Agonist-dependent Endocytosis of γ-Aminobutyric Acid Type A (GABAA) Receptors Revealed by a γ2(R43Q) Epilepsy Mutation*

Background: Modulation of GABAA receptors trafficking is critical for controlling inhibitory neurotransmission. Results: A point mutation or agonist application, both affecting the GABAA receptor extracellular domain, has an effect on receptor endocytosis. Conclusion: Endocytosis of GABAA receptors is linked to agonist-induced conformational changes. Significance: This represents one of the few reports demonstrating an influence of extracellular effectors on GABAA receptor trafficking. GABA-gated chloride channels (GABAARs) trafficking is involved in the regulation of fast inhibitory transmission. Here, we took advantage of a γ2(R43Q) subunit mutation linked to epilepsy in humans that considerably reduces the number of GABAARs on the cell surface to better understand the trafficking of GABAARs. Using recombinant expression in cultured rat hippocampal neurons and COS-7 cells, we showed that receptors containing γ2(R43Q) were addressed to the cell membrane but underwent clathrin-mediated dynamin-dependent endocytosis. The γ2(R43Q)-dependent endocytosis was reduced by GABAAR antagonists. These data, in addition to a new homology model, suggested that a conformational change in the extracellular domain of γ2(R43Q)-containing GABAARs increased their internalization. This led us to show that endogenous and recombinant wild-type GABAAR endocytosis in both cultured neurons and COS-7 cells can be amplified by their agonists. These findings revealed not only a direct relationship between endocytosis of GABAARs and a genetic neurological disorder but also that trafficking of these receptors can be modulated by their agonist.

Inhibitory transmission relies greatly on ionotropic GABA A receptors (GABA A Rs) 3 that are involved in many physiological functions and are the target of several drugs in wide clinical use (1). GABA A R trafficking is modulated by a number of different mechanisms, including surface targeting, mobility, and endocytosis. Moreover, studies in physiological and pathophysiological states have revealed that the number of GABA A Rs on the cell membrane have a profound influence on GABAergic neurotransmission (1)(2)(3). Inhibitory neurotransmission is also regulated by the exchange between surface and intracellular compartments via a constitutive clathrin-mediated dynamin-dependent endocytosis pathway (4 -6). This constitutive internalization is modulated by intracellular mechanisms and is altered in pathological conditions (4, 6 -9).
Epilepsies are complex syndromes with multiple causes and symptoms, but it is well established that alteration or modulation of GABA neurotransmission plays an important role in the disease and its treatment (10 -15). Moreover, genetic evidence has revealed a direct link between epilepsy and GABA A R dysfunction, including trafficking alteration, supporting the hypothesis that defects in GABA A Rs lead to seizures (16 -17). These mutations also offer an opportunity to obtain new insights into GABA A R structure and function as well as clues to the role of these receptors in neurological disorders (14). For example, an R43Q mutation located in the ␥2 subunit N-terminal extracellular domain is linked to childhood absence epilepsy and febrile seizure (17); heterozygous mice harboring this mutation replicate the human clinical phenotype (16). Intensive research into this mutation (18 -26) has led to controversial data on its effects on GABAergic physiology suggesting that ␥2(R43Q) might modify the dynamics of subunit trafficking (27).
Here, we analyzed ␥2(R43Q) trafficking in cultured hippocampal neurons and COS-7 cells and revealed that receptors containing the ␥2(R43Q) subunit had a shorter residence time on the plasma membrane than their wild-type counterparts.
We also showed that endocytosis of the mutated receptor was clathrin-and dynamin-dependent. However, it was surprising that a mutation in the extracellular domain (bearing binding sites for agonists and modulators) could have an influence on internalization, believed to be controlled through the intracellular domain. Moreover, endocytosis of GABA A Rs triggered by agonist exposure remains to be fully assessed (11, 28 -31). Then, by using both imaging and biochemical methods, further experiments revealed that agonist exposure triggered an increase of wild-type GABA A R endocytosis, both on native and recombinant GABA A Rs.
Cell Culture and Transfection-COS-7 and HEK 293 cells were maintained in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% fetal calf serum (Eurobio). Embryonic hippocampal neurons were obtained from E18 rat embryos, as described earlier (22). COS-7 and HEK 293 cells were transfected using the FuGENE 6 reagent (Roche Applied Science), according to the manufacturer's specifications, with equal amounts of ␣1, ␤, and ␥2 subunit cDNAs (and GFP for electrophysiology experiments on HEK 293 cells) (0.3 g/well in 24-well plates). Cells were incubated with cDNAs for 24 h before analysis (22). Hippocampal neurons were transfected in vitro at 7-11 days, using the Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer's specifications. The cells were analyzed 24 -60 h after transfection (22).
Immunocytochemistry-For living cell surface labeling, COS-7 cells and hippocampal neurons were incubated with antibodies at room temperature for 20 min in Dulbecco's modified Eagle's medium or Neurobasal medium supplemented with 10 mM HEPES, respectively. Receptors on the surface were labeled with antibodies raised in rabbits against either the ␣1 GABA A subunit N-terminal domain or the Myc tag. Sera were diluted 1:500 (anti-␣1) or 1:200 (anti-Myc) in medium. After incubation, cells were washed quickly by dipping coverslips in medium and fixed for 10 min in phosphate-buffered saline (PBS) containing 4% sucrose and 4% paraformaldehyde preheated to 37°C, washed in PBS, and blocked in 0.3% bovine serum albumin and 50 mM glycine (in PBS) for 15 min. Cells were washed in PBS containing 0.3% bovine serum albumin. After cell permeabilization using 0.3% Triton X-100, intracellular tagged ␥2 subunits were detected by incubating the cells with a mouse anti-Myc 9E10 antibody (1:1000; Roche Applied Science) for 2 h. Intracellular ␣1 subunits were detected with a mouse anti-␣1 (1:1000). Polyclonal and monoclonal antibodies were detected using an Alexa Fluor-568-coupled anti-rabbit antibody and an Alexa Fluor-488-coupled anti-mouse antibody (1:1000), respectively. Endoreticulum and cis-Golgi staining were revealed with polyclonal anti-calreticulin (1:500) and antimembrin (1:1000), respectively. The endoplasmic reticulum-Golgi intermediate compartment was revealed with ERGIC-GFP (37).
Internalization-Transfected neurons or COS cells were incubated in culture medium containing mouse monoclonal Myc antibody (1:200) at 37°C for 30 min. The medium also contained GABA A R agonists or antagonists as required. In the case of neurons, a mixture of inhibitors (6-cyano-7-nitroquinoxalene-2,3-dione (10 M), D-2-amino-5-phosphonovaleric acid (50 M), and tetrodotoxin (1 M)) was added, so as to prevent activity-dependent modulation of GABA A R trafficking (7). This mixture was added 5 min before the labeling experiments and was present throughout, both in control experiments and in GABA A R agonist/antagonist-treated neurons. Surface labeling was carried out at 20°C using an anti-mouse secondary antibody coupled to Alexa Fluor-647 for 30 min (38); alternatively, labeled receptors remaining on the COS-7 cell surface were acid-washed for 2 min at 20°C with culture medium adjusted at pH 2.5 (39,40). Cells were then fixed using 4% paraformaldehyde in 4% sucrose and permeabilized in Triton X-100. Total protein content was assessed by incubating with a polyclonal antibody directed at the Myc tag. Internalized protein was then revealed by incubating with anti-mouse antibody coupled to Alexa Fluor-488 at room temperature for 1 h, whereas total protein was revealed using anti-rabbit antibody coupled to Alexa Fluor-568. For native ␣1 expressed in hippocampal neurons, cells were incubated in culture medium containing an antibody made in rabbits directed at the N-terminal domain (extracellular) of ␣1 (1:500) at 37°C for 30 min. Surface labeling was carried as above using an anti-rabbit secondary antibody. Cells were then fixed and permeabilized as above. Total ␣1 subunit content was assessed by incubating with a mouse antibody directed at the Myc tag or at the native ␣1 at room temperature for 20 min. Internalized protein was then revealed by incubating cells with anti-mouse antibody coupled to Alexa Fluor-488 at room temperature for 1 h, whereas total protein was revealed using anti-rabbit antibody coupled to Alexa Fluor-568.
Quantitative Analysis of Fluorescence Signals-Fluorescence microscopy was performed using a Zeiss Axioplan 2 microscope, with a ϫ63, 1.4 numerical aperture oil immersion lens.
Quantification of fluorescence signals and background subtraction were performed using ImageJ (National Institutes of Health). For each image acquired, background levels were determined using the surface and intracellular signals measured in neighboring non-transfected cells and subtracted from the values obtained in transfected cells. Numerical data are presented as mean Ϯ S.E., and statistical significance was assessed using one-way analysis of variance (Origin, Originlab Corp.) (significance level, p Ͻ 0.05). Confocal microscopy was performed using an upright Leica DMR TCS SPZ AOBS, with a ϫ63, 1.4 numerical aperture Leica HPCL Fluotar oil objective. Colocalization was quantified using a plugin for ImageJ designed by F. Levet and C. Poujol (BIC (Bordeaux Imaging Center), Bordeaux, France). Briefly, two images, one containing GABA A R subunit labeling and one containing the labeling for a cellular compartment, were thresholded in the same way. The plugin calculates the percentage of pixels containing ␥2 subunit labeling that also contain specific labeling for a cellular compartment. The percentage of colocalization was normalized for total ␥2 and ␥2(R43Q) immunoreactivity, respectively. Analyses were performed in parallel cultures, blind to experimental conditions. Quantification of surface clusters or intracellular punctate labeling, blind to experimental conditions, was performed using ImageJ (National Institutes of Health). Threshold was applied to the images, and the number as well as the area of surface clusters or internalized particles were measured using the particle analyzer module of ImageJ. For COS-7 cells, the whole cell was counted. For neurons, an area of 10-m length along a dendrite was counted. For all experiments, total protein expression was assessed by antibody labeling after permeabilization of the cells and was measured for the same area to allow normalization of the values. To calculate fluorescence ratios, a stack was created for each cell in ImageJ with the image corresponding to the surface and total labeling. This allows us to draw the outline of the cell and measure the average surface and total fluorescence for the same area.
Biotinylation Assays-Biotinylation experiments were performed essentially as described previously (36,38). COS-7 cells were transfected in 6-well plates (2 wells/condition) and were incubated 24 h post-transfection. Cells were then washed two times with PBS, pH 8.0, incubated with 1 mg/ml EZ-Link Sulfo-NHS-SS-Biotin (Pierce) in PBS for 30 min at 4°C, washed three times with PBS, and scraped in lysis buffer containing 25 mm HEPES, 150 mm NaCl, 1% Triton X-100, and a mix of protease inhibitors (Roche Applied Science). After centrifugation, the supernatant was immunoprecipitated with 50 l of Immunopure immobilized streptavidin-beaded agarose overnight at 4°C and washed extensively. Surface and total proteins were separated on SDS-PAGE and revealed by Western blotting using anti-Myc antibodies at a 1:1000 dilution. Quantification of Western blots was performed using ImageJ (National Institutes of Health) or with the Chemi Doc XRSϩ under the control of the Image Lab software (Bio-Rad). To assess ␥2 internalization, plates were returned to 37°C for 30 min after biotinylation to allow endocytosis. Cells were then exposed to 50 mM MESNA, which cleaved biotin from proteins remaining on the surface. A sample was kept at 4°C (instead of 37°C) to control to ensure that the cleavage with MESNA was complete. Samples were analyzed by Western blot as above.
Electrophoretic and Western Blot Analyses-COS-7 cells were homogenized in buffer containing 20 mM HEPES, 0.15 mM EDTA, and 10 mM KCl, pH 8, supplemented with a mixture of protease inhibitors (Roche Applied Science). The buffer was then adjusted to 12% sucrose, and after four more strokes, the cells were centrifuged at 2000 rpm for 3 min to remove genomic DNA. The supernatant was centrifuged at 15,000 rpm for 30 min. The pellet was recovered, and cell membranes were solubilized with 15 strokes in a buffer containing 20 mM Tris-HCl, 0.15 mM EDTA, 150 mM NaCl, 2% Triton X-100, and 0.5% deoxycholate, pH 8, supplemented with a mixture of protease inhibitors, and then incubated for 45 min. The sample was centrifuged for 45 min at 15,000 rpm. The supernatant was supplemented with loading buffer and analyzed as described (22).
Electrophysiology-Brightly fluorescent isolated HEK 293 cells were selected for recording. Cells were bathed in a solution containing 150 mM NaCl, 2 mM KCl, 2 mM MgCl 2 , 2 mM CaCl 2 , 10 mM glucose, 10 mM HEPES, equilibrated to pH 7.4 with NaOH. Cells were recorded in whole cell mode and placed under the flow of a theta tube pulled to a final opening of ϳ100 m mounted on a piezoelectric translator (Physik Instrumente). Currents were evoked by applications of 100 M GABA for 5 s every minute at Ϫ60 mV and recorded at a sampling frequency of 2 kHz by an EPC10 amplifier (HEKA). GABA was exchanged for gabazine (100 M) and applied to the cell after 3 min, necessary for complete exchange. Thereafter, the medium was exchanged again for GABA, and we observed complete recovery of the response to the agonist. Data were analyzed with Igor 5 (Wavemetrics).
Model Building-The sequences of the human ␣1, ␤2, and ␥2 GABA receptor subunits were retrieved from the Ligand Gated Ion Channel database (41). The model of the ␣1␤2␥2 receptor was constructed by homology modeling using the structure of the glutamate-gated chloride channel as a template (Protein Data Bank code 3RIF) (42) and sequences alignments obtained with T-coffee (43). Homology modeling was performed with Modeler version 9.5 (44) using default settings. 100 models were prepared, and the best model according to the Discrete Optimized Protein Energy function (DOPE) was selected.
Because clathrin-mediated, dynamin-dependent endocytosis is the major neuronal GABA A R internalization mechanism (4, 6), we tested whether ␥2(R43Q) internalization was driven by a similar pathway. When dynamin was inhibited by incubation with 80 M dynasore (46), ␥2(R43Q) was detected at the plasma membrane of transfected COS-7 cells (Fig. 2A). The ratio of ␥2(R43Q) subunit labeled on the cell surface versus ␥2(R43Q) labeled within intracellular compartments was increased from 0.42 Ϯ 0.2 to 0.78 Ϯ 0.18 (Fig. 2B). COS-7 cells were then transfected with a ␤2(LL/AA) mutant; this mutation within the ␤2 intracellular domain reduces the interaction of GABA A R with the AP2 complex (6). In this case, the signal ratio between surface and intracellular ␥2(R43Q) increased from 0.18 Ϯ 0.08 to 0.53 Ϯ 0.1 (Fig. 2, C and D), showing that ␥2(R43Q) was present on the cell surface when co-expressed with ␣1 and ␤2(LL/AA). Taken together, these data showed that the ␥2 subunit containing the R43Q mutation increased endocytosis of GABA A Rs, hindering their detection on the cell surface.
␥2(R43Q) Endocytosis Is Inhibited by GABA A R Antagonists-This increased internalization of a ligand-gated channel, resulting from a mutation within the extracellular domain of the molecular complex, was surprising. This GABA A R domain contains binding sites for agonists and allosteric modulators, whereas the intracellular domain mediates interactions with trafficking factors (1,4,6,47). We therefore tested whether GABA A R ligands interfered with endocytosis. Incubating transfected neurons (Fig. 3, A and B 3B). In COS-7 cells transfected with ␣1-, ␤2-, and ␥2 subunits, the surface/intra ratio were 3.8 Ϯ 0.5 in control conditions and 4 Ϯ 0.9 or 3.3 Ϯ 0.4 in the presence of gabazine (100 M) or picrotoxin (100 M), respectively (Fig. 3D). In COS-7 cells transfected with ␣1, ␤2, and ␥2(R43Q) subunits, the surface/ intra ratio increased from 0.3 Ϯ 0.1 to 1.27 Ϯ 0.2 for gabazine and 1.41 Ϯ 0.23 for picrotoxin (Fig. 3D). Muscimol (10 M), a GABA A R agonist, had no detectable effect, compared with the control experiments (Fig. 3, B and D). Cell surface biotinylation The fact that picrotoxin and gabazine are both allosteric GABA A R antagonists (48 -50) indicated that, compared with their wild-type counterparts, the equilibrium between states in ␥2(R43Q)-containing GABA A Rs shifted away from the resting state toward the active or desensitized state. Our new homology model based on a glutamate-gated chloride channel (42) suggests that the Arg-43 residue of the ␥2 subunit is connected to Tyr ϭ 174 and Glu-178 from the loop B and to Asp-84 and Arg-86 of the ␤2 subunit via polar interactions (Fig. 4, B-D). In addition, this new model shows that ␥2Arg-43 and ␤2Asp-84/ Arg-86 are on loops (Allo1 and Allo2) identified as involved in the motion that opens the channel pore (51). This prompted us to test whether this alteration in the equilibrium between states favored an open channel conformation. We thus co-expressed wild-type or ␥2(R43Q) constructs with ␣1 and ␤2 subunits in HEK cells. In ␣1-, ␤2-, and ␥2(R43Q)-transfected cells, maxi-mum GABA-gated currents were 16% (p Ͻ 0.01) of currents from cells expressing wild-type receptors (Fig. 4E), in agreement with previous findings (19,23,52). Gabazine, an allosteric antagonist, did not cause a significant change in the current base line during whole cell voltage clamp recordings (Fig. 4F) in HEK cells co-expressing ␣1-, ␤2-, and either wild-type or ␥2(R43Q) subunits (n ϭ 7 and 15, respectively), showing that ␥2(R43Q)-containing receptors had no constitutive activity.
Collectively, our data suggested that ␥2(R43Q) mutation triggered GABA A R endocytosis through a structural change in the extracellular domain. Because agonist exposure triggers a long range conformational change (53,54), our finding on mutated receptors opened the interesting possibility that endocytosis of wild-type GABA A Rs may be modulated by agonists. Our data did not yet indicate significant alteration of the surface level of wild-type ␥2-containing receptors in the presence of agonists (Fig. 3). We thus decided to examine directly the impact of agonists on the internalized fraction of GABA A Rs.
GABA A R Agonist Enhances Receptor Endocytosis-To test a possible link between GABA-induced conformational change and endocytosis, suggested by our studies on ␥2(R43Q), we directly quantified internalization by measuring the level of endogenous ␣1-containing receptors internalized during agonist application on cultured neurons (Fig. 5A). In control and muscimol experiments, excitatory activity was blocked with 6-cyano-7-nitroquinoxalene-2,3-dione, D-2-amino-5-phosphonovaleric acid, and tetrodotoxin in the medium. Neurons were incubated with an antibody directed at the extracellular N-terminal domain of ␣1 subunit for 30 min at 37°C. Labeled receptors remaining at the cell surface were labeled with saturating concentration of Alexa Fluor-647 secondary antibodies, internalized receptors were labeled with Alexa Fluor-488 secondary antibodies (Fig. 5A, green), whereas total ␣1 subunits were labeled with a monoclonal antibody and Alexa Fluor-568 secondary antibodies (Fig. 5A, red). Quantification of the area of punctate labeling of internalized receptors showed a significant increase when neurons were incubated with 10 M muscimol (the area of intracellular punctate labeling/10 m increased to 3.14 Ϯ 0.36 m 2 from 1.62 Ϯ 0.17 in control condition). This increase was abolished when muscimol was co-applied with 100 M gabazine or picrotoxin (1.68 Ϯ 0.15 or 1.89 Ϯ 018 m 2 , respectively). Application of gabazine or picrotoxin alone had no effect (1.69 Ϯ 0.19 or 1.61 Ϯ 016 m 2 , respectively) (Fig. 5B).
In another set of experiments, we analyzed ␥2 subunit internalization in COS cells transfected with ␣1, ␤2, and tagged ␥2 subunits (Fig. 6A). Average values Ϯ S.E. from four independent experiments and 105 cells for each condition were plotted (Fig. 6B). These experiments showed that the number of internalized ␥2 subunits increased during agonist treatment. We also assessed ␥2 internalization by performing a biotinylation assay on COS-7 cells transfected as above. After labeling with cleavable biotin, plates were returned to 37°C for 30 min to allow endocytosis. Cells were then exposed to MESNA, which cleaved biotin from proteins remaining on the surface, allowing determination of intracellular biotinylated receptors. As shown in Fig. 6, C and D, application of GABA induced a significant increase of the internalization (212 Ϯ 40% of control, n ϭ 6, p Ͻ 0.05) that was suppressed when antagonists were co-applied with GABA (99.5 Ϯ 27%).
We next examined whether agonist-induced internalization reduced the amount of receptors on the cell surface (Fig. 6E) by surface biotinylation experiments. Comparison of the relative surface/total ratio, in the absence or the presence of GABA (Fig.  6, E and F), showed that the ligand did not change significantly the amount of receptors on the surface. Expression is normalized to 1 for control and is 1.08 Ϯ 0.11 for GABA-treated cells and 1.06 Ϯ 0.12 for antagonist-treated cells; n ϭ 3 independent experiments (Fig. 6F). These data implied that GABA A R endocytosis may be compensated for by exocytosis or reinsertion of GABA A Rs. Indeed, inhibition of receptor recycling with monensin (55, 56) (Fig. 6G) reduced significantly the surface expression of GABA A Rs following application of 10 M muscimol. Expression is normalized to 1 for control and is 0.722 Ϯ 0.034 for muscimol-treated cells (80 cells for each condition, five independent experiments), suggesting that removal and insertion of GABA A Rs on the cell surface is a use-dependent process that is tightly regulated.

. ␥2(R43Q) internalization is inhibited by GABA A R antagonists.
A and B, neurons were transfected with ␥2 Myc or ␥2 Myc (R43Q) subunits. Surface labeling was obtained after immunostaining live neurons at 20°C with a polyclonal antibody directed at the extracellular Myc tag and Alexa Fluor-568conjugated secondary antibodies. Intracellular staining of the same cells was obtained after permeabilization and labeling with a monoclonal antibody directed at the same tag and FITC-conjugated secondary antibodies. A, merged images revealed the ␥2(R43Q) subunit on the surface of transfected neurons when cells were incubated with gabazine or picrotoxin at 37°C for 1 h prior to immunostaining. Scale bar, 10 m. B, quantitative analyses of the surface cluster area formed by ␥2 constructs (n ϭ 13-32). **, p Ͻ 0.005. C and D, COS-7 cells were transiently co-transfected with ␣1-, ␤3-, and Myc-tagged ␥2 subunits, either wild-type (␥2) or bearing the R43Q (␥2R43Q) substitution. Treatment and labeling were as in A, and the surface/total expression ratio (D) was calculated as the ratio of the surface Alexa Fluor-568 fluorescence level on live cells to FITC fluorescence on permeabilized cells (n ϭ 12-25). **, p Ͻ 0.005. E and F, COS-7 cells were transfected with ␣1-, ␤2-, and Myc-tagged ␥2 or ␥2(R43Q) subunits and incubated with or without picrotoxin and gabazine 30 min before surface biotinylation. E, representative Western blot shows Myc-tagged ␥2 and ␥2(R43Q) immunoreactivity in total or surface-biotinylated extracts from transfected cells. F, means of surface/total expression ratio from three independent experiments normalized to values of total ␥2 or ␥2(R43Q), as indicated, described in E show that treatment by picrotoxin and gabazine increases the surface-biotinylated fraction of cells expressing the ␥2 subunit bearing the R43Q mutation. *, p Ͻ 0.01. Error bars, S.E.

DISCUSSION
Modulation of surface stability of GABA A Rs is essential for regulating the physiological properties of inhibitory neurotransmission. Modification of inhibitory signaling and altered receptor trafficking has been associated with several neurological diseases, including schizophrenia, substance abuse, pain, or epilepsy (2, 11, 57-61). It is also established that GABA A Rs endocytosis is regulated through intracellular signaling pathways (1, 2, 4). Here we show that a mutation in the N-terminal domain and ligand application have both an influence on receptor endocytosis. Thus, our data uncover a mechanism that links the extracellular domain of GABA A Rs to their stability on the cell surface. It must be noted that it was possible that we had detected the traffic of recombinant ␥2 subunits in  monomer form because it has been shown in some heterologous cells transfected with this subunit alone. However, we have previously shown that this is unlikely (22). Furthermore, in the present work, we found that the amount of ␥2(R43Q) detected on the cell surface increased in the presence of a ␤2 subunit bearing the LL/AA substitution or following treatment with gabazine, a competitive antagonist of the GABA binding site. Because ␤2(LL/AA) should be associated with ␥2(R43Q) to increase cell surface labeling and the gabazine binding site is within the interface between ␣ and ␤ subunits (48), these exper-iments strongly suggest that ␥2(R43Q) detected on the cell surface is part of an oligomer also containing ␣ and ␤ subunits. The same holds true for wild type ␥2 subunit endocytosis promoted by GABA (or muscimol), the binding site of which is at the ␣/␤ interface (Fig. 4A).
Analysis of ␥2(R43Q) fate in neurons showed an increased retention in the endoplasmic reticulum, in agreement with previous findings in human embryonic kidney 293-T cells (24). Additionally, we provide evidence that ␥2(R43Q) is not entirely retained within the reticulum and is transported via FIGURE 6. GABA A R internalization in COS-7 cells is increased by GABA. A, COS-7 cells were transiently co-transfected with ␣1-, ␤2-, and Myc-tagged ␥2 subunits. Internalized labeling (red) was obtained after incubating live cells with polyclonal antibody directed at the extracellular Myc tag at 37°C followed by stripping of antibodies remaining on the surface. Intracellular staining (green) of the same cells was obtained after permeabilization and labeling with a monoclonal antibody directed at the same tag (white frames detailed in lower panels). Scale bar, 10 m. B, quantification of the area of intracellular punctate labeling of internalized receptors formed by ␥2 subunit in cells incubated in culture medium (Ctrl) with GABA, gabazine (GBZ), or picrotoxin (PTX). ***, p Ͻ 0.001. C, COS-7 cells transfected as in A were labeled with biotin and returned to 37°C in control medium, GABA, gabazine, or picrotoxin, as indicated. Biotin remaining on the surface after the incubation at 37°C was removed by cleaving. Total and internalized (biotinylated) ␥2 subunits were detected with an anti-Myc antibody. GAPDH staining shows that intracellular proteins were not biotinylated. D, means of internalized/ total expression ratio normalized to total ␥2 from experiments described in C. **, p Ͻ 0.005; *, p Ͻ 0.01. E, COS-7 cells transfected as in A and C and labeled with biotin. Representative Western blot shows Myc-tagged ␥2 in total or surface-biotinylated extracts. F, means of surface/total expression ratio normalized to total ␥2 from experiments described in E show that surface expression was unchanged by treatment with GABA. G, quantitative analysis of ␣1␤2␥2 GABA A Rs surface expression in COS-7 cells transiently co-transfected with ␣1-, ␤2-, and Myc-tagged ␥2 subunits and incubated with monensin with or without muscimol for 30 min before surface and total immunofluorescence labeling. A value of 1 for cells incubated without muscimol was used. **, p Ͻ 0.005. Error bars, S.E.
the Golgi apparatus to the cell membrane, where mutated receptors are highly internalized via a clathrin-and dynamindependent mechanism. Blockade of endocytosis leads to a major increase in ␥2(R43Q) surface targeting in neurons and COS-7 cells, indicating that internalization is a major mechanism for down-regulating cell surface expression of ␥2(R43Q)-containing receptors.
We also show that gabazine or picrotoxin increases dramatically the surface expression of ␥2(R42Q) subunit. These two GABA A R antagonists are both negative allosteric modulators, acting at different sites (48 -50, 62). The gabazineand picrotoxin-sensitive internalization of ␥2(R43Q) suggested that endocytosis could be linked to a constitutive activity of the mutated receptor. Electrophysiological recordings of ␥2(R43Q)-expressing cells clearly showed that this mutated subunit did not give rise to constitutive currents. Therefore, the effect of antagonists on ␥2(R43Q) endocytosis is probably related to another conformational state (e.g. the desensitized state). Interestingly, it has been shown that the ␥2(R43Q) mutation favors desensitized states (52).
Consequently, our findings, showing that GABA A R antagonists prevent ␥2(R43Q) endocytosis, suggest that internalization is driven by a global conformational change. Molecular models show that the ␥2Arg-43 residue is at the ␥2/␤2 interface in the extracellular domain, on a loop positioned above the pocket, which is homologous to the GABA binding sites. Interestingly, many mutations in nicotinic receptors linked to diseases are at the interface between receptor subunits (63); they alter the gating allosterically (i.e. from a distance) (63)(64). A model indicates that ␥2Arg-43 and ␥2Glu-178 are connected through a bifurcated salt bridge; this model has been used to study the ␥2(R43Q) mutation (22,26,52). One of these studies has suggested that these positions have a long range allosteric effect (52). In our new GABA A R model derived from the glutamate-gated chloride channel (42), the Arg-43 residue of the ␥2 subunit is connected to Tyr-174 and Glu-178 from the loop B and to the ␤2 subunit via polar interactions that should be sensitive to the R43Q substitution and positioned on a loop thought to be involved in the channel pore opening motion (51). Moreover, electrophysiological recordings and kinetic analyses have shown that the long distance effects of ␥2(R43Q) substitution extend as far as the transmembrane domains (52). Therefore, ␥2(R43Q) mutation might have an influence on receptor endocytosis in line with the current views on pentameric ligand-gated ion channels, describing a link between extracellular, transmembrane, and intracellular domains (53,65,66).
Because ligand binding in the Cys-loop receptor family is followed by a whole chain of interconnections, including the intracellular domain (67), it is of interest to assess whether GABA binding may influence GABA A R endocytosis. Although it is established that the number of surface GABA A Rs is regulated by constitutive endocytosis and neuromodulation through intracellular signaling (5-8, 68, 69), previous findings on ligandindependent or -dependent internalization are conflicting (28 -31). Several studies have investigated GABA A receptor internalization following agonist application (29,30). However, data were obtained by analyzing the amount of receptors remaining at the surface after agonist application. Here, we used a different approach (i.e. quantification of internalized receptors). Altogether, biochemical and immunocytochemical analyses of internalized receptor fraction, both in neurons and in COS-7, showed an increased number of internalized receptors during agonist application (Figs. 5B and 6, B and D), whereas the surface/intracellular ratio or surface labeling (Figs. 3, B and D, and 6E) remained unchanged. These data show that an overall counting of receptors on the cell membrane may overlook an increased endocytosis.
It has been suggested that internalization of GABA A Rs or increase in neuronal activity is accompanied by insertion of new receptors (4, 7, 70 -72). Here, all of the experiments performed on neurons were conducted in the presence of tetrodotoxin and glutamate receptor inhibitors, suggesting that agonist binding endocytosis associated with the insertion of new receptors should instead represent an additional homeostasis mechanism. Knowing that internalized receptors are recycled back to the surface membrane or targeted for degradation and that endocytosis may be compensated for by surface targeting of distinct receptor subtypes, this balance between agonist-induced removal and insertion of receptors may regulate the number, but also the identity, of GABA A Rs on the cell surface. Because the functional and pharmacological properties of GABA A Rs depend on subunit composition, our findings showing that ligand stimulation increased endocytosis imply that this mechanism may be an important process for a fine tuning of GABAergic neurotransmission (4). It is of note that benzodiazepines (allosteric modulators of GABA A Rs) induce a subtype-specific change via enhanced degradation rather than alterations in receptor insertion or endocytosis, thus revealing another mechanism that might regulate GABAergic neurotransmission (73).
The ␥2(R43Q) mutation is directly linked to epilepsy (16,17). Thus, our findings, revealing a direct relationship between receptor endocytosis and a neurological disorder, are in line with the emerging concept that GABA A R trafficking deficiencies are key factors in initiating and maintaining several diseases, including epilepsy (4,6,15). For example, status epilepticus leads to enhanced GABA A R endocytosis (28 -30). The K289M substitution in the ␥2 subunit known to be responsible for generalized epilepsy with febrile seizures plus alters the membrane diffusion of GABA A Rs (61). It must be also noted that a shortened lifetime caused by the epilepsy mutation A322D on ␣1-containing GABA A Rs has been proposed (74) (but also see Ref. 75). Interestingly, experiments on ␥2(R43Q) knock-in mice and transfected neurons or COS-7 cells have shown that ␣1 and ␣3 subunit surface expression was not reduced, despite a dramatic decrease in ␥2(R43Q) surface labeling (16,22). Earlier data, together with our present findings, show that the consequence of the mutation is a complex and dynamic process, suggesting that the defect is an active phenomenon instead of a more static retention of mutated receptors in the intracellular compartments and that ␥2(R43Q)-containing receptor internalization is associated with a compensatory insertion of distinct GABA A R subtypes.

CONCLUSION
␥2(R43Q)-containing GABA A Rs are in a conformational state that promotes internalization, providing evidence for a direct link between GABA A R endocytosis and epilepsy. Furthermore, our data suggest that GABA A R endocytosis is usedependent, consistent with a model in which ligand binding induces a conformation of the receptor that is a substrate for the biochemical events leading to endocytosis (70). Because GABA is the main inhibitory neurotransmitter in the brain and GABA A Rs are the target of many drugs, this property may have important functional and pathophysiological implications and should therefore be fully characterized (1,4). Our data suggest that the ␥2(R43Q) mutant is a useful model for this purpose. Our findings also illustrate the fact that mutations offer insights not only into diseases but also receptor physiology (61,64,74,76). It would be also of interest to assess whether the different allosteric drugs acting on GABA A Rs have an influence on receptor trafficking.