Structure of human GABA_B receptor in an inactive state

Abstract

The human GABA_B receptor—a member of the class C family of G-protein-coupled receptors (GPCRs)—mediates inhibitory neurotransmission and has been implicated in epilepsy, pain and addiction¹. A unique GPCR that is known to require heterodimerization for function^2,3,4,5,6, the GABA_B receptor has two subunits, GABA_B1 and GABA_B2, that are structurally homologous but perform distinct and complementary functions. GABA_B1 recognizes orthosteric ligands^7,8, while GABA_B2 couples with G proteins^{9,10,11,12,13,14}. Each subunit is characterized by an extracellular Venus flytrap (VFT) module, a descending peptide linker, a seven-helix transmembrane domain and a cytoplasmic tail¹⁵. Although the VFT heterodimer structure has been resolved¹⁶, the structure of the full-length receptor and its transmembrane signalling mechanism remain unknown. Here we present a near full-length structure of the GABA_B receptor, captured in an inactive state by cryo-electron microscopy. Our structure reveals several ligands that preassociate with the receptor, including two large endogenous phospholipids that are embedded within the transmembrane domains to maintain receptor integrity and modulate receptor function. We also identify a previously unknown heterodimer interface between transmembrane helices 3 and 5 of both subunits, which serves as a signature of the inactive conformation. A unique ‘intersubunit latch’ within this transmembrane interface maintains the inactive state, and its disruption leads to constitutive receptor activity.

Main

We assembled the heterodimeric human GABA_B receptor by using baculovirus-infected mammalian cells. Each subunit was truncated at the carboxy-terminal end according to the domain boundary of the intracellular coiled-coil¹⁷, in order to eliminate flexible regions (Supplementary Fig. 1). Upon extraction and purification with detergent (Extended Data Fig. 1a, b and Supplementary Fig. 2), the complex bound to radioactive [³H]GABA with a dissociation constant comparable to the GABA affinity reported for native receptors^3,4,7 (Extended Data Fig. 1c). Functional analysis incorporating a chimaeric Gα_qi5 protein and inositol phosphate assay¹⁸ in mammalian cells showed that the agonist baclofen activated the C-terminally truncated and full-length receptor with similar potency and efficacy, indicating that the deleted regions in the GABA_B1b and GABA_B2 cytoplasmic tails are not required for ligand-mediated G-protein activation (Extended Data Fig. 1d).

We determined the structure of the heterodimeric GABA_B receptor by cryo-electron microscopy (cryo-EM) to an overall resolution of 3.3 Å (Extended Data Fig. 2a–e and Supplementary Table 1). The global density map displayed directional anisotropy owing to linker flexibility (Extended Data Fig. 2f). Performing local refinement separately on the extracellular domains (ECDs) and transmembrane domains yielded reconstructions at resolutions of 3.1 Å and 3.4 Å, respectively (Extended Data Fig. 2g, h). We used a composite map that combined the ECD and transmembrane reconstructions for model building and refinement (Extended Data Fig. 3). By applying three-dimensional variability analysis to the data, we found that the receptor is in dynamic motion, and that its functional state corresponds to a continuum of conformations along multiple dimensions (Supplementary Videos 1–4).

The heterodimeric GABA_B receptor is assembled when GABA_B1b and GABA_B2 subunits interact side-by-side while facing in opposite directions (Fig. 1a, b and Extended Data Fig. 4). Both the VFT and the transmembrane components of the two subunits are related by pseudo-twofold axes. Extracellular and intracellular loops (ECLs and ICLs) that interconnect adjacent helices within each transmembrane domain are visible in the density map, except for ICL2. The cytoplasmic tail including the coiled-coil domain is disordered, possibly because of its flexible attachment to the transmembrane domain.

Fig. 1: Cryo-EM structure of human GABA_B receptor.

a, Ribbon representation of the GABA_B-receptor structure, composed of GABA_B1b (blue) and GABA_B2 (green) subunits. Ca²⁺ is shown as a green sphere; phospholipids (PE 38:5 and PC 38:2) are shown as yellow space-filling models; N-linked glycans (NAGs) and cholesterols (CLRs) are shown, respectively, as grey and pink ball-and-stick models. Transmembrane helices 1–7 are marked for each subunit. b, Cryo-EM density map of the GABA_B receptor, composed of local reconstructions for extracellular (3.1 Å) and transmembrane (3.4 Å) domains, in an orthogonal view from a. c, Linker domains of GABA_B1b and GABA_B2, showing the main-chain and side-chain hydrogen-bonding patterns between the linker region and ECL2.

The elongated peptide linker that joins the VFT to the transmembrane domain is buttressed through its interaction with a β-hairpin formed by ECL2 (Fig. 1c). ECL2 twists across the linker, forming a united mechanical junction that transmits the conformational changes in the VFT to the transmembrane domain and vice versa. Additionally, the ECD and transmembrane domains of the receptor spontaneously flex back and forth about the linker, exhibiting the region’s intrinsic flexibility (Supplementary Video 1).

Given that no ligand was added during protein purification, we expected the receptor to be in an apo form and an inactive conformation. To our surprise, we observed several ligands bound to the receptor. GABA_B1b contains a Ca²⁺ ion at the interdomain cleft of VFT. In addition, an endogenous phospholipid is bound within the transmembrane pocket of each subunit, with a phosphatidylethanolamine (PE 38:5) in GABA_B1b, and a phosphatidylcholine (PC 38:2) in GABA_B2. Finally, ten cholesterol or cholesteryl hemisuccinate molecules, which we modelled as cholesterol (see Methods), are distributed around the exterior of the transmembrane complex, including two that interface with both subunits.

Inactive conformation of GABA_B receptor

The cryo-EM structure of the GABA_B receptor appears to occupy an inactive conformation, given its similarity to known crystal structures of the GABA_B VFT in its apo and antagonist-bound states¹⁶. First, the VFT module, composed of the lobe-shaped LB1 and LB2 domains, adopts an open interdomain conformation in both subunits (Extended Data Fig. 5a, b). Second, although an amino-terminal LB1–LB1 dimer interface is present in all functional states, the distinct lack of a heterodimer interface between the membrane-proximal LB2 domains is shared by the near full-length cryo-EM structure and inactive-state VFT structures (Extended Data Fig. 5c, d). By contrast, a hallmark of the active-state VFT structures is a unique heterodimer interface between LB2 domains, which results from agonist-induced GABA_B1b closure¹⁶.

Using conformational variability analysis, we observed that the LB2 domains of both subunits fluctuate by approaching and withdrawing from the central vertical axis of the heterodimer, yet never make contact as they do in active-state VFT crystal structures (Supplementary Videos 2, 3). The motion exhibited by the LB2 domains and their associated linkers suggests that the inactive functional state corresponds to an ensemble of conformations, where the separation between the membrane-proximal regions falls within a small range around the coordinates of our structure.

Inactive transmembrane heterodimer interface

We identified a hitherto unknown heterodimer interface between the transmembrane TM5 and TM3 helices of both subunits that embodies the signature of the GABA_B receptor’s inactive conformation (Fig. 2a). A TM5–TM5 contact has previously been detected through crosslinking of the GABA_B receptor¹⁹, but transmembrane dimer interfaces of any kind are yet to be found in other inactive class C GPCRs, including the recent inactive metabotropic glutamate receptor mGlu₅ structure²⁰ (Extended Data Fig. 5e).

Fig. 2: Transmembrane heterodimer interface of the GABA_B receptor.

a, The transmembrane heterodimer interface formed by TM5 and TM3 of both subunits. Three layers (I, II and III) of interfacial contacts are identified by dotted circles. Direct heterodimer contacts within each layer are shown on the right. b, c, Functional analysis of the ‘intersubunit latch’. Basal activity (b) and dose-dependent baclofen-stimulated receptor response (c) in cells transiently expressing a Gα_qi5 chimaera (abbreviated as Gqi) with different combinations of GABA_B wild-type subunits (WT) and mutant subunits (GABA_B1b E673R, abbreviated as E673R; GABA_B2 H579E, abbreviated as H579E). Inositol-1-phosphate (IP₁) accumulation was measured in the presence or absence of 20 μM CGP54626 (abbreviated as INV). Cells transfected with empty pcDNA3.1 vector (abbreviated as pcDNA), Gα_qi5 alone, or WT GABA_B subunits in the absence of Gα_qi5 were used as controls. Relative activity is expressed as percentage of maximum wild-type activity induced by baclofen relative to the activity of Gα_qi5 alone. Data points represent averages ± s.e.m. of multiple experiments (n values shown), each with quadruplicate measurements. ****P < 0.0001; one-way analysis of variance (ANOVA) with Bonferroni’s post hoc test was used to calculate statistical differences in basal activity (b). The level of cell surface expression was 90% for E673R/WT and 76% for WT/H579E mutants in comparison with the WT/WT heterodimer.

Positioned at 30° from the extracellular dimer interface, the transmembrane dimer interactions bury approximately 740 Å² of surface area and exhibit high shape complementarity (Extended Data Fig. 5f). The pair of TM5 helices scissor at their central residues before contacting the transverse TM3 of opposing subunits at their intracellular ends (Fig. 2a). TM5 extracellular ends display conformational variance wherein they approach and withdraw but fail to make contact (Supplementary Video 4). All direct heterodimer interactions occur near the cytoplasmic membrane surface and can be divided into three core layers (I–III) along the helical path of TM5 (Fig. 2a).

The surface layer I caps the extracellular end of the transmembrane heterodimer interface. It is composed of hydrophobic contacts between four leucine residues of both subunits. The middle layer II lies directly beneath layer I, and consists of three phenylalanine residues packing against one another, as well as their neighbouring leucine residues. Both layers solely incorporate TM5 residues.

Layer III, consisting of sections IIIa and IIIb, completes the interface at the intracellular end. IIIa possesses a network of salt bridges that tether the cytoplasmic ends of GABA_B1b and GABA_B2 transmembrane domains. This critical interaction is mediated by a quartet of charged residues from the TM3 and TM5 helices (His 572^3.55 and Glu 673^5.60 in GABA_B1b; His 579^3.55 and Glu 677^5.60 in GABA_B2; Supplementary Table 2), a feature that we refer to as the ‘intersubunit latch’ for securing the transmembrane orientation of the two subunits in the inactive conformation. One of the ‘intersubunit latch’ residues shares nonpolar contacts with a lysine of GABA_B2 TM5, establishing the accessory site IIIb.

In addition, the central layers of direct heterodimer contacts are flanked on each side by a cholesterol molecule. One mediates the interaction between TM5 helices of both subunits (CLR3), while the other bridges TM3 of GABA_B1b and TM5 of GABA_B2 (CLR6) (Extended Data Fig. 5g).

The ‘intersubunit latch’

To determine the importance of the ‘intersubunit latch’ in controlling the inactive state of the GABA_B receptor, we examined the effect of single charge-repelling mutations (GABA_B1b E673R and GABA_B2 H579E) within the motif. Wild-type GABA_B receptor exhibited basal activity in the absence of agonist (Fig. 2b), as reported previously¹⁴. Both mutants substantially increased basal activity when compared to the wild type, suggesting that each mutation promotes agonist-independent constitutive activity (Fig. 2b). The basal activity displayed by these mutants reached approximately 80–90% of the maximal agonist-dependent wild-type response (Fig. 2b, c). Application of the agonist baclofen raised receptor activity further to the highest wild-type level (Fig. 2c). Neither mutation altered the agonist potency. After each construct was treated with the inverse agonist CGP54626, the basal activity of mutants still remained higher than that of the wild type, providing further evidence that the ‘intersubunit latch’ mutations serve to shift the conformational equilibrium of GABA_B towards an active state. Taken together, our mutational data indicate that the ‘intersubunit latch’ is fundamental to maintaining the inactive state of the receptor.

Endogenous ligands bound to GABA_B1b VFT

We identified a new potential Ca²⁺-binding site in the vicinity of the orthosteric ligand-binding cleft on the LB2 surface of GABA_B1b (Fig. 3a and Extended Data Fig. 6a, b). The metal ion density has remarkable peak height in the cryo-EM density map (10σ) and is surrounded by residues that are chemically favourable for Ca²⁺ coordination. The Ca²⁺ is anchored by the carboxylate groups of three acidic residues (Asp 281, Glu 309 and Glu 423) and by the backbone carbonyl atoms of two additional residues (Gly 277 and Tyr 279) (Fig. 3b). The Ca²⁺–O bond distances are between 3.0 Å and 4.6 Å, suggesting that the ion is in a hydrated state. The Ca²⁺ location in GABA_B1b is different from any of the four sites found in the calcium-sensing (CaS) receptor²¹, another class C GPCR (Fig. 3c).

Fig. 3: Ca²⁺ binding in GABA_B1b.

a, Ribbon representation of the GABA_B1b subunit, showing the location of the Ca²⁺-binding site at the interdomain crevice of VFT. b, Specific interactions between GABA_B1b and Ca²⁺. The mesh represents the cryo-EM density map contoured at 7.5σ surrounding Ca²⁺. c, Crystal structure of the CaS receptor extracellular domain (Protein DataBank (https://www.rcsb.org) code 5K5S), highlighting its bound Ca²⁺ (black spheres, numbered 1–4 and 1′–4′ in the two protomers) and the corresponding location of the Ca²⁺-binding site in GABA_B1b (green sphere).

Using inductively coupled plasma mass spectroscopy, we detected the presence of Ca and Cu above background level in purified GABA_B receptor. The amounts of other metal elements were negligible (Supplementary Table 3). The molar ratios of Ca and Cu relative to the receptor protein (0.43:1 and 0.51:1) suggest partial occupancy of the ion-binding sites. An unmodelled density within the interdomain cleft, coordinated by tryptophan and histidine, may serve as a potential Cu²⁺-binding site but lacks sufficient signal to be labelled confidently.

In exploring the functional role of the bound ion, we found that the Ca²⁺ chelator EGTA substantially reduced GABA_B basal activity, and that mutating specific coordinating residues (GABA_B1b E309K and GABA_B1b E423R) had similar, although less drastic, effects (Extended Data Fig. 6c, d). Consistent with previous findings, these data suggest that Ca²⁺ may act as a positive allosteric modulator of the GABA_B receptor^22,23. Our structure implies that Ca²⁺ stabilizes residues adjacent to the critical agonist-binding residue Trp 278, thereby indirectly reinforcing its conformation (Fig. 3b).

We also found density in the orthosteric agonist-binding site of GABA_B1b, and its shape suggests a GABA-like endogenous ligand (Extended Data Fig. 6a, b). GABA is a potential candidate because it fits the density and was detected in the lysate of cells used to express the receptor (Extended Data Fig. 6e, f). This GABA-like endogenous ligand bound in an inactive receptor conformation may reflect a preactivation state; however, further investigation is required.

Discovery of endogenous phospholipid ligands

Our structure revealed endogenous phospholipids within the transmembrane domains of both GABA_B subunits. Using the CaS receptor as a control in mass spectrometry, we identified two phospholipids specifically bound to the GABA_B receptor, PE 38:5 and PC 38:2 (Fig. 4a–d). Both lipids consist of two long-chain fatty acyl moieties of 18 and 20 carbons. We further assigned PE 38:5 to GABA_B1b and PC 38:2 to GABA_B2 on the basis of the size difference between the phosphoethanolamine and phosphocholine head groups of the two lipids (Extended Data Fig. 7a, b). The lipid density in the GABA_B2 transmembrane domain has a bulkier head group that can better accommodate the larger choline moiety of PC 38:2.

Fig. 4: Identification of endogenous phospholipid ligands of the GABA_B receptor.

a, c, Liquid chromatography with tandem mass spectrometry (LC-MS/MS) analysis of phospholipids bound to the GABA_B receptor (n = 1). The LC traces show the abundance of phospholipids PC 38:2 (a) and PE 38:5 (c) in GABA_B preparation relative to a CaS receptor control. b, d, High-resolution MS spectra of the peaks in a, c (black) matched with standard spectra of PC 38:2 (b) and PE 38:5 (d) in red. NL, neutral loss: a fragment in a neutral charge state during collision-induced dissociation; SN1 and SN2, stereospecific numbering of glycerophospholipids, marking the fatty acyl groups at respectively the first and second carbons of the glycerol. e, f, Molecular surface of the GABA_B1b (e) and GABA_B2 (f) transmembrane domains, showing the binding pockets for PE 38:5 and PC 38:2, respectively. g–j, Three views of the specific interactions between phospholipids and each GABA_B subunit. Contacts are shown between the phospholipid head group and residues from GABA_B1b (g) and GABA_B2 (h), and between phospholipid fatty acyl chains and GABA_B1b (i) and GABA_B2 (j).

Mirroring the amphipathicity of phospholipids, the lipid-binding pocket of each GABA_B subunit retains a hydrophilic trunk and two lipophilic branches for binding the lipid polar head and nonpolar tails, respectively. The trunk consists of a negatively charged patch covering the amine moiety, and a positively charged area surrounding the phosphate (Extended Data Fig. 7c, d). The lipid-binding pockets are notably deep, extending from the extracellular membrane surface to the centre of the transmembrane domain (Fig. 4e, f). Each lipid occupies nearly the entire range of ligand-binding positions in class A, B, C and F GPCRs (Extended Data Fig. 7e–h).

Each GABA_B subunit makes extensive contacts with the bound lipid, using a majority of the transmembrane helices, including TM2, 3, 5, 6 and 7 (Extended Data Fig. 7a, b, i, j). Approximately 2,400 Å² of surface area is buried by either lipid–subunit pair. In addition, ECL2 directly contacts PE 38:5 in GABA_B1b, while the linker and all three ECLs interact with PC 38:2 in GABA_B2.

Key elements of the lipid–receptor interactions are conserved in GABA_B1b and GABA_B2 (Fig. 4g–j, Extended Data Fig. 7i, j and Supplementary Fig. 3). First, the hydrophilic head of each lipid is anchored through interactions with conserved histidine and arginine residues (His 643 of ECL2 and Arg 549^3.32 in GABA_B1b; His 647 of ECL2 and Arg 556^3.32 in GABA_B2) (Fig. 4g, h). GABA_B2 also incorporates Arg 714 of ECL3, rendering the lipid-binding pocket more electropositive than that of GABA_B1b (Fig. 4h).

Second, the 20-carbon fatty acyl chain of both lipids follows a perpendicular turn to pass between two aromatic residues (Phe 557^3.40 and Tyr 657^5.44 in GABA_B1b; Tyr 564^3.40 and Tyr 661^5.44 in GABA_B2) (Fig. 4i, j). A cis double bond in each fatty acyl chain forms π interactions with the aromatic side chains. The bend is further buttressed by extensive nonpolar contacts with the aliphatic part of a conserved lysine (GABA_B1b Lys 660^5.47; GABA_B2 Lys 664^5.47) lying parallel to the chain. Finally, the 18-carbon fatty acyl chain of both lipids is relatively straight, extending towards the cytoplasm in a binding pocket lined by small aliphatic residues on TM2, TM3 and TM7 (Fig. 4i, j).

The GABA_B receptor’s transmembrane domains are covered by the linker and ECLs, which form a lid over the lipid-binding pocket. Phospholipids may access the lipid-binding pocket of designated subunits laterally through gaps between TM5 and TM6 (Extended Data Fig. 8a–d). One of the fatty acyl tails of each phospholipid even protrudes through this opening. The lipid-binding sphingosine-1-phosphate (S1P₁) receptor possesses a similar gap between TM1 and TM7 (ref. ²⁴; Extended Data Fig. 8e, f). The endogenous lipids of other GPCRs can be readily replaced²⁴, but the size and engagement of the endogenous lipids bound to the GABA_B receptor suggest that they are essential for maintaining receptor integrity and stability.

To explore the physiological relevance of the endogenous phospholipids, we mutated residues that hydrogen bonded with the phosphate head group. We identified R714A in GABA_B2 ECL3, which displayed a small gain of function despite reduced cell surface expression (Extended Data Fig. 8g, h). This mutation is expected to enhance the movement of PC 38:2 within GABA_B2 by eliminating a critical interaction with the lid. Our results suggest that PC 38:2 may act as a negative allosteric modulator of GABA_B receptor by stabilizing the inactive conformation of GABA_B2.

Comparison of subunits with other GPCRs

The GABA_B1b and GABA_B2 subunits have highly similar VFT and transmembrane components but differ in their relative orientation (Extended Data Fig. 9a, b). Each subunit differs from all other class C GPCRs in possessing an extended peptide linker between the VFT and transmembrane domains instead of a cysteine-rich domain (Fig. 1c).

GABA_B subunits also display distinct helix positions in the seven-helix bundles among inactive GPCRs (Extended Data Fig. 9c–f). However, these differences are minor compared with the dramatic movement of TM6 in class A and B GPCRs when activated by G-protein coupling^25,26. This corroborates our conclusion that we have captured an inactive conformation of the GABA_B receptor.

The GABA_B receptor presents unique variations of conserved transmembrane motifs (Extended Data Fig. 10a, b). In most class A GPCRs such as rhodopsin²⁷, the ‘ionic lock’ tethers TM3 and TM6 to stabilize the inactive state of an individual transmembrane domain²⁷ (Extended Data Fig. 10c). The ‘ionic locks’ of both GABA_B subunits consist of an aspartate from ICL3 and a lysine from TM3 (Lys^3.50) (Extended Data Fig. 10a, b). Although only the Lys/Asp pair in GABA_B1b are within hydrogen-bond distance, their backbone Cα–Cα separations (9.3–9.4 Å) are comparable to that of rhodopsin²⁷ (8.7 Å) and inactive mGlu receptors^28,29 (10.9–11.2 Å), indicating that the ‘ionic lock’ is intact within both GABA_B subunits (Extended Data Fig. 10a–e).

The ‘ionic lock’ in each GABA_B subunit resides in close proximity to an FxPKxx sequence in TM7, which is the counterpart of the NPxxY(x)_5,6F motif in class A GPCRs²⁵. Specifically, the conserved Lys^7.51 participates in a network of hydrophilic contacts with the ‘ionic lock’ through Asn^2.39 and a serine in ICL1 (Extended Data Fig. 10a, b and Supplementary Fig. 3). These interactions unite the ‘ionic lock’ and FxPKxx motif into a larger and integral system for maintaining the inactive transmembrane conformation of individual GABA_B subunits.

Conclusion

The combination of our previous VFT structures and present cryo-EM data supports the occurrence of three essential events during activation of the GABA_B receptor: first, agonist-induced closure of the GABA_B1b VFT; second, association of membrane-proximal LB2 domains; and third, dissociation of the ‘intersubunit latch’ and ensuing rearrangement of the transmembrane heterodimer interface. This hypothesis is consistent with our finding that an inverse agonist bound to the extracellular domain can inhibit the constitutive activity stemming from the spontaneous closure of the GABA_B1b VFT, but not that resulting from direct ‘downstream’ disruption of the ‘intersubunit latch’.

The GABA_B-receptor structure also yields surprising findings regarding its endogenous ligand composition. We suspect that the phospholipids discovered well inside each transmembrane cavity are necessary structural components, as they are prebound within each subunit and their interactions with the receptor are extensive. These endogenous lipids may be unique to the GABA_B receptor, as the lipid-binding residues are not conserved among class C GPCRs (Supplementary Fig. 3). Preoccupation of the transmembrane pocket suggests that GABA_B allosteric modulators may bind to yet unknown sites, with the heterodimer interface being a potential location. An active structure of the GABA_B receptor would confirm whether the phospholipids are integral receptor components or functional modifiers.

Methods

No statistical methods were used to predetermine sample size. The experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment.

Protein expression and purification

Human GABA_B1b (UniProt (https://www.uniprot.org) code Q9UBS5-2) and GABA_B2 (UniProt code O75899) subunits were each cloned into a modified pEG BacMam vector³⁰ for coexpression in baculovirus-infected mammalian cells. GABA_B1a and GABA_B1b are two major isoforms of GABA_B1, and have identical pharmacological profiles⁷. Different C-terminal truncations of each receptor subunit were tested for heterodimeric receptor assembly. The optimal GABA_B1b expression construct consisted of residues 1–802 (GABA_B1b(1–802)), while the GABA_B2 construct included residues 1–819 (GABA_B2(1–819)). This allowed the heterodimeric construct to be transported to the cell membrane, as it retained the intracellular coiled-coil region present in the intracellular tail of each subunit^31,32. Signal peptides for GABA_B1b and GABA_B2 occupied residues 1–29 and 1–41, respectively. A Flag tag was engineered at the C terminus of each subunit to facilitate affinity purification.

Human embryonic kidney (HEK) 293 GnTI^– cells³³ were grown in suspension culture at 37 °C in 8% CO₂ using 293 freestyle media (Life Technology, Carlsbad, CA). The cells were coinfected with recombinant baculoviruses carrying the GABA_B1b(1–802) and GABA_B2(1–819) genes at 37 °C. To enhance the expression level, we added 10 mM sodium butyrate 18 h post infection, and incubated the cells for a further 72 h at 30 °C before harvest.

Cell membrane was isolated by differential centrifugation. The cells were lysed using an EmulsiFlex-C3 high pressure homogenizer (Avestin, Ottawa, Canada) in a buffer containing 50 mM HEPES, pH 7.5, 150 mM NaCl, 10% glycerol and a cocktail of protease inhibitors (Roche, Basel, Switzerland). Cell debris was removed by centrifugation of the lysed cell suspension at 10,000 r.p.m. The cell membrane was then pelleted by ultracentrifugation at 45,000 r.p.m.

The GABA_B receptor was extracted from the cell membrane with 50 mM HEPES, pH 7.5, 150 mM NaCl, 10% glycerol, 1% lauryl maltose neopentyl glycol (LMNG) (Anatrace, Maumee, OH) and 0.2% cholesteryl hemisuccinate (CHS) (MilliporeSigma, St. Louis, MO) at 4 °C overnight. After the insoluble matter was removed by centrifugation, the supernatant was applied to an anti-Flag M2 antibody affinity column (MilliporeSigma, St. Louis, MO). The column was washed stepwise with decreasing concentrations of detergent, from 0.1% to 0.002% LMNG. The heterodimeric GABA_B1b(1–802)–GABA_B2(1–819) complex was then eluted with 50 mM HEPES, pH 7.5, 50 mM NaCl, 0.002% LMNG, 0.0004% CHS and 0.2 mg ml⁻¹ Flag peptide.

The receptor was further purified by Mono Q (GE Healthcare, Chicago, IL) ion-exchange chromatography using a linear salt gradient from 50 mM to 1 M NaCl in 50 mM HEPES, pH 7.5, 0.002% LMNG and 0.0004% CHS. Finally, the assembled GABA_B receptor was subjected to Superose 6 (GE Healthcare, Chicago, IL) size-exclusion chromatography in 50 mM HEPES, pH 7.5, 50 mM NaCl, 0.002% LMNG and 0.0004% CHS.

HEK 293 GnTI⁻ cells were purchased from and authenticated by the American Type Culture Collection (ATCC catalogue number CRL-3022). Cell morphology was examined for each passage of cells. The cells were certified by ATCC to be free of mycoplasma contamination, but they were not tested again during culturing.

Cryo-EM specimen preparation and data acquisition

Specimens were composed of vitrified GABA_B protein samples occupying UltraAuFoil R 0.6/1, 300-mesh holey Au/Au grids (Quantifoil Micro Tools, Jena, Germany). The surfaces of the grids were rendered hydrophilic by glow-discharging using H₂ and O₂ for 25 s at 10 W with a Solarus 950 plasma cleaner system (Gatan, Pleasanton, CA). For vitrification, 3 μl of purified GABA_B receptor at a concentration of approximately 0.3 mg ml⁻¹ was applied to each grid, blotted for 4 s at a blot force of 3 inside a Vitrobot Mark IV (Thermo Fisher Scientific, Waltham, MA), and plunge-frozen in a liquid propane:ethane mixture (63:37, v/v) cooled with liquid nitrogen.

Data collection was performed on a Titan Krios transmission electron microscope (Thermo Fisher Scientific, Waltham, MA) equipped with a K2 Summit direct electron detection camera (Gatan, Pleasanton, CA) in counting mode and a post-column GIF Quantum energy filter (Gatan) in zero-energy-loss mode with a slit width of 20 eV. Micrographs were accrued at a calibrated pixel size of 1.06 Å and with nominal defocus range of −0.5 to −2 μm. Each micrograph consisted of 60 frames collected over a 12-s exposure at a dose rate of roughly 8 electrons per pixel per second for a total dose of roughly 85 electrons per Ų. A total of 3,435 micrographs were acquired as dose-fractionated image stacks.

Cryo-EM image processing

Image processing began with frame alignment and dose-weighting of the image stacks using the CPU-based implementation of MotionCor2 (ref. ³⁴) in Relion 3.0 (ref. ³⁵). Estimation of contrast transfer function (CTF) for each non-dose-weighted micrograph was determined by Gctf³⁶ version 1.06. After visual inspection of the micrographs and their power spectra, 3,334 were selected for further processing.

Approximately 3,000 particles were manually picked in Relion 3.0 (ref. ³⁵), and extracted from a 4× binned data set with a pixel size of 4.24 Å. These data produced an initial set of two-dimensional (2D) classes that were used as templates to select 1,048,241 particles automatically, all of which were subsequently imported into cryoSPARC version 2 (ref. ³⁷) for extensive 2D classification. After elimination of unfit classes, a total of 312,840 particles from the high-quality 2D classes were combined to produce an ab initio three-dimensional (3D) reference in cryoSPARC version 2 (ref. ³⁷). On the basis of the ab initio model, particles were re-extracted at full scale from the unbinned data set with a pixel size of 1.06 Å in Relion 3.0 (ref. ³⁵), and re-introduced into cryoSPARC version 2 (ref. ³⁷) for 3D refinement. Homogeneous refinement of the ab initio 3D model against the unbinned set of particles yielded a density map with nominal resolution of 3.6 Å according to the Fourier shell correlation (FSC) = 0.143 gold standard criterion³⁸. Heterogenous refinement of multiple models obtained before and after homogeneous refinement allowed removal of additional poor-quality particles and reduced the particle count to 233,737. Nonuniform refinement then improved the resolution to 3.5 Å. At this point, CTF refinement and Bayesian polishing were conducted in Relion 3.0 (ref. ³⁵), followed by an additional round of nonuniform refinement in cryoSPARC version 2 (ref. ³⁷), further improving the resolution to 3.3 Å.

Although the transmembrane domains of the GABA_B receptor exhibit pseudo twofold symmetry, bulky carbohydrate densities that are visible only in the ECD of the GABA_B1b subunit resulted in sufficient low-resolution asymmetry to prevent particle misalignment. These densities include partial carbohydrate densities attached to Asn 323 and Asn 365 of GABA_B1b that do not have counterparts in GABA_B2.

The global density map exhibited directional anisotropy³⁹ that is caused by interdomain movement about a flexible linker. To eliminate the adverse effect of such movement on map quality, we performed local refinement of the ECD and transmembrane domains of the GABA_B receptor independently. A mask was created covering each region, and the nonuniform refinement algorithm was used as implemented in cryoSPARC version 2 (ref. ³⁷). The resulting reconstructions for the individual ECD and transmembrane domains reached 3.1 Å and 3.4 Å resolution, respectively. A composite map was generated in UCSF Chimera⁴⁰ by taking the maximum values pointwise from the two locally refined maps after alignment to the global reconstruction (‘vop maximum’ command in UCSF Chimera⁴⁰). This composite map was used for subsequent model building and refinement.

Resolutions of cryo-EM reconstructions were determined using a cutoff value of 0.143 in gold standard half-map FSC curves³⁸.

We carried out 3D variability analysis⁴¹ in cryoSPARC version 2 (ref. ³⁷) using the 233,737 particles from global nonuniform refinement as input. Calculations were performed for the entire receptor, the ECDs and the transmembrane domains, respectively. In each case, multiple modes of variability were solved, and represented as eigenvectors along which conformational changes occur. To visualize the transformation of density, we calculated five reconstructions along each eigenvector, with a filter resolution of 4.5 Å. A movie that combines these reconstructions as frames was generated in Chimera⁴⁰ for each dimension of motion.

Model building and refinement

Model building was carried out in COOT⁴². The crystal structure of a complex of human GABA_B1b VFT and GABA_B2 VFT in the apo form (Protein DataBank code 4MQE) was used as the initial model for the extracellular domain of the receptor. The VFT modules of GABA_B1b and GABA_B2 were separately placed into density as rigid bodies. Individual residues were then adjusted to optimize the fit. The linker and transmembrane domain of each subunit were traced de novo based on the density. The final model contained residues 48–368, 377–576 and 588–747 of GABA_B1b, and 54–294, 302–376, 385–584 and 595–749 of GABA_B2.

In addition to the polypeptide chains, we built models for a Ca²⁺ ion in the extracellular domain of GABA_B1b, and for one phospholipid (PE 38:5 in GABA_B1b and PC 38:2 in GABA_B2) within the transmembrane domain of each subunit. Density for carbohydrate was observed at three N-linked glycosylation sites on GABA_B1b (Asn 365, Asn 385 and Asn 397) and one site on GABA_B2 (Asn 404). An N-glucosamine residue was modelled at each of these glycosylation sites. Continuous density was also identified for ten cholesterol or CHS molecules surrounding the transmembrane domains of both GABA_B subunits. Cholesterols were modelled to optimize the fit with density; however, these densities may belong to CHS molecules with disordered parts. Density for an endogenous ligand was found at the interdomain cleft of the GABA_B1b VFT. Although this density could be fit with GABA, it was not modelled because the origin and identity of the ligand remained ambiguous without confirmation by an independent method.

The entire structure was refined by the real-space refinement algorithm and validated with the comprehensive validation application as implemented in Phenix⁴³. Ramachandran statistics were calculated using MolProbity⁴⁴. The refined model also has an overall EMRinger⁴⁵ score of 2.7, while the extracellular and transmembrane domains have scores of 3.4 and 1.8, respectively. The final model revealed that VFT and transmembrane components are related to their counterparts in the other subunit by 177° and 179° rotations about the vertical axis, respectively.

Pairwise structural alignment was performed using LSQMAN⁴⁶. Figures were generated using Pymol Molecular Graphics System version 2.3 (Schrödinger), UCSF Chimera⁴⁰ and UCSF ChimeraX⁴⁷. Software installation support was provided by SBGrid⁴⁸.

Scintillation proximity assay

Binding of [³H]GABA (60 Ci mmol⁻¹; American Radiolabelled Chemicals, St. Louis, MO) to the GABA_B receptor was measured with a scintillation proximity assay (SPA)^49,50. Purified GABA_B1b(1–802)–GABA_B2(1–819) complex (100 ng) was extensively dialysed against the purification buffer to remove any residual endogenous ligand. The dialysed protein was then immobilized to yttrium silicate (YSi) protein A SPA beads (62.5 μg; PerkinElmer, Waltham, MA) using anti-Flag M2 antibody (12.5 pg; Millipore Sigma, St. Louis, MO), and incubated at 4 °C for 30 min in the same buffer as used for the final step of protein purification (50 mM Hepes, pH 7.5, 50 mM NaCl, 0.002% LMNG, 0.004% CHS). Increasing concentrations (ranging from 0.2 μM to 25 μM) of [³H]GABA (2.5 Ci mmol⁻¹ final specific radioactivity) were added to the protein/antibody/SPA-bead mixture, and the samples were allowed to reach equilibrium at 4 °C for 16 h. A reaction performed with an antibody/SPA-bead mixture in the absence of GABA_B receptor was used to determine the nonproximity signal originating from nonspecific interaction between the radioligand and SPA beads.

All samples were counted in a Microbeta counter (PerkinElmer, Waltham, MA) in counts per minute (c.p.m.) in the SPA mode. The efficiency of detection was calculated with a standard curve of known [³H]GABA concentrations, and this was used to transform c.p.m. into pmol. Specific binding was determined by subtracting the nonproximity signal (nonspecific binding) from the total binding signal and was plotted as a function of free radioligand concentration. Nonlinear regression fitting of the data was performed in SigmaPlot 13.0 to obtain the dissociation constant (K_d) and the molar ratio of GABA-to-receptor binding.

Identification of phospholipid ligands and GABA

Identification of bound endogenous lipids was conducted as described⁵¹ with modifications. Briefly, intact GABA_B and control CaS receptor were digested with trypsin overnight at 37 °C. Digested proteins were dried and extracted with 1 ml of ice-cold methanol:water (9:1, v/v). The supernatants were dried and resuspended with methanol:toluene (9:1, v/v) to an equivalent concentration of 2 μM. For LC-MS/MS analysis⁵², the lipid extracts were separated on a Waters Acquity UPLC CSH C18 column (100 × 2.1 mm; 1.7 μm) (Waters, Milford, MA) coupled to an Acquity UPLC CSH C18 VanGuard precolumn (5 × 2.1 mm; 1.7 μm) (Waters). The column was maintained at 65 °C at a flow rate of 0.6 ml min⁻¹. The mobile phases consisted of A, acetonitrile:water (60:40, v/v) with ammonium formate (10 mM) and formic acid (0.1%), and B, 2-propanol:acetonitrile (90:10, v/v) with ammonium formate (10 mM) and formic acid (0.1%). The 15-min separation was conducted under the following gradient: 0 min 15% B; 0−2 min 30% B; 2−2.5 min 48% B; 2.5−11 min 82% B; 11−12 min 99% B; 12−15 min 15% B. A Q Exactive HF mass spectrometer (Thermo Fisher Scientific, Waltham, MA) was operated in electrospray ionization (ESI) in positive mode with the following parameters: mass range 100−1,500 m/z; spray voltage +3.6 kV; sheath gas (nitrogen) flow rate 60 units; auxiliary gas (nitrogen) flow rate 25 units; capillary temperature 320 °C; full scan MS1 mass resolving power 60,000; data-dependent MS/MS acquisition (dd-MS/MS) four scans per cycle; dd-MS/MS mass resolving power 15,000. The mass features that were differentially higher in the GABA_B receptor were subjected to targeted MS/MS in reinjections to acquire tandem mass spectra. Thermo Xcalibur version 4.0.27.19 was used for data acquisition. Data processing and identification were performed in MS-DIAL version 3.40. Identification was conducted by matching accurate mass, tandem mass spectra and chromatographic retention time with built-in lipid library LipidBlast⁵³. The identified endogenous lipids bound to GABA_B receptor include phosphatidylcholine (PC) 38:2 (PC(18:1_20:1); International Chemical Identifier (InChI) key: QLEJPADMSQQACL-WWUFLCHTSA-N), and phosphatidylethanolamine (PE) 38:5 (PE(18:1_20:4); InChIKey VFUVYNGTMNUBMF-ZRVIQYDLSA-N). Although many isoforms exist for these lipids, both phospholipids share two long-chain fatty acyl moieties of 18 and 20 carbons based on mass spectrometry fragmentation pattern and biological relevance.

Identification of GABA was conducted with targeted LC-MS/MS. Briefly, cell supernatant, cell lysate, culture media and lysis buffer controls were dried from 1 ml and extracted with 1 ml of ice-cold methanol:water (9:1, v/v). The supernatants were dried and resuspended with 200 μl of acetonitrile:water (8:2, v/v). The extracts were separated on a Waters Acquity UPLC BEH Amide column (150 × 2.1 mm; 1.7 μm) (Waters, Milford, MA) coupled with an Acquity UPLC BEH Amide VanGuard precolumn (5 × 2.1 mm; 1.7 μm) (Waters). The column was maintained at 45 °C at a flow rate of 0.4 ml min⁻¹. Mobile phase A was water with ammonium formate (10 mM) and formic acid (0.125%) while B was acetonitrile:water (95:5, v/v) with ammonium formate (10 mM) and formic acid (0.125%). Separation was conducted using the gradient: 0−2 min 100% B; 2−7.7 min 70% B; 7.7−9.5 min 40% B; 9.5−12.5 min 30% B; 10.25−12.75 100% B, 12.75−16.75 100% B. A Q Exactive HF mass spectrometer was operated with the same parameters as above. A GABA standard was injected along with the samples to confirm its spectrum and retention time. The responses of GABA in samples were normalized to that of GABA standard with known concentration.

Cell surface expression

Full-length human GABA_B1b and GABA_B2 were each cloned into a pcDNA3.1(+) vector (Life Technologies, Carlsbad, CA), with a Flag tag inserted after the signal peptide of GABA_B1b, and a haemagglutinin (HA) tag after the signal peptide of GABA_B2. Similar constructs were also generated for the C-terminally truncated GABA_B1b(1–802) and GABA_B2(1–819). Single mutants of full-length GABA_B1b (E673R, E309K and E423R) and GABA_B2 (H579E and R714A) were constructed using the QuikChange mutagenesis system (Agilent Technologies, Santa Clara, CA).

The cell surface expression levels of wild-type and mutant GABA_B receptor were measured as described¹⁷. Briefly, HEK293 T/17 cells (ATCC) were cultured in monolayer in DMEM/F12 media (Life Technology, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS) at 37 °C in the presence of 5% CO₂. The cells were co-transfected with GABA_B1b and GABA_B2 plasmids using Lipofectamine 3000 (Life Technologies, Carlsbad, CA). Each GABA_B1b and GABA_B2 mutant was paired with its wild-type partner. Given that GABA_B1b is retained inside the cells unless it is chaperoned by GABA_B2, we used the surface expression level of GABA_B1b on intact cells to measure the amount of assembled heterodimeric GABA_B receptor on the cell surface. The amount of surface GABA_B1b protein detected for each pair of constructs was normalized with the cell count in each experiment. The cell surface expression level of each mutant is calculated as a percentage of the wild-type receptor.

After blocking with 1% BSA, the cells were incubated with mouse anti-Flag M1 antibody (MilliporeSigma, St. Louis, MO) as the primary antibody to measure GABA_B1b expression, followed by donkey anti-mouse IRDye 800-labelled antibody (Li-Cor Biosciences, Lincoln, NE) as the secondary antibody. Fluorescent signals were measured with an Odyssey Infrared Imager (Li-Cor Biosciences). Each experiment was performed in triplicate.

HEK 293 T/17 cells were purchased from and authenticated by ATCC (catalogue number CRL-11268). Cell morphology was examined for each passage of cells. The cells were certified by ATCC to be free of mycoplasma contamination, but they were not tested again during culturing.

Inositol phosphate measurement

HEK293 T/17 cells were cotransfected with plasmids encoding GABA_B1b, GABA_B2, and Gα_qi5. The Gα_qi5 chimaera was constructed by replacing the five C-terminal amino acids of murine Gα_q with those of murine Gα_i (ref. ¹⁸). Exchanging the C-terminal end of Gα_q with that of Gα_i/o permits it to couple with the GABA_B receptor and allows the functional activity of the receptor to be tracked through the activity of phospholipase C (PLC). Control experiments were conducted using cells transfected with an empty pcDNA3.1(+) vector, Gα_qi5 alone, or wild-type GABA_B1b and GABA_B2 in the absence of Gα_qi5.

Inositol phosphate accumulation was quantified with the homogenous time-resolved fluorescence (HTRF) IP-one Tb kit (Cisbio Bioassays, Codolet, France), which measures the accumulation of inositol-1-monophosphate (IP₁), a metabolite of inositol-1,4,5-triphosphate (IP₃). One day after transfection, the cells were stimulated with increasing concentrations of baclofen for one hour at 37 °C. The stimulated cells were lysed, and the native IP₁ that had been produced was incubated with a d2 fluorophore-labelled IP₁ analogue (the acceptor) to compete for binding to an Eu Cryptate-coupled anti-IP1 monoclonal antibody (the donor). Fluorescence data were collected at 620 nm and 665 nm with a PHERAstar FS plate reader (BMG Labtech, Cary, NC) after laser excitation at 320 nm. The fluorescence resonance energy transfer (FRET) signal was calculated as the fluorescence ratio (665 nm/620 nm) and is inversely proportional to the concentration of native IP₁ produced following GABA_B activation through a chimaeric Gα_qi5 G protein. The agonist-induced receptor response of each mutant was calculated as a percentage of the maximum activity of wild-type receptor relative to the activity observed for Gα_qi5 alone. Basal activity was determined in the absence of baclofen stimulation and calculated similarly to the agonist-dependent receptor response. Data analysis was performed using the nonlinear regression algorithms in Prism (GraphPad Software, San Diego, CA). Data points represent averages ± s.e.m. of multiple experiments, each consisting of quadruplicate measurements.

Application of a known antagonist, CGP54626, reduced agonist potency as expected. In addition, the compound lowered the basal activity of the GABA_B receptor, indicating that it has inverse agonist activity, as previously reported⁵⁴. Therefore, we refer to the compound as an inverse agonist.

Inductively coupled plasma mass spectrometry

Purified GABA_B receptor (200 μl, 11.1 μg μl⁻¹ or 57.7 μM) was collected in metal-free tubes and digested overnight by adding 1 ml of concentrated nitric acid (HNO₃) (Fisher Scientific, Hampton, NH; Optima grade). The digested protein was then diluted to a total volume of 10 ml with 8.7 ml of deionized water supplemented with 500 μg l⁻¹ of gold (Au) and 100 μl of an internal standard solution containing 500 μg l⁻¹ each of gallium (Ga) and rhodium (Rh) in 2% HNO₃. The protein purification buffer (200 μl), containing 10 mM HEPES, pH 7.5, 50 mM NaCl, 0.002% LMNG and 0.0004% CHS, was similarly mock digested and diluted for analysis.

Inductively coupled plasma mass spectrometry (ICP-MS) was conducted using a NexION 350S ICP-MS instrument (Perkin Elmer, Waltham, MA) equipped with dynamic reaction cell (DRC) feature and a SC-4 DX FAST Autosampler (Elemental Scientific, Omaha, NE). The DRC-ICP-MS experimental method was developed from published procedures^55,56 and a laboratory protocol for multi-element DCR-ICP-MS from the Centers for Disease Control (CDC) (https://www.cdc.gov/nchs/data/nhanes/nhanes_13_14/UM_UMS_UTAS_UTASS_H_MET.pdf). The concentrations of magnesium (Mg), calcium (Ca), manganese (Mn), iron (Fe), cobalt (Co), nickel (Ni), copper (Cu), zinc (Zn) and strontium (Sr) in the digested protein and buffer samples were measured. Data points represent averages ± coefficient of variance from eight measurements within two experiments.

One multi-element calibration standard was prepared from concentrated single-element stocks, and used for instrument calibration. The calibration standard was diluted to various concentrations using a solution containing 10% HNO₃ and 500 μg l⁻¹ of Au to cover the expected concentration range of each analyte in the protein sample: 0.01, 0.02, 0.05, 0.1, 0.2 μg l⁻¹ for Co and Sr; 0.05, 0.1, 0.25, 0.5, 1.0 μg l⁻¹ for Mn and Ni; 0.25, 0.5, 1.25, 2.5, 5.0 μg l⁻¹ for Mg and Cu; 0.5, 1.0, 2.5, 5.0, 10.0 μg l⁻¹ for Fe and Zn; and 2.5, 5.0, 12.5, 25, 50 μg l⁻¹ for Ca.

The instrument was also calibrated against a set of blank solutions, including commercially available quality controls containing digested hair samples from Public Health Expertise and Reference Center, Quebec (INSPQ, Quebec, Canada), and a water sample containing a broad range of metals from the National Institute of Technology (NIST, Gaithersburg, MD).

Special attention was paid to correction for matrix-induced interferences. Matrix suppression was compensated by the selection of suitable internal standards, which were matched to masses and, if possible, to ionization properties of the analytes. The internal standards were added to each calibration standard and quality control sample to the same final concentration as that in the protein sample and buffer (5 μg l⁻¹ each of Ga and Rh). The elements Mg, Ca, Mn, Fe, Co, Ni, Cu, and Zn were corrected with Ga, while Sr was corrected with Rh. Polyatomic interferences were suppressed with the instrument’s DRC technology feature, using ammonia as a second gas for Mn and Fe, while Mg, Ca, Sr, Co, Ni, Cu, and Zn were measured in standard mode without a second gas.

Matrix-assisted laser desorption/ionization

Purified GABA_B1b(1–802)–GABA_B2(1–819) complex (0.3 mg ml⁻¹, 1 μl) was mixed with 1 μl sinapinic acid matrix solution (Bruker Daltonics, Billerica, MA) containing 10 mg of sinapinic acid in 1 ml of 2.5% trifluoroacetic acid (MilliporeSigma, St. Louis, MO) and 50% acetonitrile (MilliporeSigma). The protein–matrix suspension (2 μl) was added to the ground-steel matrix-assisted laser desorption/ionization (MALDI) target plate and dried at room temperature. Mass spectra were collected using a UltrafleXtreme MALDI time-of-flight (TOF)/TOF mass spectrometer (Bruker Daltonics) operated with FlexControl software in linear positive mode—that is, using a mass range of 30,000–120,000 daltons. The instrument was externally calibrated with a Proteins MALDI-MS calibration kit (MilliporeSigma, St. Louis, MO). Each individual mass spectrum was analysed and adjusted for smoothness and baseline using FlexAnalysis software 3.0 (Bruker Daltonics). We determined the molecular mass of the heterodimeric GABA_B1b(1–802)–GABA_B2(1–819) complex to be 192,647.967 daltons.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Extended data figures and tables

Extended Data Fig. 1 Purification and functional analysis of the human GABA_B receptor.

a, Superose 6 size-exclusion chromatography profile of detergent-purified GABA_B1b(1–802)–GABA_B2(1–819) complex. b, SDS PAGE gel of the size-exclusion peak fraction from a under reducing conditions. For gel source data, see Supplementary Fig. 2. c, Dose-dependent [³H]GABA binding to purified GABA_B1b(1–802)–GABA_B2(1–819) complex, reaching maximum molar ratios of GABA-to-receptor binding at 0.98 ± 0.03 mol mol⁻¹. Each data point represents the mean of triplicate measurements from a typical experiment. The experiment was repeated four times with similar results. Data were subjected to nonlinear regression fitting, and the dissociation constant (K_d = 1.3 ± 0.16 µM) is reported as means ± s.e.m. of the fit. d, Functional analysis comparison of full-length and truncated WT GABA_B receptor. Shown is the dose-dependent baclofen-stimulated receptor response in cells transiently expressing Gα_qi5 (abbreviated as Gqi) with full-length GABA_B heterodimer or the C-terminally truncated GABA_B1b(1–802)–GABA_B2(1–819) complex. Cells transfected with Gα_qi5 alone were used as a negative control. Relative agonist-stimulated activity was measured by IP₁ accumulation, and is expressed as a percentage of maximum wild-type activity induced by baclofen relative to the activity of Gα_qi5 alone. Data points represent averages ± s.e.m. of multiple experiments (n), each with quadruplicate measurements. Cell surface expression level was 106% for the GABA_B1b(1–802)–GABA_B2(1–819) complex in comparison with the full-length WT/WT heterodimer.

Extended Data Fig. 2 Cryo-EM imaging of human GABA_B receptor.

a, Workflow of cryo-EM data processing. b, A representative motion-corrected cryo-electron micrograph of the GABA_B receptor. c, Reference-free 2D class averages, highlighting clear density for transmembrane helices. d, Global density map, coloured according to local resolution, in full and clipped views perpendicular to the plane of the membrane. e, Global FSC curve (purple) corrected by high-resolution noise substitution. The overall resolution as determined by an FSC cut-off value of 0.143 (blue line) is 3.3 Å. f, 3D FSC curves, measuring directional resolution anisotropy. Plots show the global half-map FSC (red solid line; left-hand y-axis), together with the spread of directional resolution values within ± 1 standard deviation of the mean (area encompassed by green dashed lines), and a histogram (turquoise; right-hand y-axis) of 100 such directional resolution values sampled evenly over the 3D FSC threshold value of 0.143 (blue line). The sphericity value reported by 3D FSC is 0.958 out of 1. g, h, Separate FSC curves for the locally refined reconstructions of ECDs g) and transmembrane domains (h). The blue lines mark the resolution corresponding to an FSC value of 0.143 (ECD, 3.1 Å; transmembrane region, 3.4 Å). The adjacent diagrams show the masks (translucent surface) used for each local refinement.

Extended Data Fig. 3 Structural model of the GABA_B receptor fit within the cryo-EM map.

a–d, Cryo-EM density maps and refined models for the LB1 interface helices (H–B and H–C) in the extracellular VFT (a), the linker between the VFT and transmembrane domains (b), all seven transmembrane helices of the GABA_B1b subunit (c), the ten modelled transmembrane cholesterols (d) and all seven transmembrane helices of the GABA_B2 subunit (e). The density map is a composite of the locally refined reconstructions for the ECD and transmembrane domains. The N- and C-terminal residues of each helix are labelled.

Extended Data Fig. 4 Architecture of the GABA_B receptor.

a, Structure of the GABA_B receptor in four views related by 90° rotations about an axis perpendicular to the membrane. GABA_B1b (blue) and GABA_B2 (green) are rendered as ribbons; Ca²⁺ is shown as a green sphere; phospholipids (PE 38:5 and PC 38:2) are presented as yellow space-filling models; the observed N-linked glycans (NAGs) and cholesterols (CLRs) are shown respectively as grey and pink ball-and-stick models. Transmembrane helices 1–7, along with N and C termini, are marked for each subunit. b, Cryo-EM density map of the GABA_B receptor, in the same orientations and colour scheme as in a. The map is composed of local reconstructions for the ECD and transmembrane domain, which were independently refined to 3.1 Å and 3.4 Å, respectively. c, Linker and transmembrane domain of the GABA_B receptor viewed from the extracellular and intracellular sides.

Extended Data Fig. 5 Heterodimer conformation and interface features of the GABA_B receptor.

a, b, Cryo-EM structure of near full-length GABA_B receptor (cyan) superimposed with the crystal structure of its extracellular VFT module in the inactive state (PDB code 4MQE; purple) (a) or active state (PDB code 4MS3; red) (b). The middle panel shows the heterodimeric receptor structures superimposed based on the LB1 domain of the GABA_B1b subunit. The two side panels show superpositions of individual GABA_B1b and GABA_B2 subunits on the basis of their respective LB1 domains. In b, the green line denotes the axis of rotation that relates the LB2 domains of near full-length and VFT structures of GABA_B1b (rotation χ = 28°; screw translation τ_χ = 0.6 Å), or near full-length and VFT structures of GABA_B2 (rotation χ = 7°; screw translation τ_χ = 0.3 Å). c, d, Extracellular LB2 domains viewed from the C-terminal end. Superposition of near full-length (cyan) and VFT structures (c, inactive state, purple; d, active state, red) based on the LB1 domain of the GABA_B1b subunit. Within each heterodimeric complex, the C termini of the LB2 domains in the GABA_B1b and GABA_B2 subunits are shown as spheres, and the distance between the two C termini is marked by dotted lines. e, Cryo-EM structure of full-length mGlu₅ in the inactive (PDB code 6N52) and active (PDB code 6N51) conformations²⁰. f, Molecular surface of the GABA_B1b–GABA_B2 complex, showing the plane of the heterodimer interface for the extracellular and transmembrane domains. Structural elements involved in heterodimer formation are highlighted in cartoon format (ectodomain, H–B and H–C helices; transmembrane domain, TM5 and TM3 helices). g, The GABA_B transmembrane domain viewed from the extracellular side, comparing the locations of core (I, II, IIIa, IIIb) versus peripheral cholesterol-mediated heterodimer contacts from different layers. Heterodimer contacts mediated by two cholesterols (CLR6 and CLR3) are shown at the bottom.

Extended Data Fig. 6 Extracellular ligand binding in GABA_B1b.

a, b, Molecular surface (a) and ribbon representation (b) of the GABA_B1b subunit, showing the location of the Ca²⁺-binding site and an unmodelled density at the interdomain crevice of VFT. c, d, Functional analysis of the impact of endogenous Ca²⁺. Basal activity (c) and dose-dependent baclofen-stimulated receptor response (d) in cells transiently expressing the Gα_qi5 subunit (abbreviated as Gqi) with different combinations of WT and mutant GABA_B-receptor subunits (GABA_B1b E309K, abbreviated as E309K; GABA_B1b E423R, abbreviated as E423R). IP₁ accumulation by the WT/WT heterodimer was measured in the presence and absence of 2.5 mM EGTA. Cells transfected with Gα_qi5 alone were used as negative controls. Relative activity in both graphs is expressed as a percentage of the maximum wild-type activity induced by baclofen relative to the activity of Gα_qi5 alone. Data points represent averages ± s.e.m. of multiple experiments (n), each with quadruplicate measurements. **P = 0.0016, ***P = 0.0002, ****P < 0.0001; one-way ANOVA with Bonferroni’s post hoc test was used to calculate statistical differences in basal activity (c). Cell surface expression level was 107% for E309K/WT and 87% for E423R/WT mutants in comparison with the WT/WT heterodimer. e, Fitting of GABA into the extra density (contoured at 7.0σ) at the orthosteric ligand-binding site and its potential interaction with GABA_B1b. f, Concentration of endogenous GABA in the supernatant and lysate of HEK 293 GnTI^– cells after recombinant expression of the GABA_B receptor, as well as cell culture medium and lysis buffer controls, as detected by mass spectrometry.

Extended Data Fig. 7 Endogenous phospholipid-binding sites of the GABA_B receptor.

a, b, Ribbon representation of the GABA_B1b (a) and GABA_B2 (b) transmembrane domains, highlighting the cryo-EM density for phospholipids contoured at 4.0σ. Phospholipids are rendered in ball-and-stick representation. c, d, Electrostatic potential surface of the lipid-binding pocket in GABA_B1b (c) and GABA_B2 (d). The phospholipids are shown as spheres. Charged residues that directly contact the phosphate group of each lipid are marked. e–h, Comparison of phospholipids bound to GABA_B subunits with ligands bound to the class A GPCRs rhodopsin²⁷ (PDB code 1F88) and S1P₁ receptor²⁴ (PDB code 3V2Y) (e), the class B GPCR corticotropin-releasing factor receptor 1 (CRF₁; ref. ⁵⁷; PDB code 4K5Y) (f), the class C GPCRs mGlu₁ (ref. ²⁹; PDB code 4OR2) and mGlu₅ (ref. ²⁸; PDB code 4OO9) (g), and the class F GPCR Smoothened⁵⁸ (PDB code 4JKV) (h). In each panel, the Cα trace of the GABA_B1b linker and transmembrane domain is shown in two orthogonal views in grey, and the superimposed GABA_B ligands PE 38:5 and PC 38:2 are shown as blue and green stick models, respectively. Various GPCRs were overlapped onto the transmembrane domain of GABA_B1b to bring their bound ligands into superposition. i, j, Schematic diagrams of the specific contacts between GABA_B1b and PE 38:5 (i), and between GABA_B2 and PC 38:2 (j). Selected contacts between residues and phospholipids are highlighted; hydrogen bonds, red dotted lines; hydrophobic contacts, black wiggled lines; polar interactions, green curved lines; pi-stacking interactions, orange box wave. Red zigzags indicate contacts that are mediated by main-chain atoms. Lipid-interacting residues that are conserved in the two subunits are highlighted in bold and include: (1) head group (GABA_B1b, His 643 and Arg 549^3.32; GABA_B2, His 647 and Arg 556^3.32); (2) 20-carbon fatty acyl chain (GABA_B1b, Phe 557^3.40, Tyr 657^5.44 and Ala 703^6.54; GABA_B2, Tyr 564^3.40, Tyr 661^5.44 and Ala 707^6.54); (3) 18-carbon fatty acyl chain (GABA_B1b, Ile 724^7.36; GABA_B2 Ile728^7.36).

Extended Data Fig. 8 Endogenous phospholipid interactions with GABA_B receptor.

a, c, e, Orthogonal views of a potential access channel in GABA_B1b, GABA_B2 and S1P₁ in molecular surface representation, along with phospholipids PE 38:5, PC 38:2 or the sphingolipid mimic ML056, in space-filling representation. Side views (left) show an opening between helices TM5 and TM6 in GABA_B1b (a) and GABA_B2 (c), and between TM1 and TM7 in S1P₁ (e), while top views (right) highlight the blocked entrance to the lipid-binding pocket from the extracellular side. In all cases, ECL1 and ECL3 (orange), ECL2 (pale brown), the linker of GABA_B-receptor subunits (pink) and the N-terminal helix of S1P₁ (pink) are distinguished by colour. b, d, f, The same information as in a, c, e but with ribbon models for GABA_B1b (b), GABA_B2 (d), and S1P₁ (ref. ²⁴) (PDB code 3V2Y) (f). Lipids are in stick format. g, h, Functional effect of a GABA_B2 lipid-binding site mutation. Basal activity (g) and dose-dependent baclofen-stimulated receptor response (h) in cells transiently expressing Gα_qi5 (Gqi) with WT GABA_B receptor or WT GABA_B1b and mutant GABA_B2 R714A (R714A) heterodimer. Cells transfected with Gα_qi5 alone were used as negative controls. Relative activity in both graphs was measured by IP₁ accumulation, and expressed as a percentage of the maximum wild-type activity induced by baclofen relative to the activity of Gα_qi5 alone. Data points represent averages ± s.e.m. of multiple experiments (n), each with quadruplicate measurements. **P = 0.0016; one-way ANOVA with Bonferroni’s post hoc test was used to calculate statistical differences in basal activity (g). Cell surface expression level was 77% for the WT/R714A mutant in comparison with the WT/WT heterodimer.

Extended Data Fig. 9 Comparison of the GABA_B transmembrane domain with other GPCRs.

a, Superposition of GABA_B1b and GABA_B2 subunits based on their VFT modules. The short purple line denotes the axis of rotation that relates the linker and transmembrane domains of GABA_B1b and GABA_B2 (rotation χ = 23°; screw translation τ_χ = 0.01 Å). b, Superposition of the linker and transmembrane domains of the GABA_B1b and GABA_B2 subunits. c, Superposition of the linker and transmembrane domains of each GABA_B subunit with the class C GPCR mGlu₅ (ref. ²⁰; PDB code 6N52) in three different views, with arrows revealing inward extracellular shifts in TM5 and TM7, as well as an inward intracellular shift in TM3, in mGlu₅ compared with either GABA_B subunit. d–f, Superposition of the transmembrane helices of each GABA_B subunit with the class A GPCR rhodopsin²⁷ (PDB code 1F88) (d), the class B GPCR CRF₁ (ref. ⁵⁷; PDB code 4K5Y) (e), and the class F GPCR Smoothened⁵⁸ (PDB code 4JKV) (f). Arrows indicate large shifts in transmembrane helix positions between GABA_B subunits and other GPCRs, such as outward extracellular movements in class A (TM2 and TM6), class B (TM1, TM2 and TM7), and class F (TM2, TM4 and TM5). There are also inward intracellular shifts in TM7 in all comparisons, and outward intracellular shifts in TM5 in class A, B and F.

Extended Data Fig. 10 Conserved motifs in GABA_B, rhodopsin and mGlu receptors.

a, b, The ‘ionic lock’ and FxPKxx motifs in the GABA_B1b (a) and GABA_B2 (b) subunits. The ‘ionic locks’ consist of Asp 684 of ICL3 and Lys 567^3.50 in GABA_B1b, and Asp 688 of ICL3 and Lys 574^3.50 in GABA_B2. The FxPKxx motifs include the conserved Lys 739^7.51 of GABA_B1b and Lys 743^7.51 of GABA_B2, which interact with the ‘ionic locks’ through Asn^2.39 (GABA_B1b N513^2.39; GABA_B2 N520^2.39) and a serine (GABA_B1b S508; GABA_B2 S515) in ICL1. c, The ‘ionic lock’ and NPxxY motifs in class A rhodopsin²⁷. d, e, The ‘ionic lock’ and FxPKxY motifs in the class C receptors mGlu₁ (ref. ²⁹; d) and mGlu₅ (ref. ²⁸; e), which are in close proximity as in GABA_B subunits. Key residues of the motifs are displayed as stick models. Hydrogen bonds are indicated by black dotted lines. Interactions between participating residues of the ‘ionic lock’ are denoted by red dotted lines with distances labelled. Distances between the Cα atoms of the ‘ionic lock’ residues are marked by brown dotted lines.