Chemical and structural investigation of the paroxetine-human serotonin transporter complex

Antidepressants target the serotonin transporter (SERT) by inhibiting serotonin reuptake. Structural and biochemical studies aiming to understand binding of small-molecules to conformationally dynamic transporters like SERT often require thermostabilizing mutations and antibodies to stabilize a specific conformation, leading to questions about relationships of these structures to the bonafide conformation and inhibitor binding poses of wild-type transporter. To address these concerns, we determined the structures of ∆N72/∆C13 and ts2-inactive SERT bound to paroxetine analogues using single-particle cryo-EM and x-ray crystallography, respectively. We synthesized enantiopure analogues of paroxetine containing either bromine or iodine instead of fluorine. We exploited the anomalous scattering of bromine and iodine to define the pose of these inhibitors and investigated inhibitor binding to Asn177 mutants of ts2-active SERT. These studies provide mutually consistent insights into how paroxetine and its analogues bind to the central substrate-binding site of SERT, stabilize the outward-open conformation, and inhibit serotonin transport.


Introduction
Serotonin or 5-hydroxytryptamine (5-HT) is a chemical messenger which acts on cells throughout the human body, beginning in early development and throughout adulthood (Berger et al., 2009). 5-HT acts as both a neurotransmitter and a hormone that regulates blood vessel constriction and intestinal motility (Berger et al., 2009). In the central nervous system, 5-HT is released from presynaptic neurons where it diffuses across the synaptic space and binds to 5-HT receptors, promoting downstream signaling and activating postsynaptic neurons (Gether et al., 2006;Kristensen et al., 2011). Thus, 5-HT is a master regulator of circuits, physiology and behavioral functions including the sleep/wake cycle, sexual interest, locomotion, thermoregulation, hunger, mood, and pain (Berger et al., 2009). 5-HT is cleared from synapses and taken into presynaptic neurons by the serotonin transporter (SERT), thus terminating serotonergic signaling (Gether et al., 2006;Kristensen et al., 2011;Rudnick et al., 2014). SERT resides in the plasma membrane of neurons and belongs to a family of neurotransmitter sodium symporters (NSSs) which also includes the dopamine (DAT) and norepinephrine transporters (NET) (Gether et al., 2006;Kristensen et al., 2011;Rudnick et al., 2014). NSSs are twelve transmembrane spanning secondary active transporters which utilize sodium and chloride gradients to energize the transport of neurotransmitter across the membrane (Rudnick et al., 2014;Navratna and Gouaux, 2019;Yamashita et al., 2005; Figure 1a).
The function of NSSs is modulated by a spectrum of small-molecule drugs, thus in turn controlling the availability of neurotransmitter at synapses. Selective serotonin reuptake inhibitors (SSRIs) are a class of drugs which inhibit SERT and are used to treat major depression and anxiety disorders (Cipriani et al., 2018). Using x-ray crystallography and cryo-EM, we have determined structures of thermostabilized variants of human SERT complexed with SSRIs, which together explain many of the common features and differences associated with SERT-SSRI interactions (Coleman and Gouaux, 2018;Coleman et al., 2016a). SSRIs are competitive inhibitors that bind with high-affinity and specificity to a central substrate-binding site in SERT, preventing 5-HT binding and arresting SERT in an outward-open conformation (Gether et al., 2006;Kristensen et al., 2011;Coleman et al., 2016a).
The central site in NSSs is composed of three subsites: A, B, and C (Wang et al., 2013;Figure 1b). In all NSS-ligand structures, the amine group of ligands resides in subsite A and interacts with a conserved Asp residue (Asp98 in SERT). The heterocyclic electronegative group of the ligand is positioned in subsite B (Navratna and Gouaux, 2019). For example, the alkoxyphenoxy groups of reboxetine and nisoxetine (Penmatsa et al., 2015) in Drosophila DAT (dDAT) structures, the halophenyl groups of cocaine analogs in dDAT and S-citalopram in SERT, and the catechol derivatives in DCP-dDAT and sertraline-SERT all occupy subsite B (Coleman and Gouaux, 2018;Coleman et al., 2016a;Wang et al., 2015a). In addition to the central binding site, the activity of SERT and NSSs can also be modulated by small-molecules which bind to an allosteric site located in an extracellular The substrate is bound at the central site (sand, triangle), near two sodium ions (purple, spheres +) and a chloride ion (green, sphere -). The light orange and light blue triangles depict pseudo two-fold symmetric helical repeats comprised of TM1-5 and 6-10, respectively. The disulfide bond (purple line) and N-linked glycosylation (red 'Y' shapes) in extracellular loop 2, along with sites of thermostable mutations (Tyr110Ala, TM1a; Ile291Ala, TM5; Thr439Ser, TM8) are also shown (cyan-filled circles). Structural elements involved in binding allosteric ligands are depicted as black-filled circles. Epitopes for the 8B6 and 15B8 Fab binding sites are in squiggly dark-blue and orange lines, respectively. (b) Schematic of the ABC pose of paroxetine bound to the central binding site, derived from the previously determined x-ray structures (Coleman and Gouaux, 2018;Coleman et al., 2016a). The transmembrane helices are shown with circles and mutated residues in subsite B are in sticks. c, The ACB pose of paroxetine bound to the central binding site of SERT predicted by molecular dynamics simulations and mutagenesis Slack et al., 2019).
vestibule, typically resulting in non-competitive inhibition of transport (Coleman et al., 2016a;Zhong et al., 2009;Wennogle and Meyerson, 1982;Plenge and Mellerup, 1985). Paroxetine is an SSRI which exhibits the highest known binding affinity for the central site of SERT (70.2 ± 0.6 pM) compared to any other currently prescribed antidepressants (Cool et al., 1990). Despite its high affinity, paroxetine is frequently associated with serious side effects including infertility, birth defects, cognitive impairment, sexual dysfunction, weight gain, suicidality, and cardiovascular issues (Nevels et al., 2016). As a result, the mechanism of paroxetine binding to SERT has been studied extensively in order to design drugs with higher-specificity and less adverse side-effects. Despite these efforts, however, the binding pose of paroxetine remains a subject of debate (Coleman and Gouaux, 2018;Coleman et al., 2016a;Abramyan et al., 2019;Davis et al., 2016;Slack et al., 2019).
Paroxetine is composed of a secondary amine which resides in a piperidine ring, which in turn is connected to benzodioxol and fluorophenyl groups (Figure 1b). X-ray structures of the SERT-paroxetine complex revealed that the piperidine ring binds to subsite A while the benzodioxol and fluorophenyl groups occupy subsite B and C in the central site, respectively (Coleman and Gouaux, 2018;Coleman et al., 2016a) (ABC pose, Figure 1b). However, recent mutagenesis, molecular dynamics, and binding studies with paroxetine analogues suggest that paroxetine might either occupy ABC pose as observed in the crystal structure, or an ACB pose where the benzodioxol and fluorophenyl groups occupy subsite C and B of the central site respectively Slack et al., 2019;Figure 1c). Paroxetine is also thought to interact with the allosteric site of SERT, albeit with low-affinity (Plenge and Mellerup, 1985). We have, however, been unable to visualize paroxetine binding at the allosteric site using structural methods. Our x-ray maps, by contrast, resolve a density feature at the allosteric site which instead resembles a molecule of detergent (Coleman et al., 2016a).
To resolve the ambiguity of paroxetine binding poses at the central binding site, we turned to paroxetine derivatives whereby the 4-fluoro group is substituted with either a bromine or an iodine group. Using transport and binding assays, anomalous x-ray diffraction, and cryo-EM, we have examined the binding poses of these paroxetine analogs and their interactions at the central site. Our studies provide key insights into the recognition of high-affinity inhibitors by SERT and the rational design of new small-molecule therapeutics.

Results
To provide a robust molecular basis for the interaction of paroxetine (1) with SERT, we devised synthetic routes for two derivatives of paroxetine where the 4-fluoro moiety is substituted with either bromo 2) or iodo (I-paroxetine, 3) groups (Figure 2a,b). We envisaged the use of a C-H functionalization strategy to access enantiopure hydroxymethyl intermediates I, from readily available N-Boc (R)-nipecotic acid 4 ( Figure 2b, Appendix 1). Transition metal-catalyzed C-H functionalization can promote the reaction of unactivated C(sp 3 )-H bonds with the aid of a directing group (He et al., 2017;Rej et al., 2020;Antermite and Bull, 2019;O' Donovan et al., 2018;Maetani et al., 2017;Chapman et al., 2016). Here, C-H functionalization enabled installation of the appropriate aryl group on the pre-existing piperidine ring (Antermite et al., 2018), providing an attractive and short route to vary this functionality with inherent control of enantiomeric excess. In contrast, common methods for (-)-paroxetine synthesis can require the aromatic substituent to be introduced before stereoselective steps or ring construction, reducing flexibility of the process Johnson et al., 2001;Hughes et al., 2003;Brandau et al., 2006;Krautwald et al., 2014;Wang et al., 2015b;Kubota et al., 2016;Amat et al., 2000). Nevertheless, during the preparation of this work, the synthesis of Br-paroxetine was reported using an asymmetric conjugate addition and its binding to SERT has been extensively studied Brandau et al., 2006).
We also employed several SERT variants and the 8B6 Fab in the biochemical and structural studies described here. The wild-type SERT construct used in transport experiments contains the fulllength SERT sequence fused to a C-terminal GFP tag ( Table 1). The ts2-active variant contains two thermostabilizing mutations (Ile291Ala, Thr439Ser) which allows for purification of the apo transporter for binding studies and has kinetics of 5-HT transport (K m : 4.5 ± 0.6 mM, V max : 21 ± 5 pmol min À1 ) that are in a similar range as wild-type SERT (K m : 1.9 ± 0.3 mM, V max : 23 ± 1 pmol min À1 ) (Coleman et al., 2016a;Green et al., 2015). The ts2-inactive variant (Tyr110Ala, Ile291Ala) (Coleman and Gouaux, 2018), by contrast, is unable to transport 5-HT but can be crystallized due to the stabilizing Tyr110Ala mutation (Green et al., 2015) and binds SSRIs with high-affinity. The DN72/DC13 SERT variant used for cryo-EM is otherwise wild-type SERT which has been truncated at the N-and C-termini (Table 1) and yet retains transport and ligand-binding activities (Coleman et al., 2019). Finally, the recombinant 8B6 Fab (Coleman et al., 2016a;Coleman et al., 2016b) was used to produce SERT-Fab complexes which were studied by X-ray crystallography and cryo-EM.
We began by assessing the functional effects of paroxetine, Br-paroxetine, and I-paroxetine on SERT activity by measuring their inhibition of 5-HT transport and S-citalopram competition binding. We assayed the ability of the Br-and I-paroxetine derivatives to inhibit 5-HT transport in HEK293 cells expressing wild-type SERT, observing that upon substituting the 4-fluoro group with 4-bromo or 4-iodo groups, the potency of inhibition of 5-HT transport in wild-type SERT decreased significantly from 4 ± 1 for paroxetine to 40 ± 20 for Br-paroxetine and 180 ± 70 nM for I-paroxetine (Figure 3a, Table 2). Next, we measured the binding of paroxetine, Br-paroxetine, and I-paroxetine (1 M), 110˚C, 24 hr; iv) Boc 2 O (four equiv), DMAP (20 mol %), CH 3 CN (0.5 M), 35˚C, 22 hr; v) LiAlH 4 (two equiv), THF, 20˚C, 0.5 hr; vi) MsCl (1.3 equiv), Et 3 N (1.4 equiv), CH 2 Cl 2 , 0 to 25˚C, 2 hr; vii) X = Br: sesamol (1.6 equiv), NaH (1.7 equiv), THF, 0˚C to 70˚C, 18 hr; viii) X = I: sesamol (2.0 equiv), NaH (2.2 equiv), DMF, 0˚C to 90˚C, 20 hr; ix) 4 N HCl in dioxane (10 equiv), 0˚C to 25˚C, 18 hr. to ts2-active and ts2-inactive SERT using S-citalopram competition binding assays, finding that the SERT variants employed in this study exhibited high-affinity for paroxetine and its derivatives ( Table 3). A decrease in the binding affinity upon substituting the 4-fluoro group of paroxetine with 4-bromo or 4-iodo groups was observed in the competition binding assays. However, the difference in the binding affinities between paroxetine variants measured by the competition binding assay was not as pronounced as the difference in the inhibition potencies observed in the 5-HT transport assays (Tables 2 and 3). For example, the ts2-inactive (Tyr110Ala, Ile291Ala) variant employed in the previous (Coleman and Gouaux, 2018) and present x-ray studies exhibited a K i of 0.17 ± 0.02 nM for paroxetine, 0.94 ± 0.01 nM for Br-paroxetine, and a further decrease in affinity to I-paroxetine (2.3 ± 0.1 nM). The ts2-active SERT variant binds with similar affinity to paroxetine and Br-paroxetine, and shows a 4-5 fold decrease in affinity to I-paroxetine ( Figure 3b, Table 3).
In the x-ray structures of SERT, paroxetine was modeled in the ABC pose such that the benzodioxol group is in subsite B (Coleman and Gouaux, 2018;Coleman et al., 2016a). A recent study suggested that binding affinity and potency to inhibit the transport of Br-paroxetine was only   Figure 1b). We recently also identified a conserved residue, Asn177 in the subsite B, which upon mutation exhibited differential effects on the inhibitory potency of ibogaine and noribogaine (Coleman et al., 2019). To further probe the role of Asn177 in subsite B, we studied the binding of paroxetine and its derivatives to selected Asn177 mutants designed in the ts2-active background ( Figure 1b). We observed that the affinity of paroxetine to ts2-active SERT decreased by three-fold when Asn177 is substituted with small non-polar or polar residues such as valine and threonine, while only a 2-fold change in K i was observed for glutamine (Asn177Gln) (Figure 3c). In the case of Br-paroxetine, the Asn177 variants (K i between 4 and 5 nM) display up to a 10-13 fold decrease in K i when compared with ts2-active SERT (0.4 ± 0.2 nM) ( Figure 3d, Table 3). The Asn177 variants show 2-4 fold decrease in affinity to I-paroxetine, with ts2-active SERT exhibiting a K i of 1.7 ± 0.3 nM and the mutants a K i of 4-7 nM. In the case of all three paroxetine variants, the reduction in affinity was the lowest for glutamine substitution. Irrespective of the SERT variant used, substitution of fluoro group with bromo or iodo group invariably decreased the affinity of paroxetine ( Figure 3e, Table 3).
To define the binding poses of paroxetine and its analogues to SERT, we solved the structures of the DN72/DC13 and the ts2-inactive SERT variants complexed with Br-and I-paroxetine using single particle cryo-EM and X-ray crystallography ( Large aromatic side-chains were well-resolved for all three complexes, also suggesting that the aromatic moieties of paroxetine and its analogues could be identified and positioned in our cryo-EM maps. In addition, the particle distribution and orientations of SERT-Fab complexes in presence of Br-and I-paroxetine were similar to paroxetine, allowing for uniform comparison between the maps. The~3.3 Å resolution map of the DN72/DC13 SERT-8B6 paroxetine complex allowed us to locate a density feature for the inhibitor at the central site ( Figure 4a). The resolution of the Br-and I-paroxetine complexes was comparatively lower at~4.1 Å and~3.8 Å , respectively (Table 4, Figure 4figure supplement 4). Nevertheless, these ligands could also be modeled into the density at the central site with a correlation coefficient (CC) of 0.75 and 0.77, respectively (Figure 4b-e). To compare paroxetine in the ABC vs. the ACB pose, we flexibly modeled paroxetine in both poses at the central site followed by real space refinement. We observed that in the ACB pose, paroxetine could be positioned with a CC of 0.70 compared with 0.84 for the ABC pose suggesting that while ABC pose is clearly preferred under the conditions we tested, the possibility of an ACB pose cannot be excluded ( Figure 4-figure supplement 5a,b). Based on the higher CC value, and the binding pose information from the ts2-inactive and ts3 SERT x-ray structures, the density in cryo-EM maps for paroxetine at the central site was interpreted to best accommodate ABC pose (Coleman and Gouaux, 2018;Coleman et al., 2016a). We also compared the reconstructed complexes by calculating difference maps, attempting to identify features associated with the scattering of bromine and iodine at the central and allosteric sites. However, the resulting difference maps did not contain any interpretable difference densities and thus did not further assist in ligand modeling. In the cryo-EM maps, the maltose headgroup of a DDM molecule could also be visualized in the allosteric site with the detergent tail inserted between TMs 10, 11, and 12. In contrast, in the X-ray maps only the head group of the octyl-maltoside detergent could be modeled due to the weak density of the hydrocarbon chain. We then explored the binding pose of paroxetine by growing crystals and collecting x-ray data of the ts2-inactive SERT-8B6 Fab complex with Br-and I-paroxetine (Table 5). Anomalous difference maps calculated from the previously determined ts2-inactive paroxetine structure (PDB ID: 6AWN) after refinement, showed clear densities for Br-and I-atoms of the paroxetine derivatives in subsite C (Figure 4f      We next compared the cryo-EM structure of the SERT-paroxetine complex to the X-ray structure of the ts3 SERT paroxetine complex. Overall comparison of the transporter revealed only minor variation between structures solved by each method, with a Ca root-mean-square-deviation (RMSD) of 0.68 Å . The most significant differences between the cryo-EM and the X-ray structures were found at the extracellular and intracellular sites of TM12 and also in EL2, while the core of the transporter (TM1-10) was largely unchanged (Figure 5a). These changes can largely be explained on the basis of a crystal packing interface formed by TM12 and a highly flexible EL2 that is bound to the 8B6 Fab. We also compared central site residues involved in paroxetine binding, finding that the best fit to the cryo-EM density revealed only minor differences in the side-chains of Asp98, Tyr176, and Phe335 when compared to the x-ray structure (all atom RMSD: 0.91 Å ) ( Figure 5b). Finally, we compared the cryo-EM structures of the SERT 15B8 Fab/8B6 scFv paroxetine complex (PDB: 6DZW) to the SERT 8B6 Fab paroxetine complex to understand if these antibodies induce changes in transporter structure. Here we found that the most significant differences occurred in the extracellular domain and involved localized regions of EL2 and EL4 that interact with the antibody (Figure 5c). The transporter core was largely unchanged, with the only other significant differences being found in EL6, TM12, and IL4.

Discussion
The binding of paroxetine to SERT has been extensively debated (Coleman and Gouaux, 2018;Coleman et al., 2016a;Abramyan et al., 2019;Davis et al., 2016;Slack et al., 2019). The first X-ray structure of the ts3-SERT variant demonstrated that the binding pose is such that the piperidine, benzodioxol, and fluorophenyl groups occupy subsites A, B, and C respectively, in the ABC pose (Coleman et al., 2016a;Figure 1b). Competition binding experiments using a variant of SERT containing a central binding site that has been genetically engineered to possess photo-cross-linking amino acids corroborated that paroxetine binds in a fashion which is similar to that observed in crystal structure (Coleman and Gouaux, 2018;Coleman et al., 2016a), where the fluorophenyl group is in proximity to Val501 (Rannversson et al., 2017). However, computational docking experiments using wild-type SERT predicted that the position of benzodioxol and fluorophenyl groups of paroxetine are 'flipped', with paroxetine occupying an ACB pose (Davis et al., 2016;Figure 1c). Subsequent studies involving wild-type and mutant SERT variants, that include modeling, mutagenesis, and Br-paroxetine docking experiments suggested that paroxetine could bind in both ABC and ACB poses. These studies also suggested that bromination of paroxetine and certain mutations near the central site, such as Ala169Asp, favored ABC pose Slack et al., 2019). Hence, the authors in these studies hypothesized that the ABC pose observed in the crystal structure could be because of the crystallization conditions and thermostabilizing mutations. One of the thermostabilizing mutations in ts3-SERT, Thr439Ser, is near the central binding site and Thr439 participates in a hydrogen bonding network in subsite B that, in turn, includes the dioxol group of paroxetine. To probe the role of the Thr439Ser mutation in modulating the binding pose of paroxetine, we solved the X-ray structure of ts2-inactive (Tyr110Ala, Ile291Ala) SERT, wherein the residue at position 439 was the wild-type threonine. Paroxetine could be modeled in the ABC pose in the X-ray structure of ts2-inactive SERT (Coleman and Gouaux, 2018). MD simulations of ts2-inactive SERT suggested that the Thr439Ser mutation weakens the Na2 site. Furthermore, MD simulations and binding and uptake kinetics experiments using wild-type SERT in presence of paroxetine and a variant of paroxetine where in the 4-fluoro group is substituted with 4-bromo group suggested that the paroxetine binding pose in SERT could be ambiguous because of the pseudo symmetry of the paroxetine molecule. It was noted that paroxetine could occupy both ABC and ACB poses with almost equivalent preference. Upon substituting the 4-fluoro with a bulkier 4-bromo group, the ABC pose was favored Slack et al., 2019). The structure of the ts2-inactive SERT-8B6 scFv/15B8 Fab paroxetine (cryo-EM, 6DZW), ts2-inactive SERT-8B6 Fab paroxetine (x-ray, 6AWN), and the SERT-8B6 paroxetine (cryo-EM, this work) complexes were superposed onto the ts3 SERT-8B6 paroxetine complex (x-ray, 5I6X) as a reference. The RMSD for Ca positions were calculated for each structure in comparison with the reference. Regions with RMSD > 3.0 Å are shown boxed in red.
Structural studies of SERT in complex with paroxetine and its analogues were thus required to resolve the uncertainty in paroxetine binding pose at the central site. Previously, we had demonstrated that cryo-EM can be used to define the position of ligands at the central site of SERT (Coleman et al., 2019). Here, we employed a similar methodology using the DN72/DC13 SERT variant complexed with 8B6 Fab to study binding of paroxetine at the central site. The density feature of paroxetine in the cryo-EM map at~3.3 Å clearly resolved the larger benzodioxol and smaller fluorophenyl groups in subsite B and C, respectively ( Figure 4b). Though this reconstruction suggests that paroxetine binds in the ABC pose, we also considered the possibility that the inhibitor density feature may represent an average of the ABC and ACB poses. We expected that if Br-and I-paroxetine were suitable surrogates for paroxetine, their binding pose would be unaffected by their reduced electronegativity and the size of the halogenated groups and therefore that they would also be associated with a comparable density feature at this site, as demonstrated by our cryo-EM maps. To further explore if there was a fraction of Br-or I-paroxetine in the ACB pose, we examined the position of the Br-or I-atoms at the central site by X-ray crystallography. If Br-and I-paroxetine were to bind in both the ABC or ACB poses, we expected to observe two anomalous peaks in our x-ray maps in subsites B and C; for both ligands, however, only a single detectable peak was observed in subsite C (Figure 4f,g). Thus, our direct biophysical observations reveal that under the conditions that we tested the ABC pose of paroxetine is preferred over the the ACB pose.
Paroxetine is stabilized at the central binding site by aromatic, ionic, non-ionic, hydrogen bonding, and cation-p interactions (Coleman and Gouaux, 2018). In the ABC pose, the amine of the piperidine ring of paroxetine binds with Asp98 (3.5 Å ) and also makes a cation-p interaction with Tyr95 of subsite A (Figure 4a). The benzodioxol group of paroxetine, a catechol-like entity, occupies a position in subsite B which is similar to the binding of catechol derivative groups of sertraline and 3,4-dichlorophenethylamine in SERT (Coleman and Gouaux, 2018) and dDAT (Wang et al., 2015a) structures, respectively. In subsite B, the ring of Tyr176 makes an aromatic interaction with the benzodioxol while the hydrogen-bonding network in subsite B formed by Asn177, Thr439, backbone carbonyl oxygens, and amides are likely responsible for stabilization of the dioxol. The side-chain of Ile172 inserts between the benzodioxol and fluorophenyl, while the rings of Phe341 and Phe335 stack on either side of the fluorophenyl, 'sandwiching' it within subsite C. The halogen group of paroxetine and its analogues reside adjacent to the side-chain of Thr497 (4.0 Å ), which may act to stabilize these groups through hydrogen bonding (Figure 4a). The larger atomic radius, the longer length of the carbon-halogen bond, and the difference in electronegativity of bromine (radius: 1.85 Å , bond-length: 1.92 Å , electronegativity: 2.96) and iodine (radius: 1.98 Å , bond-length: 2.14 Å , electronegativity: 2.66) relative to fluorine (radius: 1.47 Å , bond-length: 1.35 Å , electronegativity: 3.98) would explain why the fluorine analogue binds with greater affinity than Br-paroxetine and I-paroxetine.
We also explored the effect of conservative and non-conservative mutations in subsite B of SERT at Asn177 (Figure 3). Asn177 participates in a hydrogen-bond network with the hydroxyl group of noribogaine and with the dioxol of paroxetine. However, this network of interactions is also important for binding halogenated inhibitors in subsite B, as in the case for S-citalopram, fluvoxamine, and sertraline. All the mutants that we tested at Asn177 resulted in a loss of binding affinity to paroxetine and its analogues. Furthermore, the Ala169Asp mutation in subsite B Figure 1b,c) also reduced paroxetine inhibition and binding, likely also disrupting these interactions. Although the effects were less severe when compared to paroxetine, Br-paroxetine binding and inhibition was also reduced for Ala169Asp . Thus, these mutations highlight the importance of subsite B interactions in paroxetine binding but they cannot be used to demonstrate the inhibitor pose because, in the ABC or ACB poses, either the dioxol or fluorine of paroxetine could act as a hydrogen-bond acceptor in subsite B.
Using a combination of chemical biology, cryo-EM, and X-ray crystallography we observed that under the conditions that we studied, the SSRI paroxetine preferably occupies the ABC pose at the central site, where it is involved in numerous interactions. However, the data presented in the manuscript does not completely exclude the possibility of an ACB pose at the central site. Our studies of the mechanism of paroxetine binding to SERT provide a robust framework for the design of experiments to identify new highly specific small-molecule SERT inhibitors.
For cryo-EM of the DN72/DC13 SERT, 1 mM 5-HT was added during solubilization and affinity purification to stabilize SERT. GFP was cleaved from SERT by digestion with thrombin and the SERT-8B6 complex was made as described in the previous paragraph. The complex was separated from free Fab and GFP by SEC in TBS containing 1 mM DDM and 0.2 mM CHS, and the peak fractions were concentrated to 4 mg/ml followed by addition of either 200 mM paroxetine, Br-paroxetine or I-paroxetine.

Synthesis of Br-and I-paroxetine
All reactions were carried out under an inert atmosphere (argon) with flame-dried glassware using standard techniques, unless otherwise specified. Anhydrous solvents were obtained by filtration through drying columns (THF, MeCN, CH 2 Cl 2 and DMF) or used as supplied (a,a,a-trifluorotoluene). Reactions in sealed tubes were run using Biotage microwave vials (2-5 ml or 10-20 ml recommended volumes). Aluminum caps equipped with molded butyl/PTFE septa were used for reactions in a,a,atrifluorotoluene and toluene. Simple butyl septa were used for reactions in other solvents. Chromatographic purification was performed using 230-400 mesh silica with the indicated solvent system according to standard techniques. Analytical thin-layer chromatography (TLC) was performed on precoated, glass-backed silica gel plates. Visualization of the developed chromatogram was performed by UV absorbance (254 nm) and/or stained with a ninhydrin solution in ethanol. HPLC analyses were carried out on an Agilent 1260 Infinity Series system, employing Daicel Chiracel columns, under the indicated conditions. The high-resolution mass spectrometry (HRMS) analyses were performed using electrospray ion source (ESI). ESI was performed using a Waters LCT Premier equipped with an ESI source operated either in positive or negative ion mode. The software used was Mas-sLynx 4.1; this software does not account for the electron and all the calibrations/references are calculated accordingly, that is [M+H] + is detected and the mass is calibrated to output [M+H]. Melting points are uncorrected. Infrared spectra (FTIR) were recorded in reciprocal centimeters (cm -1 ).
Nuclear magnetic resonance spectra were recorded on 400 or 500 MHz spectrometers. The frequency used to record the NMR spectra is given in each assignment and spectrum ( 1 H NMR at 400 or 500 MHz; 13 C NMR at 101 MHz or 126 MHz). Chemical shifts for 1 H NMR spectra were recorded in parts per million from tetramethylsilane with the residual protonated solvent resonance as the internal standard (CHCl 3 : d 7.27 ppm, (CD 2 H) 2 SO: d 2.50 ppm, CD 2 HOD: d 3.31 ppm). Data was reported as follows: chemical shift (multiplicity [s = singlet, d = doublet, t = triplet, m = multiplet and br = broad], coupling constant, integration and assignment). J values are reported in Hz. All multiplet signals were quoted over a chemical shift range. 13 C NMR spectra were recorded with complete proton decoupling. Chemical shifts were reported in parts per million from tetramethylsilane with the solvent resonance as the internal standard ( 13 CDCl 3 : d 77.0 ppm, ( 13 CD 3 ) 2 SO: d 39.5 ppm, 13 CD 3 OD: d 49.0 ppm). Assignments of 1 H and 13 C spectra, as well as cis-or trans-configuration, were based upon the analysis of d and J values, analogy with previously reported compounds (Antermite et al., 2018), as well as DEPT, COSY and HSQC experiments, where appropriate. All Boc containing compounds appeared as a mixture of rotamers in the NMR spectra at room temperature. In some cases, NMR experiments for these compounds were carried out at 373 K to coalesce the signals, which is indicated in parentheses where appropriate. For NMR analysis performed at room temperature, 2D NMR experiments (COSY and HSQC) are also presented when useful for the assignments. Observed optical rotation (a') was measured at the indicated temperature (T˚C) and values were converted to the corresponding specific rotations a ½ T D in deg cm 2 g -1 , concentration (c) in g per 100 mL. Full details of the synthetic route, using enantiopure and racemic substrates are provided in Appendix 1, and NMR spectra of all reaction intermediates, 2 and 3, and HPLC analysis are cataloged in Supplementary files 1 and 2.

X-ray data collection
Crystals were harvested and flash cooled in liquid nitrogen. Data was collected at the Advanced Photon Source (Argonne National Laboratory, beamline 24-ID-C). Data for Br-paroxetine was collected at a wavelength of 0.91840 Å and at 1.37760 Å for I-paroxetine.
Anomalous difference maps X-ray data sets were processed with XDS (Kabsch, 2010); Friedel pairs were allowed to have different intensities. Molecular replacement was performed with coordinates from the previously determined ts2-inactive SERT-paroxetine structure (Protein Data Bank (PDB) code: 6AWN) (Coleman and Gouaux, 2018) using PHASER (Bunkó czi et al., 2013). B-factors were refined using PHENIX (Afonine et al., 2012) followed by generating anomalous difference maps using the phases derived from the higher resolution structures. To maximize the signal-to-noise ratio of the Br-paroxetine anomalous difference density, the high-resolution phases were blurred with a B-factor of 500 with a high-resolution cutoff of 5.5 Å . Using these optimized parameters for the Fourier analysis of the Brparoxetine diffraction data, we obtained an anomalous map with the largest difference peak being present at 6.0s and the noise level estimated at~2.5s. To maximize the signal-noise-ratio of the I-paroxetine anomalous difference density, a high-resolution and low-resolution cutoff of 6.3 and 30 Å was applied during the generation of the anomalous maps. Using these optimized parameters for the Fourier analysis of the I-paroxetine diffraction data, we obtained an anomalous map with the largest difference peak being present at 4.5s and the noise level estimated at~2.5s.

F o -F o isomorphous difference maps
Isomorphous difference (F o -F o ) maps were calculated in PHENIX by analyzing isomorphous pairs of crystals. Difference maps were calculated using the previously determined ts2-inactive SERT-paroxetine dataset and PDB (6AWN) for phasing. High-and low-resolution cutoffs of 6.0 and 30.0 Å were applied for the F o (paroxetine)-F o (Br-paroxetine) map and cutoffs of 6.3 and 30.0 Å were used for the F o (paroxetine)-F o (I-paroxetine) and F o (Br-paroxetine)-F o (I-paroxetine) maps.

Cryo-EM grid preparation
To promote the inclusion of particles in thin ice, 100 mM fluorinated octyl-maltoside (final concentration) from a 10 mM stock was added to SERT-8B6 complexes immediately prior to vitrification. Quantifoil holey carbon gold grids, 2.0/2.0 mm, size/hole space, 200 mesh) were glow discharged for 60 s at 15 mA. SERT-8B6 Fab complex (2.5 ml) was applied to the grid followed by blotting for 2 s in the vitrobot and plunging into liquid ethane cooled by liquid N 2 .

Cryo-EM data collection and processing
Images were acquired using the automated program SerialEM (Mastronarde, 2005) on a FEI Titan Krios transmission electron microscope, operating at 300 keV and equipped with a Gatan Image Filter with the slit width set to 20 eV. A Gatan K3 direct electron detector was used to record movies in super-resolution counting mode with a binned pixel size of 0.648 Å per pixel. The defocus values ranged from À0.8 to À2.2 mm. Exposures of 1.0-1.5 s were dose fractioned into 40 frames, resulting in a total dose of 54-60 e À Å À2 . Movies were corrected for beam-induced motion using MotionCor2 (Zheng et al., 2017) with 5 Â 5 patching. The contrast transfer function (CTF) parameters for each micrograph was determined using ctffind4 (Rohou and Grigorieff, 2015) and particles were picked either using DoG-Picker (Voss et al., 2009) or blob-based picking in cryoSPARC (Punjani et al., 2017). DoG or cryoSPARC picked particles were independently subjected to 3D classification against a low-resolution volume of the SERT-8B6 complex. After sorting, the DoG and cryoSPARC picked particles were combined in RELION (Scheres, 2012) and the duplicate picks were removed (particle picks that are less than 100 Å of one another were considered duplicates). Combined particles were further sorted using reference-free 2D classification in cryoSPARC, followed by refinement in RELION and further 3D classification. Particles were then re-extracted (box size 400, 0.648 Å per pixel) and subjected to non-uniform refinement in cryoSPARC. Local refinement was then performed in cisTEM (Grant et al., 2018) with a mask that excludes the micelle and Fab constant domain to remove low-resolution features. The high-resolution refinement limit was incrementally increased while maintaining a correlation of 0.95 or better until no improvement in map quality was observed. The resolution of the reconstructions was accessed using the Fourier shell correlation (FSC) criterion and a threshold of 0.143 (Rosenthal and Henderson, 2003). Map sharpening was performed using local sharpening in PHENIX.

Cryo-EM model building and refinement
A starting model was generated by fitting the X-ray structure of SERT-8B6 Fab paroxetine complex (PDB code: 6AWN) into the cryo-EM reconstruction in Chimera (Pettersen et al., 2004). Several rounds of manual adjustment and rebuilding were performed in Coot (Emsley and Cowtan, 2004), followed by real space refinement in PHENIX. For cross-validation, the FSC curve between the refined model and half maps was calculated and compared to prevent overfitting. Molprobity was used to evaluate the stereochemistry and geometry of the structures (Chen et al., 2010).

Radioligand binding and uptake assays
Competition binding experiments were performed using scintillation proximity assays (SPA) (Green et al., 2015;Coleman et al., 2016b). The assays contained~10 nM SERT, 0.5 mg/ml Cu-Ysi beads in TBS with 1 mM DDM, 0.2 mM CHS, and 10 nM [ 3 H]citalopram and 0.01 nM-1 mM of the cold competitors. Experiments were measured in triplicate. The error bars for each data point represent the s.e.m. Ki values were determined with the Cheng-Prusoff equation (Cheng and Prusoff, 1973) in GraphPad Prism. Uptake was measured as described previously in 96-well plates with [ 3 H]5-HT diluted 1:100 with unlabeled 5-HT. After 24 hr, cells were washed into uptake buffer (25 mM HEPES-Tris, pH 7.0, 130 mM NaCl, 5.4 mM KCl, 1.2 mM CaCl 2 , 1.2 mM MgSO 4 , 1 mM ascorbic acid and 5 mM glucose) containing 0.001-10,000 nM of the inhibitor. [ 3 H]5-HT was added to the cells and uptake was stopped by washing cells rapidly three times with uptake buffer. Cells were solubilized with 1% Triton-X100, followed by the addition of 200 ml of scintillation fluid to each well. The amount of labelled 5-HT was measured using a MicroBeta scintillation counter. Data were fit to a sigmoidal dose-response curve. Additional files  Appendix 1
PivOH and a,a,a-trifluorotoluene were purchased from Sigma-Aldrich Company Ltd and used as supplied.
K 2 CO 3 was purchased from Sigma-Aldrich Company Ltd and flame-dried before use as part of reaction set-up.
A flame-dried reaction tube was charged with amide (-)-S3a (102 mg, 0.20 mmol, one equiv), followed by di-tert-butyl dicarbonate (Boc 2 O, 175 mg, 0.80 mmol, four equiv) and 4-(dimethylamino) pyridine (DMAP, 4.9 mg, 0.04 mmol, 20 mol %). The reaction vessel was sealed with an aluminum cap (with molded butyl septa) and purged with argon, then anhydrous MeCN (400 mL, 0.5 M) was added by syringe. The mixture was then stirred at 35˚C for 22 hr. The reaction mixture was then allowed to cool to rt and sat. aq. NH 4 Cl (1 mL) and CH 2 Cl 2 (1 mL) were added. The phases were separated, and the aqueous layer was extracted with CH 2 Cl 2 (3 Â 5 mL). The combined organic extracts were dried over Na 2 SO 4 and filtered. The solvent was removed under reduced pressure to afford the crude N-Boc protected piperidine derivative.
A flame-dried round-bottom flask was charged with amide (+)À7a (565 mg, 1.11 mmol, one equiv), followed by di-tert-butyl dicarbonate (Boc 2 O, 969 mg, 4.44 mmol, four equiv) and 4-(dimethylamino)pyridine (DMAP, 26.9 mg, 0.22 mmol, 20 mol %). The reaction vessel was sealed with an aluminum cap (with molded butyl septa) and purged with argon, then anhydrous MeCN (3.7 mL) and anhydrous CH 2 Cl 2 (0.5 mL) were added by syringe. The mixture (0.3 M) was then stirred at 35˚C for 22 hr. The reaction mixture was then allowed to cool to rt and sat. aq. NH 4 Cl (5 mL) and CH 2 Cl 2 (5 mL) were added. The phases were separated, and the aqueous layer was extracted with CH 2 Cl 2 (3 Â 10 mL). The combined organic extracts were dried over Na 2 SO 4 and filtered. The solvent was removed under reduced pressure to afford the crude N-Boc protected piperidine derivative. This crude was solubilized in anhydrous THF (3.5 mL, 0.3 M) and the resulting solution was added dropwise to a suspension of LiAlH 4 (84.2 mg, 2.22 mmol, two equiv) in anhydrous THF (2.0 mL, 1.0 M) at 0˚C under argon atmosphere. The mixture was then stirred at 20˚C for 30 min. The reaction mixture was then quenched by slow addition of sat. aq. NH 4 Cl (5 mL) at 0˚C and stirred at rt for 30 min. The resulting suspension was filtered through a pad of Celite, eluting with EtOAc (3 Â 10 mL). The phases were separated, and the aqueous layer was extracted with EtOAc (3 Â 10 mL). The combined organic extracts were dried over Na 2 SO 4 and filtered. The solvent was removed under reduced pressure. Purification by flash column chromatography (10% to 20% acetone/hexane) afforded primary alcohol (-)À8a as a white solid (316 mg, 77% over two steps, 98.1% ee).
Alcohol (-)À8a (280 mg, 0.76 mmol, one equiv) and triethylamine (147 mL, 1.10 mmol, 1.4 equiv) were added to a flame-dried round-bottom flask, dissolved in anhydrous CH 2 Cl 2 (4.0 mL, 0.2 M) and cooled down to 0˚C. Methanesulfonyl chloride (75 mL, 0.97 mmol, 1.3 equiv) was then added by Gilson pipette. After stirring 5 min at 0˚C, the reaction mixture was stirred at 25˚C for 2 hr, then diluted with CH 2 Cl 2 (5 mL) and sat. aq. NaHCO 3 (5 mL). The phases were separated, and the aqueous layer was extracted with CH 2 Cl 2 (3 Â 10 mL). The combined organic extracts were dried over Na 2 SO 4 and filtered. The solvent was removed under reduced pressure to afford the crude mesylated alcohol derivative.
NaH (60% dispersion in mineral oil, 51.8 mg, 1.30 mmol, 1.7 equiv) was added to a solution of sesamol (168 mg, 1.20 mmol, 1.6 equiv) in anhydrous THF (4.0 mL, 0.3 M) at 0˚C. The mixture was then stirred at 25˚C for 1 hr. A solution of the crude mesylated alcohol in anhydrous THF (5.0 mL, 0.1 M) was then added dropwise to this suspension. The resulting mixture was stirred at 70˚C for 18 hr. The reaction mixture was then quenched by addition of H 2 O (5 mL) and diluted with EtOAc (5 mL). The phases were separated, and the aqueous layer was extracted with EtOAc (4 Â 10 mL). The combined organic extracts were dried over Na 2 SO 4 and filtered. The solvent was removed under reduced pressure. Purification by flash column chromatography (5% acetone/pentane) afforded piperidine (-)À9a as a white solid (225 mg, 60% over two steps).