Probing GABAA receptors with inhibitory neurosteroids

ABSTRACT &ggr;‐aminobutyric acid type A receptors (GABAARs) are important components of the central nervous system and they are functionally tasked with controlling neuronal excitability. These receptors are subject to post‐translational modification and also to modulation by endogenous regulators, such as the neurosteroids. These modulators can either potentiate or inhibit GABAAR function. Whilst the former class of neurosteroids are considered to bind to and act from the transmembrane domain of the receptor, the domains that are important for the inhibitory neurosteroids remain less clear. In this study, we systematically compare a panel of recombinant synaptic‐type and extrasynaptic‐type GABAARs expressed in heterologous cell systems for their sensitivity to inhibition by the classic inhibitory neurosteroid, pregnenolone sulphate. Generally, peak GABA current responses were inhibited less compared to steady‐state currents, implicating the desensitised state in inhibition. Moreover, pregnenolone sulphate inhibition increased with GABA concentration, but showed minimal voltage dependence. There was no strong dependence of inhibition on receptor subunit composition, the exception being the &rgr;1 receptor, which is markedly less sensitive. By using competition experiments with pregnenolone sulphate and the GABA channel blocker picrotoxinin, discrete binding sites are proposed. Furthermore, by assessing inhibition using site‐directed mutagenesis and receptor chimeras comprising &agr;, &bgr; or &ggr; subunits with &rgr;1 subunits, the receptor transmembrane domains are strongly implicated in mediating inhibition and most likely the binding location for pregnenolone sulphate in GABAARs. This article is part of the “Special Issue Dedicated to Norman G. Bowery”. HIGHLIGHTSA range of GABAA receptor subtypes are inhibited by pregenolone sulphate.Peak GABA curents are less sensitive to inhibition than steady‐state currents.Desensitised state of GABAA receptors most sensitive to neurosteroid inhibition.Inhibition increases with GABA concentration, but not strongly voltage‐dependent.Pregnenolone sulphate binding site located within subunit transmembrane domains.

These receptors are predominantly composed of abg receptors that reside at inhibitory synapses, and abg, abd and ab receptors which are found in extrasynaptic membrane domains (Farrant and Nusser, 2005;Mody, 2001;Moss and Smart, 2001). In the brain, GABA A Rs will be subject to modulation by endogenous ligands and pre-eminent amongst these are the neurosteroids (Belelli and Lambert, 2005). Two major classes have been defined; the potentiating neurosteroids exemplified by derivatives of sex and stress hormones, such as allopregnanalone (Allop) and tetrahydro-deoxycorticosterone (THDOC), and the inhibitory neurosteroids exemplified by pregnenolone sulphate (PS) and dihydroepiandrosterone (Belelli and Lambert, 2005;Seljeset et al., 2015). To date, use has been made of expression systems to study recombinant GABA A Rs and neuronal cultures which contain an array of GABA A Rs (Eisenman et al., 2003;Park-Chung et al., 1999;Shen et al., 2000;Zaman et al., 1992) all in the context of probing the mechanism of action of the inhibitory neurosteroids. Although we have a clearer idea as to where these neurosteroids bind to GABA A Rs from recent X-ray crystallography (Laverty et al., 2017;Miller et al., 2017), a systematic functional analysis of inhibitory neurosteroids at different GABA A Rs has not yet been achieved.
Previously, the receptor subtype selectivity of pregnenolone sulphate has been partly examined Zaman et al., 1992;Zhu et al., 1996), including the use of C. elegans GABA receptors to determine receptor domains that are important for PS inhibition (Wardell et al., 2006;Twede et al., 2007). However, the profiling of inhibitory neurosteroid sensitivity at the most common mammalian GABA A R subtypes thought to exist in the brain is still incomplete.
The main aim of the present study is therefore to systematically study the modulation of various GABA A R subtypes by the inhibitory neurosteroid PS using a single expression system, human embryonic kidney 293 cells (HEK cells), coupled to common analytical techniques to characterise PS inhibition. We have used chimeric receptors to probe the essential structural elements of the receptor subunits that contribute towards PS inhibition. These approaches have allowed a direct comparison between the activities of PS at different GABA A R subtypes, and provided an indication as to whether modulation is more likely to be important for the activation of synaptic or extrasynaptic GABA A Rs.

Materials and methods
2.1. Cell culture HEK cells were cultured using Dulbecco's modified Eagle medium (DMEM) supplemented with 10% v/v foetal calf serum (FCS), 100 U/ml Penicillin-G and 100 mg/ml streptomycin (Gibco). Cells were incubated at 37 C in humidified air with 5% CO 2 . When approximately 70e80% confluent, cells were washed with Ca 2þand Mg 2þ -free Hank's balanced salt solution (HBSS; Gibco) and harvested using 0.05% w/v trypsin-EDTA (Gibco). Cells were resuspended in culture medium and centrifuged at 168 Â g for 2 min (MSE Mistral, 2000 centrifuge). The cell pellet was resuspended in DMEM-based culture medium and re-plated at appropriate dilutions. For electrophysiology, cells were plated onto 22 mm glass coverslips (VWR international) pre-coated with 100 mg/ml poly-L-lysine (Sigma).

Cell transfection
HEK cells were transfected with murine DNA (except GABA r1 subunit DNA which was human) and allowed 16e40 h for expression prior to experimentation. A calcium phosphate protocol was used. cDNAs for individual receptor subunits (1 mg for each subunit) were mixed with 340 mM CaCl 2 (20 ml) and a HEPES-buffered saline (HBS; 24 mL; 50 mM HEPES, 280 mM NaCl and 2.8 mM Na 2 HPO 4 , pH 7.2). Enhanced green fluorescent protein (pEGFP-C1) was included as a marker for transfection. The total amount of DNA did not exceed 4 mg per coverslip. A transfection ratio of 1:1:1:1 was used for heteromeric receptors (e.g. a1:b2:g2L: with pEGFP-C1), whilst for homomeric receptors, a ratio of 2:1 (b3 or r1 with pEGFP-C1) was used.

Site-directed mutagenesis
All subunits were expressed in a mammalian pRK5 vector to achieve high levels of expression. Site-directed mutations were made using the QuikChange kit (Stratagene) or the Phusion kit (Thermo Fisher Scientific) with primer sequences as shown in Table 1. DNA was sequenced using the Sanger Sequencing Service (Source Bioscience, Cambridge, UK). Following successful mutagenesis, larger cultures were grown and constructs were eluted for storage in TE buffer (1 mg/ml) using the Plasmid Maxi kit (HiSpeed ® , Qiagen). Constructs were kept at À20 C for long-term storage. All chimeras used in this study were prepared as described previously (Gielen et al., 2015).

Patch-clamp electrophysiology
Coverslips with transfected HEK cells were mounted onto a recording chamber fixed to a Nikon Eclipse TE300 microscope with differential interference contrast optics. Cells were continuously superfused with Krebs solution containing (mM): 140 NaCl, 4.7 KCl, 1.2 MgCl 2 , 2.52 CaCl 2 , 11 glucose and 5 HEPES, adjusted to pH 7.4 with 1 M NaOH. Patch pipettes (resistance 2.5e4 MU) were filled with K þ -based internal solution containing (mM): 1 MgCl 2 , 120 KCl, 11 EGTA, 10 HEPES, 1 CaCl 2 and 2 K 2 ATP, adjusted to pH 7.2 with 1 M NaOH. The osmolarity of the internal solutions was measured using a vapour pressure osmometer (Wescor Inc.), and was in the range 300 ± 10 mOsm/l. All recordings were performed at room temperature.
Whole-cell membrane currents were recorded with an Axopatch 200B amplifier (Molecular Devices). HEK cells were voltageclamped between À20 and À40 mV. Data acquisition was performed with Clampex 10.3 (Molecular Devices). Currents were filtered at 2 kHz and digitised at 20 kHz via a Digidata 1440A (Molecular Devices). The series resistance was monitored and calculated throughout all recordings by measuring the membrane current responses to 10 mV hyperpolarising voltage steps of 50 ms duration at a frequency of 10 Hz. Recorded cells for which the series resistance varied by more than 30% were discarded. The series resistance was typically in the range 4e10 MU.
Control responses to GABA were obtained at regular intervals by applying GABA at a high concentration (EC 80-100 ) to obtain an estimate of membrane seal stability and any GABA current run-down over time. These responses were used to normalise subsequent responses that were used to compile the GABA concentrationresponse curves. A U-tube rapid application system was used for drug applications with a solution exchange time of <100 ms. A recovery period of 2e3 min was allowed between each application of drug to allow recovery from desensitisation and to minimise the run-down of currents.

Analysis of currents
The amplitudes of peak and steady-state GABA-activated currents were measured relative to the baseline holding current prior to GABA application using Clampfit (v10.3.1.5) software (Molecular Devices). To generate GABA concentration-response curves, the peak of each GABA response was normalised to the peak response to a saturating concentration of GABA (1 mM, unless otherwise stated) and expressed as a percentage. Similarly, steady-state GABA currents were expressed as a percentage of the steady-state current measured at a given time point during the application of GABA.
For inhibition of GABA currents, the steady-state current was defined as the current measured at 10 s following the start of GABA/ drug application. For slowly declining currents, the amplitude was Table 1 Forward and reverse primer sequences used to generate mutant r1 cDNA constructs.

Construct
Forward primer sequence (5 0 -3 0 ) Reverse primer sequence ( measured at 10 s and this was used as a proxy for the steady-state. GABA and antagonists were co-applied, unless otherwise stated. To study inhibition of a GABA response by PS, an EC 80 concentration of GABA (the concentration at which 80% of the maximal response is achieved) was co-applied with PS. Normalised GABA concentration-response curves were fitted using the Hill Equation, where I is the normalised response to GABA, Imax is the control maximum response to saturating GABA (100%), A is the concentration of applied GABA, EC 50 is the concentration of GABA producing 50% of maximal response, and n is the Hill coefficient. Inhibition curves were fitted using an inhibition equation, where I is the normalised GABA response in the presence of an antagonist, Imax is the maximal response in the absence of antagonist, B is the concentration of antagonist, n is the Hill coefficient and the IC 50 is the concentration of antagonist producing 50% inhibition of the GABA response. All data were curve fitted using a non-linear least squares algorithm in Origin 6.0 (Microcal).

Drugs
GABA was dissolved in distilled water and stock solutions (1 M) were kept at 4 C. Pregnenolone sulphate (Sigma) was dissolved in DMSO to a stock concentration of 20 mM and kept at À20 C. Picrotoxinin (Sigma) was dissolved in DMSO and stored in 100 mM aliquots at À20 C.

Statistics
The Kolmogorov and Smirnov test was used to check if data were normally distributed. For parametric data, pairwise comparisons were made using Student's t-test. Statistical comparisons between more than 2 groups were undertaken using a one-way analysis of variance (ANOVA) in conjunction with the Tukey posthoc test. All statistical analyses were performed in GraphPad InStat 3 (GraphPad Software, Inc.). The threshold for statistical significance was set at p < 0.05. Data are reported as mean ± standard error of the mean (SEM).

Recombinant GABA A Rs and pregnenolone sulphate
To assess the inhibitory activity of PS at GABA A Rs, and to determine if the neurosteroid exhibits any receptor subtype selectivity, recombinant receptors incorporating a1-6 subunits with b2 and/or g2L/d were systematically expressed in HEK cells and studied using whole-cell electrophysiology. The homomeric r1 receptor was also studied as its biophysical and pharmacological profiles are distinct from those of the heteromeric GABA A Rs, especially with regard to slower rates of receptor activation and limited desensitisation (Weiss and Chang, 1999), factors which could affect PS inhibition (Seljeset et al., 2015). These receptor subtypes were segregated into those likely to be expressed at inhibitory synapses (synaptic-type, Fig. 1A and B) and those most likely to reside extrasynaptically (extrasynaptic-type; Fig. 1C and D).
The inhibitory effect of PS was observed by co-applying increasing concentrations of PS (0.001e100 mM; Fig. 1) with GABA at an EC 80 concentration (see Table 2 for GABA concentrations).
Under these conditions, PS, only at >10 mM, inhibited GABA peak currents at all the heteromeric receptors examined; however, at <10 mM, PS caused a slowly developing greater block of steady-state currents that increased the apparent rate of desensitisation ( Fig. 1E and F). The IC 50 for PS inhibition of the steady-state GABA current at a1b2g2L was 0.4 ± 0.1 mM (n ¼ 7; Fig. 1B). At 100 mM PS, the steadystate current was completely blocked, whilst the peak currents were inhibited by only~40%, with little inhibition seen at lower (<1 mM) PS concentrations (Fig. 1A, B, E). These data suggest PS preferentially blocks the steady-state rather than peak GABA currents. This profile for PS inhibition appeared similar with each receptor subtype studied, except for the r1 receptor which was notably less sensitive to PS, with marginal inhibition of the peak and steady-state currents only observed with 100 mM PS ( Fig. 1C and D). Otherwise, for the heteromeric abg/d GABA A Rs, PS inhibition of steady-state currents yielded IC 50 values ranging between 0.4 and 1.3 mM (Table 3). For receptors containing b2 and g2L subunits, expressed with a1, 2, 3 or a5, similar IC 50 values (p > 0.05) were evident. Replacing b2 with b3 in the a1bg2L receptor had minimal impact on PS inhibition. In addition, the d-containing receptors, a4b2d and a6b2d, showed comparable sensitivity to PS compared to receptors containing ab2g2L subunits, and were similar in terms of IC 50 to a1b2 (p > 0.05). Together, these results suggest, from the subunits studied, that there is not a strong dependence on the subunit composition for PS inhibition of steady state currents, apart from receptors composed of r1 subunits (Table 3).
By comparison, a differential block by 100 mM PS at various GABA A R subtypes became evident when measuring peak currents ( Fig. 1A, C). Compared with the inhibition of peak GABA current at a1b2g2L (60.1 ± 9.6% of GABA control), inhibition by 100 mM PS was significantly increased at a6b2d (16.7 ± 3.4%; p < 0.001) and at a4b2d (31 ± 5.4%; p < 0.01). The least sensitive peak currents were those for the r1 receptor (68 ± 3.3%) where inhibition was comparable to that observed with synaptic-type a1b2g2L receptors (p > 0.05).

Access of pregnenolone sulphate to the binding site
The greater inhibition of steady-state over peak currents could arise if the PS binding (association) rate to the GABA A R is slow during co-application. This was examined by pre-applying PS for 20 s prior to its co-application with GABA. Inhibition should be increased if slow on-binding is causing the increased steady-state current inhibition. PS (10 mM) was pre-applied followed by co-applications with 30 mM GABA (EC 80 ) to the a1b2g2L receptor ( Fig. 2A). The steadystate responses to GABA were inhibited, with pre-and then coapplied PS, and remained stable with subsequent co-applications at~15% of the GABA control (n ¼ 5, Fig. 2A). Notably, the peak current response remained stable throughout also, and was similar to that observed without pre-application of PS. With preapplication, the peak response to the third application of GABA and 10 mM PS was 80.5 ± 2.9% of control, whereas without preapplication, the response was 85.9 ± 2.8% of control (n ¼ 5, p ¼ 0.2230; Figs. 1A and 2A). These data implied that access for PS to its binding site was unaffected by pre-application and thus unlikely to bind efficiently to the inactive state of the receptor.

GABA A R activation and pregnenolone sulphate block
As pre-application of PS did not affect the level of receptor block, we then assessed whether inhibition is receptor state-dependent using a1b2g2L receptors. PS (1 mM) was co-applied with GABA concentrations from 1 mM (EC 20 ) to 1 mM (EC 100 ), and peak currents and steady-state currents were measured (Fig. 2B). Whereas inhibition of peak currents by PS was similarly minimal at all concentrations of GABA, a clear increase in steady-state current inhibition was observed with higher concentrations of GABA (n ¼ 7, (A, B) Inhibition of GABA EC 80 peak (A) and steady-state (B) currents by PS at synaptic-type receptors, including: a1b2g2L (black), a1b3g2L (red), a2b2g2L (green) and a3b2g2L (blue) (n ¼ 5e7). (C,D) Inhibition of GABA EC 80 peak (C) and steady-state (D) currents by PS at extrasynaptic-type GABA A Rs, including a1b2 (black), a4b2d (red), a5b2g2L (green), a6b2d (blue) and r1 (cyan) (n ¼ 5e6). Data points represent mean ± SEM. Curves were fitted as described in the Methods. (E,F) GABA EC 80 responses and the effect of co-applications with 0.1e100 mM PS at a1b2g2L, a typical synaptic-type receptor (E), and a4b2d, a typical extrasynaptic-type receptor (F). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) p < 0.01 comparing 1 and 30 mM GABA), attaining a maximum inhibition at 30 mM GABA (EC 80 ). This GABA concentration was used subsequently for experiments regarding PS modulation of GABA A Rs.
The observation that PS is a more potent antagonist at higher GABA concentrations could be due to increased open probability of the GABA channel or increased occupancy by GABA, allowing greater access of PS to its binding site. Such a scenario is in accord with a channel or transmembrane domain (TMD) binding site, which implies PS could be a use-dependent blocker. The increase in inhibition observed at high GABA levels may also suggest the block is state-dependent. At higher GABA concentrations a larger proportion of receptors will be desensitised, and PS may only then access its binding site and act as a negative modulator.
To investigate these scenarios, current-voltage (I-V) relationships were first used to determine if antagonism by PS was voltagesensitive. This behaviour might be expected given that PS has a negatively-charged sulphate group on the C3 position of the A ring. Thus, increased block may develop at depolarised potentials if the binding site 'experiences' the membrane electric field. However, the I-V relationship revealed only weak voltage-dependence to the block exerted by 1 mM PS, with inhibition increased by 12% at þ90 compared to À90 mV ( Fig. 2C; n ¼ 8, p ¼ 0.009). The weak voltagesensitivity of PS suggests that binding is not strongly affected by the membrane electric field, arguing for a binding site located outside the channel pore or perhaps shielded within the transmembrane domain (TMD). The receptor activation-dependence of PS modulation is therefore more likely to be due to either higher agonist occupancy or a receptor state-dependent block.
Extrasynaptic GABA A Rs will experience markedly reduced GABA concentrations (i.e. nanomolar) compared to those present at inhibitory synapses (millimolar) (Farrant and Nusser, 2005;Glykys and Mody, 2007). To assess whether PS acts as an activation-or state-dependent antagonist at extrasynaptic-type a4b2d receptors, PS was co-applied with low GABA concentrations from 0.1 mM (EC 15 ) to 3 mM (EC 80 ) (Fig. 3A). As noted for synaptic-type a1b2g2L receptors, but now using much lower concentrations of GABA, 1 mM PS exerted greater inhibition as the GABA concentration increased to 1 mM for a4b2d receptors (Fig. 3B). By contrast, and again in accord with a1b2g2L receptors, the peak current was minimally affected by PS at all GABA concentrations tested. This demonstrates that PS acts as a state-dependent antagonist also at a4b2d receptors, suggesting that inhibition by PS is likely to occur at receptors located outside inhibitory synapses where ambient GABA concentrations are likely to be lower.

Internal access for pregnenolone sulphate to its binding site
Clearly PS can access its binding site to inhibit GABA A Rs when applied externally. PS has the canonical lipophilic four-ring carbon backbone common to the neurosteroids, and it is derived from cholesterol. Thus, it is possible that PS can partition into the membrane like other neurosteroid molecules (Akk et al., 2009), although this may be affected by the charged sulphate group in ring A. However, whether it can access its binding site from the cytoplasmic side of the membrane like the potentiating neurosteroids (Akk et al., 2005) is unknown. To examine this, 100 mM PS was internally-applied via the patch pipette solution and 1 mM GABA responses were recorded at 5 min intervals for 30 min (Fig. 3C) and compared to control recordings using normal PS-free internal solution. As PS inhibits steady-state currents with minimal effect on the peak current, we expected a larger run-down of the steadystate current in cells with PS-containing internal solution if the neurosteroid can access its binding site from the cytoplasm. However, the time profiles for the peak and steady-state currents were similar with or without internal solution supplemented with 100 mM PS for up to 30 min (p ¼ 0.324 and 0.833 respectively, n ¼ 8; Fig. 3C). These results imply that PS is unable to inhibit GABA A Rs from the cytosolic side of the cell membrane, and suggests that its binding site can only be accessed externally.

Pregnenolone sulphate and the GABA channel
Given the GABA activation-dependence of PS inhibition, we examined if PS can bind within the GABA channel by employing a competition protocol with picrotoxin (PTX), an antagonist that is considered to operate as an open-channel blocker of GABA A Rs and other members of the pentameric ligand-gated ion channel family (Erkkila et al., 2008;Hibbs and Gouaux, 2011;Krishek et al., 1996a). GABA was applied (EC 100 ; 1 mM) to obtain stable control responses prior to co-application with 10 mM PS to the a1b2g2L receptor. After recovery (not shown) PTX (10 mM) was then pre-applied before coapplication with GABA. As the PTX block is use-dependent (Yoon et al., 1993), it was applied 2e3 times with GABA to achieve a steady-state inhibition. Finally, GABA and PTX were co-applied with PS to determine if the PTX binding was occluding PS binding and inhibition (Fig. 4A). PS alone had minimal effect on the peak GABA current (95.2 ± 2.0% of control) whilst the steady-state current was greatly reduced (8.7 ± 1.5%; Fig. 4B). By contrast, PTX had a smaller inhibitory effect on the steady-state current (73.4 ± 1.8%, after two applications), but substantially reduced the peak current (37.3 ± 8.1% of control, Fig. 4A and B). When PS and PTX were coapplied, a profound block of both peak and steady-state currents was evident with the peak current reduced to 24.5 ± 6.4% of control and the steady-state current reduced to 6.5 ± 2.2%. Re-applying PTX alone with GABA after wash-out of PS, showed that the inhibition profile in PTX and PS was not due to an enhanced block by PTX  ( Fig. 4A). The level of block caused by PS was similar in the presence and absence of PTX, indicating that both PTX and PS exert their full inhibitory effect independently when co-applied, thus their binding sites are unlikely to overlap. Combining these data with the low voltage-sensitivity of PS argues for a binding site for the inhibitory neurosteroid located outside the ion channel.
3.6. Pregnenolone sulphate and r1 homomeric GABA A Rs As the inhibition by PS at r1 GABA receptors was distinct and less potent compared to that for other heteromeric GABA A R subtypes, we further examined its profile. Given that the extent of PS inhibition at a1b2g2L depended on the GABA concentration, we investigated whether this also applied to the r1 receptor. Coapplying 30 mM PS with GABA at concentrations between 1 mM (EC 20 ) and 300 mM (EC 100 ) revealed that the peak and steady-state currents were invariant at 80e90% of the GABA control responses (Fig. 4C) suggesting that there is no activation-dependent block at this receptor. Furthermore, pre-application of PS did not increase the level of inhibition of r1 receptor-mediated currents (data not shown). These results indicate that r1 receptor is only marginally sensitive to PS inhibition and may not contain the molecular signalling pathway necessary for full PS inhibition.

GABA A R structural domains sensitive to PS inhibition
Given the low sensitivity of r1 receptors to PS inhibition, we used this receptor as a null protein to explore receptor domains that are necessary for supporting PS inhibition. To do this we constructed receptor chimeras formed between r1 and a1, b2 or g2 subunits. Our aim was to switch the receptor sensitivities to PS inhibition depending on the heteromeric receptor assembly. The first chimera studied, r1-260-a1, contained the complete extracellular domain (ECD) of r1 up to residue 260, (i.e. the start of M1, numbered as in the mature protein), with the TMD and its associated linkers taken from the a1 subunit (Fig. 5A). This chimera, designed to determine whether PS is dependent upon the ECD or TMD of a1, was inhibited by PS (Fig. 5A). Although the potency was reduced compared to inhibition at heteromeric wild-type receptors (IC 50 range 0.4e1.3 mM for steady-state currents), prominent and similar inhibition of both peak and steady-state GABA currents was still observed with IC 50 s of 9.5 ± 1.1 mM and 7.9 ± 1.7 mM, respectively. Virtually full inhibition of the GABA peak and steady-state current was obtained at 100 mM PS. These results show that PS can antagonise a chimera where the ECD is from the largely PSinsensitive r1 subunit, and suggested that the neurosteroid is likely to rely on the TMD of a1 for inhibition. By substituting a1 for g2S in another chimera, r1-260-g2S, inhibition was again observed at concentrations of PS higher than 1 mM, and full inhibition of the steady-state current was attained at 100 mM PS (Fig. 5B). The peak current was also inhibited by PS, and reached 39.2 ± 5.1% of control at 100 mM PS. These results also imply that the TMD is important for PS inhibition, but for the g2 subunit-containing chimera, PS was less potent when compared to the potency determined at the r1-260-a1 chimera. Next we investigated a heteromeric receptor chimera using three different chimeric subunits composed of the ECD from a1, b2 and g2 separately fused to the TMD of r1 with its associated linkers (Fig. 5C). Hypothetically, we considered that PS should require the TMD of a1 and g2 and quite possibly b2, for full inhibition. So with this chimera PS inhibition was expected to be disrupted. However, PS was still able to inhibit the small GABA-activated peak and steady-state currents, though only at 10 and 100 mM (Fig. 5C).
Notably, a much larger standing current, caused by constitutive channel activity of the heteromeric chimera, was also revealed following inhibition by PS in the absence of GABA (Fig. 5D). It is likely that PS mostly mediated block of this spontaneous current rather than the GABA-mediated current, given that the outward current was 2e3 fold greater when 100 mM PS was applied in the absence of GABA.
The r1-260-b2 chimera did not functionally express in HEK cells, and thus PS block at this chimera was not determined.
Nevertheless, wild-type b3 subunits will form homomeric receptors providing currents gated by pentobarbitone (PB) but not by GABA (Davies et al., 1997;Krishek et al., 1996b;Wooltorton et al., 1997). These currents are slowly activating and followed by a rebound current after agonist wash-off (Wooltorton et al., 1997). To determine if the b3 homomers were subject to inhibition by PS, PB (500 mM) was used as an agonist, and co-applied with 100 mM PS (Fig. 6A). The neurosteroid caused a concentration-dependent block of the PB-induced peak current, and also reduced (though to a lesser extent) the magnitude of the rebound current. This observation argues that a binding site for PS must exist on the b3 subunit. Combined with the data from the chimera studies, these findings suggest that the neurosteroid has the potential to bind to more than one type of subunit of the GABA A R (i.e. a, b, g and r1), but most likely this binding involves the TMD.
To probe the TMD of the a1 subunit in more detail for PS inhibition, we created two further chimeras, r1-284-a1 ( Fig. 6B; r1 sequence up to the end of M1), and r1-352-a1 ( Fig. 6C; r1 to the start of the M3-M4 linker). Depending upon the retention or otherwise of PS inhibition with these chimeras, this would indicate the relative importance of discrete areas of the TMD for this inhibitory neurosteroid. GABA-activated currents were all inhibited by 100 mM PS co-applied with 1 mM GABA for both chimeras. The level of peak and steady-state inhibition was greater for the chimera retaining M2-M4, but also retained for the chimera containing M4 alone. These results broadly implicate a1 subunit TMDs as a critical region in PS inhibition, with binding and transduction of the PS inhibitory effect requiring synergy between M2-4.

Inhibition of GABA A Rs by PS is affected by the 2' residue in the ion channel
Although the GABA A R subunit TMD is a key region for PS inhibition, previous work has suggested an important role for the 2 0 residue near the cytoplasmic end of M2 in recombinant a1b2g2L receptors. Substituting the 2' valine in a1 (V256) and the homologous alanine in b2 (A252) for a serine residue caused the association rate for PS inhibition to reduce by 30-fold for a1 V256S b2g2L but not a1b2 A252S g2L (Akk et al., 2001). Thus the a1 subunit seems important for PS inhibition. Despite these findings, the inhibitory effects of PS are reduced or abolished for Xenopus oocytes expressing a1b2 A252S g2L or a1 V256S b2g2L (Wang et al., , 2007, showing that the mutation in either the a1 or b2 subunit can affect PS sensitivity. We examined the importance of the 2' residue for PS inhibition by examining its role in both the a1 and r1 subunits. Initially we recorded GABA whole-cell currents in HEK cells expressing a1 V256S , b2 and g2L to assess whether the mutation affected GABA potency and gating of the receptor. For the a1 V256S b2g2L receptor GABA was 6-fold more potent compared to wild-type, with the EC 50 reduced from 4.9 ± 1.4 mM (wild-type) to 0.8 ± 0.2 mM for the mutant (p ¼ 0.0104, n ¼ 5e6). We assessed the PS-sensitivity of a1 V256S b2g2L in response to GABA EC 80 application. Although this mutation has been previously reported to reduce or ablate PS sensitivity at a1 V256S b2g2L, inhibition was still evident at higher PS concentrations ( Fig. 7A and B). However, the curve for PS inhibition of steady-state currents was shifted to the right (0.4 ± 0.1 mM for wild-type a1b2g2L compared to 35.5 ± 8.2 mM for a1 V256S b2g2L; p ¼ 0.0005, n ¼ 7; Fig. 7B), and a greater level of inhibition of the peak GABA current was observed for the a1 V256S b2g2L receptor (60.1 ± 9.6% of control for wild-type; 22.6 ± 4.0% of control for a1 V256S b2g2L, p ¼ 0.0014, n ¼ 6e7). A rebound current was also present upon wash-off of 100 mM PS, suggesting the channels reentered an open state before closure (Fig. 7C). Thus, the a1 V256S b2g2L mutation shifts the steady-state PS inhibition curve to the right (lower potency), but simultaneously GABA peak currents become more susceptible to inhibition by PS.
As homomeric receptors formed of r1 subunits exhibited significantly reduced sensitivity to PS inhibition compared to synaptic and extrasynaptic heteromeric GABA A Rs ( Fig. 1C and D), we investigated whether the 2 0 residue also played a role in modulation by PS at this homomeric receptor, since it is possible that r1 lacks the transduction machinery to couple PS binding to receptor inhibition. We therefore switched the r1 proline at 2 0 for its equivalent valine found in wild-type a1 subunit to assess if this conferred increased sensitivity to PS. We also substituted the 2 0 proline for a serine (r1 P294S ) as a control since this residue reduced PS potency at the a1b2g2L receptor. Both mutations shifted the GABA concentration-response curve for r1 homomers to the right with the estimated EC 50 for wild-type r1 (2.6 ± 0.3 mM) increased by~10-fold for r1 P294V (24.2 ± 1.1 mM, p < 0.0001, n ¼ 4e8) and~3fold for r1 P294S (9.7 ± 1.2 mM, p ¼ 0.0009, n ¼ 6e8). By virtue of these curve shifts, the 2' mutations are likely to affect GABA potency and/or possibly gating kinetics for r1. To assess the PS sensitivity of the wild-type and mutant r1 receptors, EC 80 GABA was co-applied with PS at increasing concentrations (Fig. 8AeC). The r1 P294S behaved similarly to wild-type r1, with no greater inhibition of the peak or steady-state currents observed (Fig. 8 D, E). By contrast, r1 P294V was more sensitive to PS, with clear inhibition of the steady-state current observed with PS concentrations higher than 1 mM (Fig. 8C, E). The IC 50 for steadystate current inhibition by PS at r1 P294V was 6.0 ± 0.6 mM, approximately 6-fold higher than for wild-type heteromeric abg/d receptors. This result supports a role for the 2' valine in M2 for signal transduction of PS inhibition at GABA A receptors.

Discussion
Understanding the mechanism(s) by which inhibitory neurosteroids modulate GABA A Rs is important for predicting how this modulation will affect inhibition and ultimately neuronal behaviour in the brain. In addition, knowing whether an endogenous molecule exhibits GABA A R subtype selectivity is important for predicting in which regions of the brain and subcellular locations the compound is likely to have most effect. By examining a range of synaptic-and extrasynaptic-type GABA A Rs composed of abg, abd and ab, the PS IC 50 for inhibition of the steady-state currents was (A) Schematic diagram (upper panel) of the transmembrane topology for the r1-260-a1 chimera in which the ECD is taken from r1 (blue) and the TMD and linkers are from a1 (orange). The lower panel shows PS inhibition data for GABA EC 80 (30 mM) peak and steady-state currents (n ¼ 5). (B) Schematic (upper panel) for r1-260-g2S chimera in which the ECD from r1 (blue) is fused to the TMD from g2S (green). Lower panel shows PS inhibition data for GABA EC 80 (200 mM) peak and steady-state currents (n ¼ 5). (C) Upper panel depicts a heteromeric chimera composed of: a1-222-r1 þ b2-218-r1 þ g2-235-r1 subunits. ECDs are from a1 (orange), b2 (dark red) and g2 (green), and the TMD is from r1 (blue) for all chimeras. Lower panel shows PS inhibition data for GABA EC 80 (1 mM) peak and steady-state currents (n ¼ 9). The negative normalised current describes inhibition by 100 mM PS of the constitutive channel activity. Data are expressed as mean ± SEM. (D) Upper panel shows examples of PS inhibition of GABA-activated (inward) and constitutivelyactive (outward) currents for the heteromeric a1-222-r1 þ b2-218-r1 þ g2-235-r1 chimeric receptor. Lower panel shows PS inhibition concentration data for the constitutivelyactive current. between 0.4 and 1.3 mM. Although the differences in PS potency at some of the receptor subtypes are statistically different, the small difference (less than 5-fold) suggests that under physiological conditions, subtype selectivity is unlikely for PS inhibition. Nevertheless, a higher efficacy PS block was noted for peak currents at dcontaining receptors, and might imply a greater role for PS in modulating GABA-mediated tonic currents.
The potency of PS is reported to be similar at a1b2g2L and a5b2g2L receptors expressed in Xenopus oocytes when co-applied with EC 80 GABA , which supports the findings presented here. Furthermore, the potency and efficacy of PS were increased in the presence of the potentiating neurosteroid THDOC , which binds to a separate site to PS (Akk et al., 2008;Laverty et al., 2017). This corroborates the hypothesis that high levels of receptor activation or increased channel open probability promote inhibition by PS.
However, contrasting results have also been reported, with a3b2g2 receptors noted to be 10-fold more sensitive to PS inhibition than a1b2g2. Moreover, PS was more potent at a6b3g2 than at a6b3d receptors suggesting PS potency is reduced by the d subunit (Zhu et al 1996). This distinction was also noted with a4b3g2 and a4b3d, suggesting the g2 subunit increased the sensitivity to PS (Brown et al., 2002). By contrast, the human r1 receptor was notably less sensitive to PS than the heteromeric GABA A Rs (Li et al., 2007).
Similarly to the inhibitory neurosteroids, the a subunit does not influence the potency of the potentiating neurosteroids, when coexpressed with b1 and g2L subunits (Belelli et al., 2002). The b subunit isoform also did not affect allopregnanolone modulation (Belelli et al., 2002;Hadingham et al., 1993), contrasting with PS, where replacement of b2 with b3 in a1bg2L receptors reduced inhibitory potency. The efficacy of the potentiating neurosteroids does vary between receptor subtypes, with neurosteroid potentiation (macroscopic efficacy) increased at d-compared to g2containing GABA A Rs (Akk et al., 2007;Belelli et al., 2002;Wohlfarth et al., 2002). With regard to GABA occupancy of the receptor, PS inhibition was greater at higher GABA concentrations, which suggested potential activation-or use-dependent block. Inhibition also developed slowly, leading to greater block of steady-state currents than peak currents. Furthermore, applying PS before GABA to the receptors did not increase inhibition, indicating that the slowly developing block is not due to a slow forward rate constant (association) for PS binding to the receptors. Similarly slow kinetics for PS block have been noted for rat hippocampal neurons (Eisenman et al., 2003). These results are consistent with the idea that receptors have to be activated before PS can modulate the response.
However, for a1b2g2L receptors expressed in oocytes, greater inhibition of GABA currents has been observed following PS preapplication (Zaman et al., 1992).
Use-dependent block by PS was discounted since repeated GABA applications in the presence of PS did not further increase inhibition. Thus, the greater block observed at high GABA concentrations is more likely due to a state-dependent block, raising the question as to whether this is due to high receptor occupancy or to a conformation of the receptor caused by high activation levels, e.g. receptor desensitisation. A previous study assessed PS inhibitory potency in the presence of a saturating concentration of a GABA A R partial agonist, piperidine-4-sulphonic acid (P4S), which attains 30e40% of the maximal GABA response (Eisenman et al., 2003;Mortensen et al., 2004). At this maximum concentration of P4S, the potency of PS was more than 3-fold lower compared to the inhibition of a maximal GABA response. When comparing the fractional block by PS of currents produced by functionallyequivalent concentrations of GABA and P4S, the inhibition by PS of a1b2g2L receptors expressed in oocytes was similar. This showed that the level of receptor activity, rather than the extent of agonist occupancy, is likely to be a key determinant of PS potency.
The concept of activation-dependent block can be indicative of the antagonist requiring receptor activation to access its binding site, e.g., in the channel pore if located past the channel gate at 9'. For many pharmacological agents, this mode of block coincides with voltage-dependence (Cui et al., 2006;Newland and Cull-Candy, 1992). However, despite its negatively-charged sulphate moiety, PS block was not strongly voltage-dependent, with only 12% more block recorded at depolarised compared to hyperpolarised membrane potentials. Similar findings on the rate and extent of block have been reported by others in oocytes, HEK cells and rat cortical neurons (Akk et al., 2001;Eisenman et al., 2003;Majewska et al., 1988), strongly suggesting that PS does not act as an open-channel blocker and that its binding site is most likely located outside the channel pore. Also, the sulphate moiety of PS does not seem to be essential for GABA A R inhibition (Park-Chung et al., 1999;Seljeset et al., 2015) and would explain why the binding of PS is relatively unaffected by the membrane electric field. The profile of block by PS has some similarity to that of PTX at low GABA concentrations (Eisenman et al., 2003). Early studies indicated that PTX displaced PS from rat brain membranes (Majewska et al., 1990), and PS competitively inhibited the binding of the PTX-like blocker, t-butylbicyclophosphorothionate (TBPS) in rat synaptosomes (Majewska and Schwartz, 1987). Radioligand displacement is often interpreted as compounds competing for a common binding site, but for allosteric proteins like the GABA A R, this need not be the only interpretation. Furthermore, mutating the 2' channel residue in the GABA A R a and/or b subunits led to the identification of a possible binding site for PTX in the channel pore (Xu et al., 1995;Zhang et al., 1994), which is supported by the crystal structure of GluCl bound to PTX at this site (Hibbs and Gouaux, 2011). However, a single channel site for PS and PTX seems unlikely as competition experiments with a1b2g2L revealed no interaction and thus inhibitory effects on GABA currents that were additive. Moreover, block by PS and PTX is different at high GABA concentrations, with PTX having little or no effect on steadystate currents (Eisenman et al., 2003), contrasting with the marked attenuation of steady-state currents by PS at high GABA concentrations. This strengthens the hypothesis that the mechanism of block by PS and PTX are distinct and likely mediated by separate binding sites.
With regard to accessing the binding site, unlike the potentiating neurosteroids which can partition into the plasma membrane and exert their effect from the cytosol (Akk et al., 2009(Akk et al., , 2007, intracellular PS was ineffective and only reached its binding site from the external side of the membrane. This result also suggests PS has a defined binding site rather than affecting the receptor protein in a non-specific manner. This result has some similarity with another target of PS, the TRPM3 channel, which can only be activated by extracellular, but not intracellular PS (Wagner et al., 2008).
The GABA A R with a distinctive lack of PS sensitivity (below 100 mM) was the r1 receptor. This contrasts with the potentiating neurosteroids, which are positive and negative allosteric modulators at r1, though the concentrations required to achieve this are higher than at heteromeric receptors (>1 mM) (Morris et al., 1999).
In determining which receptor domains were crucial for PS inhibition, it became clear that several subunits could be involved. The sensitivity of a1b2 to PS indicates that binding does not depend solely on the g2L subunit. Furthermore, inhibition of PB-mediated currents confirmed the presence of a binding site for PS on the b3 subunit, which does not exclude potential sites on a1-6 and g2L subunits.
Various chimeras between r1 and a1, b2 or g2 subunits were examined in searching for the critical domains for PS inhibition. At first it seemed surprising that chimeras with ECD-TMD drawn from a, b or g subunits (ECD) with r1 (TMD), or vice-versa, retained some sensitivity to PS. However, for the chimeric receptor complex containing the TMD of r1 and ECD of a1, b2 and g2 (a1-222-r1 þ b2-218-r1 þ g2-235-r1) it appears likely that only the constitutive current, rather than the GABA current, was inhibited by PS implying that residues in the TMDs of the a1, b2 and/or g2 subunits are necessary for PS inhibition of the GABA-activated receptor. In support of this, chimeras containing the ECD of r1 and the TMDs of a1 or g2 (r1-260-a1/g2) were sensitive to PS, even those including only M4 of the a1 subunit. Taken together, this suggests that PS is likely to bind to the TMDs of a, b and g subunits (Fig. 9). It is also likely that the homomeric wild-type r1 receptor fails to support profound PS inhibition because of an absent signalling mechanism, since mutating the channel 2 0 residue to that found in a1 can render r1 more sensitive to PS. The residues necessary for this allosteric mechanism are likely to be present in a1, b2 and g2 subunits, and absent in the wild-type r1 receptor. This accumulated evidence for the importance of the TMDs in PS sensitivity is emphasised by recent work on a GLIC-GABA A Ra1 subunit chimera (Laverty et al., 2017), in which a PS binding site is tentatively located using X-ray crystallography to a TMD site involving a 'longitudinal Earlier studies of single channel recordings (Akk et al., 2001) revealed that the a1 subunit 2 0 mutation, a1 V256S , in abg receptors reduced the apparent association rate of PS 30-fold and prevented the reduction in GABA channel cluster duration by PS. Homologous mutations in the b2 and g2L subunits had no such effect. Our study showed that the PS IC 50 for GABA steady-state currents was decreased by 89-fold when the a1 subunit is mutated at 2 0 for a1 V256S b2g2L receptors. Conversely, for peak GABA current inhibition, the PS IC 50 is decreased. Reduced or abolished inhibitory effects of PS on GABA currents have been reported for a1 V256S b2g2L and a1b2 A252S g2L receptors (Wang et al., , 2007. This supports the notion that mutating either the a1 or b2/3 subunits at the 2 0 position reduces the potency and efficacy of PS. However, taking all data into consideration, and with the low voltage-dependence of PS, it is more likely that the 2' mutation alters an allosteric mechanism and interferes with signal transduction rather than directly affects the binding of PS. Interestingly, a1 V256S , but not b2 A252S , eliminates GABA A R inhibition by the 3b-hydroxypregnane steroids (Seljeset et al., 2015;Wang et al., 2002Wang et al., , 2007. These are diastereomers of the potentiating 3a-hydroxypregnane steroids, but are similar to the sulphated neurosteroids in that they non-competitively inhibit the GABA A R in an activity-or state-dependent manner. In the study by Wang et al. (2007), desensitisation kinetics were characterised by determining a ratio between peak and steady-state currents (I P /I SS ). In wild-type receptors, PS increased the ratio in a concentration-dependent manner, but this remained unchanged when the 2' mutation was introduced to the a1 or b2 subunit.
This suggests that PS promotes desensitisation of wild-type receptors, an effect that is removed by the mutations. Notably, the block by the 3b-hydroxypregnane steroids did not cause a concentration-dependent increase in the I P /I SS ratio in wild-type or mutant receptors, suggesting that the mechanism of block by sulphated steroids and 3b-hydroxypregnane steroid is not common (Wang et al., 2007). From this we deduce that the 2' residue is unlikely to be a common binding site for either group of steroids, and corroborates the hypothesis that this is likely to be a residue Fig. 8. PS inhibition at wild-type and mutant r1 receptors. Representative membrane currents for wild-type r1 (A), r1 P294S (B) and r1 P294V (C) in response to GABA EC 80 and following co-application with 1, 10 and 100 mM PS. (D, E) Inhibition by PS of GABA-mediated peak (D) and steady-state (E) currents at wild-type r1 (black), r1 P294V (red) and r1 P294S receptors (green) (n ¼ 6e12). Data are presented as mean ± SEM.
(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) important for allosteric signalling (Seljeset et al., 2015).
What do these data accrued from recombinant receptors mean for the actions of PS in the nervous system? Although it is difficult to be precise, by PS promoting a desensitised state, we might assume that low frequency inhibitory synaptic potentials (IPSPs) would be relatively unaffected by PS in the absence of overt receptor desensitisation. However, high frequency release, with summation of IPSPs and desensitisation, may show some degree of attenuation by PS. Moreover, given that extrasynaptic GABA A Rs can be desensitised by persistent exposure to low concentrations of GABA (Mortensen et al., 2010;Bright et al., 2011) we may expect some moderation by PS of the level of tonic inhibition.
In conclusion, PS exhibits minimal GABA A R subtype selectivity, and the extent to which this negative allosteric modulator blocks the GABA A R increases with receptor activation that is likely to promote the desensitised state. The domains that are pivotal to PS inhibition are centred on the subunit TMDs, which concurs with recent high resolution x-crystallographic data that suggests an involvement of this domain in the GABA A R for PS inhibition. Upper panel depicts a plan view of a hetero-pentameric GABA A R containing abg subunits at the level of the TMD. Subunits are labelled and a proposed binding location for PS is shown on the a subunit near the a-helices representing M3 and M4. Lower panel shows a side view of the same receptor schematic in the membrane with PS occupying a site which does not involve lipid partitioning or the channel pore, in keeping with the poor voltage sensitivity for inhibition.