Specificity of Intersubunit General Anesthetic-binding Sites in the Transmembrane Domain of the Human α1β3γ2 γ-Aminobutyric Acid Type A (GABAA) Receptor*♦

Background: General anesthetics of diverse chemical structure potentiate GABAA receptors by binding to unknown sites. Results: A photoreactive barbiturate identifies intersubunit-binding sites distinct from, but homologous to, sites identified by photoreactive etomidate analogs. Conclusion: Propofol, barbiturates, and etomidate analogs bind with variable selectivities to two classes of sites. Significance: This study helps define the diversity of GABAA receptor general anesthetic-binding sites. GABA type A receptors (GABAAR), the brain's major inhibitory neurotransmitter receptors, are the targets for many general anesthetics, including volatile anesthetics, etomidate, propofol, and barbiturates. How such structurally diverse agents can act similarly as positive allosteric modulators of GABAARs remains unclear. Previously, photoreactive etomidate analogs identified two equivalent anesthetic-binding sites in the transmembrane domain at the β+-α− subunit interfaces, which also contain the GABA-binding sites in the extracellular domain. Here, we used R-[3H]5-allyl-1-methyl-5-(m-trifluoromethyl-diazirynylphenyl) barbituric acid (R-mTFD-MPAB), a potent stereospecific barbiturate anesthetic, to photolabel expressed human α1β3γ2 GABAARs. Protein microsequencing revealed that R-[3H]mTFD-MPAB did not photolabel the etomidate sites at the β+-α− subunit interfaces. Instead, it photolabeled sites at the α+-β− and γ+-β− subunit interfaces in the transmembrane domain. On the (+)-side, α1M3 was labeled at Ala-291 and Tyr-294 and γ2M3 at Ser-301, and on the (−)-side, β3M1 was labeled at Met-227. These residues, like those in the etomidate site, are located at subunit interfaces near the synaptic side of the transmembrane domain. The selectivity of R-etomidate for the β+-α− interface relative to the α+-β−/γ+-β− interfaces was >100-fold, whereas that of R-mTFD-MPAB for its sites was >50-fold. Each ligand could enhance photoincorporation of the other, demonstrating allosteric interactions between the sites. The structural heterogeneity of barbiturate, etomidate, and propofol derivatives is accommodated by varying selectivities for these two classes of sites. We hypothesize that binding at any of these homologous intersubunit sites is sufficient for anesthetic action and that this explains to some degree the puzzling structural heterogeneity of anesthetics.

General anesthetics of diverse structures, including volatile anesthetics, propofol, etomidate, barbiturates, steroids, and alcohols, potentiate inhibitory GABA type A receptors (GABA A R) 2 in vitro with a pharmacology and concentration dependence that suggest this receptor is a major contributor to the anesthetic state (1)(2)(3). The importance of GABA A Rs for anesthesia in vivo was demonstrated by the decreased sensitivity of "knock-in" mice bearing a single substitution at position 15 in the GABA A R ␤3 subunit transmembrane helix 2 (␤3M2-15Ј), a substitution that reduced GABA A R sensitivity to propofol and etomidate in vitro (4). These mice had greatly reduced sensitivity to the immobilizing and hypnotic anesthetic effects of etomidate, propofol, and pentobarbital, with little change in sensitivity to volatile or steroid anesthetics (5)(6)(7).
The locations of anesthetic sensitivity determinants in GABA A Rs have been predicted by use of homology models derived from the structures of other members of the Cys-loop superfamily of pentameric ligand-gated ion channels, the nicotinic acetylcholine receptor (nAChR) (8), the prokaryotic homologs ELIC (9) and GLIC (10), and an invertebrate glutamate-gated chloride channel (11). Each subunit contains an N-terminal extracellular domain, a transmembrane domain made up of a loose bundle of four transmembrane helices (M1-M4), and an intracellular domain formed by the primary structure between the M3 and M4 helices. In an (␣) 2 (␤) 2 ␥ GABA A R, the transmitter-binding sites are in the extracellular domain at the ␤ ϩ -␣ Ϫ subunit interfaces, with amino acids from the ␤ and ␣ subunits forming the principal (ϩ) and complementary (Ϫ) surfaces of the binding pocket, respectively (Fig. 1). The benzo-diazepine-binding site is at an equivalent position at the ␣ ϩ -␥ Ϫ subunit interface (12,13). In the transmembrane domain, M2 helices from each subunit associate around a central axis to form the ion channel, and amino acids from the M1 and M3 helices of adjacent subunits contribute to the subunit interfaces. The etomidate-binding sites, identified by photoaffinity labeling of amino acids in ␤M3 and ␣M1, are in the two ␤ ϩ -␣ Ϫ subunit interfaces about 50 Å below the agonist sites (14,15).
Radioligand Binding Assays-[ 3 H]Muscimol binding to purified GABA A R was measured by filtration after precipitation with polyethylene glycol (14). The total concentration of sites was determined at 500 nM [ 3 H]muscimol and with 1 mM GABA to determine nonspecific binding. Anesthetic modulation of 2-3 nM [ 3 H]muscimol binding was measured as described (15,20), except that samples were incubated for 60 min at room temperature before addition of polyethylene glycol and ␥-globulins and then filtered after a 30-min incubation at room temperature. The modulation results are presented as the percentage of the specifically bound [ 3 H]muscimol over that without modulators.
GABA A R Photolabeling-Purified GABA A R in elution buffer was photolabeled on an analytical scale (40 -80-l aliquots containing ϳ3 pmol of [ 3 H]muscimol sites) to characterize photolabeling at the subunit level and to quantify the effects of nonradioactive anesthetics (or agonist) on photolabeling. To identify photolabeled amino acids, GABA A R was photolabeled on a preparative scale (1.5-2.5-ml aliquots containing ϳ90 pmol of [ 3 H]muscimol sites). Appropriate amounts of R-[ 3 H]mTFD-MPAB or R-[ 3 H]azietomidate were transferred to glass tubes, and solvent (methanol) was evaporated under an argon stream. Freshly thawed GABA A R in elution buffer was added to the tube, and the radioligand was resuspended at 4°C with gentle vortexing for 30 min to a final concentration of 0.5-1 M (ϳ1 Ci per analytical sample and 25-45 Ci per preparative sample). Drugs of interest were added to the aliquots, and samples were incubated for 30 min. Samples were then placed in the wells of a 96-well plastic microtiter plate (analytical photolabeling) or in a plastic 3.5-cm Petri dish (preparative photolabeling) and irradiated on ice for 30 min with a 365-nm lamp (Spectroline 280L) at a distance of ϳ1 cm. Sam- ples were then solubilized at room temperature in an equal volume of sample buffer (9 parts of 40% sucrose, 10% SDS, 2% glycerol, 0.0125% bromphenol blue, 0.3 M Tris, pH 6.8, and 1 part ␤-mercaptoethanol) and fractionated by Laemmli SDS-PAGE (6% acrylamide, 0.24% bisacrylamide resolving gel). The large sample volumes in preparative photolabelings (3-5 ml) necessitated the use of 1.5-mm thick slab gels that were 12 cm long and 14 cm wide, with a 5-cm stacker layer (4% acrylamide) and wells 6 cm deep and 12 cm wide. SDS-polyacrylamide gels were stained with Coomassie Blue after electrophoresis. Prior to UV irradiation, all samples were incubated in glass vials, and anesthetic additions were made using glass microcaps. Anesthetic stock solutions were prepared in methanol, and final methanol concentrations were Յ0.5% (v/v).
In analytical scale photolabeling, the 3 H incorporation into GABA A R subunits was visualized by fluorography using En 3 hance (PerkinElmer Life Sciences) and quantified by liquid scintillation counting of excised gel bands that had been incubated in 0.1 ml of water, 0.5 ml of tissue solubilizer TS-II (RPI) overnight before addition of scintillation fluid (EcoScint A, National Diagnostics). In preparative scale photolabeling experiments, the GABA A R subunit bands excised from the stained gels were eluted individually into 12 ml of buffer (100 mM NH 4 HCO 3 , 0.1% SDS, and 2.5 mM dithiothreitol, pH 8.4) for 3 days at 20°C with gentle agitation. The eluates were filtered, concentrated, acetone-precipitated, and resuspended in 100 -200 l of digestion buffer (15 mM Tris and 0.1% SDS, pH 8.5).
Photolabeled amino acids were identified in three preparative photolabeling experiments using purified ␣1␤3␥2 GABA A R (ϳ60 nM [ 3 H]muscimol sites, eluted in a buffer containing 10 mM CHAPS and 0.86 mM asolectin). GABA A Rs were photolabeled with the following: (i) H]azietomidate. Subunit photolabeling was quantified as a function of the total concentration of nonradioactive anesthetics. Because GABA A Rs were photolabeled in solutions containing 5 mM CHAPS, 0.4 mM asolectin, the anesthetic free concentrations will be substantially lower than the total concentrations and dependent upon the anesthetic lipophilicity (oil/buffer partition coefficient).
For conditions when an anesthetic only inhibited GABA A R photolabeling, the data were fit to a single site model for competitive inhibition, as shown in Equation 1, where f 1 (x) is the 3 H counts/min incorporated in a subunit at anesthetic concentration x; f 0 is the subunit counts/min in the absence of inhibitor; f ns is the nonspecific subunit photolabeling, and IC 50 is the total drug concentration reducing photolabeling by 50%. When a drug only enhanced photolabeling, data were fit to Equation 2, where f 2 (x) is the counts/min incorporated at drug concentration x; f MAX is the maximal level of photolabeling in counts/ min; f 0 is the subunit photolabeling in counts/min in the absence of drug, and EC 50 is the total drug concentration producing 50% of maximal enhancement. If an anesthetic at low concentrations produced an enhancement of photolabeling and then inhibition at high concentrations, data were fit to a model assuming anesthetic binding to independent potentiating and inhibitory sites (Equation 3), Chemical and Enzymatic Fragmentation-Aliquots isolated from gel bands enriched in either ␣1 or ␤3 subunits were digested at 20°C with either 100 g of endoproteinase Glu-C (EndoGlu-C, Worthington) for 2 days or 0.5 units of endoproteinase Lys-C (EndoLys-C, Roche Applied Science) for 2 weeks. For chemical cleavage at the C terminus of methionines, samples immobilized on PVDF sequencing filters were treated with cyanogen bromide as described (21,22). For chemical cleavage at the C terminus of tryptophans, samples on PVDF filters were treated with BNPS-skatole as described (23), except that after precipitation of the excess BNPS-skatole, the digestion solution was loaded onto a second PVDF filter, and material on the two filters was sequenced simultaneously (15).
HPLC Purification and Protein Microsequencing-Reversedphase HPLC was performed as described (24) on an Agilent 1100 binary pump system using a Brownlee Aquapore BU-300 column. Samples were eluted at 0.2 ml/min with increasing concentrations of 60% isopropyl alcohol, 40% acetonitrile, 0.05% TFA. Elution of peptides was monitored by the absorbance at 215 nm and by liquid scintillation counting of a 10% aliquot of each 0.5-ml fraction.
Samples were sequenced on an Applied Biosystems Procise 492 protein sequencer modified to collect two-thirds of each cycle for PTH-derivative detection/quantification and onethird for 3 H determination by liquid scintillation counting. For direct sequencing of intact subunits or subunit digests containing SDS, samples were loaded onto Prosorb PVDF filters (Applied Biosystems) following the manufacturer's instructions. HPLC fractions for sequence analysis were drop-loaded at 45°C onto TFA-treated glass fiber filters that were then treated with Biobrene TM . For selected samples, the sequencer was paused after the designated cycles, and the sample filter was treated with o-phthalaldehyde (OPA) before resuming sequencing as follows: (i) to block all free N termini before treatment of the filter with cyanogen bromide or (ii) to chemically isolate for further sequencing only those fragments containing a proline in the designated cycle. OPA reacts with primary amines, but not with secondary amines, and treatment with OPA prevents further sequencing of fragments not containing a proline at that cycle, thereby confirming that any subsequent peak of 3 H release originated from the proline-containing peptide (25,26). PTH-derivatives were quantitated by peak heights over background, and the actual picomole quantities and counts/min detected are plotted in the figures. The amount of a peptide sequenced was determined by fitting the individual residues detected to Equation 4, where I x is the picomoles detected in cycle x; I 0 is the initial amount of peptide, and R is the repetitive yield. Cys, Ser, His, and Trp were not used for the fit due to known problems with their quantifications. The efficiency of amino acid photolabeling in counts per min/pmol (cpm/pmol) was calculated by Equation 5, where cpm x is the counts/min in cycle x.
Molecular Modeling-Comparison of structural models for ␣1␤3 GABA A R constructed by homology with GLIC and GluCl (15) established that the positions of amino acids of ␤M3/␣M1 and ␣M3/␤M1 contributing to the ␤ ϩ -␣ Ϫ and ␣ ϩ -␤ Ϫ interfaces, respectively, were the same. Compared with the GLICderived structure, there was an increased distance in the GluCl structure between the M3 and M1 helices where the allosteric potentiator ivermectin is bound in GluCl. Because etomidate could be docked within the more constrained intersubunit pocket of the GLIC-derived model, we constructed a ␤3␣1␤3␣1␥2 GABA A R homology model based on a GLIC structure (Protein Data Bank code 3P50) as described (15), with the exception that the human ␥2 subunit sequence replaced the third ␤3 subunit sequence in that ␤3␣1␤3␣1␤3 model. In the GLIC structure, Tyr-263, the M3 residue homologous to the GABA A R-photolabeled residues ␣1Tyr-294, ␤3Phe-289, and ␥2Phe-304, is within 4 Å of the residue in the M1 helix across the interface, Pro-204. Therefore, to accommodate R-mTFD-MPAB or R-azietomidate in their interface binding pockets in the ␤3␣1␤3␣1␥2 GABA A R homology model, the aromatic side chains (␣1Tyr-294, ␤3Phe-289, and ␥2Phe-304) were rotated out of the interface. A membrane force field was calculated for the structure using the Discovery Studio molecular modeling package (Accelrys Inc.), and the model was energy-minimized with anesthetics occupying each of the five transmembrane interface pockets. R-mTFD-MPAB was placed horizontally in the ␣ ϩ -␤ Ϫ , ␥ ϩ -␤ Ϫ , and ␣ ϩ -␥ Ϫ interfaces, and R-azietomidate was placed horizontally in the two ␤ ϩ -␣ Ϫ interfaces, with their diazirines protruding between the M3 ϩ and M1 Ϫ ␣-helices in close proximity to the photolabeled residues. CDocker, a CHARMm-based molecular dynamics simulated annealing program, was used to dock R-mTFD-MPAB at each interface pocket using an 11-Å radius binding-site sphere cen-tered on each minimized anesthetic. The best 200 -300 solutions were collected for each interface starting from 50 random orientations of 50 molecular dynamics-altered anesthetic structures. Solutions were obtained at all interfaces.
For R-mTFD-MPAB (volume, 275 Å 3 ) at the ␣ ϩ -␤ Ϫ interface site, all 300 solutions were oriented similarly, with the major difference being the location of the diazirine as determined by a 180°rotation of the phenyl group around the C5-phenyl bond. The Connolly surface, determined by a probe of radius 1.4 Å, for the 300 solutions defined a volume of 535 Å 3 . The lowest energy solution was positioned with the diazirine carbon within 4.5 Å from the photolabeled residues ␤3Met-227 and ␣1Tyr-294, the phenyl ring stacked with ␣1Tyr-294, the N-methyl of barbituric acid within 4 Å of ␣1Ser-270 (␣M2-15Ј) and ␣1Tyr-294, and the C5 allyl within 4 Å of ␤3Ile-264 (␤3M2-14Ј). CDocker interaction energies overlapped for the two orientations, with the lowest 15 solutions differing by 2 kcal/mol and all 300 solutions differing by 10 kcal/mol.

Photolabeling ␣1␤3 and ␣1␤3␥2 GABA A Rs with R-[ 3 H]Azietomidate and R-[ 3 H]mTFD-MPAB-The
To begin characterizing anesthetic-binding sites in the ␣1␤3␥2 GABA A R, we photolabeled samples with R-[ 3 H]azietomidate or R-[ 3 H]mTFD-MPAB ( Fig. 2) at anesthetic concentrations and compared the patterns of subunit photolabeling to those seen for the ␣1␤3 GABA A R (15,16). When GABA A R subunits were resolved by SDS-PAGE after photolabeling, the two preparations appeared essentially the same based upon Coomassie Blue stain, with three bands migrating at ϳ56, ϳ59, and ϳ61 kDa ( Fig. 2A). For the ␣1␤3 GABA A R, the N-terminal sequence analyses had established that the ϳ56-kDa band contained the FLAG-tagged ␣1 subunit, whereas the 59-and 61-kDa bands contained ␤3 subunits differing in their glycosylation patterns, with ␤3 subunit in the ␣1 band and ␣1 subunit in the ␤3 bands at ϳ15% levels (15 (14,15) to identify photolabeling within each of the four transmembrane helices of the ␤3 subunit, the subunit with the highest 3 H incorporation. When an EndoLys-C digest of material enriched in ␤3 subunits (eluted from the 59-and 61-kDa gel bands) was fractionated by reversed-phase HPLC (Fig. 4A), all 3 H was recovered in hydrophobic fractions, consistent with  photolabeling restricted to the GABA A R transmembrane domain. N-terminal sequencing of the pool of the fractions containing the peak of 3 H identified a fragment beginning at ␤3Arg-216 near the beginning of the M1 helix as the primary sequence, with a peak of 3 H release in cycle 12 consistent with photolabeling of ␤3Met-227 (Fig. 4B). Based upon the amounts of 3 H and ␤3Met-227 released at that cycle, ␤3Met-227 was photolabeled at a calculated efficiency of 980 cpm/pmol (ϳ3% of ␤3 labeled), and photolabeling was inhibited by Ͼ80% in the presence of pentobarbital.
Because the sequenced samples also contained a fragment beginning at ␤3Ala-280, before ␤M3, at ϳ15% the level of the primary sequence, we used an alternative sequencing strategy to confirm the pentobarbital-inhibitable photolabeling of ␤3Met-227. With ␤3Met-227 positioned in the subunit primary structure 37 amino acids after ␤3Glu-190 and a proline (␤3Pro-206) in between, we took advantage of the fact that OPA, which reacts with primary amines but not proline, a secondary amine, can be used to prevent further Edman degradation of any peptide not containing a proline at the cycle of treatment (25,26). When an EndoGlu-C digest of material enriched in photolabeled ␤3 subunit was sequenced, after treatment with OPA at cycle 16, the only sequence remaining began originally at ␤3His-191. The observed peak of 3 H release in cycle 37 confirmed that ␤3Met-227 was photolabeled at 920 cpm/pmol and that 1 mM pentobarbital reduced its labeling by ϳ80% to 170 cpm/pmol (Fig. 4C).

R-[ 3 H]mTFD-MPAB Photolabels ␣1Ala-291 and ␣1Tyr-294
in ␣1M3-Although the amino acids photolabeled by R-[ 3 H]azietomidate (␤3Met-286 in ␤M3 and ␣1Met-236 in ␣M1) are located in the GABA A R structure at the ␤ ϩ -␣ Ϫ subunit interfaces (15), the amino acid photolabeled by R-[ 3 H]mTFD-MPAB, ␤3Met-227, is located within the ␤3M1 helix at the ␣ ϩ -␤ Ϫ and ␥ ϩ -␤ Ϫ subunit interfaces in proximity to amino acids from ␣1M2/M3 or ␥2M2/M3. To determine whether there was also photolabeling of amino acids in ␣1M3, we devised a strategy to sequence a fragment beginning at ␣1Asp-287 at the M3 N terminus that entailed the rpHPLC fractionation of an EndoGlu-C digest of material enriched in ␣1 subunits, the use of OPA, and digestion with cyanogen bromide to cleave after methionines (Fig. 5, A-C). When the fragment beginning at ␣1Asp-287 was sequenced (Fig. 5A), the peaks of 3 H release in cycles 5 and 8 indicated R-[ 3 H]mTFD-MPAB photolabeling of ␣1Ala-291 and ␣1Tyr-294 at photolabeling efficiencies of ϳ50 cpm/pmol. This identification was confirmed by sequencing for 50 cycles a fragment beginning at ␣Ser-251 before ␣1M2, produced by digestion with EndoGlu-C, with OPA treatments prior to cycles 3 and 28 corresponding to ␣1Pro-253 and ␣1Pro-278. Peaks of 3 H release in cycles 41 and 44 confirmed photolabeling of ␣1Ala-291 (ϳ90 cpm/ pmol) and ␣1Tyr-294 (ϳ60 cpm/pmol), and 1 mM pentobarbital inhibited incorporation into both residues by Ͼ80% (data not shown).  [25][26][27] was sequenced for four cycles, establishing that the primary sequence began at ␣1Ser-251 before ␣1M2, a fragment predicted to extend to ␣1Glu-313 near the C terminus of ␣1M3 (C). After cycle 4, the sample was treated with OPA to block all free N termini, which was confirmed by five more cycles of Edman degradation, and then treated with cyanogen bromide to cleave at methionines before sequencing for 15 additional cycles. D, 3 H (F) and picomoles of PTH-derivatives (Ⅺ) released during Edman sequencing of a GABA A R subunit fragment beginning at ␥2Asp-297 (0.6 pmol). The peak of 3 H release in cycle 5 indicated labeling of ␥2Ser-301. Material isolated by rpHPLC from an EndoGlu-C digest of 59 -61-kDa gel bands (E, fractions 28 -29)) was sequenced for 10 cycles, establishing the presence of the fragment beginning at ␥2Val-212 (F) as a secondary sequence along with the primary sequence beginning at ␤3His-191. The sample was treated with OPA after cycle 10 to block all free N termini, sequenced an additional 5 cycles to confirm block, then treated with cyanogen bromide, and sequenced for an additional 15 cycles. The efficiencies of photolabeling of the residues are tabulated in Table 5.

R-[ 3 H]mTFD-MPAB
Photolabels ␥2Ser-301 in ␥2M3-To characterize photolabeling in ␥2M3, we sequenced the fragment beginning at ␥2Asp-297 by use of a protocol similar to that used to sequence the homologous ␣1Asp-287 fragment (Fig. 5, D-F). Material recovered from an rpHPLC fractionation of an EndoGlu-C digest of labeled subunits was sequenced, N-terminally blocked, treated with cyanogen bromide, and resequenced. To maximize the amount of ␥2 and minimize the amount of ␣1 subunit, material was used from the ␤ subunit gel bands that contain more ␥2 than ␣1 subunit. Rather than use the rpHPLC fractions where the ␣1Ser-251 fragment had eluted (Fig. 5B), we used fractions eluting at higher organic solvent that contained the peak of 3 H (from photolabeled ␤3Met-227 in the ␤3His-191 fragment) and the ␥2Val-212 fragment that begins before ␥2M1 and extends through M3 (Fig. 5, E and F). When that material was sequenced after cyanogen bromide digestion (Fig. 5D), the fragment beginning at ␥2Asp-296 was present as a secondary sequence, with the primary sequence beginning at ␤3Pro-228 and no detectable ␣1 subunit sequences. There was a peak of 3 H release in cycle 5, the cycle that contained ␤3Ile-232 from the primary sequence and ␥2Ser-301 from the secondary sequence. Because there was no evidence of photolabeling of ␤3Ile-232 (Fig. 4B, cycle 17), the peak of 3 H release in cycle 5 indicated R-[ 3 H]mTFD-MPAB photolabeling of ␥2Ser-301, the amino acid in ␥M3 homologous to ␣1Ala-291. We confirmed this identification by using a protocol that took advantage of the unique distributions of Trp and Pro in the three subunits in the M2-M3 region to chemically isolate ␥2M3 during sequencing. When labeled subunits from gel bands enriched in either ␣1 or ␤3 were treated with BNPS-skatole to cleave at tryptophans and sequenced for 50 cycles with OPA treatment at cycle 7, a peak of 3 H release was seen in cycle 45 that confirmed R-[ 3 H]mTFD-MPAB photolabeling of ␥2Ser-301 at ϳ100 cpm/pmol (data not shown).

R-[ 3 H]mTFD-MPAB Binds to
Sites at the ␣ ϩ -␤ Ϫ and ␥ ϩ -␤ Ϫ Subunit Interfaces Equivalent to the Etomidate-binding Site at the ␤ ϩ -␣ Ϫ Subunit Interfaces-The high degree of amino acid sequence conservation between the GABA A R M1-M4 helices and those of GLIC or GluCl allows simple and consistent align-ment of those GABA A R regions in homology models based upon GLIC or GluCl (15). In an ␣1␤3␥2 GABA A R homology model based upon the structure of GLIC (Fig. 6), the residues photolabeled by R-[ 3 H]mTFD-MPAB are located in two different subunit interfaces (␣ ϩ -␤ Ϫ and ␥ ϩ -␤ Ϫ ) (Fig. 6C). In the ␣ ϩ -␤ Ϫ interface, ␤3Met-227 in the M1 helix is opposite both ␣1Ala-291 and ␣1Tyr-294 in the M3 helix and located between them on an axis perpendicular to the membrane, whereas in the ␥ ϩ -␤ Ϫ interface it is opposite ␥2Ser-301 in M3 and slightly below it. In both cases there is a pocket between the subunits that is large enough to accommodate R-mTFD-MPAB (volume of 275 Å 3 ). Shown in Fig. 6, D-F, are expanded views of these binding sites with R-mTFD-MPAB docked in the lowest energy orientation predicted by computational docking. R-mTFD-MPAB was predicted to bind with its reactive diazirine positioned in close proximity to the photolabeled amino acids in ␤3M1 and ␣1M3 or ␥2M3, the NCH 3 group of barbituric acid oriented toward ␣1M2-15Ј or ␥2M2-15Ј (␣1Ser-270/␥2Ser-280), and the C5 allyl group oriented toward ␤3M2-10Ј and ␤3M2-14Ј (␤3Thr-260/␤3Ile-264). ␤3Pro-228 in ␤3M1 is predicted to be a major determinant of the shape of this binding pocket, as noted previously for the homologous proline in ␣1M1 (␣1Pro-233) in the etomidate-binding site at the ␤ ϩ -␣ Ϫ interface (15).
In addition to the four intersubunit-binding sites identified by R-[ 3 H]azietomidate and R-[ 3 H]mTFD-MPAB, there is a fifth potential site in the transmembrane domain at the ␣ ϩ -␥ Ϫ subunit interface, the same interface that in the extracellular domain contains the benzodiazepine site. This site may be photolabeled by R-[ 3 H]mTFD-MPAB, because the residues it photolabeled in ␣M3 at the ␣ ϩ -␤ Ϫ interface are present in the second ␣ subunit at the ␣ ϩ -␥ Ϫ interface. We have not yet been able to characterize photolabeling in ␥M1, which is necessary to determine whether this fifth intersubunit site is also photolabeled.
Selectivities of Etomidates and Barbiturates for Intersubunitbinding Sites-To determine whether the Ͼ50-fold selectivity of R-mTFD-MPAB and ϳ10-fold selectivity of pentobarbital for the anesthetic-binding sites at ␣ ϩ /␥ ϩ -␤ Ϫ interfaces and the Ͼ100-fold selectivity of R-etomidate for the sites at the ␤ ϩ -␣ Ϫ interfaces were general properties of barbiturates and etomidates, we screened other etomidates ( Table 2) and barbi-turates ( Table 3)

as inhibitors of R-[ 3 H]azietomidate and R-[ 3 H]mTFD-MPAB photolabeling.
With this assay, only qualitative comparisons can be made of the potencies of different anesthetics, because IC 50 values were determined from total rather than free concentrations, and the anesthetics vary greatly in lipophilicity, as evidenced by the ϳ100-fold range of partition coefficients (Tables 2 and 3). However, the ratio of IC 50 values of stereoisomer pairs for a site or of each anesthetic for the two binding sites will not depend on the differences in anesthetic partition coefficients.
S-Etomidate binds preferentially to the same site as R-etomidate, but with 10-fold lower affinity. Not surprisingly, R-azietomidate and R-TDBzl-etomidate bind with Ͼ25-fold selectivity to the sites at the ␤ ϩ -␣ Ϫ interfaces. However, the presence of a bulky substituent on the etomidate phenyl ring is not tolerated, as the affinity of S-or R-pTFD-etomidate for the ␤ ϩ -␣ Ϫ sites was reduced by 50-fold compared with R-TDBzl-etomidate, and both isomers bound with 2-fold higher affinity to the ␣ ϩ /␥ ϩ -␤ Ϫ interface sites than to the ␤ ϩ -␣ Ϫ sites. Similar to pentobarbital, phenobarbital bound with ϳ10-fold selectivity to the binding sites at ␣ ϩ /␥ ϩ -␤ Ϫ interfaces, but addition of bulk to the ring, as in thiopental, reduced the selectivity to only 1.6-fold, and brallobarbital bound with 3-fold higher affinity to the "etomidate"-binding site. In contrast to the ϳ60fold binding selectivity of R-mTFD-MPAB, S-mTFD-MPAB bound with lower affinity and nonselectively to both classes of sites. We also examined the effects of stereoisomers of MPPB, as R-MPPB acts as an anesthetic and GABA A R potentiator, although S-MPPB acts as a convulsant and GABA A R inhibitor (28,29). The anesthetic isomer bound with 9-fold higher affinity to the sites at the ␣ ϩ /␥ ϩ -␤ Ϫ interfaces than at the ␤ ϩ -␣ Ϫ interfaces, and similarly to R-and S-mTFD-MPAB, the difference between and R-and S-MPPB was the decreased affinity of S-MPPB for the sites at the ␣ ϩ /␥ ϩ -␤ Ϫ interfaces.

TABLE 3 Affinities of barbiturates for GABA A R anesthetic-binding sites at the ␤ ؉ ؊␣ ؊ (R-3 Hazietomidate) and ␣ ؉ /␥ ؉ ؊␤ ؊ (R 3 HmTFD-MPAB) subunit interfaces a
a IC 50 values, the total anesthetic concentrations resulting in 50% inhibition of GABA A R photolabeling, were determined as described under "Experimental Procedures; EC 50 for anesthesia indicates tadpole loss of righting reflex. b See Ref. 16 (Fig. 7A, dashed line).
We also determined the inhibition of R-[ 3 H]mTFD-MPAB and R-[ 3 H]azietomidate photolabeling in the presence of GABA by 2,6-di-sec-butylphenol, a propofol analog that is similar in potency to propofol as a GABA A R potentiator and anes-thetic (EC 50 ϭ 2 M), and 2,6-di-tert-butylphenol, which at 300 M was inactive as a GABA A R modulator or anesthetic and did not alter responses to propofol (30). 2,6-di-sec-Butylphenol was equipotent as an inhibitor of photolabeling by both photoprobes (IC 50 ϭ 90 M), although the inactive isomer, 2,6-ditert-butylphenol, at 300 M inhibited photolabeling by Ͻ10% (Fig. 7, C and D). Because our experimental IC 50 values are determined from total, rather than free, drug concentrations and the partition coefficient of 2,6-di-sec-butylphenol (or 2,6di-tert-butylphenol) is 6-fold greater than that of propofol (Table 4), it is not possible to determine from our data whether 2,6-di-sec-butylphenol is actually more or less potent than propofol. However, differences in hydrophobicity (partition coefficient) cannot account for the capacity of 2,6-di sec-butylphenol to act as an anesthetic and bind to the GABA A R intersubunit-binding sites, although 2,6-di-tert-butyl phenol neither acts as an anesthetic nor binds to the intersubunit anesthetic-binding sites.
Interactions of Alphaxalone and Octanol with Intersubunit Anesthetic-binding Sites-In the presence of GABA, alphaxalone, a synthetic anesthetic steroid, at concentrations up to 30 M had little or no effect on photolabeling by R-[ 3 H]mTFD-

TABLE 4
Affinities of propofol analogs for GABA A R anesthetic-binding sites at the ␤ ؉ ؊ ␣ ؊ (R-3 Hazietomidate) and ␣ ؉ /␥ ؉ ؊ ␤ ؊ (R-3 HmTFD-MPAB) subunit interfaces a a IC 50 values, the total anesthetic concentrations resulting in 50% inhibition of GABA A R photolabeling, were determined as described under "Experimental Procedures"; EC 50 for anesthesia indicates tadpole loss of righting reflex. b See Ref. 30. 2,6-di-tert-Butylphenol at 200 M had no effect alone and did not alter propofol's anesthetic EC 50 . c Less than 10% inhibition at 300 M, which is the highest concentration tested (Fig. 7). MPAB (Fig. 7E) or R-[ 3 H]azietomidate (Fig. 7F). In the absence of GABA, alphaxalone increased photolabeling at both sites with EC 50 values of ϳ500 nM, similar to the potentiation of R-[ 3 H]azietomidate photolabeling of brain GABA A R by alphaxalone or neurosteroids (31). Alphaxalone binds neither to the R-[ 3 H]azietomidate nor R-[ 3 H]mTFD-MPAB-binding site, which is consistent with early studies demonstrating additive effects of alphaxalone and pentobarbital (32).
Octanol acts as an anesthetic and GABA A R potentiator with EC 50

Effects of GABA and Etomidate on R-[ 3 H]mTFD-MPAB Photoincorporation at the Amino Acid Level-For R-[ 3 H]azietomi-
date-photolabeled GABA A R purified from bovine brain, the enhancement of photolabeling seen at the subunit level in the presence of GABA or a neurosteroid, as well as the inhibition of photolabeling in the presence of propofol, was also seen at the level of the photolabeled amino acids (14,31,34). To determine whether this was also true for the R-[ 3 H]mTFD-MPAB site or whether novel amino acids were photolabeled when subunit photolabeling was enhanced, we characterized photolabeling in ␤M1, ␣1M3, ␥2M3, and ␤3M3 for ␣1␤3␥2 GABA A Rs photolabeled in three conditions as follows: control (no additional drug), ϩ1 mM GABA, or ϩ100 M etomidate ( Table 5). The incorporation at ␤3Met-277 within ␤3M1, the residue that accounts for Ͼ80% of GABA A R photolabeling, closely paralleled the labeling seen at the subunit level. GABA and etomidate increased photolabeling efficiency by ϳ50%, and no novel residues were photolabeled in ␤3M1. The complex sequencing protocols required to identify photolabeling in ␣1M3 or ␥2M3 made quantification more difficult. Qualitatively, GABA increased photolabeling of ␣1Ala-291, ␣1Tyr-294, and ␥2Ser-301, and no other amino acids were photolabeled. Additional labeling experiments would be necessary to assess the smaller effects of etomidate on those residues.

DISCUSSION
In this report we provide the first demonstration that there are two structurally related, but pharmacologically distinct, classes of intersubunit general anesthetic-binding sites in the transmembrane domain of human ␣1␤3␥2 GABA A Rs. The binding sites for R-[ 3 H]mTFD-MPAB, a photoreactive barbiturate that acts as a potent, stereoselective GABA A R potentiator and general anesthetic, are located at the ␣ ϩ -␤ Ϫ and ␥ ϩ -␤ Ϫ subunit interfaces, centered three helical turns down from the extracellular end of ␤3M3 (Fig. 6). At anesthetic concentrations, R-mTFD-MPAB does not bind at the previously characterized etomidate-binding sites (14,15), which are located at the two ␤ ϩ -␣ Ϫ subunit interfaces and are also centered three turns down from the extracellular end of ␣1M3. Conversely, R-etomidate does not bind at the R-mTFD-MPAB-binding sites. Thus, R-mTFD-MPAB binds to homologous but distinct sites from etomidate and its photoreactive derivatives.
Pharmacology of the Two Classes of General Anesthetic-binding Sites-R-mTFD-MPAB and R-etomidate each bind with Ͼ50-fold selectivity to their preferred sites, with IC 50 values similar to the EC 50 values for GABA A R potentiation in vitro or anesthesia in vivo. Displacing these ligands with nonradioactive anesthetics (see IC 50 values in Tables 2-4) lead to the conclusion that the two classes of sites are not simply etomidate or "barbiturate" sites. For example, pentobarbital and phenobarbital bound to the ␣ ϩ /␥ ϩ -␤ Ϫ sites with ϳ10-fold selectivity, whereas thiopental and S-mTFD-MPAB bound with similar affinity to both sites. Furthermore, the barbiturate brallobarbital had an ϳ3-fold higher preference for the etomidate (␤ ϩ -␣ Ϫ ) site, and pTFD-etomidate had 2-fold preference for the barbiturate (␣ ϩ /␥ ϩ -␤ Ϫ ) site. Thus, we refer to these sites by their subunit interface designations. There is precedent for a pharmacological class of anesthetics not binding to isosteric sites in the Cys loop ligand-gated ion channel superfamily. Although some barbiturates that inhibited currents in muscle type nAChRs fully displaced [ 14 C]amobarbital binding, others bound to an unidentified site (35).
Propofol bound with little selectivity at both classes of sites, suggesting it has at least four binding sites. Although the IC 50 values for R-azietomidate or R-mTFD-MPAB binding are close to anesthetic concentrations, the IC 50 values for propofol binding to either class of sites (ϳ40 M) are ϳ20-fold higher than GABA modulatory or anesthetic concentrations (36). This discrepancy might result if propofol binds with higher affinity to as

of R-͓ 3 H͔mTFD-MPAB photoincorporation into residues in the ␣1␤3␥2 GABA A R (cpm/pmol of PTH-derivative)
The efficiency of photolabeling of a residue (in cpm/pmol) was calculated using Equation 5 (see under "Experimental Procedures"). The data are presented as mean (Ϯrange) when two samples were sequenced. Other values were determined from the sequencing of single samples. ND indicates not determined. yet unidentified sites in the GABA A R. However, 2,6-di-sec-butyl phenol, which is equipotent with propofol as an anesthetic and GABA A R modulator (30), binds with potency similar to propofol to the two classes of intersubunit anesthetic-binding sites, although 2,6-di-tert-butylphenol, which is inactive as an anesthetic and GABA A R modulator, did not bind to either class of sites (Table 4). These results make it likely that the four intersubunit sites identified by R-[ 3 H]azietomidate and R-[ 3 H]mTFD-MPAB are the binding sites important for propofol's anesthetic effects. Interestingly, the potentiation and direct activation by propofol, which has little or no subunit interface selectivity, is best fit with a model that requires three equivalent binding sites, whereas etomidate only requires two (37,38). The fact that propofol binds nonselectively to four sites was unexpected, as previous mutational analyses identified propofol sensitivity determinant positions (␤M2-15Ј and ␤3Met-286) that in our GABA A R homology model are in the anestheticbinding sites at the ␤ ϩ -␣ Ϫ interfaces (39,40), although the homologous ␣ subunit substitutions in the ␣ ϩ -␤ Ϫ -binding site had little if any effect (41). However, there are two ␤ ϩ -␣ Ϫ interface-binding sites in an ␣␤␥ GABA A R and only one ␣ ϩ -␤ Ϫ interface site. In future studies it will be important to determine the effects of simultaneous substitutions at the ␣ ϩ /␥ ϩ -␤ Ϫ interface sites on the sensitivity to propofol or other anesthetics binding to those sites.
Because ␤M2-15Ј is predicted to be an important determinant of the shape of the etomidate-binding site at the ␤ ϩ -␣ Ϫ interface (Fig. 6G) (15) and pentobarbital binds with ϳ8-fold higher selectivity to the ␣ ϩ /␥ ϩ -␤ Ϫ sites, it is surprising that the anesthetic responses of pentobarbital are reduced in the ␤3N265M knock-in mouse (6). This may indicate that the ␤ ϩ -␣ Ϫ sites make a greater energetic contribution to the stabilization of GABA A R in the open state. Characterization of the anesthetic effects of R-mTFD-MPAB on the ␤3N265M GABA A R in vitro and in vivo will clarify whether the substitution prevents transduction of changes initiated by binding to the ␣ ϩ /␥ ϩ -␤ Ϫ subunit interfaces.
Intrasubunit Sites?-Propofol inhibits the nAChR and the prokaryotic homolog GLIC, and in those proteins it binds to intrasubunit-binding sites within the pocket formed by the transmembrane helix bundle (45,46 (15) provided no evidence of GABA A R intrasubunit-binding sites for those anesthetics, even though we sequenced through each of the ␣ and ␤ subunit transmembrane helices. In these peptides, we observed minor labeling of ␤3Met-286 and Phe-289 in the ␤ ϩ -␣ Ϫ anesthetic-binding site at ϳ2% the efficiency of ␤1Met-227. Thus, we can state that, if any intrasubunit labeling occurred, it must be at levels below this.
Anesthetics and GABA A R Conformational Equilibria-At lower concentrations, most general anesthetics potentiate GABA responses, and at higher concentrations, they directly activate GABA A Rs in the absence of GABA. Direct activation and potentiation of nAChRs and GABA A Rs can be well accounted for by allosteric models that assume that receptors exists in multiple, interconvertible conformational states (47)(48)(49)(50). Activators and potentiators shift the conformational equilibria toward the open channel state because they bind with higher affinity to open states than to resting, closed channel states. In purified GABA A R in detergent/lipid micelles, positive energetic coupling between the extracellular and transmembrane domains is preserved as evidenced by anesthetic enhancement of [ 3 H]muscimol binding and GABA enhancement of R-[ 3 H]azietomidate/R-[ 3 H]mTFD-MPAB photolabeling. Furthermore, in the absence of GABA, R-etomidate enhances R-[ 3 H]mTFD-MPAB photolabeling, and reciprocally, R-mTFD-MPAB enhances R-[ 3 H]azietomidate photolabeling. Our studies provide no information about the state-dependent differences in affinity for anesthetics binding at either class of sites. However, smaller differences in binding affinity between open (K o ) and closed states (K c ) are required for anesthetics binding to four rather than two sites, because the shift in conformational equilibria will be proportional to (K o /K c ) n , where n is the number of sites.
Because R-etomidate does not bind to the ␣ ϩ /␥ ϩ -␤ Ϫ sites even at 1 mM, our results provide further evidence that R-etomidate directly activates GABA A Rs (17) by binding solely to the ␤ ϩ -␣ Ϫ interfaces that also contain the agonist-binding sites in the extracellular domain (14). The selective binding of R-mTFD-MPAB to the ␣ ϩ /␥ ϩ -␤ Ϫ subunit interfaces provides the first evidence that potentiation and direct activation (16) can result from anesthetic binding at interfaces not containing the transmitter-binding site.
Conclusions-Our novel finding is that it is possible to synthesize general anesthetics that are selective for sites between specific subunits in the transmembrane domain of pentameric GABA A Rs. A wide range of general anesthetic structures target these four sites but with variable selectivity, which offers an explanation of the puzzling lack of well defined structure activity relationships among general anesthetics (51)(52)(53). These observations suggest that it may be possible to develop agents with novel intersubunit specificity that can be used to target specific nerve pathways and behaviors in a subunit-dependent manner (7). A similar strategy has recently been proposed for the extracellular domain where a potentiator site has been identified at the ␣ ϩ -␤ Ϫ interface in a pocket equivalent to the transmitter and benzodiazepine sites at the ␤ ϩ -␣ Ϫ and ␣ ϩ -␥ Ϫ subunit interfaces (13,54).