Spontaneous Cross-link of Mutated α1 Subunits during GABAA Receptor Assembly*

γ-Aminobutyric acid, type A (GABAA) receptor α1 subunits containing a cysteine mutation at a position in the channel mouth (H109C) surprisingly formed a spontaneous cross-link with each other in receptors composed of α1H109C, β3, and γ2 subunits. Cross-linking of two α1H109C subunits did not significantly change the affinity of [3H]muscimol or [3H]Ro15-1788 binding in α1H109Cβ3γ2 receptors, but GABA displayed a reduced potency for activating chloride currents. On reduction of the disulfide bond, however, GABA activation as well as diazepam modulation was similar in mutated and wild-type receptors, suggesting that these receptors exhibited the same subunit stoichiometry and arrangement. Disulfide bonds could not be reoxidized by copper phenanthroline after having been reduced in completely assembled receptors, suggesting that cross-linking can only occur at an early stage of assembly. The cross-link of α1H109C subunits and the subsequent transport of the resulting homodimers to the cell surface caused a reduction of the intracellular pool of α1H109C subunits and a reduced formation of completely assembled receptors. The formation of α1H109C homodimers as well as of correctly assembled GABAA receptors containing cross-linked α1H109C subunits could indicate that homodimerization of α1 subunits via contacts located in the channel mouth might be one starting point of GABAA receptor assembly. Alternatively the assembly mechanism might have started with the formation of heterodimers followed by a cross-link of mutated α1 subunits at the heterotrimeric stage. The formation of cross-linked α1H109C homodimers would then have occurred independently in a separate pathway.

␥-Aminobutyric acid, type A (GABA A ) receptor ␣1 subunits containing a cysteine mutation at a position in the channel mouth (H109C) surprisingly formed a spontaneous cross-link with each other in receptors composed of ␣1H109C, ␤3, and ␥2 subunits. Cross-linking of two ␣1H109C subunits did not significantly change the affinity of [ 3 H]muscimol or [ 3 H]Ro15-1788 binding in ␣1H109C␤3␥2 receptors, but GABA displayed a reduced potency for activating chloride currents. On reduction of the disulfide bond, however, GABA activation as well as diazepam modulation was similar in mutated and wild-type receptors, suggesting that these receptors exhibited the same subunit stoichiometry and arrangement. Disulfide bonds could not be reoxidized by copper phenanthroline after having been reduced in completely assembled receptors, suggesting that cross-linking can only occur at an early stage of assembly. The cross-link of ␣1H109C subunits and the subsequent transport of the resulting homodimers to the cell surface caused a reduction of the intracellular pool of ␣1H109C subunits and a reduced formation of completely assembled receptors. The formation of ␣1H109C homodimers as well as of correctly assembled GABA A receptors containing cross-linked ␣1H109C subunits could indicate that homodimerization of ␣1 subunits via contacts located in the channel mouth might be one starting point of GABA A receptor assembly. Alternatively the assembly mechanism might have started with the formation of heterodimers followed by a cross-link of mutated ␣1 subunits at the heterotrimeric stage. The formation of cross-linked ␣1H109C homodimers would then have occurred independently in a separate pathway.
GABA, 2 quantitatively the most important inhibitory neurotransmitter in the central nervous system, mediates fast synaptic inhibition by opening the chloride ion channel intrinsic to GABA A receptors (1). GABA A receptors are the targets of action of a variety of pharmacologically and clinically important drugs, such as benzodiazepines, barbiturates, steroids, anesthetics, and convulsants. These drugs modulate GABA-induced chloride ion flux by interacting with distinct allosteric binding sites at these receptors (2,3).
GABA A receptors are composed of five subunits (4,5) that can belong to different subunit classes. Recombinant receptor studies (5)(6)(7)(8) as well as studies investigating the subunit composition of GABA A receptors in the brain (9,10) indicated that the vast majority of GABA A receptors found are composed of two ␣, two ␤, and one ␥ subunit. Biochemical studies (5,11,12) as well as modeling of the GABA A receptor extracellular domain (13) according to the structure of the acetylcholinebinding protein (14) indicated the absolute arrangement of the subunits in GABA A receptors (Fig. 1A).
To achieve the correct order of subunits around the pore, each subunit must be able to discriminate between different subunits and to interact with its neighbors via specific high affinity contact sites. Several amino acid sequences have been identified that seem to be important for assembly of ␣, ␤, and ␥ subunits of GABA A receptors (15)(16)(17)(18)(19)(20)(21)(22)(23).
In a previous study, we used the comparative models developed by our group (13) for predicting amino acid residues in ␣1 and ␥2 subunits that might form direct contacts with each other (24). These residues were mutated to cysteines, and a possible disulfide bond formation between mutated subunits was investigated. Most of the cysteines introduced into ␣1 or ␥2 subunits did not cause a spontaneous cross-link between subunits, although the respective amino acid residues were found to be important for GABA A receptor assembly (24). The mutation ␣1H109C, however, caused a spontaneous cross-link of GABA A receptor subunits even in receptors composed of ␣1H109C and wild-type ␤3 and ␥2 subunits and thus in the absence of a mutated ␥2 subunit as a possible cross-link partner. 3 In the present study, this surprising observation was investigated in detail. It was demonstrated that the spontaneous disulfide bond formation occurred between two mutated ␣1 subunits during an early stage of assembly and that functional GABA A receptors could be formed despite the presence of a disulfide bond between two opposing subunits. Possible mechanisms of this cross-link during GABA A receptor assembly are discussed.
Generation of cDNA Constructs-For the generation of recombinant receptors, ␣1, ␤3, ␤2, and ␥2 subunits of GABA A receptors from rat brain were cloned and subcloned into pCDM8 expression vectors (Invitrogen) as described previously (5). Mutated subunits were constructed by PCR amplification using the wild-type subunit as template. For this, PCR primers were used to construct point mutations within the subunits by the "gene splicing by overlap extension" technique (25). The PCR primers contained XhoI and NsiI, NsiI and EcoRV, or XhoI and XbaI restriction sites, which were used to clone the ␣1, ␤2, or ␥2 fragments into pCI vectors, respectively (Promega, Madison, WI). The mutated subunits were confirmed by sequencing.
Culture and Transfection of HEK 293 Cells for Binding Studies and Western Blots-Transformed HEK 293 cells (CRL 1573; American Type Culture Collection, Manassas, VA) were cultured as described previously (5). 3 ϫ 10 6 cells were transfected with 20 g of cDNA for a single subunit transfection using the calcium phosphate precipitation method (26). For co-transfection with two different subunits, for each subunit 10 g of cDNA was used. When cells were co-transfected with three different subunits, 7 g of cDNA was used per subunit. A total of ϳ20 g of cDNA/transfection and a cDNA ratio of 1:1:1 seemed to be optimal for the expression of GABA A receptors under the conditions used as judged by receptor binding studies in cells transfected with ␣1, ␤3, and ␥2 subunits. Changing the subunit ratio by doubling the amount of cDNA for a single subunit at the cost of other subunits did not significantly change the number of [ 3 H]Ro15-1788 binding sites detected.
The cells were then harvested 48 h after transfection. At this time point the number of [ 3 H]Ro15-1788 binding sites formed per milligram of protein was at its maximum for cells transfected with ␣1, ␤3, and ␥2 subunits. Results obtained, however, did not change when cells were harvested 34 -48 h after transfection. In addition, Western blot analysis revealed that expres-sion levels of wild-type and mutated subunits were comparable at all harvesting times.
Immunoprecipitation of Wild-type and Mutated Subunits-The culture medium was removed from transfected HEK cells, and cells from four culture dishes were extracted with 1 ml of a C 12 E 10 extraction buffer (1% polyoxyethylene 10 lauryl ether (Sigma), 0.18% phosphatidylcholine (Sigma), 150 mM NaCl, 5 mM EDTA, and 50 mM Tris-HCl, pH 7.4, containing one "Mini Complete protease inhibitor mixture" tablet (Roche Diagnostics GmbH)/10 ml of extraction buffer) for 8 -12 h at 4°C. The extract was centrifuged for 40 min at 150,000 ϫ g at 4°C, and the clear supernatant was incubated overnight at 4°C under gentle shaking with 20 g of antibodies. After addition of Pansorbin (formalin-fixed Staphylococcus aureus cells, purchased from Calbiochem, EMD Bioscience Inc.) and 0.5% nonfat dry milk powder and shaking for an additional 2 h at 4°C, the precipitate was washed three times with a low salt buffer for immunoprecipitation (IP low buffer) (50 mM Tris-HCl, 0.5% Triton X-100, 150 mM NaCl, and 1 mM EDTA, pH 8.0). The precipitated proteins were dissolved in sample buffer (108 mM Tris sulfate, pH 8.2, 10 mM EDTA, 25% (w/v) glycerol, and 2% SDS with or without 3% dithiothreitol (DTT)).
Immunoprecipitation of Receptors Expressed at the Cell Surface-The culture medium was removed from HEK cells transfected with cDNA (21 g/3 ϫ 10 6 cells) of GABA A receptor subunits (cDNA ratio, 1:1:1), and the cells were washed once with PBS (2.7 mM KCl, 1.5 mM KH 2 PO 4 , 140 mM NaCl, and 4.3 mM Na 2 HPO 4 , pH 7.3). For experiments with N-ethylmaleimide, cells were preincubated with 50 mM NEM (in PBS) for 30 min at room temperature, and then unbound NEM was washed off the culture dishes. Cells with or without NEM treatment were then detached from the culture dishes by incubating with 2.5 ml of 5 mM EDTA in PBS for 5 min at room temperature. The resulting cell suspension was diluted in 6.5 ml of cold Dulbecco's modified Eagle's medium and centrifuged for 5 min at 1000 ϫ g.
For Western blot analysis, the cell pellet from two dishes was incubated with 35 g of ␣1-(1-9) antibodies in 3 ml of the same medium for 30 min at 37°C. Cells were again pelleted, and free antibodies were removed by washing twice with 10 ml of PBS buffer. Then receptors were extracted with IP low buffer containing 1% Triton for 1 h under gentle shaking. Cell debris were removed by centrifugation (30 min at 150,000 ϫ g at 4°C). After addition of Pansorbin and 0.5% nonfat dry milk powder and shaking for 2 h at 4°C, the precipitate was centrifuged for 10 min at 10,000 ϫ g at 4°C and dissolved in sample buffer (with or without DTT) and subjected to SDS-PAGE and Western blot analysis using digoxigenin-labeled antibodies.
To verify that only receptors on the cell surface were labeled by the antibodies, parallel samples were incubated with antibodies directed against the intracellular loop of GABA A receptor subunits. These antibodies could not precipitate any GABA A receptor subunits under the conditions used. A possible redistribution of the antibodies during the extraction procedure has been excluded previously (17).
Immunoprecipitation of Subunits after Depletion of Cell Surface Receptors-After immunoprecipitation of receptors expressed at the cell surface, the supernatant of the immuno- precipitate was again incubated with ␣1-(1-9) antibodies overnight at 4°C. Then Pansorbin and 0.5% nonfat dry milk powder were added, and the samples were shaken for another 2 h at 4°C. The precipitate was centrifuged for 10 min at 10,000 ϫ g at 4°C, dissolved in sample buffer (without DTT), and subjected to SDS-PAGE and Western blot analysis using digoxigeninlabeled ␣1-(1-9) antibodies.
SDS-PAGE, Western Blot, and Chemiluminescence Detection-SDS-PAGE was performed according to Neville and Glossmann (27) using gels containing 10% polyacrylamide in a discontinuous system. Proteins separated on the gels were tank-blotted onto prewetted polyvinylidene fluoride membranes. After blocking with 1.5% nonfat dry milk powder in PBS and 0.1% Tween 20 for 1 h at room temperature, the membranes were incubated overnight with digoxigenin-labeled ␣1-(1-9), ␤3-(1-13), or ␥2-(319 -366) antibodies (1 g/ml) at 4°C. The membranes were extensively washed and incubated with secondary antibodies (anti-digoxigenin-alkaline phosphatase, F(abЈ) 2 fragments, Roche Diagnostics GmbH) for 1 h at room temperature. Polyvinylidene fluoride membranes were again washed extensively as described above and then equilibrated in assay buffer (0.1 M diethanolamine and 1 mM MgCl 2 , pH 10.0) for 10 min, and secondary antibodies were visualized by the reaction of alkaline phosphatase with CDP Star (Applied Biosystems, Bedford, MA). The chemiluminescent signal was quantified by densitometry after exposing the immunoblots to the Fluor-S MultImager (Bio-Rad) and evaluated using the Quantity One quantitation software (Bio-Rad). Under the experimental conditions used, the immunoreactivities were within the linear range, and this permitted a direct comparison of the amount of antigen per gel lane between samples. Data were generated from several different gels and expressed as means Ϯ S.E.
All mutated constructs used in this study could be expressed to a comparable extent after single transfection into HEK cells. After co-transfection of different constructs, however, the stability of mutated subunits that formed non-productive assembly intermediates was reduced presumably by proteolytic degradation (17,28,29). Therefore, in all control experiments the extent of expression of mutated constructs was determined in singly transfected HEK cells.
[ 3 H]Ro15-1788 Binding Studies-Frozen membranes from transfected HEK cells were thawed, and cells were homogenized in 50 mM Tris, citrate buffer, pH 7.4, by using an Ultra-Turrax followed by three centrifugation resuspension cycles (200,000 ϫ g for 20 min at 4°C). Cell pellets were resuspended in 50 mM Tris, citrate buffer, pH 7.4, at a protein concentration in the range of 0.1-1 mg/ml as measured with the BCA protein assay kit (Pierce) using bovine serum albumin as standard.
Membranes were then incubated for 90 min at 4°C in a total of 1 ml of a solution containing 50 mM Tris, citrate buffer, pH 7.4, 150 mM NaCl, and various concentrations (range, 0.1-400 nM) of [ 3 H]Ro15-1788 (74.1 Ci/mmol, PerkinElmer Life Sciences) in the absence or presence of 100 M diazepam (Hoffmann-La Roche). Suspensions were filtered through Whatman GF/B filters, and the filters were rinsed twice with 5 ml of icecold 50 mM Tris, citrate buffer and then subjected to scintillation counting. Nonspecific binding in the presence of 100 M diazepam was subtracted from total [ 3 H]Ro15-1788 binding to result in specific binding. Scatchard analysis was performed three times using data from three different transfections and the GraphPad Prism 3.0 program (GraphPad Software, Inc., San Diego, CA).
Suspensions were then filtered through Whatman GF/B filters, and the filters were rinsed twice with 3.5 ml of ice-cold 50 mM Tris, citrate buffer and then subjected to scintillation counting. Nonspecific binding in the presence of 100 M GABA was subtracted from total [ 3 H]muscimol binding to result in specific binding. Scatchard analysis was performed three times using data from three different transfections and the GraphPad Prism 3.0 program (GraphPad Software Inc.).
Electrophysiological Studies Using Transfected HEK Cells-HEK cells for electrophysiological experiments were taken from the batches cultured for binding studies and Western blots and seeded at 1 ϫ 10 5 cells/35-mm cell culture dish. Cells were transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's recommendations. The cDNA ratio used was 1:1:1 for ␣1:␤3:␥2 subunits, respectively. pEGFP-N1 (Clontech) was co-transfected to identify transfected cells. Whole-cell recordings were performed at room temperature 1-2 days after transfection. GABA was applied using a DAD-12 superfusion system (Adams and List Associates Ltd., Westbury, NY). Extracellular solution contained 140 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 5 mM glucose, and 10 mM HEPES, pH 7.4. The pipette solution contained 140 mM KCl, 11 mM EGTA, 1 mM CaCl 2 , 1 mM MgCl 2 , and 10 mM HEPES, pH 7.2. The cells were clamped at Ϫ60 mV, and currents were filtered at 1 kHz, recorded with an Axopatch 200A amplifier (Axon Instruments, Foster City, CA), and analyzed with Clampfit software (Axon Instruments). DTT was applied immediately prior to experiments by replacing culture media with extracellular solution containing 10 mM DTT for 4 min followed by extensive washing. For experiments with copper phenanthroline, a stock solution of 100 mM CuSO 4 was generated. CuSO 4 was mixed with o-phenanthroline (Sigma) directly before use to a final concentration of 100:400 M copper:phenanthroline and applied for 10 min after extensive washout of DTT.
Density Gradient Centrifugation-Transfected HEK cells were incubated for 44 h at 37°C. Cells from four culture dishes were harvested and extracted in 1 ml of C 12 E 10 extraction buffer. The extracts were centrifuged for 40 min at 150,000 ϫ g at 4°C, and 200 l of the extracts was layered onto the top of a density gradient of 5-20% sucrose in C 12 E 10 extraction buffer. For the determination of sedimentation coefficients, 2 g of digoxigenin-labeled catalase (sedimentation coefficient, 11 s), 1.2 g of digoxigenin-labeled alkaline phosphatase (sedimentation coefficient, 6.1 s), and 1 g of digoxigenin-labeled carbonic anhydrase (sedimentation coefficient, 3.3 s) were included in the overlays. The gradients were centrifuged at 120,000 ϫ g for 23 h at 4°C and then fractionated by piercing at the tube bottom (5). Protein in individual fractions was precipitated (30) and dissolved in sample buffer for SDS-PAGE. Proteins were identified by Western blot analysis, and signal intensity per fraction was quantified as described above. In a previous study (5) it has been demonstrated that after co-transfection of HEK cells with ␣1, ␤3, and ␥2 subunits all three subunits sedimented at a single peak of 8.7 s, representing the pentameric receptor.
Immunofluorescence-HEK cells were fixed with 2% paraformaldehyde in PBS 30 -35 h after transfection followed by a 10-min wash in 50 mM NH 4 Cl in PBS. Washes between incubation steps were performed in PBS. Primary antibodies were detected with goat anti-rabbit IgG (HϩL) Bodipy FL (Molecular Probes, Eugene, OR) in 1% bovine serum albumin in PBS. Labeling was visualized using a Nikon Eclipse TE300 microscope equipped with a high pressure mercury lamp and suitable filter sets. To verify that labeling of cells without permeabilization was restricted to the cell surface, parallel samples were stained with antibodies directed against the intracellular loop of GABA A receptor subunits. These antibodies detected GABA A receptor subunits only after permeabilization of transfected cells.
Modeling-All comparative models of the GABA A receptor extracellular domain were based on the acetylcholine-binding protein structure (Protein Data Bank code 1I9B) (14). Based on seven alignments for the structurally variable regions, a total of 35 models were computed with Modeler version 6 (31). After validation, a total of seven well scoring models (one per alignment) of the GABA A receptor extracellular domain were used to predict the segments that participate in interface formation. Despite the high uncertainty that is associated with amino acid side chain positions at most interface-forming regions (13,14), predictions were made with the program WHAT IF (32) on the possible formation of disulfide bridges.

RESULTS
Cells Transfected with ␣1H109C, ␤3, and ␥2 Subunits Form GABA A Receptors Containing Cross-linked ␣1 Subunits-Preliminary experiments performed in the course of a previous study (24) demonstrated that ␣1 subunits carrying the mutation ␣1H109C formed a spontaneous cross-link either between themselves or with another protein in HEK cells transfected with ␣1H109C, ␤3, and ␥2 subunits. 3 To clarify this question, in the present study we first investigated whether the cross-linked ␣1H109C protein was associated with completely assembled receptors present at the surface of ␣1H109C-, ␤3-, and ␥2-transfected cells. Intact cells were incubated with ␣1-(1-9) antibodies, and proteins were then extracted under conditions in which the receptor-antibody complex was not drastically impaired and under which no redistribution of the antibody could be observed (17). Receptors were immunoprecipitated by addition of Pansorbin, dissolved in SDS loading buffer without DTT to allow the detection of subunit dimers formed by disulfide bonds, and subjected to SDS-PAGE and Western blot anal-ysis using digoxigenin-labeled ␣1-(1-9) antibodies. Fig. 2A shows that cells transfected with wild-type ␣1, ␤3, and ␥2 subunits under these conditions exhibited a major protein band at about 51 kDa that could be labeled by ␣1-(1-9) antibodies and thus represented ␣1 subunits. In contrast, cells transfected with ␣1H109C and wild-type ␤3 and ␥2 subunits exhibited a major protein band of about 100 kDa that could also be labeled by ␣1-(1-9) antibodies ( Fig. 2A). Staining of this 100-kDa protein band was stronger (135 Ϯ 16%) than that of wild-type ␣1 subunits and could only be observed in the absence of DTT. In the presence of DTT, a band at around 51 kDa was detected (Fig.  2D) indicating that the 100-kDa protein band ( Fig. 2A) contained ␣1H109C subunits cross-linked by disulfide bridges. Staining of this protein was still stronger than that of cells transfected with wild-type ␣1, ␤3, and ␥2 subunits. This indicates that the increase in staining was due to an increased expression of cross-linked ␣1 subunits at the surface of ␣1H109C-, ␤3-, and ␥2-transfected cells and not due to a better staining of cross-linked ␣1 subunits. In addition, this cross-link was not caused or initiated by the ␣1-(1-9) antibodies used for cell surface precipitation of receptors because it was also observed in Western blots from membranes of appropriately transfected HEK cells that were directly dissolved in SDS sample buffer without incubation with these antibodies (experiments not shown). Finally the cross-link was not induced during homogenization or solubilization because it also could be observed when free cysteines at the cell surface were alkylated with 50 mM NEM before extraction (experiments not shown). Then the same Western blot ( Fig. 2A) was stripped and reprobed using ␤3-(1-13) or ␥2-(319 -366) antibodies. Fig. 2B shows that the ␤3-(1-13) antibody, in agreement with previous results, could identify a diffuse protein band of about 52-54 kDa (21) in cells transfected with wild-type ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits. In cells transfected with ␣1H109C, ␤3, and ␥2, staining of ␤3 on average represented only 51 Ϯ 7% of that found in cells transfected with wild-type ␣1, ␤3, and ␥2 subunits. Stripping and reprobing this Western blot using ␥2-(319 -366) antibodies indicated a diffuse protein band in the range of 45-49 kDa in cells transfected with wildtype ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits that represented the ␥2 subunit (Fig. 2C). Again in cells transfected with ␣1H109C, ␤3, and ␥2, the expression of the ␥2 subunit was reduced to 44 Ϯ 5% compared with ␥2 in cells transfected with wild-type ␣1, ␤3, and ␥2. The protein smear at around 80 -90 kDa in Fig. 2C could also be detected in cells transfected with wild-type ␣1, ␤3, and ␥2 subunits. It presumably resulted from unspecific protein aggregates caused by oxidation of endogenous cysteines in these gels that were run in the absence of DTT to allow the detection of cross-linked proteins.
In contrast to ␣1-(1-9) antibodies that labeled a 100-kDa protein band in precipitates from ␣1H109C␤3␥2-transfected cells, no such protein band could be detected by ␤3-(1-13) or ␥2-(319 -366) antibodies indicating that the 100-kDa band did not contain ␤3 or ␥2 subunits. It thus either contained two ␣1H109C subunits or an ␣1H109C subunit cross-linked with an unrelated protein. However, density gradient centrifugation indicated that receptors containing cross-linked ␣1H109C subunits exhibited a maximum sedimentation coefficient similar to completely assembled, wild-type receptors (experiments not shown), excluding the possibility that an unrelated protein or additional ␣1H109C subunits were cross-linked to the pentameric receptor or that multiple receptors were cross-linked with each other. In addition, residue ␣1H109C is located at the entrance of the chloride channel formed by GABA A receptors (Fig. 1B). Any cross-link of ␣1H109C subunits with an unrelated protein would have interfered with the formation of functional GABA A receptors.
Cross-linked ␣1H109C Subunits Form Functional GABA A Receptors with ␤3 and ␥2 Subunits-To investigate the formation of functional GABA A receptors, whole-cell patch clamp experiments were performed in HEK cells transfected with ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits. GABA enhanced the chloride conductance in both sets of transfected cells (Fig.  3) indicating the formation of functional ␣1H109C␤3␥2 receptors. These results argue against a cross-link of ␣1H109C subunits with an unrelated protein and support a cross-link of two ␣1H109C subunits within the pentameric receptor. The potency of GABA for stimulating chloride flux, however, was different in ␣1␤3␥2-(EC 50 ϭ 35 M, 95% confidence interval ϭ 31-40 M) and ␣1H109C␤3␥2 (EC 50 ϭ 440 M, 95% confidence interval ϭ 365-522 M)-transfected cells (Fig. 3). To investigate whether the 13-fold reduced potency of GABA for activating ␣1H109C␤3␥2 receptors was caused by a limited conformational flexibility due to the cross-linked ␣1H109C subunits, the GABA potency was measured after treatment of transfected cells by 10 mM DTT. Under these conditions the EC 50 for GABA in ␣1H109C␤3␥2-transfected cells was shifted to the left resulting in an EC 50 of about 37 M (95% confidence interval ϭ 31-45 M), which was comparable to the EC 50 of wild-type ␣1␤3␥2 receptors. In control experiments it was demonstrated that treatment of ␣1␤3␥2-transfected cells with DTT prior to the experiment did not change the GABA doseresponse curve of wild-type ␣1␤3␥2 receptors (Fig. 3) confirming previously published data (33). However, no spontaneous reoxidation could be observed in ␣1H109C␤3␥2-transfected cells after reduction by DTT and subsequent removal of the reducing agent. Attempts to reoxidize the reduced disulfide bonds by the addition of copper phenanthroline were also not successful (experiments not shown) indicating that disulfide bond formation only was possible during assembly and not in completely assembled receptors.
Additional experiments investigated the modulation of wildtype and mutated GABA A receptors by a benzodiazepine. Diazepam was able to dose dependently modulate GABA receptors formed in ␣1␤3␥2or ␣1H109C␤3␥2-transfected cells with comparable potency but different efficacy. Although diazepam enhanced GABA-induced currents (EC 3 ) by about 3.5-fold in ␣1␤3␥2-transfected cells, enhancement was only about 2.5-fold in ␣1H109C␤3␥2-transfected cells.  . Functional properties of GABA A receptors containing mutated ␣1 and wild-type ␤3 and ␥2 subunits. GABA-elicited whole-cell currents recorded from HEK cells co-transfected with ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits in the absence (native) or presence (reduced) of DTT are shown. Due to variable transfection efficiencies, measured currents varied up to 7-fold from cell to cell even from the same transfected culture dish. Data therefore were normalized to the maximum response to GABA and represent means Ϯ S.E. of six to eight individual dose-response curves obtained from different cells derived from a total of four transfections. Despite the large variability of currents, the average whole-cell currents obtained for wild-type and mutated receptors (under native as well as reduced conditions) were in the same order of magnitude. For technical reasons, GABA concentrations Ͼ10 mM could not be applied, and thus maximum currents of the mutated receptors in the absence of DTT treatment could not be determined reliably. The EC 50 of the mutated receptors under non-reducing conditions is therefore an estimation only. Cells transfected with ␣1H109C, ␤3, and ␥2 subunits exhibited a Hill coefficient of 1.2 (before and after treatment with DTT) that is comparable to the Hill coefficient of cells transfected with wild-type ␣1, ␤3, and ␥2 subunits. from cells transfected with ␣1H109C, ␤3, and ␥2 subunits indicated a K D of 1.6 Ϯ 0.2 nM that was comparable with that of wild-type ␣1␤3␥2 receptors (1.4 Ϯ 0.1 nM). This indicated that mutated ␣1H109C subunits when co-expressed with ␤3 and ␥2 subunits displayed an intact ␣1(ϩ)/␥2(Ϫ) interface forming a benzodiazepine pocket similar to wild-type receptors. Assembly Intermediates Can Be Transported to the Cell Surface-As mentioned above, the amount of ␣1H109C subunits at the surface of cells transfected with ␣1H109C, ␤3, and ␥2 subunits was increased, whereas the amounts of ␤3 and ␥2 subunits were reduced as compared with cells transfected with ␣1, ␤3, and ␥2 subunits (Fig. 2). To investigate whether assembly intermediates containing cross-linked ␣1H109C subunits were transported to the cell surface, HEK cells were transfected with only ␣1 or ␣1H109C subunits. Wild-type or mutated ␣1 subunits possibly expressed at the cell surface were then labeled by ␣1-(1-9) antibodies and after extraction were precipitated by the addition of Pansorbin. In agreement with previous results (34), ␣1 subunits when transfected alone into HEK cells could not reach the cell surface. In contrast, however, ␣1H109C subunits could be detected at the cell surface in a protein band with an apparent molecular mass of 100 kDa (Fig. 5), suggesting that cross-linked ␣1H109C subunits could reach the cell surface.
This conclusion was confirmed by investigations using immunofluorescence microscopy. On staining with ␣1-(1-9) antibodies only mutated ␣1H109C but not wild-type ␣1 subunits could be detected at the surface of intact HEK cells transfected with either wild-type ␣1H109C or ␣1 subunits (experiments not shown).

Transport of Assembly Intermediates to the Cell Surface Resulted in Reduced Amounts of Intracellular ␣1H109C
Subunits-The finding that ␣1H109C subunits could form cross-linked complexes that could be transported to the cell surface could explain the increased expression of cross-linked ␣1H109C at the surface of cells transfected with ␣1H109C, ␤3, and ␥2 subunits (Fig. 2A). These results, however, could not explain the reduced expression of ␤3 or ␥2 subunits at the cell surface (Fig. 2, B and C). To get a more complete picture on what happened in these cells, the experiment of Fig. 2 was repeated, and the distribution of ␣1H109C subunits was investigated in total cell extracts, in receptors at the cell surface, and in extracts deprived of receptors at the cell surface. For that a pool of HEK cells transfected with ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits was generated and divided into two parts. To determine the total amount of ␣1 subunits, assembly intermediates, or receptors containing ␣1 subunits, one part was extracted, and the extracted proteins were immunoprecipitated using ␣1-(1-9) antibodies and subjected to Western blot analysis using digoxigenin-labeled ␣1-(1-9) antibodies. In the second pool of transfected HEK cells, receptors at the cell surface were labeled with ␣1-(1-9) antibodies, extracted, and precipitated by the addition of Pansorbin. The supernatant of this precipitate was then again incubated with ␣1-(1-9) antibodies, thus precipitating receptors or assembly intermediates present in the inside of the cell. As shown in Fig. 6A, in total extracts of cells transfected with ␣1, ␤3, and ␥2 subunits, the majority of ␣1 subunits migrated as a protein with an apparent molecular mass of 51 kDa as expected. The additional protein smear at around 120 -200 kDa presumably resulted from unspecific protein aggregates caused by oxidation of endogenous cysteines in these gels that were run in the absence of DTT to allow the detection of cross-linked proteins. In contrast, in cells transfected with ␣1H109C, ␤3, and ␥2 subunits, only a minor amount of ␣1H109C subunits migrated at 51 kDa, whereas the majority of ␣1H109C subunits migrated at around 100 kDa probably representing the cross-linked ␣1H109C complex. In addition, as in cells transfected with ␣1, ␤3, and ␥2 subunits some unspecific protein aggregates could be detected at Ͼ100 kDa due to the absence of DTT in these gels. The overall amount of ␣1H109C subunits (protein bands of 51 and 100 -200 kDa) from total extracts of cells transfected with ␣1H109C, ␤3, and ␥2 subunits was 106 Ϯ 7% of that of ␣1 subunits in ␣1␤3␥2-transfected cells. This indicated that wild-type and mutated ␣1 subunits were expressed to the same extent in these cells.
The second pool of cells transfected with ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits was subjected to labeling of cell surface receptors. In agreement with results from Fig. 2A, most if not all ␣1H109C subunits were cross-linked, and the overall expression of ␣1H109C subunits at the cell surface was increased (Fig. 6B). After depletion of cell surface receptors, the same extracts were again incubated with ␣1-(1-9) antibodies, and the intracellular content of ␣1 or ␣1H109C subunits from cells transfected with ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits, respectively, was investigated. As shown in Fig. 6C, cross-linked as well as non-cross-linked ␣1H109C subunits could be identified under these conditions, and the amount of ␣1H109C subunits from cells transfected with ␣1H109C, ␤3, and ␥2 subunits was drastically reduced (by 67 Ϯ 6%) compared with that of ␣1 subunits from cells transfected with ␣1, ␤3, and ␥2 subunits.
Mutations in ␤2 and ␥2 Subunits Homologous to ␣1H109C Did Not Cause Cross-linking of the Respective Subunits-To investigate whether ␤2 or ␥2 subunits mutated at positions homologous to ␣1His-109 could also form cross-linked subunit dimers, the respective amino acid residues on the ␤2 and ␥2 subunits were substituted by cysteines giving rise to mutated subunits ␤2H107C and ␥2H122C. However, no cross-link of these mutated subunits could be observed when they were either transfected individually into HEK cells or in combination with the respective wild-type subunits (experiments not shown). In addition, when ␤2H107C or ␥2H122C subunits were co-transfected with ␣1H109C and wild-type ␥2 or ␣1H109C and wild-type ␤2 subunits, respectively, only crosslinked ␣1H109C subunits could be observed (experiments not FIGURE 5. Cell surface expression of assembly intermediates containing mutated ␣1 subunits. HEK cells were transfected with ␣1 or ␣1H109C subunits. Putative assembly intermediates expressed on the surface were immunolabeled by an incubation of intact cells with ␣1-(1-9) antibodies. Labeled proteins were then extracted, precipitated by Pansorbin, and subjected to SDS-PAGE and Western blot analysis using digoxigenin-labeled ␣1-(1-9) antibodies. These experiments were performed twice with comparable results. DIG, digoxigenin. FIGURE 6. ␣1 subunit expression in total cell extracts, at the cell surface, and in intracellular compartments of cells transfected with mutated ␣1 and wild-type ␤3 and ␥2 subunits. A pool of HEK cells transfected with ␣1, ␤3, and ␥2 or ␣1H109C, ␤3, and ␥2 subunits was generated and divided into two parts. A, the total amount of ␣1 subunits was extracted, and the extracted proteins were immunoprecipitated using ␣1-(1-9) antibodies and subjected to Western blot analysis using digoxigenin-labeled ␣1-(1-9) antibodies. B, in the second pool of transfected HEK cells, receptors at the cell surface were labeled with ␣1-(1-9) antibodies, extracted, and precipitated by the addition of Pansorbin. C, the supernatant of this precipitate was then again incubated with ␣1-(1-9) antibodies, thus precipitating receptors or assembly intermediates present in the inside of the cell. These experiments were performed three times with comparable results. The respective variability (ϮS.D. values) of data is given in the text. DIG, digoxigenin.
shown). For disulfide bond formation it is crucial that the side chains of two residues display a certain distance and angle to each other. These criteria might not have been met by the ␤2H107C or ␥2H122C mutants.

The Mutation ␣1H109C Caused a Spontaneous Disulfide Bond Formation between Two ␣1 Subunits in ␣1␤3␥2
Receptors-In the present study, it was demonstrated that completely assembled GABA A receptors composed of the mutated ␣1H109C and the wild-type ␤3 and ␥2 subunits are formed at the surface of appropriately transfected HEK cells. Interestingly, however, ␣1H109C subunits that exhibit an apparent molecular mass of 51 kDa migrated as a protein with an apparent molecular mass of about 100 kDa. This 100-kDa protein could be converted to a 51-kDa protein in the presence of DTT, suggesting that ␣1H109C subunits at the cell surface were cross-linked by disulfide bridges. The cross-linked protein complex, however, did not contain ␤3 or ␥2 subunits indicating that the disulfide bond formation occurred between two ␣1H109C subunits or between an ␣1H109C subunit and an unrelated 50-kDa protein. Because residue ␣1H109C is located within the channel mouth, a cross-link with an unrelated protein would have prevented the formation of functional GABA A receptors. The demonstration of GABA-gated chloride channels in cells transfected with ␣1H109C, ␤3, and ␥2 subunits therefore indicated that in these receptors two ␣1H109C subunits were cross-linked with each other.
Receptors Composed of ␤3, ␥2, and Cross-linked ␣1H109C Subunits Exhibited the Same Subunit Stoichiometry and Arrangement as Wild-type Receptors-A cross-link of two ␣1H109C subunits could have changed subunit stoichiometry and arrangement of receptors assembled in ␣1H109C-, ␤3-, and ␥2-transfected HEK cells. The observation, however, that receptors at the surface of ␣1H109C-, ␤3-, and ␥2-transfected cells contained only cross-linked but no single ␣1H109C subunits excluded the formation of receptors containing odd numbers of ␣ subunits. The demonstration of high affinity [ 3 H]muscimol and [ 3 H]Ro15-1788 binding sites formed at the ␤(ϩ)/ ␣(Ϫ) and the ␣(ϩ)/␥(Ϫ) interface, respectively, excluded subunit stoichiometries in which any one of the ␣1H109C, ␤3, or ␥2 subunits was missing. Finally the observation that GABA A receptors composed of ␣1H109C, ␤3, and ␥2 subunits after reduction of their disulfide bond could be activated by GABA with the same potency as wild-type receptors and with a Hill coefficient of 1.2 suggested that the mutated receptor contained two GABA binding sites. Together these results can only be reconciled with GABA A receptors exhibiting wild-type subunit stoichiometry and arrangement. It thus has to be concluded that the disulfide bond between the two opposing ␣1H109C subunits bridged the channel mouth of these receptors. The segment around residue ␣1His-109 protrudes furthest toward the lumen (Fig. 1B). In addition, in contrast to the sequence of acetylcholine-binding protein, GABA A receptor subunits contain an insertion of three amino acid residues in this region that are most likely also located luminally. It is thus possible that the position of ␣1His-109 might be relatively flexible resulting in a fairly mild distortion of GABA A receptors containing cross-linked ␣1 subunits. The low potency of GABA for chloride channel opening before reduction of the disulfide bond then might have been caused by a disulfide bond-induced restriction in the mobility of residue ␣1H109C or surrounding structures, suggesting that this part of the receptor undergoes a positional or conformational change during GABA-induced opening of the chloride channel.
Cross-linking of ␣1H109C subunits, however, did not change the affinity of [ 3 H]Ro15-1788 for the benzodiazepine binding site or the potency of diazepam for stimulating GABAinduced chloride flux, indicating that residue ␣1His-109 is neither involved in direct ligand binding nor important for the benzodiazepine pocket geometry. The efficacy of diazepam for enhancing GABA currents, however, was impaired in receptors containing cross-linked ␣1His-109 subunits, indicating that the flexibility of residue ␣1H109C is important also for the allosteric modulation of the receptors by benzodiazepines.
Assembly , respectively, of that in ␣1␤3␥2-transfected cells, suggesting that the mutated receptors represented 34 -39% of wildtype receptors. Whereas the amounts of ␤3 or ␥2 subunits identified at the cell surface were consistent with the formation of 34 -39% receptors, cross-linked ␣1H109C subunits present at the cell surface were 135% of wild-type ␣1 subunits in ␣1␤3␥2transfected cells. Assuming that 34 -39% of ␣1H109C subunits contributed to ␣1H109C␤3␥2 receptors, the surplus of 95-100% ␣1H109C subunits could only be explained by a transport of cross-linked ␣1H109C subunits to the cell surface. This was actually demonstrated in cells transfected with only ␣1H109C subunits. The formation of cross-linked ␣1H109C subunits, thus, overruled the quality control mechanisms in the endoplasmic reticulum, possibly by covering an endoplasmic reticulum retention signal, thus allowing the exit of incompletely assembled intermediates. This caused an attenuation of ␣1H109C subunits in the endoplasmic reticulum resulting in an overall reduction in the formation of completely assembled receptors. Wild-type ␣1 homodimers (as well as higher oligomers) are also formed under these conditions (supplemental Fig. 1). Due to a more labile interaction of two wild-type ␣1 subunits they might be retained in the endoplasmic reticulum for a longer time period, thus enhancing their chances for interaction with another subunit, leading to completely assembled receptors.
Possible Mechanisms Allowing the Formation of Completely Assembled GABA A Receptors Containing Cross-linked ␣1H109C Subunits-The formation of ␣1H109C dimers as well as of completely assembled receptors containing crosslinked ␣1H109C subunits most directly can be explained by an interaction of two ␣1 subunits being at least one starting point of wild-type receptor formation. This possibility is supported by the fact that the ␣1 subunits are synthesized side by side by different ribosomes on the same mRNA and thus probably have a higher chance to interact with each other than with ␤3 or ␥2 subunits that are synthesized via a different mRNA-ribosome complex. In addition, ␣1H109C homodimer formation and cross-link seemed to occur more rapidly than the assembly between ␣1H109C and ␤3 or ␥2 subunits as indicated by the large surplus of cross-linked ␣1H109C dimers over ␤3 and ␥2 subunits as well as over completely assembled receptors at the cell surface. Finally formation of ␣1 homodimers is also observed in HEK cells transfected with wild-type ␣1 subunits (supplemental Fig. 1). Because no ␣1-␣1 interface is present in native ␣1␤3␥2 receptors, dimerization of ␣1 subunits via their respective (ϩ) and (Ϫ) sides would lead to a dead end of assembly. This is the reason why the formation of ␣1-␣1 dimers has never been discussed as a possible starting point of assembly. However, interaction of ␣1 subunits via the side forming the channel mouth (luminal side) offers an alternative possibility avoiding contacts via (ϩ) and (Ϫ) sides that could be weakly stabilized, e.g. via van der Waals forces to which interaction of the two sterically exposed ␣1His-109 residues might contribute. Such an assembly intermediate enhances the probability of interaction with a ␤3 subunit because this subunit can encounter the dimer from the one or the other side (Fig. 7A). Formation of two correct ␣/␤ interfaces via the (ϩ) and (Ϫ) sides of two ␣1 subunits then probably causes opening of the weak luminal interactions between the two ␣1 subunits (Fig. 7A), thus allowing the fourth and fifth subunits to enter and complete GABA A receptor assembly. To finally result in receptors with the correct subunit stoichiometry and arrangement, it is mandatory, however, that subunits assembling with the trimer preferentially bind either to the accessible ␣1(ϩ) side (␥2 subunits) or the ␣1(Ϫ) side (␤3 subunits). Overall such a mechanism is consistent with the finding that even in the presence of a covalent cross-link of two ␣1H109C subunits completely assembled receptors with a wild-type stoichiometry and arrangement are formed presumably before the cross-linked dimers could be transported to the cell surface. Covalent crosslinking with subsequent transport to the cell surface could thus have trapped a labile assembly intermediate that could not have been detected without the use of mutated subunits.
Alternatively the assembly of GABA A receptors could have started with the formation of a heterodimer either by a direct interaction of ␣1 and ␤3 (or ␥2) subunits via their respective (ϩ) and (Ϫ) sides (Fig. 7B) or via an initial contact with a subsequent rearrangement forming the final (ϩ)/(Ϫ) interaction. As soon as the correct (ϩ)/(Ϫ) interface has been established in the dimer, each additional subunit then has to assemble in the correct order to give rise to a receptor with correct subunit stoichiometry and arrangement. When this assembly mechanism is used, discrimination between subunits occurs already at the stage of the dimer. In this case, cross-linking of the nonneighboring ␣1H109C subunits could have occurred at the heterotrimeric assembly stage. Here again, ␣1H109C-␣1H109C interaction and subsequent cross-link presumably occurred via luminal contacts before correct formation of all interfaces of the ␣1-␤3-␣1 trimer because cross-linking of the two nonneighboring ␣1H109C subunits after all interfaces have been formed seems to be sterically difficult. An initial luminal contact between subunits might be more efficient in forming the correct interface than a selective contact via the final interfaces because a non-selective initial contact subsequently allows subunits to choose between the available (ϩ) or (Ϫ) side of assembly partners. If assembly of GABA A receptors started with the formation of heterodimers, then the simultaneous formation of a large surplus of cross-linked ␣1H109C homodimers represented an independent and competitive event that reduced the amounts of ␣1 subunits available for receptor assembly. Crosslinked homodimers could then have been a dead end of assembly or could have led to completely assembled receptors containing cross-linked ␣1H109C subunits before the homodimers were transported to the cell surface. The described sequential assembly mechanism (Fig. 7B) is similar to that proposed by Green and Claudio (35) for the nicotinic acetylcholine receptor. For the nicotinic acetylcholine receptor an additional assembly mechanism has been discussed in which two subunit dimers could combine with a fifth subunit to form the completely assembled pentameric receptors (36). Further experiments will have to decide between all these possibilities. FIGURE 7. Schematic drawing of possible mechanisms of GABA A receptor assembly. The mutated and sterically exposed residues ␣1H109C are indicated by a black line at the luminal side of the subunit. ϩ and Ϫ indicate the principal and complementary side of a subunit, respectively. A, in one mechanism the two ␣1H109C subunits interact via their luminal side and offer the possibility that ␤3 subunits encountering the dimer from the one or the other side could assemble by forming two correct interfaces with two different ␣1 subunits. B, in the second mechanism GABA A receptor assembly starts with the formation of a heterodimer of ␣1 and ␤3 (or ␥2) subunits via their respective (ϩ) and (Ϫ) sides. As soon as the correct (ϩ)/(Ϫ) interface has been established, each additional subunit then has to assemble in the correct order.