Introduction

Serotonin is a neurotransmitter that modulates multiple fundamental brain functions that include memory, learning, sleep, pain, mood, and appetite¹. The serotonin transporter (SERT) removes serotonin from synaptic, perisynaptic and extracellular regions by harnessing the energy from sodium and chloride transmembrane gradients, diminishing local serotonin concentrations and thus terminating serotonergic neurotransmission². Congruent with the crucial roles of serotonergic signaling in neurophysiology, dysfunction of SERT has profound consequences associated with neurological disease and disorders, including Parkinson’s disease, seizures, depression, epilepsy, and attention deficit hyperactivity disorder^2,3.

SERT is a member of the large neurotransmitter transporter family, also known as neurotransmitter sodium symporters (NSSs). Additional members of the NSS family include transporters for norepinephrine (NET), dopamine (DAT), glycine (GlyT1 and GlyT2), and γ-aminobutyric acid (GAT), as well as for betaine and creatine². The NSSs are made up of 12 transmembrane helices organized topologically into two inverted repeats that, in turn comprise a conserved three-dimensional fold known as the LeuT fold⁴. Substrate transport by NSSs is described by an alternating access mechanism⁵ in which the substrate is translocated from extracellular to intracellular spaces^6-12.

SERT is a longstanding pharmacological target for antidepressant drugs¹³, as well as for psychostimulants such as cocaine, amphetamine, and methamphetamine¹⁴. The therapeutic utility of the drugs that act on SERT, including the selective serotonin reuptake inhibitors (SSRIs), is a consequence of their specific action on SERT resulting in their relative lack of inhibition of the closely related DAT and NET transporters. By contrast, illicit, psychoactive drugs such as cocaine and amphetamines also inhibit DAT and NET, and as a consequence, they have pleotropic effects on the neurotransmitter reuptake systems, thus explaining their psychoactive and deleterious effects on neurophysiology and behavior¹³. The potent and widely abused psychostimulants amphetamine, methamphetamine, and cocaine as well as synthetic cocaine derivatives, competitively inhibit the transport of neurotransmitters and lock the transporter in a transport inactive conformation, resulting in prolonged neurotransmission in the brain or promoting neurotransmitter efflux into the synaptic space². The primary mechanism of addiction is thought to be increased monoamine signaling in the central nervous system. The X-ray structures of a transport-inactive Drosophila melanogaster DAT (dDAT) NSS transporter in complex with cocaine, amphetamine, or methamphetamine have revealed that elicit psychostimulants bind at the central substrate-binding site to the transporter, consistent with their competitive inhibition¹⁵.

Multiple studies suggest that SERT interacts with a variety of intracellular scaffolding, cytoskeletal, anchoring, and signaling proteins. A prominent example includes syntaxin1A, a vesicle fusion SNARE protein, which has been shown to interact directly with the amino terminus of SERT and regulate its cell surface expression level¹⁶. By contrast, a neuronal nitric-oxide synthase (nNOS) that has a postsynaptic density of 95/discs-large/zona occludens (PDZ) domain interacts with the carboxy terminus of SERT, reducing its surface expression level and serotonin uptake capacity¹⁷. Hydrogen peroxide-inducible clone-5 (Hic-5) is a scaffolding protein that has been linked to SERT to aid in its internalization¹⁸. While protein-protein interactions regulate SERT function and subcellular distribution, the extent to which they form stable complexes for biochemical isolation is not well understood.

SERT also has cytoplasmic domains with numerous consensus sites for post-translational modification by protein kinases, phosphatases, and other interacting proteins that modulate its transporter function and cellular distribution¹⁹. SERT possesses two consensus sites for N-linked glycosylation within the extracellular loop 2 (EL2; Asn208, Asn217) whose glycosylation is related to cell-surface expression of the transporter²⁰. Protein kinase C (PKC) phosphorylates the cytoplasmic N- and C-terminal regions of SERT. Although no specific sites have been identified for this modification, PKC phosphorylation of SERT decreases the overall transport rate by promoting SERT redistribution from the plasma membrane to intracellular compartments²¹. Modification of SERT by the cGMP-stimulated protein kinase G (PKG) occurs on threonine residues, such as Thr276 (ref. 22). Interestingly, Thr276 is located on the intracellular end of TM5 and is only partially exposed, thus providing insight into how its modification is coupled to the stabilization of specific conformational states²². Calcium/calmodulin-dependent protein kinase II (CaMKII) activity has been reported to regulate the electrophysiological properties of the transporter by modulating its interaction with syntaxin1A¹⁶. Tyr phosphorylation has also been shown to regulate SERT function, and 5-HT uptake capacity into platelets, which is positively correlated with Src-mediated Tyr phosphorylation²³. Protein kinase A (PKA) mediated in vitro phosphorylation has also been reported for the isolated C- and N-terminal domains of SERT expressed as fusion proteins²¹. Although major efforts have been directed toward understanding the role of phosphorylation of SERT, more work is needed to understand the molecular basis of transporter regulation by phosphorylation.

The oligomerization states of SERT and related NSSs have been studied in the contexts of the plasma and organellar membranes^24,25. Radiation inactivation and mutagenesis studies provided the first glimpse into SERT oligomerization²⁶. Experiments with cross-linkers additionally suggested that rat SERT can form dimers and tetramers to varying degrees²⁷. Subsequent studies investigating the oligomerization state of NSSs have employed co-immunoprecipitation^28-30, Förster resonance energy transfer (FRET) measurements^30-36, and fluorescence lifetime imaging microscopy³⁵. Many of these studies were interpreted with an oligomerization model where the transporters form a variety of quaternary arrangements, ranging from monomers to multimers, differing to some extent depending on the specific NSSs^33,35,37-39. Membrane components such as phosphatidylinositol 4,5-bisphosphate (PIP2) and other lipids also have been implicated in the formation of NSS oligomers^40,41, presumably via lipid interactions to the transmembrane part of the protein, while psychostimulants such as methamphetamine and amphetamine have been shown to influence transporter oligomerization through an unknown mechanism^34,42,43.

Despite extensive experimental data from a broad range of biochemical, biophysical and computational studies^44,45 that have been interpreted in terms of SERT oligomers, there has been no direct evidence for its oligomerization state based on the purified transporter isolated from a native source. Here, we develop methods to extract nSERT from porcine brain tissue using the high-affinity 15B8 Fab, in the presence of methamphetamine and cocaine, respectively, allowing us to study the purified complex using fluorescence-detection size-exclusion chromatography (FSEC). We then carry out high-resolution, single-particle cryo-electron microscopy (cryo-EM) reconstructions, together with computational studies, to probe the conformation of psychostimulant-bound transporter and its interaction with lipids of a native cell membrane.

Results

Purification and cryo-EM of the native transporter

To isolate nSERT, we exploited the 15B8 Fab¹⁰, an antibody fragment that binds to a tertiary epitope of human SERT (hSERT), yet does not hinder the binding of ligands or the transport activity. We hypothesized that because porcine SERT and hSERT are closely related in amino acid sequence, the 15B8 Fab would also bind to porcine SERT and could serve as a powerful tool for immunoaffinity isolation of the transporter. We thus engineered the 15B8 Fab with a carboxy terminal mCherry fluorophore and an affinity tag (Fig. 1a).

Fig. 1

To isolate nSERT from porcine brain membranes, we next explored a wide range of membrane protein solubilization conditions, aiming to extract the transporter under the mildest conditions while retaining as much surrounding native lipid as possible. We thus first attempted solubilization in the presence of styrene-maleic acid (SMA) co-polymer⁴⁶ or recently developed amphipols⁴⁷. Unfortunately, neither yielded a measurable amount of nSERT as shown by FSEC analysis (Fig. 1b). We then examined the classical non ionic detergents, n-dodecyl-β-D-maltoside (DDM) together with cholesterol hemisuccinate (CHS) or digitonin, based on their utility in extraction of recombinant SERT¹⁰. Surprisingly, a peak for the SERT-15B8 Fab-mCherry complex was only observed for the DDM/CHS mixture (Fig. 1b). We therefore utilized DDM/CHS in all subsequent studies. To isolate the nSERT from porcine brain tissue, we incubated the solubilized membranes with an excess of the 15B8 Fab-mCherry protein, as well as with saturating concentrations of either methamphetamine or cocaine. The transporter was purified by affinity chromatography, followed by FSEC and the manual collection of fractions (Fig. 1c). Analysis of the isolated material by western blot revealed a band migrating with an apparent mass of 75kDa (Fig. 1d), consistent with the estimated mass of nSERT. The well-resolved and symmetrical FSEC peak indicated that the purified nSERT 15B8 Fab-mCherry complex was monodisperse. The elution volume with FSEC of nSERT 15B8 Fab complex was consistent with the recombinant ts2-15B8 complex (Fig. 1c), indicating the purified nSERT was a monomer.

To explore the function of nSERT, we carried out saturation binding experiments using the high-affinity SSRI [³H]paroxetine and determined a dissociation constant (K_d) of 6.5 ± 1.3 nM (Fig. 1e), consistent with previous measurements⁸. To characterize methamphetamine and cocaine binding with nSERT, we next performed competition experiments, similarly employing [³H]paroxetine, and measured inhibitory constants (K_i) of 199 ± 103.4 µM (Fig. 1f) and 179 ± 68 nM (Fig. 1g), respectively, thus indicating that both methamphetamine and cocaine compete for [³H]paroxetine binding, consistent with the psychostimulants binding to the central site. The potency of methamphetamine and cocaine on nSERT differs by a factor of ∼1,000, in accord with previous studies, whereas the two ligands are equally potent on DAT⁴⁸, underscoring the differences in residue composition and plasticity of the central binding pockets of SERT and DAT. Taken together, the ligand binding data illustrate that our purification method yields native transporter fully active in ligand binding.

From one pig brain we obtained ∼20 μg of purified protein in a volume of 200 μL, which was sufficient for visualizing particles on continuous carbon-coated grids under cryogenic conditions. We then collected single-particle cryo-EM data sets and carried out reconstructions of the methamphetamine- and cocaine-bound SERT complexes, obtaining density maps that extended to approximately 2.9 and 3.3 Å resolutions, respectively (Supplementary Fig. 1-5 and Supplementary Table 1, Supplemental Fig. 6a-b). Thorough 2D and 3D classifications yielded a single class for each data set in which nSERT is found as a monomeric entity, with no evidence of dimers or higher ordered oligomers (Supplementary Fig. 1-2), consistent with the molecular size of SERT estimated by FSEC. Overall, the density maps are of sufficient quality to assign most of the amino acid side chains, identify additional non-protein density within the central binding site for bound ligands, and indicate the presence of bound CHS molecules surrounding the transporter transmembrane domains (Fig. 2).

Fig. 2

Psychostimulant occupancy of the central site

Methamphetamine binds to the central site of the nSERT complex, adopting a similar binding pose to that observed in DAT¹⁵, lodged between the aromatic groups of Tyr213 and Tyr132. The amino groups of methamphetamine interact with Ser475 and form hydrogen bonds with the carboxylate of Asp135, and the main chain carbonyl of Phe372, as seen with the hydrogen bonds formed between the amino group of methamphetamine, and the equivalent Asp46 residue, and Phe319 residue in DAT¹⁵ (PDB code: 4XP6). The side chain of Phe378 forms edge-to-face aromatic interactions with the phenyl group of methamphetamine (Fig. 3a, Supplementary Fig. 6c). Comparison of the positions of TM1, TM6, and the extracellular gate to the equivalent elements of serotonin-bound, outward-open SERT complex⁴⁹ (Protein Data Bank (PDB) code: 7LIA, RMSD: 0.606 Å), indicates that SERT-15B8 Fab-methamphetamine complex adopts an outward-open conformation (Fig. 3b, Supplementary Fig. 6d), which is consistent with previous structural studies of dDAT in complex with methamphetamine¹⁵.

Fig. 3

The structure of the SERT-cocaine complex displays an outward-occluded conformation, with cocaine occupying the entire central binding pocket with an overall similar pose to the dDAT-cocaine complex¹⁵. The nearly complete filling of the volume of the central site by cocaine perhaps accounts for the higher affinity of SERT for cocaine compared with methamphetamine, reminiscent of how increasing the volume of serotonin via methylated analogs can enhance ligand binding⁴⁹. The benzoyl moiety of cocaine is accommodated between TM3 and TM8, where it forms van der Waals interactions with Ile209, Tyr213, Phe378, and Thr476. The methyl ester group protrudes into the base of the extracellular vestibule and the tropane rings are bordered by Tyr132, Ala133, Asp135, Phe372, and Ser475. Interestingly, the side chain of Phe372 undergoes substantial displacement and moves further into the central site than seen in the dDAT complex¹⁵. This reorganization translates into the closure of the inner gate from the extracellular side, blocking the release of cocaine from the central site and ultimately occluding the binding pocket, a conformation that was not seen in the cocaine-bound dDAT structure (Fig. 3c, Supplementary Fig. 6e). The overall structure of SERT-15B8 Fab-cocaine complex is similar to 5-HT bound recombinant hSERT in its outward open structure (PDB code: 7LIA, RMSD: 0.623), except for the orientation of Phe372. Because the rotation of Phe372 closes the extracellular gate, we define the conformation of the SERT-15B8 Fab-cocaine complex as an outward-occluded state (Fig. 3d, Supplementary Fig.6f).

Taken together, the structures of SERT in complexes with methamphetamine and cocaine show how the shape, chemical composition, and plasticity of the binding site enables the transporter to recognize ligands of different shapes and sizes. The distinct orientations of Phe372 in the cocaine-bound nSERT and recombinant dDAT structures emphasize the role of the central site residues in defining ligand binding affinity, which can be expanded to understanding variations in the pharmacological profiles between biogenic amine transporters.

CHS sites surround the TMD

SERT is an integral membrane protein embedded in a complex neuronal membrane composed of phospholipids, sphingolipids, and cholesterol⁵⁰. SERT^50-52, NET^53,54, DAT^55-57, and GlyT^58,59, as well as some excitatory amino acid transporters⁶⁰ associate with cholesterol in brain tissues or in transfected cell lines. Cholesterol is implicated in a variety of biological processes, including membrane protein organization and compartmentalization within the membrane. It is also known to play a key indirect role in modulating neurotransmission via its effects on the activities of DAT⁶¹ and SERT⁵⁰. Indeed, depletion of cholesterol from membranes affects the function of neurotransmitter transporters^61,62. Previous molecular dynamics (MD) studies revealed six potential cholesterol binding sites in SERT, defined as CHOL1-6⁶³. Bound CHOL has been observed at the CHOL1 binding site in dDAT structures¹⁵. The cholesterol analog, cholesteryl hemisuccinate (CHS), has been found to bind at the CHOL2 binding site in dDAT^15,64, as well as to hSERT¹⁰, and CHS has also been observed at the CHOL3 binding site in hSERT⁸. In investigating the interactions of CHOL with nSERT, we carefully examined the density maps of methamphetamine- and cocaine-bound SERT complexes, and the quality of the density maps enabled the identification of CHS at the CHOL1 and CHOL2 binding sites in both of the structures (Fig. 4a-b, d-e), consistent with previous observations.

Fig. 4

We also identified a non-protein density in the allosteric site for both complexes (Fig. 2), a binding site for a broad spectrum of ligands^8,49,65 (Supplementary Fig. 6g-i). The overall shape of the density resembles a lipid molecule. Because the local resolutions of the density maps within the allosteric site are not sufficient for unambiguous molecular identification, we used molecular dynamics (MD) simulations to examine the binding of the most abundant lipid molecules, steroids, or fatty acids to the allosteric site. We performed all-atom MD simulations of the cocaine-SERT complex with the allosteric site occupied by either docosahexaenoic acid (DHA), in charged or neutral forms (DHA^- or DHA⁰, respectively), CHOL, or CHS, in triplicate for each lipid species. In all three independent simulations, DHA^- remained bound at the site (center-of-mass displacement ≤ 3 Å), whereas in all simulations with neutral DHA⁰, CHOL, or CHS, the ligand unbinds from its binding site within nanoseconds, as highlighted by large center-of-mass displacements (> 5 Å) (Fig. 4g, Supplementary Fig. 7a). We further performed bias-exchange umbrella sampling simulations to calculate binding free energy profiles for DHA^- and CHOL, verifying preferential binding of DHA^- to the allosteric site, whereas CHOL binding to this site appears to be accompanied with a large penalty in free energy (Fig. 4h, Supplementary Fig. 7b-c). This suggests the allosteric site is not a preferred cholesterol-binding site and instead accommodates fatty acid binding. We thus fit DHA into this site (Fig. 4c and f). The DHA molecule adopts a curled orientation, with the flexible fatty acid tail protruding between TM11 and TM12, and the carboxylate group extending into the extracellular vestibule (Fig. 4c and f). The role of lipid binding within the allosteric site awaits further elucidation.

Discussion

Despite indirect experimental results suggesting that monoamine transporters might form oligomeric quaternary complexes²⁵, there is no direct determination of the oligomerization state for these proteins using purified native transporters isolated under mild conditions. In this study, we used immunoaffinity purification to isolate the native porcine SERT, proceeding to solve its cryo-EM structure in complex with the 15B8 Fab. Together with FSEC data, we show that nSERT is best described as a monomer rather than as a dimer or a multimer (Fig. 1-2). Previous biochemical studies suggested that transmembrane helix (TM) 11 and TM12 form oligomeric interfaces in hSERT, and also suggested potential contributions by TM5 and TM6 (ref.32). By contrast, the x-ray structure of SERT indicates that the kinked TM12 and the additional C-terminal helix protruding into the membrane preclude dimerization of SERT via a LeuT-like, TM12 interface⁴. Furthermore, subsequent structural studies revealed that TM5 and TM6 are directly involved in the transporter’s conformational transitions between different functional states, and thus we speculate that their required flexibility is likely to incompatible with the formation of dimers or multimers. Thus, our findings provide support for monomeric SERT in its native environment. Nevertheless, our study does not exclude the possibility of dimeric or multimeric arrangements present in the native membrane, perhaps the loss of lipids, such as PIP2, during extraction with detergent results in an exclusively monomeric transporter. Thus, further studies are needed to address the roles that membrane constituents may play in oligomer formation.

Previous structural studies of dDAT complexed with methamphetamine, cocaine, or their analogs provided a structural framework for showing how addictive psychostimulants stabilize the transporter in an outward-open conformation¹⁵. Here, we employed native SERT and single particle cryo-EM to study amphetamine and cocaine binding. We find that compared to the corresponding dDAT structures, methamphetamine and cocaine have similar binding site locations and interactions at the central site of SERT. There are differences in the transporter conformations, however, with cocaine inducing an outward-occluded conformation of SERT, caused by Phe372 rotating ‘inward’, to cover the tropane ring of cocaine, thereby blocking the release of the ligand from the central site, a conformational change not seen in the transport inactive, cocaine-bound dDAT structures (Fig. 3).

CHOL is an important constituent of eukaryotic plasma membranes and modulates the function of neurotransmitter transporters. It is required for optimal reconstitution of the GABA transporter⁶⁶ and is implicated in the function of SERT⁵⁰ via direct CHOL-protein interactions, as well as in the activity of DAT^57,61, NET^53,54, and GlyT2⁵⁸. In the case of SERT, CHOL modulates its functional properties by enhancing substrate transport and antagonist binding⁵⁰. MD simulations show that CHOL molecules are embedded in multiple sites of SERT⁶³, three of which have been confirmed by structural studies^8,10,15,64. Here we discovered two CHOL binding sites in nSERT. In addition, we also observed a non-protein density in the serotonin allosteric site (Fig. 4). Although this region of TM10, TM11, and TM12 has been indicated to be a potential CHOL-binding site, our MD simulations suggest that CHOL cannot stably bind within this site. We instead find that DHA is well accommodated into the experimental density, and is stably bound as determined by MD simulations and free energy calculations. Conclusive determination of the native lipid molecules that bind to the allosteric site, in addition to DHA, awaits further study.

Using biochemical analysis and cryo-EM, we observed under conditions of mild non-ionic detergents, the native, mammalian nSERT is isolated as a monomeric species, without interacting proteins, yet bound with multiple CHOL and lipid molecules. Nevertheless, the data presented in this study do not completely exclude the possibility of SERT oligomers in the native membrane. We investigated amphetamine and cocaine binding to nSERT and discovered that both ligands occupy the central site, where they are involved in numerous interactions with surrounding residues. Our studies of native SERT, in complex with addictive drugs provides a strategy for the study of native monoamine transporters. In summary, the nSERT complexes demonstrate the mechanism of psychostimulant inhibition and shed light onto the modulation of NSSs by illicit substances and the interactions of lipids with the transmembrane domain, particularly within the allosteric binding site.

Methods

Antibody purification

The 15B8 Fab construct⁸ was cloned into the pFastBac-dual vector, including a GP64 signal sequence. A mCherry tag, followed by a twin Strep [TrpSerHisProGlnPheGluLys(GlyGlyGlySer)2GlyGlySerAlaTrpSerHisProGlnPheGlu Lys] and a His₁₀ purification tag, were fused to the C-terminus of the heavy chain.Baculovirus was prepared according to standard methods. The Sf9 cells were infected by the recombinant baculovirus at a cell density of 2 × 10⁶ ml^-1 at 27 °C. The culture supernatant was then collected 96 h after infection by centrifugation at 5,000 rpm for 20 min using a JLA 8.1000 rotor at 4 °C. The 15B8 Fab was purified from Sf9 supernatant by metal ion affinity chromatography followed by size exclusion chromatography.

Isolation of nSERT

One pig brain (∼150 g) was homogenized with a Dounce homogenizer in ice-cold Tris-buffered saline buffer (TBS; 20 mM Tris-HCl, pH 8.0, 150 mM NaCl) supplemented with 1 mM PMSF, 0.8 µM aprotinin, 2 µg ml^-1 leupeptin, and 2 µM pepstatin. The homogenized brain suspension was then sonicated using a sonicator equipped with a tip size of 1.27 cm, for 15 min with 3 s on and 5 s off, at medium power, on ice. The resulting solution was then clarified by centrifugation for 20 min at 10,000g at 4 °C, the supernatant was collected and applied for further centrifugation at 40,000 rpm for 1 h at 4 °C (45 Ti fixed-angle rotor, Beckman) to pellet the membranes. The membranes were resuspended in 40 ml ice-cold TBS and further homogenized with a Dounce homogenizer. The membranes were solubilized in 100 ml ice-cold TBS containing 20 mM n-dodecyl-β-D-maltoside (DDM) and 2.5 mM cholesteryl hemisuccinate (CHS) in the presence of 1 mg of 15B8 Fab, 100 µM methamphetamine or 10 µM cocaine, for 1 h at 4°C. The lysate was centrifuged at 40,000 rpm for 50 min at 4°C (45 Ti fixed-angle rotor, Beckman) and the transporter-Fab complex was isolated by affinity chromatography using Strep-Tactin resin. The complex was further purified by fluorescence detection size exclusion chromatography (FSEC)⁶⁷ on a Superose 6 Increase 10/300 column in a buffer composed of 20 mM Tris-HCl (pH 8) supplemented with 100 mM NaCl, 1 mM DDM, 0.2 mM CHS, and 100 µM methamphetamine or 10 µM cocaine. The peak fraction containing the nSERT-Fab complexes was collected and used for biochemical and single particle cryo-EM analysis.

Western blot analysis

Purified nSERT was run on a SDS-PAGE gel and subsequently transferred to a nitrocellulose membrane. Antibodies used for detection were 10F2, a monoclonal antibody generated in house, and which can recognize the linear epitope of SERT. An IRDye 680RD anti-mouse secondary antibody (LI-COR), was used for visualization. Blots were developed from the secondary antibody at a ratio of 1:10,000 and imaged by Odyssey® DLx Imaging System.

Radioligand binding assay

A saturation binding experiment using [³H]paroxetine was performed via the scintillation proximity assay (SPA)⁶⁸ using the lysate of pig brain membranes in SPA buffer (20 mM Tris-HCl, pH 8, 100 mM NaCl, 1 mM DDM, 0.2 mM CHS). The membrane lysates were mixed with Cu-YSi beads (0.5 mg ml^-1) in SPA buffer, and [³H]paroxetine at a concentration of 0.3 to 40 nM. Nonspecific binding was estimated by experiments that included 100 μM cold S-citalopram. Data were analyzed using a single-site binding function.

Methamphetamine and cocaine competition binding experiments were performed using SPA with Cu-YSi beads (0.5 mg ml^-1) in SPA buffer. For the methamphetamine competition assays, SPA was performed with Strep-purified nSERT, 10 nM [³H]paroxetine, and 1 μM to 100 mM cold methamphetamine. For the cocaine competition assays, SPA was done with Strep-purified nSERT and 10 nM [³H]paroxetine in the presence of 1 nM to 1 mM cold cocaine. Experiments were performed in triplicate. The error bars for each data point represent the standard error of the mean (SEM). K_i values were determined with the Cheng-Prusoff equation⁶⁹ in GraphPad Prism.

Cryo-EM sample preparation and data acquisition

The purified nSERT-15B8 Fab complex was concentrated to 0.1 mg ml^-1, after which either 10 mM methamphetamine or 1 mM cocaine, together with 100 μM fluorinated n-octyl-β-D-maltoside (final concentration) were added prior to grid preparation. A droplet of 2.5 μl of protein solution was applied to glow-discharged Quantifoil 200 or 300 mesh 2/1 or 1.2/1.3 gold grids covered by 2 nm of continuous carbon film. The grids were blotted for 2.0 s at 100% humidity at 20°C, followed by plunging into liquid ethane cooled by liquid nitrogen, using a Vitrobot Mark IV. The nSERT datasets were collected on a 300 kV FEI Titan Krios microscope located at the HHMI Janelia Research Campus, equipped with a K3 detector, at a nominal magnification of 105,000x, corresponding to a pixel size of 0.831 Å. The typical defocus values ranged from -1.0 to -2.5 μm. Each stack was exposed for 4.0 s and dose-fractionated into 60 frames, with a total dose of 60 e^- Å^-2. Images were recorded using the automated acquisition program SerialEM⁷⁰.

Cryo-EM image processing

The beam-induced motion was corrected by MotionCor2 (ref.71). The defocus values were estimated by Gctf⁷² and particles were picked by blob-picker in cryoSPARC⁷³. After two rounds of 2D classification, 2D classes with clear secondary structures were selected. An initial model was then generated by cryoSPARC. The initial model was further used for the following heterogeneous refinement. A round of 3D classification without image alignment was performed in RELION-3.1 (ref.74), with a soft mask excluding the constant domain of 15B8 Fab and micelle. The selected particles were imported back to cryoSPARC for homogeneous refinement, local contrast transfer function (CTF) refinement, and nonuniform refinement. The local resolution of the final map was estimated in cryoSPARC.

For the nSERT-15B8 Fab complex in the presence of (+)-methamphetamine, 7,794,907 particles were picked in cryoSPARC, which after rounds of 2D classification and heterogeneous refinement, left 348,745 particles (binned to a 200-pixel box, 1.662 Å pixel^-1). Particles were reextracted (360-pixel box, 0.831 Å pixel^-1) and subjected for homogeneous refinement and nonuniform refinement in cryoSPARC, then subjected to 3D classification with 10 classes in RELION-3.1 without image alignment (360-pixel box, 0.831 Å pixel^-1). Particles from three classes with clear TM features were combined and subjected to homogeneous refinement, local CTF refinement, and nonuniform refinement in cryoSPARC, respectively (Supplementary Fig. 2).

For the nSERT-15B8 Fab complex with cocaine, a total of 7,560,137 particles were picked from 16,094 movies in cryoSPARC with a box size of 200 pixels (1.662 Å pixel^-1). After rounds of 2D classification and heterogeneous refinement, 338,343 particles were selected, re-extracted (400-pixel box, 0.831 Å pixel^-1), and subjected to homogeneous refinement, nonuniform refinement in cryoSPARC, and further subjected to 3D classification with 10 classes in RELION-3.1 without image alignment. Two well-resolved classes with 243,207 particles were combined and further refined in cryoSPARC with homogeneous refinement, local CTF refinement, and nonuniform refinement (Supplementary Fig. 3).

Model building and refinement

Interpretation of the cryo-EM maps exploited rigid-body fitting of the SERT-antibody complex models derived from previous cryo-EM studies. The outward-open ΔN72/C13 SERT-15B8 Fab complex with a 5-HT model (PDB code: 7LIA) was used as a reference. The initial model was generated via rigid-body fitting of the homology models to the density map in UCSF ChimeraX⁷⁵. The model was then manually adjusted in Coot⁷⁶. The model was further refined using real-space refinement in PHENIX⁷⁷. Figures were prepared in UCSF ChimeraX.

System preparation for MD simulations

We performed molecular dynamics (MD) simulations of the cocaine-SERT complex in a hydrated lipid bilayer to explore the identity of the ligand in the allosteric site ligand. Triplicates of four simulation systems were studied, with the allosteric site occupied by either docosahexaenoic acid (DHA) in charged or neutral forms (DHA^- or DHA⁰, respectively), cholesterol (CHOL), or cholesteryl hemisuccinate (CHS). The initial coordinates of CHS and DHA^- were transferred from experimental modeling, while CHOL was constructed into the CHS model, and DHA⁰ was constructed by protonating the DHA^- model using the PSFGEN plugin in VMD⁷⁸. The missing side chains and hydrogen atoms in the protein were added using the PSFGEN plugin in VMD⁷⁸. The co-crystalized antibody fragment was removed. All bound Na⁺ and Cl^- ions, the cocaine molecule, and the two CHOL molecules bound to transmembrane helices were retained. The allosteric site residue Phe532 was converted to the corresponding tyrosine in wildtype hSERT. Glu173 was modeled as a protonated side chain according to pKa calculations using PROPKA 3.0 (ref. 79). A disulfide bond was introduced between Cys237 and Cys246. Neutral N-terminal and C-terminal ‘caps’ were added to the first and last residue of the protein segment, respectively. All protein models were internally hydrated using the DOWSER plugin^80,81 of VMD and externally solvated using the SOLVATE program⁸². The models were then oriented according to the Orientations of Proteins in Membranes (OPM) database⁸³ and embedded in a lipid bilayer composed of 218 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and 94 CHOL molecules from CHARMM-GUI⁸⁴. The systems were next solvated and neutralized with a 150 mM NaCl aqueous solution in VMD⁷⁸, resulting in simulation systems of ∼160,000 atoms, with approximate dimensions of 112 Å × 112 Å × 120 Å before equilibration. Each simulation system was replicated into three independent copies with lipid distributions randomized by shuffling lipid molecules within each leaflet using the VMD plugin Membrane Mixer⁸⁵.

Equilibrium MD simulations

All simulations were performed using NAMD2^86,87 and the CHARMM36m force fields⁸⁷ for proteins, CHARMM36 force fields for lipids (including CHS, CHOL, DHA^- and DHA⁰)⁸⁸, and the TIP3P model for water⁸⁹, along with the NBFIX modifications for non-bonded interactions^90,91. The force field parameters for cocaine were obtained from the CGenFF server⁹². All simulations were carried out as isothermal-isobaric (NPT) ensembles under periodic boundary conditions and simulated in a flexible cell, whose dimensions could change independently while keeping a constant ratio in the xy (membrane) plane. A constant temperature of 310 K was maintained using Langevin dynamics with a 1.0-ps^-1 damping coefficient, and a constant pressure of 1.01325 bar was maintained with the Langevin piston Nosé-Hoover method⁹³. Non-bonded interactions were calculated in a pairwise manner with the 12-Å cut-off, and a switching function applied between 10 Å and 12 Å. Long-range, non-bonded interactions were calculated with the particle mesh Ewald (PME) method⁹⁴. Bond lengths involving hydrogen atoms were constrained using the SHAKE⁹⁵ and SETTLE algorithms⁹⁶. Simulations were integrated in 2-fs time steps, and trajectories were recorded every 10 ps.

The four simulation systems, each replicated into three independent copies were simulated following these steps: (1) 3,000 steps of energy minimization; (2) 15 ns of MD equilibration, during which Cα atoms, non-hydrogen atoms of ligands, and all bound ions were restrained by harmonic potentials with decreasing force constants (k = 5, 2.5, 1 kcal mol^-1 Å^-2 for 5 ns each) to allow for protein side chain relaxation and protein hydration; (3) 150 ns production MD run, during which harmonic potentials (k = 1 kcal mol^-1 Å^-2 for 5 ns each) were applied to only Cα atoms to avoid undesired protein conformational deviation but allowing free diffusion of the allosteric site ligand.

Free energy characterization of ligand binding

The bias-exchange umbrella sampling (BEUS) method⁹⁷ was employed to characterize the binding energy profiles of CHOL and DHA^- to the allosteric site. The ligand-TM1b/TM6a distance, measured as the center-of-mass distance between the non-hydrogen atoms in the ligand and Cα atoms in the extracellular ends of TM1b and TM6a (residues 145-148 and 361-364), was chosen as the reaction coordinate to sample ligand binding. The initial distances for the modeled CHOL and DHA^- were 17.2 and 15.2 Å, respectively. We chose reaction coordinates ranging from 14 to 21.5 Å and 15 to 22.5 Å for CHOL and DHA^-, respectively, to sample ligands unbinding from the allosteric site. Each reaction coordinate was divided into 16 windows with a spacing of 0.5 Å. The initial conformations in each window were captured from steered MD simulations using the COLVAR module⁹⁸ in NAMD, in which the ligand was pulled towards the desired distances using a harmonic potential (k = 10 kcal mol^-1 Å^-2) moving at a 0.5 Å ns^-1 rate. The BEUS simulations were performed for 60 ns in each window. The Hamiltonian replica exchange was attempted every 1 ps between neighboring windows. Weighted Histogram Analysis Method (WHAM)^99,100 was used to construct the free energy profiles and perform error analysis using the Monte Carlo bootstrapping method.

Supporting information

Supplementary figures

Data availability

The 3D cryo-EM density maps and molecular coordinates have been deposited in the Electron Microscopy Data Bank (EMDB) and Protein Data Bank (PDB) for the SERT-15B8-Fab-methamphetamine outward (EMD-27384; 8DE4) and SERT-15B8-Fab-cocaine outward-occluded (EMD-27383; 8DE3) reconstructions and structures, respectively.

Acknowledgements

We thank Rui Yan at the HHMI Janelia CryoEM Facility for help in microscope operation and data collection. A portion of this research was supported by NIH grant U24GM129547 and performed at the PNCC at OHSU and accessed through EMSL (grid.436923.9), a DOE Office of Science User Facility sponsored by the Office of Biological and Environmental Research. Simulations in this study have been performed using allocations at National Science Foundation Supercomputing Centers (XSEDE grant MCA06N060). This work was otherwise funded by the NIH (R01 MH070039 and P41 GM104601). E.G. is supported by Jennifer and Bernard LaCroute and is an investigator of the Howard Hughes Medical Institute.

Author contributions

D.Y. and E.G. designed the project. D.Y. performed all the experiments. D.Y. and E.G. wrote the manuscript. Z.Z. and E.T. performed molecular dynamics simulations, and wrote sections related to computational methods. All authors contributed to editing and manuscript preparation.

Competing interests

The authors declare no competing interests.