Steric Hindrance Mutagenesis in the Conserved Extracellular Vestibule Impedes Allosteric Binding of Antidepressants to the Serotonin Transporter*

Background: The serotonin transporter contains an allosteric binding site with unknown location. Results: We use molecular modeling, mutagenesis, and zinc site engineering to locate and inhibit the allosteric binding site. Conclusion: Data are consistent with the allosteric binding site being located in the extracellular vestibule. Significance: This opens the search for a new generation of antidepressant drugs directed toward the proposed allosteric binding site. The serotonin transporter (SERT) controls synaptic serotonin levels and is the primary target for antidepressants, including selective serotonin reuptake inhibitors (e.g. (S)-citalopram) and tricyclic antidepressants (e.g. clomipramine). In addition to a high affinity binding site, SERT possesses a low affinity allosteric site for antidepressants. Binding to the allosteric site impedes dissociation of antidepressants from the high affinity site, which may enhance antidepressant efficacy. Here we employ an induced fit docking/molecular dynamics protocol to identify the residues that may be involved in the allosteric binding in the extracellular vestibule located above the central substrate binding (S1) site. Indeed, mutagenesis of selected residues in the vestibule reduces the allosteric potency of (S)-citalopram and clomipramine. The identified site is further supported by the inhibitory effects of Zn2+ binding in an engineered site and the covalent attachment of benzocaine-methanethiosulfonate to a cysteine introduced in the extracellular vestibule. The data provide a mechanistic explanation for the allosteric action of antidepressants at SERT and suggest that the role of the vestibule is evolutionarily conserved among neurotransmitter:sodium symporter proteins as a binding pocket for small molecule ligands.

The neurotransmitter serotonin (5-HT) 3 plays a critical role in the central nervous system, where it modulates physiological and psychological functions, such as mood, sleep, appetite, and sexual drive (1). The serotonin transporter (SERT) mediates rapid reuptake of 5-HT following its release from presynaptic nerve terminals and is thereby essential for maintaining 5-HT homeostasis in the brain. SERT belongs to the neurotransmitter:sodium symporter (NSS) family that also includes transporters for the neurotransmitters dopamine, norepinephrine, glycine, and ␥-amino butyric acid (2). SERT is the main pharmacological target in the treatments of major depression and anxiety disorders. Both classical tricyclic antidepressants (TCAs), including, for example, clomipramine (CMI) and imipramine, and the selective serotonin reuptake inhibitors (SSRIs), including, for example, (S)-citalopram ((S)-CIT; Lexapro), sertraline (Zoloft), and fluoxetine (Prozac), exert their actions as potent inhibitors of SERT (3).
It has been a sustained goal in the mechanistic study of these compounds to gain insight into the structural basis underlying their action at SERT. Crystallization of the amino acid transporter LeuT, a bacterial homologue to the mammalian NSS proteins, provided the first insight into the tertiary structure of this transporter family (4). Interestingly, LeuT has been crystallized not only in its substrate bound form but also together with TCAs (5,6) or with the SSRIs fluoxetine and sertraline (7), for which LeuT was found to possess low affinity. These structures revealed a binding site for TCAs and SSRIs in the LeuT located in an extracellular vestibule (termed the S2 site) that is ϳ13 Å above the central substrate binding site (S1 site). Based on these observations and further mutagenesis of the corresponding site in SERT, it was proposed that the high affinity binding site of TCAs and SSRIs is located in the aligned cavity, a putative S2 site in SERT (6 -8). However, this view was challenged by other studies strongly suggesting that both SSRIs ((S)-CIT, fluoxetine, and sertraline) and TCAs (CMI, imipramine, and amitriptyline) are classical competitive inhibitors with their primary high affinity binding site located in the S1 site (9 -14). Correspondingly, we have shown in the dopamine transporter (DAT) that the high affinity binding site for both cocaine and the benztropine class inhibitors is the S1 site (15,16).
In addition to the high affinity binding site, it has been known for almost 3 decades that SERT also possesses a low affinity allosteric site (17). Its existence was demonstrated by showing that several SERT ligands, of both the TCA and SSRI class, as well as 5-HT itself, can modulate the dissociation rates of other SERT ligands (18 -21). For example, (S)-CIT has a marked allosteric effect, resulting in dramatic inhibition of the dissociation of a high affinity bound inhibitor, such as (S)-CIT itself or other SSRIs (19,21). Notably, it has been suggested that this putative dual action of (S)-CIT at two binding sites in the SERT is responsible for the higher efficacy and faster onset observed in clinical trials for (S)-CIT as compared with racemic citalopram (CIT) (22)(23)(24)(25)(26). However, despite previous attempts to locate the binding site (27)(28)(29)(30), the molecular mechanism underlying the allosteric effect of SERT inhibitors has remained essentially unknown. The site is considered distinct from the high affinity binding site because mutations that disrupt high affinity binding of (S)-CIT to SERT did not affect the general allosteric effect of (S)-CIT (27). Conversely, a series of mutations in transmembrane segment 10 (TM10) and TM12, which impaired the allosteric effect of (S)-CIT, did not affect its high affinity binding (28 -30).
Here we provide evidence, by combining computational approaches with mutagenesis, Zn 2ϩ site engineering, and cysteine reactivity assays, that the allosteric binding site in SERT is in the vestibule located extracellular to the S1 site. Accordingly, our data support a model in which the dual action of (S)-CIT at SERT might involve the high affinity S1 site as well as binding to the S2 site. This model also identifies the S2 site as an evolutionarily conserved binding site for small molecule ligands in the NSS proteins, in line with the recent suggestion that S2 represents a secondary substrate binding site at least in LeuT (31,32).

Molecular Modeling of SERT with Ligand Complexes and
Zn 2ϩ Binding Sites-The homology model of SERT was built based on the LeuT structure in an outward-occluded state (Protein Data Bank code 2A65). Using the protocols as previously described for our DAT models (33), the SERT model was immersed in the explicit water-lipid-water environment and then relaxed with molecular dynamics simulations. The differences between the outward-occluded and outward-open conformations of LeuT revealed by crystallography are limited; their root mean square deviation of all C␣ atoms is only 1.2 Å. Importantly, the conformational rearrangements may not be relevant in addressing the divergence between LeuT and SERT in the extracellular vestibule (see supplemental Fig. 1). Therefore, in our study, enhancing the sampling of the interactions between the SERT protein and the bound antidepressants is more critical than choosing a specific LeuT structure as the starting template.
Thus, to characterize the simultaneous binding of two compounds in two binding sites, we developed an integrated and iterative protocol that combines induced fit docking (IFD) and molecular dynamics (MD). The IFD protocol (34) induces conformational changes in the binding site to accommodate the ligand and exhaustively identify possible binding modes and associated conformational changes by side chain sampling and backbone minimization. To explore the conformational changes associated with the binding and enable the sampling of structural rearrangements related to any allosteric coupling between the S1 and S2 sites, we performed MD runs using Desmond (35) on the resulting top-ranked IFD poses in order to equilibrate the complex and optimize the ligand-transporter interactions. If significant rotation and translation freedom was observed for the ligand in the MD simulation stage, we facilitated convergence by redocking the molecule into the MD-altered binding site, using IFD and re-equilibration of the complex with MD. This iterative approach, described below, was applied until the ligand binding mode was stable and converged for at least 30 ns. Its practical advantage is the increased sampling of binding modes so as to overcome the difficulties presented by the large molecular sizes of the inhibitors of SERT and their relatively low affinities for the S2 site. Moreover, the iterative relaxation and sampling through MD reduce the time required for convergence and the uncertainty associated with the use of a homology model (36) for SERT.
The iterative approach is briefly illustrated for the SERT configuration with (S)-CIT in the S1 site and CMI in the S2 site: (i) an (S)-CIT molecule was first docked into the S1 site of the SERT model (36); (ii) the top ranked ligand-transporter complex was reimmersed in an all-atom representation of a bilayerwater environment and was subjected to 48-ns MD simulations; (iii) a CMI molecule was subsequently docked in the S2 site, based on an equilibrated MD snapshot from stage (ii); and (iv) an MD simulation in explicit solvent was carried out following the protocol described in (ii). For this particular case, we observed a significant rearrangement of the interactions between the transporter and the CMI ligand that moved from its initial docking pose (ϳ2-Å translation). To efficiently explore the CMI in its altered environment, based on the snapshot at the end of 6 ns, we reiterated step (iii) and then re-equilibrated the new complex for 48 ns in step (iv), which resulted in a more stable complex. For another construct, with (S)-CIT bound in both the S1 and S2 sites, there was no major rotation or translation of the (S)-CIT in the S2 site during the 60-ns run in the initial step (iv), precluding the need for an iteration back to step (iii).
The procedures for modeling of the Zn 2ϩ binding sites and the covalent benzocaine-methanethiosulfonate (BZ-MTS) complex are described in detail in the supplemental Methods.
Site-directed Mutagenesis-The human SERT was cloned into the pUbi1z expression vector using NotI and XbaI. Muta-tions herein were generated either using the QuikChange method (adapted from Stratagene, La Jolla, CA) or ordered through GeneArt (Regensburg, Germany). All mutations were confirmed by DNA sequencing.
(S)-[ 3 H]CIT Dissociation Rate Assay-Dissociation rates were measured as previously described (27). In brief, (S)- CIT was determined using a Wallac microplate scintillation counter. Nonspecific binding was determined with 1 M paroxetine/nomifensine at 37°C for 90 min. The temperature for the dissociation rate was set for each mutant so that the precise t1 ⁄ 2 could be determined. Dissociation rates were determined in triplicate in at least three independent experiments for all constructs.
(S)-[ 3 H]CIT Binding Experiments-The affinity of (S)-CIT to the high affinity binding site in SERT WT and mutants was determined by the addition of 3-5 nM (S)-[ 3 H]CIT in binding buffer together with increasing concentrations of (S)-CIT in the concentration range from 0.01 to 2500 nM using a consecutive factor 3 dilution (10 determinations in triplicate) in 96-well plates. Subsequently, membranes expressing SERT WT or mutants were added to a total volume of 400 l. The binding mixture was incubated for 1 h at room temperature and subsequently filtered, washed, and counted as described for the dissociation rate assay. Nonspecific binding was determined by adding 5 M paroxetine.

5-[ 3 H]HT Uptake
Experiments-Uptake assays were performed using 5-[1,2-3 H]hydroxytryptamine (5-[ 3 H]HT) (28 Ci/mmol) (PerkinElmer Life Sciences). Transfected COS7 cells were plated in 24-well dishes (10 5 cells/well) coated with polyornithine to achieve an uptake level of no more than 10% of total added 5-[ 3 H]HT. The uptake assays were carried out 2 days after transfection. Prior to the experiment, the cells were washed once in 500 l of binding buffer at room temperature. 5-HT was added to the cells in 10 concentrations from 1 nM to 1 mM equally distributed around the expected IC 50 value, and uptake was initiated by the addition of ϳ10 nM 5-[ 3 H]HT in a final volume of 500 l. After 3 min (for the WT) or 5 min (for the mutants) of incubation, the cells were washed twice with 500 l of ice-cold uptake buffer, lysed in 250 l of 1% SDS, and left for 30 min at 37°C. All samples were transferred to 24-well counting plates, and 500 l of Opti-phase Hi Safe 3 scintillation fluid (PerkinElmer Life Sciences) was added followed by counting of the plates in a Wallac Tri-Lux ␤-scintillation counter (PerkinElmer Life Sciences). Nonspecific uptake was determined in the presence of 5 M paroxetine. All determinations were performed in triplicate.
Data Calculations-The allosteric potency was calculated as described previously (29). The calculated dissociation rate constants (k [drug] ) at different (S)-CIT or CMI concentrations are expressed relative to the dissociation rate constant without the presence of unlabeled ligand (k buffer ). The allosteric potency was determined as the drug concentration that impairs the dissociation rate by 50% compared with dissociation in buffer. IC 50 values were calculated from concentration effect curves of normalized dissociation ratio (k [drug] /k buffer ) versus log[drug] and are shown as mean values calculated from means of pIC 50 and the S.E. interval from the pIC 50 Ϯ S.E. All data were subjected to linear or nonlinear regression analysis using Prism version 5.0 (GraphPad Software Inc., San Diego, CA).

Computational Characterizations of (S)-Citalopram or Clomipramine Binding in the S2
Site-To test the hypothesis that the putative S2 site in SERT may be the long-sought allosteric binding site in SERT for inhibitors such as (S)-CIT and CMI, we developed an iterative IFD/MD protocol (see "Experimental Procedures"). Applying this protocol to characterize binding poses of either (S)-CIT or CMI in the S2 site with (S)-CIT bound in the S1 site, we sought to identify experimentally testable interacting residues in the S2 site. To establish a proper context for the S2 binding, we first used the protocol to characterize the binding pose of (S)-CIT bound in the S1 site (S1: (S)-CIT). Compared with results from previous efforts, our S1:(S)-CIT pose was very similar to the experimentally validated results (10, 13) (see supplemental Discussion). We then docked either (S)-CIT or CMI in the S2 site of the SERT model equilibrated with S1:(S)-CIT.
During the equilibration of our SERT model with (S)-CIT bound in both sites, when the poses of the ligands and surrounding SERT regions were stabilized, we observed the coordination of S2:(S)-CIT to include residues Leu 99 , Trp 103 , and Arg 104 in TM1; Ile 179 in TM3; Ala 486 , Val 489 , and Lys 490 in TM10; Val 236 and Leu 237 in extracellular loop 2 (ECL2); and Gly 402 in ECL4 (Fig. 1, A and B). The center-of-mass distance between the S1:(S)-CIT and S2:(S)-CIT was ϳ17 Å. In our equilibrated SERT complex with S2:CMI in the presence of S1:(S)-CIT, the binding pose of S2:CMI overlapped significantly with that of S2:(S)-CIT, resulting in a similar interaction pattern (Fig. 1C). However, CMI protrudes less toward TM10 and ECL2 and has no direct interaction with Ala 486 (Fig. 1C). Compared with the previously reported binding modes of CMI in the extracellular vestibule (8,14), both the location and ori-entation of the stabilized CMI in our SERT model are different, especially with the alkylamine pointing toward Val 489 . The differences are probably due to the presence of a bound (S)-CIT molecule in the S1 site, which is unique in our study. Note that although the S2 site in SERT is comparable with the binding site for CMI found in LeuT (5), and they share a few common positions of coordinated residues, the S2:CMI in SERT is positioned more extracellular than in LeuT (supplemental Fig. 1).
The coordination of S2:(S)-CIT and S2:CMI as seen in the results from the computations would obstruct the release of S1:(S)-CIT to the extracellular side along the proposed transport pathway (supplemental Fig. 1) (31). We note, however, that unlike the reports from molecular docking of two substrates bound in the S1 and S2 sites in DAT (32), binding modes of S1:(S)-CIT in SERT do not seem to be affected by the presence or absence of the S2:(S)-CIT/CMI. The S2-bound inhibitors had no significant structural impact either on the SERT or on the position of S1:(S)-CIT in the simulations; rather they were simply situated so as to block the entry and exit route to the high affinity S1 site.

Site-directed Mutagenesis of Proposed Interacting Residues in S2 Decreases the Allosteric Potency of (S)-CIT and CMI-To
validate the SERT models with S2:(S)-CIT/CMI, we employed mutations that would be expected to cause steric hindrance (37,38) and/or affect the conformations of residues predicted to have side chain interactions with the bound molecules. Thus, we mutated residues from beneath (L99H in TM1), from the sides (W103H and R104K in TM1 and I179H in TM3), or from the top (A486E, V489H, and K490A in TM10, V236X and L237X (where X represents His, Tyr, or Glu) in ECL2, and G402H in ECL4). The mutants were analyzed in (S)-[ 3  Because mutations of either Val 236 or Leu 237 produced complete loss of (S)-  Table 2). This suggests that the changes in volume distribution and polarity produced by the inserted histidine may cause an obstruction of the exit route from the binding site, probably analogous to the effect of an allosteric inhibitor. Interestingly, such an obstruction effect was not observed for mutations more distal to the S1 binding site like G402H and A486E (supplemental Fig. 2 and Table 2).
The potency of (S)-CIT and CMI in inhibiting the dissociation of high affinity bound (S)-[ 3 H]CIT was measured according to methods described previously (27,29) by prebinding of (S)-[ 3 H]CIT to the membrane preparations followed by deter-FIGURE 1. Binding modes of (S)-CIT and CMI in the S2 site of SERT. A, the relative locations of the S1 and S2 sites in our SERT model viewed from an angle parallel to the membrane with (S)-CIT (yellow spheres) docked into both sites. The TMs and ECL2 and -4 are indicated B and C, top views of the (S)-CIT (yellow spheres) and CMI (blue spheres) in the S2 site, respectively, which is composed of residues from TM1, TM3, and TM10, and ECL2 and -4. S1-bound (S)-CIT (yellow) as well as the putative location of the two Na ϩ (orange) and the Cl Ϫ (green) are shown.

TABLE 1 Effect of mutations in the extracellular vestibule on 5-[ 3 H]HT uptake and (S)-[ 3 H]CIT binding
The listed values were found by non-linear regression analysis of competition uptake or binding assays by either 5-[ 3 (Fig. 2, A and B). Consistent with previous observations (30) Table 3).
In agreement with our S2:(S)-CIT model, L99H, W103H, R104K, I179H, G402H, A486E, and V489H mutations significantly reduced the allosteric potency of (S)-CIT (Fig. 2, C and E and Table 3). The strongest effect was seen for G402H, leading to complete elimination of measurable allosteric effect of (S)-CIT (Fig. 2, B and C). Notably, Gly 402 is aligned to Ala 319 in LeuT that accommodates CMI binding to S2 in LeuT (5). Consistently, in our model of the G402H mutation, the introduced histidine residue protrudes into the extracellular vestibule, which should sterically hinder the binding of S2:(S)-CIT (supplemental Fig. 3). The R104K mutation also showed a drastic effect on (S)-[ 3 H]CIT dissociation ( Fig. 2C and Table 3), causing a 50-fold reduction in allosteric potency. In addition to being in the S2 site and contacting (S)-CIT directly, Arg 104 also probably forms the extracellular thin gate with Glu 493 (4,39). According to the LeuT crystal structures (4,39), the opening and closing of this gate, in association with the configuration changes of the conserved Phe 334 -Phe 335 -Tyr 175 -Tyr 176 aromatic cluster, is assumed to play a critical role in the conformational transition. To characterize the impact of Arg 104 mutation, we modeled and simulated R104K and compared it with WT in the presence of S1:(S)-CIT. The subtle but significant difference between the packing of Lys 104 -Phe 335 and Arg 104 -Phe 335 is propagated into the S2 site by disrupting the Glu 493 -Tyr 175 hydrogen bond interaction. Consequently, the Tyr 175 side chain rotates into the S2 site and partially fills the cavity (supplemental Fig. 4). Although high affinity (S)-[ 3 H]CIT and 5-[ 3 H]HT binding are retained in R104K, the transport activity (Table 1) is disrupted, with significantly reduced V max , suggesting that the mutation has limited impact on the S1 site but conformational transition is impaired. The mutation of Arg 104 to other residues (e.g. R104A) resulted in an inactive transporter ( Table 1).
The effects of the remaining mutants were significant but more modest, showing a ϳ3-12-fold decrease in allosteric potency for (S)-CIT. Because both the basal (S)-[ 3 H]CIT dissociation and the allosteric potency were decreased by the L99H and I179H mutants, we reasoned that the effect of the double mutant would be even larger if both histidines cause a steric hindrance of the allosteric bound (S)-CIT. Indeed, the double mutant produces a Ͼ20-fold decrease in allosteric potency (Table 3). A similar augmentation was observed when combining the three mutants in TM10 (i.e. A486E/V489H/K490A), with the allosteric potency found to decrease more than 20-fold to 106 [88;127] M (Table 3).
To exclude the possibility that all mutations in the extracellular vestibule would somehow affect allosteric binding, we made T178V. The side chain of Thr 178 is exposed to the vestibule but it is not in direct interaction with S2:(S)-CIT in our docking results. Consistent with this prediction, the T178V is WT-like for all tested parameters (Tables 1-3 An important agreement of the measurements with our modeling results is underscored by the parallel way in which the allosteric potency of CMI was affected by mutants that affected (S)-CIT, albeit to a lesser extent; the most pronounced effect was observed for R104K, which caused a 17-fold decrease in allosteric potency relative to SERT WT (Fig. 2D). The G402H mutation resulted in a 7-fold decrease, and a 2-4-fold decrease was observed for the L99H, W103H, and I179H mutants (Fig. 2, D and F, and Table 3). A486E caused only a minor, non-significant effect on the CMI potency ( Fig. 2F and Table 3). This is entirely consistent with the computational models where Ala 486 was found to be the only residue that coordinates S2:(S)-CIT but not S2:CMI. The different impact of A486E on the allosteric potencies of S2:(S)-CIT and S2:CMI is thus explained by the observed difference in the binding modes of the two ligands in the same pocket. Another mutation on TM10, V489H, also had no significant effect on the allosteric potency of CMI, which is consistent with the observation from the simulations that S2:CMI protrudes less toward TM10 than S2:(S)-CIT, as described above.

TABLE 2 Effect of mutations in the extracellular vestibule on basal (S)-[ 3 H]CIT dissociation
The listed values were found by linear regression analysis of dissociation time at 20°C for prebound (S)-[ 3 H]CIT on membranes prepared from COS7 cells transiently transfected with SERT WT or mutant. For the mutants with added Zn 2ϩ or BZ-MTS, the reagent was added in a concentration of 200 M and 0.5 mM, respectively. The mean Ϯ S.E. is calculated from at least three experiments performed in triplicate. Engineering of a Zn 2ϩ Binding Site in the Extracellular Vestibule Impairs the Allosteric Potency of Antidepressants-To gain further evidence for the location of the allosteric binding site and its relation to the S2 site, we engineered a Zn 2ϩ binding site in the vestibule. Zn 2ϩ binding should impair the allosteric effects of (S)-CIT and CMI in the pocket affected by the presence of the ion. The structural requirements for Zn 2ϩ binding to proteins are well defined (40), and engineered Zn 2ϩ binding sites have been used before to impose intramolecular constraints in membrane proteins, including NSS proteins (15,(41)(42)(43). An analysis of the MD simulation trajectory of SERT showed that Zn 2ϩ could be coordinated between Ile 179 and Val 489 after substitution of these residues by histidines (I179H/ V489H). The results of computational docking of Zn 2ϩ into this mutant construct, in the presence of S2:(S)-CIT (see supplemental Methods), showed that coordination of Zn 2ϩ close to the S2:(S)-CIT produced electrostatic repulsion toward the dimethylammonium group (Fig. 3A). A similar repulsion was observed between Zn 2ϩ and S2:CMI (Fig. 3B). To verify these models experimentally, we made the double mutant (I179H/ V489H) and the single mutants (I179H and V489H) and investigated the effect of Zn 2ϩ on the allosteric effects of (S)-CIT and CMI on (S)-[ 3 H]CIT dissociation. In agreement with the predictions, the application of 200 M Zn 2ϩ to I179H/V489H caused a significant decrease in the allosteric potency of both (S)-CIT (ϳ4-fold) and CMI (ϳ5-fold) (Fig. 3, C and D, and Table 4). This effect was not seen for the single mutants, showing that both histidine residues are required to obtain the effect of Zn 2ϩ , consonant with the need for both to produce a binding site for the ion (Fig. 3, C and D, insets, and Table 4).

SERT construct (S)-[ 3 H]CIT dissociation
To exclude the possibility that the effect of Zn 2ϩ at I179H/ V489H resulted from altering the affinity of (S)-  A Cysteine Reactive Bulky Reagent in S2 Impairs the Allosteric Potency of (S)-CIT-To further substantiate the role of the SERT S2 site in the allosteric effect of (S)-CIT, we introduced a steric hindrance to S2:(S)-CIT binding in the form of the bulky cysteine reagent BZ-MTS (Fig. 4A, inset). A cysteine was introduced in S2 (L99C in TM1) in a Cys-less SERT background (SERT-C) in which the only reactive cysteine on the extracellular face of SERT had been mutated (C109A) (44). As in WT SERT, the addition of (S)-CIT (40 M) to both SERT-C and SERT-C L99C decreased the dissociation rate (and thereby increased the t1 ⁄ 2 for dissociation) of (S)-[ 3 H]CIT. In SERT-C, the effect of (S)-CIT was slightly impaired by BZ-MTS, and BZ-MTS was found to have no effect by itself (Fig. 4B). In contrast, BZ-MTS impaired the allosteric effect of (S)-CIT in SERT-C L99C, as reflected by a significant decrease in t1 ⁄ 2 for (S)-[ 3 H]CIT dissociation (Fig. 4B). In addition, in the SERT-C L99C, the reactivity of BZ-MTS did have an effect by itself on (S)-[ 3 H]CIT dissociation (t1 ⁄ 2 ϭ 31.3 Ϯ 2.2 and 21.7 Ϯ 1.6 min with and without added BZ-MTS, respectively) These data are consistent with our covalent docking model of BZ-MTS in SERT L99C (Fig. 4A).
A Low Affinity Inhibitor Binding Site in the Extracellular Vestibule of the Dopamine Transporter-We then investigated whether the presence of a low affinity binding site for inhibitors in the extracellular vestibule of SERT can be generalized to other mammalian transporters. Thus, we assessed whether CMI was able to inhibit the dissociation of prebound

DISCUSSION
For almost 3 decades, it has been proposed that SERT possesses an allosteric binding site for antidepressants (17). The allosteric activity has been demonstrated for several antidepressants, including the SSRIs ((S/R)-CIT, sertraline, fluoxetine, and paroxetine) and the TCAs (CMI and imipramine) (19 -21), as well as for 5-HT itself (17). A corresponding allosteric site has also been proposed to exist in the homologous norepinephrine transporter (20). However, the molecular mechanisms behind the observations, as well as the actual binding site(s), have not been identified. In this study, we used (S)-CIT and CMI as model compounds for SSRIs and TCAs, respectively, and combined computational modeling and mutagenesis studies to locate the allosteric binding site for these compounds in SERT (Fig. 1). We found that the allosteric binding site is in the extracellular vestibule and is analogous to the S2 site in LeuT (supplemental Fig. 1), which has been found to bind inhibitors (5-7) and has been suggested to constitute a second substrate-binding site (45).   Mutation of the residues predicted to protrude into the allosteric binding site region from below (TM1, L99H and R104K), from the sides (TM1, W103H; TM3, I179H; and TM10, A486E,  V489H, and K490A), and from above (ECL4, G402H) indeed decreased the allosteric potency of both (S)-CIT and CMI to various degrees (Fig. 2). The most pronounced effect was seen for G402H in which the allosteric effect of (S)-CIT was eliminated (Fig. 2), suggesting that the side chain of histidine may be especially disruptive of an energetically acceptable positioning of the 1,3-dihydroisobenzofuran-5-carbotitrile moiety of (S)-CIT in close vicinity to position 402 (supplemental Fig. 3). This inference agrees with previous observations showing that histidines are particularly efficient in producing steric hindrance by virtue of the aromatic and polar character of the side chain (37,38). The R104K mutation also caused drastic decreases in allosteric potency for both (S)-CIT and CMI (Fig. 2). In addition to directly disrupting the binding of the S2 inhibitors, it is likely that the large decrease of allosteric potency of R104K is also due to the participation of Arg 104 in the extracellular thin gate in SERT (forming a salt bridge to Asp 493 in TM10); thus, the R104K may bias the conformational equilibrium so that it is not optimal to bind S2 inhibitors (supplemental Fig. 4).
Interestingly, we observed that substitutions with histidines at positions 99 and 179 already reduced the basal dissociation rate of the high affinity bound (S)-[ 3 H]CIT (supplemental Fig. 2 and Table 2), consistent with the high affinity binding site being located beneath the two residues (9,(11)(12)(13), which then impose a constraint on the exit route from the S1 site.
Some of our mutations in S2 (e.g. W103H and A486E) reduced high affinity binding to S1 (Table 1), but the impact on S1 affinity does not bias the comparison of allosteric potencies across different mutants because it is accommodated into the calculations (see Experimental Procedures). It is possible that the affinity change is due to an allosteric effect that propagates from the S2 site and causes minor distortions of the adjacent S1 pocket.
The structural context of the allosteric effects we proposed was further strengthened by the results obtained from constructs in which a Zn 2ϩ binding site was engineered in S2 (I179H/V489H). Zn 2ϩ binding impaired the allosteric effect of CMI and (S)-CIT, consistent with an obstruction of S2:(S)-CIT/ CMI binding without blocking the dissociation pathway of S1:(S)-CIT (Fig. 3). The possibility that Zn 2ϩ causes a major structural rearrangement by coordinating between I179H in TM3 and V489H in TM10 and thus occluding the S2 pocket is considered unlikely because the coordination requirements of the small and inert Zn 2ϩ ion are very strict, and the binding is reversible and of low affinity. Additionally, Zn 2ϩ did not exert any measurable allosteric effect in I179H/V489H by itself, which argues further against a major conformational rearrangement in the S2 pocket.
In parallel to the Zn 2ϩ experiments, we found that the cysteine-reactive compound BZ-MTS significantly impaired the allosteric effect of (S)-CIT in SERT-C L99C compared with SERT-C (Fig. 4). The modest size of the effect is possibly due to the low labeling efficiency, but we were unable to increase further the BZ-MTS concentration and the labeling time due to nonspecific effects in the background mutation (SERT-C).
Nonetheless, our data suggest that conjugation of BZ-MTS to Cys 99 in TM1 results in protrusion of the benzocaine side chain into the vestibule and thereby partial obstruction of (S)-CIT binding in S2 (Fig. 4). Notably, this is analogous to the observation of a decrease in basal (S)-[ 3 H]CIT dissociation upon steric hindrance mutagenesis in the same position (L99H).
A chimera study has previously identified five residues in TM10 and TM12 as critical for allosteric binding. Because TM12 (together with TM9) constitutes a putative dimeric interface according to the LeuT crystal structure, it was speculated that the identified allosteric residues may be associated with the dimer interface and therefore capable of transmitting signals between the binding sites in two monomers (28 -30). However, the five residues are not predicted to form a well defined binding pocket to accommodate small molecule ligands. Therefore, in light of our current observations and supported by recent simulations (46), we surmise that mutation of the five residues indirectly affects the conformation of the extracellular vestibule and thereby the allosteric potency of (S)-CIT.
It is, however, unexpected that we do not observe larger effects in several of the described mutations. If the inserted residues caused "efficient" steric hindrance, all mutations might be expected to have effects similar to G402H and R104K. Because the vestibule is relatively large and the binding is of low affinity, it is possible that the binding modes of S2 inhibitors have tolerance for changes at non-essential positions, which can reduce the impact of the introduced mutation.
The effects of some mutations on the allosteric potencies might be indirect; by binding to another site on the transporter, the ligand could potentially alter the relative abundance of different transporter conformations that in turn change accessibility to the primary binding site through the extracellular vestibule. In such a scenario, the mutations may affect the ability of the ligand to alter the conformational state of the transporter to different extents. However, our modeling, mutational analysis, Zn 2ϩ binding data, and MTS experiments argue against this possibility. It is also important to note that all critical mutations retained (S)-[ 3 H]CIT binding, supporting the notion that the mutations did not cause major changes in conformational equilibria.
Taken together with the previously accumulated data, the present observations suggest that the extracellular vestibule in NSS proteins constitutes a pocket capable of binding small molecule ligands. Crystal structures of LeuT demonstrated how TCAs and some SSRIs could be accommodated in the vestibule of this transporter, and the present data support the binding of TCAs and SSRIs, such as CMI and (S)-CIT, in the vestibule of SERT, and of CMI in DAT, although in different binding modes. The vestibule has also been suggested to be important for the translocation process by constituting a possible second substrate binding site needed for transport (45)(46)(47)(48). Our observation that some of the S2 mutations impair 5-HT transport is consistent with the suggestion that the vestibule is involved in substrate translocation. However, there is no evidence from crystallography studies of LeuT to support the biochemical findings (39, 49 -51).