Introduction

γ-Aminobutyric acid (GABA) is the major inhibitory neurotransmitter in the human central nervous system, and is crucial for the balance of neuronal excitation and inhibition1. Two types of GABA receptors have been identified based on their mechanism of action: GABAA receptors are GABA-gated chloride ion channels, while GABAB receptors are G protein-coupled receptors1. GABAA receptors belong to the pentameric ligand-gated ion channel (pLGIC) superfamily, which also contains ionotropic receptors for acetylcholine, glycine, and serotonin2. Channels in this family share a common architecture, with each of the five subunits contributing to an extracellular domain (ECD) with 10 strands (β1–β10) and a transmembrane domain (TMD) with 4 helices (M1–M4). The orthosteric ligand-binding site is located at the subunit interface in the ECD, composed of loops A, B and C from the principal subunit and D, E and F from the complementary subunit. In a generalized gating mechanism for pLGICs, binding of agonist to the unliganded resting state induces lockdown of loop C over the orthosteric site, and a long-range conformational transition spanning most of the protein2. This transition is thought to progress through a primed state with increased ligand affinity but transient kinetics, an open state with an expanded hydrophobic gate at the midpoint across the membrane, and a desensitized state with a contracted gate at the intracellular end of the pore. Although open states of common GABAA-receptor subtypes have proved difficult to resolve, a growing catalog of structures representing resting, desensitized, and intermediate states offer critical insights into conformational cycling and ligand modulation, as outlined below. In contrast, GABAB receptors mediate relatively slow and prolonged synaptic inhibition, with presynaptic GABAB receptors suppressing neurotransmitter release, and postsynaptic GABAB receptors causing hyperpolarization of neurons3.

In humans, GABAA receptors are homo- or hetero-pentamers formed from a selection of 19 different subunits (α1-6, β1-3, γ1-3, ρ1-3, δ, ε, π and θ)4. Channels formed from ρ subunits were previously named GABAC receptors due to their distinctive physiological and pharmacological properties, including higher sensitivity to but slower activation by GABA relative to classical synaptic and extrasynaptic subtypes; this subtype is also relatively insensitive to bicuculline, barbiturates and benzodiazepines5, but sensitive to phosphinic acid compounds including (1,2,5,6-tetrahydropyridin-4-yl)methylphosphinic acid (TPMPA)6, which provides possibilities to selectively modulate particular subforms of human GABAA receptors. Of the 3 types of ρ subunits found in mammals, ρ1 is expressed particularly in the retina, while ρ2 and ρ3 are widely distributed in brain7,8. ρ-type receptors play important roles in physiological processes including visual transduction9, postnatal neurodevelopment10, pain sensation11, and sleep-wake cycles12, and are potential therapeutic targets in myopia, sleep disorders, learning and memory disruption, peripheral nociception and anxiety13,14.

Several ligands act on ρ-type GABAA receptors, including the synthetic agents 4,5,6,7-tetrahydroisoxazolo[5,4-c]pyridin-3-ol (THIP, also known as gaboxadol) and 3-aminopropyl-n-butylphosphinic acid (CGP36742, also known as SGS742) and the natural product γ-amino-β-hydroxybutyric acid (GABOB, also known as buxamine) (Supplementary Fig. 1). THIP, a conformationally constrained derivative of muscimol, was developed as a non-opioid analgesic and antinociceptive agent15,16, and was a candidate in clinical trials for the treatment of insomnia17, Fragile X syndrome18 and Angelman syndrome19. CGP36742 was the first GABAB receptor antagonist in clinical trials20, but is also an orally active antagonist of ρ-type GABAA receptors21,22, and has shown therapeutic potential for the treatment of cognitive deficits23,24. As a metabolite of GABA, GABOB is found endogenously in the mammalian central nervous system, but it is also an anticonvulsant used in the treatment of epilepsy25. The hydroxyl group at the C3 position of GABOB generates a stereogenic center, resulting in R and S enantiomeric forms; although (R)-GABOB is a modestly more potent anticonvulsant, the compound is applied clinically as a racemic mixture26. Despite their therapeutic relevance, the structural foundations of these drugs’ selectivity and other functional properties at ρ-type GABAA receptors remain unclear.

Here, combining four original cryo-EM structures, electrophysiology recordings, and molecular dynamics (MD) simulations, we characterize the binding and structural impact of THIP, CGP36742 and GABOB on human ρ1 GABAA receptors. We identify a distinctive binding pose of THIP in ρ1 versus the extrasynaptic neuronal α4β3δ GABAA receptors, offering a rationale for its inverse effects on these subtypes. CGP36742 binding is similar to that of TPMPA, detailing a shared mechanism of action among phosphinic acid inhibitors. In contrast, GABOB binding is similar to that of GABA, but under equivalent conditions produces a mixture of partially-locked and desensitized states; its density is compatible with both enantiomeric forms, likely representing the racemic mixture. Together, these results elucidate detailed interactions of a ρ-type GABAA receptor with therapeutic drugs, offering mechanistic insights and a prospective basis for further drug development.

Results

ρ1 GABAA receptors resolved with antagonist and agonist drugs

We implemented a modified human ρ1 GABAA receptor construct (ρ1-EM) with truncated loops in the N-terminus and intracellular domain and an inserted fluorescent protein to facilitate expression while largely preserving wild-type function27. In Xenopus laevis oocytes expressing ρ1-EM, THIP and CGP36742 antagonized GABA activation (Fig. 1a, b), consistent with previous studies of wild-type channels28,29. GABOB functioned as an agonist, albeit at ~10-fold higher concentrations than GABA (Fig. 1c), consistent with its relatively weaker activity30.

Fig. 1: Functional and structural profiles of ρ1 with antagonist and agonist drugs.
figure 1

a–c Sample traces from two-electrode voltage-clamp electrophysiology recordings of ρ1-EM expressed in Xenopus oocytes exposed to THIP, CGP36742 or GABOB. d Cryo-EM maps of human ρ1-EM in complex with THIP (left), CGP36742 (middle) or GABOB (right), viewed from the extracellular side. Maps are colored by assigned functional states as similar to resting (green), partially locked (purple), or desensitized (blue). Low-pass filtered maps are shown in transparency to reveal lower-resolution features including the nanodisc and M3-helix intracellular-domain extension (M3 ICD). e Cryo-EM maps as in (d), viewed from the membrane plane. Insets show the drugs with corresponding densities.

Full size image

To gain insight into the binding and conformational changes associated with these drugs, we solved cryogenic electron microscopy (cryo-EM) structures of ρ1-EM in complex with THIP, CGP36742 and GABOB in saposin nanodiscs with polar brain lipids to resolutions 2.0–2.4 Å (Fig. 1, Table 1, Supplementary Figs. 2–4). We observed non-protein densities corresponding to the expected drugs in the five extracellular orthosteric ligand-binding sites of each structure; the corresponding maps enabled us to unambiguously build models for both the protein and drugs (Fig. 1, Supplementary Fig. 5).

Table 1 Cryo-EM data collection, refinement and validation statistics
Full size table

Consistent with their functional roles as competitive inhibitors, THIP- and CGP36742-bound structures of ρ1-EM adopted the resting-like state, nearly identical to our previously reported apo structure27 (Table 2). In the presence of the agonist GABOB, one class (13% of resolved particles) corresponded to the desensitized state previously reported with GABA27 (Table 2). In another class (87% of resolved particles), the ECD was not fully activated, with loop C only partially locked over the agonist site; the pore remained at rest. This structure, corresponding to neither resting nor desensitized states, aligned well with a previous so-called primed state determined in the presence of the negative modulator 17β-estradiol (E2)31 (Supplementary Fig. 4e, f, Table 2). Notably, E2 binds in a pocket located at the ECD-TMD interface, where it appears to disrupt allosteric transitions induced by GABA binding, including lockdown of loop C over the agonist site. In contrast, we observed no ligand density at the equivalent interface in the GABOB-bound structures, indicating a distinct mechanism of stabilizing the partially-locked state.

Table 2 Cɑ RMSD (Å) between ρ1-EM structures
Full size table

Distinct poses implicated in subtype-dependent THIP effects

In our cryo-EM structure, THIP was bound in the orthosteric site with its two heterocycles perpendicular to loop C (Fig. 2a). The pyridine ring faced the principal subunit, with its amino group buried in an aromatic cage involving residues Y219, Y262 and Y268. The isoxazole ring faced the complementary subunit, with its hydroxyl and amino groups making polar interactions with principal residues S264 and T265, as well as complementary residues R125 and S189 (Fig. 2a, b, Supplementary Fig. 6).

Fig. 2: Distinctive binding pose of THIP in the ρ1 GABAA receptor.
figure 2

a Zoom view of the THIP (tan) binding site in ρ1-EM (green). Residues interacting directly with THIP are shown as sticks and colored by heteroatom; residues from the complementary subunit face are denoted (-). Potential hydrogen bonds are indicated as dashed lines. b Schematic of ρ1-EM interactions with THIP, colored as in (a). Hydrogen bonds are indicated as dashed lines, hydrophobic and aromatic interactions as lashes. c, d Superimposition of the ECD of THIP-bound (green) and GABA-bound (PDB ID 8OP9, gray) ρ1-EM structures. For clarity, only two subunits are shown, aligned on the complementary subunit. Ligands are shown as sticks, with carbon atoms of THIP and GABA colored tan and gray respectively. e, f Zoom views as in (d) of the superimposition of the THIP binding site in ρ1-EM (green) and the α4β3δ GABAA receptor (PDB ID 7QND, gray), focusing on the β3-ɑ4 or δ-β3 interfaces, respectively. Structures are aligned on the ECD of the complementary subunit. Ligands are shown as sticks, with THIP carbon atoms in ρ1 and α4β3δ subtypes colored tan and gray respectively.

Full size image

Superposing the resting-like THIP complex with our previous GABA-bound desensitized structure27 (Fig. 2c, d) highlighted structural changes induced by antagonist versus agonist binding. The amino and hydroxyl groups of THIP roughly overlapped those of GABA; however, the bulky heterocycles of THIP were relatively protruded toward loop C, obstructing its lockdown over the antagonist. Moreover, computational docking of THIP into these two ρ1 structures produced more favorable binding energy scores in the resting-like versus desensitized state (Supplementary Fig. 7a, b). Thus, similar to TPMPA27, THIP appears to sterically occlude local activating transitions in the ρ-type orthosteric site.

Whereas THIP is a competitive antagonist of ρ132 (Fig. 1a), it is a partial agonist of α1β3γ2 GABAA receptors associated with the postsynapse, and a super-agonist of α4β3δ GABAA receptors found extrasynaptically33,34,35. Consistent with this behavior, a recent cryo-EM structure of the α4β3δ subtype was reported with THIP in an apparent desensitized state, with ligands at both β3(+)-α4(-) and δ(+)-β3(-) interfaces36. At both these interface types, THIP adopts a pose distinct from that in the ρ1 resting state, with the heterocycles rotated nearly 90° to lie parallel to the plane of loop C (Fig. 2e, f). This THIP pose in α4β3δ GABAA receptors is compatible with more extensive lockdown of loop C than in ρ1, apparently enabling activation. We previously observed that loop C is extended by one residue in ρ1 relative to β GABAA-receptor subunits, and that truncating the amino acid at the tip of loop C (ΔS264) results in functional properties more similar to postsynaptic subtypes27. THIP effects were similarly decreased in the ΔS264 variant, but remained inhibitory (Supplementary Fig. 8), indicating that factors other than loop C length are involved in subtype-specific modulation. Although the structural basis for the distinct binding pose is not entirely clear, several bulky groups in the ρ1 orthosteric site are substituted with smaller sidechains in β3, including Y262 (β3-F200) on the principal subunit and R125 (β3-Q64) and S189 (β3-G127) on the complementary subunit (Supplementary Fig. 6).

Structural basis for specificity among phosphinic acids

In our cryo-EM structure, CGP36742 adopted an elongated pose, with its aminopropyl tail forming a salt bridge with E217 and cation-π interactions with the aromatic cage on the principal subunit face. The central phosphinic acid group was wedged between loop C on the principal subunit (making polar interactions with S264 and T265) and residue R125 on the complementary subunit, with the butyl tail extending toward the complementary β5 strand (Fig. 3a, b). The complex was comparable to the apo and THIP structures (Table 2, Fig. 3c), and superimposable with our previous structure with TPMPA (Cα RMSD 0.035 Å) (Fig. 3d), including analogous interactions of the amino and phosphinic acid groups; the additional butyl tail only required an alternative rotamer of M177 in the β5 strand (Fig. 3c, d). As in the case of TPMPA, the bulky, electronegative phosphinic acid group appeared to prevent lockdown of loop C over the antagonist, indicating a shared mechanism as well as binding pose among phosphinic acid inhibitors (Fig. 3e, f).

Fig. 3: Common mechanism of ρ1 GABAA receptor antagonism by phosphinic acid inhibitors.
figure 3

a Zoom view of CGP36742 (tan) binding site in ρ1-EM (green), colored and labeled as in Fig.2a. b Schematic of ρ1-EM interactions with CGP36742. Hydrogen bonds and other electrostatic interactions are indicated as dashed lines, hydrophobic and aromatic interactions as lashes. c Zoom view of superimposed structures of ρ1-EM with CGP36742 (green) and THIP (gray). d Zoom view of superimposed structures of ρ1-EM with CGP36742 (green) and TPMPA (PDB ID 8OQ7, gray). e, f Superimposition of the ECD of CGP36742-bound (green) and GABA-bound (PDB ID 8OP9, gray) ρ1-EM structures. For clarity, only two subunits are shown, aligned on the complementary subunit.

Full size image

Given the ability of CGP36742 to inhibit GABAB as well as ρ1 GABAA receptors29,37, we also compared its potential interactions with these structurally distinct targets. A structure of the related phosphinic acid inhibitor CGP35348 was previously reported with the ECD of a human GABAB receptor, bound in the cleft between ligand-binding lobes (LB1 and LB2) of the GBR1 subunit38 (Fig. 4a–c). The CGP36742 pose from our ρ1-EM structure could be placed in this pocket in the GABAB receptor complex by superposition on its equivalent aminopropyl and phosphinic acid moieties (Fig. 4b), or by computational docking (Supplementary Fig. 7c), with similar poses showing no evident clashes and a favorable binding energy score. Interestingly, the amino group of CGP36742 was oriented roughly opposite to that of CGP35348 in its resolved site (Fig. 4d), indicating the rotational flexibility of these compounds could support their polymodal activities.

Fig. 4: Comparative interactions of GABAB and GABAA receptors with phosphinic acid inhibitors.
figure 4

a Chemical structure of CGP35348 (above) and its binding site in a human GABAB receptor (PDB ID 4MR8, green, below). The ligand (gray) and its direct residue contacts are colored by heteroatom. b Chemical structure of CGP36742 (above), and its superimposition into the equivalent GABAB-receptor site as in (a). The ligand pose is adopted from the structure reported here with ρ1-EM, aligned on the shared aminopropyl and phosphinic acid moieties of CGP35348. c Overview of the ECD of the human GABAB receptor as shown in (a). d Alignment of CGP36742 in ρ1-EM (tan) with CGP35348 in the GABAB receptor (gray) as implemented in b, colored by heteroatom. e Alignment of (S)−4-ACPBPA (tan) into the CGP36742 site of ρ1-EM. For perspective, principal and complementary subunits of ρ1-EM are colored purple and pink respectively; the ligand and interacting residues are colored by heteroatom. f Alignment of (S)−4-ACPBPA (tan) into the CGP35348 of the GABAB receptor site (green) shown in a. Dashed circle indicates prospective clash with residue W65.

Full size image

The most potent and selective ρ1 antagonist identified thus far, (4-aminocyclopenten-1-yl)-butylphosphinic acid ((S)−4-ACPBPA), shares the phosphinic acid and butyl groups of CGP36742 but has a conformationally restricted aminocyclopentenyl group in place of the flexible aminopropyl tail39,40 (Supplementary Fig. 1). Superposition of (S)−4-ACPBPA with CGP36742 in our ρ1-EM complex showed this ligand could be accommodated without modification (Fig. 4e). Conversely, superposing this compound with CGP35348 in the GABAB receptor resulted in a clash of the amino group with residue W65, suggesting a molecular basis for receptor specificity (Fig. 4f). Consistent with these predictions, computational docking of (S)−4ACPBPA produced more favorable binding energy scores in ρ1-EM than in the GABAB receptor (Supplementary Fig. 7d, e).

Multiple states captured with the weak racemic agonist GABOB

As described above, our cryo-EM dataset collected with GABOB contained particles in both partially-locked and desensitized states. The partially-locked state exhibited only a partial lockdown of loop C over the ligand, corresponding to a limited (1.8 Å) S264-Cα translation, and a subtle (1.2°) domain rotation relative to the resting state (Fig. 5a). Although the contribution of this partially-locked state to the receptor gating cycle remains unclear, the TMD was superimposable with that of the resting state, consistent with it representing a pre-active intermediate between resting and open. The presence of a substantial partially-locked class with GABOB may reflect the relatively low affinity and slow kinetics of this agonist, despite the application of a supersaturating concentration (4 mM) for >30 min prior to grid freezing.

Fig. 5: Partially-locked and desensitized states captured with GABOB.
figure 5

a Superimposed structures of the GABOB-bound partially-locked state (purple) with a previously reported resting state (green, PDB 8OQ6). Structures are aligned on the TMD; for clarity, all but one of the subunits are semi-transparent. Labeled distance in the zoomed figure is between S264 Cα positions in the two structures. The S264 side chain is shown as sticks, including polar hydrogens, with oxygen and hydrogen colored red and gray respectively. b Superimposed structures of GABOB-bound partially-locked state (purple) and desensitization state (blue). The structures are aligned with the TMD domain, and for clarity all but one of the subunits are transparent.

Full size image

Consistent with our previous structure with GABA27, the desensitized state with GABOB exhibited further lockdown of loop C (4.3 Å S264-Cα translation) and rotation between the ECD and TMD (7.3°) relative to the partially-locked state (Fig. 5b). Interestingly, although all structures in this work contained ligands in the extracellular orthosteric site, local resolution in the ECD was relatively higher in the GABOB-desensitized state; in resting and partially-locked structures, resolution was similar between the domains (Supplementary Fig. 4a–d), suggesting that channel activation is associated with both stabilization of the ECD and mobilization of the TMD.

Densities for GABOB were clearly resolved in both the partially-locked and desensitized states (Fig. 6a, b). Although the C3 carbon of GABOB constitutes a stereocenter, both enantiomers are ρ1 agonists, with modestly greater potency for (R)-GABOB30. As in medical practice, we used a racemic mixture in our experiments, such that our cryo-EM densities likely contained both forms; indeed, either (R)- or (S)-GABOB could be modeled into the ligand density in either structure (Fig. 6a, b). The amino end was coordinated by an aromatic cage (Y219, Y262, Y268) as well as by E217 in the principal subunit, while the carboxylate end made electrostatic contacts with R125 and S189 in the complementary subunit (Fig. 6c–f). In all cases, T265 at the tip of loop C could make a hydrogen bond with the C3 hydroxyl, consistent with the demonstrated effect of this residue on potency of both GABOB forms41.

Fig. 6: Comparable accommodation of GABOB enantiomers in the orthosteric site.
figure 6

a Refinement of (R)- (tan) and (S)-GABOB (cyan) into ligand density in the partially-locked state of ρ1-EM (purple), colored by heteroatom. b Refinement of (R)- and (S)-GABOB into ligand density in the desensitized state of ρ1-EM (blue), otherwise colored as in (a). c Protein-ligand interactions of (R)- and (S)-GABOB in the partially-locked state of ρ1-EM, colored as in (a). d Protein-ligand interactions of (R)- and (S)-GABOB in the desensitized state of ρ1-EM, colored as in (b). e Schematics of (R)- (left, tan) and (S)-GABOB (right, cyan) interactions with the partially-locked state of ρ1-EM. f Schematics of (R)- (left, tan) and (S)-GABOB (right, cyan) interactions with the desensitized state of ρ1-EM. In (e, f), hydrogen bonds and other electrostatic interactions are indicated as dashed lines, hydrophobic and aromatic interactions as lashes. g Mobility of (R)- and (S)-GABOB in MD simulations of ρ1-EM in the partially-locked (left, purple) and desensitized states (right, blue), as quantified by ligand RMSD (Å). Violin plots represent probability densities from 4 independent simulation replicates, sampled every 0.4 ns for the first 400 ns of each replicate (n = 4000), with markers indicating median and extrema. h Zoomed view of the orthosteric ligand site in resting (green), partially-locked (purple) and desensitized states (blue). Labels indicate the distance between nearest heavy atoms of Y268 and R179(-), where an apparent hydrogen bond characterizes ligand-bound versus -unbound states. i Distance between the Y268 hydroxyl and R179(-) guanidinium groups in MD simulations of ρ1-EM with (R)- and (S)-GABOB in the partially-locked (left, purple) and desensitized states (right, blue). Violin plots represent probability densities from 4 independent simulation replicates, sampled every 0.4 ns for the first 400 ns of each replicate (n = 4000), with markers indicating median and extrema. Dashed lines indicate static distances in the resting (green) and desensitized (blue) structures.

Full size image

To substantiate binding capacity for both entantiomers, we performed all-atom molecular dynamics (MD) simulations of both the partially-locked and desensitized-state structures with both (R)- and (S)-GABOB. In ≥400-ns simulations of each system, performed in quadruplicate with different initial velocities, both (R)- and (S)-GABOB remained within 2 Å median root-mean-square deviation (RMSD) relative to their starting poses in both structures (Fig. 6g, Supplementary Fig. 9), consistent with weak enantiomeric specificity on the timescale of atomistic simulations. Previously we have also observed an intersubunit hydrogen bond between residue Y268 on the principal loop C and R179(-) on the complementary loop F, which characterizes ligand-bound versus -unbound states27 (Fig. 6h). Similar to our previous simulations with GABA, either of the GABOB enantiomers sustained tighter Y268-R179 interactions than in the resting structure, though a modest trend towards even tighter contacts in the presence of (R)-GABOB appeared consistent with the slightly greater potency of this enantiomer (Fig. 6i). Interestingly, this interaction was retained in the partially-locked as well as desensitized systems, despite the limited extent of ECD activation.

Discussion

Pharmaceutical targeting of ρ-type GABAA receptors holds promise for drug development in treating visual, sleep, learning and memory disorders13,42. Given the limited sensitivity of this subtype to classical GABAA receptor modulators such as benzodiazepines, barbiturates and general anesthetics, the development of such agents will likely require a detailed understanding of ρ-specific mechanisms including binding, activation and inhibition. However, the specific mechanisms also means such drugs could limit interactions on neuronal receptors. The structural, functional and computational work presented here uncovers the binding modes of three drugs in the orthosteric site, highlighting among other things the critical role of loop C lockdown in channel gating.

Our structures highlight the critical role for loop-C lockdown in initiating ρ1 activation. Antagonists such as THIP and phosphinic acids clearly obstruct lockdown of loop C altogether, resulting in an expansive orthosteric site superimposable with that of the apo structure (Fig. 7). Among phosphinic acid inhibitors, lockdown is obstructed by a consistent binding pose of CGP36742 and TPMPA, and likely of the subtype-specific agent (S)−4-ACPBPA. Conversely, cross-reactivity of CGP36742 with GABAB receptors is attributable at least in part to rotational flexibility around the amino group. Despite the shared binding profile among phosphinic acid inhibitors in both GABA-receptor families, modification of the aminopropyl tail in (S)−4-ACPBPA, or of the butyl tail in CGP35348, successfully confers preference for GABAA and GABAB receptors respectively, indicating that selective modulators can be engineered on this scaffold.

Fig. 7: Proposed mechanisms of antagonist and agonist drugs.
figure 7

a Cartoons of ρ1 showing progressive ECD rotation between resting (green), partially-locked (purple) and activated (open or desensitized, blue) states. Loop C is highlighted, and GABA and ions are shown in circles (carbon, tan; amine, blue; carboxylate, red; chloride, green). b Cartoon zoom view of the orthosteric binding site. GABA is shown in circles as in (a). Green and blue lines depict lockdown of loop C. c Proposed dependence of orthosteric-site expansion/compaction on ligand identity, from antagonists to strong agonists.

Full size image

At the other extreme, agonists such as GABA enable substantial lockdown of loop C, resulting in compaction of the orthosteric site and rotation of the ECD relative to the TMD (Fig. 7a, b). Alongside this apparent activated-desensitized state, treatment with the weaker agonist GABOB promoted a subclass exhibiting only partial lockdown of loop C relative to the apo form (Fig. 7a, c). We previously reported a similar partially-locked structure in the presence of the inhibitor E2, which might allosterically trap the so-called primed state, or some metastable intermediate on the pathway from resting to open31. In the absence of allosteric inhibition, GABOB appears to favor a partially-locked population by different means. Given its low potency and slow kinetics relative to GABA, GABOB occupancy may be insufficient to stimulate complete rotation/activation of the ECD in a subset of ρ1 particles. It is also plausible that non-physiological sample conditions, such as embedding in a lipid nanodisc or non-instantaneous plunge-freezing, may relatively destabilize the GABOB-activated state. Although agonist binding to at least 3 of 5 subunit interfaces is thought to enable ρ1 activation43, the limited diffusive volume and high local receptor density on the cryo-EM grid may limit ligand accessibility, even in the presence of a supersaturating bulk agonist concentration. Indeed, it remains unclear why a fully activated-open structure of ρ1 remains experimentally inaccessible with either GABA or GABOB27,31.

Subtype- and agent-specific interactions, e.g. with T265 on loop C, suggest avenues for future structure-based drug design. For instance, we observed a distinctive binding pose for THIP in ρ1-EM compared to a previously reported complex with an α4β3δ GABAA receptor36, which may underlie its opposing effects in these subtypes. Interestingly, the THIP derivative aza-THIP is a more selective ρ-type antagonist, with activity comparable to THIP at ρ1 but negligible at heteromeric GABAA receptors28. The two molecules are identical except at one heavy atom in the 5-membered ring, substituting a pyrazole in aza-THIP for the isoxazolo group in THIP (Supplementary Fig. 1a). In structures with THIP, the substituted oxygen atom appears to accept a hydrogen bond from principal-subunit loop C (T265) in ρ1-EM, but from the complementary subunit (ɑ4-T163 or β3-Q64) in the α4β3δ GABAA receptor. The differential environments for this substituted atom in particular may account for discriminating activity between these two receptors. A hydrogen bond with loop-C T265 also appears to underlie potency of both enantiomeric forms of GABOB in ρ1, and may contribute to the stabilization of a partially-locked state in the presence of this weaker agonist relative to GABA. Taken together, this work details receptor-specific binding interactions of both antagonists and agonists in the orthosteric site, offering potential insights into differential pharmacology across multiple receptor subtypes in the GABAergic system.

Methods

Protein purification from mammalian cells

The expression and purification of ρ1-EM was following our earlier published methods27. Briefly, cell pellets from 2 L of Expi293F cells infected with baculovirus were resuspended in the buffer (40 mM HEPES pH 7.5, 300 mM NaCl, with cOmplete protease inhibitor tablets (Roche)) and sonicated to break cell membranes. The membrane was pelleted by ultracentrifugation then resuspended and solubilized by resuspension buffer with 2% lauryl maltose neopentyl glycol (LMNG), 0.2% cholesteryl hemisuccinate (CHS) for 3 h in the cold room. The solubilization mixture was ultracentrifuged and the supernatant was applied to 4 mL Strep-Tactin XT Superflow resin (IBA) and incubated for 90 min. Resin was washed with wash buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 0.005% LMNG, 0.0005% CHS) then protein was eluted with elution buffer (wash buffer with 10 mM d-Desthiobiotin (Sigma)). The product was further purified by size exclusion chromatography on a Superose 6 column (Cytiva) with flow buffer (20 mM HEPES pH 7.5, 100 mM NaCl, 0.005% LMNG, 0.0005% CHS). Peak fractions were pooled for nanodisc reconstitution.

Nanodisc reconstitution

The plasmid for SapA expression was a gift from Salipro Biotech AB. Purification of SapA followed the published protocols44. For the reconstitution of nanodisc, ρ1-EM, SapA and polar brain lipid (Avanti) were mixed as molar ratio 1:15:150, then incubated on ice for 1 h. Bio-Beads SM-2 resin (Bio-Rad) was added into the mixture then gently rotated overnight at 4 °C. On the next day, the supernatant was collected and further purified by gel-filtration chromatography on a Superose 6 column (Cytiva) with buffer containing 20 mM HEPES pH 7.5, 100 mM NaCl. Peak fractions were pooled and concentrated to ~5 mg/mL.

Cryo-EM grid preparation and data collection

The nanodisc sample was mixed with the compound stock solutions with volume ratio 9:1. The stock solutions were (5 mM THIP, 20 mM fluorinated foscholine 8 (FFC-8)), (20 mM CGP36742, 20 mM FFC-8), (40 mM GABOB, 20 mM FFC-8). The mixtures were incubated on ice for more than 30 minutes before freezing grids. For each grid, 3 μL of the mixture was applied to a glow-discharged grid (R1.2/1.3 300 mesh Au grid, Quantifoil), blotted for 2 s with force 0 and plunged into liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific). Cryo-EM data were collected on a 300 kV Titan Krios (Thermo Fisher Scientific) electron microscope with a K3 Summit detector (Gatan) with magnification 130k corresponding to 0.6725 or 0.6645 Å/px using the software EPU 3.5.0 (Thermo Fisher Scientific). The total dose was ~46 e−/Å2 and defocus range was −0.8 to −1.8 μm.

Cryo-EM data processing

Dose-fractionated images in super-resolution mode were internally gain-normalized and binned by 2 in EPU during data collection. Cryo-EM data processing was first done in Relion 3.1.445, including Motion correction, contrast transfer functions (CTF) estimation with CTFFIND 4.146, automatic particle picking with topaz 0.2.547, particle extraction, 2D classification, 3D classification, 3D refinement, CTF refinement and polishing. Briefly, two rounds of 2D classification were done to remove junk particles, 3D classification (3 classes) was used to analyze the structural heterogeneity. Particles from classes with protein features were centered and re-extracted, and were used for the 3D refinement with symmetry C5. Several rounds of CtfRefine and one round of polishing were executed to improve the resolution. The shiny particles were imported into CryoSPARC v4.2.1 for further processing48, including 3D classification with the PCA mode, and Non-Uniform Refinement49.

Model building and refinement

Model building was started with rigid body fitting of the previously published resting (PDB ID 8OQ6), primed (PDB ID 8RH7) or desensitized (PDB ID 8RH8) state structure into the density. The models were manually checked and adjusted in Coot 0.9.550, and chemicals, waters and lipids were also manually added. The resulting model was further optimized using real-space refinement in PHENIX 1.18.251 and validated by MolProbity52. Crystallographic information files (cif) for ligands were generated from isomeric SMILES strings using Grade253. For initial ligand comparisons across receptor families, ligands (CGP36742, (S)−4-ACPBPA) were superposed using the “align” command in UCSF Chimera54 to match corresponding non-hydrogen atoms in the target complex (CGP35348, CGP36742).

Structural analysis

Pore radius profiles were calculated using CHAP 0.9.155. Structure figures were prepared using UCSF ChimeraX 1.356. ECD rotation was calculated as the dihedral angle between a) the Cα COM of the ECD (residues 97-280) of one subunit, b) the equivalent ECD residues of all subunits, c) the TMD (residues 281-479) including all subunits, and d) the equivalent TMD residues of one subunit. RMSDs were calculated by aligning Cα atoms in two given structures using the “match” command in UCSF Chimera54. Docking was performed using Autodock Vina57, with search volumes 15 Å * 15 Å * 15 Å around each pocket.

Expression in oocytes and electrophysiology

mRNA encoding the ρ1-EM GABAA receptor was produced by in-vitro transcription using the mMessage mMachine T7 Ultra transcription kit (Ambion) according to the manufacturer protocol. Xenopus laevis oocytes (Ecocyte Bioscience) were injected with 30–50 ng mRNA and incubated 4–8 days at 13 °C in post-injection solution (10 mM HEPES pH 8.5, 88 mM NaCl, 2.4 mM NaHCO3, 1 mM KCl, 0.91 mM CaCl2, 0.82 mM MgSO4, 0.33 mM Ca(NO3)2, 2 mM sodium pyruvate, 0.5 mM theophylline, 0.1 mM gentamicin, 17 mM streptomycin, 10,000 u/L penicillin) prior to two-electrode voltage clamp measurements.

For recordings, glass electrodes were pulled and filled with 3 M KCl to give a resistance of 0.5–1.5 MΩ and used to clamp the membrane potential of injected oocytes at −60 mV with an OC-725C voltage clamp (Warner Instruments). Oocytes were maintained under continuous perfusion with Ringer’s solution (123 mM NaCl, 10 mM HEPES, 2 mM KCl, 2 mM MgSO4, 2 mM CaCl2, pH 7.5) at a flow rate around 1.5 mL/min. Buffer exchange was accomplished by manually switching the inlet of the perfusion system to the appropriate buffer. For assessing GABOB efficacy, a gravity-fed perfusion system was used to improve kinetics of solution exchange. Currents were digitized at a sampling rate of 2 kHz and lowpass filtered at 10 Hz with an Axon CNS 1440 A Digidata system controlled by pCLAMP 10 (Molecular Devices).

Molecular dynamics simulations

All-atom simulations in explicit solvent were deemed most appropriate to assess steady-state dynamics, given the relatively high precision and accuracy of atomistic interactions that can be captured compared to e.g. coarse-grained methods. Atomic coordinates of the ρ1-GABOB complex with GABOB built as either of two enantiomers were used as starting models for MD simulations. Each subunit was split into two chains for simulation, due to unresolved residues in the M3-M4 loop in the experimental structure. This approach carries the risk of enabling non-physiological dynamics around deletion endpoints; on the other hand, it avoids introducing information without direct experimental evidence, for example by modeling the missing loop, inserting a shorter sequence, or applying positional restraints. Given that the sub-microsecond timescales of our simulations are not expected to sample allosteric effects of the intracellular domain on drug-binding sites, which are located at least 70 Å away on the opposite side of the membrane, and that overall Cɑ RMSDs remained within 4 Å (Supplementary Fig. 9), this approach appeared to be a reasonable and parsimonious compromise. The simulation systems were set up in CHARMM-GUI58. The protein was embedded into a bilayer mimicking brain-lipid composition with the top leaflet containing 155 POPC, 24 POPE, and 38 cholesterol molecules and the bottom leaflet containing 65 POPC, 115 POPE, 26 POPS, and 32 cholesterol molecules. The protein-lipid complex was subsequently solvated with TIP3P water and 150 mM NaCl (Supplementary Table 1). The CHARMM36m forcefield59 was used to describe the protein. Parameters for (R)- and (S)-GABOB were generated using CGenFF60 in CHARMM-GUI. Visual and quantitative inspection of the resulting parameters (param penalty 4.600, charge penalty 10.008) indicated reasonable description of both ligands. Cation-π specific NBFIX parameters were used to maintain appropriate ligand-protein interactions in the aromatic cage in the orthosteric binding site61.

Simulations were performed using GROMACS 2022.562 at 300 K and 1 bar using the velocity-rescaling thermostat63 and Parrinello–Rahman barostat64. The LINCS algorithm was used to constrain the length of all bonds involving hydrogens65, and the particle mesh Ewald method66 was used to calculate long-range electrostatic interactions. The systems were energy minimized and then equilibrated for 20 ns, with the position restraints on the protein and neurosteroids gradually released. Four replicates, each >400 ns, were simulated for each system as final unrestrained production runs. Before analysis, MD simulation trajectories were aligned on the Cα atoms of the ECD using MDAnalysis67. Root-mean-square deviations (RMSD) of the non-hydrogen atoms of GABOB were calculated in VMD68, and data from the first 400 ns were combined and plotted as violin plots using Matplotlib69.

Reporting summary

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