Guidelines for Homology Modeling of Dopamine, Norepinephrine, and

Sep 6, 2016 - The human dopamine, norepinephrine, and serotonin transporters ... serotonin-selective reuptake inhibitors (SSRIs) have shown promise in...
0 downloads 0 Views 1MB Size
Research Article pubs.acs.org/chemneuro

Guidelines for Homology Modeling of Dopamine, Norepinephrine, and Serotonin Transporters Yazan Haddad,†,‡ Zbynek Heger,†,‡ and Vojtech Adam*,†,‡ †

Department of Chemistry and Biochemistry, Mendel University in Brno, Zemedelska 1, CZ-613 00 Brno, Czech Republic Central European Institute of Technology, Brno University of Technology, Purkynova 123, CZ-612 00 Brno, Czech Republic



S Supporting Information *

ABSTRACT: The human dopamine, norepinephrine, and serotonin transporters (hDAT, hNET, and hSERT) are carriers of neurotransmitters and targets for many drugs. Pioneering works in the past three years to elucidate experimental models of the Drosophila dDAT and human hSERT structures will rapidly impact the field of neuroscience. Here, we evaluated automated homology-based human models of these transporters, employing systematic physics-based, knowledge-based, and empirical-based check. Modeling guidelines were conveyed with attention to the central binding site (S1), secondary binding site (S2), and the extracellular loops EL2 and EL4. Application of new experimental models (dDAT and hSERT) will improve the accuracy of homology models, previously utilizing prokaryotic leucine transporter (LeuT) structure, and provide better predictions of ligand interactions, which is required for understanding of cellular mechanisms and for development of novel therapeutics. KEYWORDS: Homology-based, template-based, protein structure, dopamine transporter, norepinephrine transporter, serotonin transporter

T

splicing variants in the intracellular C-terminus.1 The human serotonin transporter (hSERT) facilitates release and recycling of serotonin (also known as 5-hydroxytryptamine).8 The first known crystal structure of a homologue transporter, elucidated by Yamashita et al. in 2005,9 was the model of prokaryotic leucine transporter (LeuT) from Aquifex aeolicus with X-ray resolution of 1.65 Å. This model played key role in understanding the conformational changes occurring in a transporter during binding/release and inhibition cycles.1 In addition, several mutant, engineered LeuT structures (LeuBAT) binding antidepressants were published.10 These structures provide experimental insight into the binding sites in monoamine transporters and identify a number of nonconserved residues as the major functional controllers of the binding sites for ligands and inhibitors. Inhibitors of hDAT, hNET, and hSERT take a different entry pathway from their exit path resulting in different rates of binding and dissociation. The entry path passes through nonconserved residues of the fourth extracellular loop (EL4), the secondary binding site (S2), and the central binding site (S1), respectively. Systematic mutational analysis revealed that six diverging amino acids in the central binding site (S1) are responsible for the selectivity of hNET inhibitor binding but not hDAT.11 The role of the

he solute carrier 6 (SLC6) gene family, also known as the neurotransmitter sodium symporter family, comprise a group of nine plasma membrane transporters for the monoamine neurotransmitters (dopamine, norepinephrine, and serotonin) and the amino acid neurotransmitters (GABA and glycine).1 The monoamine neurotransmitters are synthesized as 12-transmembrane-domain glycoproteins. N-Glycosylation on the second extracellular loop (EL2) is essential for transporter assembly and surface expression and might vary according to the type of tissue.2,3 The human dopamine transporter (hDAT) regulates uptake and efflux of dopamine into and from neurons. Similar to the rest of monoamine transporters, it is a major target for various pharmacologically active drugs and environmental toxins.4 The translocation cycle of dopamine is accompanied by cotransport of two sodium ions and one chloride ion, while hDAT needs to be reoriented or “returned” to the outward-facing (OF) state.5 According to Wang et al.,6 the central binding site in hDAT, in close proximity to bound sodium and chloride ions, undergoes at least three distinct conformational changes according to three different classes of ligands and inhibitors. The human norepinephrine transporter (hNET) mediates neurotransmission by clearing and recycling the released norepinephrine (also known as noradrenaline) to terminate its signals at noradrenergic synapses. The transporter activity and trafficking is regulated intracellularly through protein complexes and possibly through a phosphorylationbased mechanism.7 hNET is also known to have two alternative © 2016 American Chemical Society

Received: August 13, 2016 Accepted: September 6, 2016 Published: September 6, 2016 1607

DOI: 10.1021/acschemneuro.6b00242 ACS Chem. Neurosci. 2016, 7, 1607−1613

Research Article

ACS Chemical Neuroscience

map or a wrong molecular replacement solution in extreme cases. In minor cases, errors may arise from incorrect peptide orientation or misplacing of excessive water molecules. Inaccuracy might be due to limited resolution and poorly phased diffraction data. However, it is also important to note that outliers in experimental models are not necessarily errors.16 To understand the major differences in the experimental models studied here, and prior to construction of homologybased models, we performed both overall and segmented fitting of the dDAT (PDB ID 4M48) and hSERT (PDB ID 5I6Z) Xray models. Overall fitting of Cα atoms might be biased if a significant bending in the midst of the model would shift the other half of structure. To reduce the complexity of analysis, the two models were also divided into segments of ∼25−50 residues. A step by step alignment and iterative fitting of all atoms in dDAT and hSERT segments were performed, followed by visual evaluation of backbone and side-chain differences. Results are summarized in Supplementary Table S1 and Supplementary Figure S1. Insights from this analysis are added to the discussion where appropriate. The study of sequence differences, that is, nonconserved domains, and their corresponding functions is a great challenge for homology-based modeling. We believe one of the major errors that arise in homology-based modeling occur in the early stage of construction, that is, during the alignment phase. Hence, a bad alignment can result in wrong backbone structure. Studies show that the accuracy in prediction of the side-chains deteriorates when the backbone structure is not very accurate.17 In this study, homology-based models of hDAT, hNET, and hSERT using two template structures of Drosophila (dDAT) were automatically constructed by SWISS-MODEL server and further evaluated. Aside from the variable glycosylated EL2 region, we show here only three incidents of sequence “shifts” from the dDAT sequence that can drastically affect structure alignment (Supplementary Figure S1; Figure 1a). With reference to dDAT, one shared deletion region in hNET and hDAT and two insertion regions in hSERT were identified. The alignment shift 1 at dDAT’s G100−R101dDAT pair region resulted in one deleted residue in hDAT and hNET (at K133hDAT and K129hNET, respectively), while it was substituted by the R152/K153hSERT pair in hSERT. All homology models of hDAT and hNET represented these shifts accurately in the alignment phase. This shift leads to backbone difference in nearly 8 residues between experimental dDAT and hSERT models (Supplementary Figure S1) and occurs in the domain facing the cytoplasmic side in hSERT, while in dDAT the R101dDAT is covered by adjacent stacking W597dDAT (Figure 1b). On the other hand, hSERT shows two alignment shifts resulting from insertions at hSERT’s A401hSERT (shift 2) and W458hSERT (shift 3) locations. Both homology models of hSERT represented these shifts accurately in the alignment phase. The A401hSERT region is very critical and overlaps the S2 and EL4 binding sites. This insertion causes backbone shift upstream of the alanine residue (Figure 1c). The W458hSERT location (shift 3 at the intracellular loop IL4) is more complex and more divergent in hSERT, possibly resulting from insertion of W458/A459hSERT pair and deletion of one residue between R461hSERT and R462hSERT. The insertion of W458hSERT in hSERT breaks the IL4 helix displayed in dDAT but allows for stacking with L451hSERT and F465hSERT (Figure 1d). The evaluation of physical properties focuses mainly on problems resulting from protein structure outliers and steric clashes. A thorough all-atom contact analysis by MolProbity18

EL4 in neurotransmitter transport was highlighted by Rannversson et al.,12 when they showed that a mutation at L406hSERT disrupted the conformational changes by favoring an OF conformation in hSERT. Recently, Coleman et al.13 published two crystallized mutant models of hSERT, designated ts2 and ts3. However, the race for elucidating crystal structures of hDAT and hNET is still running. Researchers rely greatly on homology-based modeling of these transporters using the closest known template, which is now the Drosophila dopamine transporter (dDAT),6,14 instead of prokaryotic LeuT.9 The launch of homology-based structural modeling servers has cleared the path to future high-throughput study of human protein−protein interactions and large scale drug library screening. Qu et al.15 describe four steps in template-based structure modeling. First, a parent structure is identified by a search based on sequence similarity. Second, alignment is an important step to identify conserved and variable regions. Third, several models are constructed and refined through backbone moves and side-chain packing with attention to variable regions that can be predicted through loop modeling. Finally, evaluation of the structures is used to find the nearest model to fit experimental data. In this study, automated homology-based models of hDAT, hNET, and hSERT were constructed based on two Drosophila dDAT experimental models (PDB IDs 4M48 and 4XPA).6,14 The models were designated by the names hDAT−4M48, hDAT−4XPA, hNET−4M48, hNET−4XPA, hSERT−4M48, and hSERT−4XPA. General and deep analysis was used to evaluate the quality of the homology-based structural models: (1) the physics-based approach investigates the fitting of model to theoretical physical properties such as bond lengths and angles and protein backbone and side-chain conformations, in addition to molecular mechanics (force field) that shows energetic favorability of structure, (2) the knowledge-based approach uses statistical potential to compare the energetics of the model with previous crystal structures in the protein data bank (PDB), and (3) the empirical approach is based on fitting a new model to substantial experimental data, for example, superimposition (fitting) of the new model onto an experimental model for comparison. Data from related X-ray or nuclear magnetic resonance (NMR) studies are the most useful and direct, whereas data from functional mutant studies or ligand and substrate binding or from secondary structure characterization can also contribute to the evaluation process.



RESULTS AND DISCUSSION Understanding and utilizing the molecular structures of dopamine, norepinephrine, and serotonin transporters (DAT, NET, and SERT) is important for studying neurotransmitters and development of therapeutics. Recent pioneering work to elucidate experimental models of the Drosophila DAT and human SERT structures will rapidly affect the field of neuroscience. The Drosophila dopamine transporter (dDAT) shares pronounced sequence similarity with human hNET (63.3%), hDAT (62.3%), and hSERT (60.1%), and also its sequence identity is mostly shared with human hNET (49.6%), hDAT (46.5%), and hSERT (44.1%). When the structure is constructed by homology-based modeling from the same sequence template (in this case two dDAT experimental models), it is expected that the same quality problems will be inherited as well. The errors in the experimental model can arise from incorrect tracing of the backbone in the experimental 1608

DOI: 10.1021/acschemneuro.6b00242 ACS Chem. Neurosci. 2016, 7, 1607−1613

Research Article

ACS Chemical Neuroscience

residues. The Cβ atom deviations are another parameter, complementing Cα-based Ramachandran, for evaluation of the backbone angles.20 Cβ deviation is calculated from “ideal Cβ atom position” to the Cα atom based on information from angles and dihedrals. While the Cα atom carries the side chain and geometrically controls φ and ψ torsional angles, the C atom coordinates the Cα and nitrogen via the ψ and ω torsional angles. Due to the planer nature of the peptide bond, the ω angle is restricted to 180° (in trans case). Since proline is branched from both Cα and N atoms, it is likely that the preceded residue (preproline) is most susceptible to these deviations. On the other hand, glycine lacks Cβ torsions, and thus has more flexible φ and ψ torsional angles. The number of Cβ outliers was approximately 7−13 per homology-based model. Among the important backbone outliers, there are two residues in the S2 and EL4 regions (Supplementary Table S3). The D385/P387hDAT in hDAT−4M48 and equivalent D400/ P403hSERT in hSERT−4M48 homology models show backbone problems. In the first case, the problem arose from a fatal modeling error where the EL2 backbone passes through a proline ring, while in the second case; the proline residue torsion angle was unrealistic and can be fixed manually or by directed energy minimization. (B) Side-chain outliers. The number of side-chains rotamer outliers in all homology-based models correlated with rotamer outliers in template structures of dDAT (Table 1). An interesting case was found in the S1/S2 shared binding sites in Y151hNET from hNET−4XPA and Y175hSERT from hSERT−4XPA homology-based models, which was an inherited problem from the equivalent rotamer in 4XPA dDAT model. In such case, the problem is ignored to keep in line with experimental data. (C) All atom clashes. As expected, the number of atom−atom clashes in homology-based model was very high and correlated directly with bond lengths and bond angles. In general, clash scores in models based on 4M48 dDAT were higher than those in models based on 4XPA dDAT template although the latter had worse clash scores in the experimental model (Table 1). However, the number of atom− atom clashes was relatively similar. The clash score represents the number of unusual atom−atom contacts per 1000 atoms. The MolProbity score is directly calculated from balanced ratios for logarithms of clash score, rotamer outliers, and

Figure 1. Regions of major alignment shifts. Residues of dDAT are in green and residues of hSERT are in red: (a) dDAT model showing shift regions (in red) at residues 100−101 (shift 1), 158−213 (EL2), 384−385 (shift 2), and 440−441 (shift 3). (b) Shift 1, showing dDAT ribbon with R101 residue hidden by W597. (c) Shift 2, fitting of dDAT and hSERT showing the change in backbone at the inserted A401 residue. (d) Shift 3, hSERT ribbon with broken helix at W458; however the tryptophan residue is stabilized by L451 and F465 residues.

provides a good quality checking method to identify the physical anomalies. Details of physical properties evaluation are shown in Supplementary Table S2 and S3, while detailed list of backbone outliers for hNET and hDAT is summarized in modeling guidelines (Table 2). (A) Backbone outliers. Few Ramachandran outliers occurred in each homology model (Table 1), and most of them were related to proline residues. Due to the nature of this unique cyclic amino acid, proline’s φ angle is always locked at approximately −65°;19 therefore, errors in proline conformation can extend to neighboring Table 1. Physics and Knowledge-Based Analysis physical properties backbone model dDAT (4M48)a dDAT (4XPA)a hDAT−4M48 hDAT−4XPA hNET−4M48 hNET−4XPA hSERT (5I6Z)a hSERT− 4M48 hSERT− 4XPA a

side-chain

knowledge-based force field energy

all-atoms

statistical potential

electrostatic

total energy

QMEAN4 total score

QMEAN4 Zscore

−12969

−11015

−17347

0.608

−1.85

1.78

−14157

−10433

−17706

0.620

−1.69

135.76 93.95 84.50 81.26 7.05

2.97 2.81 3.03 2.72 1.93

20328 −330 −2596 −4864 b

−8237 −8510 −8677 −8742 −9591

33677 −77 −2482 −5435 b

0.522 0.565 0.556 0.592 0.567

−2.79 −2.29 −2.37 −1.95 −2.25

11

97.06

3.17

−2783

−10414

−5004

0.570

−2.23

6

87.91

2.89

−4673

−10439

−7175

0.557

−2.39

ramachandran outliers

Cβ outliers

rotamer outliers

0

0

10

0

0

6 4 3 5 0

clash score

molprobity score

nonbonded

3.64

1.78

6

7.64

8 7 11 7 0

5 3 10 3 9

11

13

5

8

Experimental models. bNot applicable. 1609

DOI: 10.1021/acschemneuro.6b00242 ACS Chem. Neurosci. 2016, 7, 1607−1613

Research Article

ACS Chemical Neuroscience Table 2. Guidelines for dDAT-Based Homology Modeling of Monoamine Transporters hDAT and hNET no.

guidelines

1. 1.1. a. b. c. 1.2. a. b. 1.3. a. b. 2. 2.1. a. b. 2.2. a. b. c. 3. a. b. c. d. 4. a. b. a

backbone modeling alignment shifts shift at dDAT’s G100−R101dDAT pair region results in one deleted residue in hDAT and hNET (at K133hDAT and K129hNET, respectively) alignment shifts in hSERT’s A401hSERT and W458hSERT are not affected in hDAT and hNET alignment differences in the EL2 region Ramachandran outliersa hDAT outliers: C135, P194, D196, L204, T207, T210, T211, P597 hNET outliers: K61, C131, P188, H199, K204, K206, P594 Cβ outliersa hDAT outliers: I134, N188, D196, S197, L204, N205, D206, T207, F208, T210, A214, P387, D476, A594 hNET outliers: Q54, A105, I130, K189, L190, L191, N192, N198, S203, K204, Y205, K206, T208, L238, V374, D378, P594 side-chain modeling rotamers that can be adapted from hSERT instead of dDATb hDAT residues with identity to hSERT and not dDAT are 54 amino acids hNET residues with identity to hSERT and not dDAT are 53 amino acids binding sites information from previous homology24,26,28 and experimental structures6,9,13,14,29 conserved residues in binding sites described by Koldso et al.10 have nearly the same side-chain conformations except for F319dDAT/ F320hDAT/F317hNET nonconserved residues in binding site described by Anderson et al.11 control the selectiveness of transporters and require careful modeling when comparing hDAT/hNET additional concerns one disulfide bridge in EL2 at least two sodium ion binding sites near S1 site one zinc ion binding site in EL2 atomic clashes must be resolved by minimizing the structure validation docking of known ligands/inhibitors confirmation of new findings by experiment

Problems mostly related to modeling near proline residues. bResidues described in Supplementary Figure 1.

found to show problems in planarities using MolProbity. F472hDAT residue is equivalent to F471dDAT/V489hSERT, which is one of three residues facing the inner S2 binding site that are changed from aromatic in dDAT to aliphatic in hSERT (Supplementary Table S1). F470dDAT/V488hSERT, F471dDAT/ V489hSERT, and H472dDAT/K490hSERT have some distance between them that might allow for vibrational motion. Knowledge-based evaluation by statistical potential is an energy-based method that applies qualitative model energy analysis (QMEAN)22 to compare a tested model with the available experimental structures in the database. QMEAN4 is calculated by linear combination of four statistical energy potentials; both (a) Cβ interaction energy and (b) all atoms pairwise energy are distance-based potentials, whereas (c) tortion angle energy analyzes local geometry over three successive residues, and (d) solvation energy analyzes the burial status of each amino acid. The scale for QMEAN is between zero for low quality and one for high quality structure. The SWISS-MODEL23 server already provides QMEAN information for constructed homology models. At first glance, we suspected that resulting low QMEAN scores were due to the unpredictable (i.e. difficult to model) extracellular loop (EL2), which has low similarity to the template used. To our surprise, the deletion of nonconserved EL2 had little influence on overall statistical potential scores with the exception of the all-atom pairwise energy, which had affected value but not Zscore (Supplementary Table S6). Z-scores are used to represent quality of structures assuming the average for high quality experimental models has zero Z-score. Thus, negative Z-scores

complement of Ramachandran favored percentage. A structure with zero clash score, zero rotamer outliers, and 100% Ramachandran favored will result in an ideal MolProbity score of one. A second approach for evaluation of physical properties is the evaluation of force field energy via molecular mechanics. Previously, researchers have been interested in studying the change in free energy upon interactions between molecules.21 However, the free energy of the molecule itself in vacuo can be used to evaluate unstable and unfavorable conformations (identified by high positive value). Since the free energy is calculated in residue by residue fashion, we believe this approach to be as insightful as the previous physical method. Force field energy includes covalent bond parameters and noncovalent parameters;21 the latter, which is not included in all-atom contact analysis (e.g., MolProbity), provides a new advantage for this approach. The noncovalent parameters include nonbonded (van der Waals) and electrostatic (ionic and hydrogen bonds). Since the electrostatic parameter is very biased to the content of charged residues in a protein, the nonbonded free energy provides more insight into the unfavorable interactions in a protein. The total free energy of hDAT−4M48 was most significantly unfavorable (Table 1 and Supplementary Table S4), while due to missing atoms in the hSERT experimental model, it was not applicable to compare all free energy accurately. Nevertheless, two significantly unfavorable residues were identified: the aforementioned P387hDAT in the S2/EL4 sites and also F472hDAT in the S2 site (Supplementary Table S5). The F472hDAT residue was 1610

DOI: 10.1021/acschemneuro.6b00242 ACS Chem. Neurosci. 2016, 7, 1607−1613

Research Article

ACS Chemical Neuroscience

several residues in 303−308 have missing atoms in dDAT 4M48 experimental structure. We have compiled guidelines for homology modeling of monoamine transporters based on recently published structures of dDAT and hSERT (Table 2). At least four research groups previously developed validated homology-based models of the monoamine transporters. The first hDAT homology model by Beuming et al.24 was used to predict the binding site of cocaine and cocaine analogs and was found to be overlapping with binding sites of dopamine, amphetamine, and DAT inhibitors. The model was constructed using Modeller from LeuT (PDB ID 2A65), and the work was validated through mutagenesis study and trapping of radiolabeled cocaine analog in the DAT. An alternative hDAT homology model by Huang et al.25 revealed a second site for binding of cocaine that does not overlap the dopamine binding site and can explain the different modes of transport inhibition, that is, inhibition of initial binding and reducing the turnover following binding of DAT− dopamine. This model by Huang and Zhan,26 also based on LeuT structure,9 was constructed using InsightII software by mutating nonconserved residues and relaxing their side-chains. Then the model was subjected to physiological environment simulation by insertion into a lipid bilayer and later solvated by water molecules at each side. Shan et al.27 also used the model by Beuming et al.24 to show the allosteric role of S2 binding site in triggering conformational change for transporter to transform from OF to inward-facing (IF) conformation. The hNET homology model by Schlessinger et al.28 based on LeuT structure was constructed using Modeller to predict backbone, and then side-chains were refined using Scwrl4 algorithm or manually. The structure was validated by virtual ligand screening of compounds that are known to bind hNET. Furthermore, the authors continued in virtual screening for novel compounds and then validated their discovery with in vitro experiments. The hSERT homology model by Celik et al.30 was based on LeuT structure and constructed using Modeller. Nearly 15 models were constructed and evaluated using energy-based and physical properties-based approaches. Experimental validation was performed using mutagenesis and in vitro inhibition assay for uptake of radiolabeled neurotransmitter. Comparative homology modeling of hDAT, hNET, and hSERT by Koldso et al.31 is among the most recent studies. The researchers used Modeller to construct and optimize 20 LeuT-based structures of the hDAT and hNET transporters. They performed physical properties evaluation approach using PROCHECK.32 Experimental validation was done using mutagenesis and in vitro inhibition assay for uptake of radiolabeled neurotransmitter. Recently, Koldso et al.10 evaluated the binding of ligands to their homology models including hSERT model by Celik et al.30 by comparing the binding sites to data from the recently published dDAT and engineered LeuBAT structures. Although the binding sites share many conserved residues that explain the commonality of neurotransmitters in these transporters, it was only recently that the selectivity of hDAT and hNET to certain inhibitors was explained by mutational functional analysis.11 The comparison performed by Koldso et al.10 between LeuTbased homology models and dDAT/LeuBAT experimental structures showed nearly identical superimposed rotamers in conserved residues of binding sites, for example, F43dDAT, D46dDAT, D121dDAT, Y124dDAT, V327dDAT, and S426dDAT except for F319dDAT. However, in this work, we were more interested in the evaluation of nonconserved binding site residues

represent less than average quality. When the QMEAN parameters of the original templates were calculated, it was clear that the less-than-desired QMEAN scores of homology models were because of the template models (Table 1 and Supplementary Table S6). While it sounds extreme to assume that overall statistical potential of the experimental structures in Protein Data Bank might be occasionally biased due to the additional antibody chains in crystal structures, and although QMEAN servers do not give residue per residue information, we believe it is important to take in consideration the QMEAN scores of the template after removal of excess molecules, chains, and antibodies. Main statistical potential insights for the models presented here were regarding minor deviations in all-atom pairwise energy, solvation energy, and torsion angle energy, which might be improved when energy minimization is performed and atom clashes are reduced. To apply an empirical-based evaluation, it is important to understand the major differences between the template and the target. Similar to homologue structures (Supplementary Figure S1), the identical residues between template and model are represented using almost the same backbone and side-chain rotamers. Here, the experimental structure of hSERT provides all information required to perform an evaluation of hSERT homology model based on dDAT template. Differences in the backbones of dDAT and hSERT resulted either from alignment shifts or at regions with highly nonconserved amino acids. Several backbone deviations not related to alignment shifts were found, and the most critical were before and after EL2 and at the extracellular loop EL6 (Supplementary Figure S1). Cα atom deviations at the TM9 and TM10 were not observed in segmented fitting, and it is possible that these deviations resulted from the nearby shift 3. In several cases, we observed three major changes in side-chain structures. The first representing often sets of three adjacent residues that are switched from aliphatic to aromatic or vice versa between dDAT and hSERT. When these shifts occurred in the transmembrane domains, they allowed for packing or simply were facing the lipid bilayer. Many examples of this switch involve divergence between F and V: V75dDAT/F127hSERT, F267 dDAT /V283 hSERT , F350 dDAT /V366 hSERT , V457 dDAT / F475hSERT, F470dDAT/V488hSERT, and F471dDAT/V489hSERT. Other examples involve switch from F or Y to other amino acids. The second major shift in structures involved switching to longer side chains. Examples of adjacent longer side-chains in hSERT include R307 hSERT, K314 hSERT, and Q318 hSERT. Examples of longer side-chains of dDAT include L451dDAT, F452dDAT, and Y455dDAT. The third major change in side-chains occurs in the domains facing the outer space and cytoplasmic space, due to abundance of nonconserved residues when compared to transmembrane domains. Overall fitting of hSERT (PDB ID 5I6Z) and hSERT homology models showed the same and sometimes extended deviations in backbone inherited from dDAT template (Supplementary Figure S2). In addition, several sporadic new deviations in backbone have occurred. One particular region of interest was in hSERT−4M48 at 316−320. The deviation in this region was not found in other homology models. Overall fitting of the two dDAT templates together showed that aside from the EL2 region, there was also deviation at region 303− 308. This can explain the backbone change inherited in hSERT−4M48 at 316−320 and possibly also affecting the region 305−308 (Supplementary Figure S2). Furthermore, 1611

DOI: 10.1021/acschemneuro.6b00242 ACS Chem. Neurosci. 2016, 7, 1607−1613

Research Article

ACS Chemical Neuroscience

Structure fitting between dDAT (PDB ID 4M48) and hSERT (PDB ID 5I6Z) was performed by the same method described in EmpiricalBased Evaluation using dDAT as reference. Physics-Based Evaluation. MolProbity18 was used to evaluate Ramachandran outliers, rotamer outliers, Cβ deviations (>0.25 Å), bond lengths, bond angles, chiral volumes, planar groups, and also clashes. Molecular mechanics (MM) GROMOS96 force fields37 were calculated in vacuo, via DeepView/Swiss-PDB Viewer v4.1.0.36 Nonconserved residues in binding sites (S1, S2, and EL4) as described by Anderson et al.11 were evaluated in detail. Knowledge-Based evaluation. Qualitative model energy analysis (QMEAN)22 was used to compare three general statistical potential terms covering the stability of protein structure in addition to secondary structure and solvent accessibility parameters. Empirical-Based Evaluation. An X-ray based structure of human serotonin transporter, hSERT (PDB ID 5I6Z),13 with 4.53 Å resolution was used to assess the hSERT homology-based models. Overall structure fitting was assessed via MatchMaker tool in UCSF Chimera version 1.10.2.,38 using Needleman-Wunsch alignment (BLOSUM-62 matrix). Deviations (RMSD) in Cα atom greater than 2 Å were considered. Segmented fitting was performed by dividing models into segments of ∼25−50 residues in DeepView/ Swiss-PDB Viewer v4.1.0.36 A step by step alignment and iterating magic fitting of all atoms in dDAT and hSERT segments were performed, followed by visual evaluation of backbone and side-chain differences.

described by Anderson et al.,11 which have been described to discriminate the selectiveness between monoamine transporters. Davis et al.,33 who studied the paroxetine inhibition of SERT, recently constructed SERT homology models for human, Drosophila, and chicken (which are species that have experimentally known kinetics). The models were constructed from dDAT structure template (PDB ID 4XP4). Loops were modeled without template to improve the quality of backbone dihedral angles. Approximately 2000 models were generated using Modeller, and the top 20 were chosen according to Modeller’s molpdf score. The molpdf score uses ranking based on the sum of all restraints in each model and does not have standard scale; therefore the lowest energy represents the best possible model. The authors used PROCHECK to evaluate secondary structure physical properties. Experimental analysis on mutant hSERT and radiolabeled neurotransmitter uptake in the presence of paroxetine was used to evaluate the homology structures. The latter study and the homology models developed in the past decade highlight the future trend in study of neurotransmitter transporters. The X-ray structures of dDAT and hSERT will play a main role in the study of hDAT and hNET in the near future. Understanding the differences and similarities between these proteins at the level of alignment, backbone, and side-chain will allow for easier evaluation and optimization of more accurate structures. The top priority in evaluation of structure is the law of physics. The next stage of evaluation is the confidence of nonconserved sequences, which can either be eliminated, loop modeled, or relaxed. Automated high-throughput homology-based modeling will soon become a routine procedure. Here, we employed multiple approaches for evaluation of homology-based models at the first phase of construction. Correct alignment plays a major role in prediction of structure. Physical properties relating to backbone, sidechains, and all atom bonds and angles are very important criteria for structure models. Application of new experimental models (dDAT and hSERT) will improve the accuracy of homology models, previously utilizing prokaryotic leucine transporter (LeuT) structure, and provide better predictions of ligand interactions, which is required for understanding of cellular mechanisms and for development of novel therapeutics.





ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acschemneuro.6b00242. Physical properties, knowledge-based, and empirical evaluation of homology based models and evaluation of sequence alignment of monoamine transporters (PDF)



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Tel: +420-545-133-350. Fax: +420-545-212-044. Author Contributions

Y.H. and Z.H. contributed to the design of experiments. Y.H. performed the computation and writing. Z.H. reviewed the manuscript, and V.A. was principle investigator and contributor to scheme and organization of work. All authors have given approval to the final version of the manuscript.

METHODS

Sequences and Structures. Human hDAT (Uniprot ID Q01959), human hNET (Uniprot ID P23975), and human hSERT (Uniprot ID P31645) sequences were used in all modeling and referring in this work. Drosophila dDAT (Uniprot ID Q7K4Y6) sequence was used. Sequence nomenclature was reviewed with particular emphasis on numeration skipping at positions 163 to 207 in EL2 of dDAT (PDB ID 4M48). Other numeration differences were discussed where necessary. Needle pairwise sequence alignment was used to calculate sequence similarity.34 Multiple alignment analysis was performed on PROMALS3D multiple sequence and structure alignment server.35 The homology-based structures of hDAT, hNET, and hSERT were constructed using SWISS-MODEL.23 X-ray based structures of Drosophila dopamine transporter dDAT (PDB ID 4M48)14 and dDAT (PDB ID 4XPA)6 with 2.96 and 2.95 Å resolution, respectively, were used as templates. Models with deleted EL2 loop for knowledge-based comparison were modified via DeepView/Swiss-PDB Viewer v4.1.0,36 by removing residues 190−214 (hDAT), 186−211 (hNET), and 210−229 (hSERT).

Funding

We gratefully acknowledge the Czech Agency for Healthcare Research, AZV (15-28334A), and Ministry of Education, Youth and Sports of the Czech Republic under the project CEITEC 2020 (LQ1601) for financial support of this work. Notes

The authors declare no competing financial interest.



ABBREVIATIONS dDAT, Drosophila dopamine transporter; hDAT, human dopamine transporter; hNET, human norepinephrine transporter; hSERT, human serotonin transporter; LeuT, leucine transporter



REFERENCES

(1) Kristensen, A. S., Andersen, J., Jorgensen, T. N., Sorensen, L., Eriksen, J., Loland, C. J., Stromgaard, K., and Gether, U. (2011) SLC6

1612

DOI: 10.1021/acschemneuro.6b00242 ACS Chem. Neurosci. 2016, 7, 1607−1613

Research Article

ACS Chemical Neuroscience Neurotransmitter Transporters: Structure, Function, and Regulation. Pharmacol. Rev. 63, 585−640. (2) Blakely, R. D., Defelice, L. J., and Hartzell, H. C. (1994) Molecular physiology of norepinephrine and serotonin transporters. J. Exp. Biol. 196, 263−281. (3) Li, L. B., Chen, N. H., Ramamoorthy, S., Chi, L. M., Cui, X. N., Wang, L. J. C., and Reith, M. E. A. (2004) The role of N-glycosylation in function and surface trafficking of the human dopamine transporter. J. Biol. Chem. 279, 21012−21020. (4) Chen, N. H., and Reith, M. E. A. (2000) Structure and function of the dopamine transporter. Eur. J. Pharmacol. 405, 329−339. (5) Reith, M. E. A., Xu, C., and Chen, N. H. (1997) Pharmacology and regulation of the neuronal dopamine transporter. Eur. J. Pharmacol. 324, 1−10. (6) Wang, K. H., Penmatsa, A., and Gouaux, E. (2015) Neurotransmitter and psychostimulant recognition by the dopamine transporter. Nature 521, 322−327. (7) Sung, U., Jennings, J. L., Link, A. J., and Blakely, R. D. (2005) Proteomic analysis of human norepinephrine transporter complexes reveals associations with protein phosphatase 2A anchoring subunit and 14−3-3 proteins. Biochem. Biophys. Res. Commun. 333, 671−678. (8) Torres, G. E., Gainetdinov, R. R., and Caron, M. G. (2003) Plasma membrane monoamine transporters: Structure, regulation and function. Nat. Rev. Neurosci. 4, 13−25. (9) Yamashita, A., Singh, S. K., Kawate, T., Jin, Y., and Gouaux, E. (2005) Crystal structure of a bacterial homologue of Na+/Cl– dependent neurotransmitter transporters. Nature 437, 215−223. (10) Koldso, H., Grouleff, J., and Schiott, B. (2015) Insights to ligand binding to the monoamine transporters-from homology modeling to LeuBAT and dDAT. Front. Pharmacol. 6, 208. (11) Andersen, J., Ringsted, K. B., Bang-Andersen, B., Stromgaard, K., and Kristensen, A. S. (2015) Binding site residues control inhibitor selectivity in the human norepinephrine transporter but not in the human dopamine transporter. Sci. Rep. 5, 15650. (12) Rannversson, H., Wilson, P., Kristensen, K. B., Sinning, S., Kristensen, A. S., Stromgaard, K., and Andersen, J. (2015) Importance of the Extracellular Loop 4 in the Human Serotonin Transporter for Inhibitor Binding and Substrate Translocation. J. Biol. Chem. 290, 14582−14594. (13) Coleman, J. A., Green, E. M., and Gouaux, E. (2016) X-ray structures and mechanism of the human serotonin transporter. Nature 532, 334−339. (14) Penmatsa, A., Wang, K. H., and Gouaux, E. (2013) X-ray structure of dopamine transporter elucidates antidepressant mechanism. Nature 503, 85−90. (15) Qu, X. T., Swanson, R., Day, R., and Tsai, J. (2009) A Guide to Template Based Structure Prediction. Curr. Protein Pept. Sci. 10, 270− 285. (16) Kleywegt, G. J., and Jones, T. A. (1996) Phi/psi-chology: Ramachandran revisited. Structure 4, 1395−1400. (17) Al-Lazikani, B., Jung, J., Xiang, Z. X., and Honig, B. (2001) Protein structure prediction. Curr. Opin. Chem. Biol. 5, 51−56. (18) Davis, I. W., Leaver-Fay, A., Chen, V. B., Block, J. N., Kapral, G. J., Wang, X., Murray, L. W., Arendall, W. B., Snoeyink, J., Richardson, J. S., and Richardson, D. C. (2007) MolProbity: all-atom contacts and structure validation for proteins and nucleic acids. Nucleic Acids Res. 35, W375−W383. (19) Morris, A. L., Macarthur, M. W., Hutchinson, E. G., and Thornton, J. M. (1992) Stereochemical quality of protein-structure coordinates. Proteins: Struct., Funct., Genet. 12, 345−364. (20) Lovell, S. C., Davis, I. W., Arendall, W. B., de Bakker, P. I. W., Word, J. M., Prisant, M. G., Richardson, J. S., and Richardson, D. C. (2003) Structure validation by C alpha geometry: phi,psi and C beta deviation. Proteins: Struct., Funct., Genet. 50, 437−450. (21) Huang, N., Kalyanaraman, C., Bernacki, K., and Jacobson, M. P. (2006) Molecular mechanics methods for predicting protein-ligand binding. Phys. Chem. Chem. Phys. 8, 5166−5177.

(22) Benkert, P., Tosatto, S. C. E., and Schomburg, D. (2008) QMEAN: A comprehensive scoring function for model quality assessment. Proteins: Struct., Funct., Genet. 71, 261−277. (23) Bordoli, L., Kiefer, F., Arnold, K., Benkert, P., Battey, J., and Schwede, T. (2009) Protein structure homology modeling using SWISS-MODEL workspace. Nat. Protoc. 4, 1−13. (24) Beuming, T., Kniazeff, J., Bergmann, M. L., Shi, L., Gracia, L., Raniszewska, K., Newman, A. H., Javitch, J. A., Weinstein, H., Gether, U., and Loland, C. J. (2008) The binding sites for cocaine and dopamine in the dopamine transporter overlap. Nat. Neurosci. 11, 780−789. (25) Huang, X. Q., Gu, H. H., and Zhan, C. G. (2009) Mechanism for Cocaine Blocking the Transport of Dopamine: Insights from Molecular Modeling and Dynamics Simulations. J. Phys. Chem. B 113, 15057−15066. (26) Huang, X., and Zhan, C. G. (2007) How dopamine transporter interacts with dopamine: Insights from molecular modeling and simulation. Biophys. J. 93, 3627−3639. (27) Shan, J. F., Javitch, J. A., Shi, L., and Weinstein, H. (2011) The Substrate-Driven Transition to an Inward-Facing Conformation in the Functional Mechanism of the Dopamine Transporter. PLoS One 6, e16350. (28) Schlessinger, A., Geier, E., Fan, H., Irwin, J. J., Shoichet, B. K., Giacomini, K. M., and Sali, A. (2011) Structure-based discovery of prescription drugs that interact with the norepinephrine transporter, NET. Proc. Natl. Acad. Sci. U. S. A. 108, 15810−15815. (29) Wang, H., Goehring, A., Wang, K. H., Penmatsa, A., Ressler, R., and Gouaux, E. (2013) Structural basis for action by diverse antidepressants on biogenic amine transporters. Nature 503, 141−145. (30) Celik, L., Sinning, S., Severinsen, K., Hansen, C. G., Moller, M. S., Bols, M., Wiborg, O., and Schiott, B. (2008) Binding of serotonin to the human serotonin transporter. Molecular modeling and experimental validation. J. Am. Chem. Soc. 130, 3853−3865. (31) Koldso, H., Christiansen, A. B., Sinning, S., and Schiott, B. (2013) Comparative Modeling of the Human Monoamine Transporters: Similarities in Substrate Binding. ACS Chem. Neurosci. 4, 295− 309. (32) Laskowski, R. A., Macarthur, M. W., Moss, D. S., and Thornton, J. M. (1993) PROCHECK - A program to check the stereochemical quality of protein structures. J. Appl. Crystallogr. 26, 283−291. (33) Davis, B. A., Nagarajan, A., Forrest, L. R., and Singh, S. K. (2016) Mechanism of Paroxetine (Paxil) Inhibition of the Serotonin Transporter. Sci. Rep. 6, 23789. (34) Li, W. Z., Cowley, A., Uludag, M., Gur, T., McWilliam, H., Squizzato, S., Park, Y. M., Buso, N., and Lopez, R. (2015) The EMBLEBI bioinformatics web and programmatic tools framework. Nucleic Acids Res. 43, W580−W584. (35) Pei, J. M., Tang, M., and Grishin, N. V. (2008) PROMALS3D web server for accurate multiple protein sequence and structure alignments. Nucleic Acids Res. 36, W30−W34. (36) Guex, N., and Peitsch, M. C. (1997) SWISS-MODEL and the Swiss-PdbViewer: An environment for comparative protein modeling. Electrophoresis 18, 2714−2723. (37) Scott, W. R. P., Hunenberger, P. H., Tironi, I. G., Mark, A. E., Billeter, S. R., Fennen, J., Torda, A. E., Huber, T., Kruger, P., and van Gunsteren, W. F. (1999) The GROMOS biomolecular simulation program package. J. Phys. Chem. A 103, 3596−3607. (38) Pettersen, E. F., Goddard, T. D., Huang, C. C., Couch, G. S., Greenblatt, D. M., Meng, E. C., and Ferrin, T. E. (2004) UCSF chimera - A visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605−1612.

1613

DOI: 10.1021/acschemneuro.6b00242 ACS Chem. Neurosci. 2016, 7, 1607−1613