Structures of metabotropic GABAB receptor.

Stimulation of the metabotropic GABAB receptor by γ-aminobutyric acid (GABA) results in prolonged inhibition of neurotransmission, which is central to brain physiology1. GABAB belongs to family C of the G-protein-coupled receptors, which operate as dimers to transform synaptic neurotransmitter signals into a cellular response through the binding and activation of heterotrimeric G proteins2,3. However, GABAB is unique in its function as an obligate heterodimer in which agonist binding and G-protein activation take place on distinct subunits4,5. Here we present cryo-electron microscopy structures of heterodimeric and homodimeric full-length GABAB receptors. Complemented by cellular signalling assays and atomistic simulations, these structures reveal that extracellular loop 2 (ECL2) of GABAB has an essential role in relaying structural transitions by ordering the linker that connects the extracellular ligand-binding domain to the transmembrane region. Furthermore, the ECL2 of each of the subunits of GABAB caps and interacts with the hydrophilic head of a phospholipid that occupies the extracellular half of the transmembrane domain, thereby providing a potentially crucial link between ligand binding and the receptor core that engages G proteins. These results provide a starting framework through which to decipher the mechanistic modes of signal transduction mediated by GABAB dimers, and have important implications for rational drug design that targets these receptors.

The neurotransmitter GABA is primarily responsible for synaptic inhibition throughout the nervous system via activation of GABAA ion channels and pre- and postsynaptic GABAB receptors. GABAB stimulation of the Gi/o class of heterotrimeric G proteins results in a prolonged decrease in neuronal excitability via the inhibition of adenylyl cyclase and voltage-gated Ca2+ channels, as well as the opening of G protein-coupled inward rectifying potassium channels[2,3,6]. Abnormal execution of GABAB signaling causes multiple neuropsychiatric diseases and the receptor is an attractive drug target for a range of disorders including drug addiction, pain, epilepsy, spasticity, anxiety, and gastroesophageal reflux disease[7,8].

Family C GPCRs represent ~20 receptors including GABAB, metabotropic glutamate receptors (mGlu1–8), the calcium sensing receptor, as well as two taste and several orphan receptors[9,10]. These receptors operate as obligate dimers, with each subunit composed of a bi-lobed extracellular ligand-binding domain termed the Venus flytrap (VFT) and a 7-transmembrane domain (7TM) connected via a linker region[11]. Crystallographic studies of VFTs from various Family C members, including GABAB, show that agonist binding rearranges the VFTs in a way that would bring the linker regions into close proximity[12,13]. All Family C GPCRs contain a cysteine-rich domain (CRD) within the linker region, except for GABAB which has a relatively short linker. Our recent mGlu5 structures showed that the CRD interacts with ECL2 of the transmembrane region, thereby transducing conformational changes 120Å from the ligand-binding site on the VFT to the 7TM domain where G protein activation occurs[14]. Given that GABAB lacks a CRD, the structural communication between an agonist-bound VFT and the 7TM domains remains unclear.

Furthermore, unlike other Family C GPCRs, GABAB is an obligate heterodimer of dissimilar subunits, GABAB1 and GABAB2[15]. Agonist binding occurs only on the VFT of GABAB1, while G protein coupling and activation occurs exclusively through GABAB2[4,5]. Therefore, besides its pharmacological interest, GABAB presents an ideal model system to study trans-activation mechanisms of dimeric Family C GPCRs. Such studies, however, have been limited by the lack of structural information on full-length GABAB, restricting our ability to understand how agonist binding to GABAB1 results in G protein activation on the intracellular side of GABAB2. In the present study we employed single-particle cryo-electron microscopy (cryoEM), atomistic simulations, and cellular signaling assays to obtain structural and mechanistic insights into full-length GABAB receptor complexes.

For structural studies, we co-expressed recombinant constructs of human GABAB1b and GABAB2 in insect cells and purified the GABAB heterodimer by tandem affinity-chromatography in the presence of the inverse agonist CGP55845[16]. The structure of the inactive detergent solubilized GABAB heterodimer was determined at a global indicated resolution of 3.6 Å (Fig. 1, Extended Data Figs. 1, 2a, and Table 1). The cryoEM map resolved the entire GABAB apart from the C-terminal coil-coil domain that appears to be flexible in relation to the transmembrane regions. Additionally, a focused map of the extracellular region at a resolution of 3.5 Å provided improved density for the linker regions and assisted with modeling (Extended Data Fig. 1). The asymmetric protomers, GABAB1 and GABAB2, share a similar secondary structure and arrangement but are distinguished by differential glycosylation and a well-resolved ligand density within the GABAB1 VFT domain that is absent in GABAB2 (Fig. 1, Extended Data Figs. 3 and 4a). The upper VFT lobes of GABAB form a junction, while the lower VFT lobes are separated by ~20 Å (Fig. 1, Extended Data Fig. 3). The ligand, CGP55845, adopts a horseshoe-like conformation that was confirmed by GemSpot[17], closely resembling the crystal structure of the GABAB1 VFT domain in complex with the inhibitor CGP54626 (Extended Data Fig 3)[12,16]. We also observe a small spherical density in GABAB1 that appears to be interacting with the backbone carbonyl of G277 in addition to several anionic groups and a tyrosine, raising the possibility of a divalent cation in that location (Extended Data Fig. 3e).

Figure 1.

CryoEM map and model of the full-length GABAB receptor heterodimer in inactive state.

GABAB receptor cryoEM map (a) and ribbon diagram (b) (GABAB1b: teal; GABAB2:tan; GDN micelle: white; CGP55845: gold; phospholipids: orange; glycosylation: blue and pink). Elongated densities at the 7TM interface are likely lipid or detergent molecules (yellow). c, GABA Emax increases with higher levels of co-transfected GABAB1 and GABAB2 plasmid DNA. d, ECL2 truncation of either GABAB subunit increases Emax of GABA-stimulated response when compared to co-transfected wild-type receptors of similar surface expression (Extended Data Fig. 6a), but ECL2 shortening of both protomers inhibits activity.

Extended Data Figure 1.

Sample preparation, cryoEM processing and reconstruction of GABAB heterodimer.

a, Purification scheme for GABAB. Representative cryoEM micrograph of 10,315 collected (b) and 2D class averages (c) of GABAB dimers. d, Flow chart outlining the cryoEM processing workflow using RELION[56], the global resolutions of the full-length structure and VFT focused structures were 3.6 Å and 3.5 Å, respectively, at 0.143 Fourier shell correlation (FSC) as calculated by RELION. e, f, Gold standard FSC curve of half-maps calculated using RELION and Phenix Mtriage[57] (e), and map-to-model validation curves generated through Phenix Mtriage (f). g, Local resolution of cryoEM maps. h, Angular distribution of projections used in final cryoEM reconstruction. i, Ordered cryoEM densities (light yellow), likely corresponding to GDN and/or cholesteryl hemisuccinate, are found at the TM5 interface of GABAB1 and GABAB2 7TM domains.

Extended Data Figure 2.

Agreement between cryoEM map and model.

a, EM density and model for GABAB heterodimer complex; transmembrane helices of GABAB1, transmembrane helices of GABAB2, linker region, bound PE, and ligand CGP55845. Densities visualized within UCSF Chimera[42] and zoned at 2.2 with threshold set to 0.0142, with the exception of the following: GABAB1-bound lipid,GABAB1-bound CGP55845, and GABAB2 linker in which thresholds of 0.01, 0.0189, and 0.0127 were used, respectively. b, EM density and model for GABAB1 homodimer; transmembrane helices and linker region of both protomers, 7TM-bound PE, and ligand CGP55845. Densities were zoned at 2.2 and threshold set to 0.016, apart from CGP55845 and the ECL2 in both protomers in which a threshold of 0.03 or 0.02 was used, respectively.

Extended Data Table 1.

Cryo-EM data collection, refinement, and validation statistics.

	GABAB1b/GABAb2 Heterodimer (EMDB-21533) (PDB-6W2X)		GABAB1b Homodimer (EMDB-21534) (PDB-6W2Y)
Data collection and processing	Data Set #1	Data Set #2
Magnification (x)	57,050	57,050	47,198
Voltage (kV)	300	300	300
Electron exposure (e−/Å2)	56.5	64.2	49.7
Defocus range (nm)	−1.5 – −2.7	−1.2 – −2.5	−0.9 – −2.5
Pixel size (Å)	0.8521	0.8521	1.06
Symmetry imposed	C1	C1	C1
Initial particle images (no.)	538,957	2,062,083	2,278,113
Final particle images (no.)	286,140 (Combined Data Sets)		282,811
Map resolution (Å)	3.6 Å		3.2 Å
FSC threshold	0.143		0.143
Map resolution range (Å)	3.3 – 4.4		3.0 – 4.4
Map sharpening B factor (Å2)	−183.273		−84.088
Refinement
Initial model used (PDB code)	4MR7, 6N52 Homology		GABAB1b of 6W2X (this work)
Model resolution (Å)	3.7		3.4
FSC threshold	0.5		0.5
Model composition
Non-hydrogen atoms	9860		9721
Protein residues	1345		1327
Ligands	3		4
B factors (Å2)
Protein	15.17		57.57
Ligand	28.97		59.82
R.m.s. deviations
Bond lengths (Å)	0.004		0.007
Bond angles (°)	0.608		0.763
Validation
MolProbity score	1.75		2.02
Clashscore	5.55		9.98
Poor rotamers (%)	0.00		0.12
Ramachandran plot
Favored (%)	92.92		91.66
Allowed (%)	7.08		8.34
Disallowed (%)	0.00		0.00

Extended Data Figure 3.

Binding of CGP55845 and cation to GABAB1 VFT domain.

a, Model of CGP55845 within the VFT of GABAB1b. The entire ligand is confined by W65 and W278 of GABAB1 that form hydrophobic interactions with the chlorinated ring of CGP55845. b, Schematic of interacting residues on GABAB1b with the inhibitor, CGP55845. GABAB1b residues S153 and S130 form hydrogen bonds with oxygen atoms of the phosphate group, while H170 and E349 form a hydrogen bond and a salt bridge with the amine group of the ligand, respectively. π-π stacking occurs between the chlorinated ring structure of CGP55845 and W278, while W65 provides hydrophobic packing on the opposing side of the ring. S130, H170, E349, and W65 are all substantially different residues in GABAB2, precluding ligand binding. Residues are color-coded corresponding to their properties: light blue, hydrophilic; orange, anionic; green, hydrophobic; yellow, glycine; and grey, cysteine. Interaction lines are also color-coded according to their type: light blue, side-chain hydrogen bonding; grey, backbone hydrogen bonding; blue-red gradient, salt-bridge; and green, π-π stacking. c, Overlay of CGP55845-bound GABAB cryoEM structure (tan/teal) with CGP54626-bound GABAB crystal structure (pink, PDB:4MR7) or apo-state GABAB crystal structure (grey, PDB:4MQE), resulting in RMSD values of 1.30 Å and 1.28 Å respectively[12]. d, Comparison of ligand pose between CGP55845 (yellow) and CGP54626 (blue) , which differs from CGP55845 only in a substitution of an aromatic ring in place of cyclohexane. e, Spherical density surrounded by anionic residues within the VFT supports a cation (magenta) at that site. The presence of a metal ion would be consistent with the observation that calcium and other divalent ions affect ligand affinity[58,59], and examination of the deposited scattering factors for the high-resolution VFT crystal structure (PDB:4MR7) reveals positive difference density also consistent with a cation at this site[12].

Extended Data Figure 4.

Comparison of structures across GPCR classes.

a, GABAB protomers share similar secondary and overall structure. b, Comparison of mGlu5 and GABAB 7TM and ECL2/linker shown from side view. c, Top-down view of GABAB1, GABAB2, mGlu5 (PDB: 6N52), and class A M2 acetylcholine receptor (PDB:3UON) with sidechains corresponding to the “toggle-switch” motif shown. Phospholipid space-filling model is included in gray within GABAB1 and GABAB2. d, Sequence alignment of human GABAB receptors with mGlu5 and M2 receptors comparing canonical GPCR activation motifs: TM3-TM6 ionic lock (blue), “toggle switch motif” (green), FxPKxY motif (yellow). Sequences are aligned to motifs within each TM helix and transmembrane helical secondary structure is underlined. Residues in GABAB sequences differing from canonical motifs are outlined in pink. The cysteine residue that replaces the “toggle-switch” tryptophan is highlighted in pink.

Inactive GABAB assumes a similar overall morphology to the apo-state structure of mGlu5[14], while the most substantial differences arise within the linker region (Fig. 1b). Bridging the VFT and 7TM domains, an ~20 residue linker forms a β-sheet in conjunction with the ECL2. Notably, the length of the GABAB receptor ECL2 is nearly twice that of mGlu5 (Fig.1, Extended Data Fig. 4b). In the absence of CRDs, the β-sheet structure of the GABAB linker in complex with ECL2 orders this region with the additional ECL2 length thus allowing 7TM and VFT domain coupling for signal transduction. Molecular dynamics (MD) simulations support the structural stabilization of the linker through the β-sheet formation. After 200 ns in all simulations the linker and ECL2 continued to adopt a stable structure even in the absence of VFT domains, although a slight downward rotation in β-sheet orientation was observed (Extended Data Fig. 5). To further examine the involvement of the ECL2 in receptor activation, we employed a functional assay with a chimeric Gαo/q, whereby the Gi/o-coupled GABAB receptor can couple to the PLC pathway[3,18]. Accumulation of the downstream metabolite IP1 by LiCl was measured by an established assay[19], thereby monitoring GABA-stimulated and basal receptor activity (Fig. 1c). ECL2 shortening was generally inhibitory to the proper membrane trafficking of the GABAB subunits. Accordingly, we normalized the transfected DNA to obtain similar expression levels between constructs (Extended Data Fig. 6a). The partial deletion of the GABAB1 ECL2 loop (Δ627–634), comprising the unstructured tip of the loop nearest the VFT, produced an increase in basal activity but did not affect GABA Emax when expressed with wild-type GABAB2 (Fig. 1d). These findings indicate that abrogating the ECL2-linker allows flexibility in the GABAB1 VFT relative to the rest of the receptor dimer, resulting in activation through the GABAB2 VFT/7TM route in the absence of agonist[20]. In contrast, the partial deletion of the GABAB2 ECL2 (Δ631–638) did not affect basal activity but produced an increase in GABA Emax and a decrease in GABA potency when compared to wild-type receptor expressed to similar levels at the cell surface, indicating that the GABAB2 ECL2 may be partially inhibitory (Fig. 1d, Extended Data Fig. 6a). When both receptors contain a truncated ECL2, we observed a decrease in GABA Emax, suggesting that at least one VFT domain must be structurally coupled through the extended ECL2/linker for full activation (Fig. 1d). Collectively, these data support a bimodal transactivation mechanism of GABAB in which agonist binding on the GABAB1 receptor can proceed from the GABAB1 VFT down to the GABAB1 7TM region to enact changes in GABAB2 that promote G protein activation, and also activate the GABAB2 7TM directly through the GABAB2 VFT domain.

Extended Data Figure 5.

Atomistic simulations of phospholipid structural stabilization and entry.

a, The results of three out of seven total simulations of GABAB 7TM and linker in the absence of core-bound lipid after 200 ns. The results show the extent of lipid (green, purple) entry in GABAB2 (grey) in the simulations versus the experimental structure (tan) with PE (orange). One of the trajectories was extended an additional 100 ns (rightmost panel) and lipid entry was observed to progress towards the core. b, Representative side view of the GABAB ribbon and stick model from simulations at 200 ns showing persistence of the ECL2/β-sheet structure even in absence of VFT domains. c, d, Violin plots of ensemble distances between GABAB1 (d) and GABAB2 (e) TM helices in simulations with and without core-bound lipid. Distances were measured from the Cα atoms of the following residues in GABAB1: L550 (TM3), W611 (TM4), L653 (TM5), M707 (TM6), A716 (TM7); and the following residues in GABAB2: T557 (TM3), W615 (TM4), L657 (TM5), F711 (TM6), Q720 (TM7). Simulations were run over 200 ns and violin plots represent n=37,500 data points and 5 simulations per condition. e, 7TM cavity volume measured over time for individual simulations. Average cavity volume of phospholipid-bound receptor is shown as dashed line, thick lines indicate rolling averages of 5.33 ns, and thin lines represent raw data.

Extended Data Figure 6.

Functional analysis of GABAB mutants.

a, Comparative normalized surface expression levels of constructs. Surface expression levels of wild-type and mutant GABAB receptors were determined using a direct enzyme-linked immunosorbent assay (ELISA) against the N-terminal GABAB1 FLAG tag and the N-terminal GABAB2 HA-tag. Surface expression levels were normalized to the surface expression of a wild type GABAB receptor when 6.3 ng FLAG-tagged GABAB1 and 6.3 ng HA-tagged GABAB2 DNA was used for transfection. The line x=y indicates similar surface expression of GABAB subunits as evident from transfection with lower amounts of FLAG-tagged GABAB1 and HA-tagged GABAB2. Values above the line have greater GABAB2 surface expression relative to GABAB1 and values below the line have greater GABAB1 surface expression relative to GABAB2. GABAB1 did not reach the cell surface (data point hidden behind triangle at the coordinate (0,0) in the left subpanel). b, c, Mutations of ionic residues forming the interface of GABAB1 (b) and GABAB2 (c) result in increased constitutive activity of the receptor. To achieve a range of expression levels of the mutants in b and c, the DNA amounts transfected were serially diluted 2-fold from 3.3 ng (panel b mutants) or 6.3 ng (panel c mutants) of each subunit, respectively. d, GABAB2 H579A/E677A expressed without GABAB1 shows a moderate increase in basal activity over wild-type GABAB2 expressed alone, although surface expression (e) of the wild-type GABAB2 subunit was higher than of GABAB2 H579A/E677A. f, g, Mutation of lipid coordinating residues (blue triangle) increase the constitutive activity of GABAB1 (f) and GABAB2 (g), while mutations displacing the lipid tails from the 7TM core (purple triangle) result in decreased basal activity of the receptor when compared to wild-type receptor of similar receptor surface expression. h, Key to colors and symbols used in panels a, f, and g with DNA transfection amounts indicated in parenthesis. Data are representative of one experiment performed in triplicate and repeated independently at least three times with similar results (n=3 independent experiments), error bars represent mean ± S.E.M.

Besides the VFTs and the C-terminal coiled-coil, we observe that inactive GABAB forms an additional dimer interface between TM3 and TM5 from each monomer of the heterodimer (Fig. 1, Fig. 2). The interface is formed by ionic interactions between residues on the intracellular side of each receptor, H572(TM3) and E673(TM5) on GABAB1 and H579(TM3) and E677(TM5) of GABAB2, with further stabilization through aromatic residues along the same helices (Fig. 2b, c). Observed elongated densities, likely corresponding to glyco-diosgenin (GDN) and/or cholesteryl hemisuccinate used in purification, packed between the extracellular sides of the two 7TM regions may also stabilize the interface (Fig. 1a). The dimer interface is in contrast to the inactive mGlu5 receptor where ~16Å separates the 7TM domains (Fig. 2a)[14]. To probe the significance of these interactions, we mutated the TM3/TM5 interface, and performed IP1 accumulation assays (Fig. 2d, e, Extended Data Fig. 6). Notably, mutation of either H579 or E677 on GABAB2 or mutation of both E673 and H572 on GABAB1 increased the basal activity of the receptor, suggesting that the TM3/TM5 interface is inhibitory to signaling in the absence of agonist, which further supports the transactivation mechanism described above (Fig.2, Extended Data Figs. 6b–e). When expressed at the cell surface alone, the H579/E677 double mutant of GABAB2 exhibited a slight increase in basal activity compared to GABAB2 alone (Extended Data Fig. 6d). These findings further support a role of an intra-protomer H579/E677 interaction in GABAB2 to assist stabilizing an auto-inhibited state, and are consistent with the lack of constitutive activity by the GABAB2 ECL2 (Δ631–638) deletion mutant.

Figure 2.

TM3 and TM5 stabilize an inactive-state dimer interface of GABAB 7TM domains.

a, Inactive-state GABAB 7TMs (GABAB1: teal, GABAB2: tan) are in closer proximity compared to those of inactive mGlu5 (purple; PDB:6N52), top-down view. b, The GABAB1 and GABAB2 interface is stabilized through hydrophobic interactions along TM5 helices and by polar residues on the intracellular side of TM3 and TM5 (boxed). c, Polar interaction residues from the intracellular side of the receptor. GABAB2 E677 is shown for perspective; we note, however, that the map density for that residue was insufficient for high-confidence modeling of its side chain. IP1 accumulation assays illustrate differences in GABAB TM3/5 mutants’ basal activity, d, and GABA response, e.

The most unexpected observation in the transmembrane region of both monomers of GABAB is a wishbone-shaped density occupying the extracellular half of each 7TM core. The bifurcated density, which is better resolved within the GABAB1 helical bundle, corresponds to a phospholipid (Fig. 3, Extended Data Fig. 2). Considerations of the size and shape of the density, surrounding amino acid environment within the 7TM core, and known phospholipid composition in Sf9 insect cells, led us to infer that the density in GABAB1 and likely GABAB2 corresponds to phosphatidylethanolamine (PE) (Fig. 3, Extended Data Fig. 7). However, it is possible that the observed density is the result of a heterogeneous mixture of phospholipids. The observation of the phospholipid is particularly intriguing, as no other GPCR structure has revealed a two-chained phospholipid within this space. Although a lipid-activated subfamily within Family A GPCRs exists, known ligands are single-acyl-chain lipids, eicosanoids, and sterols[21]. Notably, GABAB residues in the extracellular loops, including ECL2, and TM3 of GABAB coordinate the polar headgroup of the phospholipid, resulting in a ‘lid’ over the 7TM cavity that resembles the lipid-binding Family A GPCRs. The remaining hydrophilic atoms in the lipid appear solvent-exposed, whereas the lipid tails are buried deep into the hydrophobic portion of the transmembrane cavity (Fig. 3b, c). The presence of the 7TM lipid in our cryoEM structures suggests it is bound tightly, as also suggested by its retention during detergent solubilization of the receptor. Additionally, a conserved TM6 tryptophan, which acts as an activating “toggle-switch” in other Family C GPCRs[22,23], is replaced by cysteine in both GABAB1 and GABAB2 (Extended Data Figs. 4c, d). This replacement is essential, as any probable tryptophan rotamer at that position would clash sterically with the bound phospholipid.

Figure 3.

Phospholipid binds within the transmembrane cores of GABAB.

a, EM map clipped to show the location of phospholipid within GABAB1 of the heterodimer. Ribbon representation of PE within GABAB1, b, with boxed region presented in panel c to show the structural environment of the headgroup. d, GABA concentration response for wild-type GABAB (red), and mutants designed to displace phospholipid (green), or de-stabilize the phospholipid headgroup (blue). e, f, MD simulations show a collapse of the transmembrane cavity when lipid is absent. e, Violin plot of cavity volume ensemble data. f, Representative top-down view of the GABAB ribbon and stick model from MD simulations at 200 ns.

Extended Data Figure 7.

Modeling of phospholipid into GABAB.

a, Schematic of GABAB1 residues interacting with the polar headgroup of phosphatidylethanolamine (PE). The terminal amine (-NH3+) forms a salt bridge with residue D714ECL3, while the phosphate group is coordinated by S710ECL3, R549TM3, and the backbone nitrogen of C644ECL2. b-d, As our receptor purification did not contribute additional lipid, we considered the known lipid composition of Sf9 insect cells. Four primary phospholipids are present in Sf9 insect cells: phosphatidylcholine (43%), phosphatidylethanolamine (32%), phosphatidylinositol (23%), and cardiolipin (4%)[60]. It was apparent from the map that the lipid had only two carbon chains, immediately excluding cardiolipin as it has four hydrocarbon tails. A comparison of the map and the binding site residues led to the decision to model PE (b) into the pocket using the GemSpot pipeline[17], which produced good cross-correlation with favorable interactions. To further confirm our selection, analysis of overlays of phosphatidylcholine (c) and phosphatidylinositol (d) over the docked model revealed phosphatidylcholine is unlikely given that the interactions with the cation appear to be primarily salt bridges, rather than the cation-π interactions that more commonly coordinate choline in proteins[61]. Although phosphatidylinositol may make favorable interactions, our map does not appear to support such a large moiety in the headgroup position. Thus, PE is the most likely lipid to reside in the structure and was therefore used in the models.

The presence of lipid in the 7TM core suggests it may have a physiological role in the structural and functional integrity of the transmembrane bundles, occupying a region that corresponds to the ligand-binding site in other GPCR classes (Extended Data Fig. 4). Moreover, given the role of the ECL2 in coupling the VFT domains to the 7TM regions, it is conceivable that interactions of the phospholipid with the ECL2 are essential in increasing the stability of the linker region and also participate in relaying structural transitions from the VFTs to the 7TM core. Markedly, mutations designed to displace lipid tails in the 7TM core, GABAB1 L553W and GABAB2 L560W, decrease basal activity and potency of GABA while increasing GABA Emax, whereas mutation of the arginine coordinating the headgroup of the phospholipids, GABAB1 R549A and GABAB2 R556A, results in increases in both GABAB basal activity and receptor response to GABA (Fig. 3d, Extended Data Figs. 6f and g). Collectively, these results indicate that destabilization of phospholipid in the core affects receptor activation, and it is thus conceivable that the lipid directly modulates receptor activity as a result of ECL2 movements in native GABAB. To further probe the role of phospholipid we employed a range of atomistic simulations with the GABAB receptor system. We initiated simulations of the GABAB 7TM and linker with the VFTs removed, both with and without lipid. The simulations revealed that without lipid the transmembrane helices begin to collapse into the core with mean decreases in cavity volume of ~30% (Fig. 3, Extended Data Fig. 5, Supplementary Data 1 and 2), suggesting a strongly coupled interplay between the phospholipid and 7TM core. Alternatively, multiple simulations that started without lipid in the TMs showed lipid tails from the bilayer entering the receptor hydrophobic cavity below residue Y661, at a site akin to where the tail protrudes in our cryoEM structure. Remarkably, peripheral lipid tail insertion in simulations correlated with circumvention of the 7TM cavity collapse (Extended Data Fig. 5). In one simulation, after 200 ns we observed a lipid that had inserted both of its hydrocarbon tails into the receptor core. Extending the simulation by an additional 100 ns revealed the headgroup has moved over the top of the receptor, indicating lipid entry may occur as a stepwise process, with lipid tail entry sliding between TM5 and TM6 and followed by the entire lipid (Extended Data Fig. 5). Although it is probable that lipid insertion into the 7TM core occurs concurrently with helix insertion into the membrane during protein folding, these results suggest a clear tendency for phospholipids to insert at that position in a mature receptor.

Pharmaceuticals targeting the GABAB receptor and other Family C GPCRs are proposed to function either at the orthosteric ligand binding site or allosterically within the transmembrane core, like Family A orthosteric ligands. However, the bound-phospholipid would appear to occlude the binding of allosteric modulators analogous to those used to target other Family C receptors[24]. Consequently, potential GABAB allosteric modulators would need to either displace the core lipid or bind at an alternative site peripheral to the bundle.

While pure GABAB heterodimer was obtained by tandem affinity purification of receptor constructs with distinct tags, when both subunits contain an N-terminal FLAG tag we also purified a significant fraction (> 40%) of GABAB1 homodimers. In cells, an ER retention signal within the GABAB1 coil-coil domain prevents GABAB1 from reaching the plasma membrane unless masked by the GABAB2 C-terminus or other interacting partners[25,26]. Thus, the presence of GABAB1 homodimer in our preparation is likely due to the persistence of internal membranes during the purification procedure. However, multiple studies have suggested a physiologic role for GABAB1 homodimers within some cell types of the nervous system and gastrointestinal tract that express GABAB1 isoforms in the absence of GABAB2[27-32]. Although it is yet unclear how homodimeric GABAB1 functions in the absence of GABAB2 G protein coupling, we sought to obtain further mechanistic insights into the GABAB system and obtained the structure of the GABAB1b homodimer at a global indicated resolution of 3.2 Å (Fig. 4, Extended Data Figs. 2, 8, and Table 1).

Figure 4.

Inhibitor-bound GABAB1 homodimers adopt similar VFT orientation as active GABAB heterodimer and similar 7TM interface as active mGlu5.

a, VFT overlay of crystal structures of GABAB heterodimer (tan, teal) in inactive-state (top) and active-state (PDB:4MS4)[12] (bottom) with the model of GABAB1 homodimer (blue). The upper lobe interaction interface (boxed) of the GABAB1 homodimer matches that of the active-state heterodimer. b, The GABAB1 subunit of the heterodimer (teal) and homodimer (blue) were aligned to the 7TM domains, revealing differences in VFT and linker positioning. c, Top-down view of superposed 7TM domains of GABAB heterodimer (tan, teal) and homodimer (blue) showing the dramatic difference in protomer interface. d, Top-down view of superimposed 7TM domains of agonist-bound mGlu5 (purple, PDB:6N51[14]) and GABAB1 homodimer (green).

Extended Data Figure 8.

CryoEM processing workflow of GABAB1b homodimer.

a, Flow chart outlining the cryoEM processing of the GABAB homodimer. b, c, Representative micrograph of 5,602 collected (b) and 2D class averages (c). d, Local resolution of cryoEM maps. e, Angular distribution of projections employed in the final cryoEM reconstruction. f, g, Gold standard FSC curve of half-maps calculated using RELION and Phenix Mtriage[57] (f), and map-to-model validation curves generated through Phenix Mtriage (g). Global indicated resolutions of the full-length structure and VFT structure were 3.2 Å and 3.1 Å, respectively, at 0.143 Fourier shell correlation (FSC) as calculated by RELION. h, EM maps of active mGlu5 (purple, EMD-0345), and homodimeric GABAB1 (green). The GABAB1 homodimer adopts a similar overall architecture as active mGlu5.

The structure of the GABAB1 homodimer shows that both VFTs are liganded and the monomers assume a roughly 2-fold symmetric arrangement. The individual lobes of each VFT domain are in an open-conformation and superimposable with the GABAB1 subunit VFT in the inactive heterodimer (Fig. 4b). Remarkably, despite the individual protomer conformation, the VFT homodimer adopts the same overall conformation as the crystal structure of agonist-bound GABAB VFTs (PDB:4MS4)[12] due to an almost identical upper lobe interaction interface (Fig. 4b). Correspondingly, the lower lobes of the receptor are brought into proximity and the receptor adopts a more compact state shortening the distance between the VFTs and the 7TMs (Fig. 4). Compared to the heterodimer, the 7TM domains in the GABAB1 homodimer are rotated to form a TM6-TM6 interface (Fig. 4c). This is in agreement with cross-linking experiments that identified a TM6-TM6 interaction upon agonist-induced stimulation of the heterodimer[33], and agrees with structural rearrangements of mGlu5 upon agonist stimulation (Fig. 4d)[14]. Furthermore, the relative arrangement of the VFT and transmembrane regions are strikingly similar to the activated state of near full-length mGlu5[14] (Extended Data Fig. 8). Thus, the conformation observed in the GABAB1 homodimer may provide valuable hints into the architecture of the active state GABAB receptor (Fig. 4). Along the same lines, GABAB chimera studies show that replacement of one of the 7TM domains of the GABAB1 homodimer with that of the GABAB2 receptor, thus enabling G protein coupling, produces a constitutively active receptor that is not additionally stimulated by GABA[34,35]. Hence, the relative positioning of the VFT domains, as observed in the GABAB1 homodimer structure, appears to be sufficient to stabilize an active conformation, presumably through the linkage of VFT to 7TM domains to reorient the helical interface. Since GABAB1 in the homodimer is bound to an inverse agonist, the adoption of an ‘active-like’ global conformation of the VFT domains must be irrespective of the ligand, likely due to differences in the interaction interface of the VFT N-terminal lobes in comparison to the same interface in the heterodimer.

Collectively, our structures, cellular signaling assays, and atomistic simulations provide crucial insights into mechanistic aspects of GABAB signaling. The extended ECL2 and its interaction with the linker region appears to compensate for the lack of CRD in GABAB compared to all other Family C receptors, and thus also transduce conformational changes from the VFT to the 7TM domains. The phospholipid occupying the 7TM core is important for the integrity of the transmembrane domains in both GABAB subunits, while also structurally coordinating the critical ECL2 region. Our findings support a model in which agonist binding to GABAB1 results in VFT dimer compaction that reorients the protomers via the linker/ECL2. Such conformational change would drive the 7TM domains to twist away from an auto-inhibited state mediated by the inactive TM3/TM5 interface, thereby forming a new TM6 helical interface. Although currently unclear, the activating transitions within the GABAB2 7TM are likely mediated both through the newly formed TM6/TM6 interface and propagation through its own ECL2, as supported by our functional assays. An intriguing possibility is that the phospholipid may act as a sensor of changes in VFT and ECL2 conformations that would result in activating transitions within the 7TM domain of GABAB2 to prime it for G protein engagement. Addressing these questions will require detailed structural studies of active state heterodimer alone and in complex with G protein engaging the receptor core. The present work, along with our recent studies on mGlu5, form a starting structural framework to decipher the enigmatic signal transduction mechanism of GABAB and Family C GPCRs in the context of full-length receptors.

Materials & Methods

Cloning

The cDNA clone for human GABAB2 receptor (Accession: NM_005458) in pcDNA3.1+ was obtained from the cDNA Resource Center (www.cdna.org); and the cDNA clone for human GABAB1b was purchased from Horizon Discovery (Accession: BC050532, Clone ID: 5732186). Primers were designed to include a hemagglutinin (HA) signal sequence[36] in the place of authentic signal sequences of each receptor, thus removing the first 29 residues of GABAB1 and 41 residues of GABAB2. Both the authentic signal sequence and the HA sequence are cleaved during processing; therefore, the substitution does not result in a change of sequence in the mature receptor. For purification of receptors for cryoEM studies, GABAB constructs were subcloned into the pFastBacDual vector (Invitrogen) with N-terminal Flag epitope (DYKDDDD) following the HA signal sequence and/or C-terminal hexa-histidine (His6) tags, so that the following constructs were produced: HA-FLAG-GABAB1(30–844)-His6, HA-GABAB1(30–844)-His6, and HA-FLAG-GABAB2(41–941). For signaling assays the HA-FLAG-GABAB1(30–844)-His6 and a HA-HA-GABAB2(41–941)-construct (where FLAG tag was replaced by the HA epitope tag YPYDVPDYA) were subcloned into pcDNA3.1(+). The primers (Integrated DNA Technologies) used to sub-clone GABAB included: EcoRI-HA-Flag-GABAB1,5’-GCGCGCGAATTCATGAAGACGATCAT

CGCCCTGAGCTACATCTTCTGCCTGGTGTTCGCCGATTACAAGGACGACGATGACAAGTCCACTCCCCCCATCTCCCG-3’ ; GABAB1-His6-SalI, 5’- GCGCGCGTCGACTTAATGATGATGATG

ATGGTGCTTATAAAGCAAATGCAC-3’; GABAB1-SalI 5’- GCGCGCGTCGACTTACTTATAAAG

CAAATGCAC-3’; EcoRI-HA-FLAG-GABAB2, 5’-GCGCGCGAATTCATGAAGACGATCATCGCC

CTGAGCTACATCTTCTGCCTGGTGTTCGCCGATTACAAGGACGACGATGACAAGTGGGCGCGGGGCGCCCCC-3’; GABAB2-His6-SalI, 5’- GCGCGCGTCGACTTAATGATGATGATGATGGTG

CAGGCCCGAGACCATGAC-3’; GABAB2-SalI, 5’- GCGCGCGTCGACTTACAGGCCCGAGACCA

TGAC-3’. To generate GABAB mutants PCR reactions were conducted with either PfuTurbo or Q5 polymerase using the following pairs of primers and pcDNA3.1+neo containing either HA-FLAG-GABAB1b or HA-HA-GABAB2: GABAB1b H572A, 5’-GGTGGGTCGCCACGGTCTTC-3’ and 5’-GAAGACCGTGGCGACCCACC-3’; GABAB1b E673A, 5’- CTTGCTTATGCTACCAAGAG-3’ and 5’-CTCTTGGTAGCATAAGCAAG-3’; GABAB2 H579A, 5’-CTGGAGAGTCGCTGCCATCTTCAA-3’ and 5’- TTGAAGATGGCAGCGACTCTCCAG-3’; GABAB2 E677A, 5’- CTTAGCTTGGGCTACCCGCAAC-3’ and 5’- GTTGCGGGTAGCCCAAGCTAAG-3’; GABAB1 (Δ627–634), 5’- CTTGGCAAATGTCTCAATGGTC-3’ and 5’-GTCTCTATTCTGCCCCAGC-3’; GABAB2 (Δ631–638), 5’-ATCTCCATCCGCCCTCTCC-3’ and 5’-CATGCTGTACTTCTCCACTG-3’; GABAB1 R549A, 5’- CTGCCAGGCCGCCCTCTGGCTCCTG-3’ and 5’- ACGAAAGGGAACTGG-3’; GABAB2 R556A, 5’- GCACCGTCGCTACCTGGATTCTC-3’ and 5’- AAA GTG TTT CAA AGG-3’; GABAB1 L553W, 5’- CTCTGGCTCTGGGGCCTGGGCTTTAG-3’ and 5’-GCGGGCCTGGCAGACG-3’; GABAB2 L560W, 5’-GGACCTGGATTTGGACCGTGGGCTAC-3’ and 5’-TGACGGTGCAAAGTG-3’.

Expression and Purification

Spodoptera frugiperda (Sf9) insect cells (Expression Systems) were co-infected at a density of ~2.0 × 106 cells/mL with HA-FLAG-GABAB2 baculovirus and either HA-FLAG-GABAB1b-His6 or HA-GABAB1b-His6 baculovirus at a multiplicity of infection (M.O.I.) between 3.0 – 5.0. During expression, cells were treated with 5 μM CGP55845 (Hello Bio, Inc.). At 48 hours post-infection cells were harvested by centrifugation, washed once with phosphate-buffered saline containing protease inhibitors (leupeptin, soybean trypsin inhibitor, N-p-Tosyl-L-phenylalanine chloromethyl ketone, Tosyl-L-lysyl-chloromethane hydrochloride, phenylmethylsulfonyl fluoride, aprotinin, bestatin, pepstatin) and 5 μM CGP55845. Cell lysis was achieved through nitrogen cavitation in buffer containing 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM EDTA, 10 μM CGP55845, 2 mM MgCl2, nuclease, and protease inhibitors. The whole-cell lysate was centrifuged at 1,000 xg to remove nuclei and unbroken cells. The supernatant was centrifuged at 100,000 xg to isolate the membrane fraction. Membranes were resuspended by Dounce homogenization in buffer containing 20 mM HEPES, pH 7.5, 150 mM NaCl, 2 mM MgCl2, 1 mM EDTA, 2 mg/mL iodoacetamide, 10 μM CGP55845, 1% n-Dodecyl β-D-maltoside (DDM), 0.2% Sodium cholate, 0.2% Cholesterol hemisuccinate (CHS), nuclease, and protease inhibitors. Solubilized membranes were clarified by centrifugation at 100,000 xg, and the supernatant was loaded onto a pre-equilibrated column of anti-DYKDDDDK G1 affinity resin (Genscript). The resin was washed with Buffer A (20 mM HEPES, pH7.5, 150 mM NaCl, 10 μM CGP55845, and protease inhibitors) with 0.1% DDM and 0.02% CHS. Protein was eluted with Buffer A containing 0.1% DDM, 0.02% CHS, and 0.2 mg/mL DYKDDDDK peptide. The eluate was then loaded onto a pre-equilibrated Nickel-NTA column. Resin was washed with Buffer A containing 0.1% DDM, 0.02% CHS; and the buffer was exchanged in six steps to Buffer A supplemented with 0.2% GDN, 0.02% CHS, followed by a two-step exchange into Buffer A containing 0.004% GDN, 0.0004% CHS. Protein was eluted from the Ni-NTA resin with Buffer A containing 0.004% GDN, 0.0004% CHS, and 500 mM Imidazole. The resulting eluate was concentrated by centrifugal filtration with a 50 kDa molecular weight cut off, and subsequently run on a Superose 6 size exclusion column (GE Healthcare). Samples were pre-screened for sample quality by negative stain transmission electron microscopy and then immediately prepared on cryoEM grids.

CryoEM Data Collection

For the GABAB1b/GABAB2 heterodimer, 3.5 μL of sample was applied at a concentration of 3–5 mg/mL to glow-discharged holey carbon grids (Quantifoil R1.2/1.3). The grids were blotted using an FEI Vitrobot Mark IV (Thermo Fisher Scientific) at 18 °C and 100% humidity, and plunge frozen into liquid ethane. Two data sets were used to produce the final structure. For both data collections cryoEM imaging was performed on a Titan Krios (Thermo Fisher Scientific) electron microscope equipped with a K3 Summit direct electron detector (Gatan). The microscope was operated at 300 kV accelerating voltage, at a magnification of 57,050x in counting mode resulting in a magnified pixel size of 0.8521 Å. For the first data set, movies were obtained at an exposure rate of 14.19 electrons/ Å2/sec with defocus ranging from −1.5 to −2.7 μm. The total exposure time was 3.985 sec over 57 frames per movie stack. For the second data set, movies were obtained at an exposure rate of 21.43 electrons/ Å2/sec with defocus ranging from −1.2 to −2.5 μm. The total exposure time was 2.996 sec including 50 frames per movie stack.

CryoEM grids for the GABAB1b homodimer at a concentration 5.0 mg/mL were prepared similar to the heterodimer. CryoEM imaging was performed on a Titan Krios electron microscope equipped with a post-column energy filter and a K2 Summit direct electron detector (Gatan). The microscope was operated at 300 kV accelerating voltage, at a magnification of 47,198x in counting mode resulting in a pixel size of 1.06 Å. Movies were obtained at an exposure rate of 6.212 electrons/Å2/sec with defocus ranging from −0.9 to −2.5 μm. The total exposure time was 8.0 sec over 40 frames per movie stack. Automatic data acquisition was performed using SerialEM[37] for all data sets.

Image Processing and 3D Reconstruction

Dose-fractionated image stacks were subjected to beam-induced motion correction and dose-weighting using MotionCor2[38]. Contrast transfer function parameters for each non-dose weighted micrograph were determined by Gctf[39] for the homodimer and data set #1 of the heterodimer, and by CtfFind-4.1[40] for data set #2 of the heterodimer. For all data sets; particle selection, 2D and 3D classification were performed on a binned dataset (pixel size 1.72Å and 4.24Å for the heterodimer and homodimer, respectively) using RELION (versions 3.0 and 3.1)[41]. The two data sets for the heterodimer were processed individually before being combined following a Bayesian polishing step. A total of 538,957 particles from 1,324 micrographs and 2,062,083 particles from 8,991 micrographs were extracted using semi-automated particle selection for the heterodimer data set #1 and #2, respectively. Both particle sets were then separately subjected to three rounds of 2D classification and two rounds of 3D classification. Particles in both sets were subjected to Bayesian polishing individually and then combined for a total of 286,140 particles. The merged dataset was fit for Ctf parameters (per particle defocus and astigmatism, per micrograph B-factor) and estimated for anisotropic magnification and beam-tilt. A final 3D refinement was followed by post-processing using a mask that excluded the GDN micelle density. A focused refinement was also carried out using a mask encompassing the VFT and linker regions of GABAB. For the GABAB1b homodimer structure, a total of 2,278,113 particles were extracted from 5,602 micrographs using semi-automated particle selection. Particles were subjected to multiple rounds of 2D and 3D classification until a subset of 282,811particles were selected for the final map. The particle set underwent multiple rounds of Ctf parameter fitting and was subjected to Bayesian polishing before 3D Refinement and post-processing of the final map. UCSF Chimera[42] was used for map/model visualization.

Model Building

The initial model for the VFT domain was taken from the inactive state VFT crystal structure PDB:4MR7[12] and the initial structure of the transmembrane domain of GABAB1b was generated as a homology model from the inactive cryoEM structure of mGlu5 (PDB:6N52, 38% sequence similarity to GABAB1) using Schrödinger’s Prime homology modeling[14]. Both components were placed into the GABAB cryoEM map using Chimera’s ‘fit-in-map’ function. The linker, intracellular loops, and extracellular loops of GABAB1 were interactively adjusted into the EM map using Coot (version 0.8.9.1el)[43] and the resulting model of the GABAB1 Linker/7TM was then used to generate a homology model of GABAB2 using Schrodinger’s Prime homology modeling, which was also placed into the map in Chimera. Iterative rounds of interactive model adjustment in Coot followed by real-space refinement in Phenix (version 1.17.1–3660)[44] employing secondary structure restraints in addition to the default restraints were performed to improve the modeling. The resulting structure of GABAB1 was sufficiently different than the original homology model such that that the RMSD between the homology model and the final model is non-trivial (~2.0Å in the transmembrane helices alone). Once confidence in the sidechain placement was reached for the ligand-binding cleft on GABAB1, the GemSpot pipeline[17] (Schrödinger) was used to model the inhibitor, CGP55845, into the map. After further improvement, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE) was modeled in the transmembrane pocket of GABAB1 and GABAB2 with the GemSpot pipeline. Final refinement was performed with Phenix[44].

Molecular Dynamics Simulations and Analysis

To prepare the system for molecular dynamics simulations, the low-resolution features of the map were used to manually build ICL2 into the model of the GABAB1/GABAB2 inactive heterodimer using Coot. The system was then prepared in Maestro, version 2019–4 (Schrödinger) to build any stubbed sidechains and determine protonation states. The VFTs were removed from the heterodimer to produce a truncated construct starting at residues T461 for GABAB1b and T468 for GABAB2, thus containing only the linkers and the TM domains. The Orientations of Proteins in Membranes (OPM)[45] webserver was used to orient the system with respect to a membrane plane and the CHARMM-GUI[46] was employed to generate a pdb of the system in either a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and cholesterol bilayer or a 3:1 POPC:1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE) and cholesterol bilayer. Approximate dimensions for the system were 105 × 105 × 110 Å for a total of 240 lipid and 7 cholesterol molecules. This bilayer was then solvated in TIP3P water with 150 mM sodium chloride ions balanced to achieve charge neutrality. The salt concentration of 150 mM is consistent with the conditions at which the receptor was purified. POPE was used for the lipid in the TM binding sites of GABAB1 and GABAB2.

The PDB file for the full solvated system was prepared in VMD (version 1.9.3)[47] for simulation in NAMD (version 2.13)[48] to produce a protein structure file (psf). The OPLS-AA/M[49] force field was used for the protein, while OPLS-AA[49] was used for the lipids, cholesterol, and ions. Disulfide bonds were placed between C546 and C644 in GABAB1 and C553 and C648 in GABAB2 and both the N- and C- termini were blocked with capping groups, acetylated N-termini, and N-methylamide C-termini. NAMD was used to run molecular dynamics simulations, where all phases employed periodic boundary conditions with non-bonded interactions smoothed starting at 10 Å to 12 Å, with long range interactions treated with the particle mesh Ewald method. Systems were minimized for 2000 steps and then slowly heated in the NPT ensemble with a Langevin thermostat and a Nosé-Hoover Langevin piston barostat set at 1 atm with a period of 50 fs and a decay of 25 fs. A 2 fs time-step was used with the SHAKE[50] and SETTLE[51] algorithms. Heating occurred from 0 K to 310 K in increments of 20 K with 0.4 ns of simulation at each increment. Harmonic restraints of 1 kcal/mol/Å2 were used during heating on all non-hydrogen atoms of the protein and lipids. The system was then equilibrated with 1 kcal/mol/Å2 harmonic restraints on all protein and lipid non-hydrogen atoms for 10 ns followed by another 10 ns of equilibration with 1 kcal/mol/Å2 harmonic restraints on non-hydrogen backbone atoms. Finally, 1 kcal/mol/Å2 harmonic restraints were applied to only C alpha atoms for 2 ns before being stepped down to 0.5 kcal/mol/Å2 for 2 ns, 0.3 kcal/mol/Å2 for 2 ns, and then removed. The first 30 ns of unrestrained molecular dynamics were also discarded as equilibration.

All trajectories were down sampled by 10x for analysis. Cavity volume was calculated with Epock (1.0.5)[52] in VMD[47] on trajectories that had been aligned to either GABAB1 or GABAB2 from the starting structure. The cavity region was defined to include the binding region of the hydrophobic tails of the lipid. TM-TM distances were calculated in VMD based on the CA position of residues: 3.33, 4.50, 5.40, 6.54, and 7.28 in the Ballesteros–Weinstein[53] numbering scheme.

Transfection and seeding of cells for signaling assays

HEK293 cells (ATCC® CRL-1573™) were transfected with expression vector DNAs encoding the two GABAB receptor protomers and a chimeric Gαq/o5 subunit (five C-terminal amino acids of Gαq were exchanged with those of Gαo) to allow the Gαi/o-coupled GABAB receptor to activate PLC and induce IP3 and intracellular Ca2+ release[3]. Prior to transfection cells were brought into suspension by trypsinization and resuspension to 0.18 million cells/mL in growth medium (D-MEM, Gibco 10566016; supplemented with 10% Fetal Bovine Serum, Gibco 10270106; 1% Sodium Pyruvate, Gibco 11360039; 1% MEM Non Essential Amino Acids, Gibco 11140068; and 1% Penicillin-Streptomycin Solution, Gibco 15140122).

For each 1 mL of cell suspension transfected, a total of 1 μg DNA in 25 μL OptiMEM (Gibco 51985) was incubated for 20 min with a mixture of 57 μL OptiMEM and 3 μL FuGene6 (Promega E2692). After FuGene6/DNA complex formation, the mixture was added directly to the cell suspension, mixed thoroughly and cells seeded with 100 μL cell suspension in appropriate 96-well plates. Of the 1 μg DNA/mL cell suspension, the amount of expression vector DNA encoding the chimeric Gαq/o5 was 0.5 μg/mL cell suspension in all experiments. The amount of GABAB encoding DNA was varied between 7.8 ng and 0.25 μg for each of the GABAB receptor protomer DNAs depending on the assay and mutants tested. Empty vector DNA was added to give a total amount of 1 μg DNA/mL cell suspension transfected. For characterization of basal activity of the TM3/5 protomer mutants, a typical gene dose experiment was performed. DNA corresponding to 62.5 ng DNA/mL cell suspension of each of the protomers (wild-type or mutants) were mixed and serially diluted 2-fold 5 times, typically down to 3.9 ng DNA/mL cell suspension. The transfected cell suspension was seeded at 100 μg/mL both in clear poly-L-lysine coated 96-well plates for IP1 accumulation assays and in white poly-L-lysine coated 96-well plates for cell surface ELISA assays.

IP1 accumulation assays

The IP1 assays for wild-type and mutant receptors was performed essentially as described[14]. Forty-eight hours after transfection, the growth medium was replaced with HBSS buffer (HBSS (Gibco 14025), 20 mM HEPES pH 7.5, 1 mM CaCl2, 1 mM MgCl2 and 0.1% BSA) supplemented with BSA to 0.5% and incubated at 37 °C for 3–4 hours.

For characterization of the TM3/TM5 protomer interface mutants for basal activity, the HBSS + 0.5% BSA buffer was replaced with 100 μL HBSS buffer, followed by addition of 50 μL HBSS buffer containing LiCl (150 mM) to give a final concentration of 50 mM LiCl. After incubation for 1 hour at 37°C the IP1 accumulation was stopped by addition of 40 μL CisBio IP-One Tb HTRF Kit (CisBio, 62IPAPEC) lysis buffer. The accumulated IP1 levels were determined according to the manufacturer’s instructions and as described[14].

For the generation of GABA concentration response curves, the compounds were diluted in three times the final concentration in HBSS buffer containing 60 mM LiCl. The assay was started, first by replacing the HBSS + 0.5% BSA buffer with 100 μL HBSS buffer, followed by addition of 50 μL of the above compound dilutions to give a final LiCl concentration of 20 mM. The IP1 accumulation assay was stopped and assayed as described above after incubation for 1 hour at 37 °C. Data were calculated as the amount of IP1 formed per well or normalized to the basal IP1 level and fitted by non-linear regression using GraphPad Prism. Results and description of statistical analyses used in this manuscript are found in Supplementary Information Table 1.

Cell surface ELISA assay

Surface expression levels of wild-type and mutant GABAB receptors were determined using a direct enzyme-linked immunosorbent assay (ELISA) against the N-terminal GABAB1b FLAG tag and the N-terminal GABAB2 HA-tag, as described[54]. Transfected cells were seeded in white Poly-D-Lysine-coated 96-well plates. Forty-eight hours after transfection, cells were washed once with 100 μL/well DPBS + 1 mM CaCl2 (wash buffer). Following fixation with 50 μL/well 4% paraformaldehyde solution for 5 minutes at room temperature, cells were washed twice with 100 μl wash buffer and blocked with 100 μL/well blocking solution (3% dry-milk, 1 mM CaCl2, 50 mM Tris-HCl, pH 7.5) for 30 minutes at room temperature, followed by addition of 75 μL/well HRP-conjugated anti-FLAG antibody (Sigma Aldrich, A8592), or HRP-conjugated anti-HA antibody (R&D systems HAM0601), both diluted 1:2000 in blocking solution, and allowed to incubate for 1 hour at room temperature. The plates were then washed four times with 100 μl/well blocking solution followed by four washes with wash buffer. The amount of surface expressed receptors was detected by adding 60 μL wash buffer and 20 μL HRP substrate (Bio-Rad, 170–5060) per well, incubating for 10 minutes and measuring of luminescence in an EnVision plate reader (Perkin Elmer).

Statistics and Reproducibility

Detailed results and description of statistical analyses used in this manuscript are found in Supplementary Table 1. Data and error bars in Figs. 1c and 1d represent mean ± S.E.M. from at least five independent experiments. In Fig. 1c, data of co-expressed GABAB1-WT (6.3 ng) with GABAB2-WT (6.3 ng), GABAB1-WT (3.1 ng) with GABAB2-WT (3.1 ng), and GABAB1-WT (1.6 ng) with GABAB2-WT (1.6 ng) represent n=7 independent experiments; data of co-expressed GABAB1-WT (0.8 ng) with GABAB2-WT (0.8 ng), solely expressed GABAB1-WT (6.3 ng), solely expressed GABAB2-WT (3.1 ng), and untransfected HEK293 cells represent n=6 independent experiments. In Fig. 1d, data of co-expressed GABAB1 (Δ627–634) with GABAB2-WT, and GABAB1-WT with GABAB2 (Δ631–638) represent n=7 independent experiments; and co-expressed GABAB1 (Δ627–634) with GABAB2 (Δ631–638) represent n=5 independent experiments. Data in Fig. 2d are representative of one experiment performed in triplicate and repeated independently at least three times with similar results (n=3), data points and error bars represent mean ± S.E.M. Data in Fig. 2e represent mean ± S.E.M. from at least three independent experiments; data for co-expressed GABAB1 (H572A/E673A ) with GABAB2-WT, and GABAB1-WT with GABAB2 (H579A/E677A) represent n=3 independent experiments; and data for GABAB1-WT with GABAB2-WT represent n=7 independent experiments. Data in Fig. 3d represent mean ± S.E.M. from at least four independent experiments; co-expressed GABAB1 (L553W) with GABAB2 (L560W) represent n=5 independent experiments, and GABAB1 (R549A) with GABAB2 (R556A) represent n=4 independent experiments, and data for GABAB1-WT with GABAB2-WT represent n=7 independent experiments. Ensemble data in Fig. 3e represent distribution over a 200 ns time course, for 5 simulations per condition, each violin represents n=37,500 time points.