Positive and Negative Allosteric Modulation of an α1β3γ2 γ-Aminobutyric Acid Type A (GABAA) Receptor by Binding to a Site in the Transmembrane Domain at the γ+-β− Interface*

Background: For some chiral barbiturates, one isomer potentiates and the other inhibits GABA responses by binding to unknown sites. Results: A photoreactive convulsant barbiturate identifies a transmembrane intersubunit-binding site between the γ and β subunits. Conclusion: Positive and negative allosteric modulators can bind to a common intersubunit site. Significance: This study defines a novel mode of regulation of GABAAR responses. In the process of developing safer general anesthetics, isomers of anesthetic ethers and barbiturates have been discovered that act as convulsants and inhibitors of γ-aminobutyric acid type A receptors (GABAARs) rather than potentiators. It is unknown whether these convulsants act as negative allosteric modulators by binding to the intersubunit anesthetic-binding sites in the GABAAR transmembrane domain (Chiara, D. C., Jayakar, S. S., Zhou, X., Zhang, X., Savechenkov, P. Y., Bruzik, K. S., Miller, K. W., and Cohen, J. B. (2013) J. Biol. Chem. 288, 19343–19357) or to known convulsant sites in the ion channel or extracellular domains. Here, we show that S-1-methyl-5-propyl-5-(m-trifluoromethyl-diazirynylphenyl) barbituric acid (S-mTFD-MPPB), a photoreactive analog of the convulsant barbiturate S-MPPB, inhibits α1β3γ2 but potentiates α1β3 GABAAR responses. In the α1β3γ2 GABAAR, S-mTFD-MPPB binds in the transmembrane domain with high affinity to the γ+-β− subunit interface site with negative energetic coupling to GABA binding in the extracellular domain at the β+-α− subunit interfaces. GABA inhibits S-[3H]mTFD-MPPB photolabeling of γ2Ser-280 (γM2–15′) in this site. In contrast, within the same site GABA enhances photolabeling of β3Met-227 in βM1 by an anesthetic barbiturate, R-[3H]methyl-5-allyl-5-(m-trifluoromethyl-diazirynylphenyl)barbituric acid (mTFD-MPAB), which differs from S-mTFD-MPPB in structure only by chirality and two hydrogens (propyl versus allyl). S-mTFD-MPPB and R-mTFD-MPAB are predicted to bind in different orientations at the γ+-β− site, based upon the distance in GABAAR homology models between γ2Ser-280 and β3Met-227. These results provide an explanation for S-mTFD-MPPB inhibition of α1β3γ2 GABAAR function and provide a first demonstration that an intersubunit-binding site in the GABAAR transmembrane domain binds negative and positive allosteric modulators.

For over 150 years, drug screens assessing in vivo animal responses have led to the identification of a structurally diverse group of compounds, including simple volatile ethers, alcohols, barbiturates, and steroids, that produce the complex physiological responses desirable for clinical anesthesia (1). In the process of identifying novel anesthetics, comparison of the actions of geometric isomers of certain volatile fluorinated ethers and barbiturate stereoisomers sometimes revealed that one isomer acted as an anesthetic and the other as a convulsant (2)(3)(4)(5). Anesthetic barbiturates and other intravenous anesthetics (propofol, etomidate, and steroids), as well as volatile ethers, potentiate inhibitory GABA type A receptors (GABA A R) 2 in vitro at concentrations producing anesthesia in vivo (6 -8), and the importance of GABA A Rs for anesthesia is demonstrated by the decreased sensitivity of "knock-in" mice bearing a single amino acid substitution in a GABA A R ␤ subunit to the immobilizing and hypnotic effects of pentobarbital, etomidate, and propofol (9 -12).
The convulsant effects of some barbiturates may be mediated by targets other than GABA A Rs (13). However, the convulsant S-1-methyl-5-phenyl-5-propyl barbituric acid (S-MPPB) inhibits GABA A R responses at the same concentration at which the anesthetic isomer, R-MPPB, potentiates responses (14,15).
Neither anesthetic nor convulsant barbiturates bind directly to the GABA or benzodiazepine-binding sites (16), and the differential effects of R-and S-MPPB on the binding of a GABA A R channel blocker suggest that the convulsant and anesthetic isomers may bind to distinct sites in a GABA A R (14).
Recently, two classes of general anesthetic-binding sites have been identified in the transmembrane domain (TMD) of ␣1␤3␥2 GABA A Rs based upon the locations of residues photolabeled by analogs of etomidate, mephobarbital, and propofol in homology models based on published structures of several homologous members of the Cys-loop superfamily of pentameric ligand-gated ion channels (17)(18)(19), including one of a human ␤3 GABA A R ( Fig. 1) (20). Photoreactive etomidate analogs identify a high affinity binding site for etomidate at the ␤ ϩ -␣ Ϫ subunit interfaces, based upon the photolabeling of amino acids in the ␤ subunit M3 and ␣ subunit M1 transmembrane helices (21,22). A mephobarbital analog, R-[ 3 H]mTFD-MPAB, photolabeled amino acids in the ␤M1, ␣M3, and ␥M3 transmembrane helices, identifying a second homologous class of anesthetic-binding sites at the ␣ ϩ -␤ Ϫ and ␥ ϩ -␤ Ϫ subunit interfaces (23). Although etomidate and R-mTFD-MPAB bind with Ͼ50-fold selectivity to the ␤ ϩ -or ␤ Ϫ -containing interface sites, respectively, the sites are not strictly etomidate-or barbiturate-specific. Propofol binds with little selectivity to both classes of sites, and a propofol analog photolabels residues in both classes of sites (23,24). These allosteric anesthetic-binding sites in the TMD show positive energetic coupling to each other and to the GABA-binding sites at the ␤ ϩ -␣ Ϫ subunit interfaces in the extracellular domain (ECD), as etomidate and GABA each enhanced R-[ 3 H]mTFD-MPAB photolabeling.
In this report, we prepare S-[ 3 H]mTFD-MPPB and use it as a photoaffinity reagent to determine where a convulsant barbiturate binds in an ␣1␤3␥2 GABA A R. Based upon the direct identification of photolabeled amino acids and the pharmacological specificity of photolabeling, we find that S-[ 3 H]mTFD-MPPB binds in the same ␥ ϩ -␤ Ϫ interface pocket as R-mTFD-MPAB. However, it binds in a different orientation, and its binding is inhibited allosterically by GABA, indicative of negative energetic coupling between the sites. S-[ 3 H]mTFD-MPPB also binds with lower affinity to the other intersubunit anesthetic sites, with positive energetic coupling to the GABA site. Our results provide a first demonstration that, similar to the benzodiazepine-binding site at the ␣ ϩ -␥ Ϫ interface in the ECD (27), at least one intersubunit-binding site in the GABA A R's TMD is a target for negative as well as positive allosteric modulators.
Electrophysiology-Whole-cell patch clamp recordings were obtained from induced HEK293-TetR cells expressing either ␣1␤3 or ␣1␤3␥2L GABA A receptors using methods described previously (28,29). Briefly, cells were seeded on a glass coverslip, and protein expression was induced with tetracycline (2 g/ml) for 5-26 h before recordings. All experiments were performed at room temperature (20 -22°C). The recording chamber was continuously perfused with the bath solution (in mM) as follows:145 NaCl, 5 KCl, 10 HEPES, 2 CaCl 2 , 1 MgCl 2 , and 10 glucose, pH 7.4 (pH adjusted with NaOH). The electrode solution contained (in mM) the following: 140 KCl, 10 HEPES, 1 EGTA, and 2 MgCl 2 at pH 7.3 (pH adjusted with KOH). Open pipette resistances ranged from 1.9 to 3 megohms. Cells were voltage-clamped at Ϫ50 mV using the patch clamp amplifier (Axopatch 200A, Molecular Devices Corp., Sunnyvale, CA). Whole-cell membrane capacitances and series resistances were compensated electronically by more than 85% with a lag of 10 s. Series resistances ranged from 0.5 to 2.5 megohms and cell capacitances from 16 to 18.5 picofarads. GABA A receptors were activated using 8-s pulses of GABA delivered via a multichannel superfusion pipette coupled to piezo-electric elements that switched solutions in less than 1 ms. Currents were filtered at 5 kHz and digitized at 10 kHz using pCLAMP version 8.1 (Molecular Devices Corp., Sunnyvale, CA) for off-line analysis with Clampfit 9 (Molecular Devices Corp., Sunnyvale, CA). Statistical analysis was performed in GraphPad Prism version 6 software (GraphPad Software, Inc., San Diego). All data are expressed as mean Ϯ S.D.
Purification of Expressed Human ␣1␤3␥2 GABA A Rs-␣1␤3␥2 L and ␣1␤3 GABA A Rs containing a FLAG epitope at the N terminus of the mature ␣1 subunit (MRK…SYGDYKDDDDKQPS…) were purified from tetracycline-inducible, stably transfected HEK293S cell lines using an anti-FLAG affinity resin as described previously (23,24,28,29). GABA A R was solubilized in 30 mM n-dodecyl ␤-D-maltopyranoside, and column wash and elution buffers contained 5 mM CHAPS and 0.2 mM asolectin. After elution with 1.5 mM FLAG peptide, aliquots from the eluted fractions were characterized for the number of GABA A R-binding sites, using [ 3 H]muscimol, and for etomidate modulation of [ 3 H]muscimol binding. Starting from membrane fractions containing 4 -8 nmol of [ 3 H]muscimol-binding sites, typical purification yields were 0.5-1.5 nmol of purified ␣1␤3␥2 GABA A R (30 -60 nM binding sites) and 1.5 nmol of ␣1␤3 GABA A R (60 nM binding sites), each in 15-25 ml of elution buffer. Fractions were flash-frozen in liquid N 2 and stored at Ϫ80°C until use.
GABA A R Photoaffinity Labeling-Aliquots of purified FLAG-␣1␤3␥2 GABA A Rs in elution buffer were photolabeled at analytical and preparative scales (40 -80 l or 1-2.5 ml of ␣1␤3␥2 GABA A R, per condition, respectively) to either characterize photoincorporation at the subunit level or to identify individual photolabeled amino acids using protein sequencing methods, respectively. Required volumes of stock solutions of radiolabeled, photoreactive anesthetic in methanol were transferred to glass test tubes, and solvent was evaporated under an argon stream before addition of GABA A R. The radioligand was resus-pended with occasional vortexing for 30 min on ice. Stock solutions of nonradioactive S-mTFD-MPPB (60 mM), R-mTFD-MPAB (60 mM), propofol (1 M), pentobarbital (60 mM), picrotoxinin (60 mM), and etomidate (60 mM) were prepared in methanol. Bicuculline methochloride (6 mM) was prepared in water. Methanol was present in all samples during photolabeling at a concentration of 0.5% (v/v). Photolabeled samples were immediately solubilized in SDS-sample buffer (23) and incubated for 30 -60 min at room temperature before SDS-PAGE.
SDS-PAGE and Enzymatic/Chemical Digestion of GABA A Rs-GABA A R subunits in SDS sample buffer were resolved by SDS-PAGE on 6% Tris-glycine gels, which were constructed as described (23), to accommodate the 150-l and ϳ1.5-ml sample volumes generated in analytical and preparative scale photolabelings, respectively. After electrophoresis, gels were stained with Coomassie Brilliant Blue. In analytical scale experiments, 3 H incorporation into subunits was determined by liquid scintillation counting or fluorography, and in preparative scale experiments, subunits were eluted from excised subunit gel bands as described (23). Material eluted from the gel bands was filtered, concentrated, acetone-precipitated (Ϫ20°C), and resuspended in 100 -200 l of digestion buffer (15 mM Tris and 0.1% SDS, pH 8.5). Aliquots (90 or 180 l) of resuspended subunits were digested at room temperature with EndoLys-C (0.5 units) for 14 days or with EndoGlu-C (2.5 g) for 2-4 days. Enzymatic digests were fractionated by reversed-phase HPLC (rpHPLC) as described (30), and fractions containing radiolabeled fragments were pooled for N-terminal sequencing or for further chemical fragmentation. Incorporation in ␥M2 was determined by sequencing a fragment beginning at ␥2Asp-260 from an EndoLys-C digest and by sequencing a parallel sample treated after immobilization on the sequencing filter with CNBr as described (31,32) for cleavage of peptides at the C termini of methionines. Photolabeling in ␤M2 was determined by sequencing a fragment beginning at ␤3Ile-242, produced by treating intact subunit immobilized on a sequencing filter with BNPS-skatole as described (22,24,33) to cleave at the C terminus of tryptophans. To characterize photolabeling in ␣M2, we sequenced the fragment beginning at ␣1Ser-251 at the N terminus of ␣M2 that can be isolated by rpHPLC fractionation of EndoGlu-C digests of ␣1 subunit and treatment with OPA at cycle 3 during sequencing (23). Photolabeling in ␤M1, ␤M3, ␣M1, and ␣M3 was determined by sequencing appropriate rpHPLC fractions from EndoLys-C digests of ␣1 or ␤3 subunits (21,22) Quantification of Inhibition of Photolabeling-The concentration dependence of inhibition of photolabeling by nonradioactive barbiturates or other drugs was determined in analytical photolabeling experiments. 3 H incorporation was determined in the following three stained subunit bands: a 56-kDa band, enriched in the ␣ subunit, and bands of 59 and 61 kDa, enriched in the ␤ subunit but differentially glycosylated (22). The ␥2 subunit was distributed more diffusely but centered in the 56-kDa band. For [ 3 H]azietomidate, parameters for the concentration dependence of inhibition were determined for the 56-kDa gel band that reflects photolabeling of ␣1Met-236 at the ␤ ϩ -␣ Ϫ subunit interface. For R-[ 3 H]mTFD-MPAB, parameters were determined for the 59-and 61-kDa gel bands that reflect photolabeling of ␤3Met-227 at the ␤ Ϫ subunit interfaces (23). For S-[ 3 H]mTFD-MPPB, parameters were determined for the 56-kDa gel band, which in this case reflects photolabeling of ␥2Ser-280 (see under "Results"). The concentration dependence of inhibition of subunit photolabeling was fit using nonlinear least squares by SigmaPlot 11.0 (Systat Software) to a single or two-site model using Equations 1 and 2, respectively, where B(x) is the 3 H in counts/min (cpm) incorporated into a subunit gel band when the total inhibitor concentration is x, B 0 is the specific 3 H incorporation in the absence of inhibitor; IC 50 is the total concentration of inhibitor that reduces the incorporated 3 H by 50%, with H and L denoting the high and low affinity binding sites; n H is the Hill coefficient; and B ns is the nonspecific 3 H incorporation in the presence of maximal concentrations of a competitor. Data were fit initially to Equation 1 with variable IC 50 values; B 0 was equal to the difference between total binding and nonspecific binding, and n H was equal to 1 or variable. Reversed-phase HPLC and N-terminal Sequence Analysis-Enzymatic digests of GABA A R subunits were fractionated by rpHPLC and subjected to N-terminal sequencing as described (23,30). Briefly, rpHPLC fractionation was performed using an Agilent 1100 binary HPLC system with a Brownlee Aquapore column. Fractions of 0.5 ml were collected at a flow rate of 200 l/min, and peptide elution was monitored by the absorbance at 215 nm. Aliquots (10%) of each fraction were counted to determine the 3 H distribution. Fractions containing peaks of 3 H were pooled and loaded onto Micro TFA glass fiber filters (Applied Biosystems) at 45°C. Total digests of intact GABA A R subunits and rpHPLC fractions, where indicated, were loaded directly onto Prosorb PVDF filters (Applied Biosystems) according to the manufacturer's directions.
Samples were sequenced using a Procise 492 protein sequencer (Applied Biosystems), with 2/3 of the material from each cycle of Edman degradation used for PTH-derivative quantification and 1/3 collected to measure 3 H release by scintillation counting. In some cases, we used o-phthalaldehyde (OPA) treatment during sequencing, as described (34), to chemically isolate a fragment of interest known to contain a proline at a particular cycle of Edman degradation or to test for the presence of a proline. Because OPA reacts with primary amines but not secondary amines (35), OPA treatment at a cycle containing a proline in the peptide of interest allows continued sequencing of that peptide while blocking further sequencing of other peptides not containing a proline at that cycle.
The amount of PTH-derivative released in a given sequencing cycle (in picomoles) was determined by comparing the peak height for the amino acid derivative in the chromatogram to the height of its standard peak. I 0 , the initial amount of a peptide in a sequencing sample (in picomoles), was determined from the amounts of PTH-derivative in each cycle by nonlinear least squares fit to Equation 3, where I x is the background-subtracted mass of the peptide residue in cycle x (in picomoles), and R is the average repetitive yield. For samples containing multiple fragments, only PTHderivatives unique to the fragment of interest were included in the fit. Amino acid derivatives whose amounts cannot be accurately estimated (His, Trp, Ser, Arg, and Cys) were omitted from the fit. E(x), the efficiency of photolabeling (in cpm/ pmol) of the amino acid residue in cycle x was calculated by Equation 4, where cpm x is the 3 H released in cycle x (in cpm).
Molecular Modeling-The locations of the photolabeled residues were visualized in an ␤3␣1␤3␣1␥2 GABA A R homology model based upon the structure of the homomeric human ␤3 GABA A R (PDB code 4COF (20)) that was made (Discovery Studio 4.0 (Accelrys, Inc.)) as described for the ␣1␤3 GABA A R (24) with the substitution of the ␥2 subunit for the ␤3 subunit designated E in the PDB model. After construction, the receptor was placed in a membrane force field and minimized (10 cycles) to ease strained interactions. To determine whether the pocket at the ␥ ϩ -␤ Ϫ interface can accommodate S-mTFD-MPPB, computational docking was performed using the CDocker module. Four randomly oriented S-mTFD-MPPB molecules were placed within the pocket in a binding site sphere of 11 Å radius centered at the level of ␥2Ser-280 (␥M2-15Ј), ␥2Ser-301 in ␥M3, and ␤3Met-227 in ␤M1. The 100 lowest interaction energy orientations (simulated annealing with full potential minimization) were collected for each molecule from 50 random conformations (high temperature molecular dynamics) and 50 randomized orientations within the spheres (i.e. 2500 initial conditions tested per molecule). 213 of 400 collected solutions predicted stable binding (CDocker interaction energies Ͻ0 kcal/mol). The 10 most favored binding solutions had CDocker energies from Ϫ35.6 to Ϫ38.9 kcal/mol and included orientations with the S-mTFD-MPPB diazirine directed toward ␥2Ser-280 and others with the diazirine oriented toward ␥M3/ ␤M1. This procedure was repeated for the equivalent pockets at the ␤ ϩ -␣ Ϫ and ␣ ϩ -␤ Ϫ interfaces. At the ␤ ϩ -␣ Ϫ interface adjacent to the ␥ ϩ -␤ Ϫ interface, all 400 collected solutions had interaction energies Ͻ Ϫ6 kcal/mol. The 10 lowest energies ranged from Ϫ33.4 to Ϫ37.2 kcal/mol, and for each of these the diazirine was oriented toward ␤M3/␣M1. At the ␣ ϩ -␤ Ϫ interface, all 400 solutions had interaction energies Ͻ Ϫ25 kcal/mol, with the 10 lowest energy solutions ranging from Ϫ38.2 to Ϫ40.6 kcal/mol. All 10 orientations were similar, with the diazirine projecting between ␣1M3 and ␤3M1.

S-mTFD-MPPB Inhibits ␣1␤3␥2 and Potentiates ␣1␤3 GABA A R Responses-We compared effects of S-mTFD-MPPB
on GABA responses in cell lines expressing ␣1␤3␥2 or ␣1␤3 GABA A Rs. Reponses were measured using approximate EC 10 GABA concentrations of 1 M for ␣1␤3 and 10 M for ␣1␤3␥2, because at these concentrations sufficiently robust currents are elicited for studying inhibition, while leaving ample room for the observation of current enhancement. As shown in Fig. 2, in ␣1␤3␥2 GABA A Rs, S-mTFD-MPPB at 46 M inhibited peak GABA-induced current amplitudes by 72 Ϯ 1.3% (n ϭ 4), whereas in ␣1␤3 GABA A Rs it enhanced them by 49 Ϯ 18% (n ϭ 6). These results suggest that inhibition by S-mTFD-MPPB requires the presence of the ␥2 subunit. The inhibition of ␣1␤3␥2 GABA A R responses by S-mTFD-MPPB is similar to the inhibition of GABA responses in cortical neurons seen for S-MPPB (15) and contrasts with the effects of mTFD-MPAB on ␣1␤3␥2 GABA A Rs, for which both isomers potentiate responses (25).  (23), [ 3 H]azietomidate photoincorporated primarily into a 56-kDa band, reflecting photolabeling of ␣1Met-236 in ␣M1 at the ␤ ϩ -␣ Ϫ subunit interface, and R-[ 3 H]mTFD-MPAB incorporated primarily into 59-and 61-kDa bands, reflecting photolabeling of ␤3Met-227 in ␤M1 at the ␤ Ϫ subunit interfaces. At both sites, photolabeling was enhanced by GABA but not by bicuculline. In contrast, S-[ 3 H]mTFD-MPPB incorporated most efficiently into a diffusely distributed GABA A R subunit band with mobility of ϳ56 kDa. Photolabeling in that band was inhibited by GABA but not by bicuculline, indicating that the GABA-inhibitable photolabeling was not within the GABAbinding site. S-[ 3 H]mTFD-MPPB was photoincorporated at lower levels into the 59-and 61-kDa bands, with that photolabeling enhanced by GABA. All S-[ 3 H]mTFD-MPPB photola-  That picrotoxinin did not inhibit S-[ 3 H]mTFD-MPPB photolabeling suggested that the GABA-inhibitable labeling is unlikely to be in a site within the ion channel, whereas the inhibition of photolabeling by anesthetic barbiturates and etomidate suggested that the photolabeling may be in the known intersubunit anesthetic-binding sites.

mTFD-MPPB Binding to Known General Anesthetic Sites-In initial photolabeling experiments, we tested S-and
Initial Localization of S-[ 3 H]mTFD-MPPB-binding Sites within ␣1␤3␥2 GABA A R-Because the ␥2 subunit is poorly stained and broadly distributed in the 56/59-kDa region of the SDS-polyacrylamide gel (23), and S-[ 3 H]mTFD-MPPB photolabeling in the ϳ56-kDa band appeared more diffusely distributed than the [ 3 H]azietomidate-photolabeled ␣ subunit band, experiments were designed to determine whether S-[ 3 H]mTFD-MPPB was photoincorporated primarily into the ␣1 or ␥2 subunit. Comparison of the distributions of 3 H when EndoLys-C subunit digests were fractionated by rpHPLC provided evidence that the GABA-inhibitable photolabeling in the 56-kDa gel band originated from the ␥2 subunit rather than the ␣1 subunit (Fig. 5). For the 56-kDa band from ␣1␤3␥2 GABA A R photolabeled by S-[ 3 H]mTFD-MPPB, there was a GABA-inhibitable hydrophilic peak of 3 H eluting at ϳ40% organic solvent (Fig. 5A). This peak was not observed in digests of the 59/61-kDa bands (Fig. 5B) or in digests of 56 or 59/61-kDa gel bands from S-[ 3 H]mTFD-MPPB-labeled ␣1␤3 GABA A Rs (Fig. 5, C and D). S-mTFD-MPPB was photoincorporated into a subunit fragment not labeled by R-[ 3 H]mTFD-MPAB, because there was no hydrophilic peak of 3 H in digests of 56 or 59/61-kDa bands derived from ␣1␤3␥2 GABA A R photolabeled by that anesthetic (Fig. 5, E and F). However, in all samples there was a broad peak of 3 H in the hydrophobic fractions (60 -70% organic) known to contain most of the ␣1 and ␤3 subunit transmembrane helices (22,23).  (Fig. 5A) led us to examine the differences in predicted subunit fragmentation patterns for EndoLys-C digests of ␥2 compared with the ␣1 subunit. In the regions of primary structure containing transmembrane helices, the presence of ␥2Lys-259 in the M1-M2 loop was notable, because EndoLys-C cleavage there and at either Lys near the C terminus of ␥M2 would generate a fragment containing only ␥M2. In contrast, EndoLys-C digestion of ␣1 or ␤3 subunits can only produce the fragments beginning before M1 and extending through M2 that had been identified previously in the hydrophobic HPLC fractions (22,23).

GABA Inhibits S-[ 3 H]mTFD-MPPB
GABA-inhibitable photolabeling of ␥2Ser-280 (␥M2-15Ј) was established by N-terminal sequence analyses of material from the hydrophilic rpHPLC peak of 3 H from the EndoLys-C digests of the 56-kDa gel band. Because the primary sequences in these fractions originated from the ␣1 subunit ECD, we used radiochemical sequencing strategies taking advantage of the fact that the ␥M2 fragment beginning at ␥2Asp-260 contains a Pro in cycle 4 and a Met in cycle 17 of Edman degradation. First, two identical samples from GABA A R photolabeled in the absence of GABA were sequenced with sequencing of one sample interrupted at cycle 4 for treatment with OPA to prevent further sequencing of any fragments not containing a Pro at that cycle. For both samples, there was a peak of 3 H release in cycle 21 of Edman degradation, consistent with photolabeling of ␥2Ser-280 (Fig. 6A). Based on the detected PTH-derivatives, the major fragments present originated from the ECD of the ␣1 subunit and included peptides beginning at ␣1Thr-43 and ␣1Ser-107 at 1-2 pmol. Treatment with OPA reduced sequencing of those fragments by Ͼ80%, and the EndoLys-C fragment beginning at ␥2Asp-260 was present at a low level (I 0 ϳ0.2 pmol). No peaks of 3 H release were seen when the corresponding fractions were sequenced from GABA A R photolabeled in the presence of GABA (Fig. 6A).
Photolabeling of ␥2Ser-280 was confirmed by sequencing 2 equivalent samples from another photolabeling experiment, with one sample pretreated with CNBr to cleave at the C ter- mini of methionines. Pretreatment with CNBr shifted the peak of 3 H release from cycle 21 to cycle 4 (Fig. 6B), consistent with cleavage at ␥2Met-276 in ␥M2. These radiochemical sequencing strategies established that the GABA-inhibitable photolabeling in the 21st cycle of Edman degradation was in a GABA A R subunit with a defined distribution of Lys, Pro, and Met residues, Lys-Xaa 3 -Pro-Xaa 12 -Met-Xaa 3 , where Xaa is any amino acid. Inspection of the ␣1, ␤3, and ␥2 subunit sequences revealed only one other fragment consistent with that distribution, a fragment from the ␥2 subunit cytoplasmic domain beginning at ␥2Asn-336 that also contains a Pro in cycle 2. Because OPA treatment at cycle 2 fully inhibited subsequent release of 3 H in cycle 21 (data not shown), the combined radiochemical sequencing strategies established GABA-inhibitable photolabeling of ␥2Ser-280 (␥M2-15Ј).
Based on the peak of 3 H release in cycle 21 and the mass of the ␥2Asp-260 fragment detected after OPA treatment, S-[ 3 H]mTFD-MPPB photolabeled ␥2Ser-280 at ϳ4000 cpm/ pmol, and GABA inhibited that photolabeling by Ն90%. The calculated efficiencies of photolabeling of ␥2Ser-280 in different pharmacological conditions and of the amino acids photolabeled in other subunits are tabulated in Table 1. The locations of ␥2Ser-280 at the ␥ ϩ -␤ Ϫ interface and of the other amino acids photolabeled by S-[ 3 H]mTFD-MPPB are depicted in Fig.  7, based upon their locations in a GABA A R homology model described below.

Evidence for Additional S-mTFD-MPPB-binding Sites, GABA Enhances S-[ 3 H]mTFD-MPPB
Photolabeling of ␤3Phe-289 (␤M3) and ␤3Thr-262 (␤M2-12Ј) at the ␤ ϩ -␣ Ϫ Interface-We next turned to the identification of S-mTFD-MPPBbinding sites that differ in their GABA sensitivity from the ␥ ϩ -␤ Ϫ site. In ␣1␤3␥2 GABA A Rs, GABA enhanced S-[ 3 H]mTFD-MPPB photolabeling in the ␤3 subunit ( Fig. 3; 59/61-kDa band), and rpHPLC fractionation of EndoLys-C digests of ␣1␤3␥2 GABA A Rs of this gel band enriched in ␤3 subunits established that all 3 H was recovered in the hydrophobic fractions that contain fragments beginning at the N termini of the M1 and M3 helices (Fig. 5B). To identify photolabeled amino acids in ␤M1 and ␤M3, we sequenced material from the appropriate rpHPLC fractions isolated from GABA A Rs photolabeled in the absence or presence of GABA or bicuculline. Representative sequencing data are shown in Fig. 8, A and B, and the calculated efficiencies of amino acid photolabeling in the different pharmacological conditions are tabulated in Table 1.

S-mTFD-MPPB Inhibition of S-[ 3 H]mTFD-MPPB and R-[ 3 H]mTFD-MPAB Photolabeling-To characterize S-mTFD-
MPPB binding affinity at the ␥ ϩ -␤ Ϫ interface, we compared the  Table 2). These results indicate that S-mTFD-MPPB binds in the presence of bicuculline to the ␥ ϩ -␤ Ϫ site with ϳ10-fold higher affinity than it binds to other intersubunit sites in the presence of bicuculline or GABA.

TABLE 1 Pharmacological specificity of photolabeling of residues in ␣1␤3␥2 and ␣1␤3 GABA A Rs by S-͓ 3 H͔mTFD-MPPB, a convulsant, and R-͓ 3 H͔mTFD-MPAB, an anesthetic (cpm/pmol of PTH-derivative)
The efficiency of photolabeling of a residue (in cpm/pmol) was calculated using Equation 4 (see under "Experimental Procedures"). N, the number of samples sequenced. The data are presented as the mean and individual values when two samples were sequenced or as mean (Ϯ S.D.) when three or four samples were sequenced. Other values were determined from the sequencing of single samples, with estimated uncertainties of Ͻ25%. The radiochemical specific activities of S-͓ 3 H͔mTFD-MPPB and R-͓ 3 H͔mTFD-MPAB are 50 and 38 Ci/mmol, respectively. ND means not determined.

Subunit-interface
Amino acid  Fig. 6A. For the control condition, the sample was sequenced with OPA treatment in cycle 4 to allow mass determination. For the ϩGABA condition, the sample was sequenced without OPA, and the cpm/pmol was calculated from the 3 H released in cycle 21 and the mass of the control sample. Residues labeled most efficiently in the absence or presence of GABA are highlighted by red or purple backgrounds, respectively. In the absence of GABA, ␥2Ser-280 (␥M2-15Ј) is labeled Ͼ10-fold more efficiently than any other residue (Table 1). GABA reduces photolabeling of ␥2Ser-280 by Ͼ90% and enhances photolabeling of ␤3Thr-262 (␤M2-12Ј) and ␤3Phe-289 in ␤M3 by 3-5-fold. The locations of the residues are approximated based upon their locations in an ␣1␤3␥2 GABA A R homology model (see Fig. 12).

Discussion
In this study, we demonstrate that the S-isomer of mTFD-MPPB, an ␣1␤3␥2 GABA A R inhibitor, stabilizes the receptor in a closed channel state by binding with high affinity to a TMD site in the ␥ ϩ -␤ Ϫ interface previously identified as a binding site for the anesthetic barbiturate R-mTFD-MPAB (23). S-mTFD-MPPB binding at this site shows negatively energetic coupling to GABA binding in the ECD. In contrast, R-mTFD-MPAB binding is positively coupled to GABA. Our results provide the first demonstration that subtle changes in structure determine whether a drug acts as a positive or negative GABA A R allosteric modulator when binding at a TMD intersubunit site, as occurs at the ECD ␣ ϩ -␥ Ϫ interface benzodiazepine-binding site (27,37). S-[ 3 H]mTFD-MPPB binding at the ␥ ϩ -␤ Ϫ interface was identified by the efficient and GABA-inhibitable photolabeling of ␥2Ser-280 in ␥M2. No other GABA-inhibitable photolabeling was observed either in ␣1␤3␥2 or ␣1␤3 GABA A Rs (Table  1). This observation, together with the inhibition detected in ␣1␤3␥2 but not ␣1␤3 GABA A Rs (Fig. 2), ties this site to the inhibition seen in vitro and convulsant activity seen in vivo (26). Interestingly, ␥2Ser-280 is homologous to ␤3Asn-265 (both M2-15Ј residues), a determinant of etomidate and propofol potency as GABA A R potentiators in vitro and as anesthetics in vivo (9,38). In a GABA A R homology model based upon the  Fig. 5, A and B, EndoLys-C digests of the 59-kDa gel bands were fractionated by rpHPLC, and materials in fractions 26 and 27 (A) or fractions 28 -30 (B) were sequenced. A, primary sequence began at ␤3Ala-280 (I 0 ϭ 1.6 pmol), with a secondary sequence beginning at ␤3Arg-216 (I 0 ϭ ϳ1 pmol). The major peak of 3 H release at cycle 10, consistent with photolabeling of ␤3Phe-289, was enhanced by 100% in the presence of GABA (Table 1). B, primary sequence began at ␤3Arg-216 (I 0 ϭ 4.5 pmol), with a secondary sequence beginning at ␤3Ala-280, present at levels below 1 pmol before OPA treatment in cycle 13 and undetectable after treatment. The peaks of 3 H release in cycles 12 and 16 indicated GABA-enhanced photolabeling of ␤3Met-227 and ␤3Leu-231 in ␤M1. The peak of 3 H release in cycle 10 corresponds to photolabeling of ␤3Phe-289 in ␤M3 of the secondary sequence. C, to identify photolabeling in ␤M2, aliquots from the 61-kDa gel bands from GABA A Rs photolabeled in presence of GABA (E) or bicuculline (F) were sequenced after treatment of the sequencing filter with BNPS-skatole to cleave at the C termini of tryptophans. The sequence beginning at ␤3Ile-242 was present (I 0 ϭ 2.3 pmol), along with fragments at 1-4 pmol each beginning at the ␤3 subunit N terminus, ␤3Arg-68, ␤3Val-93, and ␤3Arg-169, 49 amino acids before ␤M1, ␤3Ser-427, and ␤3Leu-444 in ␤M4 that are 17 and 4 amino acids in length, respectively. The peak of 3 H release in cycle 21 indicated GABA-enhanced photolabeling of ␤3Thr-262 (␤M2-12Ј). No 3 H release was seen when intact ␤ subunit was sequenced, and if cycle 21 of the ␤3Arg-169 fragment had been photolabeled, it would have been recovered by rpHPLC from EndoLys-C digests as a hydrophilic fragment from the ECD.  Table 1). A, when rpHPLC fractions 26 -27 were sequenced from an EndoLys-C digest of the 61-kDa gel band, the fragment beginning at ␤3Ala-280 was present at 6.8 pmol, and the peaks of 3 H release in cycles 7 and 10 indicated photolabeling of ␤3Met-286 and ␤3Phe-289. B, when rpHPLC fractions 28 and 29 were sequenced with OPA treatment at cycle 13, corresponding to ␤3Pro-228 in ␤M1, the primary sequence began at ␤3Arg-216 (I 0 ϭ 20 pmol) and a secondary sequence began at ␤3Ala-280, present at 2 pmol before OPA and undetectable after treatment. The peaks of 3 H release in cycles 12 and 16 indicated photolabeling of ␤3Met-227 and ␤3Leu-231 in ␤M1. The peaks of release in cycles 7 and 10 resulted from the photolabeling of ␤3Met-286 and ␤3Met-289 in the secondary sequence present before OPA treatment in cycle 13. C, to identify photolabeling in ␤M2, aliquots from the 59-kDa gel bands were sequenced after treatment of the sequencing filter with BNPS-skatole to cleave at the C termini of tryptophans. The sequence beginning at ␤3Ile-242 was present (I 0 ϭ 7pmol), along with fragments beginning at the ␤3 subunit N terminus, ␤3Arg-68, ␤3Val-93, ␤3Arg-169, and ␤3Ser-427 at 4 -10 pmol each. The peak of 3 H release in cycle 21 indicated photolabeling of ␤3Thr-262 (␤M2-12Ј). D, 2 aliquots of rpHPLC fractions 26 -29 from an EndoLys-C digest of the 56-kDa subunit gel band were sequenced with (f) or without (ࡗ) OPA treatment in cycle 13 (at ␣1Pro-233). For the untreated sample, the fragments beginning at ␣1Arg-221 (data not shown) and ␣1Val-280 (छ) were present at 11 and 8 pmol, respectively. For the OPA-treated sample, ␣1Arg-221 (Ⅺ) and ␣1Val-280 (data not shown) were initially present at 5 pmol. After OPA treatment, sequencing of the ␣1Arg-221 fragment continued, although the ␣1Val-280 fragment was reduced by Ͼ90%. The peak of 3 H release in cycle 15, not seen after treatment with OPA, indicated photolabeling of ␣1Tyr-294 in ␣M3. After treatment with OPA, the small peak of 3 H release in cycle 16 indicated photolabeling of ␣1Met-236 in ␣M1. Efficiencies of residue photolabeling in the absence or presence of GABA or bicuculline are included in Table 1.  Table 2.
In addition to the ␥ ϩ -␤ Ϫ site, S-mTFD-MPPB also bound with ϳ10-fold lower affinity to the intersubunit TMD site in the ␤ ϩ -␣ Ϫ interface, photolabeling residues that overlap with those photolabeled by [ 3 H]azietomidate (Fig. 12, D and F). Similar to etomidate, S-mTFD-MPPB binding at the ␤ ϩ -␣ Ϫ site is positively coupled to GABA binding. However, inhibition of [ 3 H]azietomidate photolabeling establishes that even in the presence of GABA, S-mTFD-MPPB binds weakly to those sites.
State Dependence of S-mTFD-MPPB and R-mTFD-MPAB Binding-The differences in the IC 50 values of S-mTFD-MPPB in the presence of GABA or bicuculline provide evidence that agonist/antagonist binding at the orthosteric site in the purified ␣1␤3␥2 GABA A R shifts the receptor conformational equilibrium, presumably between desensitized and closed states in our photolabeling assays. Our results provide a simple explanation for why S-mTFD-MPPB inhibits ␣1␤3␥2 and potentiates ␣1␤3 GABA A Rs. S-mTFD-MPPB binds at the ␥ ϩ -␤ Ϫ site with Ն10fold higher affinity in the bicuculline-stabilized state than the GABA-stabilized state or to the ␤ ϩ -␣ Ϫ site in its preferred GABA-stabilized state. Hence, S-mTFD-MPPB binding at the ␥ ϩ -␤ Ϫ site will result in negative allosteric modulation of GABA responses in the ␣1␤3␥2 GABA A R. In an ␣1␤3 GABA A R that has no ␥ ϩ -␤ Ϫ -binding site, S-[ 3 H]mTFD-MPPB photolabeling of residues at the ␤ ϩ -␣ Ϫ and ␣ ϩ -␤ Ϫ intersubunit sites is enhanced in the presence of GABA. This enhanced photolabeling is consistent with positive energetic coupling between S-mTFD-MPPB and GABA binding, with S-mTFD-MPPB acting as a positive allosteric modulator of GABA responses.
Our results also indicate that the positive energetic coupling between R-mTFD-MPAB and GABA binding is mediated primarily by strong state dependence of binding at the ␥ ϩ -␤ Ϫ site.
In the presence of bicuculline, S-mTFD-MPPB binds with high affinity at the ␥ ϩ -␤ Ϫ site (IC 50, H ϭ 1.7 M), but it inhibits R-[ 3 H]mTFD-MPAB photolabeling with an IC 50 of 130 M. This discrepancy indicates that in the presence of bicuculline, R-[ 3 H]mTFD-MPAB photolabels primarily ␤3Met-227 at the ␣ ϩ -␤ Ϫ site, and the 50% increase of ␤3Met-227 photolabeling in the presence of GABA compared with bicuculline results primarily from enhanced binding affinity at the ␥ ϩ -␤ Ϫ site. In the presence of GABA, R-mTFD-MPAB binds to both ␤ Ϫ sites with similar affinity (IC 50 ϭ 0.7 M). Further studies will be required to quantify the asymmetry of R-mTFD-MPAB state dependence between the ␥ ϩ -␤ Ϫ and ␣ ϩ -␤ Ϫ sites, similar to the asymmetry seen for agonist binding at the nonequivalent transmitter-binding sites in the muscle-type nicotinic acetylcholine receptor (39 -41).
the ␥ ϩ -␤ Ϫ -binding site in an orientation with the reactive diazirine in proximity to ␥2Ser-280. As seen previously in computational docking studies of TDBzl-etomidate or R-mTFD-MPAB (22,23), S-mTFD-MPPB is also predicted to bind with similar energies at each of the intersubunit interfaces in homology models based upon other homomeric pentameric ligand-gated ion channels. Thus, docking studies cannot yet provide any explanation for the preferential binding of S-mTFD-MPPB at the ␥ ϩ -␤ Ϫ interface or of the observed state dependence. Pharmacological Specificity of Binding at the ␥ ϩ -␤ Ϫ Site-Etomidate, at a concentration where it binds selectively at the ␤ ϩ -␣ Ϫ interface, allosterically inhibited photolabeling by ϳ90%. This allosteric inhibition is predicted because GABA and etomidate stabilize the same receptor state (42,43). Because R-mTFD-MPAB also binds to the ␥ ϩ -␤ Ϫ -intersubunit pocket with highest affinity in the presence of GABA, it may also inhibit S-mTFD-MPPB binding allosterically. However, in view of the proximity of the residues in the ␥ ϩ -␤ Ϫ -intersubunit pocket photolabeled by S-mTFD-MPPB and R-mTFD-MPAB, competitive inhibition is the simplest interpretation. Picrotoxinin, which binds at the cytoplasmic end of the ion channel (17,44), did not inhibit S-[ 3 H]mTFD-MPPB photolabeling. This establishes that in the purified ␣1␤3␥2 GABA A R, picrotoxinin binds preferentially to the same closed channel state as bicuculline, a result consistent with its allosteric inhibition of [ 3 H]muscimol binding to rat brain membrane fractions (45) and with recent mutational analyses (46). The strong negative coupling between S-mTFD-MPPB binding at the ␥ ϩ -␤ Ϫ site and GABA binding in the ECD or anesthetic binding at the ␤ ϩ -␣ Ϫ site necessitates care in the use of S-[ 3 H]mTFD-MPPB to identify other drugs that bind preferentially in the closed channel state to sites in the TMD. However, the differential binding properties of S-[ 3 H]mTFD-MPPB, R-[ 3 H]mTFD-MPAB, and [ 3 H]azietomidate now allow the development of assays to determine whether drugs such as the volatile convulsant and GABA A R inhibitor fluorothyl (bis[2,2,2-trifluoroethy] ether) or its anesthetic isomer and GABA A R potentiator "isofluorothyl" (1,1,1,3,3,3-hexafluoro-2-methoxypropane) (4,47,48) bind selectively to intersubunit sites in the presence of bicuculline or GABA, respectively.
Conclusions-Our novel finding is that in a ␣1␤3␥2 GABA A R the binding pocket in the TMD at the ␥ ϩ -␤ Ϫ interface is the binding site for S-mTFD-MPPB, a negative allosteric modulator in vitro and a convulsant in vivo, although R-mTFD-MPAB, an anesthetic, binds with high affinity to the same intersubunit pocket but with a different orientation and with positive coupling to GABA binding. Intersubunitbinding sites in the TMD for positive and negative allosteric modulators have been identified in nicotinic acetylcholine receptors and serotonin 5-HT 3 receptors containing cationselective channels (49,50). Also, general anesthetics of diverse chemical structure that act as GABA A R-positive allosteric modulators bind with varying selectivities to each of the intersubunit sites in the GABA A R TMD. Further studies are required to determine whether the ␥ ϩ -␤ Ϫ binding pocket has unique structural features that result in negative as well as positive allosteric modulation or whether other drugs can inhibit GABA responses by binding to the homologous sites at the other subunit interfaces.
Author Contributions-J. B. C. and K. W. M. conceived and coordinated the study. S. S. J. and J. B. C. designed and analyzed the experiments of Figs. 3-8 and 10 -11 that were performed by S. S. J. X. Z. expressed and purified GABAARs. P. Y. S. and K. S. B. synthesized chemical reagents used in the study. D. C. C. conducted the homology modeling and computational docking studies. R. D. and K. W. M. designed, performed, and analyzed electrophysiology experiments shown in Fig. 2. J. B. C. and S. S. J. wrote the paper with input from all authors. All authors approved the final version of the manuscript.