The N-terminal domain of human GABA receptor rho1 subunits contains signals for homooligomeric and heterooligomeric interaction.

gamma-Aminobutyric acid type C (GABAC) receptors identified in retina appear to be composed of GABA rho subunits. The purpose of this study was to localize signals for homooligomeric assembly of rho1 subunits and to investigate whether the same region contained signals for heterooligomeric interaction with rho2 subunits. In vitro translated human rho1 was shown to be membrane-associated, and proteinase K susceptibility studies indicated that the N terminus was oriented in the lumen of ER-derived microsomal vesicles. This orientation suggested the involvement of the N terminus of rho1 in the initial steps of subunit assembly. To test this hypothesis, mutants were created containing only N-terminal sequences (N-rho1) or C-terminal sequences (C-rho1) of rho1. Co-immunoprecipitation studies revealed that N-rho1, but not C-rho1, interacted with rho1 in vitro. When coexpressed in Xenopus oocytes, N-rho1 interfered with rho1 receptor formation. Together, these data suggested that signals for rho1 homooligomeric assembly reside in the N-terminal half of the subunit. Sequential immunoprecipitations were then performed upon cotranslated rho1 and rho2 subunits which demonstrated that rho1 and rho2 interacted in vitro. Co-immunoprecipitation indicated that N-rho1 specifically associated with rho2. Therefore, the N-terminal regions of rho subunits contain the initial signals for both homooligomeric and heterooligomeric assembly into receptors with GABAC properties.

␥-Aminobutyric acid type C (GABA C ) receptors identified in retina appear to be composed of GABA subunits. The purpose of this study was to localize signals for homooligomeric assembly of 1 subunits and to investigate whether the same region contained signals for heterooligomeric interaction with 2 subunits. In vitro translated human 1 was shown to be membrane-associated, and proteinase K susceptibility studies indicated that the N terminus was oriented in the lumen of ERderived microsomal vesicles. This orientation suggested the involvement of the N terminus of 1 in the initial steps of subunit assembly. To test this hypothesis, mutants were created containing only N-terminal sequences (N-1) or C-terminal sequences (C-1) of 1. Coimmunoprecipitation studies revealed that N-1, but not C-1, interacted with 1 in vitro. When coexpressed in Xenopus oocytes, N-1 interfered with 1 receptor formation. Together, these data suggested that signals for 1 homooligomeric assembly reside in the N-terminal half of the subunit. Sequential immunoprecipitations were then performed upon cotranslated 1 and 2 subunits which demonstrated that 1 and 2 interacted in vitro. Co-immunoprecipitation indicated that N-1 specifically associated with 2. Therefore, the N-terminal regions of subunits contain the initial signals for both homooligomeric and heterooligomeric assembly into receptors with GABA C properties.
In the brain and retina, inhibitory neurotransmission is predominantly achieved through activation of receptors for GABA 1 (1). A novel class of receptors for GABA, termed type C or GABA C , has been identified in the retinae of a variety of species including white perch, catfish, hybrid bass, salamander, rat, and cow (2)(3)(4)(5)(6)(7). GABA C receptors share a number of similarities with the well-characterized GABA A receptors including an integral Cl Ϫ channel that can be inhibited by picrotoxin and sensitivity to divalent metal cations such as Zn 2ϩ (2)(3)(4). However, these novel receptors are insensitive to compounds that have classically been used to define GABA receptor types (2)(3)8). Furthermore, the ligand-binding site of GABA C receptors is distinguished from GABA A and GABA B receptors in its apparent preference for GABA analogues that can adopt a folded conformation, such as cis-4-aminocrotonic acid, over those that are locked in an extended conformation (9 -12). These observations indicate that GABA C receptors may have a molecular structure similar to their closest relative, GABA A receptors, but with distinct physical differences that confer unique biophysical and pharmacologic properties.
Two GABA receptor subunits, 1 and 2, have been isolated from a human retina library that have amino acid and structural similarity to members of the ligand-gated ion channel superfamily and to the GABA A receptor subunits in particular (13)(14)(15)(16). However, human subunits form homooligomeric receptors in Xenopus oocytes that are insensitive to bicuculline, barbiturates, and benzodiazepine (13,(15)(16). Also, receptors have a higher affinity for GABA and little desensitization compared with GABA A receptors (16). The pharmacologic similarities between GABA C receptors and subunits lead to the suggestion that subunits are the molecular components of GABA C receptors (12,15). The localization of both subunits by in situ hybridization and immunochemistry to rat rod bipolar cells where GABA C responses have been detected further supports the inclusion of subunits in GABA C receptors (17)(18). Although human 1 and 2 are each capable of forming homooligomeric receptors, there is functional evidence suggesting rat 1 and 2 may form heterooligomeric receptors (19). Thus, it is possible that GABA C receptors in vivo are homooligomeric or heterooligomeric combinations of 1 and 2 subunits.
In this study, in vitro translated protein and Xenopus oocyte expression were used to identify sequences involved in subunit assembly. The in vitro translation system was chosen for two reasons: the ability of protein interaction from signals in the primary amino acid sequence can be examined independent of cell-specific factors, and the availability of each protein for interaction can be verified. A critical question was the orientation of the in vitro translated subunit protein relative to the membrane. Using protease-susceptibility studies, we demonstrated that the orientation of 1 is consistent with the predicted topology of the ligand-gated ion channel superfamily, with an extracellular N-terminal domain and a membraneassociated C-terminal domain. Specific antibodies were developed to test the interaction of 1 with truncated forms of 1 and with 2. Our results demonstrate that 1 assembles with 2 in vitro, implying that they are capable of forming a single receptor in vivo, and that the N terminus of 1 contains signals for homooligomeric and heterooligomeric assembly.

cDNA Constructs
1FLAG-An 8-residue synthetic epitope (FLAG: DYKDDDDK) recognized by the mouse monoclonal M2 antibody (Eastman Kodak) was inserted in-frame at the C terminus of 1 by a three-way ligation of two annealed oligonucleotides with MaeI and NcoI sites (sense: 5Ј-TAC-GACTACAAGGACGACGATGACAACAAGTAGC-3Ј; antisense: 5Ј-CATGGCTACTTGTCATCGTCGTCGTCCTTGTAGTCG-3Ј), a KpnI/ MaeI fragment containing the 1 cDNA, and a KpnI/NcoI-digested pBluescript vector (Stratagene). The amber termination codon of 1 was replaced by an amber codon in the oligonucleotides. Oocyte injection and current recordings demonstrated that 1FLAG formed homooligomeric GABA-gated receptors with properties indistinguishable from native 1 (22).
2HA-A triplicated 9-residue epitope of the influenza virus hemagglutinin (HA tag: TPYDVPDYA) recognized by the monoclonal antibody 12CA5 (20) was ligated into either the N terminus or C terminus of the human 2 subunit. Kpn2I recognition sites were included in PCR primers complementary to the 5Ј and 3Ј ends of the HA tag (5ЈHA-Kpn2I 5Ј-ATCCGGATCTTTTACCCATAC-3Ј, restriction site is underlined; 3ЈHA-Kpn 2I 5Ј-TTCCGGAGAGCAGCGTAATCTGG-3Ј). Following PCR and subcloning into pCRII (Invitrogen), a 120-bp Kpn2I fragment with the correct HA tag sequence was ligated into the Kpn2I site of 2 at nucleotide 229 in the N terminus of 2, 30 residues downstream from the predicted signal peptide cleavage site. Inserting the tag at this position was predicted to not alter 2 channel function on the basis of a chimera domain-swapping strategy which excluded this region from being necessary for robust homooligomeric expression (21). To introduce the HA epitope to the C terminus of 2, StyI sites were included in PCR primers HA-forward (5Ј-TTCCAAGGGCCGCATCTTTTAC-3Ј) and HA-reverse (5Ј-ATCCTTGGCTAGCACTGAGCAGCGTAATC-3Ј). The 118-bp PCR product was subcloned into the StyI site at the 2 termination codon (nucleotide 1474). The resulting construct was sequenced in entirety. Both 2-5ЈHA and 2-3ЈHA were tested for interaction with 1 and will be referred to as 2HA for simplicity.
1 Mutants-N-1 was created by replacing the tyrosine residue of 1 at codon 256 with an amber termination codon (TAG) using the mutagenic oligonucleotide 5-CTGTGCTGCTCTAGAAAGCCA-3Ј. The resulting mutant 1 is truncated 18 residues prior to the predicted first transmembrane domain (TM) and represents the N-terminal half of 1. C-1 was created by deleting the 700-bp NlaIV-RsaI fragment containing the entire N terminus of 1, leaving the four putative hydrophobic segments and the sequences between each. This strategy retained the native canonical methionine and the putative signal peptide of 1. The mutations were confirmed by sequencing.

In Vitro Transcription and Translation
cRNA was synthesized from linearized 1 and 2 cDNA using the appropriate RNA polymerase and the Megascript In Vitro Transcription Kit (Ambion). The cRNA was purified by phenol/chloroform extraction, precipitated by ethanol, and resuspended in diethyl pyrocarbonate-treated water. The integrity of the cRNA was verified on a 1.5% agarose-formaldehyde gel. GABA subunits were synthesized using the Flexi Rabbit Reticulocyte Translation System (Promega), in the presence of canine pancreatic microsomal membranes (Promega, Boehringer Mannheim). Synthesized protein was detected by incorporation of 4 mCi/ml (specific activity, 1180 mCi/mmol) [ 35 S]methionine and [ 35 S]cysteine (DuPont NEN), and resolved on SDS-PAGE (10%) gels using one-fifth of each translation reaction mixed with an equal volume of sample buffer (15% glycerol, 5% ␤-mercaptoethanol, 4.5% SDS, 100 mM Tris-Cl, pH 6.8, 0.03% bromphenol blue). Gels were fixed, dried, and exposed to Hyper-film (Amersham Corp.) for 16 h to 4 days. The molecular masses (in kDa) of the translated proteins were derived using standard curves generated from protein size standards (Amersham Corp.).

Topological Analysis of in Vitro Translated 1
Membrane Association-In vitro translated 1 was transferred to polycarbonate tubes and centrifuged at 80,000 rpm in a Beckman TLA-100 rotor for 2 h at 4°C to pellet the microsomal membranes. The pellet and an aliquot of the supernatant were resuspended in sample buffer and subjected to electrophoresis. As a control for membrane association the hydrophobic water channel aquaporin-1 (cRNA provided by Dr. J. Reeves, Johns Hopkins University) was translated and centrifuged in parallel.
Protease Protection-One-half of the translate was incubated with 0.1 mg/ml proteinase K (Life Technologies, Inc.) alone, and the other half was incubated with proteinase K and 1% Triton X-100. Both digestions were carried out for 60 min at 0°C and then terminated by the addition of the protease inhibitor phenylmethylsulfonyl fluoride (1.0 mM) for 10 min. The reactions were mixed with sample buffer and loaded onto a 10% polyacrylamide gel.

Development of 1-Specific Antibodies
hASH-1-The 1-specific peptide (QRQRREVHEDAHK) representing a 13-residue epitope at the N terminus of 1 was conjugated at its C terminus to the carrier protein keyhole limpet hemocyanin prior to emulsification and injection. Peptide synthesis, keyhole limpet hemocyanin conjugation, and rabbit handling were performed by Research Genetics, Huntsville, AL. All experiments described here were performed with 10-week bleed antisera.
hASH-2-PCR primers 5Ј TM3-TM4-BamHI (5Ј-ACGGATCCGT-CAACTACCTGACC-3Ј) and 3Ј TM3-TM4-EcoRI (5Ј-CAGAATTCAAT-GGCGTGGGTGTC-3Ј) were used to amplify the TM3-TM4 loop of 1 from residues 358 to 445. The 280-bp product was subcloned into pCRII (Invitrogen) and sequenced and then shuttled into pGEX-2T (Pharmacia Biotech) using BamHI and EcoRI sites to create a glutathione S-transferase fusion protein. A 200-ml HB 101 bacterial culture (Life Technologies, Inc.) expressing the fusion protein was pelleted and lysed in 10 ml of lysis buffer (1% Nonidet P-40 in PBS) with 0.5 mg/ml lysozyme (Sigma) for 30 min at room temperature. The lysate was sonicated with a 5-mm diameter probe for 4 ϫ 10 s and then clarified by centrifugation. The clarified sonicate was mixed overnight with 1 ml of a 50% slurry of glutathione-Sepharose (Sigma) with 1 mM phenylmethylsulfonyl fluoride at 4°C. The fusion protein bound to Sepharose beads was pelleted at 500 ϫ g for 2 min at 4°C, washed four times with cold 1% Triton X-100 in PBS, and then washed three times with cold PBS. The fusion protein was eluted from the beads using an equal volume of 15 mM reduced glutathione, 50 mM Tris-Cl, pH 8.0, for 1 h at room temperature or with sample buffer for 10 min at 100°C. The eluted product was cut from a preparative 15% SDS-PAGE gel and sent to Research Genetics for purification and injection of rabbits. Antisera obtained 10 weeks post-immunization were used in all experiments.
The specificity of the anti-1 antibodies was tested by cell staining. A mammalian cell line (HEK 293) that stably expresses human 1 (22) was used as the protein source. Wild-type HEK 293 cells and 1expressing cells were seeded onto glass coverslips (VWR) coated with purified collagen (Collagen Corp.) and grown to approximately 75% confluency. The cells were washed 2 or 3 times with PBS and then fixed in 3% paraformaldehyde solution for 20 min at room temperature. The fix was removed, and the cells were rinsed in PBS and then permeabilized in 1% Triton X-100/PBS for 5 min. The cells were immunostained with the polyclonal antisera, and 1 protein was detected using a horseradish peroxidase staining method. In the first method, endogenous horseradish peroxidase was inhibited by incubating the cells in 1 ϫ blocking solution for 10 min at room temperature and removed by washing in two changes of PBS. The primary antibody (hASH-1 or hASH-2) diluted 1:50 -1:500 was added to the cells for 20 min and then the cells were washed three times in PBS. The secondary antibody, biotin-conjugated anti-rabbit IgG (Boehringer Mannheim), was added at 1:100 -1:1000 dilution for 20 min and then washed as above. A 1:1000 dilution of streptavidin-horseradish peroxidase (Boehringer Mannheim) was added for 20 min, followed by the True-Blue substrate (Kirkegaard & Perry Laboratories) for 10 min. The cells were washed in water and then air-dried before being analyzed.

Immunoprecipitation
In vitro translation reactions were diluted to 200 l with buffer A (10 mM Tris-Cl, 140 mM NaCl, 0.5% Triton X-100, and 0.1% bovine serum albumin) and 10 l of a 50% slurry of protein-A Sepharose CL-4B (Pharmacia) was added. The resulting solution was mixed for 3 h at 4°C and then centrifuged; antibody was added to the clarified supernatant at the following concentrations: 2 g of anti-FLAG mouse monoclonal antibody M2 (Eastman Kodak), 1:40 dilution of polyclonal antisera hASH-1 or hASH-2, or 1:1000 dilution of anti-HA mouse monoclonal antibody 12CA5 (Boehringer Mannheim). The solution was mixed for 16 h at 4°C and then 30 l of 50% protein A-Sepharose was added. The antibody-antigen complex was pelleted by centrifugation at 200 ϫ g, and the supernatant containing unbound proteins was saved. The pellet was washed twice with buffer A, twice with buffer B (10 mM Tris-Cl, 140 mM NaCl), and once with 50 mM Tris-Cl, pH 6.8. The proteins were eluted in sample buffer by heating at 56°C for 10 min prior to SDS-PAGE electrophoresis. To evaluate the synthesis reaction, aliquots of the supernatant, or the translate prior to precipitation, were included for each immunoprecipitation. For sequential immunoprecipitations, the immunoprecipitation pellet was washed as above and then eluted in 0.1 M glycine, pH 3.0, neutralized by 0.25 volume of 1 M Tris-Cl, pH 7.4, and brought up to a volume of 200 l with buffer A. The second antibody was added to precipitate complexed proteins by the standard protocol described above.

Oocyte Expression of Wild-type and Mutant 1
Cytoplasmic injection of cRNA and the protocol for electrophysiologic studies were as previously published (16). For assembly studies, oocytes were co-injected with 5 ng of wild-type 1 cRNA and an equimolar amount or a 2-and 5-fold molar excess of N-1 or C-1 cRNA. Seventytwo hours after injection, whole-cell current in response to 5 M GABA was measured by two microelectrode voltage clamp in ND96 solution and compared with maximum currents obtained in oocytes injected with 1 alone. ND96 solution contains 96 mM NaCl, 2 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , and 5 mM Hepes-NaOH, pH 7.4. Student's t test was used to determine the statistical significant difference between two sample groups. Independent sampling and comparison were used, and p Ͻ 0.01 was set as a significant level.

RESULTS
In Vitro Translated 1 Is Membrane-associated with the N Terminus Residing in the Microsomal Vesicle Lumen--A cellfree synthesis system was utilized to produce sufficient levels of subunit protein for study. The predominant in vitro translated 1 protein resolved at approximately 62 kDa (range 52-65 kDa), which is 10 kDa larger than the size predicted from amino acid composition (Fig. 1). In addition to the major 1 62-kDa polypeptide, less intense bands of 58, 56, and 48 kDa were occasionally present (see Fig. 5). In vitro translated 2 protein resolved at approximately 58 kDa (range 50 -60 kDa), which is 7 kDa larger than the predicted size for this protein. A second, less intense, band at 54 kDa was occasionally observed. To aid in the analysis of the subunits, 1 and 2 were tagged with epitopes: 1 was tagged with FLAG at the C terminus (20), and two 2 constructs were made with the influenza hemagglutinin antigen tag (HA) (23) at either the N or C termini. The electrophoretic migration patterns of in vitro translated 1FLAG and 2HA were indistinguishable from 1 and 2, respectively. Two 1 mutants were used in this study. N-1 included the 1 N terminus sequence up to residue 256, and C-1 contained the signal peptide, the native initiating methionine, and the four hydrophobic domains (Fig. 2). N-1 and C-1 synthesized in vitro migrated as single bands close to their predicted sizes of 30 and 26 kDa, respectively (Fig. 1).
By analogy to other members of the ligand-gated neurotransmitter receptor family, GABA subunits are predicted to have four hydrophobic domains that are associated with membranes ( Fig. 2). We evaluated membrane association of the 1 subunit in vitro by translating proteins in the presence of microsomal vesicles derived from the endoplasmic reticulum of canine pancreatic cells. Microsomal membranes and their associated proteins were then pelleted by high-speed centrifugation. The water channel aquaporin-1, which is heavily hydrophobic and has been shown to be membrane-associated (24), was used as a control in this study. As expected, in vitro translated aquaporin-1 was found predominantly in the membranous pellet fraction (Fig. 3A). Similarly, in vitro translated 1FLAG partitioned predominantly in the membrane fraction, with a minor portion in the soluble fraction (Fig. 3A).
To assess the topology of 1FLAG relative to the microsomal vesicle membranes, proteinase K susceptibility studies were performed. The lipid bilayer of the vesicle membrane acts as a physical barrier, protecting portions of polypeptides translocated into the vesicle lumen from digestion by proteinase K. Regions of a protein that have an extracellular orientation in vivo will be enclosed within the lumen of the microsomal vesicle in this in vitro assay. 1FLAG was reduced in size to about 32 kDa following treatment with proteinase K (Fig. 3B, 2nd lane). Upon disruption of the microsomal membrane, 1FLAG was completely degraded (3rd lane). Thus, about half of the 1 protein was protected by the membrane of the microsomal vesicle. To determine which part of 1FLAG was protected from degradation, protease susceptibility studies were performed upon the N-1 and C-1 mutants. In the presence of microsomal vesicles, N-1 was completely protected from proteinase K degradation (Fig. 3B, 5th lane). Disruption of membranes with detergent resulted in degradation of N-1 (6th lane) indicating that the protease was able to cleave N-1. Conversely, microsomal membranes did not protect C-1 from digestion (Fig. 3B,  8th lane). Together, these results indicate that the N-terminal half of full-length 1 is located in the lumen of the microsome and predicts that this region will have an extracellular orientation in vivo.
Polyclonal Anti-Antibodies Specifically Immunoprecipitated in Vitro Translated 1 Subunits-The nucleic analysis software program PCGene (IntelliGenetics, Inc.) was used to identify the most antigenic region specific to 1 in the N terminus of the protein (see Fig. 2). A 13-residue peptide corresponding to amino acids 37 to 49 of 1 was synthesized, conjugated to keyhole limpet hemocyanin, and injected into rabbits to produce the polyclonal antisera hASH-1. A second polyclonal antisera, hASH-2, was obtained from rabbits injected with a glutathione S-transferase fusion protein containing the putative intracellular loop between TM3 and TM4 of 1 (Fig. 2). The specificity of both antisera was tested by immunodetection of FIG. 1. Cell-free synthesis of 1, 2, N-1, and C-1 FIG . 2. Predicted orientation of a single 1 subunit in the membrane. This structure is based upon hydropathy analysis of 1 and by analogy to the ligand-gated ion-channel superfamily. The 2 subunit, with 74% amino acid identity to 1, is predicted to have a very similar structure (14). Both the N and C termini are extracellular; the Nterminal domain is predicted to contain the ligand binding site, a disulfide bridge (two Cs), and four putative asparagine-linked glycosylation sites (asterisks). The four hydrophobic domains (represented by rectangles [1][2][3][4] in the C-terminal domain are predicted to be inserted into the cell membrane. The bracketed regions indicate the extent of the 1 truncation mutants. The regions used to create the 1-specific antibodies hASH-1 and hASH-2 are shaded. 1 protein expressed in a mammalian cell line. An HEK 293 cell line transformed with an expression vector containing the human 1 cDNA (293-1) (22) was incubated with an antibody specific for subunits of several vertebrate and invertebrate species (18). The 1-expressing cells, and not the parental HEK 293 cells, were immunostained (Fig. 4A, left panel). Next, the polyclonal anti-1 antibodies were tested for their ability to detect 1 protein in the 293-1 cells. The hASH-1 and hASH-2 antisera specifically immunostained 293-1 cells but not the parental HEK 293 cells (Fig. 4A, middle and right panels). The signal was distributed mainly at the periphery of the cells; perinuclear staining observed in some cells may represent protein within the ER. Incubation with primary antibody alone or secondary antibody alone did not immunostain 293-1 cells. Furthermore, immunostaining was markedly reduced by preabsorbing hASH-1 with 10 g/ml peptide used to create the antisera or by preabsorbing hASH-2 with in vitro translated 1 protein (data not shown).
When tested against in vitro translated proteins, hASH-1 immunoprecipitated 1 (Fig. 4B, 2nd lane). The N terminus mutant N-1, which contains the hASH-1 epitope, was also precipitated (Fig. 4B, 6th lane). Proteins that did not contain the hASH-1 epitope, 2 and C-1, were not immunoprecipitated by hASH-1 (4th and 8th lanes). Furthermore, precipitation could be blocked by preabsorbing the hASH-1 antisera with 2 g/ml of the 13-residue peptide. As expected, hASH-2 antisera immunoprecipitated 1 and C-1 but not N-1 or 2 (Fig. 4B). Preimmune sera did not immunoprecipitate any of the proteins. We also confirmed that the epitope-tagged subunits, 1FLAG and 2HA, were immunoprecipitable by the appropriate monoclonal antibodies, M2-antiFLAG and 12CA5-antiHA, respectively. As noted above, in vitro translated 1 occasionally had minor electrophoretic forms (58, 56, and 48 kDa) in addition to the major form at 62 kDa. The two 1-specific polyclonal antibodies and the M2-antiFLAG monoclonal antibody precipitated all electrophoretic variants.
The N Terminus of 1 Contains Signals for Homooligomeric Assembly of 1 Receptors-To localize the sequences that specify 1 subunit homooligomeric assembly, 1 mutants were tested for interaction with full-length 1 subunits by immunoprecipitation of cotranslated proteins and by recordings of GABA-gated currents of co-injected Xenopus oocytes. The initial interaction between most proteins occurs in the lumen of the ER (25). Thus, the localization of the N terminus of in vitro translated 1 to the lumen of microsomal vesicles, a region analogous to the lumen of the endoplasmic reticulum (ER), suggested that the N terminus of 1 was involved in subunit assembly. We therefore investigated whether the N-terminal half of 1 (N-1) interacted with either 1 or 2 subunit. The ]cysteine, and the membranes were pelleted by high-speed centrifugation. Equal aliquots of the 1 and Aq-1 membrane pellets and equal aliquots of the 1 and Aq-1 supernatant fractions were electrophoresed in a 10% SDS-PAGE gel and then autoradiographed. The two panels were exposed for the same amount of time. Different conformations of the 1 subunit produce the heterogeneous migration pattern observed in the membrane fraction. B, autoradiogram demonstrating the susceptibility of 1FLAG, N-1 and C-1 to proteinase K when translated in vitro in the presence of microsomal vesicles (microsomes). Equal aliquots of each translation reaction were exposed to proteinase K in the absence (Ϫ) or presence (ϩ) of Triton X-100. Other regions of 1FLAG less than 14 kDa in size may have been protected but were not resolved due to limitations of our electrophoretic system. This study has been repeated four times with identical results .   FIG. 4. Characterization of 1-specific antibodies. A, wild-type HEK 293 (top row) and 1/HEK 293 (bottom row) cells were incubated with a 1:100 dilution of anti-rat 1 (18) or 1:500 dilutions of hASH-1 (middle) and hASH-2 (right) polyclonal anti-1 antibodies. The protein was detected by peroxidase substrate formation using the True-Blue staining kit. B, autoradiograms that demonstrate the specificity of hASH-1 and hASH-2 antisera. 1, 2, N-1, and C-1 were translated in vitro, detergent-solubilized, and immunoprecipitated with a 1:40 dilution of either antisera. Supernatant fractions containing nonprecipitated proteins are marked S, and the pellet fractions containing specifically immunoprecipitated proteins are marked P. Some residual protein remained in the supernatant fractions of 1, N-1, and C-1. The heterogeneous migration of the subunits in this experiment represents different conformers of the proteins (see "Results"). C-1 construct, containing the four transmembrane domains, was used as a control.
The epitope-tagged version of 1 (1FLAG) was used to facilitate its identification since the M2-antiFLAG monoclonal antibody immunoprecipitated only 1FLAG and did not crossreact with the 1 truncation mutants. 1FLAG was cotranslated with N-1 or C-1 and then immunoprecipitated with M2-antiFLAG. 1FLAG and mutant 1 proteins were identified on the basis of size; co-immunoprecipitation of the proteins indicated an interaction. Immunoprecipitation of cotranslated N-1 and 1FLAG using the M2-antiFLAG monoclonal antibody precipitated both proteins (Fig. 5, 2nd lane). However, when N-1 and 1FLAG were individually translated, solubilized, and then mixed, only 1FLAG was immunoprecipitated with the M2-antiFLAG antibody (Fig. 5, 4th lane). The dependence of this interaction upon cotranslation suggests that both proteins had to be present in the same microsomal vesicle for oligomerization to occur and excludes the possibility that the coprecipitation of N-1 and 1FLAG was the result of nonspecific protein aggregation. Furthermore, C-1 was not precipitated by M2-antiFLAG antibody when cotranslated or mixed with 1FLAG (Fig. 5, 6th and 8th lanes). Together, these results indicate that the precipitation of N-1 with 1 was due to a specific interaction.
The effect of N-1 and C-1 upon the ability of full-length 1 to form functional GABA receptors was investigated in Xenopus oocytes. Neither N-1 nor C-1 formed GABA-gated chloride channels when the respective cRNAs were injected singly into oocytes. However, co-injection of N-1 with full-length 1 resulted in a dose-dependent inhibition of 1 current. A 2:1 ratio of N-1 to 1 cRNA decreased the current by about 90% (n ϭ 5) maximal 1 current, and a 5:1 ratio eliminated the 1 current entirely (n ϭ 5) (Fig. 6). In contrast, co-injection of C-1 with full-length 1 did not inhibit 1 current at any ratio (n ϭ 5 for each) (Fig. 6). Based upon evidence that N-1 physically interacts with in vitro, we propose that the same phenomenon occurred in Xenopus oocytes, and the resulting N-1/1 oligomers were either non-functional or re-routed to a degradation pathway. In this scenario, 1 subunits that did not interact with N-1 were able to form functional homooligomeric GABA receptors. Increasing the ratio of N-1 to 1 RNA favored the formation of non-functional N-1/1 heterooligomers, thereby reducing the number of functional 1 homooligomeric receptors and decreasing the whole-cell currents elicited by GABA. Therefore, the N terminus of 1 is critical for 1 receptor assembly, based on its luminal location, its co-immunoprecipitation with 1, and its ability to interfere with 1 receptor formation.
1 and 2 Subunit Interaction in Vitro Is Mediated by Signals in the N Terminus-To determine whether 1 physically interacts with 2, 1 was cotranslated with 2 and then sequentially immunoprecipitated. Two immunoprecipitation steps were necessary because the in vitro translated 1 and 2 proteins could not easily be distinguished on the basis of size. Since 2-specific antibodies are not available, 2 with the HA epitope (2HA) was used. As noted above, 2HA is specifically immunoprecipitated by the anti-HA antibody 12CA5. In the first study, 1-specific antisera (hASH-2) was used in the first immunoprecipitation, and the anti-HA monoclonal antibody (12CA5) was used in the second immunoprecipitation (Fig. 7A). Since the hASH-2 antisera was highly specific for 1, we expected protein to be precipitated by 12CA5-antiHA only if 2 was associated with 1 in the first immunoprecipitation. Indeed, protein was not present following double immunoprecipitation of 1 translated alone (Fig. 7A, right panel, 1st lane), but proteins of the size appropriate for both subunits were present after double immunoprecipitation of 1 cotranslated with 2 (Fig. 7A, right panel, 2nd lane). In the second experiment, the HA-specific antibody (12CA5) was used in the first immunoprecipitation, and the 1-specific antibody (hASH-2) was used in the second immunoprecipitation (Fig. 7B). Again, proteins of the size expected for 1 and 2 subunits were present following the second immunoprecipitation step only in the cotranslation (Fig. 7B).
The next step was to assess whether the N termini of subunits were involved in the interaction between 1 and 2 subunits. The 1 mutants were tested for interaction with 2 containing the HA epitope (2HA). When 2 was cotranslated with N-1 and immunoprecipitated with the anti-HA antibody, both 2 and N-1 were precipitated (Fig. 8, 2nd lane). Furthermore, 2 and N-1 were both immunoprecipitated when antisera specific for N-1 (hASH-1) was used (Fig. 8, 4th lane). The same result was obtained when untagged 2 was cotranslated with N-1 and precipitated with antisera specific for 1 (hASH-1) (data not shown). This result indicated that the HA tag in 2 was not responsible for the interaction between N-1 and 2. Conversely, only 2 was precipitated with the anti-HA antibody when 2 was cotranslated with C-1 (Fig. 8, 6th lane). Together, these data indicated that N-1 interacts specifically with 2.

DISCUSSION
GABA and GABA A subunits have similar amino acid sequences and predicted structures (12)(13). However, there are distinct differences between receptors formed from subunits and those formed from GABA A subunits. The pharmacologic and biophysical properties of GABA receptors are more similar to retinal GABA C receptors than GABA A receptors (12,(15)(16). Furthermore, human 1 and 2 subunits both efficiently form homooligomeric GABA-gated receptors (13,15), whereas GABA A subunits must heterooligomerize for efficient receptor formation (26). Interestingly, the overlapping expression patterns of 1 and 2 in the rat retina (17) indicate that subunits may also heterooligomerize into a single receptor. This concept is supported by functional studies of rat subunits (19). Using in vitro translated protein and immunoprecipitation with subunit-specific antibodies, we provide evidence that the primary amino acid sequences of 1 and 2 contain signals for subunit interaction, independent of cell-specific factors. The following observations indicate that this interaction is not the result of nonspecific aggregation facilitated by the hydrophobic regions of each protein. First, the signals for interaction were mapped to a region of the protein shown to be in the appropriate location for oligomerization to occur (microso-mal vesicle lumen). Second, the interaction between 1 and N-1 was dependent upon cotranslation, consistent with assembly studies of other multimeric ion channels (27)(28). Third, the hydrophobic portion of 1 (C-1) did not co-immunoprecipitate with either 1 or 2 under any condition. We therefore conclude that the N-terminal region of 1 contains signals for homooligomeric and heterooligomeric interactions.
The in vitro translation system has been used to evaluate assembly of a variety of proteins that form multimeric structures (29 -30). In vitro studies of K ϩ channels have demonstrated the in vivo relevance of this approach for integral membrane proteins that function as ion channels (27,31). The in vitro system has not been used extensively to study assembly of ligand-gated ion channels. One early report failed to detect association of nACh receptor subunits translated in vitro that were known to interact in vivo (32). However, as the authors acknowledged, interaction may not have been observed due to low levels of each subunit expressed from total RNA (estimated to be 0.5% total translate) (32). In the present study, the use of in vitro synthesized RNA transcripts ensured high levels of protein expression. In addition, examination of the lysate prior to immunoprecipitation studies enabled us to confirm that each protein was synthesized and therefore available for interaction. The apparent molecular mass of the predominant in vitro translated subunit protein was larger than predicted from the amino acid sequences (10 kDa greater for 1 and 7 kDa for 2). Similar discrepancies between observed and predicted sizes have been noted for GABA A subunit proteins (33). The in vitro translated subunits also displayed heterogeneous migration. Each of the electrophoretic variants were immunoprecipitated by antibodies against the N terminus (hASH-1), an internal domain (hASH-2), and the C terminus (M2-anti-FLAG) of 1, indicating that each variant was an intact subunit protein. Therefore, the multiple bands may represent different conformers of the full-length proteins or proteolytic fragments generated after immunoprecipitation. The same phenomenon was observed for in vitro translated nACh receptor ␣ subunits (34).
Centrifugation studies indicated that in vitro translated 1 was associated with membranes. Interactions among multimeric integral membrane proteins generally take place in the lumen of the ER (25). In the in vitro system, the interior of microsomal vesicles provides an environment analogous to the  (2). The anti-1 polyclonal antibody hASH-2 was used in the first immunoprecipitation, and the anti-HA monoclonal 12CA5 was used in the second immunoprecipitation. B, in this experiment, the immunoprecipitations were performed in the reverse order. The anti-HA antibody 12CA5 was used in the first immunoprecipitation, and the 1-specific antibody hASH-1 was used in the second immunoprecipitation. The electrophoretic variants of 1 and 2 in these experiments represent multiple conformers of each protein (see "Results"). lumen of the ER. It was therefore important to determine the topology of the in vitro translated subunits relative to the microsomal vesicle membrane. Protease digestion of 1 and truncated 1 proteins synthesized in the presence of microsomal vesicles indicated that the N-terminal half of 1 was protected from degradation. Our results suggest that this region of 1 is inserted into the lumen of the vesicle and predict an extracellular location in vivo. In contrast, the C-terminal half of 1 was not protected from degradation. If the C-1 protein was inserted in the membrane in the same manner proposed for full-length 1 (Fig. 2), the majority of the peptide would have been exposed to proteinase K resulting in digestion of C-1. Although C-1 was designed with the native signal peptide sequence, it is possible that this mutant had an aberrant structure that prevented insertion into the membrane. Determination of the membrane topology of the C-terminal half of 1 will require additional studies. Placement of the N-terminal half of in vitro translated 1 in an extracellular orientation is consistent with the identification of an N-terminal histidine residue (His-156) critical for the inhibitory effect of extracellular Zn 2ϩ upon functional 1 receptors (35). Furthermore, two agonist binding domains have been localized to the N terminus of 1 between the cysteine loop and TM1 (36). Therefore, the orientation of GABA subunits in vitro appears to be similar to the topology of subunits within functional receptors and corresponds to the predicted topology of other subunits in the ligandgated ion channel superfamily (37)(38).
Although the association of 1 and 2 in vitro predicts that the subunits will also interact in vivo, we cannot determine in this study whether the 1/2 heterooligomeric receptors are functional. Indeed, oligomerization has been demonstrated among certain GABA A subunits that do not form functional receptors (39). However, a functional interaction of 1 and 2 was suggested using rat subunits, in which picrotoxin-insensitive receptors were formed when picrotoxin-sensitive 1 subunits were coexpressed with 2 subunits (19). The localization of signals for subunit interaction to the N-terminal region is consistent with studies of other members of the ligand-gated receptor superfamily, the nACh receptor, and glycine receptors (40 -42). It is possible that the C-terminal half of 1 contains signals for subunit interaction, but such signals could have been masked by aberrant conformation of the C-1 protein. Future studies will clarify whether there is only one assembly motif in 1 that enables both homooligomerization as well as heterooligomerization, as seen in the Shaker-like potassium channels (43). If the signals for homooligomeric receptor formation are the same as those for heterooligomeric receptors, and operate with the same efficiency, then the proportion of homooligomeric and heterooligomeric receptors should be related to the amount of 1 and 2 polypeptides present. Distinct signals may indicate that homooligomeric and heterooligomeric receptors are formed in mixed ratios, regardless of the relative abundance of each subunit. Identifying the precise amino acid residues for oligomerization of subunits would be intriguing as these signals may also be involved in determining the stoichiometry of 1 and 2 subunits in heterooligomeric GABA C receptors. This mechanism could be similar to the "assembly boxes" in the N-terminal luminal domain that determine the 3␣:2␤ subunit stoichiometry of glycine receptors (42). Unfortunately, the similar size of each subunit did not allow determination of the relative proportion of 1 and 2 in the immunoprecipitations of heterooligomeric complexes. It is also possible that other factors influence receptor assembly. These may include neuronal proteins, such as the transmembrane chaperone calnexin involved in nACh receptor ␣ subunit folding (44), or signals for later stages of assembly, as reported for the C-terminal regions of nACh receptors (45).
Isolation and subunit analysis of GABA C receptors from retina tissue will confirm whether the receptors exist as heterooligomers in vivo. The identification of a third subunit in the rat (46) and evidence that there are five subunits in the perch (47) suggest that a wide variety of GABA C receptor compositions probably occur. This study indicates that the assembly properties of cloned subunits that have sequence homology to (such as rat 3 (46)) or functional similarity to GABA C (such as the Drosophila Rdl subunit (48)) could be tested in vitro which will guide in vivo studies of GABA C receptor composition. Finally, manipulation of the cDNAs encoding the subunits, starting with the N terminus, can be used to identify specific amino acids involved in GABA C receptor assembly.