Structure of tetrameric forms of the serotonin-gated 5-HT3_A receptor ion channel

We purified the full-length murine 5-HT3_AR in a saposin lipoprotein-stabilized lipid bilayer and performed single particle cryo-EM, as we previously reported (Zhang et al, 2021). We imaged the receptor in the apo state and in the presence of serotonin, corresponding to the desensitized state (Appendix Fig. S1). 2D classification of the resulting particles surprisingly revealed that a fraction of proteins had tetrameric features (Fig. 1A). Consecutive 3D classification by CryoSPARC’s heterogeneous refinement (Punjani et al, 2017) without imposition of symmetry revealed that while _˜50% of particles adopted an expected previously described C5-symmetric arrangement (Zhang et al, 2021), other particles unexpectedly formed a tetramer (Fig. 1C,E). Cryo-EM maps of tetrameric 5-HT3_AR in apo closed-state and serotonin-bound conformations were obtained at resolutions of 3.5 Å and 3.4 Å (Appendix Fig. S1).

Further classification of the tetrameric particles, without the application of symmetry, resulted in two structures: a near-C2-symmetric (which we will refer to as symmetric) and an asymmetric tetramer (Fig. 1B,D). The differences in the conformations were observed mostly in the extracellular domain (ECD): in the asymmetric tetramer, the ECD is organized similarly to the pentameric form of the receptor with one missing subunit (Fig. 1B), while the second conformation assembles as an apparent dimer of dimers (Fig. 1D). Both conformations showed tightly packed transmembrane domains (TMD) with a C4 symmetry between the membrane leaflets. In all the maps of the tetramers, the local resolution of the TMD reached 2.8 Å and the ECD around 3.6 Å, while the density of the intracellular domain (ICD) was poorly ordered due to the higher flexibility of this domain compared to the pentameric form of the channel (Fig. 1B,D; Appendix Figs. S2 and S3). In the apo-dataset 60% of the particles represent an asymmetric tetramer (Fig. 1B; Appendix Fig. S2) and 40% the symmetric one (Fig. 1D; Appendix Fig. S2) while in the serotonin-bound dataset, the ratio was 53:47 (Appendix Fig. S3). We observed densities resembling N-linked glycosylation in the extracellular domains of the tetramers in correspondence to N82, N164, and N148 as it was previously reported for pentameric 5-HT3_AR (Hassaine et al, 2014) (Fig. 1B,D, in red).

Treatment of 5-HT3_AR with SDS revealed progressive dissociation of pentamers into lower oligomeric states, with more monomers at higher SDS concentrations (Fig. 1F). Interestingly, the circular dichroism (CD) spectrum of the mostly monomeric fraction is nearly-identical to the untreated pentameric form (Fig. 1G). While our cryo-EM data demonstrates that the tetramers maintain the overall conformation of monomers, the unfolding experiments suggest that lower-order oligomers maintain a stable conformation even in aggressive environments.

Our cryo-EM maps allowed refinement of the available atomic model (PDB ID: 6Y59 (Zhang et al, 2021)) into the transmembrane and extracellular domains, but not into the intracellular domains. Comparison of the chains constituting the symmetric apo tetramer to the atomic model of the C5-symmetric apo-state (PDB ID: 6Y59 (Zhang et al, 2021)) showed a very similar structure of the domains with the average root mean squared deviation of the carbon atoms in the main chain (Cα RMSD) of 0.64 Å to 1.07 Å (Fig. 2A–C). The highest deviation of ˜2 Å was detected in the helices M3 and in the generally more flexible horizontal MX helices (Fig. 2A).

The differences between the tetrameric conformations were larger in the extracellular domain (Fig. 2F). The extracellular domain of the chains C and D of the asymmetric tetramer are displaced by 6–14 Å outwards from the receptor’s central axis, leaving a larger space between the extracellular domains of chains A and D (Fig. 2D–F). The transmembrane domains of the tetrameric conformations were less variable, with displacements of 1–4 Å (Fig. 2F,G). Given the new unexpected tetrameric forms, we wondered whether the tetramers are assembly intermediates for the pentamer or are they able to perform their own functions? To address the function of tetramers, we performed molecular calculations based on our structures.

To assess whether the transition from one tetrameric form to another is energetically possible, we computed the translation free-energy surface (FES) expressed as Gibbs free energy (kJ/mol) at 310 K (Fig. 3A) using the “flexible axis” approach (Kutzner et al, 2011). The analysis of FES suggested that the transition from a symmetric (6.4 kJ/mol) to an asymmetric tetramer (3.6 kJ/mol) is energetically possible and is likely to happen spontaneously (ΔG = −2.8 kJ/mol ≈ −0.67 kcal/mol). We next probed the transition of the symmetric to the asymmetric form. For this we used guided MD simulations, sampling the time interval of 1000 ps (Fig. 3B). Molecular dynamics (MD) simulations showed that the solvent-accessible surface area reduces during the transition from the symmetric conformation to the asymmetric one from 690 to 650 nm² (Fig. 3B). The reduction of the solvent-accessible area and of the gyration radius for the asymmetric state highlights that the asymmetric state is more compact than the symmetric one (Fig. 3A). During the progression from the symmetric to the asymmetric form we observed a redistribution of the numbers of hydrogen bonds between the chains with a decrease of the number of H-bonds between chains A-B and C-D and an increase between chains B-C (Fig. 3C).

Both the symmetric and asymmetric tetramers have a narrower pore compared to the previously reported pentameric structures (PDB ID: 6Y5A = asymmetric 5-HT-bound, PDB ID: 6Y59 = symmetric apo and PDB ID: 6Y5B = asymmetric apo). The range of pore width of 1.2–3.0 Å (Fig. 3D) suggests that water molecules should not penetrate the pore. We tested the ability of the tetrameric forms to hydrate the pore with all-atom molecular dynamics simulations and measured the number of water molecules in the pore. While few water molecules were observed close to both the intracellular and extracellular exits from the pore, the central ion pore region (corresponding to the residue L260) of both the symmetric and the asymmetric tetramers did not contain any water molecules during the whole long-time scaled MD simulations (Fig. 3E, top panels). This analysis suggests that the captured structures are not permeable to water and ions.

Our dataset of 5-HT3_AR, recorded in the presence of serotonin and without extracellular calcium, should correspond to a desensitized state of the receptor. Structural analysis revealed densities for serotonin in the ligand binding pockets (LBPs): for the symmetric tetramer two pockets were occupied—between the chains A and B (LBP1) and C and D (LBP3) (Appendix Fig. S4). The distance between the extracellular domains for the other chain pairs B/C and A/D was too wide to form an LBP (Fig. EV1A; Appendix Fig. S4D). For the asymmetric tetramer, two bound serotonin molecules were observed—between the chains A and B (LBP1) and between the chains B and C (LBP2) (Figs. EV3H–J and 4A,C–E). The interface between the chains C/D at the extracellular side had a lower resolution (~4 Å) which was not sufficient to build an atomic model, therefore, we could not conclude if there is a serotonin molecule in the corresponding binding pockets. The distance between the chains D and A was too far to form a contact (Fig. 4A).

Upon serotonin binding, the receptor’s LBP side chains compactify, the loop C “caps” the LBP, and side chains of W63, R65, Y126, W156, and Y207 interact with the serotonin molecule (Zhang et al, 2021; Polovinkin et al, 2018; Basak et al, 2018a). In tetramers, the displacement of the C-loop and of the interacting residues were minimal for the LBPs of the asymmetric tetramer and LBP1 of the symmetric tetramer. The serotonin poses were significantly different compared to fully functional pentamers (Figs. 4G,H and EV1G,H). Only for LBP2 of the asymmetric tetramer, the serotonin pose was similar to the fully assembled pentamer (Fig. 4H). Side chains in LBP2, specifically W63, W156, Y207, and F199, rearranged closer to the serotonin molecule, showing high similarity to the LBP of the functional pentamer (Fig. 4H; Appendix Fig. S4).

We evaluated the stability of the interaction between serotonin (5-HT) and the LBPs using the PDBePISA server (Krissinel and Henrick, 2007). We calculated the solvation-free energy (ΔiG, kcal/mol) gain upon forming interfaces between a specific chain and the ligand, and the ΔiG P-value, an indicator of an interaction-specific interface (Figs. 4F and EV1F). ΔiG P-values < 0.5 indicate a specific molecular interaction due to higher hydrophobicity than expected for a non-specific interface. The previously reported C5-symmetric pentamer had an average ΔiG of −3 kcal/mol and a ΔiG P-value of 0.45. Similar values were only obtained for LBP2 of the asymmetric tetramer (LBP1 ΔiG −2.7 kcal/mol, ΔiG P-value 0.46, Fig. 4F,H). In contrast, the ΔiG for LBP1 of the asymmetric tetramer and both LBPs of the symmetric tetramer suggested less efficient binding (Figs. 4F,G and EV1F–H). Only for the asymmetric tetramer the calculated ΔiG P-value was significant (P = 0.48, Fig. 4F), suggesting a specific interaction. The ΔiG P-value calculated for LBP1 of the symmetric tetramer was >0.5 (i.e., 0.59, Fig. EV1F), suggesting a non-specific molecular interaction. No significant conformational changes in the transmembrane domains of the tetrameric forms were observed upon ligand binding (Figs. 4B and EV1B). Significant differences between the symmetric and asymmetric tetramer 5-HT-bound forms were identified primarily within the extracellular domain (Fig. EV2A–C). Chains A and B form the conserved LPB1, maintaining a stable conformation (Fig. EV2B,C). Notable alterations are observed within the extracellular domain of chains C and D (Fig. EV2B,C). The transmembrane domain exhibits a consistently stable arrangement across both serotonin-bound tetrameric forms (Fig. EV2D).

Structural analysis of the tetramers in the presence of serotonin and CaCl₂, which inhibits desensitization and allows the formation of the active state of the receptor, showed an overall resemblance to the ligand-unbound and -bound tetramers at 6 Å resolution (Fig. EV3). No noticeable displacements of the transmembrane helices were revealed, suggesting that the tetrameric receptors represent a closed ion pore (Fig. EV3E).

We investigated whether tetrameric forms could be delivered to the plasma membrane, potentially playing a functional role. Using a previously described plasma membrane-derived microvesicle system in human cell lines (Pick et al, 2005), we imaged 5-HT3_AR in microvesicles by cryo-ET. Tomograms contained distinguishable 5-HT3_AR-shaped densities (Fig. 5A–D). All observed 5-HT3_AR-containing vesicles in 309 tomograms showed the expected orientation when inserted in the plasma membrane. On the contrary, receptors localized at ER membranes would contain inverted protein topology (Duran and Meiler, 2013) and vesicles would occasionally contain ribosomes. The receptors were located in clusters, similar to previous findings in the plasma membrane of live HEK cells (Pick et al, 2003), with both pentameric and tetrameric receptors present (Fig. 5E). Examination of the inter-receptor distances indicated a clustering of pentamers in close proximity, while tetramers exhibited a more dispersed distribution with an average separation in the range of 30 nm (Fig. 5E). The tendency of pentamers to cluster aligns with previous observations in vivo within the plasma membrane of HEK cells (Pick et al, 2003).

We next performed subtomogram classification and averaging of the receptor-like densities without applying symmetry (Fig. EV4A–E). From the manually picked thirty thousand particles, a subset of 4431 particles generated a density map resembling the asymmetric tetramer observed in our single-particle analysis at a resolution of 25.2 Å (Figs. 5F,H and EV4E). Another subset of 3031 particles produced a density map depicting an apparently symmetrical pentameric receptor at a resolution of 19.6 Å (Figs. 5G–I and EV4D). Despite more particles contributing to the tetrameric isoform, the resolution was higher for the pentamer, suggesting the tetramer is more flexible. Both atomic models of tetramers from our single-particle cryo-EM analysis fitted well into the tetrameric map from subtomogram averaging (Fig. 5H).

We next computationally probed if the tetrameric forms of 5-HT3_AR could serve as intermediates for the assembly process of the pentameric receptors. For this, we used coarse-grained atomistic MD simulations in combination with metadynamics simulations, a widely used method for the sampling of rare biological events by analyzing free energy landscapes of biomolecules (Barducci et al, 2011, 2008; Bussi and Laio, 2020). Starting from monomers, we simulated the possible oligomerization steps that can lead to the formation of a final pentameric 5-HT3_AR (i.e., Monomer + Monomer = Dimer, Monomer + Dimer = Trimer, Trimer + Monomer = Tetramer, and so on, as shown in Fig. EV5A–F). We calculated free energy surfaces (FES) for given oligomerization states as a function of distance between extracellular regions and transmembrane regions (Fig. EV5A–F). FES analysis for the interaction between two monomers and between a dimer and a monomer, showed energy barriers of 10.2 and 6.9 kcal/mol respectively (Figs. 6 and EV5A, B). Two scenarios for tetramer formation are possible: a symmetric tetramer forms from two dimers with a barrier of 16.8 kcal/mol, or a potentially less symmetric tetramer forms from a trimer and a monomer with a barrier of 12.3 kcal/mol (Figs. 6 and EV5C,D). Overall, the global minimum of FES appeared at low distances between the subunits (ECD < 50 Å and TMD < 40 Å) for all the calculations, suggesting that the formation of lower oligomeric states is energetically favorable.

Interestingly, there were alternative local minima in the FES plots for the formation of tetramers and pentamers (Fig. EV5D,E). The FES for the formation of the tetramer from a trimer and a monomer shows two energy minima of −33.6 and −33.5 kcal/mol (Fig. EV5D, white and red circles). The structure of the symmetric tetramer observed in our cryo-EM data (Fig. EV5D, black circle) is a few Angstroms away from both those energy minima. The formation of pentamers from trimers and dimers also has two minima of FES which are further away from each other (Fig. EV5E, white and red circles). The minima closer to the observed structure (Fig. EV5D, black circle) corresponds to the energy minima of −46.2 kcal/mol. The major differences observed between the two alternative energy minima were observed in the ECD domain (Appendix Fig. S5). The energy barrier for the formation of a pentamer from a tetramer and a monomer is 11.1 kcal/mol (Figs. 6 and EV5F) and it is lower than the addition of a dimer to a trimer. Therefore, from the energy landscape point of view, the lowest energy valley for the assembly of a pentamer is the addition of monomers to the existing complex one by one (Fig. 6).

We next conducted guided molecular dynamics simulations to form oligomers from specific combinations of precursors. For this, we computationally pulled away one subunit from a particular assembly over a simulation time of 500 ns. Reversing this process on the time axis corresponds to the formation of a particular oligomer of the receptor starting by forming a dimer from the interaction of two monomers and then proceeding with the formation of higher-order oligomers (Fig. 7A,B). During the formation of different oligomeric forms, the initial contact area was similar across all simulations: the hydrophobic TM domains of the interacting subunits approached each other first, followed by a structural rearrangement of the ECD region (Fig. 7C–H). In dimer formation, the first contact between the two monomers occurred over the entire length of their M2 helices (Fig. 7C). For higher-order oligomers, the first contact between interacting subunits was at the inner leaflet of the membrane (Fig. 7D–H). A simulation of pentamer formation from an asymmetric tetramer is presented in Movie EV1.

Murine wild-type 5-HT3A receptor containing four N-terminal StrepII tags was expressed by a stable T-Rex-293 cell line. As previously performed and described by us (Zhang et al, 2021), protein expression was induced by adding tetracycline to the cultures and by lowering the shaker’s temperature to 30 °C. In order to increase protein expression levels (Chaudhary et al, 2012; Gorman et al, 1983), sodium butyrate was added shortly after the induction. The following steps were performed at 4 °C, unless differently specified. For the purification of the receptor, cells were resuspended in lysis buffer (10 mM HEPES, 1 mM EDTA, pH 7.4 plus 2 μg/mL leupeptin, 2 μg/mL pepstatin A, 0.2 mg/mL benzamidine-HCl, 20 μg/mL AEBSF, and 3 μg/mL aprotinin) using a dounce homogeniser and subsequently disrupted using a Microfluidizer processor (Microfluidics M-110L) at 80 psi. Cell debris was removed by centrifugation (10,000 × g, 45 min) and the membrane pellet was isolated from the clarified lysate by ultracentrifugation for 6 h at 130,000 × g. The membrane pellet was mechanically resuspended in solubilization buffer (final concentrations: 50 mM Tris, 500 mM NaCl, C12E9 (Anatrace) 0.75% (w/v), pH 8) and subsequently, it was left under gentle stirring for 2 h. Non-solubilised material was removed by centrifugation for 1 h at 49,000 × g. The supernatant was then passed through a 0.45-μm-pore filter and applied to a 5 mL Streptactin Superflow high-capacity column (IBA) (equilibration and washing buffer: 50 mM Tris, 150 mM NaCl, 0.01% C12E9, pH 8, flow rate of 1 mL/min). After the washing of the column with 250 mL of washing buffer, the receptor was eluted with the above-mentioned buffer supplemented with 10 mM D-desthiobiotin. The peak fractions were pooled and concentrated to ~300 μL for the further reconstitution of the channel into saposins and porcine brain polar lipids (BPL, Avanti Polar Lipids) (i.e., 5-HT3R-Salipro). Briefly, Once the lipids were prepared as previously described (Zhang et al, 2021), the purified channel was first incubated for 15 min at room temperature on an orbital rocker with a 450-fold molar excess of detergent-destabilized BPL (650 g/mol). Secondly, saposin was added to 30-fold molar excess relative to 5-HT3_AR and the sample was incubated for 15 min at room temperature. Subsequently, the sample was incubated for 15 min at room temperature with 100–200 mg of wet BioBeads SM2 resin (Bio-Rad), and this last step was repeated in total two times. The sample was then filtered through a 0.22 μm-pore centrifugal filter and injected into a Superose 6 Increase 10/300 column (GE Healthcare) (SEC buffer 25 mM HEPES, 125 mM NaCl, pH 7.4). SEC peak fractions were pooled together and sample concentration was adjusted upon checking by negative staining using a Tecnai Spirit Biotwin at 120 kV (ThermoFisher Scientific).

Blue native PAGE and circular dichroism (CD) measurements were performed as described previously (Tol et al, 2013). Electrophoresis: Pre-casted 5–20% gradient gels (Bio-Rad) were loaded with 10 μg of solubilized protein in sample buffer (final concentrations 10% glycerol, 0.2% Coomassie brilliant blue G-250 and 20 mM 6-aminohexanoic acid). NaCl concentration was kept below 25 mM, and the nonionic detergent concentration was not higher than 10 mM. Electrophoresis was performed at 4 °C, starting with 20 min at 25 V, until all samples entered the gel, followed by 2 h at 125 V, until the Coomassie front was leaking in the anode buffer. When the gel run was finished, excess Coomassie dye from empty detergent micelles was removed by destaining, and protein bands were accentuated by normal staining. The oligomers of BSA (ranging from 66 to 264 kD) were used as a molecular mass marker. CD measurements were performed as described before, on a model 62DS CD spectrometer (Aviv Biomedical, USA). Solubilized 5-HT3R was diluted to a concentration of 150 μg/ml in KPi buffer (10 mM potassium phosphate pH 7.4, with 0.025% (w/v) C12E9). In case the stock concentration was smaller than 2 mg/ml, the sample was dialyzed for several hours against the same buffer to prevent high UV-absorption buffer components. CD-spectra were recorded in 1 mm quartz cuvettes at a sampling interval of 1 nm, a bandwidth of 1.5 nm, and an average time of 5 s. Generally, spectra were measured from 260 to 190 nm. Spectra were smoothed by Savitzky-Golay filtering with 7 points and a second-order polynomial using Igor Pro (Wavemetrics) and converted to molar ellipticity values.

Cryo-EM grids were prepared in the presence (i.e., holo) and absence (i.e., apo) of natural agonist serotonin (5-HT). In this case, we used photo-caged serotonin (RuBi-5-HT, Aabcam) that was activated by exposure to light. To observe the active form of the channel (i.e., holo active), the receptor was incubated in ice for 1 h before freezing with 100 μM 5-HT and 2 mM CaCl₂. 3 µL (for apo and holo active samples) or 2 µL (for holo deactivated samples) of 5-HT3_AR reconstituted in Salipro (Frauenfeld et al, 2016) at a final concentration of 0.01–0.07 mg/mL were applied onto freshly glow-discharged (15 mA for 45 s in a PELCO easiGlow system) Cu/C, R2/2, 300 or 400 mesh copper grids (Quantifoil) coated with a ~2 nm thick carbon layer. For the holo-deactivated sample, 1 µL of 600 μM caged serotonin (RuBi-5-HT, Abcam, final concentration on grid 200 μM) was directly added to 2 μL of the sample directly on the grid. Caged serotonin was used for further research reasons. For this experiment, the two components were incubated in dark conditions for 1 min on the grid and the ligand was further released upon controlled illumination with a harmless LED light source operating at 445 nm (blue) inserted in one of the lateral apertures of Vitrobot Mark IV (Thermo Fisher Scientific) right before the grid fell into liquid ethane.

All samples were blotted for 4.5–5.5 s with blot force 20 (595 Whatman paper) at 4 °C in 100% humidity and plunge-frozen into liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific). Cryo-EM data were collected automatically using EPU software (Thermo Fisher Scientific) on a Titan Krios G3i microscope at 300 kV, equipped with a K3 (Gatan) detector operating in electron counting mode and a Bioquantum energy filter (Gatan). Movies were acquired at a nominal magnification of 105,000×, resulting in a pixel size of 0.837 Å. Total accumulated exposure of all the specimens ranged from 40 to 60 e^-/Ų. For both the apo and the holo-deactivated samples two data sets were collected from different preparations while for the holo-active sample, two data sets were collected on different dates from the same grid. Further data collection details are provided in Appendix Table S1.

5-HT3R was expressed in the T-Rex-293 cell line as described earlier in Methods. After cells were collected, the supernatant of the growth medium was centrifuged at 3000 × g for 30 min to remove large vesicles and cell fragments that were out of interest. Microvesicles were collected from the pellet after supernatant centrifugation at 10,000 × g for 30 min. Microvesicles were resuspended in 500 μl PBS and hydrated overnight at 4 °C. Microvesicles were incubated with 2 mM CaCl₂ and 100 μM serotonin for 40 min on ice before vitrification. The concentration of microvesicles was checked via negative staining. Grids of Quantifoil Au/Au R2/2 were glow-discharged on the PELCO easiGlow system using 15 mA for 45 s. Four μl of microvesicle solution was applied to the grid and then vitrified by Vitrobot Mark IV (Thermo Fisher Scientific) adopting blot time of 4 s and blot force of 10.

Tilt series were recorded on the Thermo Fisher Scientific Titan Krios G3i in Core Facility for Cryo Electron Microscopy, Charité – Universitätsmedizin Berlin. Tilt series of +/−48° were acquired at a nominal magnification of 64,000× (0.69 Å per pixel in super-resolution mode) using SerialEM with a Hybrid-STA script (Sanchez et al, 2020) or PACEtomo (Eisenstein et al, 2023). The electron dose was 29.3 and 4.9 e^-/Ų for non-tilted and tilted projections, respectively. Frame stacks were recorded on a Gatan K3 electron detector at the defocus range of −3 ~ −5 μm.

Raw images output by electron microscope were directly imported to TomoBEAR (Balyschew et al, 2023), a workflow engine for streamlined processing of cryo-ET data for subtomogram averaging. The raw images were sorted according to different tilt series. MotionCor2 (Zheng et al, 2017) was used for correcting beam-induced motion, IMOD (Kremer et al, 1996) was used for generating tilt stacks, the fiducial beads were detected by Dynamo Tilt Series Alignment (Coray et al, 2024) and GCTF (Zhang, 2016) was used for defocus estimation. IMOD was then used for fiducial model refinement and tomographic reconstruction. The coordinates of particles were detected by template matching implemented in TomoBEAR. Candidate particles (30,623) were cropped and contaminations were excluded by Dynamo multi-reference classification (Castaño-Díez et al, 2012) adopting a local angular search in the pixel size of 11.04 Å. After several times of cleaning, classes of tetramer and pentamer were observed during classification in Dynamo when using a tight mask and searching for 360° for the in-plane Euler angle in the pixel size of 5.52 Å. Pentamer classes and tetramer classes were imported separately to the subtomogram averaging pipeline of Relion 4.0 (Zivanov et al, 2022). For pentamer and tetramer classes, duplicated particles (distance <50 Å) were removed. Relion 3D classifications were executed using local search to exclude contaminations and heterogeneity particles. Finally, 3D refinements were executed at the pixel size of 5.52 Å, and resolutions were estimated by Fourier shell correlation. The data flow is presented in Fig. EV3.

During data collection all data sets were pre-processed using CryoSPARC Live (Punjani, 2021) for frame alignment, contrast transfer function (CTF) estimation, particle picking, and extraction with a box size of 330 pixels (288 Å). Particularly, 2D classification “on-the-fly” helped to monitor data quality including particle orientation and oligomeric state of the channel. In CryoSPARC (Punjani et al, 2017), further 2D classification and 3D heterogeneous refinement (without imposing symmetry) allowed to separate the tetrameric particles from the pentameric. After an initial 3D non-uniform refinement, tetramers’ 3D refined particles together with a real-space mask excluding the Salipro belt and the solvent were used as inputs for 3D variability analysis (3DVA) (Punjani and Fleet, 2021). 3DVA was run using three variability components and a low-pass filter resolution of 6 Å. By using the cluster mode function in CryoSPARC in both the apo and holo deactivated data sets, particles were resolved into two main clusters that yielded both the symmetric and asymmetric maps of the tetramer. The obtained particle sets were again subjected to a non-uniform refinement in CryoSPARC, which resulted in maps of 3.5 Å resolution for the apo asymmetric tetramer (~385 K particles), of 3.4 Å resolution for the apo symmetric tetramer (~230 K particles), of 3.3 Å resolution for the holo deactivated asymmetric tetramer (~338 K particles) and of 3.4 Å resolution for the holo deactivated symmetric tetramer (~296 K particles). Local resolution estimation was performed in CryoSPARC. All resolutions were estimated according to the Fourier shell correlation (FSC) 0.143 cut-off criterion of two independently refined half maps (Rosenthal and Henderson, 2003).

Minimal neighbor distances of pentamers and tetramers were calculated separately. For pentamers or tetramers, coordinates were extracted from the final .star files from the Relion 3D refinement jobs and split according to tomogram indexes. Foreach tomogram, Euclidean distances between each particle and every other particle were calculated, and the lowest value was chosen. The distribution range of minimal neighbor distances (0 to 687 nm) was split into 100 intervals and the probabilities of each interval were calculated and plotted. The probabilities of distances larger than 400 nm were low and were truncated from the graphs for clarity.

The initial model was built manually in UCSF Chimera (Pettersen et al, 2004) with single subunits from the published Salipro-5-HT3R (PDB accession code 6Y5A and 6Y59) for guidance. ISOLDE (Croll, 2018) was used for the initial model fitting. Densities corresponding to N-linked glycosylation adjacent to N82, N148, and N164 at the ECD were resolved but not included in the model (Fig. 1B,D, in red). COOT (Emsley et al, 2010) together with PHENIX (Liebschner et al, 2019) were used for the iterative refinement of the models. Model quality was assessed with Molprobity (Chen et al, 2010). Structural Figures were prepared using UCSF ChimeraX (Meng et al, 2023). The figures in this paper were made using BioRender.com.

PDB files were processed with the PISA (Protein Interfaces, Surfaces, and Assemblies) program (Krissinel and Henrick, 2007) which calculates interface and different components of the binding free energy, including the solvation-free energy gain upon formation of the interface (ΔiG, kcal/mol) and the associated ΔiG P-value, which is a probability measure of the specificity of the interface.

All protein models were prepared with the Protein Preparation Wizard (Madhavi Sastry et al, 2013) included in Maestro under the OPLS4 force field (Lu et al, 2021): hydrogen atoms were added to the prepared cryo-EM structures at physiological pH (7.0) with the PROPKA tool (Bas et al, 2008) to optimize the hydrogen bond network. All water molecules were removed; C- and N-terminal cappings were added; disulfide bonds were assigned and constrained energy minimizations were carried out on the full-atomic models until the RMSD of the heavy atoms converged to 0.3 Å.

We used three MD flavors customized for the particular problems. For the analysis of the transition between the tetrameric forms (Fig. 3A–C), we performed an enhanced rotation MD simulation (Kutzner et al, 2011). We performed all-atom MD simulations to analyze the atomic details for the pore permeation by water molecules (Fig. 3E). In this case, we took advantage of the relatively small size of the simulation system. To study the oligomerization processes, we performed the coarse-grained MD combined with sampling by metadynamics. We chose this combination of methods because the whole system is extremely large after we separate the molecules to a large distance.

Membrane systems were built using the membrane building tool CHARMM-GUI (Jo et al, 2008) and the OPM webserver (Lomize et al, 2012) was used to align the experimental structures in the lipid bilayer. To mimic the lipid composition of brain polar lipid, we used POPC, POPS, POPE, and cholesterol at a molar ratio of 15:22:39:24. Symmetrical and asymmetrical tetramer models were embedded in a lipid bilayer of 300 lipids, coupled to TIP3P water molecules comprising 0.15 M NaCl. The final systems contained 176,555 atoms and 177,295 atoms, respectively, with a volume of 100 × 100 × 180 ų. All systems were simulated using CHARMM36m force field (Huang et al, 2017). The MD simulations were performed in GROMACS (Berendsen et al, 1995). The systems were minimized with a 50,000-step energy minimization using the steepest descent algorithm. The systems were subjected to temperature equilibrating in the NVT ensemble at 310 K for 200 ps and then to density equilibrating in the NPT ensemble at 310 K and 1 atm for 10 ns. The heavy atoms were constrained using a harmonic restraint with a force constant set to 1000 kJ mol⁻¹ nm⁻² in the equilibrating steps. The production runs lasted 1 μs and three replicas were used to ensure reproducibility. The convergence of the simulation was confirmed by the analysis of RMSD of trajectories over time.

Since the subunit rotation is a rare biological event, the unbiased MD simulation might take an unexpectedly long time to sample. Therefore, for the simulations in Fig. 3A–C, we introduced a novel method to enforce the rotation changes between the symmetric and the asymmetric structure (Kutzner et al, 2011). It allows flexible adaptations of both the rotating subunit as well as the rotation axis during the simulation. Apart from the rotation itself, it imposes minimal constraints on the rotating group, allowing conformational adaptations to the surroundings. The radial profiles of proteins obtained from MD simulations were analyzed using CHAP (Klesse et al, 2019). The density of lipids was analyzed in the GROmaps tool. All Figures from MD simulations have been prepared in PyMOL (The PyMOL Molecular Graphics System, Version 1.2r3pre, Schrödinger, LLC). The parameters for MD simulations are shown in Appendix Tables S2–4.

To probe how pentamers could form, we applied well-tempered metadynamics simulations, which is a reliable and widely used method for sampling rare biological events, such as protein-protein binding (Barducci et al, 2011, 2008). Metadynamics simulations were conducted using GROMACS-2021.5 (Berendsen et al, 1995) patched with plumed-2.7.3 (Bonomi et al, 2009). Coarse-grained (CG) simulation systems were built to investigate the assembly process of the 5-HT3R. We combined an Elastic Network with the Martini3force field (Souza et al, 2021) to obtain the initial atomic models. All CG systems were embedded in a phosphatidylcholine lipid bilayer (as described above) and were solvated in Martini CG water boxes, containing 0.15 M NaCl. In the CG simulations, a 5000-step energy minimization was then followed by a 5-ns NPT pre-equilibration simulation in which the positional constraints of the protein backbone beads were gradually relaxed. In the simulations, the temperature was set to 310 K using the V-rescale thermostat, and the pressure was set to 1 bar using a semi-isotropic coupling method. Free-energy profiles of the systems were calculated. Metadynamics simulations use a history-dependent potential V (s, t) to accelerate the sampling of the collective variables (CVs), s (s₁, s₂, …, s_m). V (s, t) is usually constructed as the sum of multiple Gaussians centered along the trajectory of the CVs:

To achieve this, a Gaussian-shaped potential is added to bias the system at the current position of the CVs, periodically during the simulation. The height of the Gaussian is decreased with the amount of bias already deposited according to

In our simulations, the distance between the mass center of the TMD and that of the ECD for each subunit was assigned as the CVs, while the width of Gaussians, σ, was set as 0.3. The time interval τ was 1.0 ps. Well-tempered metadynamics involves adjusting the height, w_j, in a manner that depended on V (s, t), where the initial height of Gaussians w₀ was 1.5 kcal/mol, and the simulation temperature was 310 K Each metadynamics simulation lasted for 500 ns. The convergence of the simulations was confirmed by the analysis of distance plots over time. More details of metaMD simulations can be found in our previous works (Yuan et al, 2014; Chan et al, 2018).