Molecular Basis of the γ-Aminobutyric Acid A Receptor α3 Subunit Interaction with the Clustering Protein Gephyrin*

Background: The molecular basis of GABAA receptor α3 subtype-specific synaptic localization is unknown. Results: GABAAR α3 interacts with the gephyrin E domain via defined intracellular motifs that partially overlap with glycine receptor binding determinants. Conclusion: GABAAR subtypes containing α3 are clustered at postsynaptic specializations via direct interactions with gephyrin. Significance: Distinct binding properties of GABAAR and GlyRs to gephyrin may govern mixed glycinergic/GABAergic transmission. The multifunctional scaffolding protein gephyrin is a key player in the formation of the postsynaptic scaffold at inhibitory synapses, clustering both inhibitory glycine receptors (GlyRs) and selected GABAA receptor (GABAAR) subtypes. We report a direct interaction between the GABAAR α3 subunit and gephyrin, mapping reciprocal binding sites using mutagenesis, overlay, and yeast two-hybrid assays. This analysis reveals that critical determinants of this interaction are located in the motif FNIVGTTYPI in the GABAAR α3 M3–M4 domain and the motif SMDKAFITVL at the N terminus of the gephyrin E domain. GABAAR α3 gephyrin binding-site mutants were unable to co-localize with endogenous gephyrin in transfected hippocampal neurons, despite being able to traffic to the cell membrane and form functional benzodiazepine-responsive GABAARs in recombinant systems. Interestingly, motifs responsible for interactions with GABAAR α2, GABAAR α3, and collybistin on gephyrin overlap. Curiously, two key residues (Asp-327 and Phe-330) in the GABAAR α2 and α3 binding sites on gephyrin also contribute to GlyR β subunit-E domain interactions. However, isothermal titration calorimetry reveals a 27-fold difference in the interaction strength between GABAAR α3 and GlyR β subunits with gephyrin with dissociation constants of 5.3 μm and 0.2 μm, respectively. Taken together, these observations suggest that clustering of GABAAR α2, α3, and GlyRs by gephyrin is mediated by distinct mechanisms at mixed glycinergic/GABAergic synapses.

Inhibitory transmission in the central nervous system is mediated by GABA A and glycine receptors (GlyRs) 4 consisting of pentameric combinations of subunits (␣1-6, ␤1-3, ␥1-3, ␦, ⑀, , and for GABA A Rs and ␣1-4 and ␤ for GlyRs), each comprising an extracellular domain, four membrane-spanning helices (M1-M4) and a large intracellular loop between M3 and M4. Cell type-specific gene expression patterns, subunit stoichiometry, and interplay between presynaptic and postsynaptic specializations are thought to underlie the spatial and temporal localization of these receptors. In particular, a postsynaptic matrix of receptor-associated proteins is essential for the dynamic localization of both GABA A Rs and GlyRs and also recruits components of specific signaling cascades to synapses (1). The multifunctional protein gephyrin (2) is a key player in the clustering of both GlyRs and GABA A Rs. For example, studies using gephyrin knock-out mice or mRNA knockdown (3)(4)(5)(6)(7)(8) have shown a loss of postsynaptic clustering of GlyRs as well as GABA A Rs containing ␣2 and ␥2 subunits. However, GABA A R ␣1 and ␣5 subunit clustering is unaltered in gephyrin knockout mice (4 -7), demonstrating that gephyrin-independent clustering mechanisms also exist in vivo. GABA A Rs containing the ␣4, ␣5, and ␣6 subunits are located preferentially at extrasynaptic sites (9) and are likely to be localized by other clustering factors, such as the actin-binding protein radixin (10). Hence, although the majority of GlyRs are likely to be clustered by gephyrin, only certain GABA A R subtypes are subject to gephyrin-dependent clustering. Curiously, the subcellular localization of gephyrin is in turn dependent on the presence of certain GABA A R subtypes. For example, targeted deletion of the GABA A R ␣1, ␣3, and ␥2 subunit genes results in a loss of synaptic gephyrin clusters (11)(12)(13)(14)(15)(16), resulting in nonsynaptic dendritic gephyrin aggregates.
The interaction of the GlyR ␤ subunit with the gephyrin E domain has been characterized in detail (17)(18)(19)(20)(21). By contrast, a direct interaction of gephyrin with GABA A R subunits has proven elusive, perhaps because of the number of individual GABA A R subunits, splice variants, accessory proteins, and post-translational modifications that could influence these interactions. However, recent studies (22,23) have demonstrated that the GABA A R ␣2 subunit interacts directly with both gephyrin and the RhoGEF collybistin (18,24). Here, we report a detailed characterization of the interaction between the GABA A R ␣3 subunit and gephyrin in recombinant systems and neuronal cultures, revealing overlapping binding determinants on gephyrin for GABA A R ␣2, GABA A R ␣3, and collybistin that are distinct from the E domain GlyR ␤-subunit binding site.

EXPERIMENTAL PROCEDURES
Expression Constructs and Site-directed Mutagenesis-GABA A R ␣1, ␣3, GlyR ␤ subunit, and gephyrin cDNAs were amplified from rat spinal cord or whole brain first-strand cDNA using Pfx DNA polymerase (Invitrogen) and cloned into the yeast two-hybrid vectors pYTH16 or pACT2. Cloning resulted in an in-frame fusion of the GAL4 DNA binding domain (GAL4BD; vector pYTH16) (25) or GAL4 activation domain (GALAD; vector pACT2) to the N termini of all expressed proteins. Mutations were introduced using the QuikChange sitedirected mutagenesis kit (Stratagene), and all constructs were verified by Sanger DNA sequencing. A hemagglutinin (HA) tag (YPYDVPDYA) was inserted between amino acids 32 and 33 of the rat GABA A R ␣3 subunit (Uniprot P20236; NCBI Entrez Gene ID 24947) using the GeneSOEing (Gene Splicing by Overlap Extension) technique. Because the ␣3 signal peptide comprises amino acids 1-28, the HA tag is located between amino acids 4 and 5 of the mature polypeptide. There are two published versions of the rat GABA A R subunit ␣3 intracellular loop in the protein data base Uniprot with regard to amino acid 381, which is either a leucine (nucleotides TTG) or lysine (nucleotides AAG). Because this amino acid is near the critical region for ␣3 subunit-gephyrin interactions, we compared both constructs in our experiments. Deletions were made in ␣3 L381 using the GeneSOEing technique. GST fusion proteins were constructed by cloning the intracellular loops into an engineered pGEX vector, which provided a C-terminal His 6 tag for purification of the fusion protein.
Culture and Transfection of Primary Hippocampal Neurons-Hippocampal cultures were made from E18 rats from Charles River as described previously (26). Transfections were made using the Amaxa System with GABA A R ␣3 and mutant expression constructs in the vector pCI (Promega). Transfected neurons were plated onto poly-L-lysine-coated glass coverslips and maintained in Neurobasal/B27 medium for 18 days. Plasmid DNA used for transfection was prepared with the Endofree maxi kit (Qiagen).
Transfection of HEK293 Cells-HEK293 cells (ATCC CRL-1573) were co-transfected with pCI expression constructs encoding HA-tagged GABA A R ␣3 and deletion mutants together with the GABA A R ␤3 subunit. Cells were initially plated on poly-L-lysine-coated glass coverslips in DMEM containing 10% FCS. For transfection we used the TurboFect transfection reagent (Fermentas) and 0.5 g of pCI GABA A R ␤3 DNA together with 0.5 g of pCI GABA A R ␣3 DNAs (deletions 1-4) according to the manufacturer's protocol using serumfree medium. Three hours after transfection, the culture medium was changed to one also containing 10% FCS. Cells were fixed and stained 48 h after lipofection.
Immunocytochemistry-Neurons grown on glass coverslips were fixed with 4% paraformaldehyde and 4% sucrose in PBS. HEK293 cells were fixed with 4% paraformaldehyde without sucrose. Nonspecific binding was blocked by incubation with 5% BSA in PBS. Primary and secondary antibodies were diluted in 1% BSA/PBS. Secondary antibodies included FITC and Cy3 anti-rabbit and anti-mouse IgGs. For detecting intracellular epitopes, cells were permeabilized with 0.05% Triton X-100 for 5 min prior to blocking with BSA. After incubation with secondary antibodies, coverslips were washed with PBS and finally with water and mounted using Mowiol mounting solution (Polysciences Inc., Warrington, PA).
Overlay Assays-GST fusion proteins were purified from Escherichia coli BL21 extracts under denaturing conditions on a nickel-nitrilotriacetic acid-Sepharose column as described (Qiagen Expressionist Manual). The protein concentration of purified proteins was quantified using the BCA protein assay (Pierce). Approximately 5 g of each fusion protein was separated on two 10% polyacrylamide gels (NuPAGE, Invitrogen) using MOPS SDS running buffer (50 mM MOPS, 50 mM Tris-HCl, 0.1% SDS, 1 mM EDTA, pH 7.7). One gel was stained with Coomassie Blue, the other gel was blotted onto a PVDF membrane (Immobilon, Millipore). The membrane was incubated in 7 M guanidinium chloride in renaturation buffer (10 mM HEPES, pH 7.0, 70 mM KCl, 80 mM NaCl, 5 mM EDTA, 1 mM ␤-mercaptoethanol) for 1 h at 4°C. This solution was diluted every hour with renaturation buffer to a final concentration of 6, 5, 4, 3, 2, and 1 M guanidinium hydrochloride. Finally the membrane was incubated in renaturation buffer without guanidinium hydrochloride for 1 h, blocked with 5% BSA and 0.03% Triton X-100 in renaturation buffer for 1 h and another 1 h in 1% BSA in detergent-free renaturation buffer. In vitro translation and [ 35 S]methionine labeling of gephyrin was performed using the TNTT7 Quick Coupled Transcription/ Translation kit (Promega) and [ 35 S]methionine (1000 Ci/mmol at 10 mCi/ml). 0.5 g of the pRK5 gephyrin P1 plasmid was mixed with 2 l of [ 35 S]methionine, 6 l of nuclease-free ddH 2 O, and 40 l of TNT T7 Master Mix and incubated for 90 min at 30°C. The blot was incubated overnight at 4°C in 20 ml of 1% BSA in detergent-free renaturation buffer with 50 l of in vitro-translated 35 S-labeled gephyrin. Subsequently, the blot was washed with 1% BSA in detergent-free renaturation buffer for 1 h and air-dried. The dry membrane was exposed to a PhosphorImager screen for 4 h and analyzed using the Quantity One software. To determine relative gephyrin binding affinities to different GST fusion proteins, the Coomassie-stained gel was analyzed by densitometry, and variations of intensities were used as correction factors when quantifying the overlay assays. The assays were repeated three times under identical conditions. For statistical evaluation, the intensities of bands were compared with the wild-type construct using an unpaired Student's t test with a confidence interval of p Ͻ 0.01.
Electrophysiology-Electrophysiological properties of ␣3␤3␥2 GABA A Rs expressed in Xenopus oocytes were measured using the two-electrode voltage clamp technique. Methods for isolating, culturing, injecting, and defolliculating of oocytes were as described previously (28). In brief, mature female Xenopus laevis (Nasco, Fort Atkinson, WI) were anesthetized in a bath of ice-cold 0.17% Tricain (ethyl-m-aminobenzoate; Sigma) before decapitation and removal of the ovaries. Stage 5-6 oocytes with the follicle cell layer intact were removed from the ovary using a platinum wire loop. Oocytes were stored and incubated at 18°C in modified Barth's medium (88 mM NaCl, 10 mM HEPES-NaOH, pH 7.4, 2.4 mM NaHCO 3 , 1 mM KCl, 0.82 mM MgSO 4 , 0.41 mM CaCl 2 , 0.34 mM Ca(NO 3 ) 2 ) that was supplemented with 100 units/ml penicillin and 100 g/ml streptomycin. Oocytes with an intact follicular cell layer were subjected to nuclear injection with a total of 3 ng of cDNA in aqueous solution per oocyte. The subunit ratio was 1:1:5 for ␣3␤3␥2 receptors consisting of wild-type or mutant ␣3 subunits together with ␤3 and ␥2 subunits. After injection of cDNAs, oocytes were incubated for at least 24 h before the enveloping follicle cell layers were removed. Collagenase treatment (type IA; Sigma), and mechanically defolliculating of the oocytes was performed as described previously (29).
For electrophysiological recordings, oocytes were placed on a nylon grid in a bath of Xenopus Ringer solution (XR; containing 90 mM NaCl, 5 mM HEPES-NaOH, pH 7.4, 1 mM MgCl 2 , 1 mM KCl, and 1 mM CaCl 2 ). The oocytes were constantly washed by a flow of 6 ml/min XR, which could be switched to XR containing GABA and/or diazepam. Diazepam was diluted into XR from dimethyl sulfoxide solutions resulting in a final concentration of 0.1% dimethyl sulfoxide perfusing the oocytes. Diaz-epam was preapplied for 30 s before the addition of GABA, which was then co-applied with the diazepam until a peak response was observed. Between two applications, oocytes were washed in XR for up to 15 min to ensure full recovery from desensitization. For current measurements, the oocytes were impaled with two microelectrodes (1-3 megaohms) which were filled with 2 M KCl. Maximum currents measured in cDNA-injected oocytes were in the microampere range for all GABA A receptor subtypes. To test for modulation of GABAinduced currents by diazepam, a concentration of 3 M GABA was co-applied to the cell with 1 M diazepam. All recordings were performed at room temperature at a holding potential of Ϫ60 mV using a Dagan CA-1B Oocyte Clamp or a Dagan TEV-200A two-electrode voltage clamp (Dagan Corporation, Minneapolis, MN). Data were digitized, recorded, and measured using a Digidata 1322A data acquisition system (Axon Instruments, Union City, CA) and analyzed using GraphPad Prism. Data for GABA-dependent dose-response curve were fitted to the equation Y ϭ Bottom ϩ (Top Ϫ Bottom)/(1 ϩ 10 (LogEC50ϪX) * n H ), where EC 50 is the concentration of the compound that increases the amplitude of the GABA-evoked current by 50%, and n H is the Hill coefficient. Data are given as mean Ϯ S.E. from at least three oocytes and two oocyte batches. Statistical significance was calculated using one-way ANOVA with Bonferroni post hoc test.
Isothermal Titration Calorimetry (ITC) and Native PAGE-Partial GABA A R ␣3 subunit variants were PCR-amplified and cloned into the NcoI/NotI sites of pETM11. M3-M4 fusion proteins were expressed in the E. coli strain BL21 (Stratagene) as His 6 -tagged proteins. Cells were grown in LB medium at 30°C, and protein expression was induced following addition of 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside for 18 h. Cells were then resuspended in lysis buffer (50 mM Tris-HCl, 500 mM NaCl, pH 8.0), passed through a French pressure cell, and centrifuged (4000 ϫ g). Proteins were initially purified using a 5-ml HisTrap FF crude column according to the instructions of the manufacturer (GE Healthcare). Protein-containing fractions were collected, concentrated, and applied to a 26/60 Superdex 200 size exclusion column (Amersham Biosciences) equilibrated with buffer (10 mM Tris-HCl, 250 mM NaCl, pH 8.0). Pure fractions were pooled, concentrated to 5-100 mg/ml, flash-frozen in 0.5-ml aliquots, and stored at Ϫ80°C. The gephyrin E domain was prepared as described previously (20,21). Prior to all ITC experiments, gephyrin and GABA A R ␣3 M3-M4 loop variants were extensively dialyzed against 10 mM Tris-HCl, 250 mM NaCl, 1 mM ␤-mercaptoethanol, pH 8.0, at 4°C overnight, followed by filtration and degassing. A 200 M solution of the GABA A R ␣3 variants was titrated as the ligand into the sample cell containing 9 M gephyrin E domain. A volume of 10 -15 l of ligand was added at a time with a total number of 20 -30 injections, resulting in a final molar ratio of ligand to protein of 4:1. All experiments were performed using a VP-ITC instrument (MicroCal, Northampton, MA) at 25°C. Buffer-to-buffer titrations were performed as described above, so that the heat produced by injection, mixing, and dilution could be subtracted prior to curve fitting. The binding enthalpy was measured directly, whereas affinity (K D ) and stoichiometry (N) were obtained by data analysis using the ORIGIN software.
Native PAGE gels containing 1% agarose and Tris/glycine, pH 8.4, were used to separate the E domain of gephyrin from the E domain-GABA A R ␣3 complex. 10 l of the E domain (5 M) mixed with the respective ligand (80 M) was loaded in each lane.

RESULTS
Mapping Determinants of GABA A R ␣3 Subunit Binding to Gephyrin Using Overlay and Yeast Two-hybrid Assays-To assess the possible interaction of the GABA A R ␣3 subunit with gephyrin we performed overlay assays using GST-tagged GABA A R ␣3 fusion proteins and recombinant 35 S-labeled gephyrin, using the abundant P1 isoform that lacks the C3 and C4 cassettes (1). Because the most likely binding site is located in the large intracellular domain between membrane-spanning domains 3 and 4, we split the 99-amino acid domain (constructs HL1 and HL2) and engineered several deletion constructs (⌬1-⌬4; Fig. 1A). The HL1 construct was generated in two versions, due to an apparent polymorphism at amino acid 381 (Leu-381 versus Lys-381). Quantitative analysis (Fig. 1, B and C) revealed that the intact GST-GABA A R ␣3 M3-M4 domain fusion protein (but not GST alone) showed a robust interaction with recombinant gephyrin (Fig. 1, B and C). This interaction was retained by both versions of the N-terminal portion of the M3-M4 domain (HL1 L381 and HL1 K381 ; amino acids Asn-332-Asp-382), but binding of the C-terminal HL2 fragment was negligible (amino acids Thr-383-Lys-430). This suggested that the polymorphism at amino acid 381 does not influence GABA A R ␣3-gephyrin interactions and, more importantly, that a gephyrin-binding motif is likely to reside in the N-terminal 51 amino acids of the M3-M4 domain. Consistent with this hypothesis, further analysis revealed that ␣3-gephyrin interactions were most severely reduced by the overlapping deletions ␣3⌬1 (Pro-357-Ile-377) and ␣3⌬4 (Phe-368 -Ile-377) that lie within the N-terminal HL1 fragment, although deletion ⌬3 (Thr-361-Thr-367) also appeared to reduce ␣3-gephyrin interactions, albeit to a lesser extent than ⌬1 and ⌬4.
As a second confirmatory assay, we utilized the yeast twohybrid system to test interactions between the M3-M4 intracellular loops of the GABA A R ␣1 and ␣3 subunits and GlyR ␤ (control) with full-length P1 gephyrin and ␣3 deletion mutants. Although all bait proteins are expressed in yeast (23) the GABA A R ␣3 and GlyR ␤ baits, but not the GABA A R ␣1 bait, interact with full-length gephyrin and two N-terminal deletion mutants (Geph276 -736 and Geph305-736), as assessed by LacZ freeze-fracture assays (Fig. 2). Using deletion and domain swap mutations (Fig. 2), we found that consistent with overlay assays, deletion of the Thr-361-Ile-377 (␣3⌬5) and Phe-368 -Ile-377 (␣3⌬4) abolished the interaction of the GABA A R ␣3 subunit with gephyrin (Fig. 2, A and B). By contrast, deletion of Thr-361-Thr-367 (␣3⌬3) revealed that this motif was dispensable for GABA A R ␣3-gephyrin interactions in yeast. Because both overlay and yeast two-hybrid assays suggest that the motif FNIVGTTYPI (Phe-368 -Ile-377) contains key determinants of GABA A R ␣3-gephyrin interactions, we inserted Phe-368 -Ile-377 from ␣3 into a wild-type GABA A R ␣1 subunit bait, which does not normally interact with gephyrin (23) in yeast (Fig. 2C). This bait (named ␣3in␣1) was able to interact with gephyrin ( Fig. 2C), confirming that we have identified key residues involved in the GABA A R ␣3-gephyrin interaction.
Clustering Properties of HA-tagged GABA A R ␣3 and Bindingsite Mutants in Transfected Hippocampal Neurons-To verify the nature of the identified gephyrin binding motif on GABA A R ␣3 in a neuronal context, we transfected hippocampal neurons with wild-type HA-tagged GABA A R ␣3 and selected deletion mutants by nucleofection before plating (Fig. 3). At 17 days in vitro, neurons were immunostained with HA antibodies under nonpermeabilizing conditions and after permeabilization with mAb7a, which recognizes gephyrin. We first examined the subcellular distribution of wild-type HA-␣3, which formed clusters on the surfaces of both neuronal processes and the cell body (Fig. 3A) that co-localized with endogenous gephyrin (Fig. 3,  B-D). Consistent with our overlay and yeast two-hybrid data, construct HA-␣3⌬1 (removing Pro-357-Ile-377) exhibited a diffuse distribution on the plasma membrane and showed little or no co-localization with gephyrin (Fig. 3, E-H). By contrast, HA-␣3⌬3 (lacking Thr-361-Thr-367) showed robust co-localization with gephyrin (Fig. 3, I-L), confirming that this motif is dispensable for both gephyrin binding and synaptic targeting. Finally, HA-␣3⌬4 (lacking the minimal gephyrin binding motif Phe-368 -Ile-377) showed a diffuse cytoplasmic distribution with some small puncta that did not co-localize with endogenous gephyrin (Fig. 3, M-P).
To control for possible alterations in the assembly of the GABA A Rs containing ␣3, their capacity to access the plasma membrane and form functional benzodiazepine-responsive GABA A Rs was assessed using recombinant expression in HEK293 cells (Fig. 4). Using immunofluorescence under permeabilizing and nonpermeabilizing conditions, it was evident that all deletion constructs except ␣3⌬2 localized to the cell membrane (Fig. 4A). This mutant also showed changes in the maximal response to GABA (Fig. 4B), but normalization of dose-response curves indicated no significant changes in EC 50 values (Fig. 4C). Importantly, the deletion mutants affecting the gephyrin binding site (␣3⌬1 and ␣3⌬3; see Fig. 1) had no significant influence on membrane localization or electrophysiological properties of recombinant GABA A Rs, including benzodiazepine responsiveness (Fig. 4D).
Mapping a Binding Site for the GABA A R ␣3 Subunit on Gephyrin-The yeast two-hybrid system was also used to map potential determinants of the GABA A R ␣3 subunit binding site on the gephyrin G, C, or E domains (Fig. 5A), to determine overlap with previously reported GABA A R or GlyR binding motifs (20,21,23). Sequential deletion analysis revealed that the minimal gephyrin prey interacting with the GABA A R ␣3 bait encompassed amino acids 305-736 (Geph305-736) encoding the E domain and the C-terminal eighteen amino acids of the linker region. Shorter constructs, such as Geph323-736 and Geph305-704, did not mediate this interaction as observed previously for GABA A R ␣2 and collybistin baits (18,23). We therefore used alanine-scanning mutagenesis to locate determinants of GABA A R ␣3 subunit binding to gephyrin (Fig. 5B). Two alanine block mutants (A5 and A6) completely disrupted interactions of the wild-type GABA A R ␣3 subunit, ␣3⌬3 and ␣3in␣1 baits with gephyrin, while leaving GlyR ␤ subunitgephyrin interactions unaffected (Fig. 5, B and C). It is also noteworthy that GABA A R ␣3 subunit and ␣3⌬3 interactions with the Ala-7 mutant ( 335 EMPTV 339 to 335 AAAAA 339 ) were weak compared with interactions with the A2-A4 and A8 -A9 mutants, suggesting that robust GABA A R ␣3 binding might also require one or more amino acids in this sequence. Curiously, however, the GABA A R ␣3in␣1 chimeric bait showed a stronger interaction with this gephyrin mutant. Despite this caveat, the core GABA A R ␣3 subunit binding motif within the N-terminal part of the gephyrin E domain (SMDKAFITVL; Fig.  1) shows partial overlap with the previously determined collybistin (PFPLTSMDKA) and GABA A R ␣2 (SMDKAFITV-LEMPTVLGTE) binding motifs on gephyrin (18,23). Interestingly, two residues (Asp-327 and Phe-330) in the minimal GABA A R ␣2 and ␣3 binding sites on gephyrin have been previously implicated in GlyR ␤ subunit-E domain interactions (21) (Fig. 5A). However, mutation of Asp-327 and Phe-330 in mutants A5 and A6 respectively, was not sufficient to disrupt GlyR ␤-E domain interactions in the YTH system (Fig. 5C).
Comparison of GABA A R ␣3 and GlyR ␤ Subunit Binding to Gephyrin-To quantify GABA A R ␣3 subunit binding to gephyrin and to compare this with GlyR ␤ subunit-gephyrin interac- , and ⌬4 (M-P) were stained with HA antibodies under nonpermeabilizing conditions (without Triton X-100; green) and after permeabilization with 0.05% Triton X-100 with mAb7a against gephyrin (red). Note that wild-type GABA A R ␣3 and mutant ⌬3 display good co-localization with gephyrin, whereas for mutants ⌬1 and ⌬4 little co-localization is observed. The third panel in each row represents the merged images, and enlargements of the respective dendrites (indicated by white boxes) are displayed in the fourth panel. Scale bars, 25 m. tions, we used ITC. GABA A R ␣3 and the ␣3⌬4 deletion variant were expressed in E. coli as His 6 fusion proteins to determine interactions with a gephyrin E domain fragment (amino acids 318 -736) (21). This gephyrin construct has previously been used to characterize GlyR ␤ subunit binding to gephyrin (20,21). ITC revealed that the intracellular domain of the ␣3 subunit bound in an exothermic reaction (⌬H ϭ Ϫ4.9 Ϯ 1.2 kcal/ mol) to the gephyrin E domain with a dissociation constant in the low micromolar range (K D ϭ 5.3 Ϯ 1.5 M), and a stoichiometry of 0.77 Ϯ 0.18 mol/mol (Fig. 6A). As expected, no binding was observed for ␣3⌬4 in ITC. The previously analyzed E domain-GlyR ␤ loop interaction was significantly stronger (20,21) with a K D of 0.2 M but also displayed a two-site binding behavior with the second lower affinity binding site exhibiting a K D of 11 M. To further confirm the mapping of the binding site on GABA A R ␣3 we employed native PAGE to investigate the behaviors of ␣3 and ␣3⌬4 with and without the gephyrin E domain. Due to the high isoelectric point of the GABA A R ␣3 construct (calculated pI of ϳ10) it is positively charged at the pH (8.4) at which this experiment was conducted and hence does not enter into the gel. By contrast, the E domain of gephyrin is negatively charged (calculated pI of ϳ7) and migrates into the gel. Native PAGE assays confirmed complex formation of the gephyrin E domain with ␣3, but not ␣3⌬4 (Fig. 6B).

DISCUSSION
It is becoming increasingly apparent that GABA A R ␣ subunits not only influence GABA A R physiology, pharmacology, and biological function, but also mediate the synaptic versus extrasynaptic localization of these receptor subtypes via distinct protein-protein interactions. For example, the GABA A R ␣2 subunit co-localizes with gephyrin in vivo and interacts with FIGURE 4. Cell surface expression and functional properties of benzodiazepine-sensitive GABA A Rs containing ␣3 deletions. A, HEK293 cells were transfected with HA-tagged GABA A R ␣3 and deletion constructs (␣3⌬1-4) together with the ␤3 subunit. Two days after lipofection, cells were fixed and stained with anti-HA polyclonal antibody without cell permeabilization. All constructs directed formation of cell surface ␣3␤2 GABA A Rs with the exception of ␣3⌬2, which lacks a binding site for the ubiquitin-like protein Plic-1, which presumably causes impaired membrane insertion (35). As the ␣3⌬2 mutant did not show cell surface expression, cells were permeabilized to allow detection of intracellular antigens. B and C, GABA dose-response curves for ␣3␤3␥2 and ␣3⌬1-4 deletion constructs together with ␤3␥2 are shown. C, data are normalized to the maximum GABA current. Data points represent mean Ϯ S.E. (error bars) from at least three oocytes derived from Ն2 batches. EC 50 values were compared using one-way ANOVA (with Bonferroni post hoc test) and found to be not significantly different. D, Xenopus oocytes expressing recombinant GABA A receptors containing HA-tagged GABA A R ␣3 and deletion constructs (␣3⌬1-4) together with ␤3 and ␥2 in the presence of 3 M GABA were challenged with 1 M diazepam. Stimulation was normalized to the control current at 3 M GABA. Control current represents 100% stimulation. Data represent means Ϯ S.E. of at least three oocytes. Mean values were determined from recordings of three or four cells.
both gephyrin and the RhoGEF collybistin via overlapping binding sites (22,23). For this reason, we considered that other GABA A R ␣ subunits might also be clustered at synapses by these molecules. The ␣3 subunit was selected for analysis because (i) GABA A Rs containing this subunit co-localize with gephyrin, e.g. in cerebellar cortex (30), the thalamic reticular nucleus (16), or at perisomatic synapses in the globus pallidus (31) and (ii) gephyrin is mislocalized in GABA A R ␣3 subunit knock-out mice (15,16).
Using a combination of deletion mutagenesis, overlay, and yeast two-hybrid assays we have determined that GABA A R ␣3 specifically interacts with gephyrin via a critical motif (FNIVGTTYPI) that overlaps with sequences that bind gephyrin and collybistin in GABA A R ␣2. Curiously, very few amino acids are conserved between the equivalent regions of the ␣2 and ␣3 subunits, suggesting that either the nature of the amino acids in these motifs is crucial or that conserved amino acids within (Tyr-375) or directly flanking these minimal motifs (e.g. Asn-378) are crucial determinants of gephyrin binding. Certainly, deletion of the minimal gephyrin binding motif prevented synaptic clustering and co-localization of recombinant GABA A R ␣3 with endogenous gephyrin in cultured neurons.
Using the yeast-two hybrid system, we also mapped crucial determinants of GABA A R ␣3 binding to gephyrin to the start of the E domain. These residues overlap with previously characterized determinants of GABA A R ␣2 binding on gephyrin and show partial overlap with key determinants (Asp-327 and Phe-330) of GlyR ␤ binding to gephyrin (21) (Fig. 7) This suggests that binding of GABA A Rs containing ␣2, ␣3, and GlyRs to gephyrin could be mutually exclusive. However, given that mutation of Asp-327 and Phe-330 in mutants A5 and A6 did not appear to disrupt GlyR ␤-E domain interactions in the YTH system, key differences may exist between GABA A R and GlyR binding to gephyrin. Consistent with this hypothesis, ITC revealed that the GABA A R ␣3 bound to the gephyrin E-domain in a 1:1 ratio and displayed a lower affinity (K D ϭ 5.3 Ϯ The potential GABA A R ␣3 subunit binding site on gephyrin (amino acids 325-334) is shown together with motifs vital for GABA A R ␣2 subunit (amino acids 325-343) (23), collybistin (amino acids 320 -329) (18), and two residues (Asp-327 and Phe-330) implicated in GlyR ␤-gephyrin interactions (17) (purple lettering, top sequence). LacZ freeze-fracture assay rankings: ϩϩϩϩ, strong interaction; ϩϩ, moderate interaction; ϩ, weak interaction; Ϫ, no detectable interaction. C, GABA A R ␣3, ␣3⌬3, ␣3in␣1, and GlyR ␤ subunit intracellular loop baits were tested for interactions with Geph276 -736 and alanine substitution mutants created in this prey (A2-A9). LacZ freeze-fracture assays demonstrate that the GABA A R ␣3 bait does not interact with mutants A5 and A6 and is weakened in mutant A7, whereas the GlyR ␤ subunit bait interacts with all of these preys.
1.5 M) than previously observed for GlyR ␤ (two binding sites with K D values of 0.2 M and 11 M) (20,21). Collectively, these observations may explain why GABA A R-gephyrin interactions have been difficult to characterize by immunoprecipitation and are prone to the effects of detergents (22). GABA A Rs containing the ␣3 subunit co-localized with gephyrin in both dendritic and perisomatic locations in different brain regions (16,30,31). This may reflect different roles for the ␣3 subunit in mediating dendritic inhibition, affecting the efficacy and plasticity of excitatory synaptic inputs of principal , the E domain hardly enters the gel when bound to the GABA A R ␣3 intracellular loop (PI ϳ10), and GephE is fully complexed with the GABA A R ␣3 loop. However, no binding is seen for ␣3⌬4 even when using a 16-fold molar excess of the GABA A R ␣3 fragment over gephyrin. cells, versus perisomatic inhibition, controlling output by synchronizing action potential firing of larger groups of principal cells. Certainly, although GABA A receptors containing the ␣3 subunit are thought to represent only 10 -15% of all GABA A receptors, they are the major GABA A receptor subtype expressed in brain stem monoaminergic nuclei (32,33). Consistent with these findings, GABA A Rs containing ␣3 have been linked to sensorimotor gating and affective and cognitive functions phenotypes (32,34). A molecular understanding of the basis of synaptic localization of GABA A R ␣3-containing receptors by gephyrin adds to our knowledge of this interesting receptor subtype.