Cryo-EM structure of the human α1β3γ2 GABA_A receptor in a lipid bilayer

Abstract

Type A γ-aminobutyric acid receptors (GABA_ARs) are pentameric ligand-gated ion channels (pLGICs) and the main drivers of fast inhibitory neurotransmission in the vertebrate nervous system^1,2. Their dysfunction is implicated in a range of neurological disorders, including depression, epilepsy and schizophrenia^3,4. Amongst the numerous assemblies theoretically possible, α1β2/3γ2 GABA_ARs are most prevalent in the brain⁵. The β3 subunit plays an important role in maintaining inhibitory tone and expression of this subunit alone is sufficient to rescue inhibitory synaptic transmission in a CRISPR/Cas9 derived β1-3 triple knockout⁶. To date, efforts to generate accurate structural models for heteromeric GABA_ARs have been hampered by the use of engineered receptors and the presence of detergents^7–9. Significantly, some recent cryo-EM reconstructions report “collapsed” conformations^8,9 which disagree with the prototypical pLGIC, the Torpedo nicotinic acetylcholine receptor^10,11, the large body of structural work on homologous homopentameric receptor variants¹², and the logic of a ion channel architecture. To address this problem, here we present a high-resolution cryo-EM structure of the full-length human α1β3γ2L, a major synaptic GABA_AR isoform, functionally reconstituted in lipid nanodiscs. The receptor is bound to a positive allosteric modulator megabody and in a desensitised conformation. Unexpectedly, each GABA_AR pentamer harbours two phosphatidylinositol 4,5-bisphosphate (PIP2) molecules, whose head groups occupy positively-charged pockets in the intracellular juxtamembrane regions of α1-subunits. Beyond this level, the intracellular M3-M4 loops are largely disordered, possibly because interacting post-synaptic proteins were not included. This structure illustrates the molecular principles of heteromeric GABA_A receptor organization and provides a reference framework for future mechanistic investigations of GABA-ergic signalling and pharmacology.

We isolated the full-length human α1β3γ2L receptor from a stable cell line¹³, reconstituted it into lipid nanodiscs, and solved its structure by single-particle cryo-EM (Fig. 1a-e, Extended Data Fig. 1 & 2a-b, Extended Data Table 1, Supplementary Video 1). Western blot analysis of reconstituted receptors revealed minimal proteolysis of receptor subunits (Extended Data Fig 2c). Radioligand binding assays demonstrate that the reconstituted receptor retains functional properties and allosteric interactions as expected in a membrane context¹³ (Fig. 1f, Extended Data Fig. 2d). The agonist, GABA, enhances binding of a benzodiazepine, [³H]flunitrazepam (Extended Data Fig 2d), while allosteric interactions between extracellular (ECD) and transmembrane (TMD) domains are confirmed by enhancement of agonist, [³H]muscimol, binding in the presence of the anaesthetic etomidate (Fig. 1f). The α1 subunit-specific nanobody Nb38⁷, enlarged by fusion to the extracellular adhesin domain of H. pylori (HopQ) to generate the “megabody” $\text{M}{\text{b}}_{\text{Nb}38}^{\text{cHopQ}}$ (called here Mb38 for simplicity), was used to randomize GABA_AR orientation in ice and facilitate particle alignment.

Figure 1. Architecture of full-length α1β3γ2L GABA_AR in lipid nanodiscs.

a-c. Side (a), top (b) and bottom (c) views of the sharpened cryo-EM map of α1β3γ2L GABA_AR-Mb38 complex in lipid nanodisc. Mb38 and glycans are colored green and orange, respectively. Density contributed by the nanodisc is colored in pale blue. d, e. Side (d) and top (e) views of atomic model of the α1β3γ2L GABA_AR in ribbon representation and glycans and lipids in ball-and-stick representation. Subunit coloring reflects that in (a-c). f. Etomidate enhancement of [³H]muscimol binding at α1β3γ2L GABA_AR-nanodisc complexes in the presence or absence of Mb38. Each point represents average of at least 3-4 independent measurements and error bars correspond to one standard deviation, with the exception of points for 0.1 and 0.3 μM etomidate with Mb38 which represent two independent measurements.

Three-dimensional reconstruction of the α1β3γ2L-Mb38 complex led to a cryo-EM map at 3.2 Å nominal resolution, which allowed model building for the ECD and TMD of all GABA_AR subunits, and for the nanobody domain of Mb38 (Fig. 1, Extended Data Fig. 1a and d-g, Extended Data Fig. 3a-l, Supplementary Video 2). Intracellular domains (ICD), largely disordered in the absence of interacting post-synaptic proteins, could only be partially modelled (Extended Data Fig. 3m, Supplementary Video 2). The tri-heteromeric α1β3γ2L receptor has a five-fold quasi-symmetrical architecture throughout (Fig. 1d, e). The intact organization of the TMDs observed here (Fig. 1, Extended Data Fig. 3) is in marked contrast with previous EM reconstructions of GABA_ARs in detergent ^8,9 (Extended Data Fig. 4). Additional density, corresponding to the helical belt formed by the nanodisc MSP2N2 proteins and the enclosed lipid bilayer, surrounds the TMD (Fig. 1a, c). Unlike the complex we previously reported in the presence of excess Nb38⁷, Mb38 was added to lower concentrations and is bound only at a single, presumably higher affinity site: the α1⁺/β3^– interface in the receptor ECD (Fig. 1 a-e, Extended Data Fig. 5 and Supplementary Information).

N-linked glycans observed at the periphery of β3 (Asn80, β3-strand & Asn149, β7-strand) and γ2 (Asn208, β9-strand) subunits project into the extracellular environment (Fig. 1 a-b and d-e). In contrast, glycans linked to α1 subunits at Asn111 occupy the receptor vestibule and remain in an immature form, modelled here as Man₄GlcNAc₂ units (Extended Data Fig. 6a, b), as observed in other heteromeric GABA_AR structures^7–9. The arrangement of α1 subunit N-linked glycans likely serves to regulate receptor assembly and stoichiometry, limiting the number of α subunits per pentamer to maximum two^7,8. We propose that, together with the unique properties of subunit interfaces, this ensures an ordered assembly whereby an α1β3 dimer combines with an α1β3γ2L trimer to form a hetero-pentameric receptor^7,8 (Extended Data Figs. 6, 7 and Supplementary Information).

In addition to annular lipids on the extracellular side of the bilayer, modelled as POPC moieties (Extended Data Fig. 8a), we discovered distinct EM densities associated with the cytosolic side of α1 subunits within electropositively-charged pockets (Fig. 2a, b, Supplementary Videos 1 and 2). The high EM map quality allows unambiguous assignment to phosphatidylinositol 4,5-bisphosphate (PIP2) (Fig. 2a), confirmed by all our high-resolution, small molecule-bound, GABA_AR structures¹⁴. The inositol 1,4,5-trisphosphate (IP3) head groups form extensive interactions with the protein backbone, as well as charged and polar residues on the α1 M3 (Lys312 and Arg313) and M4 helices (Ser388, Ser390 and Lys391; Fig 2c, d). Sequence alignment reveals strict conservation of these residues amongst α-subunits of synaptic-type GABA_ARs (Fig. 2e, Extended Data Fig. 8b). The IP3 phosphate group at position 1 is also coordinated by the side chain of Arg249 protruding from the α1 M1-M2 loop (Fig. 2c, d). This loop directly affects receptor desensitization^15,16 and arginine residues at this position are conserved amongst the synaptic α-subunits (α1-3,5) or replaced by lysine residues in extrasynaptic α-subunits (α4,6) (Fig. 2e).

Figure 2. PIP2 binding sites in α1β3γ2L GABA_AR.

a. PIP2 bound at the base of α1 subunit TMDs. EM density map contoured around PIP2 (at α1 chain A binding site). b. Electrostatic surface potential shown at the ‘cytosolic’ end of α1 subunit and bound PIP2, shown in stick representation. c, d. PIP2 binding site when viewed approximately parallel to the plane of the membrane (c) and when rotated around the vertical axis (d). e. Sequence alignment of GABA_AR α-subunits for PIP2 binding regions. α1 residues forming hydrogen-bonds or salt-bridge interactions with PIP2 are identified by yellow hexagons, and those which are conserved amongst α-receptor subunits are highlighted orange (identical) and yellow (similar).

PIP2, an important component of the inner leaflet of plasma membranes, might have two functional roles in GABA_ARs. First, it might directly regulate channel function¹⁷, as it does in inwardly-rectifying potassium channels¹⁸. Second, it might be involved in the internalization of synaptic, but not extrasynaptic, receptors. To test the first hypothesis, we performed inside-out patch-clamp electrophysiology on the same cell-line used in structural studies. As receptors expressed on HEK293 cells were likely pre-bound to endogenous PIP2, we assessed the effects of PIP2 depletion on currents evoked by the allosteric activator etomidate. The PIP2 scavenger poly-L-lysine (PLL)¹⁷ elicited a minor enhancement of peak currents amplitudes (Extended Data Fig. 8c, d). Previous studies, assessing the effect of PIP2 depletion on steroidal modulation of GABA_ARs did not reveal a functional effect on channel gating either¹⁹. Therefore, we speculate that PIP2 plays roles in regulating receptor trafficking, rather than modulating the channel function.

The agonist GABA binds at the two interfaces between adjacent β⁺ and α^– subunits (Extended Data Fig. 9a). Although our structure was solved in the absence of exogenous GABA, we observe serendipitous non-protein densities at both β⁺/α^– agonist binding sites (Extended Data Fig 9b, c). Relative to competitive antagonist-bound structures¹⁴ these sites adopt compact “aromatic box” configurations, lined by side chains of Phe200 and Tyr205 from loop C on the β⁺ side and Phe46 and Phe65 on the α^– side (Fig. 3a,b), as observed across pLGICs¹². While we cannot reliably determine the contaminating ligand identity, we carried out in silico molecular docking of GABA within these sites. The most energetically favourable pose (estimated free energy of binding -6.5 kcal/mol) reveals a binding mode coordinated primarily via hydrophilic interactions (Fig. 3a, b and Extended Data Fig. 9d). In each site, the α^– Arg67 guanidinium moiety coordinates the GABA carboxylate (Fig. 3a, b). In this pose, matching those we determined experimentally in GABA/benzodiazepine and GABA/picrotoxin α1β3γ2L complexes¹⁴, the GABA amino group forms cation-π interactions with Tyr205 (Fig. 3c, d). Substitution experiments in β2-containing receptors corroborate the importance of interactions identified by docking and their effects on GABA sensitivity²⁰. Despite differences in detail (Fig. 3), this GABA binding mode is broadly consistent with those proposed recently for α1β1γ2 and α1β2γ2 heteromers^8,9.

Figure 3. Neurotransmitter binding pocket and docked GABA.

a, b. Close-up view of the binding site when viewed approximately parallel to the plane of the membrane (a) and from the extracellular side (b). GABA, in its most energetically favored pose from computational docking analysis, is shown in grey ball-and-stick representation. Hydrogen bonds, salt bridges and cation-π interactions are shown as dotted lines. c-h. Superposition of the orthosteric binding site of α1β3γ2L (grey) with the alprazolam/GABA bound α1β3γ2L (c, d. PDB 6HUO) GABA-bound α1β1γ2 (e, f. PDB 6DW1) and α1β2γ2 (g, h. PDB 6D6U) GABA_AR cryo-EM structures. Experimentally derived GABA (yellow ball-and-stick) poses and interactions (dotted lines) are shown.

Superposition of complementary subunit ECDs illustrates structural distinctions between β⁺/α^– (agonist) and analogous sites at other interfaces (Extended Data Fig. 9f). Loops-C at α⁺/β^–, α⁺/γ^– and to a lesser extent γ⁺/β^– interfaces protrude marginally further into the corresponding pockets. While there is a general conservation of residue type, the loss of critical aromatic (β3⁺ Phe200/α1 Ser205/γ2 Thr215) and charged/polar residues (α1^- Arg67/β3 Gln64/γ2 Ala79) prevents agonist binding outside canonical β⁺/α^– sites. Accordingly, docking experiments at the α⁺/β^–, α⁺/γ^– and γ⁺/β^– interfaces found GABA binding less favoured energetically (-5.5, -5.2 and -4.8 kcal/mol, respectively; Extended Data Fig. 9e).

The α1β3γ2L TMD region is fully ordered and its 5-fold symmetry is only broken by inherent sequence variations between subunit types. This is markedly different to some truncated heteromeric GABA_AR structures solved in detergent (Extended Data Fig. 4), where the pore is collapsed (Extended Data Fig. 4o-r), causing long-range receptor damage^8,9. The TMD forms the ion channel and displays and electropositive ring at its cytosolic portal, a determinant of ionic selectivity in pLGICs^12,21(Fig. 4a). Relative to the central vertical axis, M2 helices tilt outward by ~9-11° (Fig. 4c-d). As in the cryo-EM structures of diazepam and alprazolam-bound α1β3γ2L, GABA_AR β3, chimeric GLIC-GABA_AR_α1 and GABA_AR_β3-α5^14,22–24, the tightest pore constriction is observed at the level of the desensitization gate (-2’ Pro/Ala)¹⁵, with a radius of 1.45 Å (Fig. 4b-d). At the activation gate level, the 9’ Leu side chains are rotated away from the channel lumen compared to the bicuculline-bound α1β3γ2L structure¹⁴ and the pore radius is 2.38 Å (Fig. 4 b-d). Thus, the channel structure delimits a closed, non-conducting state (given the 1.8 Å Pauling radius of a dehydrated chloride ion). The concept of distinct hydrophobic gates controlling transitions between resting, active and desensitized states has been extensively studied at GABA_ARs and homologous receptors^15,25,26. Here the principal impediment to the flow of ions is formed by the cytoplasmic desensitization gate, thus our structure illustrates a desensitized channel conformation.

Figure 4. Conductance and permeation pore structure of the α1β3γ2L GABA_AR.

a. Cutaway of the receptor showing electrostatic surface potential along the ion conducting pathway. b. Asymmetry in the channel at the level of the activation and desensitization gates. -2’ and 9’ residues are shown in ball-and-stick representation. Distance between Cα of -2’ and 9’ residues are given in Å. c. M2 α-helices from opposing α1 and β3 subunits with sidechains shown for pore lining residues. Spheres represent the solvent accessible volume of the ion channel. Red spheres delimit the narrowest aspect of the channel. d. Profile of pore radius of the α1β3γ2L-Mb38 complex, alprazolam/GABA-bound α1β3γ2L (ALP; PDB ID: 6HUO) and bicuculline-bound α1β3γ2L (BCC; PDB ID: 6HUK) and the benzamidine-bound β3 GABA_A (BEN; PDB ID: 4COF).

The M1, M3 and M4 α-helices delineate binding sites for a range of compounds, including neurosteroids and general anaesthetics^22,23,27. The β⁺/α^– interface has been extensively studied with regard to binding and modulation by anaesthetics and alcohol^2,27. Accordingly, we observe a binding cavity lined by residues identified in photolabelling studies with anaesthetic derivatives²⁸ and by Asn285 (β3 15’ M2), mutation of which strongly reduces propofol and etomidate sensitivity²⁹ (Fig. 5a, b). Additionally, transmembrane residues labelled by the barbiturate photolabels mTFD-MPAB or mTFD-MPP^27,30 map to a cavity formed at the γ2⁺/β3^- interface, lined at the back by γ2 Ser280 (15’ M2) (Fig. 5c, d). In contrast, the α1⁺/γ2^– interface appears locked by a continuous network of interactions, primarily hydrophobic (Extended Data Fig. 6f). This is consistent with the α⁺/γ^– TMD interface remaining an ‘orphan site’, to date, with respect to receptor pharmacology²⁷.

Figure 5. Anesthetic binding sites in the α1β3γ2L GABA_AR TMD.

a. Exposed surface of the β3-α1 subunit interface coloured according to electrostatic surface potential. b. Close up view of a cavity formed at the transmembrane β-α interface. Residues identified in photolabeling studies with etomidate and propofol derivates are outlined in magenta and blue respectively. c. Exposed surface of the γ2-β3 subunit interface coloured according to electrostatic surface potential. d. Close up view of a cavity formed at the transmembrane γ-β interface. Residues identified in photolabeling studies with propofol and barbiturate derivates are outlined in blue and orange respectively.

While sequence analyses do not predict significant secondary structure in the ICD for GABA_AR subunits, this domain is known to form a scaffold for auxiliary proteins^16,31. We observed protein density extending from the intracellular ends of M3 and M4 helices, most noticeably for the α1 subunits, where we were able to model part of the post-M3 (Arg313-Val323) and pre-M4 regions (Thr385-Ser390; Extended Data Fig. 9g, h). Beyond their contribution to the PIP2 binding sites described above, the α1^– post-M3 regions extend towards the neighbouring β3⁺ subunits where interactions between α1 Trp317, β3 Phe307 and β3 Pro311 further enhance the stability of β⁺/α^– interfaces (Extended Data Fig. 9g, h).

We present here the first structure of a GABA_AR in a lipid bilayer. The agonist-binding site and the ECD adopt activated conformations, while the ion channel resides in a non-conducting, desensitised state. Importantly, this structure is different to those of related GABA_ARs reported recently in the presence of detergents, which appear to significantly affect inter and intra-subunit interfaces^8,9. We also report an α-subunit specific interaction between the GABA_AR and inner leaflet lipid PIP2, which merits a detailed investigation in follow-up studies. In a physiological context, this interaction may serve to sequester receptors to specific lipid microdomains, where receptor trafficking can be precisely regulated. This hypothesis draws parallels to the proposed coupling of the Torpedo nAChR and cholesterol-rich microdomains³². Our results open avenues for future functional and structural studies of GABA_AR modulation by lipids and small molecules, such as anaesthetics and barbiturates, that target the transmembrane domain.

Methods

GABA_A receptor expression and purification

Human α1β3γ2L receptors were expressed using a stable tetracycline-inducible HEK293 cell line¹³ adapted to suspension culture. The cell line was not authenticated or tested for mycoplasma contamination. Briefly the cell line comprises the full-length human α1 (Uniprot ID P14867), β3 (Uniprot ID P28472) and γ2L (Uniprot ID P18507-2; ‘long’ isoform) genes, each under individual antibiotic selection. For purification purposes a FLAG-tag was inserted at the N-terminus of the α1 subunit and a (GGS)₃GK-1D4 tag inserted at the C-terminus of the γ2L subunit. Cells were grown in suspension in Freestyle293 medium supplemented with fetal bovine serum (Invitrogen) and selection antibiotics (geneticin, hygromycin-B, zeocin and blasticidin) at 37°C and 8% CO₂. At a cell density of ~1.5-2x10⁶ cells ml^-1 expression was induced by addition of doxycycline and sodium butyrate (Sigma) added at 5 mM to enhance expression. Cells were harvested by centrifugation after ~36 h. Cell pellets from ~1 l of suspension culture were resuspended in 20mM HEPES pH 7,2, 300mM NaCl, 1 % (v/v) mammalian protease inhibitor cocktail (Sigma-Aldrich). For solubilization a 10% (w/v) solution of lauryl maltose neopentyl glycol (LMNG, Anatrace) and cholesterol hemisuccinate (CHS, Anatrace) was prepared at 10:1 molar ratio respectively, and added at 1% (v/v) to the cell suspension. Cells were solubilized for ~1.5 h at 4°C and cell debris and insoluble material removed by centrifugation (10,000 g, 30 min). The supernatant was incubated with 1D4 affinity resin²² with gentle mixing at 4°C for ~2 h. Resin was recovered by centrifugation and washed extensively with buffer (20mM HEPES pH7.2, 300mM NaCl) containing 0.1% (v/v) LMNG before on-bead reconstitution into nanodiscs.

Nanodisc reconstitution

MSP2N2 (Addgene, ID 29520) was expressed and purified from E. coli as previously described³³. Owing to the tendency for detergent purified receptors to aggregate upon elution, α1β3γ2L receptors were incorporated into nanodiscs while attached to 1D4 resin. Briefly, receptors attached to 1D4 resin were mixed with ~1 bed volume of buffer containing phosphatidylcholine (POPC, Avanti; prepared as a 10mg/ml stock in 3% n-dodecyl-β-D-maltoside, DDM, Anatrace – POPC final concentration ~1.2mM) supplemented with bovine brain extract (Sigma-Aldrich – Type I,Folch Fraction I; prepared as a 20mg/ml stock in 3% DDM) at 4°C for ~0.5 h. Lipids were added at a relative ratio of 85:15% POPC:brain extract. Excess lipids were removed by centrifugation before addition of an excess of MSP2N2 (~0.6mg/ml final concentration) to the receptor-bound resin-lipid mixture. The mixture of receptor-bound resin and MSP2N2 were pre-mixed gently at 4°C for ~0.5 h before addition of bio-beads SM2 (~60mg/ml; Bio-rad) to initiate reconstitution. At this point the mixture was incubated at 4°C for ~4 h with constant rotation. The resin/bio-bead mixture was recovered by centrifugation and washed extensively to remove empty nanodiscs; first with 20mM HEPES pH7.2, 300mM NaCl, next with 10mM HEPES pH7.2, 150mM NaCl and finally 10mM HEPES pH7.2, 100mM NaCl. Bound receptor-nanodisc complexes were eluted by addition of 1mM 1D4-peptide in final wash buffer at 4°C for ~16 h. The mixture was applied to EconoPac column (Bio-rad) to remove bio-beads and resin, the receptor-nanodisc complex transferred to 100 kDa Amicon Ultra concentrator (Millipore) and concentrated to 0.5 mg/ml for cryo-EM grid preparation. Sample monodispersity was assessed by analytical Size Exclusion Chromatography on a Shimadzu HPLC system and negative-stain EM. On the basis of this analysis, further purification of the sample before grid preparation was not required. Western blot analysis of cell lysates and purified receptors was performed using the following primary antibodies; rabbit anti-GABA_A α1 (Millipore; 06-868), mouse anti-GABA_A β2/3 (Millipore; MAB341) and mouse anti-GABA_A γ2 (Synaptic Systems; 224003). Immunoreactivity was detected with secondary antibodies conjugated to horseradish peroxidase.

Generation of Mb38

Megabodies (Mbs) are rigid antibody chimeras, built from nanobodies (Nbs) that are grafted onto large scaffold proteins (examples and detailed descriptions of which are to be published in a dedicated methods paper). The particular construct used here, $\text{M}{\text{b}}_{\text{Nb}38}^{\text{cHopQ}}$ (called Mb38 throughout the text for simplicity), is a circular permutant of the extracellular adhesin domain of H. pylori (HopQ, UniProt B5Z8H1)³⁴ which was inserted into the first β-turn connecting β-strands A and B of Nb38⁷. Specifically, the Mb38 open reading frame contains the DsbA leader sequence, followed by a consensus sequence encoding the conserved β-strand A of the Nanobody fold (residues 1-13), followed by the C-terminal part of HopQ (residues 227-449), followed by a three-residues linker connecting the C and the N-termini of HopQ, followed by the N-terminal part of HopQ (residues 49-221). The construct has C-terminal His6 and EPEA tags in frame for affinity purification. Mb38 is monomeric, with a molecular weight of ~58 kDa. Similar to Nbs, Mb38 was expressed as a secreted protein in the periplasm of E. coli WK6 bacteria and purified using nickel affinity chromatography followed by size-exclusion chromatography on a Superdex 200 16/60 column (GE Healthcare) in 10 mM Tris, pH 7.3,140 mM NaCl. The stock solutions were concentrated to 15 mg/mL, snap-frozen in liquid nitrogen and stored at -80°C.

Cryo-EM sample preparation

Purified α1β3γ2L-nanodisc complexes at 0.5mg/ml were mixed with Mb38 (at 1.2μM) and incubated on ice for 0.5 h prior to cryo-EM grid preparation. 3.5 microlitres of sample was applied to glow-discharged gold R1.2/1.3 300 mesh UltraAuFoil grids (Quantifoil) before blotting for 5 s at ~100% humidity and 14°C, and then plunge-frozen in to liquid ethane using a Vitrobot Mark IV (FEI).

Cryo-EM image collection and processing

Data from a single grid was collected over two 24 h sessions on a FEI Titan Krios (ThermoFisher) operated at 300kV and equipped with a Falcon 3EC direct-detector and Volta phase plate. A total of 784 micrographs were collected in electron-counting mode at a nominal magnification of 75,000 corresponding to a calibrated pixel size of 1.07Å. Each micrograph was collected as a 60 s exposure, comprised of 75 movie frames at a dose rate of 0.55e^-/p/s for a total accumulated dose of 30.84 e^-/Ų. Drift correction, beam induced motion and dose-weighting were performed with MotionCor2³⁵ using a 5 x 5 patch. CTF fitting and phase shift estimation were performed on motion corrected sums without dose weighting with Gctf-v.1.18³⁶. Micrographs were manually checked and a total of 737 micrographs (motion corrected sums with dose weighting) used for all other image processing. An initial round of auto-picking on small subset of micrographs was performed in RELION 2.1³⁷ using a Gaussian blob as a template. Picked particles were subjected to reference-free 2D classification and good classes used for auto-picking on the entire dataset. Autopicked particles were manually inspected and false positives removed, resulting in a total of 370,757 particles. Particles were 2 x binned (pixel size 2.14 Å) for initial processing. Two rounds of 2D classification were carried out in RELION 2.1 and 197,673 particles within classes exhibiting recognizable GABA_AR channel features selected for further processing. An initial 3D reference model was generated in RELION and low-pass filtered to 50Å. 3D classification into six classes resulted in four good classes (106,039 particles) and two classes comprised principally of top and bottom views of the receptor-Mb38 complex. Particles from the four best classes were combined, re-extracted (1.07 Å pixel) and refined in RELION 2.1. Particle polishing (RELION 2.1) and CTF-Refine (RELION 3.0) further improved the resolution of the initial reconstruction. A second round of 3D classification, with no image alignment, yielded a single class containing 55,449 particles. 3D refinement of this class yielded a reconstruction at 3.75 Å before post-processing. Applying a soft mask in RELION Post-processing, yielded a final EM map of 3.22 Å. Notably, the calculated soft mask also accounted for density contributed by the nanodisc. Resolution was estimated using Fourier Shell Correlation (FSC) = 0.143 criterion. Local resolution was calculated using the MonoRes implementation in Scipion^38,39.

Model building, refinement and validation

An initial rigid body fit of coordinates for the truncated α1β3γ2⁷ was carried out using UCSF Chimera⁴⁰. Owing to the improved resolution of the reconstruction and apparent stability of the receptor complex in the nanodisc, a complete model of the receptor, incorporating nearly all residue side chains, could be built with high accuracy in local geometry. The model was first manually adjusted and refined using COOT⁴¹ and then subjected to global refinement and minimization in real space using PHENIX (‘phenix.real_space_refine’)⁴². Geometric constraints for lipids were generated using GradeServer (Global Phasing Ltd.) and the ligand coordinates docked into densities and refined using COOT⁴¹.

The ECD and TMD regions in all subunits were well resolved in EM density map, as were the Mb38 and N-linked glycans. However, the HopQ scaffold protein fused to Mb38 was comparatively flexible, with no apparent EM density in the final reconstruction. While density contributed by the nanodisc was evident in the EM reconstruction, the resolution did not permit modelling of the helical belt formed by MSP2N2 scaffold proteins. The final construction allowed for unambiguous assignment of α1, β3 and γ2 subunits and main-chain and sidechain model building. Strong density was observed protruding from the “intracellular” side of the nanodisc proximal to the α1 subunit TM3 and TM4 helices. We modelled this as ~10 residues of the M3-M4 loop of the α1 subunit. Strong density, unaccounted for by the protein backbone, was also apparent at the base of α1 TM3 and M4 helices. This was modelled as PIP2 on the basis of its characteristic shape and orientation at the inner leaflet of the lipid bilayer. Additional density in the “extracellular” layer of the nanodisc was modelled as POPC moieties. Cryo-EM map densities observed within the ion channel and in the (pseudo-)agonist sites were not modelled as we could not reliably determine the identity of a bound ligand.

During model building and refinement, model geometry was assessed using MolProbity⁴³. For cross-validations, the final model was refined against the half maps generated by 3D auto-refine in RELION. Subunit interfaces and associated free energies were analysed using the PDBePISA server⁴⁴. Pore diameters were calculated using the HOLE implementation in COOT. Structural figures were made using UCSF Chimera⁴⁰ and Pymol (Schrodinger, LLC), including the APBS⁴⁵ electrostatics plugin. Structural alignments were performed using the Superpose command in CCP4mg⁴⁶ or MatchMaker⁴⁰. Tunnel radii calculations were carried out using the Caver plugin⁴⁷ in Pymol.

Density maps shown in figures were contoured at the following levels. Fig. 1, sharpened maps for the receptor and glycans contoured at 0.03 and unsharpened map contoured at 0.02. Fig. 6b density around PIP2 at 0.021. Extended Data Fig. 2a-c, subunits, density maps contoured at 0.03; d, e, glycans, density maps contoured at 0.025; f-h, loop C, density maps contoured at 0.03; i-k, TM helices, density maps contoured at 0.03; l density maps contoured at 0.03. Extended Data Fig. 3b, glycan, density maps contoured at 0.025. Extended data Fig. 7b, density map contoured at 0.045. Extended Data Fig. 9a, b density map contoured at 0.02.

Computational docking

Computational docking of small molecules to GABA_AR cryo-EM structures was performed using AutoDock Vina⁴⁸. Structures of small molecules were optimised using Grade webserver (Global Phasing Ltd.). GABA_AR was kept rigid during docking. The whole agonist-binding pocket (~20×20×20 ų) centered under loop-C was selected for docking.

Electrophysiology

HEK 293 cells stably transfected with tetracycline inducible human (N)-FLAG-α1β3γ2L-(GGS)3GK-1D4 GABA_A⁴⁹, were grown on glass coverslips, and expression was induced by 0.1-2 μg/ml doxycycline for 14 to 28 hours depending on the level of current required. For pulling inside-out macro-patches, we used poly-L-lysine coated glass coverslips (Becton Dickinson, Franklin Lakes, NJ). Cells were also treated with kifunensine (5 μg/ml) at the time of induction.

Currents were recorded in either whole-cell or inside-out configuration of patch clamp as previously described⁵⁰ using an Axopatch 200A amplifier (Molecular Devices, San Jose, CA). Ligands were applied via a quad-channel super-perfusion pipette coupled to a piezoelectric element that switched the super-perfusion solution in <1 ms as described previously⁵¹. Data were acquired using Clampex 8.2 (Molecular Devices, San Jose, CA). Data were acquired at either 10, 5 or 2 kHz and filtered at 5, 2 or 1 KHz respectively depending on the duration of the pulse, with longer the recording, the slower the data acquisition rate. Cells were continually perfused with a bath solution consisting of (in mM): 145 NaCl, 5 KCl, 10 HEPES, 2 CaCl₂, 1 MgCl₂ and 10 glucose, pH 7.4 (adjusted with NaOH). The pipette solution consisted of (in mM): 140 KCl, 10 HEPES, 1 EGTA, 2 MgCl₂ and 2 Mg-ATP, pH 7.3 (adjusted with KOH). For whole-cell recordings, the open pipette resistances ranged from 1.3 – 2.5 MΩ, cell capacitance ranged from 5 – 13 pF and series resistance ranged from 0.2 – 2.5 MΩ. Series resistances were electronically compensated by >85% with a lag of 10 μs. The liquid junction potential between the bath and pipette solution was −2 mV and was corrected post recordings. Cells and patches were voltage-clamped at −52 mV. For inside-out patches, the solutions were reversed, i.e., the pipettes were filled with the bath solution and the patches were perfused with the pipette solution. The pipette resistance for inside-out patches ranged from 4 – 6 MΩ. In all patch protocols, initially currents were repeatedly elicited with pulses of etomidate until stable currents were obtained before proceeding with the application of modulatory agents as described in the figure legend. For analysis 2 – 5 identical traces from the same patch were averaged, and the ones before the patch was stabilized were disregarded. The peak current amplitudes obtained with inside-out patches ranged between 7 – 44 pA.

Electrophysiology Data Analysis

Current traces were analyzed using Clampfit 9.0 (Molecular Devices, San Jose, CA). Statistical analysis (paired or unpaired Students t test) was done in Graphpad Prism 6 (Graphpad Software, La Jolla, CA). P < 0.05 was considered statistically significant. Data are shown as mean ± SD. Each n represents a different patch or cell. Current traces were normalized in Origin 6 for presentation. The enhancement data for Mb38 was fitted by nonlinear least squares in Igor Pro (Wavemetrics, Lake Oswego, OR).

Radioligand binding

The functionality of the receptors reconstituted in nanodiscs, in the presence or absence of Mb38 megabody, was assessed using radioligand binding assays. Modulation of 2 nM [³H]flunitrazepam or [³H]muscimol binding by varying concentrations of GABA or etomidate, respectively, was measured using WGA coated YSi scintillation proximity assay beads (SPA, PerkinElmer). Background binding was determined by displacement of [³H]flunitrazepam or [³H]muscimol with 100 μM diazepam or with 10 mM GABA respectively. Each reaction contained ~5 μg/ml purified receptor and 1 mg/ml SPA beads suspended in 10 mM HEPES pH7.5 and 100 mM NaCl. Data normalization, graph preparation and nonlinear least squares curve fitting to a single-site binding model were performed in Igor Pro 7 (WaveMetrics).

Extended Data

Extended Data Fig 1. Cryo-EM image processing procedure.

a. Overview of cryo-EM data collection and image processing procedure (see Methods). b. Representative micrograph of the α1β3γ2L-Mb38-nanodisc complex obtained using Falcon3 detector and VPP. c. Representative 2D class averages for downscaled data (box size of 240 Å). d. FSC curves for the reconstruction before and after applying a soft mask. e. The unsharpened map from refinement coloured by local resolution estimate (calculated using MonoRes³⁸) and shown at a low isosurface level to enable visualization of the nanodisc (left) and at a high isosurface level (right). f. Angular distribution histogram of particle used in calculating the final 3D reconstruction for the receptor assembly. g. FSC curves for cross-validation between maps and models: model versus summed map (black), model versus half map 1 (used in test refinement, green), model versus half map 2 (not used in test refinement, blue).

Extended Data Fig 2. Sequence alignment of GABA_AR α1, β3 and γ2 subunits, biochemical characterization and binding assays.

a. Alignment of wild-type GABA_AR subunit sequences, where number one represents first residue of the mature protein. α-helices (grey cylinders), β-strands (black arrows) and associated loops are indicated. Glycosylation sites are indicated by blue pentagon and the associated subunit residue highlighted in blue. Residues identified as coordinating PIP2 binding are highlighted in yellow and indicated by yellow hexagons. The alignment graphic was prepared on the ESPript 3.0 server (http://espript.ibcp.fr/ESPript/ESPript/). b. Structure of a single α1 subunit. c. Western blot analysis of cell lysates from LMNG solubilized control HEK293 cells and α1β3γ2L GABA_AR cells, and purified α1β3γ2L GABA_ARs in nanodiscs. The arrowhead denotes the band corresponding to the full-length GABA_AR subunits which migrates as a species of ~51-55 kDA. With the exception of the α1 subunit (which displays a small degree of proteolysis following reconstitution – denoted by asterisk), GABA_AR subunits do not display apparent proteolysis during solubilization, purification and reconstitution. Western blots were repeated twice independently with similar results. d. GABA enhanced displaceable [³H]flunitrazepam binding to purified receptors in a concentration–dependent manner in the presence or absence of Mb38. Points represent individual samples from two separate experiments.

Extended Data Fig 3. α1β3γ2 model-map validation and EM density.

a-m. EM density segments for representative regions of each subunit and Mb38. Sharpened map contoured as detailed in Methods.

Extended Data Fig 4. Comparison of TMD architecture with α1β2γ2 and α1β1γ2 GABA_AR structures in detergent.

a-f. Superposition of subunit chains of α1β3γ2L GABA_AR (grey) onto equivalent chains of α1β2γ2 GABA_AR in DDM (Conformation B: PDB 6D6T) and the γ2 subunit of α1β2γ2 GABA_AR (Conformation A: PDB 6D6U). RMSD values are for Cα atoms over entire subunit. g. Overview of the TMD of α1β3γ2 in nanodisc. h-j. Superposition of TMD for α1β2γ2 GABA_AR conformation A (h), conformation B (i), α1β1γ2 GABA_AR (PDB: 6DW0 - j) and α1β3γ2-nanodisc complex (grey ribbon). TM helices of the γ2 subunit are labelled. The TM helices of the γ2 show significant distortion in detergent-bound complexes. M4 helices in α1 and γ2 subunits were not modelled in the α1β1γ2 GABA_AR. k-n. Superposition of β^- subunits reveals conformational differences of γ2⁺ subunit (α1β3γ2-nanodics complex in grey). Differences of distance (Å) between selected residue Cα atoms (spheres) is indicated by lines. Disruption of the γ2 TMD induces substantial displacement of loop 7, loop 2 and the M2-M3 loop at the ECD-TMD interface in the detergent bound α1β2γ2 structures (l, m), and to a lesser extend in α1β1γ2 GABA_AR (n). o-r. Close up view of M2 helices at level of -2’ proline/alanine residues (Cα atoms shown as sphere) in nanodisc (o) and detergent bound structures (p-r).

Extended Data Fig 5. Mb38 binding and function.

a. Side and b. top-down view of neighboring α1⁺ and β3^- subunits bound to Mb38. c. Representative normalized current traces obtained in cells expressing α1β3γ2L receptor exposed to GABA (3 μM) alone or with varying concentrations of Mb38 (n = 3-6 cells), applied for 4 s. Currents were normalized to peak current amplitude obtained with GABA (3 μM) alone during the first 1s phase of the trace. The concentration of Mb38 is color-coded as indicated in the legend. d-f. Close-up view of the binding site when viewed approximately parallel to the plane of the membrane. CDR loops 1 (d), 2 (e) and 3 (f) of the Mb38 are colored in turquoise, teal and dark-green respectively and residues involved in interactions shown in ball-and-stick representation. Polar interactions are shown as dotted lines. g. Representative current trace obtained in cells expressing α1β3γ2L receptor exposed to 3 μM Mb38. Mb38 (3 μM) opened 16 ± 11 % (Mean ± SD; n = 5 cells) of the receptors gated by EC₁₀ GABA alone (therefore ~2% of the total receptors expressed).

Extended Data Fig 6. Vestibular glycans and interface classes in the α1β3γ2L GABA_AR.

a. Side view of the receptor shows the position of vestibular α1 N-linked glycans. For clarity, the near α1 and β3 subunits have been removed. b. View across the extracellular vestibule reveals the stacking of α1 N-linked glycans. Receptor surface is coloured according to electrostatic surface potential and reveals an electropositive ring in middle portion of the ECD vestibule. c-f. Paired views of the interface between principle (+) and complementary (-) subunits viewed from the pore axis outwards (left) and open book view of each subunit when viewed from the receptor periphery (right). Residues involved in forming interactions (defined using PDBePISA⁴⁵) are colored according to the type of interaction and mapped onto the isosurface representation; polar – cyan, electrostatic/salt bridges – magenta and van der Waals – orange. Arrowheads reveal the openings of defined tunnels between adjacent subunits g. Calculated interfacial buried surface areas and solvation energy gain at complex formation (both calculated using PDBePISA⁴⁴). The asterisk denotes the second β3 - α1 (chain E-chain A) interface in the pentameric assembly. Radii of tunnels, denoted by arrowhead in panel c-f, were also calculated (see Methods). Open arrowheads in panels c, e denotes cavities forming proposed anesthetic binding sites (as discussed in the text).

Extended Data Fig 7. Disease mutations associated with α1, β3 and γ2, lateral tunnels and fenestrations at the subunit interfaces.

a, b. Disease mutations associated with GABA_A α1, β3 and γ2 subunits are mapped onto the structure and shown in sphere representation. The receptor is viewed parallel to the membrane plane (a) and from the extracellular aspect (b). Outlined boxes highlight position of mutations shown in the panels (c-e). c-e. Close up view of disease mutations associated with the α1 and β3 subunits. Polar interactions between residues are shown as dotted lines. f. Table summarising a number of disease mutations identified in genes for α1, β3 and γ2. Functional effects as determined from experimental studies of channels properties are summarised. g. Exposed surface of the γ2-β3 subunit interface coloured according to electrostatic surface potential. h. Close up view of an electronegative fenestration formed at the γ2-β3 extracellular domain interface. The continuous tunnel leading from extracellular space to the receptor vestibule is outlined. i. Exposed surface of the α1-β3 subunit interface coloured according to electrostatic surface potential. j. Close-up view of the α1-β3 extracellular interface reveals an upper tunnel leading to the mid-portion of the ECD vestibule. A lower tunnel (denoted by arrow) opens into the upper aspect of the ion channel at the level of β3His267, a residue implicated in mediating the effects of propofol⁵⁹.

Extended Data Fig 8. Lipid binding sites and functional modulation of GABA_AR by PIP2.

a. Well resolved density for POPC lipid moiety (yellow, ball-and-stick representation) at the extracellular aspect of the lipid nanodisc. EM density is shown in chicken wire representation and contoured around the lipids. b. Sequence alignment of GABA_AR and GlyR subunits for PIP2 binding regions; the M1-M2 loop, post-M3 and pre-M4 segments. α1 residues forming hydrogen-bonds or salt-bridge interactions with PIP2 are identified by yellow hexagons, and those which are conserved amongst receptor subunits are highlighted in orange (identical) and yellow (similar). The alignment graphic was prepared on the ESPript 3.0 server (http://espript.ibcp.fr/ESPript/ESPript/). c. Representative normalized current traces from the same patch, obtained in a two-pulse protocol, where inside-out patches were exposed to two 5 s etomidate (100μM) pulses 7.5 s apart. During the second pulse, etomidate was either applied alone or co-applied with PLL (250 μg/ml). Currents traces were normalized to the peak current amplitude obtained during the first etomidate pulse. d. Dot plot of peak current amplitudes obtained during the second pulse (co-application of PLL) normalized to the peak current amplitudes obtained with first pulse of etomidate (center value represents mean ± SD; n = 9 patches). Unpaired and paired Students t test (Two-tailed) values obtained are given in the figure.

Extended Data Fig 9. Comparisons of agonist sites and analogous pockets at other subunit interfaces.

a. View of the receptor from across the synaptic cleft with the agonist binding sites highlighted. b. Cut-away view of (a) at the level of loops C, reveals EM density (shown as magenta chicken wire representation) at four inter-subunit pockets. c. EM density in the orthosteric binding pocket shown (grey surface representation). For comparison, the top binding pose for GABA is displayed in grey ball-and-stick representation. d, e. Overlay of GABA binding poses from molecular docking calculations at the β3-α1 (d) and α1-β3 (e) binding pockets. The range of estimated free energies of binding (kcal/mol) is given. f. Comparison of the orthosteric binding pocket at the β3-α1 interface (grey), with the 3 unique interfaces observed in the α1β3γ2-Mb38 receptor (coloured as in a.) Superposition of the (-) subunit ECD reveals the relative movement of the (+) subunit ECD. g, h. Modelling of the intracellular end of M3 and M4 helices, contributing to the receptor intracellular domain, shown approximately parallel (g) and perpendicular (h) to the plane of the membrane.

Extended Data Table 1. Cryo-EM data collection, refinement and validation statistics.

	α1β3γ2-Mb38 Nanodisc (EMDB-4411) (PDB 6I53)
Data collection and processing
Magnification	75,000
Voltage (kV)	300
Electron exposure (e–/Å2)	30.84
Defocus range (μm)	-0.5 to -0.7
Pixel size (Å)	1.07
Symmetry imposed	C1
Initial particle images (no.)	370,757
Final particle images (no.)	55,559
Map resolution (Å)	3.22
FSC threshold	0.143
Refinement
Map sharpening B factor (Å2)	-68
Model composition
Non-hydrogen atoms	15,366
Protein residues	1,821
Ligands	558
B factors (Å2)
Protein	156
Ligand	180
R.m.s. deviations
Bond lengths (Å)	0.06
Bond angles (°)	0.801
Validation
MolProbity score	1.43 (100th percentile)
Clashscore	3.53 (100th percentile)
Poor rotamers (%)	0
Ramachandran plot
Favored (%)	95.8
Allowed (%)	4.2
Disallowed (%)	0

Supplementary Material

Supplementary Information is available in the online version of the paper.