The binding sites for cocaine and dopamine in the dopamine transporter overlap

Binding models for dopamine and the cocaine analog CFT

To achieve insight into the molecular basis for the interaction between cocaine and DAT, we developed a molecular model of DAT from the high-resolution structure of the bacterial NSS member LeuT (Protein Data Bank accession code 2A65)¹² and a recently refined alignment of NSS proteins¹⁰. The ligand-docking protocol¹⁸ was validated with the successful redocking of leucine into LeuT, which reproduced the crystal structure. Thus, the lowest-energy model identified by the set of criteria established for the docking protocol used throughout had a r.m.s. deviation of ~1 Å relative to the crystal structure. Notably, this low-energy binding mode was populated multiple times during the docking simulation, indicating that the protocol yields converged solutions. Following this same protocol, we determined the binding modes of dopamine and the cocaine analog CFT ((−)-2β-carbomethoxy-3β-(4-fluorophenyl)tropane or WIN 35,428) in the DAT model. CFT has pharmacological properties that are similar to those of cocaine, and was chosen for its high affinity and structural rigidity (Fig. 1b)¹⁹. The models resulting from the initial docking were refined with molecular dynamics simulations of the complex immersed in an environment that included water and the membrane in explicit atomic details.

The final docking models revealed binding sites for dopamine and CFT in the center of DAT that were composed of residues in transmembrane domains (TMD) 1, 3, 6 and 8. Despite the differences in molecular size and structure between dopamine and CFT, we found that most of the protein side chains that interact with CFT were identical with those in dopamine binding (Fig. 1 and Table 1). A recent modeling study proposed a similar binding mode for dopamine²⁰. The protonated amine of dopamine formed a salt-bridge with the Asp79 side chain (Fig. 1c). Our models also predicted a polar interaction between the amine of CFT and Asp79 (Fig. 1d). The aromatic ring of dopamine and the tropane ring of CFT interacted with hydrophobic and aliphatic residues in TMD 1 (Phe76), 3 (Val152) and 6 (Phe320 [omitted for clarity], Phe326, Val328) (Fig. 1c,d). Dopamine further engaged in hydrogen bonds with exposed backbone carbonyls of the unwound regions of TMDs 1 and 6 (Fig. 1c and Table 1) and the N-methyl substituent of CFT was inserted in between TMDs 1 and 6, where it interacted with Phe76 and with the backbone of TMD 6 (Fig. 1d and Table 1). The para-hydroxyl group of the catechol ring in dopamine interacted the with backbone atoms of Ser149 and Val152, whereas the meta-hydroxyl group interacted with Ser422, Ala423 and G425. In the case of CFT, these residues interacted with the halogenated phenyl ring (Fig. 1c,d and Table 1).

Important differences also became apparent in the binding sites of the two ligands. One marked difference involved the network of interactions surrounding Tyr156 in TMD 3. Notably, this residue is 100% conserved in all NSS transporters¹⁰, and, in LeuT, the hydroxyl group of the tyrosine binds the carboxylate group of leucine¹². In the dopamine-binding model, the network of interactions involving Tyr156 included an aromatic-aromatic stacking interaction with the catechol ring of dopamine, a hydrogen bond between the hydroxyl group and Asp79 in TMD 1 and a hydrophobic-aromatic interaction with the side chain of Leu80 on the extracellular side (Fig. 1c). CFTalso interacted with Tyr156 through the 2β-methylester substituent (Fig. 1d) in the CFT-binding model, but stabilized a different conformation of the side chain. This reorientation disrupted the hydrogen bond between Tyr156 and the key Asp79 residue (Fig. 1c,d). Another key difference between CFT and dopamine binding involved a unique interaction between the nitrogen of Asn157 and the fluoride atom of CFT (Fig. 1d and Table 1).

Experimental validation of binding models

We sought to experimentally validate the binding models of CFT and dopamine by verifying commonalities and differences between their binding modes. Previous reports have suggested a role for Asp79 in binding both CFT and dopamine^21,22, which is consistent with the interactions that we observed between the protonated amines of the ligands and this residue (Fig. 1c,d). Further supporting our model, earlier studies of DAT have indicated that Phe76, Phe320 and Phe326 (ref. ²³) are involved in the binding. Val152, Tyr156, Val328 and Ser422 (Table 1) were also predicted to be involved in binding both CFT and dopamine, and we decided to analyze them in further detail.

In TMD 3, Val152 is predicted to form hydrophobic interactions with dopamine and CFT (Fig. 1c,d). It has been suggested previously that this residue is involved in cocaine binding from analysis of bovine and human DAT chimeras²⁴ and by analysis of the corresponding residue in SERT (Ile172)²⁵. To further assess the role of this position in DAT, we mutated the residue to alanine (V152A), isoleucine (V152I) and methionine (V152M). Consistent with our predictions, the K_M value for [³H]dopamine uptake increased 5–18-fold in these mutants (Fig. 2a and Supplementary Table 1 online). Moreover, consistent with a role of this residue in binding both dopamine and CFT, the K_D value for [³H]CFT binding increased 2–4-fold in V152A and V152I, and we were unable to measure binding in V152M, suggesting a larger loss of binding affinity (Supplementary Table 1). The [³H]dopamine uptake–inhibition assay confirmed that there was a more pronounced effect on CFT binding of the methionine substitution as compared with V152A and V152I (Fig. 2b and Supplementary Table 1). For Tyr156, which is situated one α-helical turn above Val152 in TMD 3, our models predicted an essential role for the aromatic ring in binding both dopamine and CFT (Fig. 1c,d). Correspondingly, mutation of Tyr156 to Ala (Y156A) and Cys (Y156C) resulted in inactive transporters with neither measurable [³H]dopamine uptake nor [³H]CFT binding (data not shown). Surface biotinylation and immunofluorescence experiments showed similar plasma membrane expression of Y156C compared to controls (Supplementary Fig. 1 online). Thus, the lack of measurable binding and uptake was probably the result of disruption of critical interactions with the ligands and not of impaired folding and retention in the endoplasmic reticulum. Note that previous studies in SERT and in the GABA transporter-1 (GAT-1) have suggested that this highly conserved tyrosine is involved in substrate binding in these transporters as well^26,27.

According to our model, Val328 in TMD 6 is another important residue in the hydrophobic pocket that surrounds the ligands, and thus its mutation would be predicted to affect both recognition of CFT and dopamine (Fig. 1c,d). Consistent with this prediction, we could not measure [³H]CFT binding on mutation of Val328 to isoleucine (V328I; data not shown). This was supported by [³H]dopamine uptake–inhibition assays showing more than tenfold loss in CFT potency (Fig. 2b and Supplementary Table 1). In contrast, we did not observe an effect on the K_M value for [³H]dopamine uptake in the V328I construct, but the V_MAX value was substantially decreased (Fig. 2a and Supplementary Table 1), whereas a less conservative substitution (V328F) made [³H]dopamine uptake and [³H]CFT binding unmeasurable (data not shown). The differences in effects of V328I on dopamine and CFT recognition were analyzed in the structural context with molecular dynamics simulations of the V328I mutant. We found that dopamine could accommodate the added bulk in V328I through a small shift in the position of the ligand that was made possible by its relatively small size; in contrast, CFT occupied a larger portion of the binding-site volume, and its interactions did not permit such a shift to avoid the steric clash (data not shown).

Finally, we analyzed Ser422 (situated in TMD 8), which, according to our model, interacts with the hydroxyls of the catechol ring in dopamine and the aromatic ring in CFT (Fig. 1c,d and Table 1). Substitution with alanine (S422A) increased the K_M value for [³H]dopamine uptake ~fivefold and decreased CFT inhibition potency by 30-fold (Fig. 2a,b and Supplementary Table 1). Consistent with the model, we did not detect [³H]CFT binding in S422A (data not shown). In summary, by showing effects on both dopamine and CFT recognition, the mutational analysis of Val152, Tyr156, Val328 and Ser422 supports our binding models and, thus, an overlapping binding site for these compounds in DAT.

To investigate the predicted differences between the binding modes of dopamine and CFT, we also mutated Tyr156 to phenylalanine (Y156F). The removal of this hydroxyl group would be predicted to destabilize the dopamine-binding pocket by disrupting the hydrogen bond between Tyr156 and Asp79, while preserving the CFT binding scheme in which this hydrogen bond is already missing (Fig. 1c,d). Consistent with this, the K_M value for [3H]dopamine uptake was markedly increased in Y156F and the V_MAX for [3H]dopamine uptake was markedly decreased (Fig. 2c–f and Supplementary Table 2 online). We also observed a decrease in estimated dopamine affinity (~13-fold) in [³H]CFT competition binding experiments (Supplementary Table 3 online). In contrast, we observed no measurable change in CFT affinity as determined by direct [³H]CFT binding and by measuring inhibition of [³H]dopamine uptake (Fig. 2c–f and Supplementary Tables 2 and 3).

In the dopamine-binding model, Tyr156 not only formed a hydrogen bond with Asp79, but also made a hydrophobic-aromatic interaction with Leu80 in TMD 1 (Fig. 1c). Accordingly, we predicted that mutation of Leu80 would destabilize Tyr156, and thereby selectively impair the interaction of DAT with dopamine, as in Y156F. Indeed, mutation of Leu80 to Cys (L80C) decreased the potency of dopamine, whereas [³H]CFT binding was unaffected (Supplementary Fig. 2 and Supplementary Table 4 online). [³H]dopamine uptake was not measurable in L80C (data not shown), which is consistent with the decreased potency of dopamine to inhibit [³H]CFT binding, but which also suggests an additional effect on uptake efficacy as compared with Y156F. Note that mutation in SERT and GAT-1 of the residues corresponding to Leu80 also resulted in marked impairment of substrate transport, supporting a possible general role of this position in NSS proteins^28,29.

As a final test to validate the differences between the dopamine- and CFT-binding modes, we mutated Asn157 in TMD 3, which forms a polar interaction with the fluorine substituent of CFT and therefore has the only predicted unique side chain interaction with this compound compared with the dopamine-binding mode (Fig. 1d and Table 1). Indeed, the mutation (N157C) did not affect the K_M value for [³H]dopamine uptake, but the K_D for [³H]CFT binding was increased by ~15-fold, supporting the specific interaction in the model (Fig. 2c–f and Supplementary Table 1).

A buried CFT-binding pocket is supported by cross-linking

To exclude indirect effects from the inferences suggested by the mutagenesis and to further validate the buried position of the CFT binding site, we attempted to trap radiolabeled CFT in its binding pocket by introducing structural constraints at positions predicted to lie extracellular to the bound ligand. First, we undertook to engineer a Zn²⁺-binding site, as such constructs have been used previously to impose intramolecular structural constraints in membrane proteins, including this transporter class^30–32. Our model predicted that a bidentate (two coordinating residues) Zn²⁺-binding site might be formed above the CFT binding pocket between position 80 (Leu80) and 159 (Ile159) if these positions were substituted with a histidine and a cysteine, respectively (Fig. 3a). According to the model, Zn²⁺ would coordinate His80 and Cys159 in a tetrahedral conformation with Cys-S–to-Zn²⁺ distances of 2.6 Å, and His-N–to-Zn²⁺ distances of 2.2 Å. Binding of Zn²⁺ to this site should constrain the His80 side chain and place the imidazole ring immediately above the CFT-binding pocket, thereby potentially restricting release of bound CFT (Fig. 3a). It is interesting to note that the Zn²⁺-binding site position corresponded to an extracellular vestibule that was recently shown to represent a noncompetitive binding site for antidepressants in LeuT^13,14 (illustrated by a spherical model of clomipramine bound to LeuT; Fig. 3a). Thus, the two residues in LeuT corresponding to Leu80 and Ile159 (Leu25 and Ile111) both line this allosteric site^13,14, making the binding of Zn²⁺ and of the antidepressant mutually exclusive (Fig. 3a).

To validate the model experimentally, we generated the Zn²⁺-binding double mutant L80H-I159C in a DAT background (E2C H193K) in which endogenous Zn²⁺ sensitivity was eliminated by mutation of His193 to lysine^30,33. Two extracellular cysteines (Cys90 and Cys306) were also mutated in this construct, thereby rendering the transporter insensitive to modification with extracellularly applied sulfhydryl-reactive reagents³³. This background was chosen to allow direct comparison with the cysteine cross-linking experiments described below (see Supplementary Note online). Consistent with the model, application of 100 μM Zn²⁺ to cells expressing the resulting construct E2C H193K–L80H-I159C caused a decrease in the off rate of prebound [³H]CFT by ~fourfold (Fig. 3a). In contrast, Zn²⁺ had no effect on [³H]CFT dissociation in the background construct (E2C H193K) or in the following control mutants L80H, I159C, L80H-I159A and L80K-I159C (all made in E2C H193K) (Fig. 3a). Unfortunately, we were unable to perform uptake experiments in E2C H193K–L80H-I159C because of the importance of Leu80 for dopamine binding (Supplementary Table 4). Note, however, that Zn²⁺ did not affect [³H]CFT binding affinity per se (K_D = 3.1 [2.4, 4.1] nM with Zn²⁺ versus 3.4 [2.7, 4.2] nM without Zn²⁺, means with calculated s.e.m. interval in square brackets, n = 3 and 8, respectively). Thus, the decreased off rate cannot be the result of stabilizing the transporter in a high-affinity [³H]CFT conformation, but rather is the result of steric occlusion of a buried binding site with the Zn²⁺ ion affecting both the on and off rates. This was confirmed in on-rate experiments showing that the rate of [³H]CFT binding to E2C L80H-I159C-H193K was decreased ~threefold in the presence of Zn²⁺ (k_0.5 = 15.4 [13.6, 17.9] min with 100 μM Zn²⁺ versus 5.8 [5.3, 6.4] without Zn²⁺ (95% confidence interval, n = 2)).

In a second strategy we used the bifunctional sulfhydryl-reactive cross-linking reagent 1,3-propanediyl bismethane-thiosulfonate (MTS-3-MTS) to crosslink cysteine residues with Cβ atoms separated by <10 Å. We introduced cysteines in the background construct E2C (see Supplementary Note) at positions 84 (Trp84) and 159 (Ile159), which would be just above the CFT binding site with a predicted distance of 8.5 Å between the Cβ atoms of Cys84 and Cys159 (Fig. 3b). In addition, these positions have been cross-linked in the related GAT-1 (ref. ³⁴). We observed a >threefold decrease in [³H]CFT off-rate on treatment of the resulting double mutant E2C W84C-I159C with 0.5 mM MTS-3-MTS after prebinding of [³H]CFT (Fig. 3b), consistent with cross-linking of the two engineered cysteines and subsequent partial trapping of [³H]CFT in its binding site. There was no effect of MTS-3-MTS on [³H]CFT dissociation in the background construct E2C or in the control mutants W84C, I159C, W84C-I159A or W84A-I159C (all made in E2C; Fig. 3b). We also attempted cross-linking in E2C W84C-I159C mutants with copper phenanthroline and cadmium; however, we did not observe an effect of these, suggesting that the two cysteines did not come sufficiently close to form a disulfide bridge or to be cross-linked by cadmium. Finally, we used the bulky MTS reagent benzocaine-methanethiosulfonate (BZ-MTS) to test whether imposing a simple steric hindrance above the binding pocket was sufficient to slow [³H]CFT dissociation (Fig. 3c). Consistent with this notion, BZ-MTS (0.5 mM) markedly decreased the off-rate of prebound [³H]CFT in both E2C I159C and in E2C W84C (Fig. 3c), whereas no effect on [³H]CFT dissociation was observed in the background construct E2C or in the control mutants I159A or W84A (both made in E2C; Fig. 3c). Although the effect of the reagents on CFT dissociation was similar to that of Zn²⁺, we were unable to reliably assess the effect of MTS-3-MTS and BZ-MTS pretreatment on [³H]CFT association because of interference from cysteines still present in the E2C background (Supplementary Note).

Binding modes for cocaine, amphetamine and benztropines

The binding models of CFT and dopamine were first compared with that of cocaine, which was predicted to bind similarly to CFT (Fig. 4). Consistent with this, we observed no changes in affinity for cocaine in Y156F as determined in [³H]CFT competition binding experiments and [³H]dopamine uptake–inhibition experiments, as predicted by the disruption of the Tyr156-Asp79 hydrogen bond by cocaine binding (Fig. 4a and Supplementary Tables 2 and 3). This similarity in binding modes was further supported measurements of the ability of cocaine to inhibit [³H]dopamine uptake in V152M, V328I and S422A mutants (Supplementary Table 1). Mutation of Asn157 (N157C) also affected cocaine binding (fivefold decrease in potency; Supplementary Table 1); however, the interaction of CFT was through the fluorine substituent (Fig. 1d), whereas the interaction of cocaine was through the aromatic ring (Fig. 4a). Calculation of the nonbonded interaction energy between the side chain of Asn157 and the ligands agreed with these data, with CFT showing a favorable interaction energy of ~1.9 kcal mol⁻¹, cocaine showing 0.8 kcal mol⁻¹ and dopamine having virtually no interaction (~0.1 kcal mol⁻¹).

We also compared the binding models of dopamine, CFT and cocaine with those of a series of structurally diverse DAT ligands (Fig. 5a). We found it interesting that the Tyr156-Asp79 hydrogen bond that we observed in dopamine binding was preserved in all modeled complexes except those of cocaine and CFT (Fig. 4 and Supplementary Fig. 3 online). Accordingly, it would be expected that the removal of the hydrogen bond by mutation of Tyr156 to phenylalanine would affect binding for all these ligands, as we observed for dopamine. Consistent with this, we observed, on the basis of [³H]CFT competition binding experiments, 19- and 6-fold decreases in affinity for the substrates amphetamine and 3,4-methylenedioxy-N-methylamphetamine (‘ecstasy’), respectively (Fig. 4d, Supplementary Fig. 3 and Supplementary Table 4). For three different benztropines, including benztropine itself, JHW007 and MFZ 2-71 (refs. ^35,36), the experimental observations were also consistent with the models; hence, the binding affinities assessed in [³H]CFT competition assays decreased between 7- and 35-fold for the three compounds on mutation of Tyr156 to phenylalanine, supporting the idea that the Asp79-Tyr156 hydrogen bond is important for interaction (Fig. 4f, Supplementary Fig. 3 and Supplementary Table 3). The analog JHW007 was of particular interest to us, as it has been shown to fully antagonize the response to cocaine in behavioral models³⁶. Note that we observed mainly similar effects when assessing [³H]dopamine uptake–inhibition potency in Y156F as compared with calculating affinity from [³H]CFT competition binding, although the effects appeared to be more modest for MFZ 2-71 and methylenedioxy-N-methylamphetamine (Supplementary Tables 2 and 3).

The stability and time-dependent behavior of the discriminant Tyr156-Asp79 hydrogen bond was explored with molecular dynamics simulations. All of the modeled complexes were simulated for 1 ns at 310 K, and the average distance between the oxygen atoms of Tyr156 and Asp79 was monitored in the second (500 ps) half of the trajectory. As illustrated by plotting the observed affinity ratios (K_I of Y156F:K_I of wild type) against the donor-acceptor distance of the hydrogen-bond (Fig. 5b), the hydrogen bond was continuously present (distance <3.5 Å) in all trajectories, except for those of CFT and cocaine, in agreement with the inference from experimental data. Finally, the CFT and dopamine models were simulated in a realistic solvated membrane environment, showing that disruption of the Tyr156-Asp79 hydrogen bond was associated with water permeating deep into the CFT-binding site (Fig. 5c). In the dopamine-bound state, however, the Tyr156-Asp79 hydrogen bond effectively closed the binding site from the extracellular side and prevented water from entering (Fig. 5c). Notably, the increased hydration stabilized the CFT-binding site and thereby compensated for the loss of intramolecular interactions. This explains mechanistically the unaltered affinity of CFT seen on removal of the Tyr156-Asp79 hydrogen bond in Y156F, and also suggests how the recently shown ability of cocaine and CFT, but not dopamine, to stabilize an open, outward-facing conformation of the transporter can be achieved³⁷.

Modeling of DAT and ligand complexes

Our modeling protocol is described in detail in the Supplementary Methods online. Briefly, ligands were docked to the LeuT-based model using a simulated annealing–based protocol¹⁸, implemented in CHARMM⁴⁶. Initial positions were generated with a Monte Carlo algorithm. Parameters for the ligands were generated with structures from the Cambridge Structure Database and ab initio calculations at the HF/6-31G* level. The restrained electrostatic potential–fit method⁴⁷ was used to derive the atomic charges. Ligand-transporter interactions were assessed by simulating the complexes in a realistic membrane environment model for ~50 ns (Supplementary Methods).

Modeling of chemical cross-links

The Zn²⁺-binding site was modeled in I159C-L80H and the side chains were optimized using the rotamer-optimization module of InsightII (Accelrys). Zn²⁺ was initially placed in between the Cys-S and His-N atoms and the complex was minimized for 1,000 steps, without any restraints on Zn²⁺. The ε-rotameric state of His80 produced the most stable complex. For the BZ-MTS and MTS-3-MTS models, mutants I159C and W84C-I159C were generated in the DAT model, respectively. The structure of BZ-MTS was built in an extended fashion and covalently linked to I159C. Torsional angles of the I159C/BZ-MTS unit were modified to minimize steric overlap with other residues. The MTS-3-MTS molecule was placed in between the rotamer-optimized Cys84 and Cys159 by minimizing the distance between the sulfur atoms of extended MTS-3-MTS and Cys84/Cys159. The BZ-MTS and MTS-3-MTS complexes were minimized to generate the final models.

Molecular biology

Synthetic cDNAs encoding the human DAT (synDAT) and E2C were subcloned into pcDNA3 (Invitrogen)³³. All mutations were generated by the QuickChange method (Stratagene) and confirmed by DNA sequencing.

Cell culture and transfection

COS7 cells were grown as described previously and transiently transfected with wild-type and mutant constructs using the calcium phosphate–precipitation method³³.

[³H]Dopamine-uptake experiments

Uptake assays were carried out as described previously³³ using 2,5,6-[³H]dopamine (9–13 Ci mmol⁻¹, Amersham Pharmacia Biotech), and transfected COS7 cells were plated in either 24-well (10⁵ cells per well) or 12-well dishes (3 × 10⁵ cells per well). The uptake assays were carried out 2 d after transfection for 5 min at room temperature (20–22 °C) in uptake buffer (25 mM HEPES, 130 mM NaCl, 5.4 mM KCl, 1.2 mM CaCl₂, 1.2 mM MgSO₄, 1 mM L-ascorbic acid, 5 mM D-glucose and 1 μM of the catechol-O-methyltransferase inhibitor Ro 41-0960 (Sigma), pH 7.4).

[³H]CFT-binding experiments

Binding assays were carried out on intact COS7 cells as described previously³⁰ using 2–4 nM [³H]CFT (WIN 35,428 from PerkinElmer, 87 Ci mmol⁻¹) as a radioligand. The assay was performed 2 d after transfection in uptake buffer on cells in either 24-well (10⁵ cells per well) or 12-well (3 × 10⁵ cells per well).

[³H]CFT off-rate experiments

Binding of [³H]CFT to transfected COS7 cells was initiated by the addition of 4 nM [³H]CFT in 450 μl of uptake buffer, and the cells were kept at 4 °C. After 1 h, either Zn²⁺ (100 μM), MTS-3-MTS (0.5 mM) or BZ-MTS (0.5 mM) were added and the plates were incubated for 10 min at 21 °C before washing twice in uptake buffer (21 °C). At t = 0, 0.5 ml of uptake buffer (21 °C) with 100 μM CFT (for the Zn²⁺ experiments, the uptake buffer also contained 100 μM Zn²⁺) were added and the reaction was stopped by aspiration of uptake buffer at the indicated time points. Cells were lysed and counted as described previously³⁰.

[³H]CFT on-rate experiments

Transfected cells were washed and preincubated with buffer or Zn²⁺ (100 μM) before the addition of [³H]CFT. The [³H]CFT incubations were stopped at the indicated time points, washed twice in ice cold uptake buffer, transferred and counted as described previously³⁰.

Data calculations

Data from uptake and binding experiments were analyzed by nonlinear regression analysis using Prism 5.0 (GraphPad Software). The IC₅₀ values used in the estimation of K_M values for [³H]dopamine uptake and K_D values for [³H]CFT binding were calculated from means of the obtained pIC₅₀ values and the s.e. interval from the pIC₅₀ ± s.e.m. values. The K_I values were calculated from the IC₅₀ values using the equation K_I = IC₅₀/(1+(L/K_M)) for [³H]dopamine uptake and K_I = IC₅₀/(1+(L/K_D)) for [³H]CFT binding (L = concentration of [³H]dopamine or [³H]CFT).