GABAA Receptor Assembly IDENTIFICATION AND STRUCTURE OF γ2 SEQUENCES FORMING THE INTERSUBUNIT CONTACTS WITH α1 AND β3 SUBUNITS

GABAA receptors are ligand-gated chloride channels composed of five homologous subunits that specifically recognize one another and assemble around an aqueous pore. To identify domains responsible for the specificity of subunit association, we constructed C-terminal truncated γ2subunits, as well as mutated and chimeric fragments. From their ability to interfere with α1β3γ2receptor assembly and to associate with full-length subunits, we concluded that amino acid sequences γ2-(91–104) and γ2-(83–90) form the sites mediating assembly with α1 and β3 subunits, respectively. Neural network-based secondary structure prediction, Monte Carlo optimization, and hydrophobicity analysis led to the conclusion that these sites also form the intersubunit contacts in the completely assembled receptor and provided important information on the benzodiazepine-binding site and structure of GABAA receptors.

GABA A receptors are ligand-gated chloride channels composed of five homologous subunits that specifically recognize one another and assemble around an aqueous pore. To identify domains responsible for the specificity of subunit association, we constructed C-terminal truncated ␥ 2 subunits, as well as mutated and chimeric fragments. From their ability to interfere with ␣ 1 ␤ 3 ␥ 2 receptor assembly and to associate with full-length subunits, we concluded that amino acid sequences ␥ 2 -(91-104) and ␥ 2 -(83-90) form the sites mediating assembly with ␣ 1 and ␤ 3 subunits, respectively. Neural network-based secondary structure prediction, Monte Carlo optimization, and hydrophobicity analysis led to the conclusion that these sites also form the intersubunit contacts in the completely assembled receptor and provided important information on the benzodiazepine-binding site and structure of GABA A receptors.
␥-Aminobutyric acid (GABA), 1 the quantitatively most important inhibitory neurotransmitter in the central nervous system, mediates fast synaptic inhibition by opening the chloride ion channel intrinsic to the GABA A receptor (1). This receptor is the site of action of various pharmacologically and clinically important drugs, such as benzodiazepines, barbiturates, steroids, anesthetics, and convulsants. These drugs modulate GABA-induced chloride ion flux by interacting with separate and distinct allosteric binding sites (2).
The GABA A receptor is a hetero-oligomeric protein consisting of five subunits (3,4). So far at least 20 GABA A receptor subunits belonging to several subunit classes (six ␣, four ␤, four ␥, one ␦, one ⑀, one , and three ) have been identified (5). In situ hybridization and immunocytochemical studies indicate a distinct but overlapping temporal and regional expression of these subunits. The finding that multiple receptor subunits are expressed within single neurons (6,7), raises the possibility for the formation of an extremely large variety of GABA A receptor subtypes. However, not all receptors that can be formed theoretically are formed in the cells (8,9). Thus, GABA A receptor heterogeneity is limited by the temporal and spatial pattern of subunit expression and by the selective oligomerization mediated by receptor assembly.
The assembly of hetero-oligomeric receptors is a complex, multistep process that generally occurs in the endoplasmatic reticulum (10,11). Each subunit must recognize its neighbors via specific high affinity contact sites. To achieve the correct order of subunits around the pore, in addition selective discriminations must be made between different subunits. So far, little is known on the molecular mechanism of GABA A receptor assembly. Recently it was demonstrated that after transfection of human embryonic kidney (HEK) 293 cells with dual combinations of ␣ 1 , ␤ 3 , and ␥ 2 subunits, ␣ 1 ␥ 2 and ␤ 3 ␥ 2 dimers as well as ␣ 1 ␤ 3 tetramers and pentamers were formed (4). These results suggested that assembly of one of the major GABA A receptor subtypes composed of ␣ 1 , ␤ 3 , and ␥ 2 subunits could start with the formation of each of the three possible dimers and led to the determination of the stoichiometry and subunit arrangement of ␣ 1 ␤ 3 ␥ 2 receptors ( Fig. 1) (4,12,13). Since in this receptor only a single ␥ 2 subunit is present and situated between an ␣ 1 and a ␤ 3 subunit, distinct binding sites for ␣ 1 and ␤ 3 subunits should exist on the ␥ 2 subunit.
In the present study truncated ␥ 2 subunits as well as mutated and chimeric constructs were used to identify amino acid sequences on ␥ 2 subunits mediating assembly with ␣ 1 and ␤ 3 subunits. The neural network-based prediction PHDsec, Monte Carlo optimization as well as hydrophobicity analysis provided important additional information on the structure of GABA A receptors and indicated that the identified ␥ 2 sequences form the intersubunit contact sites with ␣ 1 and ␤ 3 subunits.
Generation of cDNA Constructs-For the generation of recombinant receptors, ␣ 1 , ␤ 3 , and ␥ 2 subunits of GABA A receptors from rat brain were cloned and subcloned into pCDM8 expression vectors (Invitrogen, San Diego, CA) as described previously (4,16). Truncated subunits were constructed by polymerase chain reaction amplification using the fulllength subunit as template. The polymerase chain reaction primers contained EcoRI and HindIII restriction sites, which were used to clone the fragments into pCDNAIAmp vectors (Invitrogen). The truncated subunits were confirmed by sequencing. Chimeras were constructed using the "gene SOEing" technique (17) and were cloned into pCDNAIAmp vectors using the EcoRI and HindIII restriction sites of the primers.
HEK 293 cells (3 ϫ 10 6 ) were transfected with a total amount of 20 g of subunit cDNAs via the calcium phosphate precipitation method (18). The cells were harvested 44 h after transfection. Immunoprecipitation of Receptors Expressed on the Cell Surface and Receptor Binding Studies-The culture medium was removed from HEK 293 cells transfected with ␣ 1 , ␤ 3 , and ␥ 2 subunits together with a truncated ␥ 2 construct (cDNA ratio 1:1:1:1) and the cells were washed once with phosphate-buffered saline (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). Cells were then detached from the culture dishes by incubating with 2.5 ml of 5 mM EDTA in phosphate-buffered saline 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.
The pellet from two dishes was incubated with 30 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 phosphate-buffered saline buffer. Then receptors were extracted with low IP buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, pH 8.0) containing 1% Triton X-100 for 1 h under gentle shaking. Cell debris was removed by centrifugation (30 min; 150,000 ϫ g; 4°C). After addition of Immunoprecipitin (Life Technologies, Gaithersburg, MD; see Ref. 4) and 0.5% nonfat dry milk powder and shaking for 3 h at 4°C, the precipitate was centrifuged for 10 min at 10,000 ϫ g and washed three times with low IP buffer. The precipitated receptors were then suspended in 1 ml of a solution containing 0.1% Triton X-100, 50 mM Tris citrate buffer, pH 7. 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 (experiments not shown). These antibodies could not precipitate any GABA A receptor subunits.
To investigate a possible redistribution of the antibodies during the extraction procedure, in other experiments the extract containing the cell surface-labeled receptors was divided into two fractions. One fraction was kept at 4°C for 2 h, whereas the other fraction was incubated with additional ␣ 1 -(1-9) antibodies at 4°C for the same time period. Immunoprecipitin was added to both fractions, the resulting precipitates were centrifuged, washed, dissolved in sample buffer (108 mM Tris sulfate, pH 8.2, 10 mM EDTA, 25% (w/v) glycerol, 2% SDS, and 3% dithiothreitol) and were subjected to SDS-PAGE and Western blot analysis (4) using digoxygenized ␥ 2 -(1-33) antibodies. In both ␣ 1 -(1-9) precipitates full-length ␥ 2 subunits could be detected. The truncated ␥ 2 subunit, however, could only be detected in the fraction where additional ␣ 1 -(1-9) antibodies had been added after cell lysis (experiments not shown). Since aggregates consisting of full-length ␣ 1 and truncated ␥ 2 subunits are not transported to the cell surface and are present only intracellularly (11), this experiment demonstrated that there was no redistribution of antibodies during extraction and that only receptors present on the cell surface were detected by this procedure.
In other experiments membranes from HEK cells transfected with full-length ␣ 1 and truncated ␥ 2 subunits were incubated with 10 nM [ 3 H]flunitrazepam (83.0 Ci/mmol; Amersham Pharmacia Biotech) in the absence or presence of 10 M diazepam. After incubation for 90 min at 4°C, the suspensions were filtered through Whatman GF/B filters and subjected to scintillation counting.
Purification and Co-immunoprecipitation of Complete and Truncated Subunits-44 h after co-transfection of HEK cells with the cDNA of a full-length and a truncated subunit (cDNA ratio 1:1), cells from 4 culture dishes were extracted with 800 l of a Lubrol extraction buffer (1% Lubrol PX, 0.18% phosphatidylcholine, 150 mM NaCl, 5 mM EDTA, 50 mM Tris-HCl, pH 7.4, containing 0.3 mM phenylmethylsulfonyl flu-oride, 1 mM benzamidine, and 100 mg/ml bacitracin) for 8 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 15 g of antibodies directed against the full-length subunit. After addition of Immunoprecipitin and 0.5% nonfat dry milk powder and shaking for additional 3 h at 4°C, the precipitate was washed three times with low IP buffer. The precipitated proteins were dissolved in sample buffer. SDS-PAGE and Western blot analysis with digoxygenized antibodies was performed as described in Ref. 4.
Secondary Structure Prediction-The EMBL PredictProtein server (19) was used to align ␥ 2 sequences to homologous sequences available in the SwissProt data base and then to predict the secondary structure based on the set of aligned sequences by the PHDsec method (20,21). The significance of this prediction for short sequences of interest was evaluated by force field calculations within a Dynamic Monte Carlo (DMC) scheme (22). In this procedure the structure optimization started with the sequence in extended conformation. Within one DMC step a single , , or side chain dihedral angle was updated in the range [Ϫ180,180], the conformational energy was calculated within the ECEPP/3 force field (23) and the free energy of solvation was computed based on a continuum solvation model. 2 ϫ 10 5 DMC steps were performed. The final acceptance probability was given by a modified Metropolis criterium considering both, conformational and solvation energies. Details on the optimization algorithm are given in Ref. 24. Hydrophobicity analysis was performed according to the relative hydrophobicity scale for amino acids given by Eisenberg et al. (25).

RESULTS
Truncated ␥ 2 Subunits Reduce the Expression of Recombinant ␣ 1 ␤ 3 ␥ 2 Receptors on the Cell Surface-Experiments investigating the assembly of the nicotinic acetylcholine receptor have indicated that truncated subunits bearing assembly signals form unproductive complexes with full-length subunits and thus inhibit receptor assembly and its subsequent expression on the cell surface (11).
In the present study, HEK cells were transfected with fulllength ␣ 1 , ␤ 3 , and ␥ 2 subunits together with a truncated ␥ 2 subunit. The truncated ␥ 2 subunits used are shown in Fig. 2A. GABA A receptors expressed on the surface of transfected HEK cells were labeled with subunit-specific antibodies. Antibodylabeled receptors were extracted and precipitated by addition of Immunoprecipitin. The amount of receptors precipitated was quantified using a [ 3 H]Ro 15-1788 binding assay and was compared with that precipitated from cells transfected in the absence of truncated subunits.
As shown in Fig. 2B the construct ␥ 2 -(1-234) consisting of the complete extracellular N-terminal domain was able to reduce the expression of ␣ 1 ␤ 3 ␥ 2 receptors on the cell surface by 50%. When higher amounts of ␥ 2 -(1-234) cDNA were co-transfected with full-length ␣ 1 , ␤ 3 , and ␥ 2 subunits (cDNA ratio 2:1:1:1) the amount of receptors expressed on the cell surface was reduced to 40% (data not shown). A subsequent reduction in the size of the truncated ␥ 2 subunit indicated that ␥ 2 -(1-113) was the smallest N-terminal fragment that could reduce the expression of ␣ 1 ␤ 3 ␥ 2 receptors on the cell surface. Control experiments indicated that the failure of ␥ 2 -(1-110) to reduce the expression of ␣ 1 ␤ 3 ␥ 2 receptors on the cell surface was not due to a weak expression of this construct (Fig. 2C).
As shown for cells co-transfected with ␥ 2 -(1-113) and with ␣ 1 subunits, three protein bands (apparent molecular mass 14, 17, and 20 kDa) were precipitated by ␣ 1 -(1-9) and detected by ␥ 2 -(1-33) antibodies (Fig. 3A). The molecular mass of the smallest protein band (14 kDa) corresponds with that expected for the unglycosylated ␥ 2 -(1-113) fragment. Since two glycosylation sites are present in ␥ 2 -(1-113) (28), the 17-and 20-kDa bands presumably represent partially and fully glycosylated ␥ 2 -(1-113) fragments, respectively. The fragment ␥ 2 -(1-110), in contrast to ␥ 2 -(1-113) could not be co-precipitated with ␣ 1 subunits from appropriately co-transfected HEK cells (Fig. 3A), although the actual expression of both truncated constructs could be confirmed by immunoprecipitation and subsequent detection of the proteins with ␥ 2 -(1-33) antibodies (Fig. 3B). The inability to co-precipitate the fragment ␥ 2 -(1-110) indicated that ␣ 1 -(1-9) antibodies did not cross-react with truncated ␥ 2 subunits and support the conclusion of an interaction between ␥ 2 fragments and ␣ 1 subunits. As with ␥ 2 -(1-113), three protein bands were observed for ␥ 2 -(1-110) fragments corresponding to differentially glycosylated proteins. The slight differences in the apparent molecular mass of the unglycosylated and partially glycosylated ␥ 2 -(1-113) and ␥ 2 -(1-110) frag-FIG. 2. Truncated ␥ 2 subunits decrease the relative amount of ␣ 1 ␤ 3 ␥ 2 receptors expressed on the surface of transfected HEK cells. A, schematic drawing of the ␥ 2 subunit and C-terminal truncated ␥ 2 constructs. The ␥ 2 subunit consists of the N-terminal extracellular domain with the typical cystein-loop, of four transmembrane domains (TM1-4) and the large cytoplasmic loop between TM3 and TM4. The sequences of the C-terminal truncated ␥ 2 constructs are indicated by the amino acid numbers given in parentheses. 1 represents the first amino acid of the mature subunit (28). B, HEK cells were co-transfected with full-length ␣ 1 , ␤ 3 , and ␥ 2 subunits together with truncated ␥ 2 constructs (cDNA ratio 1:1:1:1) or ␤-galactosidase, a protein, that does not interfere with GABA A receptor assembly (100% value). The relative amount of receptors expressed on the cell surface was determined as described under "Experimental Procedures." The values represent the mean Ϯ S.E. from five individual determinations performed in three different experiments. Data obtained after co-transfection with . C, HEK cells expressing full-lenth ␣ 1 , ␤ 3 , and ␥ 2 subunits together with a truncated ␥ 2 construct as indicated were extracted, proteins were precipitated by methanolchloroform treatment, and subjected to SDS-PAGE and Western blot analysis using ␥ 2 -(1-33) antibodies. The strongly labeled protein with apparent molecular mass 44 -49 kDa represents the full-length ␥ 2 subunit (4). The proteins with apparent molecular mass of 30 kDa and those slightly above 20 kDa represent the respective ␥ 2 fragments. The protein bands with 43 kDa represent a degradation product of the full-length ␥ 2 subunit. Degradation of full-length ␥ 2 subunits is more prominent under conditions where assembly is reduced by interfering ␥ 2 fragments, since unassembled subunits are less stabilized against degradation than completely assembled ␣ 1 ␤ 3 ␥ 2 receptors. ments (Fig. 3B) might have been due to incomplete denaturation by SDS and a corresponding atypical migration of these proteins with low molecular mass and distinct secondary structure (see "Discussion").
Amino Acids 111-113 of the ␥ 2 Subunit Do Not Form the Binding Site for ␣ 1 or ␤ 3 Subunits-The co-precipitation of ␥ 2 -(1-113) with ␣ 1 or ␤ 3 subunits indicated that this fragment is able to directly bind to these subunits. Since ␥ 2 -(1-110) could not be co-precipitated with these subunits, it was possible that amino acids 111-113 form the binding site responsible for interaction with ␣ 1 and ␤ 3 subunits. In order to investigate this possibility, amino acids 111-113 of the ␥ 2 subunit were replaced by amino acids present at homologous positions in subunits of the same receptor superfamily. As shown in Fig. 4A, the 1 subunit of the GABA A receptor contains a hydrophobic methionine at a position homologous to the hydrophilic threonine 111 of the ␥ 2 subunit. Since it has been demonstrated that the 1 subunit is unable to assemble with ␣ 1 or ␤ 1 subunits (29), the ␥ 2 -(1-113)mut1 fragment was cloned, containing the T111M mutation (Fig. 4A). This construct was transfected into HEK cells together with ␣ 1 or ␤ 3 subunits. Cell extracts were precipitated with ␣ 1 -(1-9) or ␤ 3 -(345-408) antibodies, respectively, and the precipitate was subjected to SDS-PAGE and Western blot analysis using digoxygenized ␥ 2 -(1-33) antibodies. As shown in Fig. 4B, ␥ 2 -(1-113)mut1, similar to ␥ 2 -(1-113), was able to bind to ␣ 1 as well as to ␤ 3 subunits. In order to investigate whether the two phenylalanines at position 112 and 113 of the ␥ 2 subunit were responsible for binding to ␣ 1 and ␤ 3 subunits in the ␥ 2 -(1-113) construct (Fig. 4A), ␥ 2 -(1-113)mut2 was cloned in which amino acids 111-113 of the ␥ 2 subunit were replaced by the corresponding amino acids of the ␦ subunit of the acetylcholine receptor (Ile, Val, and Leu), that are completely different from those of the ␥ 2 subunit (Thr, Phe, and Phe). As shown in Fig. 4B, ␥ 2 -(1-113)mut2 was also able to bind to ␣ 1 and to ␤ 3 subunits in appropriately transfected HEK cells.
The observation, that binding of the ␥ 2 -(1-113) fragment to ␣ 1 or ␤ 3 subunits was not dependent on the structure of the amino acid residues 111-113, indicated that amino acids 111-113 do not form the site(s) responsible for specific interaction with ␣ 1 and ␤ 3 subunits. Since the fragment ␥ 2 -(1-110) did not bind to ␣ 1 or ␤ 3 subunits, the presence of the amino acids 111-113 might have been necessary for stabilizing the conformation of the actual binding sites in the ␥ 2 -(1-113) construct.
Identification of Amino Acid Sequences of the ␥ 2 Subunit That Are Important for Binding to ␣ 1 and ␤ 3 Subunits-In order to identify amino acid sequences of the ␥ 2 subunit that are essential for the binding to ␣ 1 or ␤ 3 subunits, it would have been possible to replace additional amino acids of the ␥ 2 -(1-113) fragment by mutagenesis until its ability to bind to ␣ 1 or ␤ 3 subunits was lost. The replacement of a single amino acid, however, could change the conformation of the sites responsible for binding to ␣ 1 or ␤ 3 subunits even if the replaced amino acid is not located in the respective binding site. In order to avoid this possibility, a different strategy was used: instead of aiming to eliminate the binding sites by mutagenesis, it was investigated which ␥ 2 amino acid sequences could induce binding to ␣ 1 or ␤ 3 subunits after incorporation into a fragment that originally could not bind to these subunits. A comparison of the N-terminal sequences of the ␣ 1 and ␥ 2 subunits indicated that the first 100 amino acids of the ␣ 1 subunit are homologous to ␥ 2 -(1-113) (Fig. 5). But in contrast to ␥ 2 (1-113), ␣ 1 -(1-100) could not be co-precipitated with ␣ 1 or ␤ 3 subunits after co-expression in HEK cells. In order to incorporate binding sites of the ␥ 2 subunit, several chimeras were constructed by replacing the C-terminal part of the ␣ 1 -(1-100) fragment with the corresponding ␥ 2 sequences (Fig. 5). These chimeras were transfected into HEK cells together with fulllength ␣ 1 or ␤ 3 subunits. Expressed subunits were precipitated from cell extracts with ␣ 1 -(328 -382) or ␤ 3 -(345-408) antibodies. The epitopes recognized by these antibodies were present on the full-length ␣ 1 or ␤ 3 subunits, respectively, but not on the truncated chimeras. The precipitate was subjected to SDS-PAGE and the proteins were detected with digoxygenized ␣ 1 -(1-9) antibodies in Western blots. The actual expression of the FIG. 4. Binding of the fragment ␥ 2 -(1-113) to ␣ 1 or ␤ 3 subunits is not dependent on the structure of amino acid residues 111-113. A, sequence alignment of the GABA A receptor ␥ 2 and 1 subunits, the acetylcholine receptor ␦ subunit, and mutated GABA A receptor constructs in the region homologous to ␥ 2 -(104 -113). Boxed amino acids are identical with those of the ␥ 2 subunit. B, HEK cells were transfected with full-length ␣ 1 or ␤ 3 subunits together with ␥ 2 -(1-113), ␥ 2 -(1-113)mut1, or ␥ 2 -(1-113)mut2, as indicated. Cell extracts were immunoprecipitated with ␣ 1 -(1-9) or with ␤ 3 -(345-408) antibodies, respectively. The precipitate was subjected to SDS-PAGE and Western blot analysis using digoxygenized ␥ 2 -(1-33) antibodies. The experiment was performed three times with similar results. chimeras was confirmed by precipitation and detection with ␣ 1 -(1-9) antibodies (data not shown).
Although in Chim1 the nine C-terminal amino acids of the ␣ 1 -(1-100) fragment were replaced by amino acids 105-113 of the ␥ 2 subunit, due to the homology of sequences this chimera differed from ␣ 1 -(1-100) in one amino acid (T95I) only. As indicated in Fig. 5, this chimera could not be co-precipitated with full-length ␣ 1 or ␤ 3 subunits from appropriately co-transfected HEK cells, demonstrating the specificity of the ␣ 1 -(328 -382) and ␤ 3 -(345-408) antibodies used and indicating that amino acids ␥ 2 -(105-113) are not able to induce the formation of binding sites for ␣ 1 or ␤ 3 subunits in ␣ 1 -(1-100). In Chim2, the amino acid sequence ␣ 1 -(78 -100) was replaced by ␥ 2 -(91-113). This construct was able to bind to full-length ␣ 1 , but not to full-length ␤ 3 subunits in appropriately co-transfected cells (Fig. 5). Since amino acids ␥ 2 -(105-113) were not sufficient to induce binding as discussed above, this indicated that amino acids 91-104 of the ␥ 2 subunit are important for binding to ␣ 1 subunits. Finally Chim3 was constructed, in which the amino acid sequence ␣ 1 -(70 -100) was replaced by ␥ 2 -(83-113) (Fig. 5). Chim3 not only was able to bind to full-length ␣ 1 , but also to full-length ␤ 3 subunits in appropriately transfected HEK cells. These results indicated that the additionally incorporated amino acids 83-90 of the ␥ 2 subunit are able to induce the formation of the site responsible for binding to full-length ␤ 3 subunits.
Interestingly, three protein bands representing differentially glycosylated constructs were observed in the experiments with Chim3. This was similar to previous experiments (Fig. 3) performed with ␥ 2 -(1-113), but was in contrast to experiments with ␣ 1 -(1-100), Chim1, or Chim2, where only two protein bands were observed. The only amino acid that could be glycosylated and differed between Chim3 and Chim2 was the newly introduced Asn 90 of the ␥ 2 subunit. These results suggest that Asn 90 of the ␥ 2 subunit is actually glycosylated, confirming a previous prediction on glycosylation of this amino acid (28).
To further demonstrate the importance of the sequences ␥ 2 -(83-92) and ␥ 2 -(91-104) for the assembly of ␣ 1 ␤ 3 ␥ 2 receptors, it was investigated whether Chim4 or Chim6 could inhibit the expression of ␣ 1 ␤ 3 ␥ 2 receptors on the cell surface. For this Chim4 or Chim6 were transfected into HEK cells together with full-length ␣ 1 , ␤ 3 , and ␥ 2 subunits and the amount of receptors on the cell surface was determined. In contrast to ␣ 1 (1-100), Chim4 and Chim6 were able to significantly reduce expression of ␣ 1 ␤ 3 ␥ 2 receptors on the cell surface and the percent reduction was similar to that of the ␥ 2 -(1-113) construct (Fig. 5).
Secondary Structure Prediction for Amino Acids 71-113 of the ␥ 2 Subunit-To investigate the secondary structure of the amino acid sequence ␥ 2 -(71-113), a neural network-based prediction using PHDsec (20,21) was performed. As shown in Fig.  6A, the probability for the formation of an ␣-helix (solid line) was relatively high for amino acids 83-88. The probability for the formation of an extended structure (dashed line) was highest for amino acids 72-77 and 93-98, whereas the probability for the formation of a loop (dotted line) was highest for amino acids 80 -83, 89 -93, 98 -103, and 108 -113. Overall, from this prediction the probability for the formation of long range ␣-helices or ␤-sheet structures is low. Interestingly, however, the predicted short ␣-helix for amino acids ␥ 2 -(83-88) is located within the experimentally identified region important for binding to both ␤ 3 and ␣ 1 subunits.
To evaluate the significance of the prediction in this region, force field calculations within a DMC optimization scheme were performed. A pool of structurally equivalent low energy conformations could be calculated that were separated by 8 -12.5 kJ/mol from other local minimum structures and represented an unconstrained ␣-helix (backbone dihedral angles in the range of Ϫ60°) comprising one and a half turns. The ␣-helical conformation in this region was defined by electro- FIG. 5. Identification of ␥ 2 sequences that induce binding of the ␣ 1 -(1-100) fragment to ␣ 1 or ␤ 3 subunits. C-terminal sequences of the fragments ␥ 2 -(1-113) and ␣ 1 -(1-100) and of different chimeras are shown. Amino acids of the ␥ 2 subunit are boxed. HEK cells were co-transfected with these constructs together with ␣ 1 or ␤ 3 subunits and a possible co-precipitation was investigated as described in the text. ϩ indicates binding, and Ϫ indicates absence of binding between these constructs and full-length ␣ 1 or ␤ 3 subunits. The experiments were performed three times with similar results. The reduction of ␣ 1 ␤ 3 ␥ 2 receptors on the cell surface by the respective constructs was investigated as described in the legend to Fig.  2. The values represent the mean Ϯ S.E. from at least four individual determinations. Reduction obtained for constructs Chim4 (p ϭ 0.010, unpaired Student's t test) and Chim6 (p ϭ 0.018, unpaired Student's t test) was significant. n.d., not determined. static interactions along the side chains as well as by formation of a backbone hydrogen bond between amino acid residues Asp 84 and Lys 88 . This ␣-helical element should therefore maintain its conformation also in the context of tertiary and quaternary structure of the receptor. Data base knowledge, as well as energetic considerations thus indicated that amino acids 83-88 of the ␥ 2 subunit presumably form a short ␣-helix.
For amino acids 91-104 of the ␥ 2 subunit, that are sufficient to induce binding to ␣ 1 but not to ␤ 3 subunits, two loops (dotted line) and a short extended region (dashed line) are predicted by PHDsec (Fig. 6A). A DMC optimization for this sequence resulted in a large number of energetically equivalent, but structurally different conformations. Based on the results provided by PHDsec and the DMC optimization no defined secondary structure could be predicted for this sequence. This sequence, thus, should be influenced strongly by folding events induced by the tertiary and quaternary structure of the protein.
Interestingly, two of these regions are identical with the ␥ 2 -(83-90) or ␥ 2 -(91-104) sequences identified to be important for binding to both ␤ 3 and ␣ 1 or ␣ 1 subunits, respectively. Thus, amino acid residues 83-90 form a hydrophilic structure and amino acid residues 91-104 form a structure with intermediate hydrophobicity. In contrast, amino acid residues 105-113 that have been demonstrated to stabilize the conformation of the binding sites, are highly hydrophobic.

DISCUSSION
The N-terminal 113 Amino Acids of the ␥ 2 Subunit Contain Recognition Sites Important for Assembly with ␣ 1 and ␤ 3 Subunits-The present study demonstrated that the N-terminal domain of the GABA A receptor ␥ 2 subunit (␥ 2 -(1-234)) could be co-precipitated with full-length ␣ 1 or ␤ 3 subunits after co-expression in HEK cells. This indicated that ␥ 2 -(1-234) was able to bind to ␣ 1 as well as ␤ 3 subunits. These results are consistent with in vitro translation experiments, demonstrating that Nterminal sequences without transmembrane domains of GABA A receptor subunits (30) or of K ϩ channel subunits (31) can bind to and be co-precipitated with full-length subunits. Binding between ␥ 2 -(1-234) fragments and full-length ␣ 1 or ␤ 3 subunits seemed to be fairly stable, because it survived extraction by detergent, immunoprecipitation, and several washing steps. In addition, heterodimers consisting of ␥ 2 -(1-234) fragments and full-length ␣ 1 subunits were able to form specific high affinity benzodiazepine-binding sites assumed to be formed at the interface of ␣ 1 and ␥ 2 subunits in GABA A receptors (26). This indicated that binding between truncated ␥ 2 constructs and the full-length subunits was not caused by unspecific interaction, but by a specific assembly process. This conclusion was strengthened by the observation that ␥ 2 -(1-234) on co-transfection with ␣ 1 , ␤ 3 , and ␥ 2 subunits was able to decrease the amount of ␣ 1 ␤ 3 ␥ 2 receptors expressed on the cell surface, suggesting an interference of the truncated subunit with receptor assembly (11).
A subsequent reduction in the size of the truncated ␥ 2 subunit indicated, that ␥ 2 -(1-113) was the smallest N-terminal fragment that could be co-precipitated with ␣ 1 or ␤ 3 subunits and could interfere with ␣ 1 ␤ 3 ␥ 2 receptor assembly. These results demonstrated that the N-terminal 113 amino acids of the ␥ 2 subunit contain binding sites for ␣ 1 as well as for ␤ 3 subunits and that these sites are important for GABA A receptor assembly.
The observation that the longest N-terminal fragment exhibits the strongest interference with GABA A receptor assembly (Fig. 2B), can be explained by the possibility that longer fragments are more able to stabilize their binding sites for ␣ 1 or ␤ 3 subunits. Alternatively, additional contact sites for these subunits could be located between ␥ 2 -(113-234).
Amino Acid Sequences ␥ 2 -(91-104) and ␥ 2 -(83-90) Mediate Assembly with ␣ 1 and ␤ 3 Subunits-In contrast to ␥ 2 -(1-113), the fragment ␥ 2 -(1-110) could not interfere with ␣ 1 ␤ 3 ␥ 2 receptor assembly and this fragment could not be co-precipitated with full-length ␣ 1 or ␤ 3 subunits, indicating that the presence of amino acids 111-113 is necessary for assembly with ␣ 1 and ␤ 3 subunits. After replacement of amino acid residues 111-113 in the ␥ 2 -(1-113) fragment by site-directed mutagenesis, the resulting fragments were still able to bind to ␣ 1 and ␤ 3 subunits. These results indicate that the structure of amino acid residues 111-113 of the ␥ 2 subunit is not essential for binding to ␣ 1 and ␤ 3 subunits. These amino acid residues therefore do not directly form these binding sites. A hydrophobicity analysis for amino acid residues 70 -135 (Fig. 6B) indicated that amino acid residues 105-113 form a highly hydrophobic structure. Presumably, an appropriate length of this sequence is required for an efficient interaction with other hydrophobic regions in order to stabilize the conformation of the actual binding sites located more N-terminal.
Predictions on the Structure of Amino Acid Sequences ␥ 2 -(83-90) and ␥ 2 -(91-104)-Hydrophobicity analysis, a neural network-based secondary structure prediction (PHDsec), as well as force field optimization within a Dynamic Monte Carlo scheme, indicated that the sequence ␥ 2 -(83-90), that induces binding to ␤ 3 and ␣ 1 subunits, forms a hydrophilic structure that contains a short ␣-helix. Since electrostatic interactions and a hydrogen bond stabilize the ␣-helix as a structural element, this helix is ideally suited as a specific recognition site and presumably interacts with another hydrophilic structure on the ␤ 3 or ␣ 1 subunit. The ␣-helix is terminated by an aromatic phenylalanine and the asparagine 90. Since glycosylation of Asn 90 is essential for the formation of the binding site for ␤ 3 and ␣ 1 subunits mediated by ␥ 2 -(83-90), the introduced carbohydrate chains might stabilize the three-dimensional position of this hydrophilic structure in the subunit. It is reasonable to assume that the sugar chains bound to Asn 90 are located outside the ␥ 2 /␤ 3 (or ␥ 2 /␣ 1 ) contact site, and exhibit a position that does not impede free access of ions to the channel in the completely assembled receptor (Fig. 7).
Interestingly, at the transition between ␥ 2 -(83-90) and ␥ 2 -(91-104) a quite abrupt shift in hydrophobicity is observed (Fig. 6B). Amino acid sequence ␥ 2 -(91-104) that is sufficient to induce binding to ␣ 1 subunits forms a structure with intermediate hydrophobicity. PHDsec and a force field optimization did not result in a defined energetic minimum conformation. It is thus reasonable to assume that the conformation of this hydrophobic ␥ 2 sequence is strongly influenced by binding to a complementary hydrophobic structure on the ␣ 1 subunit, indicating an induced fit mechanism of ␣ 1 /␥ 2 assembly mediated by the sequence ␥ 2 -(91-104).
Amino Acid Sequence ␥ 2 -(91-104) Forms Part of the Interface to ␣ 1 Subunits-Amino acid residues Phe 77 and Met 130 of the ␥ 2 subunit have been shown to be involved in the formation of the benzodiazepine-binding site of GABA A receptors that is assumed to be located at the interface of ␥ 2 and ␣ 1 subunits (26). In order to contribute to this site, these residues not only should be located close to the ␣ 1 /␥ 2 interface, but should also be located in close proximity to each other. Interestingly, the amino acid sequence ␥ 2 -(91-104) that is sufficient to induce binding to the ␣ 1 subunit is located between these two amino acid residues in the primary structure of this subunit. Since amino acid Phe 77 is located in the center of a highly hydrophobic region (Fig. 6B), it is reasonable to assume that Phe 77 is not located on the surface but in the interior of the ␥ 2 subunit. The suggested interaction of Phe 77 with the phenyl substituent of diazepam during binding to the benzodiazepine site of GABA A receptors (32) will therefore strongly affect the conformation of the ␥ 2 subunit.
In contrast, Met 130 is located at the end of the sequence ␥ 2 -(121-130) that forms a structure with a hydrophobicity comparable to that of ␥ 2 -(91-104) (Fig. 6B). It is tempting to speculate that amino acids ␥ 2 -(121-130) form another part of the ␣ 1 /␥ 2 contact site, the core of which is formed by ␥ 2 -(91-104). Interaction of diazepam with Met 130 might then increase the cleft between ␣ 1 and ␥ 2 subunits and thus cause an additional change in the conformation of the GABA A receptor.
In addition to Phe 77 and Met 130 of the ␥ 2 subunit, amino acid residues His 101 , Tyr 159 , Thr 206 , and Tyr 209 of the ␣ 1 subunit also contribute to the formation of the benzodiazepine binding pocket (26). The observation that these ␣ 1 amino acid residues (data not shown), as well as Phe 77 and Met 130 of the ␥ 2 subunit are located in hydrophobic regions of the respective subunits, further supports the conclusion that the whole interface between ␣ 1 and ␥ 2 subunits is formed by hydrophobic interactions and that the sequence ␥ 2 -(91-104) contributes to this interface. In the completely assembled receptor, Phe 77 and Met 130 of the ␥ 2 subunit must be in a position close to His 101 , Tyr 159 , Thr 206 , and Tyr 209 of the ␣ 1 subunit to form the benzodiazepine binding pocket. The additional observation that a benzodiazepine-binding site is formed by heterodimers consisting of full-length ␣ 1 and truncated ␥ 2 -(1-234) constructs, thus, supports the conclusion that amino acid residues forming this site exhibit the same relative position in heterodimers as in intact ␣ 1 ␤ 3 ␥ 2 receptors. This indicates that the sequence ␥ 2 -(91-104) identified to form the primary contact between ␥ 2 and ␣ 1 subunits also forms part of the interface between these subunits in the completely assembled receptor (Fig. 7).
Amino Acid Sequence ␥ 2 -(83-90) Forms Part of the Interface to ␤ 3 Subunits-The observation that amino acids ␥ 2 -(83-92) not only mediate binding to ␤ 3 but also to ␣ 1 subunits could be explained by the possibility that this sequence forms the primary binding site to ␣ 1 and ␤ 3 subunits and that further assembly steps and conformational changes are necessary for the formation of the final interface between subunits. The robust binding between the truncated ␥ 2 subunits and ␤ 3 or ␣ 1 subunits, however, indicates that a shift of contact sites would only be possible by a change in the conformation of the ␥ 2 -(83-90) sequence mediating this binding. This is highly unprobable. Force field calculations within a Dynamic Monte Carlo optimization scheme indicated that the ␣-helix formed by ␥ 2 -(83-90) is stable enough to be retained in the tertiary and quaternary structure of the receptor. In addition, the hydrophilic sequence neighboring sequences and thus, a conformational change of ␥ 2 -(83-90) is not easily possible. Therefore it can be assumed that the sequence ␥ 2 -(83-90) not only forms the primary contact site but also the final interface between ␥ 2 and ␤ 3 or ␥ 2 and ␣ 1 subunits in the completely assembled receptor.
Evidence discussed above indicates that the ␣ 1 /␥ 2 interface is formed by hydrophobic interactions partially mediated by the sequence ␥ 2 -(91-104). The hydrophilic sequence ␥ 2 -(83-90) then presumably forms the interface between ␥ 2 and ␤ 3 subunits in completely assembled ␣ 1 ␤ 3 ␥ 2 receptors (Fig. 7). In the absence of ␤ 3 subunits, however, the sequence ␥ 2 -(83-90) possibly could be used to accommodate a second ␣ 1 subunit. Assembly of a ␥ 2 with two ␣ 1 subunits is consistent with the observation that ␣ 1 and ␥ 2 subunits under certain conditions are able to form chloride ion channels that can be opened by GABA (2,33). Conflicting results on the existence of pentameric ␣ 1 ␥ 2 receptors (2, 33), however, and recent studies indicating that the ␣ 1 ␥ 2 subunit combination predominantly formed subunit dimers (4) that could not be identified on the surface of transfected cells (10,15), indicate that ␣ 1 ␥ 2 receptors form less readily than ␣ 1 ␤ 3 ␥ 2 receptors. As with the nicotinic acetylcholine receptor (34), additional sequences C-terminal to ␥ 2 -(1-113) might exist that could contribute to a selective assembly with ␣ 1 or ␤ 3 subunits. Further experiments will have to identify these sequences as well as the respective counterparts of the ␥ 2 -binding sites on ␣ 1 and ␤ 3 subunits.