The endogenous NS, 3α5αP is known to potentiate GABA-elicited currents (Figure 1A) and enhance [³H]muscimol binding to α₁β₃ GABA_ARs (Figure 1E; Chen et al., 2019; Harrison et al., 1987a). We examined a series of NS analogues with different stereochemistries or substituents in the 3- and 17-positions: 3β5αP, KK148, and KK150 (structures shown in Figure 1B–D) for their ability to potentiate GABA-elicited currents and enhance orthosteric agonist ([³H]muscimol) binding. 3β5αP is the 3β-epimer of 3α5αP. KK148 and KK150 are NS analogue photolabeling reagents, which have a 3-diazirinyl moiety instead of the 3-OH, and differ from each other by the stereochemistry of the 17-ether linkage (Jiang et al., 2016). We observed a discrepancy between the ability of these compounds to potentiate GABA-elicited currents and their ability to enhance [³H]muscimol binding in α₁β₃ GABA_ARs. None of the NS analogues lacking a 3α-OH potentiated GABA-elicited currents (Figure 1B–D). However, both 3β5αP and KK148 significantly enhanced [³H]muscimol binding (Figure 1E). KK150, in contrast, did not potentiate GABA-elicited currents and minimally enhanced [³H]muscimol binding (Figure 1D–E). Collectively, these data show that, NS analogues with different stereochemistry or substituents at the 3- and 17-positions show distinct patterns in modulation of α₁β₃ GABA_AR currents and orthosteric ligand binding. We hypothesized that these patterns are a consequence of the various NS analogues stabilizing distinct conformational states of the GABA_AR, possibly by binding and acting at different sites. Notably, the compounds with a 3-OH (3α5αP, 3β5αP) are 10-fold more potent than those with a 3-diazirine (KK148, KK150) in enhancing [³H]muscimol binding (Figure 1E), suggesting that the 3-OH is an important determinant of binding affinity to the site(s) mediating these effects.

To determine why 3β5αP and KK148 enhance [³H]muscimol binding but do not potentiate α₁β₃ GABA_AR currents, we first considered the possibility that 3β5αP- and KK148-induced enhancement of [³H]muscimol binding is a selective effect on intracellular GABA_ARs, since the radioligand binding assay was performed on total membrane homogenates, whereas the electrophysiological assays report only from cell surface channels. NS are known to have effects on intracellular GABA_ARs and have been shown to accelerate GABA_AR trafficking (Abramian et al., 2014; Comenencia-Ortiz et al., 2014; Smith et al., 2007). To test this possibility, we examined [³H]muscimol binding in intact cells (i.e. binding to receptors only in the plasma membrane) (Vauquelin et al., 2015; Bylund et al., 2004; Bylund and Toews, 1993) compared to permeabilized cells (plasma membranes plus intracellular membranes). Notably, [³H]muscimol binding was twofold greater in permeabilized cells than in intact cells, indicating a significant population of intracellular GABA_ARs. KK148 enhanced [³H]muscimol binding in intact cells as much or more than in permeabilized cells, indicating that this effect is not a result of selective NS actions on intracellular receptors (Figure 1—figure supplement 1).

A second possibility is that 3β5αP and KK148 selectively bind to and stabilize a high-affinity non-conducting state, such as a pre-active (Gielen and Corringer, 2018) or a desensitized conformation of the GABA_AR. This is expected to result in inhibition of receptor function; however, the magnitude of the effect may be small under the experimental conditions used to generate the traces in Figure 1. To examine the inhibitory effect of these NS analogues, we activated α₁β₃ GABA_AR with a saturating concentration (1 mM) of GABA and tested the effect of the NS on steady-state currents (Germann et al., 2019a). KK148 and 3β5αP both decreased steady-state currents (Figure 2A and C), whereas KK150 did not (Figure 2B). To further delineate the electrophysiological effects of these compounds, we focused on 3β5αP, since it is an endogenous NS and we had limited availability of KK148. Co-application of 3β5αP with 1 mM GABA preferentially inhibited steady-state rather than peak currents (Figure 2—figure supplement 1). While this result is consistent with stabilization of a desensitized state rather than a pre-active state, it is ambiguous because it is possible that the steroid has a slower onset than GABA, thus minimizing the effect on peak current. Additional evidence that 3β-NAM-NS stabilize a desensitized state includes studies examining their effects on inhibitory post-synaptic currents (Wang et al., 2002) and single channel currents (Akk et al., 2001). The evidence that NAM-NS stabilize a desensitized rather than a pre-active state is more thoroughly explored in the Discussion. In the ensuing text, we refer to the inhibition of steady-state current as desensitization.

The inhibitory effect of 3β5αP was not observed in receptors with the α₁(V256S) TM2 pore-lining mutation, which was previously shown to remove the inhibitory effects of sulfated steroids (Akk et al., 2001; Wang et al., 2002; Figure 2D). Although both 3α5αP and 3β5αP enhance [³H]muscimol binding, the former predominantly results in receptor activation, whereas the latter results in inhibition. Consistent with this, the α₁(V256S)β₃ mutation which abolishes NS-induced inhibition (Akk et al., 2001; Wang et al., 2002) eliminated [³H]muscimol binding enhancement by 3β5αP but not 3α5αP (Figure 3). We infer that 3α5αP increases [³H]muscimol binding by stabilizing an active state of the receptor. In contrast, 3β5αP increases [³H]muscimol binding by stabilizing a desensitized state of the receptor; this effect is eliminated in the α₁(V256S)β₃ receptor. The mechanisms of enhancement of [³H]muscimol binding by allosteric activators and inhibitors are described in detail in our recent publication (Akk et al., 2020). Collectively, these data indicate that 3β5αP and KK148 enhance orthosteric ligand affinity by stabilizing a desensitized state of the GABA_AR.

While there is qualitative agreement between the relative effects of the various NS analogues on orthosteric ligand binding and receptor desensitization, there is a quantitative discrepancy in the magnitude of the effects. For example, 3β5αP enhances [³H]muscimol binding by two-fold (Figure 1E), whereas it reduces steady-state current by only ~25% (Figure 2C). To address this difference, we considered that the radioligand binding and electrophysiological assays are performed under different experimental conditions. The radioligand-binding studies are performed using low [³H]muscimol concentrations to allow for sufficient dynamic range of ligand binding. In contrast, the desensitization experiments are performed at high orthosteric ligand (GABA) concentration to achieve high peak open probability and steady-state receptor desensitization, thus minimizing the number of channels in the resting state. To address the quantitative differences in results from the two assays, we analyzed the electrophysiological data in the framework of the three-state Resting-Open-Desensitized model (Germann et al., 2019a; Germann et al., 2019b). We assumed that both the open and desensitized states had higher affinity for muscimol than the resting state, and that the affinities were similar and could be treated as equal. We then calculated the predicted occupancy of the high-affinity states (P_open + P_desensitized) using parameters derived from the functional responses, to compare to the observed changes in binding. The raw current amplitudes of peak and steady-state responses were converted to units of open probability as described previously in detail (Eaton et al., 2016), and the probabilities of being in the open (P_open) or desensitized (P_desensitized) states were calculated for different experimental conditions (see Materials and methods).

Application of 1 mM GABA elicited a current response that had a peak P_open of 0.71 ± 0.25 (mean ± SD; n = 16). The P_open of the steady-state response was 0.121 ± 0.033 (n = 7), that was reduced to 0.077 ± 0.013 (n = 5) with 3 µM 3β5αP. Analysis of steady-state currents using the Resting-Open-Desensitized model indicates that the steady-state P_desensitized is 0.829 in the presence of GABA, and 0.892 in the presence of GABA + steroid. The relatively small increase in the sum of (P_open + P_desensitized) (from 0.95 to 0.97) is due to the use of saturating GABA in these experiments.

To compare the data from the radioligand binding and electrophysiology experiments, we exposed oocytes containing α₁β₃ GABA_ARs to 20 nM muscimol and recorded currents before and after co-application of 3 μM 3β5αP. The percent reduction in steady-state current following 3β5αP exposure was measured and used to estimate the relative probabilities of resting, open and desensitized receptors. The application of 20 nM muscimol elicited a peak response with P_open of 0.012 ± 0.004 (n = 6). The steady-state P_open was 0.011 ± 0.004. In the same cells, subsequent exposure to 3 µM 3β5αP reduced the steady-state P_open to 0.009 ± 0.004 (p=0.0174; paired t-test). The calculated steady-state P_desensitized was 0.1001 in the presence of muscimol, and 0.2168 in the presence of muscimol + 3β5αP. Thus, there is a predicted two-fold increase in the sum of (P_open + P_desensitized) when the steroid is combined with muscimol, consistent with the doubling of muscimol binding caused by 3β5αP in the [³H]muscimol binding experiments (Figure 1E). While the measured changes in current are small, they are precise because each experiment served as its own control; a steady-state current was achieved during continuous agonist administration and the response to 3β5αP was then measured. Overall, these data indicate that when P_resting is high (low orthosteric ligand concentration), an agent that stabilizes desensitized receptors may produce a small decrease in steady-state current, but a relatively large increase in the occupancy of desensitized state, at the expense of resting receptors. Conversely, with high orthosteric ligand concentrations (low P_resting), a desensitizing ligand produces a relatively larger change in steady-state current as open receptors are converted to desensitized receptors with minimal effect on the sum occupancy of high-affinity states.

To determine whether KK148 and 3β5αP stabilize a desensitized conformation of the GABA_AR by selectively binding to one or more of the identified NS-binding sites on the GABA_AR (Chen et al., 2019), we first determined which of the identified NS sites they bind. We have previously shown that the 3α5αP-analogue photolabeling reagent, KK200 labels the β₃(+)–α₁(-) intersubunit (β₃G308) and α₁ intrasubunit (α₁N408) sites on α₁β₃ GABA_ARs (Figure 4A), and that photolabeling can be prevented by a 10-fold excess of 3α5αP (Chen et al., 2019). As a first step to determine the binding sites for 3β5αP, KK148 or KK150, we examined whether a 10-fold excess of these compounds (30 μM) prevented KK200 (3 μM) photolabeling of either binding site. Photolabeling was performed on membranes from HEK293 cells transfected with epitope-tagged α_1His-FLAGβ₃ receptors, mimicking the conditions used in the [³H]muscimol binding assays and photolabeled residues were identified and labeling efficiency was determined using middle-down mass spectrometry (Chen et al., 2019). KK148, KK150, 3α5αP and 3β5αP all prevented KK200 photolabeling of α₁N408 in the α₁ intrasubunit site (Figure 4B), consistent with their binding to this site. In contrast, KK148, KK150 and 3α5αP but not 3β5αP prevented labeling of β₃G308 in the intersubunit site (Figure 4C), indicating that 3β5αP does not bind to the intersubunit site. Similarly, 3β5αP did not prevent labeling of the intersubunit site by a similar NS-analogue photolabeling reagent in detergent-solubilized GABA_ARs (Jayakar et al., 2020).

The KK148- and KK150-photolabeled (30 μM) samples were also analyzed to directly identify the sites of adduction. In both the KK148- and KK150-labeled samples, photolabeled peptides were identified from the TM4 helices of both the α₁ and β₃ subunits. The labeled peptides had longer chromatographic elution times than the corresponding unlabeled peptides and corresponded with high mass accuracy (<20 ppm) to the predicted mass of the unlabeled peptides plus the add weight minus N₂ of KK148 or KK150 (Figure 4—figure supplement 1). Product ion (MS2) spectra of the KK148- and KK150-labeled peptides from the α₁ subunit identified the labeled residue as Y415 for both KK148 and KK150 with photolabeling efficiencies of 0.77% and 0.62%, respectively (Figure 4D–E, Figure 4—figure supplement 2); Y415 is the same residue labeled by KK123 at the α₁ intrasubunit site (Chen et al., 2019). The KK148 and KK150 labeled peptides in TM4 of the β₃ subunit and corresponding unlabeled peptide were identified by fragmentation spectra as β₃TM4 I426-N445. These data support labeling of the β₃ intrasubunit site by KK148 and KK150. Fragmentation spectra of the peptide-sterol adducts were not adequate to determine the precise labeled residue because of low photolabeling efficiency (0.13% for KK148; 0.19% for KK150, Figure 4—figure supplement 1). No photolabeled peptides were identified in the β₃(+)–α₁(-) intersubunit site. This is likely because KK148 and KK150, similar to KK123, utilize an aliphatic diazirine that preferentially labels nucleophilic residues (Sugasawa et al., 2019; Budelier et al., 2017; Das, 2011); such residues are not present in the intersubunit site.

We have also shown that KK123 labeling of the α₁ intrasubunit (α₁Y415) and β₃ intrasubunit (β₃Y442) sites (Figure 5A) can be prevented by a 10fold excess of 3α5αP (Chen et al., 2019). We thus examined whether 3β5αP (30 μM) inhibited photolabeling by KK123 (3 μM). 3β5αP completely inhibited KK123 photolabeling at both intrasubunit sites (Figure 5B–C). Collectively, the data show that KK148, KK150 and 3α5αP bind to all three of the identified NS-binding sites. In contrast, 3β5αP selectively binds to the two intrasubunit binding sites, but not to the canonical β₃(+)–α₁(-) intersubunit site.

To determine which of the previously identified binding sites contributes to NS enhancement of [³H]muscimol binding, we performed site-directed mutagenesis of the NS-binding sites previously determined by photolabeling (Figure 6A; Chen et al., 2019). Specifically, α₁(Q242L)β₃ targets the β₃(+)–α₁(-) intersubunit site, α₁(N408A/Y411F)β₃ and α₁(V227W)β₃ the α₁ intrasubunit site, and α₁β₃(Y284F) the β₃ intrasubunit site. None of these mutations produced a significant change in [³H]muscimol K_d (Figure 7B and Figure 7—source data 1). Accordingly, concentration-dependent NS effects were assayed at a fixed concentration of [³H]muscimol (3 nM; ~EC₅). It should be noted that earlier studies showed two-component binding curves for [³H]muscimol in brain membranes, with NS causing an increase in the B_max of the high-affinity component (Harrison and Simmonds, 1984). In contrast, our results with expressed α₁β₃ GABA_ARs show a single-component [³H]muscimol binding curve with NS producing an increase in muscimol affinity. Our results are similar to results reported with expressed α₁β₃γ₂ GABA_ARs, where allosteric modulators increased the affinity of a single-component [³H]muscimol binding curve (Dostalova et al., 2014). Whether the complex [³H]muscimol binding curves observed in brain is the result of heterogeneity of receptor subtypes or multiple states of the GABA_AR is unresolved.

Mutations in the β₃(+)–α₁(-) intersubunit and α₁ intrasubunit sites decreased 3α5αP enhancement of [³H]muscimol binding by ~80%, while mutation of the β₃ intrasubunit site led to a small decrease (Figure 6B, Table 1). The residual enhancement of [³H]muscimol binding observed in receptors with mutations in the intersubunit or α₁ intrasubunit site occurs at 10fold higher concentrations of 3α5αP than wild-type (WT) and receptors with mutations in the β₃ intrasubunit site (Table 1), suggesting that 3α5αP binds to the β₃ intrasubunit site with lower affinity. In contrast, mutations in the α₁ and β₃ intrasubunit sites, but not the intersubunit site decreased the enhancement of [³H]muscimol binding by 3β5αP and KK148 (Figure 6C–D, Table 1). To confirm that the effect of these mutations on NS effect are steroid-specific, we also tested their effect on etomidate, which enhances [³H]muscimol binding in α₁β₃ GABA_ARs and acts through a binding site distinct from NS (Li et al., 2006; Jayakar et al., 2019). The mutations targeting NS-binding sites resulted in modest decreases in [³H]muscimol binding enhancement by etomidate; however, the α₁β₃(M286W) mutation which abolishes etomidate potentiation and activation of GABA_ARs (Stewart et al., 2008; Ziemba et al., 2018), also abolished [³H]muscimol binding enhancement (Figure 6E).

We did not test the effects of mutations on KK150 action because it minimally enhances [³H]muscimol binding. However, KK150 binds to all three of the identified NS-binding sites, and may thus be a weak partial agonist or antagonist at the sites mediating NS enhancement of [³H]muscimol binding. Consistent with this prediction, KK150 inhibited enhancement of [³H]muscimol binding by 3α5αP and KK148 (Figure 8).

Collectively, these results show that multiple NS-binding sites contribute to enhancement of [³H]muscimol affinity and that potentiating NS (3α5αP) and non-potentiating NS (3β5αP, KK148 and KK150) have both common and distinct sites of action. Specifically, 3α5αP enhances [³H]muscimol binding through all three sites but predominantly through the intersubunit and α₁ intrasubunit sites, which we have previously shown mediate PAM-NS potentiation (Chen et al., 2019). In contrast, 3β5αP and KK148 enhance [³H]muscimol binding exclusively through the α₁ and β₃ intrasubunit sites. KK150 antagonizes the effects of KK148 on [³H]muscimol binding, presumably via the intrasubunit sites, and antagonizes the effects of 3α5αP, possibly via all three sites. These data indicate that NS binding to both the intersubunit and intrasubunit sites contributes to 3α5αP enhancement of [³H]muscimol binding, but that only the intrasubunit binding sites contribute to the effects of 3β5αP and KK148. The data (Figure 6B–D) are consistent with NS producing their effects by independent action at each of the binding sites and our interpretation is based on that assumption. We cannot, however, rule out the possibility of an allosteric interaction between the NS binding sites (Chen et al., 2019).

It is important to note that the [³H]muscimol binding curves in Figure 6 are normalized to control. The raw data show that membranes containing WT receptors have 10–20-fold higher [³H]muscimol binding (B_max) than membranes containing α₁(N408A/Y411F)β₃ receptors (Figure 7A), whereas the B_max of membranes containing α₁(Q242L)β₃, α₁β₃(Y284F) or α₁(V227W)β₃ receptors was the same as for WT α₁β₃ GABA_ARs (Figure 7—source data 1). The lower total [³H]muscimol binding observed in α₁(N408A/Y411F)β₃ membranes is likely a consequence of decreased receptor expression. To assure that differences in NS effect between WT and α₁(N408A/Y411F)β₃ are not due to different muscimol affinities, we examined [³H]muscimol binding at a full range of concentrations. The α₁(N408A/Y411F)β₃ mutations did not have a significant effect on [³H]muscimol affinity (Figure 7B), but eliminated the modulatory effects of NS (3α5αP and KK148) on [³H]muscimol affinity (Figure 7C–D and Figure 7—source data 2). To assure that the effect of α₁(N408A/Y411F)β₃ was specific to NS, we also examined the effect of etomidate (a non-steroidal GABA_AR PAM) on muscimol affinity. Etomidate enhanced [³H]muscimol affinity in both the WT and α₁(N408A/Y411F)β₃ receptors, indicating that the effect of these mutations are specific to NS action (Figure 7C–D).

To further explore the relationship between desensitization and enhancement of [³H]muscimol binding, we examined the consequences of mutations to these sites on physiological measurements of desensitization induced by NS. Again, these experiments were performed with 3β5αP because it is the endogenous 3β-OH NS and because of limited quantities of KK148. Desensitization was quantified by defining the baseline steady-state current at 1 mM GABA as 100% and measuring percent reduction of the steady-state current elicited by a NS (Figure 9A inset). While GABA was at a saturating concentration for all receptors, the peak P_open it elicited was <<1.0 for several of the mutated receptors. To normalize P_open, mutated receptors with low peak P_open were activated by co-application of 1 mM GABA with 40 μM pentobarbital (PB) (Steinbach and Akk, 2001) prior to application of 3α5αP. This was done because the magnitude of negative allosteric modulation varies as a function of P_open (Germann et al., 2019a). To ensure that PB did not influence NS negative allosteric modulation, control experiments were performed in WT α₁β₃ GABA_ARs and showed no significant difference in the desensitization elicited by 3β5αP between receptors activated by GABA vs. GABA plus PB. The maximum P_open for the mutant receptors varied between 0.55 and 0.95; some of the NS effects on macroscopic currents may be influenced by these differences.

3β5αP reduced the steady-state current (i.e. enhanced desensitization) by 23.0 ± 2.8% (% of desensitization: mean ± SEM, n = 5, Figure 9A). Mutations in the α₁ and β₃ intrasubunit sites [i.e. α₁(N408A/Y411F)β₃ and α₁β₃(Y284F), respectively] prevented 3β5αP-enhanced desensitization by ~67% (Figure 9B), whereas mutation in the β(+)–α(-) intersubunit site [α₁(Q242L)β₃] was without effect (Figure 9B). Receptors with mutations in both the α₁ and β₃ intrasubunit sites [α₁(N408A/Y411F)β₃(Y284F)] showed less NS-enhancement of desensitization than receptors with mutations in either of the intrasubunit sites alone, indicating that both intrasubunit sites contribute to the desensitizing effect (Figure 9B). Although the desensitizing effect of 3β5αP is completely eliminated by the V2’S mutation α₁(V256S)β₃, it is not completely eliminated by combined mutations of all three binding sites [α₁(Q242L/N408A/Y411F)β₃(Y284F)] (Figure 9B). This suggests either that the effects of the mutations are incomplete or there are additional unidentified NS-binding sites contributing to desensitization. Since mutations of the α₁ and β₃ intrasubunit sites also disrupt 3β5αP-enhancement of [³H]muscimol binding (Figure 6C), we conclude that 3β5αP binding to these intrasubunit sites stabilizes the desensitized state of the GABA_AR and thus enhances [³H]muscimol binding. Furthermore, KK148 increased GABA_AR desensitization (% of desensitization = 27.2 ± 6.0: mean ± SEM, n = 3, Figure 2A) and the α₁(V256S)β₃ mutation abolished the effect (% of desensitization = 0, n = 1). These observations support the idea that binding of certain NS analogues to α₁ and β₃ intrasubunit sites, increases GABA_AR desensitization. In contrast, KK150 showed a very small effect on desensitization (% of desensitization = 2.1 ± 0.7: mean ± SEM, n = 5, Figure 2B), consistent with the small increase in [³H]muscimol binding by KK150 (Figure 1E).

3α5αP binds to all three of the NS-binding sites on α₁β₃ GABA_AR, and mutations in all three sites reduce 3α5αP enhancement of [³H]muscimol binding (Figure 6B). This suggests the possibility that activation by 3α5αP (mediated primarily by the β₃(+)–α₁(-) intersubunit site) masks a desensitizing effect mediated through the β₃ and/or α₁ intrasubunit binding sites. To determine whether intrasubunit binding sites mediate increased desensitization by 3α5αP, we examined the effect of 3α5αP on steady-state currents in receptors with mutations in the α₁ or β₃ intrasubunit site. Mutations in the intrasubunit sites were prepared with a background α₁(Q242L)β₃ mutation to remove 3α5αP activation (Chen et al., 2019; Sugasawa et al., 2019; Akk et al., 2008; Bracamontes and Steinbach, 2009) and focus on the effects of 3α5αP on the equilibrium between the open and desensitized states.

3α5αP produced a small reduction in steady-state current in α₁(Q242L)β₃ receptors with mutations in neither of the intrasubunit sites (Figure 10A). This inhibitory effect was eliminated by α₁(V256S)β₃, indicating that it was due to receptor desensitization (Figure 10D). In receptors with combined mutations in the intersubunit and α₁ intrasubunit sites [i.e. α₁(Q242L/N408A/Y411F)β₃], 3α5αP significantly inhibited the steady-state current (Figure 10B), an effect that was markedly reduced by mutations in the β₃ intrasubunit site [α₁(Q242L)β₃(Y284F)] (Figure 10C). These data suggest that 3α5αP exerts a desensitizing effect by binding to the β₃ intrasubunit site and that 3α5αP binding to the α₁ intrasubunit site does not promote desensitization (Figure 10D). Notably, 3α5αP exerted only a modest inhibitory effect in α₁(Q242L)β₃ receptors in which occupancy of the β₃ intrasubunit site should promote inhibition. This may be due to a counterbalancing action at the α₁ intrasubunit site, where 3α5αP binding contributes more to receptor activation as demonstrated by our previous observation that mutations in the α₁ intrasubunit site significantly reduce 3α5αP potentiation of GABA-elicited currents (Chen et al., 2019). These results suggest that in addition to activation, 3α5αP enhances receptor desensitization. Enhanced desensitization by the PAM-NS 3α5αP (Haage and Johansson, 1999) and 3α5α-THDOC (Zhu and Vicini, 1997; Bianchi and Macdonald, 2003) has been observed in prior studies supporting the current finding with 3α5αP.

The human α₁ and β₃ GABA_AR subunits were subcloned into pcDNA3 for molecular manipulations and cRNA synthesis. Using QuikChange Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA), a FLAG tag was first added to the α₁ subunit then an 8xHis tag was added to generate the following His-FLAG tag tandem (QPSLHHHHHHHHDYKDDDDKDEL), inserted between the fourth and fifth residues of the mature peptide. The α₁ and β₃ subunits were then transferred into the pcDNA4/TO and pcDNA5/TO vectors (Thermo Fisher Scientific), respectively, for tetracycline inducible expression. Transient expression was done using the GABA_AR subunits rat α₁FLAG (Ueno et al., 1996) and human β₃ obtained from Geoffrey White (Neurogen, Branford, CT), each were subcloned into pcDNA3 for molecular manipulations and cRNA synthesis. Point mutations were generated using the QuikChange Site-Directed Mutagenesis Kit (Agilent) and the coding region was fully sequenced prior to use. The cDNAs were linearized with XbaI (NEB Labs, Ipswich, MA), and the cRNAs were generated using T7 mMessage mMachine (Ambion, Austin, TX).

Cell culture was performed as described in previous reports (Chen et al., 2019). The tetracycline inducible cell line T-REx-HEK293 (Thermo Fisher Scientific) was cultured under the following conditions: cells were maintained in DMEM/F-12 50/50 medium containing 10% fetal bovine serum (tetracycline-free, Takara, Mountain View, CA), penicillin (100 units/ml), streptomycin (100 g/ml), and blasticidin (2 μg/ml) at 37°C in a humidified atmosphere containing 5% CO₂. Cells were passaged twice each week, maintaining subconfluent cultures. Stably transfected cells were cultured as above with the addition of hygromycin (50 μg/ml) and zeocin (20 μg/ml). A stable cell line was generated by transfecting T-REx-HEK293 cells with human α₁-8xHis-FLAG pcDNA4/TO and human β₃ pcDNA5/TO, in a 150 mm culture dish, using the Effectene transfection reagent (Qiagen, Germantown, MD). Two days after transfection, selection of stably transfected cells was performed with hygromycin and zeocin until distinct colonies appeared. Medium was exchanged several times each week to maintain antibiotic selection. Individual clones were selected from the dish and transferred to 24-well plates for expansion of each clone selected. When the cells grew sufficiently, about 50% confluence, they were split into two other plates, one for a surface ELISA against the FLAG epitope and a second for protein assay, to normalize surface expression to cell number. The best eight clones were selected for expansion into 150 mm dishes, followed by [³H]muscimol binding to examine the receptor density. Once the best expressing clone was determined, the highest expressing cells of that clone were selected through fluorescence activated cell sorting. Transient transfections were done in HEK293S GnTI^- cells obtained from ATCC (CRL-3022) using Effectene (Qiagen). The identity of the cell lines has been authenticated using short tandem repeat analysis. Mycoplasma test performed on the cells used for these experiments was negative.

Stably transfected cells were plated into dishes. After reaching 50% confluence, GABA_A receptors were expressed by inducing cells with 1 μg/ml of doxycycline with the addition of 5 mM sodium butyrate. Cells were harvested 48 to 72 hr after induction. HEK cells, after induction, were grown to 100% confluence, harvested and washed with 10 mM potassium phosphate, 100 mM potassium chloride (pH 7.5) plus protease inhibitors (Sigma-Aldrich, St. Louis, MO) two times. The cells were collected by centrifugation at 1,000 g at 4°C for 5 min. The cells were homogenized with a glass mortar and a Teflon pestle for ten strokes on ice. The pellet containing the membrane proteins was collected after centrifugation at 20,000 g at 4°C for 45 min and resuspended in a buffer containing 10 mM potassium phosphate, 100 mM potassium chloride (pH 7.5). The protein concentration was determined with micro-BCA protein assay and membranes were stored at −80°C.

The syntheses of neurosteroid photolabeling reagents (KK148, KK150, KK200, KK123) are detailed in previous reports (Jiang et al., 2016; Cheng et al., 2018). For all the photolabeling experiments, 10–20 mg of HEK cell membrane proteins (about 300 pmol [³H]muscimol binding) were thawed and resuspended in buffer containing 10 mM potassium phosphate, 100 mM potassium chloride (pH 7.5) and 1 mM GABA at a final concentration of 1.25 mg/ml. For the photolabeling competition experiments, 3 μM KK200 or KK123 in the presence of 30 μM competitor (3α5αP, KK148, KK150, and 3β5αP) or the same volume of ethanol was added to the membrane proteins and incubated on ice for 1 hr. The samples were then irradiated in a quartz cuvette for 5 min, by using a photoreactor emitting light at >320 nm. The membrane proteins were then collected by centrifugation at 20,000 g at 4°C for 45 min. The photolabeled membrane proteins were resuspended in lysis buffer containing 1% n-dodecyl-β-D-maltoside (DDM), 0.25% cholesteryl hemisuccinate (CHS), 50 mM Tris (pH 7.5), 150 mM NaCl, 2 mM CaCl₂, 5 mM KCl, 5 mM MgCl₂, 1 mM EDTA, 10% glycerol at a final concentration of 1 mg/ml. The membrane protein suspension was homogenized using a glass mortar and a Teflon pestle and incubated at 4°C overnight. The protein lysate was centrifuged at 20,000 g at 4°C for 45 min and supernatant was incubated with 0.5 ml anti-FLAG agarose (Sigma) at 4°C for 2 hr. The anti-FLAG agarose was then transferred to an empty column, followed by washing with 20 ml washing buffer (50 mM triethylammonium bicarbonate and 0.05% DDM). The GABA_ARs were eluted with aliquots of 200 μg/ml FLAG tag peptide and 100 μg/ml 3X FLAG (ApexBio) in the washing buffer. The pooled eluates (9 ml) containing GABA_ARs were concentrated to 100 μl using 100 kDa cut-off centrifugal filters.

The purified α₁β₃ GABA_AR (100 μl) was reduced with 5 mM tris(2-carboxyethyl)phosphine for 1 hr, alkylated with 5 mM N-ethylmaleimide (NEM) for 1 hr, and quenched with 5 mM dithiothreitol (DTT) for 15 min. These three steps were done at RT. Samples were then digested with 8 μg of trypsin for 7 days at 4°C to obtain maximal recovery of TMD peptides. The digestions were terminated by adding formic acid in a final concentration of 1%, followed directly by LC-MS analysis on an Orbitrap Elite mass spectrometer. 20 μl samples were injected into a home-packed PLRP-S (Agilent, Santa Clara, CA) column (10 cm ×75 μm, 300 Å), separated with a 145 min gradient from 10% to 90% acetonitrile, and introduced to the mass spectrometer at 800 nl/min with a nanospray source. MS acquisition was set as a MS1 Orbitrap scan (resolution of 60,000) followed by top 20 MS2 Orbitrap scans (resolution of 15,000) using data-dependent acquisition, and exclusion of singly charged precursors. Fragmentation was performed using high-energy dissociation with normalized energy of 35%. Analysis of data sets was performed using Xcalibur (Thermo Fisher Scientific) to manually search for TM1, TM2, TM3 or TM4 tryptic peptides with or without neurosteroid photolabeling modifications. Photolabeling efficiency was estimated by generating extracted chromatograms of unlabeled and labeled peptides, determining the area under the curve, and calculating the abundance of labeled peptide/(unlabeled + labeled peptide). Analysis of statistical significance comparing the photolabeling efficiency of KK200 and KK123 for α₁β₃ GABA_AR was determined using one-way ANOVA with Bonferroni's multiple comparisons test and paired t-test, respectively (Prism 6, GraphPad Software, San Diego, CA). MS2 spectra of photolabeled TMD peptides were analyzed by manual assignment of fragment ions with and without photolabeling modification. Fragment ions were accepted based on the presence of a monoisotopic mass within 20 ppm mass accuracy. In addition to manual analysis, PEAKS (Bioinformatics Solutions Inc, Waterloo, ON, Canada) database searches were performed for data sets of photolabeled α₁β₃ GABA_AR. Search parameters were set for a precursor mass accuracy of 20 ppm, fragment ion accuracy of 0.1 Da, up to three missed cleavages on either end of the peptide, false discovery rate of 0.1%, and variable modifications of methionine oxidation, cysteine alkylation with NEM and DTT, and NS analogue photolabeling reagents on any amino acid.

[³H]muscimol-binding assays were performed using a previously described method (Chen et al., 2019). HEK cell membrane proteins (100 μg/ml final concentration) were incubated with 3 nM [³H]muscimol (30 Ci/mmol; PerkinElmer Life Sciences), neurosteroid (3 nM–30 μM) or etomidate (30 nM–200 μM) in different concentrations and binding buffer (10 mM potassium phosphate, 100 mM potassium chloride, pH 7.5) in a total volume of 1 ml. Assay tubes were incubated for 1 hr on ice in the dark. Nonspecific binding was determined by binding in the presence of 1 mM GABA. Membranes were collected on Whatman/GF-C glass filter papers using a Brandel cell harvester (Gaithersburg, MD). To perform [³H]muscimol binding isotherms, 100 μg/ml aliquots of membrane protein were incubated with 0.3 nM–1 μM [³H]muscimol for 1 hr on ice in the dark. The specific activity for [³H]muscimol concentrations from 30 nM to 1 μM was reduced to 2 Ci/mmol by dilution with nonradioactive muscimol. The membranes were collected on Whatman/GF-B glass filter papers using a vacuum manifold. Raw concentration-dependent total and nonspecific binding and calculated specific binding data from a representative experiment (WT receptors, no NS) are shown in Figure 7—figure supplement 1. For [³H]muscimol-binding experiments examining competitive interactions between neurosteroids, the combined neurosteroids (0.3 μM 3α5αP or 3 μM KK148 ±30 μM KK150) or the same volume of dimethyl sulfoxide (DMSO) were added to the membranes which were then incubated with 3 nM [³H]muscimol on ice for 1 hr. Time courses of neurosteroid [³H]muscimol binding enhancement were examined by adding 10 µM of neurosteroids (3α5αP, 3β5αP) to membranes that had been fully equilibrated with 3 nM [³H]muscimol for 1 hr on ice; binding was then measured as a function of time at 1, 3, 10, 30, 60 min. The membranes were collected on Whatman/GF-B glass filter papers using a vacuum manifold. Radioactivity bound to the filters was measured by liquid scintillation spectrometry using Bio-Safe II (Research Products International, Mount Prospect, IL).

HEK cells were harvested by gently washing dishes with buffer containing 10 mM sodium phosphate (pH 7.5), 150 mM sodium chloride twice. The cells were collected by centrifugation at 500 g at 4°C for 5 min, and resuspended in isotonic (10 mM sodium phosphate, 150 mM sodium chloride, pH 7.5) or hypotonic (10 mM sodium phosphate, pH 7.5) buffer to prepare two different conditions for radioligand binding to intact cells [isotonic buffer for cell surface receptors; hypotonic buffer for total receptors (cell surface receptors + intracellular receptors)]. The cells were incubated on ice for 2 hr, after which the sodium chloride concentration was adjusted to be isotonic before the radioligand binding procedure. HEK cells were aliquoted to assay tubes (20 samples/150 mm dish) in a total volume of 1 ml, and incubated with 3 nM [³H]muscimol ±10 μM KK148 for 1 hr on ice in the dark. Nonspecific binding was determined by binding in the presence of 1 mM GABA. The membranes were collected on Whatman/GF-B glass filter papers using a vacuum manifold. Radioactivity bound to the filters was measured by liquid scintillation spectrometry using Bio-Safe II.

The wild-type and mutant α₁β₃ GABA_AR were expressed in oocytes from the African clawed frog (Xenopus laevis). Frogs were purchased from Xenopus 1 (Dexter, MI), and housed and cared for in a Washington University Animal Care Facility under the supervision of the Washington University Division of Comparative Medicine. Harvesting of oocytes was conducted under the Guide for the Care and Use of Laboratory Animals as adopted and promulgated by the National Institutes of Health. The animal protocol (No. 20180191) was approved by the Animal Studies Committee of Washington University in St. Louis. The oocytes were injected with a total of 12 ng cRNA. The ratio of cRNAs was 5:1 ratio (α₁:β₃) to minimize the expression of β₃ homomeric receptors. Following injection, the oocytes were incubated in ND96 (96 mM NaCl, 2 mM KCl, 1.8 mM CaCl₂, 1 mM MgCl₂, 5 mM HEPES; pH 7.4) with supplements (2.5 mM Na pyruvate, 100 U/ml penicillin, 100 μg/ml streptomycin and 50 μg/ml gentamycin) at 16°C for 2 days prior to conducting electrophysiological recordings. The electrophysiological recordings were conducted at room temperature using standard two-electrode voltage clamp. The oocytes were clamped at −60 mV. The chamber (RC-1Z, Warner Instruments, Hamden, CT) was perfused with ND96 at 5–8 ml/min. Solutions were gravity-applied from 30 ml glass syringes with glass luer slips via Teflon tubing. The current responses were amplified with an OC-725C amplifier (Warner Instruments, Hamden, CT), digitized with a Digidata 1200 series digitizer (Molecular Devices), and stored using pClamp (Molecular Devices). Current traces were analyzed with Clampfit (Molecular Devices). Activation by steroids (Figure 1) was tested by coapplying a steroid with 0.3 µM GABA (P_open = 0.05–0.1). The desensitizing effects of steroids (Figures 9–10) were tested by coapplying a steroid with 1 mM (saturating) GABA, alone or in the presence of PB, during the steady-state phase of the current response. The combination of GABA and PB was used to activate some combinations of mutations to maintain a consistent, high peak open probability (0.55–0.95). In control experiments in WT α₁β₃ GABA_ARs, pentobarbital had no effect on 3β5αP inhibition of steady-state current (data not shown). The stock solution of GABA was made in ND96 at 500 mM, stored in aliquots at −20°C, and diluted on the day of experiment. The stock solution of muscimol was made at 20 mM in ND96 and stored at 4°C. The steroids were dissolved in DMSO at 10–20 mM and stored at room temperature.

The raw amplitudes of the current traces were converted to units of open probability through comparison to the peak response to 1 mM GABA + 50 µM propofol, that was considered to have a peak P_open indistinguishable from 1 (Chen et al., 2019). The level of constitutive activity in the absence of any applied agonist was considered negligible and not included in this calculation. The converted current traces were analyzed in the framework of the three-state Resting-Open-Desensitized activation model (Germann et al., 2019a; Germann et al., 2019b). The model enables analysis and prediction of peak responses using four parameters that characterize the extent of constitutive activity (termed L; L = Resting/Open), affinity of the resting receptor to agonist (K_C), affinity of the open receptor to agonist (K_O), and the number of agonist binding sites (N). Analysis and prediction of steady-state responses requires an additional parameter, termed Q, that describes the equilibrium between open and desensitized receptors (Q = Open/Desensitized).

The P_open of the peak response is expressed as:Popen,peak=11+LΓand the P_open of the steady-state response as:Popen,steady−state=11+1Q+LΓwhereΓ=[ (1+[X]/KC)(1+[X]/KO) ]N

[X]is the concentration of agonist present, and other terms are as described above. In practice, the value of LΓ was calculated using the experimentally determined P_open of the peak response, and then used as a fixed value in estimating Q from P_{open,steady-state}.

The P_desensitized was calculated using:Pdesensitized=11+Q+QLΓ

The effect of 3β5αP on 1 mM GABA-elicited steady-state current was expressed through a change in the value of Q. The modified Q (termed Q*) was then used to predict changes in P_open and P_desensitized at low [GABA]. Calculated probabilities (e.g. P_open, P_desentitized) are reported as mean ± SD.

A model of the α₁β₃ GABA_AR was developed using the crystal structure of the human β₃ homopentamer (PDB ID: 4COF) as a structural template (Miller and Aricescu, 2014). In this structure, the cytoplasmic loop was replaced with the sequence SQPARAA (Jansen et al., 2008). The pentamer subunits were organized as A α₁, B β₃, C α₁, D β₃, E β₃. The α₁ sequence was aligned to the β₃ sequence using the program MUSCLE (Edgar, 2004). The pentameric alignment was then used as input for the program Modeller (Sali and Blundell, 1993), using 4COF as the template; a total of 25 models were generated. The best model as evaluated by the DOPE score (Shen and Sali, 2006) was then submitted to the H++ server (http://biophysics.cs.vt.edu) to determine charges and optimize hydrogen bonding. The optimized structure was then submitted to the PPM server (https://opm.phar.umich.edu/ppm_server) for orientation into a lipid membrane. The correctly oriented receptor was then submitted to the CHARMM-GUI Membrane Builder server (http://www.charmm-gui.org) to build the fully solvated, membrane bound system oriented into a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine (POPC) bilayer. The system was fully solvated with 40715 TIP3 water molecules and ionic strength set to 0.15 M KCl. The NAMD input files produced by CHARMM-GUI (Lee et al., 2016) use a seven-step process of gradually loosening constraints in the simulation prior to production runs. A 100 ns molecular dynamics trajectory was then obtained using the CHARMM36 force field and NAMD (Lee et al., 2016). The resulting trajectory was then processed using the utility mdtraj (McGibbon et al., 2015), to extract a snapshot of the receptor at each nanosecond of time frame. These structures were then mutually aligned by fitting the alpha carbons, providing a set of 100 mutually aligned structures used for docking studies. The docking was performed using AutoDock Vina (Trott and Olson, 2010) on each of the 100 snapshots in order to capture receptor flexibility. 3α5αP and 3β5αP were prepared by converting the sdf file from PubChem into a PDB file using Open Babel (O'Boyle et al., 2011), and Gasteiger charges and free torsion angles were determined by AutoDock Tools. Docking grid boxes were built for the β₃–α₁ intersubunit, and the α₁ and β₃ intrasubunit sites with dimensions of 15 × 15 × 15 Ångströms encompassing each binding pocket. Docking was limited to an energy range of 3 kcal from the best docking pose and was limited to a total of 20 unique poses. The docking results for a given site could result in a maximum of 2000 unique poses (20 poses × 100 receptor structures); these were then clustered geometrically using the program DIVCF (Meslamani et al., 2009). The resulting clusters were ranked by Vina score and cluster size, and then visually analyzed. A comparison of proposed NS-binding sites between the modeled α₁β₃ GABA_AR TMD and the experimentally determined α₁β₃γ₂ cryo-EM structure PDB ID: 6I53 (Laverty et al., 2019) was performed. The two structures were read into UCSF Chimera and mutually aligned using MatchMaker (Meng et al., 2006). Using the same Vina docking configuration files discussed above, 3α5αP and 3β5αP were then docked into the respective sites of the α₁β₃γ₂ cryo-EM structure. The results are shown in Figure 11 and Figure 11—source data 1; there was very little difference in the results between the modeled α₁β₃ and the α₁β₃γ₂ cryo-EM structure.