GABAA Receptor α and γ Subunits Shape Synaptic Currents via Different Mechanisms*

Background: GABAAR α2 and γ1 subunits are highly expressed in amygdala but their influence on synaptic currents is unknown. Results: α2 subunits increased GABA affinity thereby slowing current deactivation; γ1 subunits reduced synaptic receptor clustering. Conclusion: These subunits may differentially shape synaptic kinetics. Significance: Understanding how α2 and γ1 subunits shape synaptic currents may help us understand amygdala processing mechanisms. Synaptic GABAA receptors (GABAARs) mediate most of the inhibitory neurotransmission in the brain. The majority of these receptors are comprised of α1, β2, and γ2 subunits. The amygdala, a structure involved in processing emotional stimuli, expresses α2 and γ1 subunits at high levels. The effect of these subunits on GABAAR-mediated synaptic transmission is not known. Understanding the influence of these subunits on GABAAR-mediated synaptic currents may help in identifying the roles and locations of amygdala synapses that contain these subunits. Here, we describe the biophysical and synaptic properties of pure populations of α1β2γ2, α2β2γ2, α1β2γ1 and α2β2γ1 GABAARs. Their synaptic properties were examined in engineered synapses, whereas their kinetic properties were studied using rapid agonist application, and single channel recordings. All macropatch currents activated rapidly (<1 ms) and deactivated as a function of the α-subunit, with α2-containing GABAARs consistently deactivating ∼10-fold more slowly. Single channel analysis revealed that the slower current decay of α2-containing GABAARs was due to longer burst durations at low GABA concentrations, corresponding to a ∼4-fold higher affinity for GABA. Synaptic currents revealed a different pattern of activation and deactivation to that of macropatch data. The inclusion of α2 and γ1 subunits slowed both the activation and deactivation rates, suggesting that receptors containing these subunits cluster more diffusely at synapses. Switching the intracellular domains of the γ2 and γ1 subunits substantiated this inference. Because this region determines post-synaptic localization, we hypothesize that GABAARs containing γ1 and γ2 use different mechanisms for synaptic clustering.

is known about the impact of ␥1-containing GABA A Rs on inhibitory synaptic transmission.
Here we describe the kinetic and synaptic properties of GABA A Rs containing ␣2 and ␥1 subunits and compare them to those containing ␣1 and ␥2 subunits. By providing new insights into the functional properties of ␣2and ␥1-containing GABA A Rs, our study facilitates investigations into whether these GABA A Rs contribute to synaptic currents in brain regions that mediate anxiety-related disorders such as fear, depression, and post-traumatic stress.

EXPERIMENTAL PROCEDURES
Cell Culture and Molecular Biology-Human ␣1 (pCIS2), ␣2 (pCIS2 or pcDNA3.1), ␤2 (pcDNA3.1ϩ or pcDNA3.1Zeo), ␥1 (pcDNA3.1ϩ), and ␥2L (pcNDA3.1ϩ) subunits were transfected in a subunit plasmid ratio of 1␣:1␤:3␥ (total DNA was 0.2-2.0 g), into HEK293 cells using Ca 2ϩ phosphate-DNA coprecipitation. This transfection ratio ensured the incorporation of the ␥ subunit into the receptors. GABA A Rs comprised only of ␣ and ␤ subunits were produced by transfecting these subunits at a plasmid ratio of 1:1. Cotransfecting the neuroligin splice variant neuroligin 2A (with HA tag), which was obtained from Addgene (USA) (22), facilitated the formation of heterosynapses. Enhanced GFP and CD4 were also transfected and acted as expression markers. Interchanging the intracellular domain (ID) and fourth transmembrane domain (TM4) domain of one ␥ subunit isoform with the other produced two ␥ subunit chimeras, which were transfected with ␣2 and ␤2 subunits. The two ␥ subunit chimeras were: 1) the ␥2L-␥1, which expresses the ␥2L subunit sequence from the N terminus up to the end of TM3 (up to Leu-317) and the ID and TM4 of the ␥1 subunit sequence (from His-320), and 2) the ␥1-␥2L, which contains the ␥1 sequence from the N terminus to the end of TM3 (up to Leu-319) and the ID and TM4 of the ␥2L sequence (from His-318). In a separate set of transfections we co-transfected the ␣2-containing GABA A Rs along with rat gephyrin (with and without an N terminus GFP tag), and the human collybistin homologue, hPEM.
Primary neuronal cultures were prepared using standard protocols (23). The cortices of E18 rat embryos were triturated and plated at ϳ80,000 cells per 18-mm poly-D-lysine-coated coverslip in DMEM with 10% fetal bovine serum. After 24 h the entire medium was replaced with Neurobasal medium including 2% B27 and 1% GlutaMAX supplements; a second feed after 1 week replaced half of this medium. Neurons were grown for 3 to 5 weeks in vitro and the heterosynapse co-cultures were prepared by directly introducing transfected HEK293 cells onto the primary neuronal cultures. Recordings of synaptic currents were done 1-3 days later.
Immunofluorescent Labeling-Coverslips with cells were fixed for 5-10 min in 4% paraformaldehyde in phosphate-buffered saline, then blocked and permeabilized in 3% bovine serum albumin with saponin (0.05%) for 30 min. HA-tagged neuroligin 2A was labeled with rabbit anti-HA (Santa Cruz, 1/100) and GABAergic terminals were labeled for the GABA synthesizing enzyme GAD65 (mouse anti-GAD65, Chemicon/Millipore, 1/10,000). Primary antibodies were added to blocking solution overnight at room temperature, the cells were washed and sec-ondary antibodies were applied at 1/500 for 30 min. Coverslips were mounted using DAKO fluorescent mounting medium and imaged on upright fluorescent and confocal microscopes.
Electrophysiology-All experiments were performed at room temperature in either the whole cell or outside-out patch configuration of the patch clamp technique, at a holding potential of Ϫ70 mV. The intracellular solution was composed of (in mM): 145 CsCl, 2 CaCl 2 , 2 MgCl 2 , 10 HEPES, and 10 EGTA, adjusted to pH 7.4 with CsOH. Cells and patches were continuously perfused with extracellular solution made up of (in mM): 140 NaCl, 5 KCl, 2 CaCl 2 , 1 MgCl 2 , 10 HEPES, and 10 D-glucose, adjusted to pH 7.4 with NaOH. The liquid junction potential between the intra-and extracellular solutions was calculated to be 4.0 mV (24). A double-barreled glass tube was mounted onto a piezo-electric translator (Siskiyou) to achieve rapid solution exchange (Ͻ1 ms) over outside-out patches by lateral movement of the glass tube. Synaptic currents were filtered (Ϫ3 dB, 4-pole Bessel) at 4 kHz and sampled at 10 kHz, whereas the macropatch recordings were filtered at 10 kHz and sampled at 30 kHz. Synaptic and macropatch data were recorded using a Multiclamp 700B amplifier and pClamp 9 software. Single channel currents were recorded using an Axopatch 200B amplifier, pClamp 10 software, filtered at 10 kHz and sampled at 50 kHz. Current traces were filtered off-line at 5 kHz for making figures.
Stock solutions of flunitrazepam and diazepam were kept frozen and diluted to the desired concentration in extracellular solution on the day of recording. Typically, at least 3 min of spontaneous activity was recorded before and during drug application. To preserve network activity for spontaneous recordings, the drug solution was targeted to the recorded cell, whereas the extracellular solution was washed over the surrounding area. Drug washout was obtained in about half of the cells recorded, and was averaged with the baseline data to minimize time-dependent effects.
Analysis-Data are presented as mean Ϯ S.E. Exponential equations were fit to the rising phase (10 -90%) and current decay (weighted double-or mono-exponentials) of macropatch and synaptic currents as previously described (7) using Axograph X. Each current from a recorded cell or patch was analyzed separately and then averaged for that record. These averages were then pooled into data sets, from which means were calculated. Currents containing double events or artifacts in current rise and decay were manually excluded. Current-voltage (I-V) experiments were done by measuring single channel current amplitude at the corresponding voltage, for voltages of (in mV): Ϯ70, Ϯ35, Ϯ15, and 0. The current reversal potential was read directly from the I-V plots.
Single channel kinetic analysis was done using QuB software. Current records were idealized at a cut-off resolution of 70 s. The idealized records were then divided into discrete, single channel active periods by applying a tcrit shut duration. Tcrit values were determined for each patch and selected so as to retain the three briefest shut components (common to all records) in the dwell distributions as previously outlined (7,25). Clusters (3 mM GABA) and bursts (2 M GABA) of activity were accepted for deriving an activation mechanism if they contained Ͼ10 or 3 events, respectively (for estimating the mean burst duration at 2 M GABA, bursts that contained Ն2 events were also included). This resulted in open dwell distributions that were also composed of three components, when fitted using the "star" function in QuB. Three shut and three open components were taken to represent the minimum number of corresponding states for constructing activation schemes. Mechanisms were then postulated and used to generate fits to the dwell distributions by maximum likelihood fitting (26,27). The procedure optimized the rate constants and produced a goodness of fit value (log likelihood) that was used to evaluate the schemes. Data obtained at 3 mM GABA were first analyzed for determining the best consensus scheme for all four GABA A Rs. The rate constants thus obtained were averaged across records for each GABA A R. To estimate the rate constants for the binding (k ϩ1 ) and unbinding (k Ϫ1 ) of GABA, the averaged rate constants for activation at 3 mM GABA were fixed. Binding steps were then appended to the first shut state in the scheme(s) (A 2 R 1 ), and the scheme was re-fitted to data sets that included low (2 M) data, allowing k ϩ1 and k Ϫ1 to vary freely in the fitting. Combining several records at 2 M GABA was required to increase the number of total events for that concentration. These were then combined with data obtained at 3 mM GABA to produce a data set for simultaneous fitting to the mechanism. The binding affinity (K d ϭ k Ϫ1 /k ϩ1 ) was then calculated for each data set and averaged for each GABA A R. Macropatch simulations were generated by the finalized mechanism (with all rate constants). The "dose-response" function in QuB was used to simulate macropatch currents, after setting the number of channels to 1000 and the K d values of ␣1and ␣2-containing GABA A Rs to 25 and 100 M, respectively. Exponential fitting to the rise and decay phases of the simulated currents was done in QuB or pClamp 10 (Clampfit).

RESULTS
Incorporation of the ␥-Subunit into GABA A Rs-On the basis of conductance and kinetic properties, GABA A Rs comprising of ␣, ␤, and ␥ subunits are clearly distinguishable on the single channel level from those composed of ␣ and ␤ subunits. ␣␤␥ receptors activate with a predominant unitary conductance of ϳ26 pS (at Ϫ70 mV) and exhibit complex bursting behavior with relatively long burst durations. In contrast, ␣␤ receptors under similar recording conditions have a conductance of ϳ15 pS and exhibit simple, relatively short periods of activity (28,29). We wished to investigate the presence of GABA A Rs comprised only of ␣ and ␤ subunits in our standard ␣␤␥ receptor transfections to determine whether our transfections produced pure populations of ␣␤␥ receptors. To facilitate the identification of ␣␤ receptors we transfected ␣1 with ␤2 or ␣2 with ␤2 at an ␣:␤ plasmid ratio of 1:1, and recorded the resulting single channel activity. ␣␤ receptors comprised of ␣1 and ␤2 subunits opened to 1.0 pA (␥ ϭ 12.7 pS, n ϭ 7 pooled), whereas ␣2␤2 receptors opened to a mean amplitude of 1.1 pA (␥ ϭ 14.0 pS, n ϭ 8 pooled). No activations were observed that exceeded these levels (Fig. 1A). We then looked for these ␣␤ receptor activations in patches excised from cells transfected with an ␣:␤:␥ plasmid ratio of 1:1:3. To obtain an estimate of the incidence of ␣␤ (ϳ1 pA) versus ␣␤␥ (ϳ2 pA) receptor activity we conducted a count of discrete (well separated) single channel activations mediated by both receptor types. Activations (burst or clusters) that were due to a single receptor were determined as outlined under "Experimental Procedures." Counting the relative numbers of well separated periods of activity minimized the false positive detection of ␣␤ receptor activity, as it is well known that ␣␤␥ channels can transition to sublevels within activations (7). The appearance of ␣␤ channel activations in all four ␣␤␥ receptor transfections was minimal. Transfections that included ␣2, ␤2, and ␥2L subunits exhibited ␣2␤2 receptor activations that constituted 10 Ϯ 2% (n ϭ 3) of the total activity, whereas those that included ␣2, ␤2, and ␥1 subunits produced ␣2␤2 receptor activations that were only 12 Ϯ 3% (n ϭ 5) of the total activity (Fig. 1B). In patches expressing ␣1, ␤2, and ␥2L subunits the incidence of ␣1␤2 receptor-mediated activity was 11 Ϯ 2% (n ϭ 4) of the total measured. Similarly, when expressing ␣1, ␤2, and ␥1 subunits, 6 Ϯ 1% (n ϭ 3) of the activations were of the ␣1␤2 phenotype (Fig. 1C). Hence, our standard transfection ratio produced mainly signature ␣␤␥ channel activations, ranging from 88 to 94% of the total number. This result is consistent with a study that deduced that ␣␤␥ receptors are the almost exclusively preferred assembly, even with a transfection ratio of 1:1:1 (28). Rapid GABA Application onto Macropatches-To understand the impact of ␥ (␥1 and ␥2L) and ␣ (␣1 and ␣2) subunits on the intrinsic properties of GABA A Rs we recorded ensemble currents from outside-out patches excised from HEK293 cells expressing ␣1␤2␥2L, ␣1␤2␥1, ␣2␤2␥1, or ␣2␤2␥2L GABA A Rs in response to brief (Ͻ1 ms, Fig. 2, A and B) saturating GABA (3 mM). Receptors containing ␣1 subunits activated relatively rapidly as compared with those containing ␣2 subunits. ␣1␤2␥2L and ␣1␤2␥1 GABA A Rs activated with 10 -90% rise times of 0.49 Ϯ 0.05 ms (n ϭ 10) and 0.30 Ϯ 0.04 ms (n ϭ 6, Fig. 2, C and D), respectively, whereas, ␣2␤2␥2L and ␣2␤2␥1 GABA A Rs activated with rise times of 0.53 Ϯ 0.10 ms (n ϭ 7) and 0.58 Ϯ 0.07 ms (n ϭ 9, Fig. 2, C and D), respectively. A two-way ANOVA revealed a correlation between rise time and the ␣ subunit (p ϭ 0.02), but not the ␥ subunit isoform. The deactivation phase of the currents was also substantially slower for GABA A Rs containing the ␣2 subunit (Fig. 2, C and E). The weighted deactivation time constants for ␣1␤2␥2L and ␣1␤2␥1 GABA A Rs were 5.9 Ϯ 0.5 (n ϭ 10) and 9.1 Ϯ 0.9 ms (n ϭ 6), One of the barrels contained a standard extracellular solution, whereas the other contained a barrel diluted by 50% with distilled water. Open pipette responses were used to optimize agonist application onto macropatches. C, averaged sweeps of macropatch currents recorded from patches expressing, ␣1␤2␥2L, ␣1␤2␥1, ␣2␤2␥2L, and ␣2␤2␥1 GABA A Rs in response to ϳ1 ms application of 3 mM GABA (arrowhead). The currents for all four GABA A Rs develop rapidly, with a 10 -90% rise times of Ͻ1 ms. Current deactivation has a slower time course and shows a clear ␣-subunit isoform correlation, with ␣1-containing receptors deactivating more rapidly than those containing the ␣2 subunit. Averaged data for the 10 -90% rise time (D) and deactivation rate (E) for the macropatch data.
respectively. The presence of the ␣2 subunit dramatically slowed current decay with the mean decay time constant of ␣2␤2␥2L GABA A Rs, being 44.9 Ϯ 3.9 ms (n ϭ 7), and that of ␣2␤2␥1 GABA A R-mediated currents being 33.4 Ϯ 4.2 ms (n ϭ 9). Again, a two-way ANOVA revealed a highly significant correlation between the ␣ subunit and current decay (p Ͻ 0.0001), but not the ␥ subunit isoform. These results confirm previous results showing that ␣1␤2␥2 GABA A Rs (30, 31) display significantly faster activation and deactivation kinetics, as compared with those containing ␣2 subunits (8,32). Thus, whereas the ␣ subunit isoform has a profound affect on ensemble current kinetics, mainly by slowing current deactivation, replacing ␥2L subunits with ␥1 has no effect on the kinetics of expressed receptors.
Single Channels Analysis and Activation Mechanisms-We next asked how the ␣2 subunit enables the current to persist as the GABA concentration drops to zero. Single channel currents were recorded at saturating (3 mM) and low (2 M) concentrations of GABA, which mimic the concentration profile at the onset and near the end of a synaptic event, respectively. The initial analysis focused on the durations of discrete activations (bursts and clusters of bursts) that define the activity of a single ion channel, the open state occupancy within activations (P o ) and current voltage (I-V) relationships. All four GABA A Rs exhibited single channel currents that were ϳ2 pA in amplitude at Ϫ70 mV and had I-V with mild inward rectification (Fig. 3). Single channel conductance were calculated at Ϫ70 mV after correcting the driving force for reversal (4.5-5.0 mV) and liquid junction (4.0 mV) potentials. The calculations yielded conductance values of 26.6 (␣1␤2␥2L), 26.9 (␣1␤2␥1), 25.7 (␣2␤2␥2L), and 26.7 pS (␣2␤2␥1). All receptors showed at least 2 gating modes, which were equally prevalent among the receptors. This phenomenon has been observed in other GABA A Rs (25,30), but as we were ultimately interested in determining the factors that slowed the deactivation phase of ␣2-containing receptors, the different modes of activity for each GABA A R were pooled for further analysis. Table 1 summarizes the durations of the activations and the P o values for the four channel types. At 3 mM GABA, the mean durations of clusters of activity ranged between 148 and 206 ms, with a small, but non-significant trend toward longer activations for GABA A Rs harboring the ␣2 subunit. The same rank order of, ␣1␤2␥2L Ͻ ␣1␤2␥1 Ͻ ␣2␤2␥1 Ͻ ␣2␤2␥2L was observed for mean burst durations elicited by 2 M GABA, but the differences here were more dramatic. Burst durations for ␣1-containing GABA A Rs ranged between 23 and 27 ms (Fig. 3, A and B). This was ϳ3-4-fold briefer than those  for ␣2␤2␥2L receptors that activated for a mean duration of 99 ms (Fig. 3C), whereas bursts of activity mediated by ␣2␤2␥1 receptors were of intermediate durations, being 56 ms (Fig. 3D, Table 1). The time spent in conducting configurations was similar for all four receptors, especially at 3 mM GABA, yielding P o values of ϳ0.6 -0.7. At 2 M GABA, the P o values mirrored the rank order of burst durations, but the absolute differences were smaller. It is notable, however, that the P o value for the ␣2␤2␥1 and ␣2␤2␥2L receptors at 2 M GABA were indistinguishable from those of ␣1␤2␥2L and ␣1␤2␥1 receptors at 3 mM GABA, suggesting that ␣2-containing GABA A Rs dwell in conducting states for longer intervals. Overall, the most noteworthy difference between the receptor types was the mean duration of bursts elicited by 2 M GABA. This likely underlies the longer deactivation times for receptors harboring the ␣2 subunit. In support of this inference, synaptic currents mediated by other ligand-gated ion channels have also been shown to deactivate as a function of the durations of single channel bursts of activity (33)(34)(35).
We then proceeded to analyze the open and shut dwell time distributions for the purpose of deriving a consensus mechanism for channel activation. A mechanism that accounted for the salient properties of agonist affinity and gating kinetics would allow us to determine the underlying kinetic factors that give rise to the differential ensemble and single channel currents between the four GABA A Rs, within the same quantitative framework. This would facilitate a direct comparison between receptors. We commenced this analysis by plotting shut dwell histograms to activations elicited by 3 mM and 2 M GABA. These histograms were then fitted to mixtures of exponentials to determine the minimum number of individual components that were apparent across patches and at both concentrations of GABA, and the tcrit values required to preserve them. Clusters and bursts of activity divided by this method yielded shut and open dwell histograms with three components each, as shown in Fig. 4A. This was consistent across all four receptor types suggesting that, in kinetic terms, they were all broadly similar.
We first considered clusters of activity at saturating (3 mM) GABA because this ensures binding site saturation, allowing us to omit the binding steps in the initial analysis. The number of components in the shut and open histograms was taken to represent the minimum number of functional states in the underlying activation mechanism. Mechanisms with three shut and open states were connected in various schemes and used to fit the dwell histograms to mixtures of exponentials by maximum likelihood fitting (26,27). The fitting method uses the (apparent) open and shut dwell distributions to compute the likeli-hood that the data are represented by a postulated sequence of open and shut times. The free parameters to be fitted, for each postulated mechanism, are the rate constants governing the transitions between states, which are optimized to maximize the probability of observing the data. Mechanisms that best described the activity included schemes that were linear with some branching and schemes containing looped connections (Fig. 4, B and D). The schemes were then evaluated and ranked on the basis of a goodness of fit measure (log likelihood) and how accurately the schemes recapitulated the time constants and fractions of the initial "star" fit of the data. The three linearbranched schemes (Fig. 4, B and D, Schemes 1-3) that generated the best fits to the data and the single best, looped scheme (Scheme 4 as shown in Fig. 4). Similar linear-branched schemes have previously been reported for GABA A R activation (7,25,36). Scheme 3 has previously been reported as an activation mechanism for ␣1␤2␥2S and ␣3␤3␥2S GABA A Rs (7). This scheme also fit the activity for ␥2L-containing GABA A Rs. However, we found that Scheme 1 produced higher log likelihood values for ␥1-containing channels and was competitive with Scheme 3 for ␥2L-containing channels. Summing the likelihood (⌺LL) values for each scheme over all four GABA A Rs revealed Scheme 1 as the best overall arrangement. Schemes that contained loops did not generally fit the data as well as linear-branched schemes, but Scheme 4 (Fig. 4D) adequately described most of the data, even though it was not as competitive as Schemes 1-3. On the basis of the ⌺LL and most accurate reproduction of individual components, in terms of time constants and fractions of the dwell distribution, Scheme 1 was chosen as the consensus mechanism for further analysis of rate constants for GABA activation. Rate constants were computed for each patch, averaged for each receptor subtype (Table 2), and the equilibrium constant for each state transition was determined (Table 3). Equilibrium constants were broadly similar across receptor types. One consistent difference was the constant between the first and second shut states, A 2 R 1 and A 2 R 2 (⌽). GABA A Rs expressing the ␥2 subunit had ⌽ constants that were Ͼ1, whereas those for ␥1-containing receptors were Ͻ1. ⌽ was subunit specific, suggesting that the ␥ subunit is not only involved in the activation process, but its contribution to activation is ␥ isoform dependent. The mean lifetime of A 2 R 2* was also prolonged by the presence of the ␣2 subunit, consistent with the higher P o values for these channels. However, none of the equilibrium constants differed to an extent that would adequately account for the longer burst durations for ␣2-containing receptors at 2 M GABA.
Bursts of activity at 2 M GABA were used to estimate the rate constants for GABA binding. Sequential, identical binding steps were appended to A 2 R 1 (red arrows in Fig. 4) and fitted to dwell time histograms derived from data obtained at high and low GABA, which constituted a single data set. The rate constants for the transitions downstream of the binding steps were fixed to the mean values obtained at 3 mM GABA for each receptor subtype (Table 2), allowing only the GABA association and dissociation rate constants to vary during the fitting. More consistent binding rate constants were obtained when data from multiple patches exposed to 2 M GABA were combined. Three or more data sets were used for each GABA A R, and mean values for GABA binding affinity (K d ) were obtained (Table 3). This analysis revealed clear differences in affinity that closely correlated with the ␣ subunit isoform, but not the ␥ isoform, and is consistent with the lack of involvement of ␥ subunits in GABA binding. For ␣1-containing receptors the GABA associ-ation rate constants (k ϩ1 ) varied between 2.2 ϫ 10 6 and 3.6 ϫ 10 6 M Ϫ1 s Ϫ1 and the dissociation rate constant (k Ϫ1 ) varied between 350 and 450 s Ϫ1 , yielding a mean K d of ϳ100 M for both receptors. In contrast, ␣2-containing receptors had a 3-4fold greater affinity for GABA. The k ϩ1 values estimated for these two GABA A Rs ranged between 4.0 and 4.5 ϫ 10 6 M Ϫ1 s Ϫ1 , whereas the k Ϫ1 values varied between 75 and 130 s Ϫ1 , producing mean K d values of ϳ25-30 M for GABA. As an independent (and non-equilibrium) test for Scheme 1 as a suitable consensus mechanism for activation of multiple types of GABA A Rs, we used this scheme with the respective mean rate constants for gating for the four channels, and K d  Table 2). The macropatch currents were generated by setting the channel number to 1000 and the agonist application time to 1 ms in QuB. The time constants for activation and deactivation are shown for each current. D, other postulated schemes that fit the data adequately, including Scheme 2 (3) and a scheme containing looped connections (Scheme 4). Note that all schemes have at least one shut-shut transition between the binding steps (red arrows) and the open states.  values of 100 and 25 M for ␣1and ␣2-containing GABA A Rs, respectively, to generate simulated macropatch ensemble currents (Fig. 4C). The simulated ensemble currents all activated rapidly (ϳ1 ms), being only marginally slower than the measured macropatch currents (Fig. 2). For ␣1-containing GABA A Rs the simulated ensemble currents were similar, but not identical. The deactivation phase of these currents, fitted to two exponential equations, produced single weighted time constants of ϳ10 ms, which was also close to the measured values of ϳ6 -9 ms. Similarly, Scheme 1 produced simulated ensemble currents that activated with 10 -90% rise times of ϳ1 ms for both ␣2-containing receptors and deactivation time constants of ϳ40 ms for ␣2␤2␥1 GABA A Rs (measured ϳ33 ms) and ϳ50 ms for ␣2␤2␥2L GABA A Rs (measured ϳ45 ms). These estimates corresponded closely with the measurements from experimental currents (Fig. 2), again validating Scheme 1 as an accurate general descriptor of both single channel activations and macropatch currents for the four synaptic GABA A Rs considered here. Synaptic Currents Mediated by ␣1␤2␥2L, ␣1␤2␥1, ␣2␤2␥1, and ␣2␤2␥2L GABA A Rs-We have shown that ␣1␤2␥2L, ␣1␤2␥1, ␣2␤2␥1, and ␣2␤2␥2L GABA A Rs can be described by a single kinetic mechanism with the key difference being that receptors containing ␣2 subunits have a significantly higher affinity for GABA, resulting in slower current deactivation times. In contrast, the ␥ subunit has little or no impact on the kinetics of ensemble currents. We therefore predicted that the kinetics of synaptic currents mediated by these receptors would be dominated by the ␣ subunit. This prediction was tested in engineered heterosynapses formed between HEK293 cells and cultured cortical neurons, enabling us to examine the properties of synaptic currents mediated by populations of GABA A Rs of defined subunit composition. Importantly, synaptic currents at these engineered synapses should not be affected by errors due to voltage clamp or electronic distortions commonly present when recording synaptic currents from neurons. Mature cortical neurons readily formed GABAergic synaptic contacts on HEK293 cells transfected with the desired GABA A R. The synapses were observable as GAD65-positive contacts on the surface of the HEK293 cells (Fig. 5A). Higher resolution confocal images of cells where neuroligin 2A had been labeled to represent the postsynaptic density showed a close correspondence between neuroligin 2A and GAD-65 positive synaptic contacts confirming assembly of GABAergic synapses on HEK293 cells (Fig. 5B).
Whole cell recordings from transfected HEK293 cells in coculture with cortical neurons exhibited spontaneous synaptic currents of variable amplitude that ranged between ϳ20 and 200 pA for all four receptor types (Fig. 5C). IPSCs mediated by the well characterized ␣1␤2␥2L GABA A Rs activated rapidly, with mean 10 -90% rise times of 1.2 Ϯ 0.2 ms and decayed with a mean time constant of 4.0 Ϯ 0.8 ms (n ϭ 3 cells). These values are similar to rise time and offset time constants for the same receptors expressed in macropatches (Fig. 2). Moreover, they are similar to previously reported recordings of synaptic currents at synapses expressing ␣1␤2␥2 GABA A Rs (4,9), including studies on neuronal types that are not susceptible the distorting effects of cable filtering (20). Together, these results show that synapses that form in co-cultures faithfully recapitulate functional synapses.
As compared with those mediated by ␣1␤2␥2L receptors, synaptic currents mediated by the other three GABA A Rs showed markedly different activation and deactivation profiles (Fig. 5, D and E). The rise times for these synaptic currents were all slower than their respective activation rates in macro- patches. ␣1␤2␥1 and ␣2␤2␥2L receptor synaptic currents had mean 10 -90% rise times of 4.0 Ϯ 0.7 (n ϭ 4) and 4.0 Ϯ 0.5 ms (n ϭ 7), respectively. The rise time of the ␣2␤2␥1 receptormediated currents was exceptionally slow, being 8.2 Ϯ 1.1 ms (n ϭ 5). A two-way ANOVA revealed that both ␣ and ␥ subunit isoforms had a significant effect on current activation (p Ͻ 0.001). Similarly, as compared with macropatches, the deactivation of IPSCs mediated by ␣1␤2␥1 and ␣2␤2␥1 GABA A Rs were substantially slower (Fig. 4E), with mean time constants of 19.8 Ϯ 3.0 (n ϭ 4) and 67.1 Ϯ 7.6 ms (n ϭ 7), respectively. The ␣2␤2␥2L GABA A R generated IPSCs that deactivated with an intermediate time constant (38.7 Ϯ 3.0 ms, n ϭ 7). Here too, a two-way ANOVA test indicated that both ␣ and ␥ subunit isoforms had a significant effect on current deactivation (p Ͻ 0.001). Synaptic currents mediated by ␣2-containing receptors had the slowest decay time constants, but this could only partially be explained by the macropatch and single channel data. These data suggest that ␣2 and ␥2L subunits play distinct roles in determining the kinetics of GABA A R-mediated IPSCs.
Receptors incorporating the ␣2 subunit mediate currents with slower activation and deactivation kinetics, whereas the presence of the ␥2L subunit tended to accelerate both activation and deactivation. The antagonistic effect between ␣2 and ␥2L is best illustrated in ␣2␤2␥2L GABA A Rs, whose currents activated more slowly than macropatch currents, but deactivated at about the same rate.
In contrast, the slowing of current decay for GABA A Rs incorporating the ␥1 subunit cannot be attributed to this subunits' contribution to the intrinsic properties of the receptors, as both macropatch and simulated ensemble currents for ␥1-containing GABA A Rs had rapid onsets and decays (Figs. 2 and 4). Clearly, then, factors other than the intrinsic kinetic properties of the receptors are responsible for the slower kinetics of the synaptic currents mediated by receptors expressing ␥1 subunits. One revealing observation was the reciprocal deactivation pattern for macropatch versus synaptic currents between ␣2␤2␥1 and ␣2␤2␥2L GABA A Rs. The deactivation rate for ␣2␤2␥1 receptors was marginally faster than ␣2␤2␥2L receptors in macropatch currents but synaptic currents mediated by ␣2␤2␥1 GABA A Rs were significantly slower than those mediated by ␣2␤2␥2L GABA A Rs, suggesting the ␥ subunit has a prominent effect on synaptic current kinetics. One possible explanation is that as with the ␣ subunit (37,38), the ␥ subunit isoform may also affect receptor clustering at synapses. GABA A Rs that are only loosely clustered at synapses would exhibit slow deactivation kinetics due to slower changes in GABA concentration, whereas GABA A Rs that were more tightly concentrated post-synaptically would give rise to faster current kinetics. Synaptic currents with the slowest kinetics were those generated by ␣2␤2␥1 GABA A Rs, likely because of a combination of the ␣2 subunit on mean burst duration and the "de-clustering" effect of both ␣2 and ␥1 subunits.
The analysis of ␣␤ receptors in our transfections suggests that, due to their small conductance (ϳ13-14 pS) and infrequent activation (ϳ10% of total), their presence would not make a substantial impact on ensemble currents (macropatch and synaptic) that included the ␥ subunit. Nevertheless, we also recorded currents in co-cultures transfected only with ␣1 and ␤2 or ␣2 and ␤2 subunits to examine if ␣␤ receptors can assemble at synapses. Pure populations of ␣␤ receptors exhibited synaptic currents with rise and decay kinetics that were broadly similar to those of ␣␤␥ receptors. ␣1␤2 receptors produced 10 -90% rise times of 3.0 Ϯ 0.1 ms and decayed with a mean time constant of 11.0 Ϯ 1.1 ms (n ϭ 3). These values were intermediate between those mediated by ␣1␤2␥2L and ␣1␤2␥1 receptors, and an ANOVA test showed no significant difference (p Ͼ 0.05) between ␣1␤2 receptors and either of their ␥-containing counterparts. ␣2␤2 receptors produced mean rise and decay times of 10.5 Ϯ 1.9 and 72.0 Ϯ 15.4 ms, respectively (n ϭ 4). As revealed by an ANOVA test, ␣2␤2-mediated synaptic currents were only significantly slower in rise and decay times (p Ͻ 0.05 for both) to the corresponding measurements of ␣2␤2␥2L-mediated currents. This result is consistent with the ␥2L subunit having a clustering effect on receptors, whereas the incorporation of the ␣2 subunit tended to de-cluster the receptors to produce slower activation rates. The slow decay times in ␣2␤2-mediated currents are also consistent with ␣2-containing receptors having a higher affinity for GABA. These data demonstrate that ␣␤ receptors can assemble at synaptic sites, as has been demonstrated for ␣2␤3 and ␣6␤3 receptors (38). However, as in our transfections, ␣␤ receptors only constitute about 10% of the overall activity (Fig. 1), their impact on the kinetics of synaptic currents will be minimal.
␥-Chimeras Reveal Differential Clustering Properties in Synaptic GABA A Rs-In ␣ subunits, the intracellular domain between TM3 and TM4 has been shown to play a role in clustering GABA A Rs to the synapse via interactions with gephyrin (37,39). An association between gephyrin and the ␥2 subunit was suggested to contribute to synaptic targeting of GABA A Rs (10). Although this has not been confirmed by other studies (14), it remains possible that gephyrin and ␥2 interact in mammalian systems, as has been recently shown for the ␤ subunit (13). Interactions with other proteins must mediate gephyrinindependent clustering (5), and the ␥ subunit could also contribute to these interactions. We tested whether gephyrin and collybistin affected the kinetics of synaptic currents by co-expressing both of these proteins along with either ␣2␤2␥1 or ␣2␤2␥2L receptors. The rise and decay times for the ␣2␤2␥1 and ␣2␤2␥2L receptors in combination with these proteins were, respectively, 7.6 Ϯ 0.6 (n ϭ 7) and 4.7 Ϯ 0.4 ms (n ϭ 4) and 52.9 Ϯ 3.9 and 41.0 Ϯ 2.2 ms. t tests showed that gephyrin and collybistin expression had no significant effect on synaptic current rise times (p Ͼ 0.1 for both receptors) or decay times (p Ͼ 0.1; for both receptors). These results demonstrate that gephyrin (and collybistin) have little effect on GABA A R-mediated synaptic currents, as has been suggested by some studies (38,40). Alternatively, because HEK293 cells endogenously express gephyrin (38), recombinantly expressed gephyrin may have no additional effect on current kinetics.
The IDs of the ␥1 and ␥2L show considerable sequence divergence and their TM4 domains vary at sites that correspond to those shown to be essential for ␥2-mediated receptor clustering in cultured neurons (12). Given these observations, we tested the possibility that the ␥ subunit isoform was also affecting synaptic clustering, by making chimeras of the ␥1 and ␥2L subunits that replace the ID and TM4 of one isoform with that of the other. This produced two ␥-chimeric subunits, ␥2L-␥1 and ␥1-␥2L (Fig. 7A), which were then cotransfected with ␣2 and ␤2 subunits. These transfections also produced robust spontaneous synaptic activity, of comparable frequency and amplitude to the wild-type receptors. Synaptic currents mediated by the ␣2␤2␥1-␥2L GABA A Rs activated with a mean 10 -90% rise time of 4.4 Ϯ 0.5 ms (n ϭ 5) and deactivated with a mean weighted time constant of 38.2 Ϯ 2.4 ms (Fig. 7B). This current profile was indistinguishable from that of the wild-type ␣2␤2␥2L receptors (Fig. 7C). Similarly, the ␣2␤2␥2L-␥1 receptors exhibited activation and deactivation rates of 7.4 Ϯ 1.1 and 53.5 Ϯ 7.2 ms (n ϭ 5), respectively, and these too were similar to wild-type ␣2␤2␥1 GABA A Rs (Fig. 7, B and C). A two-way ANOVA confirmed that the ID plus the TM4 region had a significant effect on activation and deactivation rates (p Ͻ 0.001 for both), whereas the extracellular domain and TM1-3 did not (p Ͼ 0.1 for both). These observations show that the ␥ subunit isoform is a major contributor to the kinetic profile of synaptic currents and the ID and TM4 likely mediates this effect.

DISCUSSION
In this study we have shown that the presence of the ␣2 subunit slows the deactivation phase of the IPSC by increasing the receptors' affinity for GABA, whereas inclusion of the ␣2 and ␥1 subunits slows both the activation and deactivation phases of the IPSC by conferring loose clustering properties to the receptors. The presence of the ␥1 subunit results in IPSCs with markedly slower activation and deactivation phases, and the kinetics of chimeras of ␥1 and ␥2 subunits are in agreement with this proposal. Together, these data suggest that GABA A Rs containing ␥1 and ␥2 subunits use different mechanisms for synaptic clustering.
We first determined the kinetic properties of four subtypes of GABA A Rs that vary in their ␣ (␣1 or ␣2) and/or ␥ (␥1 or ␥2L) subunit isoform, whereas keeping the ␤ subunit constant. Brief (Ͻ1 ms) GABA application onto macropatches elicited currents that mimic those at synapses, but are unaffected by factors that are not related to the inherent properties of the receptors. The receptor kinetic properties were further investigated on a single channel level, and within the framework of a single activation mechanism, facilitating a correlation between subunit isoform, GABA affinity, and the efficacy with which GABA activated the receptors (41). Macropatch currents mediated by all four GABA A Rs activated with sub-millisecond rates, with ␣2-containing receptors activating marginally more slowly. The inclusion of the ␣2 subunit also slowed current deactivation by almost an order of magnitude.
An analysis of the discrete activations (clusters and bursts) showed that the durations of these activations was ␣ subunit dependent. At a low GABA concentration, ␣1-containing receptors activated for mean durations of 23-27 ms, whereas the presence of ␣2 subunits lengthened the bursts to 60 -100 ms. Single channel data were also used to derive an activation  FIGURE 6. Benzodiazepine pharmacology of ␥2Land ␥1-containing GABA A Rs. Averaged and normalized current traces from multiple cells expressing either ␣2␤2␥2L or ␣2␤2␥1 GAB A Rs before (black) and during (gray) continuous perfusion of diazepam (A) and flunitrazepam (B). The accompanying bar plots are pooled data for current decay and peak amplitude. Note that both benzodiazepines markedly slowed the decay rate of ␣2␤2␥2L GABA A Rs (*, p Ͻ 0.05; ****, p Ͻ 0.0001), but had no significant (ns) effect on ␣2␤2␥1 GABA A Rs. Neither drug significantly altered the peak amplitude of the currents. mechanism that accurately described the single channel and macropatch data of all four GABA A Rs. This scheme comprised two sequential, equivalent binding steps for GABA followed by three shut and three open functional states (Fig. 4B). Given similar models have previously been applied to other isoforms of GABA A Rs (7, 25), our activation scheme may be generally applicable to other synaptic GABA A R stoichiometries. This consensus mechanism suggests that the essential contribution made by the ␣2 subunit is to enhance the GABA binding affinity 3-4-fold, thereby increasing the durations of bursts. A similar result was observed for ␣3-containing GABA A Rs (7). The discrepancy in ligand affinity between ␣1and ␣2-containg GABA A Rs is compatible with the significant primary sequence divergence at the GABA binding domains of these two subunits. A common feature of all schemes that were tested here, and indeed for mechanisms derived for other pentameric ligand-gated ion channels (42)(43)(44)(45)(46) is the presence of at least one shut-to-shut state transition immediately following the binding reaction steps. The equilibrium constant describing the transition between these two shut states was denoted as ⌽ and it is intriguing that macropatch and single channel analysis failed to detect any kinetic parameter that could be attributed to the ␥ subunit isoform other than ⌽. This constant was Ͻ1 only if the receptor expressed the ␥1 isoform and may pertain to GABA A R modulation by benzodiazepines, which a recent study has shown to manifest as an enhancement of ⌽ in ␥2-containing GABA A Rs (47). Our data are consistent with the notion that ⌽ is ␥ isoform dependent, the lower value of ⌽ for ␥1-containing receptors might suggest a reduced capacity for enhancement by benzodiazepine modulators.
Transfecting HEK293 cells with GABA A R subunits together with neuroligin 2A, and co-culturing these on a bed of neurons induces the formation of functional synapses between neurons and HEK293 cells (48), demonstrating that all of the essential pre-and post-synaptic elements are present in the artificial system, including neurexin, which is endogenously expressed in neurons, and gephyrin, which is present in HEK293 cells (38).
At these synapses notable pharmacological differences were observed between ␥1-containing and ␥2L-containing GABA A Rs. Experiments using flunitrazepam and diazepam demonstrate that benzodiazepines are ineffective at enhancing synaptic currents mediated by ␥1-containing GABA A Rs. This result is consistent with whole cell peak current measurements of ␥1-containing GABA A Rs (18), and key differences in the amino acid sequence between ␥2L and ␥1 that have been shown to affect the potency with which benzodiazepine-site ligands modulate currents (49 -52).
In addition, our results show that the ␣2 and ␥1 subunits have de-clustering effects when expressed at synapses. Using chimeric constructs we show that the ID (plus TM4) is responsible for this difference in the ␥-subunit. The ID and TM4 of GABA receptor subunits is crucial for clustering of receptors at post-synaptic sites (12), and our results suggest that, at these engineered synapses, ␥1and ␥2L-containing GABA A Rs have different synaptic kinetics because of differences in their clustering properties. Thus, at neuronal synapses in situ, it is possible that GABA receptors containing ␥1 and ␥2 subunits may also be differentially targeted (18,53). Subunit-specific clustering mechanisms have already been noted for ␣ subunits in neurons. For example, dystrophin is currently thought to be involved in anchoring dendritic clusters of ␣1 in specific cortical layers (15), and radaxin has been shown to selectively anchor ␣5 subunits (54). Differential clustering properties have also been demonstrated for ␣1 and ␣2 subunits, such as the lower affinity of the ␣2 subunit for gephyrin (37) and the recruitment of ␣2, but not ␣1 subunits to the axon initial segment (55).
Postsynaptic GABA A Rs are dynamic, diffusing in and out of the synaptic active zone, which is ϳ200 -300 nm in diameter (56,57), with a diffusion coefficient that ranges from 0.01 to 0.05 m 2 s Ϫ1 (56). Quantum dot and Immunogold-labeled GABA A Rs show sub-micrometer separations between GABA A Rs that appose the presynaptic density and those that are perisynaptic (57,58), whereas extrasynaptic GABA A Rs, such as those containing the ␦ subunit, are generally located hundreds of nanometers to several micrometers further (38,57). These observations are consistent with a concentration gradient of receptors that is an . ␥1 subunits slow synaptic current kinetics. A, schematic representation of the subunit chimeras used to investigate ␥ subunit related synaptic clustering. Swapping the intracellular and TM4 domains made the chimeras. B, averaged synaptic currents recorded from heterosynapses expressing the wild-type ␣2␤2␥2L and ␣2␤2␥1 (reproduced from Fig. 5) and chimeric ␣2␤2␥1-␥2L and ␣2␤2␥1-␥2L GABA A Rs. Note the similar activation and deactivation between wild-type and corresponding chimeric GABA A Rs. C, averaged values for activation and deactivation for the wild-type and chimeric GABA A Rs showing that both activation and deactivation are strongly dependent on the ID plus TM4 domains of the ␥ subunit.
inverse function of receptor diffusion mobility. We interpret our data as being consistent with a differential, ␥-isoform-dependent gradient of receptors, rather than mutually exclusive zones delineating synaptic receptors from those beyond the synaptic perimeter. The slower rise and decay times for ␥1-containing GABA A Rs suggest that these receptors are more mobile and at a higher density outside the synapse than ␥2L-containing receptors, whereas the converse would apply for ␥2L-containing receptors. Within this context we refer to ␥2L-containing GABA A Rs as being more tightly clustered at synapses where a higher proportion are perfused with high GABA prior to significant GABA diffusion. Our findings evince key factors that determine the profile of synaptic currents mediated by GABA A Rs containing ␣1, ␣2, ␥1, and ␥2L subunits, and provide a solid basis for future studies to establish whether GABA A Rs containing ␣2 and ␥1 subunits contribute to GABAergic synapses in key brain regions that mediate fear and anxiety (59).