γ-Aminobutyric Acid Type A (GABAA) Receptor α Subunits Play a Direct Role in Synaptic Versus Extrasynaptic Targeting*♦

Background: GABAA receptor γ2 and δ subunits are thought to be responsible for synaptic and extrasynaptic targeting. Results: We demonstrate here that α2 and α6 subunits can target δ/γ2 chimeras to synaptic and extrasynaptic sites. Conclusion: The α subunits play a direct role in GABAA receptor targeting. Significance: Different subunits of GABAA receptors encode intrinsic signals to control subcellular targeting. GABAA receptors (GABAA-Rs) are localized at both synaptic and extrasynaptic sites, mediating phasic and tonic inhibition, respectively. Previous studies suggest an important role of γ2 and δ subunits in synaptic versus extrasynaptic targeting of GABAA-Rs. Here, we demonstrate differential function of α2 and α6 subunits in guiding the localization of GABAA-Rs. To study the targeting of specific subtypes of GABAA-Rs, we used a molecularly engineered GABAergic synapse model to precisely control the GABAA-R subunit composition. We found that in neuron-HEK cell heterosynapses, GABAergic events mediated by α2β3γ2 receptors were very fast (rise time ∼2 ms), whereas events mediated by α6β3δ receptors were very slow (rise time ∼20 ms). Such an order of magnitude difference in rise time could not be attributed to the minute differences in receptor kinetics. Interestingly, synaptic events mediated by α6β3 or α6β3γ2 receptors were significantly slower than those mediated by α2β3 or α2β3γ2 receptors, suggesting a differential role of α subunit in receptor targeting. This was confirmed by differential targeting of the same δ-γ2 chimeric subunits to synaptic or extrasynaptic sites, depending on whether it was co-assembled with the α2 or α6 subunit. In addition, insertion of a gephyrin-binding site into the intracellular domain of α6 and δ subunits brought α6β3δ receptors closer to synaptic sites. Therefore, the α subunits, together with the γ2 and δ subunits, play a critical role in governing synaptic versus extrasynaptic targeting of GABAA-Rs, possibly through differential interactions with gephyrin.

swap the fragment containing the large intracellular loop (IL), the TM4, and the C terminus of the ␦ subunit with the corresponding one of ␥2S (␥2ILTM4). In the first step, cmyc-␥2S in pCDNA3.1 was used as a template in a PCR to amplify a 447-bp fragment. In the second step, this fragment was combined with methylated full-length cmyc-␦ (in pcDNA3.1) as a template, using the Invitrogen GeneTailor TM site-directed mutagenesis system. The ␦/␥2ILTM4 chimeric subunit was sequenced, and the full-length chimeric open reading frame fragment was amplified by PCR using specific primers containing 5Ј-NheI and 3Ј-XhoI sites and inserted into pCDNA 3.1 to exclude possible modifications of pcDNA3.1 during the previous steps, followed by sequencing.
To construct the ␣6 ␣2IL chimera, the ␣6 subunit coding sequence was amplified by PCR from pCMVneo-␣6 and inserted into pcDNA3.1 ϩ between BamHI and XbaI sites. The coding sequence of the ␣2 subunit IL (amino acids 307-391 of the mature polypeptide) was amplified from the ␣2 construct and inserted between the two EcoRI sites just outside the coding region of the ␣6 subunit IL (amino acids 306 -400). For the ␣2 ␣6IL chimera, the rat ␣2 subunit coding sequence was amplified from the ␣2 construct using HindIII and ApaI restriction site-containing primers and inserted into pcDNA3.1 ϩ between HindIII and ApaI sites. An EcoRI site was engineered just upstream of the IL coding region through synonymous mutagenesis at amino acid 303. The EcoRI and XbaI sites around amino acid 16 and 17 were eliminated in the same way to ensure successful insertion of the ␣6 subunit IL-coding sequence. The coding sequence of the ␣6 subunit IL was amplified and inserted between the engineered EcoRI site and an XbaI site just downstream from the ␣2 subunit IL coding region. For both constructions, the sequences outside the ILs were not changed.
␦ GBS and ␣6 GBS chimeras were constructed in pCMVneo by insertion of the 18-amino acid gephyrin-binding site (GBS) of the glycine receptor ␤ subunit (38) into the IL of the ␦ subunit (after amino acid 341 of the mature polypeptide) and the IL of the ␣6 subunit after amino acid 340. Two DNA fragments, one including coding sequences of the target protein from the N terminus to the insertion site and the other including that from the insertion site to the C terminus, were amplified from pCMVneo-␣6 and pCM-Vneo-␦. Both fragments also contained partial and overlapping insertion sequences and were fused into one fragment by PCR. The resulting fragment was inserted back into pCM-Vneo vector.
The murine HA-tagged NL2A expression vector (pNiceNLG-2, referred to as NL2 in this work) was obtained from Dr. P. Scheiffele (University of Basel) (39). The HA tag was inserted between the signal peptide and the N terminus of the mature protein. The gephyrin-GFP construct encodes human gephyrin with GFP fused to the C terminus of gephyrin (40). The collybistin constructs encode two isoforms of human collybistin: CB3 SH3ϩ /hPEM2 SH3ϩ containing the SH3 domain and CB3 SH3Ϫ /hPEM2 SH3Ϫ lacking the SH3 domain (41).
Electrophysiology-Whole-cell recordings were performed in voltage clamp mode by using Multiclamp 700A amplifier (Molecular Devices, Palo Alto, CA) as described before (42). Patch pipettes were pulled from borosilicate glass and fire-pol-ished to a resistance of 3-6 megohms. The recording chamber was continuously perfused with a bath solution containing 128 mM NaCl, 30 mM glucose, 25 mM HEPES, 5 mM KCl, 2 mM CaCl 2 , and 1 mM MgCl 2 (pH 7.3, adjusted with NaOH, ϳ320 mosM). The pipette solution contained 135 mM KCl, 10 mM HEPES, 2 mM EGTA, 10 mM Tris-phosphocreatine, 4 mM MgATP, 0.5 mM Na 2 GTP (pH 7.3, adjusted with KOH, ϳ300 mosM). Data were acquired using the pCLAMP9 software (Molecular Devices), sampled at 5 kHz, filtered at 1 kHz, and analyzed with Clampfit 9.0 (Molecular Devices). Drugs were applied through a fast drug application system (VC-6; Warner Instruments, Hamden, CT) to assess the pharmacological properties of the reconstituted GABA A -Rs, as indicated by the rapid rise phase of whole-cell GABA and THIP currents in the pharmacological study (Fig. 1). Spontaneous IPSCs were recorded with normal bath perfusion. Spontaneous events were analyzed by MiniAnalysis software (Synaptosoft). The 20 -80% rising time and the weighted time constant ( weighted ϭ (1 ϫ A1 ϩ 2 ϫ A2)/(A1 ϩ A2)) of the IPSCs were analyzed to compare the kinetics of the events. Pooled data were presented as means Ϯ S.E., and n represents the number of the cells recorded. One-way ANOVA was employed to analyze multiple groups of data, followed by Bonferroni's pairwise comparison.
Ultrafast GABA application and outside-out patch recording were employed to assess the onset kinetics of GABA A -Rs composed of different subunits. The ultrafast drug application system (ALA Inc., Long Island, NY) consists of solution reservoirs, manual switching valves, a solenoid-driven four-way pinch valve, and two tubes (inner diameter 500 m) oriented at 50°for rapid solution exchange (43,44). One tube contains normal bath solution and the other contains 10 mM GABA to maximally activate GABA A -Rs. The solution exchange rate was estimated to be within 1 ms (20 -80% rise time), using an open tip electrode to detect the junction potential caused by different salt concentrations (75 mM versus 150 mM NaCl). Typically, six pulses of GABA were applied to each patch. The duration of GABA application was sufficient (200 -500 ms) to reach the peak current value. Data were sampled at 10 kHz and low pass filtered at 4 kHz (8-pole Bessel filter). Individual traces were aligned and averaged, and the 20 -80% rising time was analyzed with MiniAnalysis software.
For the electron microscopy experiments, HEK cells were co-transfected with the following: 1) NL2, ␣6, ␤3, and ␦; 2) NL2, ␣2, ␤3, and ␥2-GFP and co-cultured with hypothalamic neurons for 2 days. The cells were briefly fixed with 4% paraformaldehyde ϩ 0.05% glutaraldehyde (10 min at room temperature followed by 20 min in 4°C), quenched in 0.15% glycine for 10 min, and incubated in blocking solution (3% normal goat serum plus 2% normal donkey serum in bath solution) for 1 h at 4°C. Primary antibodies were diluted in blocking solution (rabbit anti-␦-Nterm (1:100); rabbit anti-GFP (1:200, Invitrogen)) and applied to the samples at 4°C overnight. The cells were then incubated with secondary antibodies (1.4 nM Nanogold goat anti-rabbit (1:50; Nanoprobes, Yaphank, NY)) for 1 h at room temperature, fixed with 1% glutaraldehyde for 20 min, and processed with the HQ silver enhancement kit (Nanoprobes, Yaphank, NY) according to the instructions. After developing with the silver enhancer, the cells were submerged in 2% glutaraldehyde, scraped off from the coverslips, and centrifuged at 8000 relative centrifugal force for 10 min to collect the cells. The pellets were further fixed with 2% glutaraldehyde for 1 h at room temperature before EM processing. The cell pellets were post-fixed in 1% OsO 4 for 1 h. The cells were then dehydrated in a serial of graded ethanol solutions and embedded in Eponite 12. Thin sections (80 nm) were cut with a Leica UC6 ultramicrotome, contracted with uranyl acetate and lead citrate, and examined in a TEM JEOL JEM 1200 EXII (Peabody, MA) at 80 kV. Hetero-synapses were identified by nerve terminals (filled with synaptic vesicles) apposing HEK cells that showed immunogold puncta on the plasma membrane. The edge of synapses was defined as the point where the plasma membrane of nerve terminals starts to diverge from HEK cell membrane. The localization of silver-enhanced gold labeling of GABA A -Rs was characterized into three categories as follows: 1) synaptic, inside a synapse, and more than 30 nm away from the edges; 2) perisynaptic, less than 30 nm away from the synaptic edges; and 3) extrasynaptic, outside synapses, and over 30 nm away from the edges of synapses (8).
Co-immunoprecipitation-HEK cells were transfected with either ␦ or ␦ GBS together with gephyrin-GFP. Gephyrin-GFP single transfection served as the control. Rabbit anti-␦-Nterm was used for the immunoprecipitation, and rabbit anti-GFP was used in the immunoblotting to detect the gephyrin-GFP.

RESULTS
Distinct Pharmacological Properties of Heterologously Expressed GABA A -Rs-Neurons express a broad spectrum of GABA A -Rs composed of different subunits, making it difficult to identify the critical factors important for the targeting of a specific receptor subtype. We therefore employed our recently established hetero-synapse system to investigate the targeting of different subtypes of GABA A -Rs (34). When HEK cells were transfected with GABA A -R subunits and a cell adhesion molecule neuroligin-2 (NL2) and then co-cultured with neurons, both spontaneous and action potential-evoked GABAergic events were detected (34). With this system, we can precisely control the subunit composition of GABA A -Rs and their potential interacting proteins to investigate the targeting mechanism of GABA A -Rs.

Distinct Kinetic Properties of GABAergic Events Mediated by Different Subtypes of GABA A -Rs-
We previously demonstrated that NL2-transfected HEK cells receive GABAergic innervation from surrounding neurons in the HEK cell neuron co-culture system (34). Orthogonal views of Z-stack confocal images showed GABAergic terminals labeled by GAD staining (green) wrapping around a transfected HEK cell ( Fig. 2A). Interestingly, GAD puncta were found not only at the bottom of the HEK cell, where the initial contact with neurons took place, but also on side and top surfaces of HEK cells. This observation suggests that following initial contact with transfected HEK cells, neuronal axons have ramified to innervate large portions of the cell surface.
Rapid Onset Kinetics of GABA A -Rs Composed of Different Subunits-We wondered whether the onset kinetics of different receptors might explain such a drastic difference in the IPSC rise phases. To answer this question, we employed a high speed solution exchange system to apply GABA (10 mM) to outside-out patches excised from transfected HEK cells (Fig.  4A). Ultrafast GABA application was achieved by starting GABA perfusion and stopping bath solution simultaneously. We found that ␣2␤3␥2-Rs were activated rapidly upon GABA application (Fig. 4, B and G, T 20 -80%Rise ϭ 1.0 Ϯ 0.2 ms, n ϭ 8), faster than the rise phase of ␣2␤3␥2-mediated IPSCs in HEK cells but comparable with neuronal IPSCs. This result suggests that GABA A receptors in HEK cells are not clustered as tightly as in neuronal cells. The rise phase of ␣6␤3␥2-Rs was indistinguishable from that of ␣2␤3␥2-Rs (Fig. 4, B and G, T 20 -80%Rise : ␣6␤3␥2 ϭ 1.0 Ϯ 0.2 ms, n ϭ 8; p Ͼ 0.9). However, the rise phase of ␣6␤3␦-mediated GABA currents was significantly slower than that of ␣2␤3␥2 or ␣6␤3␥2 receptors (Fig. 4, B and G, T 20 -80%Rise : ␣6␤3␦ ϭ 2.3 Ϯ 0.3 ms, n ϭ 10; p Ͻ 0.01 for both comparisons, one-way ANOVA followed by Bonferroni's pairwise comparison), yet it was still an order of magnitude faster than that of ␣6␤3␦-IPSCs in HEK cells. Because the difference in receptor kinetics is too small to explain the drastic 10-fold difference between the rise phase of ␣2␤3␥2 and ␣6␤3␦ IPSCs, the slow ␣6␤3␦-IPSCs is likely a result of the extrasynaptic localization of ␣6␤3␦ receptors.
Ultrastructural Localization of GABA A -Rs-We further carried out immunoelectron microscopic studies to reveal the ultrastructural localization of ␣6␤3␦ and ␣2␤3␥2 receptors in neuron-HEK cell co-cultures. HEK cells expressing ␣6␤3␦ or ␣2␤3␥2 receptors were identified by silver-enhanced gold par-ticles immunolabeling the ␦ or myc ␥2 subunit. Nerve terminals containing synaptic vesicles were found in close contact with HEK cells. Importantly, gold particles immunopositive for ␦ receptors were localized mostly at extrasynaptic membranes, whereas ␥2-positive particles were mainly at synaptic cleft (Fig.  5, A and B). To quantify the detailed receptor localization, 7 randomly selected sections with a total of 34 synapses and 55 gold particles labeling ␦-receptors were analyzed. The majority of particles (80%) were localized at extrasynaptic membranes, whereas only 12.7 and 7.3% were localized perisynaptically or synaptically (Fig. 5C). For comparison, 6 sections containing 18 synapses from ␣2␤3␥2-expressing HEK cells were assessed. Among 81 ␥2-immunoreactive particles analyzed, 63% were synaptic and 16% perisynaptic, with the remaining 21% being extrasynaptic (Fig. 5C). The immuno-EM results confirmed that the ␣6␤3␦ GABA A -Rs are preferentially localized at extrasynaptic membranes in the hetero-synapse model. Together with the kinetics analysis (Figs. 2-4), the IPSC rise phase seems to be a faithful indicator of receptor localization in our heterosynapse model; fast rise phase indicates synaptic localization, and slow rise phase indicates extrasynaptic or perisynaptic localization.
The targeting of ␣6 GBS and ␦ GBS subunits was further analyzed in neurons co-transfected with ␣6 GBS , ␤3, and ␦ GBS subunits. Transfected neurons were double immunolabeled to visualize the co-localization of the ␦ subunit and gephyrin or the ␦ subunit and GAD. As control, neurons transfected with ␣6␤3␦ or ␣2␤3␥2 were also examined. We found that ␦ subunitcontaining receptors were diffusely localized throughout the neuronal membrane surface, without obvious enrichment at synaptic sites apposed to GAD-labeled presynaptic terminals (Fig. 9A). By contrast, the immunostaining of the ␥2 subunit revealed punctate labeling along the dendrites, with many clusters juxtaposed to GAD puncta (Fig. 9B). Intriguingly, neurons overexpressing ␦ GBScontaining receptors showed punctate staining, which was also co-localized with punctate gephyrin staining (Fig. 9C). More importantly, some of the ␦ GBS -containing puncta were found jux- FIGURE 8. Interaction with gephyrin targets ␣6␤3␦-containing GABA A -Rs to synaptic sites of reconstituted synapses. A, schematic representation of the ␦ GBS and ␣6 GBS chimeras. B, GFP-gephyrin (geph) was co-precipitated with the ␦ GBS chimera. C, ␣6␤3␦ GBS and ␣6 GBS ␤3␦ receptors were found to co-localize with gephyrin-GFP in big intracellular aggregates when co-expressed in HEK cells, demonstrating the interaction between ␦ GBS and ␣6 GBS subunits and gephyrin; ␣6␤3␦ receptors showed no co-localization with gephyrin-GFP. Scale bar, 10 m. D, whole-cell GABA current in HEK cells expressing the ␦ GBS and/or ␣6 GBS chimeras. E, sample traces of sIPSCs mediated by ␣6␤3␦ or ␣6␤3␦ GBS receptors. The events were scaled to the same amplitude and aligned according to the initial rise time. F, pooled data of sIPSC rise time in different groups. The ␣6␤3␦ GBS receptor-mediated sIPSCs showed a rise phase significantly faster than that of ␣6␤3␦ receptors. Co-expression of gephyrin or gephyrin plus collybistin did not further change the IPSC rise time. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001 (one-way ANOVA followed by Bonferroni's pairwise comparison).
taposed to GAD puncta (Fig. 9D), suggesting that these GBS-containing chimeric receptors were recruited to postsynaptic sites in neurons, likely through interaction with gephyrin.

DISCUSSION
In this study, we demonstrate that different subtypes of GABA A -Rs are distinctly targeted to synaptic and extrasynaptic sites in neuron-HEK cell hetero-synapses. With this unique synapse model, we found that ␣2 and ␣6 subunits target the same ␦/␥2 chimeric subunit to synaptic and extrasynaptic sites, respectively, suggesting a direct role of ␣ subunits in GABA A -R targeting. Furthermore, forced interaction of the ␣6 or ␦ subunit with gephyrin can recruit normally extrasynaptic ␣6␤3␦ receptors closer to synaptic sites, suggesting that gephyrin can stabilize any interactive GABA A -Rs at synaptic sites. Fig. 10 is a schematic diagram illustrating the relative subcellular localiza- FIGURE 9. Incorporation of the gephyrin-binding site induces clustering of ␣6␤3␦-receptors at postsynaptic sites in neurons. A, hypothalamic neurons co-transfected with ␣6, ␤3, and ␦ subunits were double immunolabeled for the ␦ subunit and GAD. Immunoreactivity for the ␦ subunit was diffusely localized on the neuronal surface. B, neurons were co-transfected with ␣2, ␤3, and myc ␥2 subunits, followed by GAD and surface myc ␥2 double staining. The ␥2 subunit-containing receptors formed puncta along the dendrites, some of which were apposed to GAD puncta. C and D, neurons transfected with ␣6 GBS , ␤3, and ␦ GBS subunits were double immunolabeled for the ␦ GBS subunit and gephyrin (Geph) (C) or ␦ GBS and GAD (D). The ␦ GBS subunit-containing receptors formed clusters in neurons. A portion of the ␦ GBS subunit-containing receptors were co-localized with gephyrin clusters (right panels in C) or with GAD puncta (right panels in D), suggesting a synaptic localization.
tions of different subtypes of GABA A -Rs investigated in this study. Importantly, the intermediate rise and decay phases of ␣6␤3and ␣6␤3␥2-IPSCs suggest that these receptors are most likely localized at perisynaptic sites, different from the synaptic ␣2␤3␥2 or extrasynaptic ␣6␤3␦ receptors. Such distinct IPSC events with graded changes of rise and decay phases are difficult to distinguish in neurons, underscoring the advantage of our model synapses in pinpointing the precise targeting mechanisms of specific subtype receptors.
Molecularly Engineered Synapses as a Model System to Study Receptor Targeting-The hetero-co-culture system is often used to study synaptogenesis induced by cell adhesion molecules, such as neuroligins, SynCAM, netrin-G ligand, and LRRTM (39,(55)(56)(57)(58)(59)(60)(61). We have previously shown that functional GABAergic synapses can be formed in HEK 293T cells by coexpressing NL2 and ␣2␤3␥2 GABA A -Rs (34). Here, we further developed the hetero-synapses as a model system to study GABA A -R targeting. The advantage of this system is the precise control of the expression of specific receptor subtypes, avoiding the complexity of GABA A -Rs in neurons. For example, if a neuron contains both ␣2␤3␥2 and ␣6␤3␥2 receptors, it will be difficult to know whether recorded IPSCs are mediated by ␣2␤3␥2 or ␣6␤3␥2 receptors or both. Our model synapses offer clear distinction between synaptic events mediated by ␣2␤3␥2 and ␣6␤3␥2 receptors (Fig. 3), providing an important research tool for future studies on different subtypes of receptors. Furthermore, we have recently demonstrated that such a model system is useful for the screening of human disease-related gene mutations by co-expressing GABA A -Rs with wild type or mutant NL2 identified from patients with schizophrenia (62). Our previous and current studies suggest that molecularly engineered hetero-synapses are a versatile model system that can be used to study not only synaptogenesis but also receptor targeting and functional deficits of gene mutations.
␣ Subunits Are Sufficient to Target GABA A -Rs-Previous studies suggest that ␥2 subunit-containing GABA A -Rs are mainly concentrated at postsynaptic sites (28 -30), whereas ␦ subunit-containing GABA A -Rs are mostly distributed in extrasynaptic membranes (2,5,7,8,31,63). Based on the present analyses of ␦/␥2 chimeras, it seems that there is no single domain in the ␦ subunit responsible for the slow IPSC kinetics, because the rise phases became increasingly slower with chimeras containing a greater portion of the ␦ subunit (Fig. 6B). As for the role of different ␣ subunits, recent studies found that targeted deletion of ␣1, ␣2, or ␣3 subunit abolishes ␥2-containing postsynaptic receptor clusters in selective subcellular regions (24 -27). Conversely, deletion of ␣4, ␣5, or ␣6 subunit greatly reduced tonic currents, suggesting an extrasynaptic localization (64 -66). These knock-out experiments suggest that the ␣ subunit is required for functional assembly of synaptic (␣1-3)and extrasynaptic (␣4 -6)-GABA A -Rs, but they did not address whether the ␣ subunit is involved in receptor targeting.
In this work, we directly investigated the role of ␣2 and ␣6 subunits in GABA A -R targeting. We first observed a slower rise phase of ␣6␤3␥2-IPSCs than that of ␣2␤3␥2-IPSCs. Similarly, ␣6␤3-IPSCs were also slower than ␣2␤3-IPSCs, a clear indication of differential functions of the two ␣ subunits. The direct role of the ␣ subunit in receptor targeting was discovered by co-assembling with a series of ␦/␥2 chimeras. We demonstrated that when combined with the ␣2 subunit, the ␦/␥2 chimeras always mediated fast IPSCs, similar to that mediated by synaptic ␥2-containing receptors; but when combined with the ␣6 subunit, the same ␦/␥2 chimeras always mediated very slow IPSCs, reminiscent of that by extrasynaptic ␦-containing receptors (Fig. 6). Because the ␣2 and ␣6 subunits do not affect the onset kinetics of GABA A -Rs (Fig. 4), the drastic difference in IPSC rise phase likely reflects the difference in receptor localization. Thus, the same ␦/␥2 chimera can be targeted to either synaptic or extrasynaptic membrane, depending on the ␣ subunit with which it is co-assembled. These experiments suggest that different ␣ subunits directly play a targeting role in guiding GABA A -Rs to synaptic versus extrasynaptic sites.
Gephyrin and GABA A -R Targeting-Synaptic GABA A -Rs are thought to be first inserted to extrasynaptic membranes and then laterally diffused into postsynaptic sites, where they are stabilized by the scaffolding protein complex (40,50,67,68). Both knock-out and knockdown of gephyrin expression disrupted the clustering of a major subset of synaptic GABA A -Rs and resulted in decreased GABAergic neurotransmission (28,40,50,69,70). However, not all GABA A -R clusters are dependent on gephyrin (70,71). For example, ␣1-containing receptors in pyramidal neurons are likely stabilized by the dystrophin-glycoprotein complex (27).
The ␣1-3 subunits, but not the ␣6 subunit, have been shown to directly bind with gephyrin through their large IL (48,52,53). By swapping the IL domain between ␣2 and ␣6 subunits, we generated ␣2 ␣6IL and ␣6 ␣2IL chimeras to test their targeting role. However, the ␣6 ␣2IL ␤3␥2-IPSCs did not show faster kinetics but rather slightly slower than the IPSCs mediated by ␣6-containing receptors (Fig. 7). Thus, the ␣2 IL domain alone is not sufficient for the synaptic targeting of GABA A -Rs. In agreement with our finding, a recent study showed that the interaction between GABA A -R ␣2 subunit and gephyrin is much weaker than that between GlyR ␤ subunit and gephyrin (54).
We hypothesized that the extrasynaptic localization of ␣6␤3␦ receptors is due to the lack of interaction with gephyrin. To test this hypothesis, we inserted a high affinity gephyrinbinding site into the IL domain of ␣6 and ␦ subunits to force an interaction with gephyrin (38,72). We showed that ␣6␤3␦ GBS IPSCs in HEK cells (with or without gephyrin overexpression) have faster kinetic properties than ␣6␤3␦ IPSCs, suggesting that the ␦ GBS subunit-containing receptors are localized closer to synaptic sites than native ␦ subunit-containing receptors (Fig. 8). Furthermore, immunostaining in neurons demonstrated that ␣6 GBS ␤␦ GBS receptors form clusters that co-localized with gephyrin and GAD at synapses (Fig. 9). These results suggest that forced interaction with gephyrin is capable of bringing extrasynaptic ␣6␤3␦ receptors close to synaptic sites.
To our surprise, gephyrin or collybistin co-expression was not required for the ␦ GBS -Rs to mediate faster IPSC events (Fig.  8F). We hypothesize that endogenous gephyrin in HEK cells is sufficient to interact with ␦ GBS and target the receptors closer to synaptic membranes. Indeed, we found that a subpopulation of HEK cells expressed a high level of gephyrin, although the rest showed a low level of expression. Interestingly, the HEK cells expressing high levels of gephyrin usually showed compact chromatin structures as revealed by DAPI staining (supplemental Fig. 1). Because gephyrin is a microtubule-binding protein, we suspect that such a high level of expression might indicate a potential role of gephyrin during cell division, which is worthy of future study but is beyond the scope of this work.
Besides GABA A -Rs, recent studies suggest that collybistin and NL2 also interact with gephyrin (1). NL2 has been suggested to interact with gephyrin and collybistin to target GABA A -Rs to perisomatic membranes (73). NL2 overexpression may also change GABA A -R subunit composition as shown in cerebellar granule cells (74). Collybistin can facilitate gephyrin localization to submembrane sites (75) and increase synaptic GABA A -R accumulation (76). Collybistin deficiency results in region-specific loss of gephyrin and a subset of GABA A -Rs, as well as altered synaptic plasticity and increased levels of anxiety (77,78). Moreover, collybistin and gephyrin may form a complex that is particularly important for interaction with the ␣2 subunit (79). In this study, we have co-expressed collybistin (CB3 SH3ϩ or CB3 SH3Ϫ ) with ␣6␤3␦ GBS and gephyrin as well as NL2 in HEK cells. Interestingly, GABA current amplitudes were increased by collybistin (data not shown), but the IPSC kinetics were not changed. This may suggest that collybistin contributes to GABA A -R trafficking to the membrane surface but does not affect receptor localization.
Conclusion-Our studies suggest that different GABA A -R subunits encode intrinsic targeting information, and the subcellular localization of a particular subtype of receptor is determined by the integral effect of not only the ␥2 and ␦ subunits but also different ␣ subunits (e.g. ␣2 versus ␣6 subunit). Thus, ␣ subunits not only are required for the assembly of functional receptors but also carry a direct targeting signal for subcellular localization. Our hetero-synapse system provides a unique model for further studying the targeting mechanisms of GABA A receptors with a variety of subunit partnership.