GABAA Receptor Composition Is Determined by Distinct Assembly Signals within α and β Subunits*

Key to understanding how receptor diversity is achieved and controlled is the identification of selective assembly signals capable of distinguishing between other subunit partners. We have identified that the β1–3 subunits exhibit distinct assembly capabilities with the γ2L subunit. Similarly, analysis of an assembly box in α1-(57–68) has revealed an absolute requirement for this region in the assembly of αβ receptors. Furthermore, a selective requirement for a single amino acid (Arg-66), previously shown to be essential for the formation of the low affinity GABA binding site, is observed. This residue is critical for the assembly of α1β2 but not α1β1 or α1β3 receptors. We have confirmed the ability of the previously identified GKER signal in β3 to direct the assembly of βγ receptors. The GKER signal is also involved in driving assembly with the α1 subunit, conferring the ability to assemble with α1R66A on the β2 subunit. Although this signal is sufficient to permit the formation of β2γ2 receptors, it is not necessary for β3γ2 receptor formation, suggesting the existence of alternative assembly signals. These findings support the belief that GABAA receptor assembly occurs via defined pathways to limit the receptor diversity.

However, relatively few functionally distinct receptor compositions are thought to exist in vivo (1). It is possible that multiple receptor types may exist that are functionally equivalent. Their distinct subunit compositions may provide subtle functions such as modulation by endogenous ligands such as neurosteroids (2) or second messenger systems (3,4), subcellular localization (5), or long term differences in the regulation of receptor surface expression (6,7). Despite these caveats, GABA A receptor heterogeneity occurs via defined pathways to limit receptor diversity (3,6). Possible mechanisms include brain region-specific (8) and temporal expression (9). However, many neuron types often express multiple receptor subunit mRNAs simultaneously (8), suggesting that subcellular mechanisms for differential receptor assembly may also exist. Potential processes could include the discrete sites of subcellular receptor assembly (10,11) and/or the presence of assembly signals capable of differential interaction with other subunits.
In support of the existence of differential assembly signals, GABA A receptor assembly appears to be strictly controlled, producing receptors with a fixed stoichiometry of 2␣, 2␤, and 1␥ (12)(13)(14)(15)(16)(17). Furthermore, GABA A receptor assembly signals have been identified in the ␣1 (18,19), ␤2/3 (19,20), and ␥3 (21) subunits. Although the regions identified in these studies may exhibit subunit class-specific interactions, to date no studies have investigated the ability of GABA A receptors to discriminate between subunits of the same class.
Consistent with the location of these assembly signals to intersubunit contact points, the ␣/␥ signals (18,19,21) are located proximal to the GABA and benzodiazepine binding sites (22,23) formed at subunit interfaces between the ␣-␤ and ␣␥ subunits, respectively, and also the ␤2 high affinity GABA site (24). Similarly, the homologous region in 1 is an important component of the GABA binding domain (25).
Given the high degree of homology between the ␣ and ␥ subunits in this region combined with their differential ability to assemble with ␤ subunits (␣1 with ␤1-3, ␥2 with ␤3, possibly ␤1, but not ␤2) (20), we sought to investigate the role of these sequences in the differential assembly of ␣/␥ subunits with ␤ subunits. Using site-directed mutagenesis, we have determined that the conversion of a single amino acid in ␣1 to that of ␥2 (R66A) is sufficient to alter the assembly profile of the ␣1 subunit to that of the ␥2, as determined by immunofluorescence and cell surface ELISA. These results demonstrate that the identity of a single amino acid may be critical in determining assembly with particular receptor subunits and may identify a basis for subunit-specific GABA A receptor assembly. Furthermore, we present evidence for the existence of alternative assembly signals, which may permit the formation of diverse receptor types, dependent upon subunit availability.
DNA Constructions-Murine ␣1, ␤1-3, and ␥2L subunit cDNAs containing either the Myc or FLAG epitope tags (between amino acids 4 and 5 of the mature polypeptide) have been described previously and shown to be functionally silent with respect to receptor pharmacology and physiology (5,18,20,26). The mutant constructs ␣1S, ␣1(1), ␤2 GKER , and ␤3 DNTK were generated by site-directed mutagenesis as reported previously (18,20). The remaining mutant ␣1/␥2 constructs were generated by site-directed mutagenesis using the oligonucleotides: The fidelity of the final expression constructs was verified by DNA sequencing.
Antibodies-The 9E10 antibody was obtained from 9E10 hybridoma cells (27) and used directly as supernatant without purification. The secondary antibodies, goat anti-mouse Alexa Fluor 568 and goat antimouse Alexa Fluor 488, were purchased from Molecular Probes and goat anti-mouse horseradish peroxide was from Amersham Biosciences.
Immunofluorescence-COS7 cells were fixed in 3% paraformaldehyde (in PBS), washed twice in 50 mM NH 4 Cl (in PBS), and blocked (10% fetal bovine serum, 0.5% bovine serum albumin in PBS) for 30 min. Subsequent washes and antibody dilutions were performed in PBS containing 10% fetal bovine serum and 0.5% bovine serum albumin. After surface immunofluorescence cells were permeabilized by the addition of 0.5% Triton X-100 (10 min), and the immunofluorescence protocol was repeated from the NH 4 Cl step. Cells were examined using a confocal microscope (Zeiss LSM510).
Quantification of Cell Surface Expression-COS7 cells were plated onto 3-cm wells of a 6-well dish. Two transfections were pooled and used to seed 6 wells ("surface" and "total" in triplicate). Cells were fixed in 3% paraformaldehyde (in PBS). Cell surface detection was performed in the absence of detergent, and total expression levels were determined following Triton X-100 (0.5%, 15 min) treatment. Cells were washed twice in 50 mM NH 4 Cl (in PBS) and blocked (5% fat-free powdered milk (Marvel), 10% fetal bovine serum , 0.5% bovine serum albumin in PBS) for 1 h. Subsequent washes were performed in block. Receptor expression was determined using a horseradish peroxide-conjugated secondary antibody and assayed using 3,3Ј,5,5Ј-tetramethylbenzidine (Sigma) as the substrate, with detection at 450 nm after 30 min, after the addition of 0.5 M H 2 SO 4 . The reaction rate was determined to remain linear for up to 1 h (results not shown).

RESULTS
Previous studies have revealed that the subunits of the most commonly expressed GABA A receptor (␣1␤2␥2) cannot reach the cell surface when expressed alone nor when ␣1␥2L or ␤2␥2L combinations are co-expressed. Only when ␣1␤2 or ␣1␤2␥2L subunits are co-expressed are functional cell surface receptors produced (5). Non-productive subunit monomers or ␣1␥2L or ␤2␥2L dimers are retained in the endoplasmic reticulum (ER) by interactions with BiP or calnexin.
Receptor Homology within the Region of a Putative Assembly Signal-Recently, a putative assembly signal within the ␣1 and ␣6 subunits (see ␣6S, Fig. 1A) was identified as being required for assembly and cell surface expression with ␤3 (18). This region exhibits homology between all the GABA A receptor subunits (Fig. 1). The highest homology exists within subunit classes (␣ values ϭ 63.2%, ␤ values ϭ 78.9%, and ␥ values ϭ 71.1%), whereas homology between the subunit classes (including ␦ and ⑀) is 26.3%. Within this region, both the glutamine (Gln-67) and the tryptophan (Trp-69) have been shown to be essential for assembly (18,28). However, the glutamine is completely conserved among all GABA A receptors, and the tryptophan is completely conserved among all members of the ligand-gated ion channel superfamily, indicating a general role in subunit architecture or folding.
To implicate residues that may be involved in subunit classspecific assembly of GABA A receptors, we examined the sequences in this region to identify residues completely conserved within a subunit class but absent from the other subunit classes. Surprisingly, despite the high degree of homology evident, only one residue (Arg-66 in ␣1) fits this criterion. At this position, an arginine (␣s), glutamine (␤s), or alanine (␥s), is always present (histidine in ␦, serine in ⑀). Interestingly, this residue in ␣1 has been shown to be critical for GABA binding (22).
To assess the role of this region in GABA A receptor assembly we investigated ␣1/␥2 subunit chimeras and mutants for their ability to assemble with the ␤ subunits. The constructs used are ␣1 myc , ␣1S myc (lacking residues 57-68), ␣1 (␥2)myc (residues 57-68 replaced with the homologous residues from ␥2), ␣1 (R-A)myc (arginine at position 66 mutated to alanine), ␥2L myc , ␥2L (␣1)myc (residues replaced with homologous residues from ␣1), and ␥2L (A-R)myc . Subunit detection was performed using 9E10 antibodies that recognize the Myc epitope present in all the ␣1/␥2L subunits. All ␤ subunits used were epitope-tagged with the FLAG epitope. When expressed alone, all the above ␣1/␥2L subunits do not reach the cell surface but are retained within the ER (5) (results not shown). Thus, we measure the ability of the ␣1/␥2L subunits to be rescued from the ER and expressed on the surface as heteromeric receptors with ␤ subunits.
Cell Surface Expression of ␣1/␥2 Subunits with ␤2-COS7 cells were transfected with ␣/␥ myc ␤2 FLAG , and immunofluorescence was performed using 9E10 antibodies. As shown previously (5,26), ␣1 and ␤2 subunits are able to assemble and form heteromeric receptors on the cell surface ( Fig. 2A). Consistent with the presence of an assembly signal existing between residues 57 and 68 of the ␣1 subunit, ␣1S (lacking this region) cannot reach the cell surface despite the presence of the ␤2 subunit. The ␣1S is retained within the ER, as evidenced by the classic reticular staining pattern observed ( Fig. 2A). To address the possibility that this subunit might misfold, we examined an ␣1 (1) chimera in which residues 57-68 in the ␣1 were replaced with the homologous residues from GABA C receptor 1 subunit (18). Given the expected structural similarities between 1 and GABA A receptor subunits combined with the recent observation that this region in 1 also contributes to the GABA binding site (25), this chimera would be less likely to misfold. In keeping with the inability of the 1 subunit to assemble with the GABA A receptor ␤ subunits (29,30), coexpression of ␣1 (1) with ␤2 did not lead to cell surface expression (results not shown).
Receptors composed of ␤2␥2 do not access the cell surface but are retained in the ER (5) (Fig. 2). Therefore, we assessed the ability of ␣1 (␥2) to assemble with ␤2. Consistent with a role for ␣1 residues 57-68 in assembly with ␤2, cell surface expression of ␣1 (␥2) is abolished.
Given the critical role of arginine at position 66 of ␣1 in GABA binding (22) and the complete conservation of this position within all subunit classes (Fig. 1), we mutated this arginine (in ␣1) to the corresponding residue (alanine) in ␥2 creating ␣1 (R-A) . This mutation completely abolished the ability of ␣1 to assemble with ␤2.
To determine whether this putative ␣1 assembly signal might be transplanted into the ␥2 subunit, we generated a ␥2L (␣1) construct containing the residues 57-68 from ␣1 in place of the homologous ␥2 sequence. When expressed with ␤2, ␥2L (␣1) is still incapable of assembling with ␤2 ( Fig. 2A). Thus, although the ␣1 region 57-68 is essential for the assembly of ␣1 and ␤2 and critically dependent upon an arginine at position 66, this region is insufficient to direct this assembly event. This is corroborated by the ␥2L (A-R) construct, which cannot assemble with ␤2.
Quantification of these observations was performed by whole cell ELISA to determine receptor cell surface expression (no detergent) versus total (detergent) receptor expression. The values presented here do not reflect a true percentage of surface-expressed receptors for two reasons; first, the ␣1␤2 values have been normalized to 100% and, second, surface values greater than 100% have been detected for ␤3 homomers (results not shown), suggesting a possible inhibitory influence of prior detergent treatment on this assay. In these experiments (Fig. 2B, Table I) it can be seen clearly that the assembly of ␤2 with ␣1 requires residues 57-68 of ␣1 (␣1S ϭ 0.9 Ϯ 1.3%), most notably an arginine at position 66 (␣1 (R-A) ϭ 5.0 Ϯ 4.3%). In agreement with the results observed by immunofluorescence, although this region is critically involved in assembly with ␤2, transplantation of this signal to ␥2 does not confer the ability to assemble with ␤2 (␣1 (␥2) ϭ 3.7 Ϯ 3.1%).
Cell Surface Expression of ␣1/␥2 Subunits with ␤1-COS7 cells were transfected with ␣/␥ myc ␤1 FLAG , and immunofluorescence was performed using 9E10 antibodies. As shown previously (26) ␣1 and ␤1 subunits are able to assemble and form heteromeric receptors on the cell surface (Fig. 3). Consistent with the presence of an assembly signal existing between residues 57-68 of the ␣1 subunit, ␣1S cannot reach the cell surface despite the presence of the ␤1 subunit (2.7 Ϯ 3.1%). Similar results were observed for the ␣1 (1) (results not shown).
In this study, we could detect only very low levels of ␥2L␤1 surface receptors by immunofluorescence (Fig. 3A) that could not be resolved from background (mock-transfected cells) when analyzed quantitatively (10.2 Ϯ 11.8%, Fig. 3B). To determine whether the robust surface expression observed upon ␣1␤1 expression results from the use of the ␣1 57-68 assembly signal, we assessed the ability of ␣1 (␥2) to assemble with ␤1. In contrast to the results observed with the ␤2 subunit, cell surface expression of ␣1 (␥2) is not significantly different (108.6 Ϯ 23.4%) from that observed for wild-type ␣1 (100 Ϯ 7.6%) but higher than that observed for the wild-type ␥2L (10.2 Ϯ 11.8%). These findings suggest that, in contrast to the ␤2 subunit, the ␤1 subunit does not exhibit the same requirements for this region of the ␣1 for the assembly of ␣1␤1 heteromeric receptors.  This possibility is corroborated by the efficient assembly of ␤1 with ␣1 (R-A) (83.9 Ϯ 16.1%, Fig. 3). However, this region is still essential for ␣1␤1 receptor assembly, as evidenced by the inability of ␣1S (Fig. 3) and ␣1 (1) (results not shown) to assemble with ␤1. Both of the ␥2 mutants (␥2 (␣1) and ␥2 (A-R) ) were indistinguishable from the wild-type ␥2L with respect to cell surface expression with ␤1, i.e. weak surface immunofluorescence and non-detectable surface levels by ELISA (Fig. 3).
Cell Surface Expression of ␣1/␥2 Subunits with ␤3-COS7 cells were transfected with ␣/␥ myc ␤3 FLAG , and immunofluorescence was performed using 9E10 antibodies. As shown previously (26) ␣1 and ␤3 subunits are able to assemble and form heteromeric receptors on the cell surface (Fig. 3). Consistent with the presence of an assembly signal existing between residues 57 and 68 of the ␣1 subunit, ␣1S cannot reach the cell surface (3.5 Ϯ 3.6%) despite the presence of the ␤3 subunit. Similar results were observed for the ␣1 (1) (results not shown).
Interestingly, the ␤3 subunit (20,31) and to a lesser extent the ␤1 subunit (32,33) can form functional homomeric ion channels (not GABA-gated), whereas the ␤2 subunit is incapable of exiting the ER (5). A four-amino acid signal (GKER) that controls ␤3 homooligomerization has been identified (20). When transferred to ␤2 (␤2 GKER ), this subunit is capable of cell surface expression as functional ion channels. The reciprocal construct (␤3 DNTK ) can no longer assemble into homomeric receptors but is retained in the ER (20). To determine whether the assembly signal identified in the ␣1 may also recognize the homomeric assembly signal present in the ␤3 subunit, we analyzed the assembly of ␣1/␥2 subunits with the ␤2 GKER and ␤3 DNTK subunits.
To determine the ability of the ␣1/␥2 polypeptides to oligomerize with ␤2, cDNAs were cotransfected into COS7 cells, [ 35 S]methionine-labeled, and immunoprecipitated via the Myc epitope tag on the ␣1/␥2 subunits. Only the extracellular domain of ␤2 was used to eliminate any contribution from other subunit interactions (34) and events at the cell surface such as receptor turnover (6). Bands corresponding to the ␣1/␥2 and ␤2 extracellular fragments were excised and quantified using a scintillation counter. The ratio of ␤2 co-immunoprecipitated with the ␣1/␥2 subunit was normalized to that of wild-type ␣1 (39%) such that the ratio (␤2:␣1) of ␤2 co-immunoprecipitated by ␣1 represents 100%. No significant reduction in binding was evident for any of the ␣1/␥2 subunits (Fig. 7). This is not surprising, because each subunit must possess at least two interfaces (and presumably assembly signals) with other subunits (Fig. 8).

DISCUSSION
To date 16 different GABA A receptor cDNAs have been isolated from a variety of vertebrates (4). 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 (8,35). A major challenge in trying to analyze the diversity of GABA A receptor structure in the brain is determining what processes control receptor assembly. The determination of receptor composition may arise from temporal and spatial regulation, subcellular subunit segregation, and/or differential assembly/stability. Al-though there is extensive evidence for a role of temporal (9) and spatial (8,35) regulation of subunit expression in determining receptor composition, these mechanisms cannot explain how receptor diversity is limited in neurons co-expressing multiple GABA A receptor subunits simultaneously (8,35). To date, there is no supporting evidence for discrete subcellular assembly sites for GABA A receptors, although emerging evidence for the localized translation of proteins at synapses (11) may be relevant.
The potential for hierarchical assembly signals is supported by the strict control of receptor stoichiometry (12)(13)(14)(15)(16)(17) and the observations that ␣1␤3 (versus ␤3 homomers) and ␣1␤2␥2 (versus ␣1␤2) receptors form to the exclusion of the other possible combinations (3,31,36). In vivo evidence for hierarchical receptor assembly has been provided by the analysis of ␣6 knockout mice, which determined that the ␦ subunit was concomitantly "knocked-out" by a partial ␣6 polypeptide product that remains able to associate with the ␦ subunit but cannot produce functional receptors (37).
The discovery of putative GABA A receptor assembly signals began with the identification of a natural splice variant of ␣6 that lacked 10 amino acids in the N-terminal extracellular domain (38). This splice variant (termed ␣6Short) was determined to be incapable of assembling into functional receptors (18,38). The high degree of homology between all GABA A receptor subunits within this region suggests a common role in receptor assembly, a possibility vindicated for the ␣1 (18) (but see Ref. 19) and ␥3 (21). Putative assembly signals have also been discovered adjacent to this region in ␣1 (for binding to ␥2) (39) and ␥2 (for binding to ␣1 and ␤3) (19). Furthermore, two invariant tryptophans within this region have been shown to be essential for GABA A receptor assembly and benzodiazepine, but not GABA, binding (28). Indeed these tryptophans are completely conserved in all members of the ligand-gated ion channel superfamily and may provide some common structural feature necessary for receptor assembly (28).
Given the similarity between the homologous regions between ␣ (MEYTIDVFFRQSW) and ␥ (MEYTIDIFFAQTW) subunits, we examined the possibility that the sequence differences between these subunit classes might be responsible for the differential ability of the ␥2 subunit to assemble with ␤1 and ␤3 (20, 33), but not ␤2 (5,20), whereas the ␣1 can assemble with all three ␤ subunits (26).
Consistent with previous findings, ␣1S and ␣1 (1) (18) are not able to assemble with ␤1-3. In addition, ␥2L cannot assemble with ␤2 (20) and only weakly with ␤1 (Table I) (3). In keeping with the possibility that ␣1-(57-68) constitutes an assembly signal determining oligomerization with ␤ subunits (18), ␣1 (␥2) is also unable to assemble with ␤2. In fact, a single site (Arg-66) was found to be critical for the assembly of ␣1␤2 receptors. However, the failure of ␥2L (␣1) to assemble with ␤2 suggests that this putative assembly signal is necessary but not sufficient to direct this oligomerization step.
Interestingly, a striking overlap between the ␣A site and the GABA binding site (loop D) exists for ␣1␤2 receptors (22). Boileau et al. (22) showed that residues Phe-64, Arg-66, and Ser-68 were found to be part of (or close to) the GABA binding site, with Phe-64 and Arg-66 critical for modulation by GABA. In keeping with other studies (18,28), these authors found that Gln-67 and Trp-69 were essential for the production of functional receptors. Moreover, this region is not only important in the ␣1 subunit contribution to GABA binding, but the high affinity GABA binding site has been shown to be produced by the homologous region in ␤2 (␤2A interface to ␣1) (24). In addition, the benzodiazepine binding site on the ␥2 subunit resides within the same homologous region, with A79 (homologous position to Arg-66 in ␣1) lining the benzodiazepine binding pocket (23). A similar overlap between assembly signal and benzodiazepine binding is observed for the opposing interface on the ␣1 subunit, with residues 74 -123 providing the binding site (40) and residues 81-100 (MTVLRLNNLMASKIWTP-DTFF, see Fig. 1) involved in oligomerization with ␥2 (39).
In contrast to the tight correlation between receptor assembly and the formation of the GABA/benzodiazepine binding site in ␣1␤2␥2 receptors, little overlap exists for the opposing side of this interface, the ␤B site. This side of the interface has been determined to be constructed from 3 distinct regions; loop A between residues 93 and 101, loop B between residues 157 and 160, and loop C between residues 202 and 209 to generate the GABA binding domain on the ␤2 subunit (41,42). The residues required for assembly (GDKAVTGVER in ␤3) fall between loops B and C.
In contrast to the findings for the ␤2 subunit, although ␤1, ␤3, ␤2 GKER , and ␤3 DNTK all require the presence of residues 57-68 from either ␣1 or ␥2, they do not exhibit the same dependence upon this sequence as ␤2. The most likely explanation for these results is a general requirement for this region to either induce correct folding or subunit architecture but not to provide an assembly signal. This would be consistent with the role of these sequences in the formation of GABA and benzodiazepine binding sites. Alternatively, assembly signals for ␤1 and ␤3 may reside within this region, as for assembly with ␤2, but be interchangeable between ␣1 and ␥2. However, this seems unlikely given that ␤1 assembles only poorly with ␥2L, yet robustly with ␣1(␥2). Thus, it seems more likely that ␤1 and ␤3 utilize unique assembly signals in ␣1 and ␥2. This is consistent with previous findings (19) in which assembly signals just downstream from the ␣1 57-68 region have been identified for ␥2 binding to ␤3 (Figs. 1 and 8). However, unlike the ␥2, ␥3 does utilize the homologous assembly signal (MEY-QIDIFFAQTW) used by ␣1 (to assemble with ␤2) to assemble with ␤3 (21).
A more complex scenario is also possible. Perhaps ␤1 and ␤3, but not ␤2, may possess and utilize alternative assembly signals dependent on availability. In this way GABA A receptors may form their preferred receptor compositions depending on subunit availability. In other words, a process of hierarchical assembly may operate, as observed previously (3,31,36). Such a possibility is supported by the observation that although "GKER" in ␤3 is sufficient to drive assembly with ␣1 and ␥2 (see ␤2 GKER ), it is not necessary to do so (see ␤3 DNTK ). Thus, it seems possible that multiple, alternative assembly signals may exist for GABA A receptor formation, providing a flexible mechanism for the construction of GABA A receptors. In such a scenario, a ␤ subunit would be able to select its most desirable companion available without committing itself to just one partner. This is observed in ␦ knockout mice, in which ␣4 and ␣6 subunits, normally assembling with the ␦ subunit, are now free to assemble with ␥2 (35,43,44). Thus, the selection of a second choice would not be expected to be possible in the presence of a favorite. This is well illustrated by the fatal attraction suffered for the ␦ subunit in cerebellar granule neurons of ␣6 knockout mice (37) (that retain the ␣6 assembly signal), implying a complete devotion of the ␦ subunit for ␣6 assembly signal within this environment.
In summary, we have identified a single amino acid within the proposed assembly signal of ␣1 for ␤ subunits that is absolutely required for ␣1␤2, but not ␣1␤1 or ␣1␤3, receptor formation. ␤1, ␤2, and ␤3 exhibit distinct assembly profiles with ␣1 and ␥2, utilizing distinct assembly signals in ␣1/␥2. We have identified an assembly signal in ␤3 that is sufficient to drive assembly with both ␣1 and ␥2. However, it is not necessary, implicating the presence of multiple assembly signals capable of fulfilling the same function, selecting a subunit with which to assemble.