Asymmetric gating of a human hetero-pentameric glycine receptor

Hetero-pentameric Cys-loop receptors constitute a major type of neurotransmitter receptors that enable signal transmission and processing in the nervous system. Despite intense investigations in their working mechanism and pharmaceutical potentials, how neurotransmitters activate these receptors remain unclear due to the lack of high-resolution structural information in the activated open state. Here we report near-atomic resolution structures in all principle functional states of the human α1β GlyR, which is a major Cys-loop receptor that mediates inhibitory neurotransmission in the central nervous system of adults. Glycine binding induced cooperative and symmetric structural rearrangements in the neurotransmitter-binding extracellular domain, but asymmetrical pore dilation in the transmembrane domain. Symmetric response in the extracellular domain is consistent with electrophysiological data showing similar contribution to activation from all the α1 and β subunits. A set of functionally essential but differentially charged amino-acid residues in the transmembrane domain of the α1 and β subunits explains asymmetric activation. These findings point to a gating mechanism that is distinct from homomeric receptors but more compatible with heteromeric GlyRs being clustered at synapses through β subunit–scaffolding protein interactions. Such mechanism provides foundation for understanding how gating of the Cys-loop receptor members diverge to accommodate specific physiological environment.

GlyRs are major Cys-loop receptors that mediate fast inhibitory neurotransmission in the spinal cord and brain 13,21,22 . Dysfunction of GlyR causes the congenital disorder hyperekplexia 7,8,23−25 . It is a therapeutic target in neuropathic pain 8,26−28 , and related to autism and other neurological disorders 4,13,22,29 . The widely studied stereotypical homomeric GlyRs contain only the α subunits and practically do not exist in adult animals. On contrary, the heteromeric GlyRs containing both the α and the β subunit are the found throughout the central nervous system 13 . Structural information of heteromeric GlyRs has only become available very recently, revealing an unexpected 4:1 α2:β subunit stoichiometry 18, 30 . Unfortunately, none of the structures reported was in the activated open state, although a "semi-open" state in one study 30 had its ion conduction pore asymmetrically expanded to a size just shy of the predicted range of open GlyR pores 22,30,31 .
In this study, we resolved near-atomic resolution structures of the α1β GlyR with native function in all principle functional states throughout its gating cycle: apo (closed), glycine-bound desensitized and glycine-bound open. Combining with electrophysiology and mutagenesis experiments, we show that glycine binding to the orthosteric site results in highly cooperative and pseudo-symmetrical conformational changes in the extracellular domains (ECD), explaining the cooperativity of glycine in activation of GlyRs. However, due to the distinct characteristics of the α and β subunit transmembrane domains (TMD), symmetric activation in ECD resulted in asymmetrical structural rearrangements in the TMD, leading to asymmetrical open conformations that is distinct from homomeric GlyRs. Such differential responses in ECD and TMD reconciliates the cooperativity in glycine activation and differences in subunit properties in an asymmetric gating mechanism. This mechanism ensures that the clustering of GlyRs at synapses through its β subunit does not cause aberrant channel activation independent of glycine. In addition, it explains how mutations of a charged amino-acid residue pointing away from the ion conduction pore diminishes GlyR conductance and cause hyperekplexia.

Results
Overall architecture of the human α1β GlyR To understand gating mechanisms of the human α1β GlyR without complications from functionmodifying antibodies and limitations in puri cation from native tissues, we generated constructs of the α1 and β subunits (α1em and βem, respectively) that form α1emβem GlyR and exhibit indistinguishable function as wild type. α1em contained a partial truncation of the unstructured M3-M4 loop ( Supplementary Fig. 1a). βem has been described previously 32 , which contained substitution of GFP for part of the unstructured M3-M4 loop that does not bind with synaptic scaffolds (Supplementary Fig. 1b).
Through single-particle cryo-EM analysis, we resolved structures of α1emβem GlyR both in the absence and in the presence of glycine. The density arising from GFP fusion on the βem subunit served as ducial marker to differentiate the β subunit from the structurally similar α subunits 30 ( Fig. 1c; Fig. 2a, b, i, j), without the use of available subunit-speci c antibodies that alter function 18 . The overall resolution ranged from 3.6 -4.1 Å, with local resolutions extending well beyond 3.0 Å for many regions of interest ( Fig. 2d-f, l-n). These density maps allowed for unambiguous model building for most part of the protein ( Fig. 2d; Supplementary Fig. 3; Supplementary Table 1). In all the structures determined, either with or without glycine, 4 α1 subunits and 1 β subunit were identi ed (Fig. 1c, d), consistent with the recent discovery of 4:1 α2:β stoichiometry of human α2β GlyR and heteromeric GlyR puri ed from porcine tissue 30 . These structures suggest that 4:1 α1:β stoichiometry is an intrinsic property of α1β GlyR, not dependent on the presence of other GlyR subunits (such as α2) or unidenti ed factors/chaperons in native tissues 18 .
α1β GlyR in the closed, open and desensitized states In the absence of glycine (apo state), a single conformation was resolved that corresponds to the closed states (( Fig. 2a-c). All the orthosteric sites are empty (Fig. 2a). The activation gate 11 at 9' is tightly constricted with the 9' Leu sidechains pointing toward the pore (Fig. 2b, c). The minimal pore radius is 1.8 Å, too narrow for Clto pass (Fig. 2m black). This conformation resembles that of homomeric GlyR in the closed state 11 , as well as the pseudo-symmetrical α2β-strychnine complex in the closed state 30 .
In the presence of glycine, two structures were resolved with widened pores allowing Clto pass through ( Fig. 2d-i). In one structure, clear glycine densities are found in all 5 orthosteric sites (Fig. 2d). Compared to the apo state, 9' Leu side chain ipped away from the pore, combined with outward movement of the M2 helix, resulted in an open activation gate 11 . The -2' position dilated in an apparently asymmetrical manner ( Fig. 2 e, f). The minimal radius along the pore is ~ 2.9 Å (Fig. 2m yellow), su cient to allow partially hydrated Clthrough, and within expected range of physiologically pore sizes of open GlyR 12,31 . This structure should represent a conformation of α1β in the open state.
In the other structure, the orthosteric sites at all 3 α/α interfaces have clear glycine densities, while sites at α(+)/β(-) and β(+)/α(-) interfaces are empty (Fig. 2g, Supplementary Fig. 3). Nonetheless, the pore also dilated in an asymmetrical manner, but to a larger extent, resulting in a minimal radius of 3.5 Å (Fig. 2h, i and m cyan), which is too large for a physiologically open GlyR. Over-widened pore is not likely a result of truncation in the unstructured M3-M4 loops considering the functional equivalence of α1emβem GlyR with wild type ( Fig. 1; Supplementary Fig. 1). In addition, a similar "expanded open" conformation has been recently reported of full-length homomeric GlyR 12 .
A third structure was resolved in the α1emβem GlyR -glycine complex that corresponds to the desensitized state. 2 out of 5 orthosteric sites are occupied with glycine ( Fig. 2j; Supplementary Fig. 3). The activation gate around 9' Leu is open. The desensitization gate is constricted to a radius of ~ 2 Å (( Fig. 2k, l, m pink), rendering the channel non-conductive. The pore shape in this structure is similar to that of α2β GlyR 30 , and GlyR puri ed from porcine tissue in the desensitized state 18 (Fig. 2m pink and purple), retaining mostly 5-fold pseudo-symmetric shape in the TMD.
The above structures in all the major functional states depict a gating mechanism that has resemblance but is also very different from homomeric GlyRs. Homomeric GlyRs maintain the 5-fold symmetric structure throughout the gating cycle, with all 5 orthosteric pockets in an identical ligandbinding state [10][11][12]17,33,34 . α1β GlyR changes its symmetry during gating and exhibit varying glycine occupancy in the 5 orthosteric sites. This provides a unique opportunity to understand how asymmetrical gating works in a multimeric ligand-gated ion channel containing non-equivalent subunits.
Symmetric glycine-induced response in the ECD Glycine binding induced similar conformational rearrangements across all 5 subunits. Binding of agonists is known to induce a compact conformation in the orthosteric site 10,11,18,35,36 . In our apo structure (Fig. 2a), all 5 binding pockets are empty and in the same apo conformation ( Supplementary   Fig. 4a). In the open state structure, glycine is found in all 5 pockets, resulting in the same compact conformation across all pockets (Fig. 3a, Supplementary Fig. 4b). However, in the expanded-open and desensitized states, orthosteric pockets are only partially occupied (3 and 2 out of 5, respectively) ( Fig. 2g, i). Nonetheless, the same compact conformation was observed across all 5 pockets, irrespective of whether glycine is bound (Fig. 3b, c). In addition, ECDs from all 5 subunits showed similar rocking motions that propagate to TMD during channel activation (Fig. 3h). Such symmetric change in the ECD upon glycine binding suggests cooperativity among orthesteric pockets, which is indicated by Hill slopes ranging from ~ 2-5 in our and previously reported glycine-dose response curves ( All orthosteric pockets contribute similarly to the gating of α1β GlyR. This is indicated by the same conformation across all binding pockets irrespective of subunit type. To further test this, we mutated one amino acid residue (α:F207 or β:Y231, see Fig. 3a) that is important for glycine binding, and performed glycine titration (Fig. 3d). Mutating α:F207 resulted dramatic decrease in apparent glycine a nity, to the point where only a lower limit of EC 50 ~ 0.8 mM can be estimated (Fig. 3d, blue), while mutating the homologous β:Y231 only resulted in ~ 2 fold increase in EC 50 (Fig. 3d red). This is because there is only 1 β subunit but 4 α subunits in each GlyR. Consistent with this, mutating a subset of α:F207 resulted in a similar effect as the β:Y231 mutation (Fig. 3d, light blue). This phenomenon is consistent across extensive mutagenesis experiments in glycine binding pockets ( Supplementary Fig. 5). The equivalence in response to glycine binding of all subunits is expected considering the cooperativity among these sites, which ensured a pseudo-5-fold symmetric structure in the extracellular domain throughout the gating cycle ( Supplementary Fig. 6a, c, e, g).
Interestingly, glycine-induced structural changes in the ECD are similar across the open, expanded-open and desensitized states ( Fig. 3e-g). Orthosteric pockets at the β(+)α(-) (Fig. 2e), α(-)β(+) (Fig. 2f), and α(+)α(-) (Fig. 2g) subunit interfaces showed similar contraction upon glycine binding, which in turn induced a rotational motion of the ECD (Fig. 3h) 10,11 . ECD rotation moved the directly connected extracellular end of M1 helix away from the conduction pore, resulting in an outward symmetric expansion at the ECD-TMD interface ( Supplementary Fig. 6i, j). 5-fold pseudosymmetry is maintained through all functional states in ECD and the extracellular end of TMD. However, the conformations of TMD near the intracellular side become more distinctive between different subunits (Fig. 3h). Apparently, the TMD of heteromeric GlyR converts the same symmetric input from ECD into different, sometimes structurally asymmetrical, functional states.
Asymmetrical gating in the α1β GlyR TMD α1β GlyR in the apo and desensitized states, but not in the open states, retained 5-fold pseudo-symmetry. As described above, the ECD remains largely symmetric throughout the gating cycle. Such symmetry is retained in the TMD in the apo and desensitized states ( In the open state, the extracellular end of all 5 TMDs moved radially away from the conduction pore, resulting in a symmetric expansion similar to the desensitized state (the pore-lining M2 helices are shown in Fig. 4a, b, d top panel). However, the expansion became asymmetrical near the intracellular end -in addition to small radial expansion, one of the α1 subunits (chain B) moved away in both radial and tangential directions, creating a wider spacing from one adjacent α subunit (chain C) (Fig. 4b lower   panel). This movement resulted in an asymmetrically widened desensitization gate that allows the conduction of Cl -. The expanded-open state showed resembling, but more extended structural rearrangements near the desensitization gate -two of the α1 subunits (chain B and C) moved in both radial and tangential directions, creating an asymmetrical wide-open pore. Intriguingly, none of the α1 subunit adjacent to the β subunit experienced such large movements, which we believe is resulting from amino-acid residues unique to each subunit types, as discussed below.
Differentially charged amino acid residues in the α1 and the β M2 helices promote asymmetrical gating. The pore-lining M2 helix is one of the least conserved regions between the α and β subunits in amino acid sequences (Fig. 4e), harboring disease-causing mutagenetic sites and resulting in distinctive functional features including glycine dose-response, single-channel conductance and picrotoxin sensitivity in homoand hetero-meric GlyRs 13,22,30,37,39 . A set of unique amino acid residues resulted in opposite electrostatic elds at subunit interfaces in TMD of α1 and β subunits (Fig. 4f-j). The combination of two neutral and one positively charge residues, α1:S273/R271/Q266, makes the TMD of α1 positively charged at both the (+) (Fig. 4g) and (-) (Fig. 4j) sides. The β subunit has two negatively charged residues and one hydrophobic residues at corresponding positions; β:E297/A295/ E290, which renders negative potentials at both (+) and (-) sides (Fig. 4h, i). Opposite potentials promote the interaction between the α1 and β subunit TMDs (Fig. 4g binds with h, i binds with j), especially in the hydrophobic environment inside the membrane, leading to a stable α1-β-α1 assembly throughout the gating cycle ( Fig. 4b-d, Fig. 2c-l). However, the same positive potential contributes to repulsion between α1-α1 interfaces, increasing the likelihood of widening in α1-α1 interfaces. This explains the widening of some α1-α1 but not the α1-β-α1 interfaces in both the open (Fig. 4k) and expanded-open states.

Discussion
We have resolved the structures of human α1β GlyR in all its major functional states, which depicts the structural rearrangements through the gating cycle (Fig. 5). In the apo state, 5-fold pseudo-symmetry is maintained in the whole α1β GlyR, with tightly constricted ion conduction pore (Fig. 5a). When glycine binds, the ECD of all 5 subunits experienced similar rotational motion, maintaining pseudo-symmetry. ECD rotation pulls on the extracellular end of TMD, resulting in the same radial motion away from the pore in all 5 subunits (Fig. 4b). However, due to electrostatics repulsion, the pore lining M2 helix of one α1 subunit moves tangentially away from the adjacent α1 subunit near the intracellular end. Since the β subunit carries opposite electrostatic charges, it remains attracted to neighboring α1 subunits. In this way, the pore is dilated in an asymmetrical manner, representing one open state of α1β GlyR. When desensitization happens, the intracellular end of M2 helices of all subunits collapse back into a more pseudo-symmetrical conformation that is thermodynamically stable 10,11,22,40 , stopping Cl − conduction (Fig. 5c).
The ECD and TMD of α1 and the β subunits contributes differently to the assembling and gating of α1β GlyR. Although the unique ECD of the β subunit dictates the 4:1 α:β stoichiometry in heteromeric GlyRs 30,38,41 , it contributes similarly to α1β GlyR gating by glycine. All 5 orthorsteric pockets experience similar conformational changes from the close to the open/desensitized states, regardless of subunit types. This leads to identical structural changes in the extracellular end of TMD for all the α1 and β subunits (Fig. 2). The TMDs, on the other hand, do not affect the assembly, but instead determines functional properties of heteromeric GlyRs and contain multiple disease-causing mutagenesis sites 13,37,39 . Mutation of one residue to remove the positive charge, R271L/Q/P, of the α1, is known to diminish Cl − conduction and cause hyperekplexia in a dominant manner through unclear mechanism 7,23,25 . Recently it was shown that R271 (19') allows conduction not through locally concentrating Cl − , but creating electrostatic repulsion between subunits to allow the opening of the pore 42 . This is consistent with our observation of widened α1-α1, but not α1-β subunit distances in the open state because α1 and β subunits are oppositely charged at equivalent positions (Figs. 4 and 5). We believe asymmetrical expansion of pore is the most likely mechanism for a heteromeric GlyR to open.
Considering the multiple single channel conductance states reported of heteromeric GlyRs 37,43 , it is tempting to speculate that multiple open state conformations exist, and some remain to be identi ed.
The β subunit being tightly tethered to its neighboring α1 subunits throughout the gating cycle is compatible with its role in cellular organization of GlyRs. Post-synaptic scaffolding protein gephyrin (Fig. 5 green oval), binds to the intracellular M3-M4 loop of the β subunit and clusters GlyRs at postsynaptic membranes [44][45][46] . Tight β-α1 interaction ensures when β subunit experiences force originating from relative motion with respect to the tethered scaffold, GlyR responds as a rigid body without changes in the ion conduction pore geometry, which may cause aberrant conduction independent of glycine binding. Since many Cys-loop receptors are clustered in post-synaptic densities, gating mechanisms that avoids excessive crossover between higher-order assembly and neurotransmitter activation may be also relevant for other receptor members, depending on speci c interaction contexts.

Plasmid constructs
The human glycine receptor α1 (NCBI: NP_001139512.1) and β (NCBI: NP_000815.1 ) sequence were ampli ed from cDNA clones (McDermott Center, UT Southwestern Medical center). The α1em sequence was derived by substitution of M3/M4 loop (residues R316-P381) s by GSSG peptide. For the βem construct, we used the previously described βem construct 30 . The α1em and α1 wild type sequence was subcloned into a BacMam expression vector 47 . The β wild type sequence were introduced into pLVX-IRES-ZsGreen1 vector (Clonetech) for electrophysiology. All α1em and βem mutants was generated using sitedirected mutagenesis.

Protein expression and puri cation
The α1em and βem constructs were transformed into DH10BacY competent cells (Geneva Biotech) to produce bacmids. The bacmids were transfected into Sf9 cells to generate baculovirus. Recombinant baculovirus titer was determined as described before 30,47 . Virus was added at a multiplicity of infection (MOI) of 2(at 3βem:1α1em ratio) to the cell cultures, at a density of 2.5×10 6 cells/ml. To increase the expression level, 10 mM sodium butyrate was added, and culture temperature was changed to 30˚C after transduction 12h. Cells were collected after induction 60h by centrifugation at 30,000 g for 20 minutes at 4°C and stored at −80 °C until further use. Cryo-EM sample preparation, data collection and image processing For apo α1emβem GlyR, the sample was vitri ed without any ligand. For glycine-bound α1β GlyR, the sample was incubated for 1h with 2 mM glycine on ice. 1 × CMC nal concentration (~ 3 mM) of Fluorinated fos-choline 8 (Anatrace) was added into sample immediately before freezing. Grids (Quantifoil R1.2/1.3 400-mesh Au holey carbon grid) were glow-discharged. An FEI Vitrobot Mark IV Vitrobot (Thermo Fisher) was employed to plunge freeze the grids after application of 3 µl sample at 4℃ under 100% humidity.
Micrographs were collected using a Titan Krios microscope (Thermo Fisher) with a K3 Summit direct electron detector (Gatan) operating at 300 kV using the SerialEM data acquisition software. The GIF-Quantum energy lter was set to a slit width of 20 eV. Images were recorded in the super-resolution counting mode with the pixel size of 0.415 Å. Micrographs were dose-fractioned into 50 frames with a dose rate of 1.4 e -/Å/frame. 2-fold binning (0.83 Å pixel size after binning), motion correction and dose weighting of the movie frames were performed using the Motioncorr2 program 49 . CTF correction was carried out using the CTFFIND 4 program 50 . The following image processing steps were carried out in RELION 3 51 , as illustrated in Supplementary Fig. 2. Particles were initially picked using the Laplacian-of-Gaussian blobs and subjected to 2D classi cation to obtain good class-averages, which was then used as template for reference-based autopicking. Resulting particles were extracted with 4-fold binning for a further round of 2D classi cation ( Supplementary Fig. 2a, i). Good 2D classes were selected and subjected to 3D classi cation using an initial model downloaded from EMDB database (EMD-23148) 30 . For both the apo-and glycine-bound samples, 1 out of 6 classes in 3D classi cation appeared with good density for the entire channel ( Supplementary Fig. 2b, j). A single density blob for GFP was identi ed for both the apo and glycinebound samples. A further 3D classi cation into 4 classes with non-binned particles (0.83 Å pixel size) without particle alignment was performed. For the apo-sample, partial signal subtraction 52 was performed to focus on the TMD. 2 indistinguishable good classes were pooled, which resulted in a nal of 29, 850 particles ( Supplementary Fig. 2c). After reverting particles to un-subtracted version, CTF re nement, Bayesian polishing in RELION and non-uniform re nement 53 in cryoSPARC 54 , an overall resolution of 3.6 Å was achieved, with local resolutions exceeding 3.0 Å in many regions ( Supplementary  Fig. 2d, f). For the glycine-bound sample, second 3D classi cation was performed using a mask excluding GFP and micelle, resulting in three good classes with distinct conformations ( Supplementary   Fig. 2k). After CTF re nement, Bayesian polishing in RELION and non-uniform re nement in cryoSPARC, overall resolutions of 3.6 Å (21, 676 particles), 3.9 Å (24, 487 particles) and 4.1 Å (30, 723 particles) were achieved for the open, expanded-open and desensitized states, with local resolutions exceeding 3.0 Å in many regions (Supplementary Fig. 2l, n). Resolutions were estimated by applying a soft mask around the protein densities with the Fourier Shell Correlation (FCS) 0.143 criterion. Local resolutions were calculated using Resmap 55 .
Model building and re nement Models of GlyR α1β heteromer were bulit by tting the structure of Rattus norvegicus α1β homomer glycine-bound state (PDB ID: 7mly) 56 into the Cryo-EM density maps of GlyR α1β heteromer using Chimera 57 and Coot 58 . The atomic model was manually adjusted in Coot. The nal models were re ned with real-space re nement module and validated with comprehensive validation module in PHENIX package 59,60 . Fourier shell correlation (FSC) curves were calculated between re ned atomic model and the work/free half maps as well as the full map to assess the correlation between the model and density map ( Supplementary Fig. 2e and m). Statistics of cryo-EM data processing and model re nement are listed in Table S1. Pore radii were calculated using the HOLE program 61  Fluorescence-Detection Size-Exclusion Chromatography (FSEC) expression assay In the FSEC assay, uorescence was detected using the RF-20Axs uorescence detector for HPLC (Shimadzu, Japan) (for EGFP, excitation: 480 nm, emission: 512 nm) as EGFP was fused into βem construct. Using 2μl of Lipofectamine 3000 (Thermo Fisher Scienti c, US), 1ug of plasmid (at 1α1:3β ratio) was transfected into HEK293T cells for each well of 12 well plate. Cells were incubated in a CO 2 incubator (37 °C, 8% CO2) for 48 h after transfection and solubilized with 50μl buffer B for 1 h. After centrifugation (40,   (c) Side view of cryo-EM map of 1β GlyR in complex with glycine.
(f) The position of key amino acids in the helix. The corresponding amino acids are marked in panel E.
(k) The cartoon representation of the TMD interfaces between (blue) -β (salmon), and -subunits in the open state. Red arrow indicates increased distance between adjacent subunits.

Figure 5
Proposed gating mechanism of the 1β GlyR.
Adjacent (blue) and β (orange) subunits are shown with respectively charged amino acid residues and surrounding electrostatics (positive: blue; negative: red). Post-synaptic scaffold is shown as green oval.
(a) In the apo state, heteromeric 1β GlyR is pseudo-symmetrical with a closed pore.
(b) Upon glycine (yellow spheres) binding, conformational changes in ECD cause the widening of the extracellular end of the TMD. Electrostatic repulsion between adjacent subunits makes them easier to separate (red arrow). On contrary, the opposition electrostatics of and β subunits ensures that they stay