Assembly and cell surface expression of heteromeric and homomeric gamma-aminobutyric acid type A receptors.

The ability of differing subunit combinations of gamma-aminobutyric acid type A (GABAA) receptors produced from murine alpha 1, beta 2, and gamma 2L subunits to form functional cell surface receptors was analyzed in both A293 cells and Xenopus oocytes using a combination of molecular, electrophysiological, biochemical, and morphological approaches. The results revealed that GABAA receptor assembly occurred within the endoplasmic reticulum and involved the interaction with the chaperone molecules immunoglobulin heavy chain binding protein and calnexin. Despite all three subunits possessing the ability to oligomerize with each other, only alpha 1 beta 2 and alpha 1 beta 2 gamma 2L subunit combinations could produce functional surface expression in a process that was not dependent on N-linked glycosylation. Single subunits and the alpha 1 gamma 2L and beta 2 gamma 2L combinations were retained within the endoplasmic reticulum. These results suggest that receptor assembly occurs by defined pathways, which may serve to limit the diversity of GABAA receptors that exist on the surface of neurons.

The ability of differing subunit combinations of ␥-aminobutyric acid type A (GABA A ) receptors produced from murine ␣1, ␤2, and ␥2L subunits to form functional cell surface receptors was analyzed in both A293 cells and Xenopus oocytes using a combination of molecular, electrophysiological, biochemical, and morphological approaches. The results revealed that GABA A receptor assembly occurred within the endoplasmic reticulum and involved the interaction with the chaperone molecules immunoglobulin heavy chain binding protein and calnexin. Despite all three subunits possessing the ability to oligomerize with each other, only ␣1␤2 and ␣1␤2␥2L subunit combinations could produce functional surface expression in a process that was not dependent on Nlinked glycosylation. Single subunits and the ␣1␥2L and ␤2␥2L combinations were retained within the endoplasmic reticulum. These results suggest that receptor assembly occurs by defined pathways, which may serve to limit the diversity of GABA A receptors that exist on the surface of neurons.
␥-Aminobutyric acid type A (GABA A ) 1 receptors are believed to be the major sites of fast synaptic inhibition in the brain and are also the sites of action for many psychoactive drugs including the benzodiazepines and barbiturates (Olsen and Tobin, 1990). Molecular cloning has revealed a number of GABA A receptor subunits that can be divided by sequence homology into subunit classes with multiple members: ␣ (1-6), ␤ (1-4), ␥ (1-3), and ␦ (1), creating considerable potential for structural diversity. Additional diversity of receptor structure is generated by alternative splicing of some of these subunit mRNAs (Burt and Kamatchi, 1991). In situ hybridization and immunocytochemical methodologies suggest a large temporal and spatial diversity of receptor structure in the brain with many neuron types often expressing multiple receptor subunits (Wisden and Seeburg 1992;Fritschy et al., 1992). It is believed that GABA A receptors are pentameric in their final assembled plasma membrane form (Nayeem et al., 1994); however, the precise subunit composition and stoichiometry of a single population of native GABA A receptors remains unknown.
Accordingly, the expression of cDNA clones has been used to examine the minimal subunit composition required to produce functional GABA A receptors, determined by electrophysiological methodologies. Expression of unitary subunits has produced conflicting results; some subunits expressed alone appear to be able to produce GABA-gated ion channels (Blair et al., 1988;Pritchett et al., 1988) or channels that are sensitive to inhibition by picrotoxin, a GABA A receptor channel blocker (Sigel et al., 1989), whereas other studies demonstrate that some single subunits do not produce functional receptors (Sigel et al., 1990;Angelotti and Macdonald, 1993;Krishek et al., 1994). Expression of some binary subunit combinations have also produced conflicting data. For example, GABA-gated channels have been reported upon co-expression of either ␣1␥2 or ␤2␥2 subunits Draguhn et al., 1990). In contrast, the failure of co-expressed ␤1␥2 and ␣1␥2 subunits to produce functional GABA A receptors has also been reported (Sigel et al., 1990;Krishek et al., 1994;Angelotti and Macdonald, 1993). There is, however, general agreement that coexpression of ␣ and ␤ subunits is sufficient for the production of GABA-gated chloride currents, and the co-expression of ␣ and ␤ with either the ␥2 or ␥3 subunits produce GABA A receptors that are sensitive to modulation by benzodiazepines (Pritchett et al., 1988(Pritchett et al., , 1989Burt and Kamatchi, 1991) To further investigate these observations and attempt to seek an explanation for the failure of certain subunit combinations to produce functional GABA A receptors, we have examined the assembly and surface expression of homomeric and heteromeric GABA A receptors using biochemical, immunological, and electrophysiological methodologies. We have studied the assembly of GABA A receptors of varying subunit composition produced from ␣1, ␤2, and ␥2L subunits expressed in both Xenopus oocytes and transiently transfected A293 cells. From in situ hybridization and immunochemical analyses these subunits are co-localized in many adult brain regions and comprise up to 30% of all benzodiazepine-sensitive GABA A receptors in the adult brain (Benke et al., 1994;Fritschy et al., 1992).
In this study we demonstrate that GABA A receptor assembly occurs in the endoplasmic reticulum (ER), where interactions with the molecular chaperones immunoglobulin heavy chain binding protein (BiP) and calnexin were detected. Access to the cell surface was, however, limited to receptors composed of ␣1␤2 and ␣1␤2␥2L subunits. Single subunits and the binary combinations of ␣1␥2L and ␤2␥2L, although capable of oligomerization, were retained within the ER, presumably via interactions with BiP and calnexin. Receptor assembly and transport to the cell surface was not dependent on N-linked glycosylation, although its efficiency was enhanced. These results suggest that GABA A receptor assembly occurs by defined mechanisms, which may serve to regulate the diversity of GABA A receptors expressed on the surface of neurons.

EXPERIMENTAL PROCEDURES
Cell Culture and Transfection-Human embryonic kidney 293 cells (ATCC CRL 1573) were maintained in Dulbecco's modified Eagle's medium/Ham's F-12 medium (Life Technologies Ltd.) supplemented with 10% fetal bovine serum. Exponentially growing cells were seeded at 2 ϫ 10 6 cells/10-cm dish and transfected by calcium phosphate precipitation as described previously . 20 g of DNA was used per 10-cm plate of A293 cells using equimolar ratios of expression constructs. Cells were analyzed 12-18 h (immunofluorescence and immunoprecipitation) or up to 24 h (pharmacology and electrophysiology) after transfection. Nuclear injection of Xenopus oocytes with murine GABA A receptor subunit constructs was performed as described by Krishek et al. (1994).
Antibodies-The 9E10 antibody was obtained from 9E10 hybridoma cells (Evan et al., 1985) and used directly as supernatant without purification. Anti-FLAG M2 mouse monoclonal antibody was purchase from IBI Ltd. The anti-BiP antibody was purchased from Cambridge Bioscience (clone 5A5), and the anti-calnexin antibody (AF8) was a kind gift from Michael Brenner (Harvard Medical School) (Hochstenbach et al., 1992). Rabbit anti-horseradish peroxidase was purchased from DAKO (Denmark). The secondary antibodies, goat anti-rabbit rhodamine, and goat anti-mouse fluorescein isothiocyanate were purchased from Pierce.
Immunofluorescence-A293 cells on poly-L-lysine (10 g/ml)-coated coverslips were fixed in 3% paraformaldehyde (in phosphate-buffered saline) and washed twice in 50 mM NH 4 Cl (in phosphate-buffered saline). Subsequent washes and antibody dilutions were performed in phosphate-buffered saline containing 10% fetal bovine serum and 0.5% bovine serum albumin. When cells were permeabilized, 0.05% Triton X-100 was added to all solutions after fixation. 9E10 supernatant was diluted 1:5, and rabbit anti-horseradish peroxidase used at 0.5 g/ml. Both secondary antibodies were used at 1 g/ml. Cells were examined using a confocal microscope (MRC1000, Bio-Rad).
DAB Quenching-After fixation, cells were washed in 50 mM Tris⅐HCl (pH 7.5) and incubated in the same buffer containing 0.15% diaminobenzidine, 0.02% hydrogen peroxide, and 0.1 M imidazole for 1 h at 37°C in the dark. Cells were then processed for immunofluorescence.
Reimmunoprecipitation was performed on immunoprecipitates in reducing sample buffer. These were diluted 20-fold in lysis buffer containing 2% Triton X-100 prior to the second immunoprecipitation with either 9E10 supernatant, anti-calnexin, or anti-BiP as described above.
Electrophysiological Analysis-Whole cell recordings from transfected A293 cells and analysis of membrane currents from Xenopus oocytes were performed as described previously (Krishek et al., 1994). Currents from transfected cells were analyzed up to 24 h after transfection, wherase Xenopus oocytes were examined at 48 h after nuclear injection.

GABA A Receptor Subunit Combinations: Capability of Surface Membrane
Expression-To study the assembly of murine GABA A receptors, heterologous expression of GABA A receptor ␣1, ␤2, and ␥2L (a differentially spliced variant of ␥2 containing an 8-amino acid insert in the predicted major intracellular domain, which is lacking from the other splice variant, ␥2S) (Kofuji et al., 1991;Whiting et al., 1990) subunit cDNAs in both transiently transfected A293 cells (CRL 1573) and Xenopus oocytes was utilized.
To aid biochemical and morphological analyses, GABA A receptor subunits were tagged using the epitopes of 9E10 or FLAG. These epitopes were added to the ␣1, ␤2, and ␥2L subunits by site-directed mutagenesis between amino acids 4 and 5 of the mature subunits to create ␣1 9E10 , ␤2 9E10 , and ␥2L 9E10 . The functional effects of these additions were tested by electrophysiological analysis in A293 cells and Xenopus oocytes. Receptors incorporating 9E10-tagged subunits produced GABA-activated responses, which were indistinguishable from receptors comprised of wild-type subunits Smart et al., 1991;Angelotti and Macdonald, 1993) with regard to zinc insensitivity and benzodiazepine modulation ( Fig. 1). Similar results were obtained with subunits containing the FLAG epitope. 2 Therefore, the addition of these small epitopes to the extreme N-terminal domains of GABA A receptor subunits appears to be "functionally silent." The subcellular distribution of these tagged GABA A receptors expressed in A293 cells was determined by immunofluorescence of the FLAG epitope followed by confocal microscopy. Expression of ␣1 FLAG , ␤2 FLAG , or ␥2L FLAG alone revealed an ER-like staining pattern similar to that for horseradish peroxidase containing the C-terminal ER retention signal Lys-Asp-Glu-Leu (horseradish peroxidase-KDEL) ( Fig. 2A), which is almost exclusively localized to the ER (Connolly et al., 1994). To further investigate the localization of these GABA A receptor subunits, immunofluorescence was performed on cells in which the peroxidase reaction had been performed using diaminobenzidine as the substrate. This results in the production of a dark insoluble precipitate, which cross-links all proteins that colocalize with horseradish peroxidase (Courtoy et al., 1984;Ajioka and Kaplan, 1987). The reaction product should therefore prevent the detection of fluorescence (DAB quenching) if the candidate protein shares the same intracellular localization as horseradish peroxidase.
To confirm this, we examined the ability to detect endogenous markers in the presence of DAB reaction product produced from horseradish peroxidase-KDEL. After completion of the DAB reaction, under conditions identical to those shown in Fig. 2B, fluorescence was performed on horseradish peroxidase-KDEL-expressing cells using either antibodies against BiP and calnexin (localized to the ER; Fig. 3A) or fluorescentlabeled wheat germ agglutinin and lens culinaris lectins (markers for the Golgi apparatus; Fig. 3B). In cells expressing horseradish peroxidase-KDEL, no fluorescence was detected using antibodies to the ER markers, BiP, and calnexin, al-though fluorescence was detected in untransfected cells. In contrast, high levels of fluorescence for the Golgi markers, wheat germ agglutinin, and lens culinaris were observed in horseradish peroxidase-KDEL-transfected cells. These results confirm DAB quenching as a valid protocol for the detection of components within the same compartment.
Using this methodology we analyzed all three binary subunit combinations ␣1␤2, ␣1␥2L, and ␤2␥2L (Fig. 2B). Immunofluorescence was rare (approximately 0.01%) in ␣1␥2Lor ␤2␥2Ltransfected cells and was never coincident with cells exhibiting DAB staining, suggesting co-localization of these binary subunit combinations with horseradish peroxidase-KDEL. Similar results were also seen with single subunits utilizing DAB quenching. 2 These results are consistent with the restriction to an ER-like staining pattern observed for these combinations in the absence of horseradish peroxidase-KDEL ( Fig. 2B; ␣␥ and ␤␥) and their failure to reach the cell surface, as determined in nonpermeabilized cells. 2 Together these results demonstrate that single subunits and the ␣1␥2L and ␤2␥2L binary subunit combinations are restricted to the ER.
In contrast, the ␣1␤2 combination was capable of leaving the ER as determined by the co-existence of immunofluorescence and DAB staining within the same cells, consistent with the apparent surface staining (Fig. 2B, ␣␤) and confirmed in nonpermeabilized cells (Fig. 2C, ␣␤ and ␤␣). Only when all three subunits were co-expressed could the ␥2L subunit be detected on the cell surface (Fig. 2C, ␥␣␤). These results correlated with electrophysiological recordings made from A293 cells expressing all subunit combinations. Consistent with the results from the immunofluorescence experiments, only cells expressing either ␣1␤2 or ␣1␤2␥2L receptor subunits exhibited resolvable GABA-gated currents (Table I). Similar results were obtained from expression studies performed using Xenopus oocytes up to 72 h after injection. Thus it appears that only the combinations ␣1␤2 and ␣1␤2␥2L can access the cell surface. The accumulation of all nonsurface receptor combinations (␣, ␤, ␥, ␣␥, and ␤␥) in the ER strongly suggests that receptor assembly occurs in this intracellular compartment.
Immunoprecipitation of GABA A Receptor Subunits-Immunoprecipitation from detergent extracts of [ 35 S]methionine-labeled A293 cells revealed that the ␣1 9E10 subunit exists in three forms with molecular masses of 48, 50, and 52 kDa (  , 1992). The ␣1 subunit contains two consensus sequences (Asn-Xaa-Ser/Thr) for N-linked glycosylation at positions 10 and 110. Treatment of the ␣1-expressing cells with tunicamycin (a potent inhibitor of N-linked glycosylation) produced a single band coincident with the lowest form of ␣1 9E10 subunit of 48 kDa (Fig. 4, ␣, ϩ). Thus, the three ␣1 forms differ in their extent of N-linked glycosylation, and their sizes are consistent with the presence of zero, one, and two sites of N-linked glycosylation.
The ␤2 9E10 subunit exhibits apparent molecular masses of 53 and 56 kDa plus a weak band at 50 kDa (Fig. 4, ␤, Ϫ). Again, this subunit contains two consensus sequences for N-linked glycosylation at positions 8 and 80. Tunicamycin treatment produced a major band at 50 kDa (Fig. 4, ␤, ϩ), as predicted by cDNA cloning (Kamatchi et al., 1995). Thus, the ␤2 forms are also consistent with the presence of zero, one, and two sites of N-linked glycosylation. The ␥2L 9E10 subunit migrated as a broad band at approximately 42 kDa (Fig. 4, ␥, Ϫ) and, following tunicamycin treatment, as a broad band of around 30 kDa (Fig. 4, ␥, ϩ); this shift is consistent with the predicted presence of three consensus sites for N-linked glycosylation at positions 13, 90, and 208. The apparent molecular masses of both glycosylated and unglycosylated forms of ␥2L are much smaller than its 48-kDa predicted molecular mass derived from cDNA cloning (Pritchett et al., 1989). This may be due to proteolysis, a common observation for this subunit Haddingham et al., 1992) and may explain the appearance of a smear rather than discreet bands.
Oligomerization of GABA A Receptor Subunits-A possible explanation for the differential ability of GABA A subunit combinations to reach the cell surface as functional receptors may be found in their respective abilities to oligomerize with each other, as is often a prerequisite for exit from the ER (Hurtley and Helenius, 1989). The ability of GABA A receptor subunits to oligomerize was therefore analyzed by co-immunoprecipitation of binary combinations (␣1␥2L, ␤2␥2L, and ␣1␤2).
Consistent with the surface expression of the ␣1␤2 subunit combination, it was found that ␤2 co-immunoprecipitates both the 48-kDa unglycosylated and the 50-kDa partially glycosylated but not the 52-kDa fully glycosylated forms of the ␣1 subunit (Fig. 5A, lane 3). The reciprocal co-immunoprecipitation was not so clear and may be complicated by the co-migration of both the ␣1 and ␤2 bands at 50 kDa (Fig. 5A, lane 4). Despite being transport incompetent, the ␣1 and ␥2L subunits could also co-immunoprecipitate each other (Fig. 5B, lanes 3  and 4). As seen with the ␤2 subunit, the ␥2L also binds exclusively to the two lower forms of ␣1 (Fig. 5B, lane 3). Thus, neither ␤2 nor ␥2L show detectable oligomerization with the 52-kDa fully glycosylated form of the ␣1 subunit, even though it is the major species present. Similarly, the ␤2 and ␥2L subunits are also capable of oligomerizing despite their transport incompetence. In this case, ␥2L binds predominantly to the nonglycosylated 50-kDa form of ␤2 (Fig. 5C, lane 3). It also appears that both ␣1 and ␤2 bind preferentially to lower molecular mass ␥2L (Fig. 5, B, lane 4, and C, lane 4, respectively) compared to the major species present when immunoprecipitated directly (Fig. 5, B, lane 2, and C, lane 2). These preferences for subunit binding occur in the presence of the nonbind-

FIG. 3. Localization of ER and Golgi markers using DAB
quenching from horseradish peroxidase-KDEL. A293 cells were transfected with the horseradish peroxidase-KDEL cDNA expression construct. 12-18 h after transfection, the cells were fixed and exposed to 0.02% H 2 O 2 , 0.15% diaminobenzidine, and 0.1 M imidazole for 1 h at 37°C. The cells were then processed for immunohistochemical analysis using antisera directed against either ER or Golgi markers. A, ER markers: BiP (Clone 5A5) and calnexin (AF8). B, Golgi markers detected by rhodamine-labeled wheat germ agglutinin (WGA) and fluorescein isothiocyanate-labeled lens culinaris lectin (LC). The left-hand panels are phase images, whereas the right-hand panels represent fluorescent images recorded from the same field. The scale bar represents 10 microns.
ing forms, which exist within 5 min of [ 35 S]methionine labeling, a period in which oligomerization has occurred. 2 Similar patterns of subunit oligomerization were seen under nonreducing conditions, suggesting that they did not result from intermolecular disulfide bridges, as well as after solubilization in a range of other detergents including CHAPS (1%) and digitonin (1%). 2 To test the oligomerization of the triple subunit combination ␣1␤2␥2L immunoprecipitation utilizing ␥2L 9E10 subunit was performed. These results demonstrated association of the ␤2 and ␣1 subunits with the ␥2L subunit as expected. 2 Taken together with the immunolocalization studies, these experiments detailing the oligomerization of receptor subunits demonstrate that GABA A receptor assembly is primarily localized to the ER.
GABA A Receptor Assembly Is Independent of N-Linked Glycosylation-To examine the significance of N-linked glycosylation in receptor assembly, subunit oligomerization in the presence of tunicamycin was analyzed. Transfected cells expressing receptor cDNAs were treated with tunicamycin to prevent Nglycosylation and labeled with [ 35 S]methionine. Subunit oligomerization was then examined by co-immunoprecipitation using 9E10 antibodies directed against the ␣1 9E10 subunit. Precipitation of this subunit from cells co-expressing the ␤2 and ␥2L subunit (Fig. 6, lane 3) demonstrated that subunit oligomerization could occur in the absence of N-linked glycans, because the nonglycosylated ␣1 subunit (identified in Fig. 6, lane 1) co-precipitated the nonglycosylated ␤2 subunit (identified in Fig. 6, lane 2) as well as the nonglycosylated ␥2L subunit (30 kDa). Furthermore the oligomerization of the binary subunit complex ␣1␥2L was also apparently unaffected by tunicamycin treatment as determined by co-precipitation (Fig. 6,  lane 4).
Although the inhibition of N-linked glycosylation did not affect subunit oligomerization, N-linked glycosylation may be important for subsequent maturation. Therefore, we examined cell surface receptor expression in the presence of tunicamycin. This treatment caused other effects, most notably a reduction in cell number, reduced transfection efficiency, and changes in morphology. However, cell surface expression was not pre- F IG. 4. Immunoprecipitation and glycosylation of GABA A receptor ␣1, ␤2, and ␥2L subunits from transfected A293 cells. A293 cells expressing ␣1 9E10 (␣), ␤2 9E10 (␤), ␥2L 9E10 (␥), or untransfected (C) were [ 35 S]methionine-labeled in the absence (Ϫ) or presence (ϩ) of 5 g/ml tunicamycin (present 2 h prior to and during labeling). Receptor subunits were then immunoprecipitated using 9E10 antibody coupled to protein G-Sepharose. Immune complexes were then separated by SDS-polyacrylamide gel electrophoresis using 8% gels. The molecular masses of marker proteins (Bio-Rad) are indicated on the right. A co-immunoprecipitating band migrating at approximately 75 kDa (*) was consistently observed.
FIG. 5. Oligomerization of GABA A receptor subunits expressed in A293 cells. A293 cells expressing single or binary combinations of GABA A receptor subunits were labeled with [ 35 S]methionine, cells were lysed, and GABA A receptor subunits were purified using 9E10 antibody coupled to protein G-Sepharose. GABA A receptor subunits were then separated by SDS-polyacrylamide gel electrophoresis using 8% gels. Control cells show the presence of a contaminating band at approximately 40 kDa, which is sometimes observed regardless of the antibody used (arrowhead vented, as determined by immunofluorescence for receptors consisting of ␣1␤2␥2L subunits (Fig. 7A) but not for intracellular markers such as BiP, although the efficiency was significantly reduced from 95 Ϯ 7.5% in the absence to 37 Ϯ 15.52% in the presence of tunicamycin (Fig. 7B). This is consistent with the report of the ␣1 subunit (lacking N-linked glycosylation sites by site-directed mutagenesis) in the presence of ␤1 and ␥2, which produced functional GABA A receptors with pharmacological properties similar to those observed for the wild-type receptor but with reduced levels (Buller et al., 1994).
GABA A Receptor Subunits Interact with ER Chaperone Proteins-The correct folding of many proteins have been shown to involve the interaction of BiP (Pelham, 1989). In addition, another ER chaperone, calnexin, appears to be involved exclusively with glycoproteins (Ou et al., 1993). As seen earlier, immunoprecipitation of either ␣1, ␤2, or ␥2L GABA A receptor subunits routinely co-immunoprecipitated a protein of approximately 75 kDa (Fig. 4, asterisk), which may represent BiP. We therefore sought to determine if these two chaperone proteins participate in the assembly of GABA A receptors by retention of unassembled subunits within the ER, as observed for the ␣1, ␤2, ␥2L, ␣1␥2L, and ␤2␥2L combinations. Cells expressing the ␣1 9E10 subunit alone (which is transport-incompetent and therefore retained in the ER) were first immunoprecipitated with 9E10 antibody. This precipitate was then reprecipitated with either anti-BiP, anti-calnexin, or 9E10 antibodies, and bands migrating with expected molecular masses for both BiP and calnexin were evident (Fig. 8A). In addition, weak bands co-migrating with the nonprecipitated proteins (e.g. BiP and calnexin when performed via 9E10) were also present (weakly visible in Fig. 8A).
This interaction with BiP and calnexin is not unique to the ␣1 9E10 subunit, because the ␣1 9E10 , ␤2 9E10 , and ␥2L 9E10 subunits can also be co-immunoprecipitated by anti-calnexin antibody ( Fig. 8B) with no apparent preference for different forms. In all three cases, a protein migrating at the same molecular mass as BiP is also co-immunoprecipitated, suggesting some overlap in the binding abilities of BiP and calnexin. This overlap appears to occur with some endogenous proteins in A293 cells as evidenced by the untransfected control lane (Fig. 8B), in which anti-calnexin antibody co-immunoprecipitated a band coincident with BiP. When these apparent complexes were immunoprecipitated by 9E10, calnexin is only weakly observed, consistent with its low turnover rate and subsequently low [ 35 S]methionine incorporation (Hammond et al., 1994). Upon expression of all three, subunits BiP is still immunopre- FIG. 7. Effect of tunicamycin treatment on cell surface expression. A, immunofluorescence for the ␣1 9E10 subunit in the presence (ϩ) or the absence (Ϫ) of TX100 was performed on A293 cells transfected with ␣1 9E10 ␤2␥2L (Ϫtunicamycin). In a duplicate experiment, the cells were treated with tunicamycin constantly post-transfection, as well as 2 h pretransfection (ϩtunicamycin). B, quantitation of the frequency of surface (ϪTX100) fluorescence as a percentage of the frequency of total (ϩTX100) fluorescence. In the absence of tunicamycin 38.23 Ϯ 3.03% (n ϭ 519) were positive in the absence of TX100, with 40.3 Ϯ 7.84% (n ϭ 445) in the presence of TX100. In the presence of tunicamycin 6.05 Ϯ 2.53% (n ϭ 843) were positive in the absence of TX100, with 16.3 Ϯ 7.2% (n ϭ 336) in the presence of TX100 (where n ϭ total number of cells counted). ]methionine, immunoprecipitated using 9E10 antibody coupled to protein G-Sepharose, and resolved by SDS-polyacrylamide gel electrophoresis using 8% gels. Cells were treated with 5 g/l tunicamycin for 2 h prior to and during labeling with [ 35 S]methionine. C represents immune precipitations from control untreated cells. The presence of a contaminating band of 40 kDa, which is sometimes observed regardless of antibody used, is indicated (*). cipitated (Moss et al., 1995). However, whether this is due to interactions occurring between unitary, binary, or tertiary subunit complexes under these conditions is difficult to ascertain. DISCUSSION To date, 15 different GABA A receptor cDNAs have been isolated from a variety of vertebrates (Burt and Kamatchi, 1991). Many of these subunits exhibit differing patterns of both spatial and developmental expression in the CNS, with many neurons often expressing multiple numbers of receptor subunits (Wisden and Seeburg, 1992). A major challenge in trying to analyze the diversity of GABA A receptor structure in the brain is determining what processes control receptor assembly. Regulation of receptor assembly could occur at numerous stages, including subunit oligomerization or export to the cell surface. Unfortunately, to date there is little experimental data on these important questions. To address this, we have examined the assembly of GABA A receptors of differing subunit compositions constructed from those of ␣1, ␤2 and ␥2L subunits expressed in both A293 cells and Xenopus oocytes using immunological, biochemical, and electrophysiological methodologies. These subunits are thought to co-localize in many brain regions and comprise up to 30% of all benzodiazepine-sensitive GABA A receptors in the adult brain (Fritschy et al., 1992;Benke et al., 1994). For immunological and biochemical analyses the 9E10 (Evan et al., 1985) or the FLAG epitopes were added between amino acids 4 and 5 of each of these subunits. As demonstrated by electrophysiological analyses, these additions appeared to be functionally silent.
Using immunolocalization, epitope tagged ␣1, ␤2, or ␥2L subunits expressed alone are incapable of leaving the ER, as are the binary subunit combinations of ␣1␥2L and ␤2␥2L, demonstrated by co-localization with horseradish peroxidase-KDEL (Connolly et al., 1994). The only combinations of receptors produced from these subunits that are capable of exiting the ER and accessing the cell surface are ␣1␤2 and ␣1␤2␥2L. In agreement with this only the latter two subunit combinations exhibited resolvable GABA-gated currents when expressed in either A293 cells or Xenopus oocytes.
Previous studies have produced conflicting data on the expression of single subunits and the combinations of ␣1␥2L and ␤2␥2L, as determined by electrophysiological analysis. Blair et al. (1988) reported the production of GABA-gated channels on the expression of single ␣1 or ␤1 subunits in Xenopus oocytes, and Sigel et al. (1989) reported finding chloride currents that could be blocked by picrotoxin on the expression of the rat ␤1 subunit. Moreover, Verdoorn et al. (1990) and Draguhn et al. (1990) found robust receptor expression from ␣1␥2 subunits and smaller currents from ␤2␥2L subunits in A293 cells. In contrast to these results Angelotti and Macdonald (1993) and Krishek et al. (1994) found no GABA-gated currents when expressing ␣1␥2 or ␤1␥2 subunits in L929 or A293 cells. These discrepancies could be due to differences in the expression systems used or differences in the type of ␤ subunit used in some of these experiments (␤1 versus ␤2). Expression of murine ␤1 subunits alone in both A293 cells and Xenopus oocytes can produce low levels of surface expression, 3 in common with the observations of Blair et al. (1988) and Sigel et al. (1989). Therefore some of this variability in expression may be subunitspecific. A second reason for such discrepancies may result from over-expression. It is possible that the longer expression periods used in many previous studies (48 -72 h in A293 cells and 2-6 days in oocytes (Blair et al., 1988;Verdoorn et al., 1990) compared with 12-24 h and 2-3 days for 293 cells and Xenopus oocytes, respectively, used in this study) may have resulted in the escape of normally transport-incompetent receptor complexes through saturation of the ER retention system, resulting in low levels of surface expression.
In spite of the apparent inability of the ␣1␥2L and ␤2␥2L subunit combinations to leave the ER, they were still capable of rapid oligomerization. In fact, all oligomerization events were not dependent on N-linked glycosylation and could occur with unglycosylated subunits. Furthermore, N-linked glycosylation was not required for the transport of receptor heteroligomers to the cell surface. Interestingly, the ␤2 subunit appeared to oligomerize more efficiently to lower molecular mass forms of the ␣1 (and possibly the ␥2L subunits), and the ␥2L subunit appeared to oligomerize more efficiently to lower molecular mass forms of both the ␣1 and the ␤2 subunits. The significance of these interactions is uncertain; they may represent preferred patterns of subunit oligomerization if these incompletely processed forms are represented in cell surface receptors in the presence of glycosylated forms. Whether these interactions are important in controlling the final assemblies of GABA A receptors produced warrants further investigation. GABA A receptor subunits interacted with at least two ER chaperone proteins, BiP and calnexin, whose function is to retain misfolded and unassembled proteins. Interactions with BiP are thought to result from the exposure of hydrophobic domains in incorrectly folded proteins (Pelham, 1989), whereas calnexin appears to show specificity for glycoproteins containing partially "glucose-trimmed" carbohydrate side chains (Hammond et al., 1994). Calnexin is thought to hold glycoproteins in the ER until folding/assembly is complete, thus preventing their aggregation and/or premature exit from the ER (Hochstenbach et al., 1992;Ou et al., 1993;Hammond et al., 1994). Presumably single GABA A receptor subunits and the binary combinations of ␣1␥2L and ␤2␥2L are retained in the ER via the interaction with chaperone proteins such as calnexin and BiP.
The demonstration of the intracellular localization of GABA A receptor heteroligomers has important consequences for our understanding of GABA A receptor structure. To date, the method of choice for examining the complexity of GABA A receptor structure in the brain has been the use of subunitspecific antisera to immunoprecipitate receptor complexes from solubilized brain tissue (e.g. Duggan and Stephenson, 1990;Mertens et al., 1993;Mckernan et al., 1991;Mossier, 1994). These procedures are complex and often yield contradictory results (cf. Pollard et al. (1993), Quirk et al. (1994), and Khan et al. (1994)). It is possible that some of the interactions seen following immunoprecipitation in these experiments and thereby proposed to represent cell surface receptor subunit combinations may in fact represent ER-retained forms of GABA A receptors. Such complexes as demonstrated in our study may have differing subunit combinations from GABA A receptors expressed on the surface of neurons.
Finally, the inability of GABA A receptor heteroligomers consisting of ␣1␥2L and ␤2␥2L subunits to reach the cell surface suggests that assembly of GABA A receptors might share similar mechanisms to those employed by muscle acetylcholine receptors. Assembly of these receptors is believed to utilize intracellular dimer or trimer intermediates (Green and Miller, 1995). Further studies will elucidate whether the ER-retained subunit heteroligomers described in our work are important intermediates in the production of fully assembled pentameric GABA A receptors. This may be of significance because the consensus of opinion (Burt and Kamatchi, 1991;Pritchett et al., 1988Pritchett et al., , 1989, derived from a combination of molecular, pharmacological, and electrophysiological methodologies, suggests that in vivo benzodiazepine-responsive GABA A receptors are composed of ␣␤␥ in unknown stoichiometries. The relevance of these ␣1␥2L and ␤2␥2L intracellular subunit complexes to the final functional GABA A receptors produced is currently under investigation.