compromised in its ability to utilize Na1 during substrate transport.85 In some fashion, D79 appears to be contributing electrostatic interactions that enhance substrate recognition. Finally, the DAT model suggests favorable interactions between the TM 3 V152 side chain and either the aromatic ring or the lipophilic hydrocarbon portion of both dopamine and amphetamine. This residue is critical to DAT substrate transport.79 The analogous SERT residue is found to be on the ligand-accessible face of TM 3, in or near the serotonin binding site.77,78 Two of the ligand docking approaches used with the DAT model, ASEDock and MOE-Dock 2005.06, yielded a consensus secondary substrate binding pocket distinct PROTEINS

1043

M. Indarte et al.

from the primary substrate pocket (Fig. 9). It is conceivable that this secondary site is a temporary ‘‘waiting room’’ for the substrate, and the substrate is ushered to its primary binding pocket in the presence of Na1, Cl2, or the appropriate outward-facing DAT conformation. The secondary site may be important for recognition of cocaine and other dopamine uptake inhibitors (unpublished data). Amphetamine and perhaps other uptake inhibitors may directly compete with dopamine for occupancy of this secondary pocket if the current conformation resembles a bioactive conformation able to recognize and bind inhibitors. The DAT models remained fixed during the docking process and therefore conformational flexibility of the macromolecule upon binding was not addressed. This should be noted as a limitation of the approach given that the three DAT models are based on the LeuTAa structure in only one configuration—the transporter with a tightly bound substrate. Moreover, the modeling effort may well miss potential ligand binding sites or overstate minor sites that would be recognized as such if more flexibility were introduced in the model. Despite the considerable progress achieved in the past few years, accurate docking methods that predict macromolecular conformational changes upon ligand binding still remain computationally challenging.24 This novel DAT model will continue to provide new DAT mutagenesis targets. The pharmacology from these mutants will in turn refine the DAT model, affording high resolution mapping of DAT substrate and inhibitor binding sites. At that point, the DAT model may be used for QSAR analysis of putative DAT ligands, involving in silico screening of structural libraries containing millions of compounds. The more promising compounds would be screened at the bench, and then in preclinical and clinical settings. In this way, rational design of novel DAT pharmacotherapeutic ligands should be possible. Such ligands may interfere with actions of abused psychostimulants including cocaine and the amphetamines while largely sparing normal DAT function. Novel medications for treating depression, anxiety disorders, attention deficit hyperactivity disorder, narcolepsy, Parkinson’s disease, and other DAT-related disorders may also result from rational drug design afforded by this DAT model. CONCLUSION Using the LeuTAa crystal structure as a template, three comparative modeling approaches were used to create three DAT models. Although quite similar, the nonidentical sequence alignments led to subtle but significant differences between the models. Three docking methods were applied to the three DAT models to identify potential binding sites for the substrates dopamine and the psychostimulant d-amphetamine. The docking calculations identified two discrete DAT binding regions: a pri-

1044

PROTEINS

mary substrate binding site correlating with the binding site observed in the LeuTAa crystal structure, and a broad secondary substrate site closer to the extracellular interface. The secondary site may act as a potential staging area for substrate translocation through the cell membrane. The proposed binding pockets and their function are consistent with published and unpublished mutagenesis data. The DAT models coupled with ligand docking simulations are refining mutagenesis and other structurefunction investigations, and should aid in the development of QSAR as well as pharmacophore models toward development of novel medications. ACKNOWLEDGMENTS Chemical Computing Group is acknowledged for providing MOE software, especially for access to a beta release version of MOE. M.I. thanks the technical support scientists at CCG, especially Dr. Suzanne Schreyer, Dr. Alain Deschenes and Dr. Andrew Henry for their assistance. Dr. Barry Honig is thanked for helpful comments and discussions. Dr. Junichi Goto is thanked for granting the Ryoka Institute docking program ASEDock. REFERENCES 1. Giros B, Jaber M, Jones SR, Wightman RM, Caron MG. Hyperlocomotion and indifference to cocaine and amphetamine in mice lacking the dopamine transporter. Nature 1996;379:606–612. 2. Ritz MC, Lamb RJ, Goldberg SR, Kuhar MJ. Cocaine receptors on dopamine transporters are related to self-administration of cocaine. Science 1987;237:1219–1223. 3. Caine SB, Koob GF. Modulation of cocaine self-administration in the rat through D-3 dopamine receptors. Science 1993;260:1814– 1816. 4. Caine SB, Koob GF. Pretreatment with the dopamine agonist 7OH-DPAT shifts the cocaine self-administration dose-effect function to the left under different schedules in the rat. Behav Pharmacol 1995;6:333–347. 5. Fischer JF, Cho AK. Chemical release of dopamine from striatal homogenates: evidence for an exchange diffusion model. J Pharmacol Exp Ther 1979;208:203–209. 6. Sitte HH, Farhan H, Javitch JA. Sodium-dependent neurotransmitter transporters: oligomerization as a determinant of transporter function and trafficking. Mol Interv 2004;4:38–47. 7. Saier MH, Jr. A functional-phylogenetic system for the classification of transport proteins. J Cell Biochem 1999; Suppl 32/33:84–94. 8. Rudnick G. Mechanisms of biogenic amine neurotransmitter transporters. In: Reith MEA, editor. Neurotransmitter transporters: structure, function, and regulation. Totowa, NJ: Humana Press; 1997. pp 73–100. 9. Goldberg NR, Beuming T, Soyer OS, Goldstein RA, Weinstein H, Javitch JA. Probing conformational changes in neurotransmitter transporters: a structural context. Eur J Pharmacol 2003;479:3–12. 10. Surratt CK, Ukairo OT, Ramanujapuram S. Recognition of psychostimulants, antidepressants, and other inhibitors of synaptic neurotransmitter uptake by the plasma membrane monoamine transporters. AAPS J 2005;7:E739–E751. 11. Henry LK, Adkins EM, Han Q, Blakely RD. Serotonin and cocainesensitive inactivation of human serotonin transporters by methanethiosulfonates targeted to transmembrane domain I. J Biol Chem 2003;278:37052–37063.

DOI 10.1002/prot

Comparative Model of the Dopamine Transporter

12. Henry LK, Field JR, Adkins EM, Parnas ML, Vaughan RA, Zou MF, Newman AH, Blakely RD. Tyr-95 and Ile-172 in transmembrane segments 1 and 3 of human serotonin transporters interact to establish high affinity recognition of antidepressants. J Biol Chem 2006;281:2012–2023. 13. Loland CJ, Granas C, Javitch JA, Gether U. Identification of intracellular residues in the dopamine transporter critical for regulation of transporter conformation and cocaine binding. J Biol Chem 2004;279:3228–3238. 14. Yamashita A, Singh SK, Kawate T, Jin Y, Gouaux E. Crystal structure of a bacterial homologue of Na1/Cl2-dependent neurotransmitter transporters. Nature 2005;437:215–223. 15. Petrey D, Honig B. Protein structure prediction: inroads to biology. Mol Cell 2005;20:811–819. 16. Bonneau R, Tsai J, Ruczinski I, Chivian D, Rohl C, Strauss CE, Baker D. Rosetta in CASP4: progress in ab initio protein structure prediction. Proteins 2001; Suppl 5:119–126. 17. Bradley P, Chivian D, Meiler J, Misura KM, Rohl CA, Schief WR, Wedemeyer WJ, Schueler-Furman O, Murphy P, Schonbrun J, Strauss CE, Baker D. Rosetta predictions in CASP5: successes, failures, and prospects for complete automation. Proteins 2003;53 (Suppl 6):457–468. 18. Bradley P, Malmstrom L, Qian B, Schonbrun J, Chivian D, Kim DE, Meiler J, Misura KM, Baker D. Free modeling with Rosetta in CASP6. Proteins 2005;61 (Suppl 7):128–134. 19. Misura KM, Chivian D, Rohl CA, Kim DE, Baker D. Physically realistic homology models built with ROSETTA can be more accurate than their templates. Proc Natl Acad Sci USA 2006;103:5361– 5366. 20. Rohl CA, Strauss CE, Chivian D, Baker D. Modeling structurally variable regions in homologous proteins with rosetta. Proteins 2004;55:656–677. 21. Esposito EX, Tobi D, Madura JD. Comparative protein modeling. In: Lipkowitz KB, editor. Reviews in computational chemistry, Vol. 22. Hoboken, NJ: Wiley; 2005. pp 57–167. 22. Pieper U, Eswar N, Stuart AC, Ilyin VA, Sali A. MODBASE, a database of annotated comparative protein structure models. Nucleic Acids Res 2002;30:255–259. 23. Visiers I, Ballesteros JA, Weinstein H. Three-dimensional representations of G protein-coupled receptor structures and mechanisms. Methods Enzymol 2002;343:329–371. 24. Brooijmans N, Kuntz ID. Molecular recognition and docking algorithms. Annu Rev Biophys Biomol Struct 2003;32:335–373. 25. Geschwend DA, Good AC, Kuntz ID. Molecular docking towards drug discovery. J Mol Recognit 1996;9:175–186. 26. Zhou Z, Fisher D, Spidel J, Greenfield J, Patson B, Fazal A, Wigal C, Moe OA, Madura JD. Kinetic and docking studies of the interaction of quinones with the quinone reductase active site. Biochemistry 2003;42:1985–1994. 27. Zhou Z, Madrid M, Madura JD. Docking of non-nucleoside inhibitors: neotripterifordin and its derivatives to HIV-1 reverse transcriptase. Proteins 2002;49:529–542. 28. Perola E, Walters WP, Charifson PS. A detailed comparison of current docking and scoring methods on systems of pharmaceutical relevance. Proteins 2004;56:235–249. 29. Kuntz ID, Blaney JM, Oatley SJ, Langridge R, Ferrin TE. A geometric approach to macromolecule-ligand interactions. J Mol Biol 1982;161:269–288. 30. Al-Lazikani B, Jung J, Xiang Z, Honig B. Protein structure prediction. Curr Opin Chem Biol 2001;5:51–56. 31. Chivian D, Kim DE, Malmstrom L, Bradley P, Robertson T, Murphy P, Strauss CE, Bonneau R, Rohl CA, Baker D. Automated prediction of CASP-5 structures using the Robetta server. Proteins 2003;53 (Suppl 6):524–533. 32. Chivian D, Kim DE, Malmstrom L, Schonbrun J, Rohl CA, Baker D. Prediction of CASP6 structures using automated Robetta protocols. Proteins 2005;61 (Suppl 7):157–166.

DOI 10.1002/prot

33. Kim DE, Chivian D, Baker D. Protein structure prediction and analysis using the Robetta server. Nucleic Acids Res 2004;32: W526–W531. 34. Kilty JE, Lorang D, Amara SG. Cloning and expression of a cocaine-sensitive rat dopamine transporter. Science 1991;254:578– 579. 35. Chemical Computing Group C. Molecular Operative Enviroment (MOE), 2006.0706. 1255 University St., Suite 1600, Montreal, Quebec, Canada, H3B 3x3; 2006. 36. Ponder JW, Case DA. Force fields for protein simulations. Adv Protein Chem 2003;66:27–85. 37. Eisenberg D, Luthy R, Bowie JU. VERIFY3D: assessment of protein models with three-dimensional profiles. Methods Enzymol 1997; 277:396–404. 38. Luthy R, Bowie JU, Eisenberg D. Assessment of protein models with three-dimensional profiles. Nature 1992;356:83–85. 39. Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ. Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res 1997;25:3389– 3402. 40. Jones DT. Protein secondary structure prediction based on position-specific scoring matrices. J Mol Biol 1999;292:195–202. 41. Bates PA, Kelley LA, MacCallum RM, Sternberg MJ. Enhancement of protein modeling by human intervention in applying the automatic programs 3D-JIGSAW and 3D-PSSM. Proteins 2001; Suppl 5:39–46. 42. Levitt M. Accurate modeling of protein conformation by automatic segment matching. J Mol Biol 1992;226:507–533. 43. Fechteler T, Dengler U, Schomburg D. Prediction of protein threedimensional structures in insertion and deletion regions: a procedure for searching data bases of representative protein fragments using geometric scoring criteria. J Mol Biol 1995;253:114–131. 44. Petersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, Ferrin TE. UCSF Chimera—a visualization system for exploratory research and analysis. J Comput Chem 2004;25:1605– 1612. 45. Clark AM, Labute P, Santavy M. 2D structure depiction. J Chem Inf Model 2006;46:1107–1123. 46. Halgren TA. Merck molecular force field. I. Basis, form, scope, parameterization, and performance of MMFF94. J Comput Chem 1996;17:490–519. 47. Halgren TA. Force fields: MMFF94. In: Schleyer PVR, editor. Encyclopedia of computational chemistry, Vol. 2. West Sussex, UK: Wiley; 1998. p 1033. 48. Maple JR. Force fields: a general discussion. In: Schleyer PVR, editor. Encyclopedia of computational chemistry, Vol. 2. West Sussex, UK: Wiley; 1998. p 1015. 49. Goto J, Kataoka R, Hirayama N. Ph4Dock: pharmacophore-based protein-ligand docking. J Med Chem 2004;47:6804–6811. 50. Morris GM, Goodsell DS, Huey R, Olson AJ. Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4. J Comput Aided Mol Des 1996;10:293–304. 51. Goodsell DS, Morris GM, Olson AJ. Automated docking of flexible ligands: applications of AutoDock. J Mol Recognit 1996;9:1–5. 52. Hart TN, Read RJ. A multiple-start Monte Carlo docking method. Proteins 1992;13:206–222. 53. McElvain JS, Schenk JO. A multisubstrate mechanism of striatal dopamine uptake and its inhibition by cocaine. Biochem Pharmacol 1992;43:2189–2199. 54. Chen N, Sun L, Reith ME. Cationic interactions at the human dopamine transporter reveal binding conformations for dopamine distinguishable from those for the cocaine analog 2 a-carbomethoxy3a-(4-fluorophenyl)tropane. J Neurochem 2002;81:1383–1393. 55. Li LB, Cui XN, Reith MA. Is Na(1) required for the binding of dopamine, amphetamine, tyramine, and octopamine to the human dopamine transporter? Naunyn Schmiedebergs Arch Pharmacol 2002;365:303–311. PROTEINS

1045

M. Indarte et al.

56. Forrest LR, Tang CL, Honig B. On the accuracy of homology modeling and sequence alignment methods applied to membrane proteins. Biophys J 2006;91:508–517. 57. Rychlewski L, Fischer D. LiveBench-8: the large-scale, continuous assessment of automated protein structure prediction. Protein Sci 2005;14:240–245. 58. Reddy Ch S, Vijayasarathy K, Srinivas E, Sastry GM, Sastry GN. Homology modeling of membrane proteins: a critical assessment. Comput Biol Chem 2006;30:120–126. 59. Tai CH, Lee WJ, Vincent JJ, Lee B. Evaluation of domain prediction in CASP6. Proteins 2005;61 (Suppl 7):183–192. 60. Chen IJ, Neamati N, MacKerell AD, Jr. Structure-based inhibitor design targeting HIV-1 integrase. Curr Drug Targets Infect Disord 2002;2:217–234. 61. Hancock CN, Macias A, Lee EK, Yu SY, Mackerell AD, Jr, Shapiro P. Identification of novel extracellular signal-regulated kinase docking domain inhibitors. J Med Chem 2005;48:4586–4595. 62. Del Carpio CA, Takahashi Y, Sasaki S. A new approach to the automatic identification of candidates for ligand receptor sites in proteins: (I). Search for pocket regions. J Mol Graph 1993;11:23–29. 63. Edelsbrunner H, Facello M, Liang J. On the definition and the construction of pockets in macromolecules. Pac Symp Biocomput 1996:272–287. 64. Edelsbrunner H, Koehl P. The weighted-volume derivative of a space-filling diagram. Proc Natl Acad Sci USA 2003;100:2203–2208. 65. Goodford PJ. A computational procedure for determining energetically favorable binding sites on biologically important macromolecules. J Med Chem 1985;28:849–857. 66. Hendlich M, Rippmann F, Barnickel G. LIGSITE: automatic and efficient detection of potential small molecule-binding sites in proteins. J Mol Graph Model 1997;15:359–363. 67. Liang J, Edelsbrunner H, Fu P, Sudhakar PV, Subramaniam S. Analytical shape computation of macromolecules. I. Molecular area and volume through alpha shape. Proteins 1998;33:1–17. 68. Liang J, Edelsbrunner H, Woodward C. Anatomy of protein pockets and cavities: measurement of binding site geometry and implications for ligand design. Protein Sci 1998;7:1884–1897. 69. Miranker A, Karplus M. Functionality maps of binding sites: a multiple copy simultaneous search method. Proteins 1991;11:29–34. 70. Peters KP, Fauck J, Frommel C. The automatic search for ligand binding sites in proteins of known three-dimensional structure using only geometric criteria. J Mol Biol 1996;256:201–213. 71. Reynolds CA, Wade RC, Goodford PJ. Identifying targets for bioreductive agents: using GRID to predict selective binding regions of proteins. J Mol Graph 1989;7:103–108. 72. Wade RC, Clark KJ, Goodford PJ. Further development of hydrogen bond functions for use in determining energetically favorable binding sites on molecules of known structure. I. Ligand probe groups with the ability to form two hydrogen bonds. J Med Chem 1993;36:140–147. 73. Canutescu AA, Shelenkov AA, Dunbrack RL, Jr. A graph-theory algorithm for rapid protein side-chain prediction. Protein Sci 2003; 12:2001–2014.

1046

PROTEINS

74. Henry LK, Defelice LJ, Blakely RD. Getting the message across: a recent transporter structure shows the way. Neuron 2006;49:791– 796. 75. Adkins EM, Barker EL, Blakely RD. Interactions of tryptamine derivatives with serotonin transporter species variants implicate transmembrane domain I in substrate recognition. Mol Pharmacol 2001;59:514–523. 76. Barker EL, Perlman MA, Adkins EM, Houlihan WJ, Pristupa ZB, Niznik HB, Blakely RD. High affinity recognition of serotonin transporter antagonists defined by species-scanning mutagenesis. J Biol Chem 1998;273:19459–19468. 77. Chen J-G, Rudnick G. Permeation and gating residues in serotonin transporter. Proc Natl Acad Sci USA 2000;97:1044–1049. 78. Chen J-G, Sachpatzidis A, Rudnick G. The third transmembrane domain of the serotonin transporter contains residues associated with substrate and cocaine binding. J Biol Chem 1997;272:28321– 28327. 79. Lee SH, Chang MY, Lee KH, Park BS, Lee YS, Chin HR. Importance of valine at position 152 for the substrate transport and 2bcarbomethoxy-3b-(4-fluorophenyl)tropane binding of dopamine transporter. Mol Pharmacol 2000;57:883–889. 80. Melamed N, Kanner BI. Transmembrane domains I and II of the g-aminobutyric acid transporter GAT-4 contain molecular determinants of substrate specificity. Mol Pharmacol 2004;65:1452– 1461. 81. Ukairo OT, Bondi CD, Newman AH, Kulkarni SS, Kozikowski AP, Pan S, Surratt CK. Recognition of benztropine by the dopamine transporter (DAT) differs from that of the classical dopamine uptake inhibitors cocaine, methylphenidate, and mazindol as a function of a DAT transmembrane 1 aspartic acid residue. J Pharmacol Exp Ther 2005;314:575–583. 82. Wang W, Sonders MS, Ukairo OT, Scott H, Kloetzel MK, Surratt CK. Dissociation of high-affinity cocaine analog binding and dopamine uptake inhibition at the dopamine transporter. Mol Pharmacol 2003;64:430–439. 83. Zomot E, Kanner BI. The interaction of the g-aminobutyric acid transporter GAT-1 with the neurotransmitter is selectively impaired by sulfhydryl modification of a conformationally sensitive cysteine residue engineered into extracellular loop IV. J Biol Chem 2003; 278:42950–42958. 84. Lin Z, Wang W, Kopajtic T, Revay RS, Uhl GR. Dopamine transporter: transmembrane phenylalanine mutations can selectively influence dopamine uptake and cocaine analog recognition. Mol Pharmacol 1999;56:434–447. 85. Barker EL, Moore KR, Rakhshan F, Blakely RD. Transmembrane domain I contributes to the permeation pathway for serotonin and ions in the serotonin transporter. J Neurosci 1999;19:4705– 4717. 86. Kitayama S, Shimada S, Xu H, Markham L, Donovan DM, Uhl GR. Dopamine transporter site-directed mutations differentially alter substrate transport and cocaine binding. Proc Natl Acad Sci USA 1992;89:7782–7785.

DOI 10.1002/prot