Amino acid substitutions in the human homomeric β3 GABAA receptor that enable activation by GABA

GABAA receptors (GABAARs) are pentameric ligand-gated ion channels that mediate synaptic inhibition throughout the central nervous system. The α1β2γ2 receptor is the major subtype in the brain; GABA binds at the β2(+)α1(−) interface. The structure of the homomeric β3 GABAAR, which is not activated by GABA, has been solved. Recently, four additional heteromeric structures were reported, highlighting key residues required for agonist binding. Here, we used a protein engineering method, taking advantage of knowledge of the key binding residues, to create a β3(+)α1(−) heteromeric interface in the homomeric human β3 GABAAR that enables GABA-mediated activation. Substitutions were made in the complementary side of the orthosteric binding site in loop D (Y87F and Q89R), loop E (G152T), and loop G (N66D and A70T). The Q89R and G152T combination enabled low-potency activation by GABA and potentiation by propofol but impaired direct activation by higher propofol concentrations. At higher concentrations, GABA inhibited gating of β3 GABAAR variants containing Y87F, Q89R, and G152T. Reversion of Phe87 to tyrosine abolished GABA's inhibitory effect and partially recovered direct activation by propofol. This tyrosine is conserved in homomeric GABAARs and in the Erwinia chrysanthemi ligand-gated ion channel and may be essential for the absence of an inhibitory effect of GABA on homomeric channels. This work demonstrated that only two substitutions, Q89R and G152T, in β3 GABAAR are sufficient to reconstitute GABA-mediated activation and suggests that Tyr87 prevents inhibitory effects of GABA.

GABA A Rs 2 are members of the pentameric ligand-gated ion channel family and mediate fast synaptic inhibition (1). Consequently, they are important pharmacological targets (2,3).
The orthosteric binding site is located between the ␣ and ␤ subunits that comprise the complementary (Ϫ) and principal (ϩ) components, respectively. The site contains seven noncontiguous binding loops (A-G): A-C belong to the principal side, whereas loops D-G belong to the complementary side (7)(8)(9).
There are 19 different GABA A R subunits that form at least 14 distinct combinations in vivo (10,11), accounting for the physiological versatility and pharmacological selectivity of these channels (2). The major subtype in the central nervous system is the ␣ 1 ␤ 2 ␥ 2 GABA A R. The ␤ 1 , ␤ 3 , and subunits can form homomers when recombinantly expressed in vitro. Although the homomeric ␤ 3 has not been identified in vivo, it is of considerable interest as the first GABA A R to yield to high-resolution structural analysis (12) and for functional studies because histaminergic ligands and propofol activate the receptor (13)(14)(15)(16). Recently, four heteromeric GABA A R structures were published, including the major subtype (17)(18)(19). These studies determined the important residues for GABA binding and suggest that variability on the complementary subunit influences ligand selectivity (19). The homomeric ␤ 3 cannot be activated by GABA (16,20). This raises questions about which residues in the complementary side are required to reconstitute activation. The availability of the ␤ 3 structure provides an opportunity to locate candidate residues.
In the present study, we investigated whether substituting amino acids in the complementary side of the ␤ 3 GABA A R to corresponding residues in the ␣ 1 subunit would reconstitute activation by GABA. Four ␤ 3 mutants were designed and used for patch-clamp electrophysiology. We analyzed the activation by GABA and propofol, potentiation of GABA-evoked currents by propofol, and the kinetics of GABA-evoked currents. Comparative modeling and molecular docking calculations were used to predict the orientation of GABA at the orthosteric site of the mutant ␤ 3 GABA A R. Using these approaches, we demonstrated that Q89R and G152T substitutions reconstituted GABA activation of GABA A R ␤ 3 and potentiation by propofol.
In addition, we found that the Y87F substitution caused GABA to inhibit receptor function.

Designing the constructs
We modified the ␤ 3 GABA A R by replacing the ICD (residues 346 -396) with the SQPARAA sequence to mimic the construct used to crystallize the ␤ 3 GABA A R (12), referred to from this point as ␤ 3 -cryst (Table 1).

Docking GABA into the ␤ 3 orthosteric site
Docking calculations were performed between GABA and the ␤ 3 -cryst model (Fig. 1A). As far as we are aware, there are no prior reports of docking GABA into the homomeric ␤ 3 receptor. The best GABA pose presented an energy of Ϫ38 kcal/mol, suggesting binding. Examination of residues within the orthosteric binding domain on the ␣ 1 subunit revealed amino acids that are not shared by ␤ 3 at key locations known to affect activation by GABA (Fig. 1B). Substitution of these residues into the ␤ 3 -cryst model (GABA A R ␤ 3 C1) improved the binding energy of GABA as evidenced by docking calculations (Ϫ46 kcal/mol). The model suggests that the GABA amino group forms a salt bridge with Glu 180 on the (ϩ) interface of the ␤ 3 subunit (Fig. 1C) and that the GABA carboxyl makes a bidentate interaction with Arg 89 and a hydrogen bond with the Thr 152 hydroxyl group, substituted in the (Ϫ) interface. These interactions are in agreement with the cryo-EM structures of the human GABA A R ␣ 1 ␤ 2 ␥ 2 and rat GABA A R ␣ 1 ␤ 1 ␥ 2 (18,19). In addition, they were described by other studies using docking calculations with human GABA A R ␣ 1 ␤ 2 ␥ 2 (21) and insect GABA A R models (21,22).
We subsequently determined the concentration-response relationship of ␤ 3 C1 to characterize the potency of GABA (Table 2). GABA was applied at increasing concentrations to cells expressing ␤ 3 C1. A representative example of these currents is shown in Fig. 2D. GABA-evoked current amplitudes were expressed as a percentage of the maximum and plotted as a concentration-response relationship (Fig. 2E). The data indicate that GABA exhibits a biphasic concentration-response relationship, which suggests two effects: activation and inhibition. We therefore fitted a two-component logistic function to the data (see "Experimental procedures"). GABA, up to 10 mM, activates ␤ 3 C1 with an EC 50 of ϳ3 mM ( Table 2). Higher concentrations of GABA caused a reduction in current amplitude with an IC 50 of ϳ50 mM. This inhibitory effect has not been observed previously in GABA A Rs (9,(23)(24)(25)(26) or in the bacterial pentameric ligand-gated ion channel ELIC (27), which, like ␤ 3 C1, also requires high concentrations of GABA for its activation (Fig. S1). We also observed a lack of inhibitory effect in heteromeric GABA A Rs formed from ␤ 3 -cryst and ␤ 3 C1 subunits (Fig. S2).

Kinetics of ␤ 3 C1
In addition to the biphasic nature of the GABA concentration-response relationship, the representative currents shown in Fig. 2D also display unusual kinetics. Therefore, we analyzed the current activation and deactivation rates by measuring the 10 -90% rise time and by fitting a two-component exponential function, respectively (see "Experimental procedures"). The mean values of rise times and weighted were plotted (Fig. 2, F and G). The individual components of the double-exponential fits for deactivation can be found in Table 3. Currents evoked by lower concentrations of GABA (0.1 and 0.3 mM) were excluded from the analysis due to their small amplitudes. Consistent with the concentration dependence of peak current activation (Fig.  2E), the concentration dependence of activation and deactivation also appears to be biphasic (Fig. 2, F and G).

GABA does not cause a voltage-dependent channel block
The inhibitory effect of GABA at higher concentrations could be due to binding at a lower-affinity site, which blocks the channel pore. We therefore examined whether GABA causes a voltage-dependent block of ␤ 3 C1 by comparing the currentvoltage (I-V) relationships of currents evoked by 1 and 100 mM GABA. These concentrations were chosen because the inhibitory effect was observed at 100 mM but not at 1 mM GABA. Representative examples of the currents evoked by GABA at voltages ranging from Ϫ60 to 60 mV are shown in Fig. 3A. GABA (1 mM) produced an outwardly rectifying I-V relationship, consistent with previous observations of currents mediated by ␤ 3 and ␣ 1 ␤ 3 GABA A Rs (16), as did 100 mM GABA. We quantified outward rectification by expressing the current amplitudes as a ratio of those evoked at Ϫ60 mV (Fig. 3B). The rectification indexes calculated (I 60 mV /I Ϫ60 mV ) were 3.5 Ϯ 0.7 and 3.6 Ϯ 0.5 (n ϭ 4) for 1 and 100 mM GABA, respectively (p ϭ 0.8, t test, n ϭ 4). These results suggest that the inhibitory effect of 100 mM GABA was not caused by voltage-dependent channel block.
Mutations enabling GABA activation of GABA A ␤3 homomers

Substitutions in loop G do not affect the activation of ␤ 3 C1 by GABA
Mutagenesis studies in GABA A R ␣ 1 ␤ 2 ␥ 2 indicate that the identities of ␣ 1 loop G residues at positions 71 and 75 influence gating and, thereby, the apparent potency of GABA (9,23,25,28). In an attempt to increase the apparent potency of GABA, substitutions in loop G were made, introducing ␣ 1 residues into ␤ 3 C1 N66D and ␤ 3 C1 A70T GABA A Rs (residues equivalent to those at ␣1 positions 71 and 75, respectively). Neither the potency nor the efficacy of GABA was affected (one-way ANOVA post hoc Dunnett's, p ϭ 0.8, F(2,12) ϭ 0.28). This is perhaps not surprising due to the conservative nature of the N66D and A40T substitutions ( Fig. S3 and Table S1).

A loop D Tyr conserved in homomeric receptors prevents block by GABA
An amino acid sequence alignment of the pentameric ligandgated ion channel subunits that form homomeric GABA-activated receptors, including ELIC, reveals conservation of the Tyr at the position equivalent to ␤ 3 amino acid 87 (Fig. S4A). We investigated whether replacement of ELIC Tyr 38 with Phe affects activation by GABA. Interestingly, GABA failed to evoke currents mediated by ELIC Y38F despite the conservative nature of this substitution (Fig. S4C). These data suggest that the Phe is detrimental to ELIC function. Because ELIC, GABA A , and GABA A ␤ all contain a Tyr, this residue may be necessary for preventing block of homomeric receptors by GABA. We tested the hypothesis that the Tyr is required in ␤ 3 receptors to prevent inhibitory effects of GABA at high concentrations by creating the ␤ 3 C1 F87Y in which the Phe 87 was reverted back to the tyrosine found in WT ␤ 3 .
Cells expressing ␤ 3 C1 F87Y were voltage-clamped at Ϫ60 mV, and GABA-evoked currents were recorded. Representative examples are shown in Fig. 4A. The current amplitudes were expressed as a percentage of maximum and plotted as a concentration-response relationship, which was fitted with a single-component logistic function (Fig. 4B). The potency of activation by GABA was similar when compared with ␤ 3 C1 (p ϭ 0.2, n ϭ 4, t test; Table 2) as was the maximum current density (p ϭ 0.6, n ϭ 4, t test; Table 2). However, the inhibition by 100 mM GABA was absent in ␤ 3 C1 F87Y with a significant difference in the current amplitude evoked by 100 mM GABA compared with that mediated by ␤ 3 C1 (p ϭ 0.003, n ϭ 4, t test). Similarly, higher concentrations of GABA (300 mM) did not reduce GABA-mediated current amplitude (Fig. S5), indicating that the inhibitory component was abolished with the F87Y substitution. Furthermore, the rate of activation of ␤ 3 C1 F87Y increased with GABA concentration and was not biphasic (Fig.  4C). There was also no apparent influence of GABA concentration on deactivation (Fig. 4D), consistent with the previous data for GABA A Rs (29) and ELIC (Fig. S1, C and D).
Potentiation, activation, and blockade of GABA A Rs occur at different propofol concentrations, consistent with the possibility of distinct sites with differing affinities (30 -32). The substitutions introduced in ␤ 3 C1 and ␤ 3 C1 F87Y may have affected  A and B, examples of currents recorded when GABA (10 mM) was applied to cells expressing GABA A R ␤ 3 -cryst (A) and GABA A R ␤ 3 C1 (B) indicate that the latter is functional and activated by the neurotransmitter. C, mean Ϯ S.D. current densities evoked by GABA (10 mM), with an asterisk indicating significant differences between the proteins (n ϭ 7, p ϭ 0.003, t test). D, examples of currents mediated by GABA A R ␤ 3 C1, evoked by increasing concentrations of GABA. Currents in gray are declining due to inhibition by GABA (Ͼ10 mM). The bar indicates GABA application (5 s). E, concentrationresponse relationships obtained using the percentage of the maximum amplitude recorded for each cell (n ϭ 5). Logistic equations were fitted to the data points (see "Experimental procedures"). From the double-logistic fit, two distinct potencies were observed for activation (EC 50 ϭ 2.9 mM) and inhibition (IC 50 ϭ 50.5 mM). A summary of the data is in Table 2. F, graph of mean current 10 -90% rise time. Activation rates are slowed somewhat by increasing the GABA concentration in ␤ 3 C1 (n ϭ 6, F(4,25) ϭ 42.2, one-way ANOVA post hoc Dunnett's, p ϭ 0.04 comparing 10 with 1 mM GABA), whereas currents evoked by 100 mM GABA were activated faster (n ϭ 6, p Ͻ 0.0001, F(4,25) ϭ 42.2, one-way ANOVA post hoc Dunnett's, comparing 100 with 1 mM GABA). G, values for weighted of deactivation exhibited a similar trend with increasing GABA concentration (p ϭ 0.04, one-way ANOVA, n ϭ 6, F(4,19) ϭ 3.2), although there was no significant difference comparing 1 mM with the other GABA concentrations tested. Detailed information about the components is in Table 3. Error bars represent S.D.

Mutations enabling GABA activation of GABA A ␤3 homomers
the gating mechanism or induced structural rearrangements that disrupt the binding site of propofol responsible for the direct activation of the receptor.
We investigated whether propofol can potentiate GABA-induced currents mediated by ␤ 3 C1 and ␤ 3 C1 F87Y. Cells were stepped from GABA (1 mM) to a solution of GABA (1 mM) plus propofol (10 or 30 M) and back to GABA (1 mM). This concentration of GABA corresponds to EC 25 according to the concentration-response relationship, allowing ample scope for enhancement (Fig. 2E).

Discussion
This study demonstrates that the replacement of two key residues in the orthosteric binding site of the ␤ 3 subunit (Gln 89 and Gly 152 ) by the equivalent ECD loci in the ␣ subunit, Arg and Thr, respectively, enables gating of ␤ 3 receptors by GABA. Docking to the ␤ 3 C1 model, which includes these substitutions plus the additional F87Y substitution, confirmed the interaction of GABA with all three of these binding residues. The favored GABA-binding pose was similar to that of heteromeric GABA A R structures (18,19) and to observations in previous docking studies using the mammalian heteromeric and the insect homomeric GABA A Rs (21,22) and in general agreement with the literature (21,33,34). The GABA carboxyl makes a bidentate interaction with Arg 89 and a hydrogen bond with the Thr 152 hydroxyl group in ␤ 3 C1. The same interactions were reported in heteromeric GABA A R structures solved in the presence of the agonist (18,19). In addition, site-directed mutagenesis studies demonstrate that substitution of these residues in the ␣ subunit affects GABA potency in GABA A R ␣ 1 ␤ 2 ␥ 2 and GABA A R ␣ 1 ␤ 2 (24,35,36). Taken together, the results of docking and functional analysis are consistent with the idea that the introduction of Q89R and G152T substitutions into ␤ 3 generates a heteromeric ␤ 3 (ϩ)␣ 1 (Ϫ)-like interface capable of activation by GABA albeit at high concentrations (Ͼ300 M).
GABA concentrations above 10 mM caused a blocking effect in ␤ 3 C1. This has not been observed in other physiologically relevant heteromeric GABA A Rs (9,(23)(24)(25)(26) or in ELIC (27). The effect was abolished when the phenylalanine in ␤ 3 C1 was reverted back to tyrosine, F87Y. Interestingly, this effect was also abolished in heteromeric GABA A Rs formed from ␤ 3 C1 (where position 87 is a Phe) and ␤ 3 -cryst subunits (where position 87 is a Tyr). The apparent potency of GABA-mediated activation is not altered in these heteromeric GABA A Rs. Although the stoichiometries of heteromeric GABA A Rs formed from ␤ 3 C1 and ␤ 3 -cryst subunits are not known, our data suggest that the incorporation of one or more Tyr 87 is sufficient to prevent GABA-mediated blockade while preserving GABA-   Mutations enabling GABA activation of GABA A ␤3 homomers mediated activation, highlighting the importance of this residue in GABA A R function. The kinetics of GABA-evoked currents mediated by ␤ 3 C1 GABA A Rs were also unusual. Activation and deactivation became slower and then faster with increasing concentrations of GABA, whereas the kinetics in ␤ 3 C1 F87Y GABA A Rs were more consistent with those of heteromeric GABA A Rs (29) and ELIC (Fig. S1). Interestingly, activation and deactivation rates of GABA-evoked currents mediated by ␤ 3 C1 GABA A Rs appear similar to those described for GABA A Rs activated in the presence of modulators, such as propofol (37) and benzodiazepines (38). In addition to its role as an agonist and an inhibitor of ␤ 3 C1 GABA A Rs, GABA may also act as a positive allosteric modulator. In the homomeric ␤ 3 C1 GABA A Rs, GABA may bind to all five subunit interfaces, and the Hill slope of 1.3 suggests cooperativity between at least two of these sites. It is possible that binding to additional orthosteric sites may result in potentiation, similar to the effect of benzodiazepines (38). However, our data with ␤ 3 C1 and ␤ 3 -cryst heteromeric GABA A Rs suggest that GABA-mediated activation does not require GABA binding to all interfaces.
Moreover, bell-shaped concentration-response curves have been described for allosteric activators and modulators of GABA A Rs, such as propofol (16), valerenic acid (39), and pentobarbital (40 -42). Pentobarbital, at low concentrations (low micromolar), can potentiate GABA A R currents by increasing the mean open duration. Higher concentrations (high micromolar) of pentobarbital can activate GABA A Rs, and millimolar concentrations can inhibit the channel, slowing deactivation (42). Similarly, GABA may act as an agonist, modulator, and inhibitor of ␤ 3 C1. However, the inhibition is not through a voltage-dependent channel block. Instead, there may be a lower-affinity inhibitory site for GABA. A similar mechanism has been proposed for the inhibitory effect observed with high concentrations of propofol (32).
Although the potentiation of GABA-evoked currents was unaffected, propofol's direct activation of ␤ 3 C1 was impaired compared with ␤ 3 -cryst. There was partial recovery of propofol-activated current medicated by ␤ 3 C1 F87Y GABA A Rs. It is clear that substitutions in the orthosteric site can influence direct activation by propofol despite its binding site being in the TM region. In keeping with a need for conformational rearrangement in the orthosteric binding site during gating by propofol, the activation is also inhibited by bicuculline (43). Furthermore, we recently demonstrated faster deactivation of propofol-evoked currents with ␣ 1 loop D (F64C) and loop G  Table 2. C, mean 10 -90% rise times showed no significant change with increasing GABA concentrations in ␤ 3 C1 F87Y except comparing 100 with 1 mM (p ϭ 0.008, n ϭ 4, F(4,15) ϭ 5.2, one-way ANOVA post hoc Dunnett's). D, mean deactivation weighted was also independent of GABA concentrations (p ϭ 0.5, one-way ANOVA, F(4,15) ϭ 0.96). Detailed information about the components is in Table 3. Error bars represent S.D.

Mutations enabling GABA activation of GABA A ␤3 homomers
(T47R) substitutions in GABA A R ␣ 1 ␤ 2 ␥ 2 , which adds additional support for a role of residues in or near the orthosteric binding site in the efficacy of gating by an allosteric agonist (26). Several studies suggest that gating by both orthosteric and allosteric agonists involves an interaction of the loops in the ECD with the TMD, particularly the loops between the ␤1-␤2 strands and TM2-TM3 helices (44 -47) and between ␤6-␤7 strands and TM2-TM3 helices (12). It is important to note that loop G is located in ␤1 strand, loop D is in ␤2, and loop E is in ␤6. The substitutions in GABA A R ␤ 3 C1 are located in loops D and E; therefore, they may affect a concerted gating mechanism.
It is not yet clear why the substitution Y87F causes GABA to act as an inhibitor of ␤ 3 C1 GABA A Rs at high concentrations and impair propofol direct activation. The substitution may affect channel gating, consistent with previous mutagenesis studies of homologous residues in GABA A R 1 that produced spontaneous opening and affected GABA, trans-4-aminocrotonic acid, and imidazole-4-acetic acid potencies (48) and in GABA A R ␣ 1 ␤ 1,2 ␥ 2 that affected GABA potency and kinetics (9,49).
The tyrosine is found in all GABA A R ␤ and subunits and in ELIC. The latter two form homomers that can be activated by GABA (20,27,50). Tyrosine may prevent an inhibitory effect of GABA in homomeric receptors, and its substitution to phenylalanine may enable GABA to bind at another lower-affinity site and inhibit gating.
In summary, this study demonstrated that only two substitutions (Q89R and G152T) were required to reconstitute activation by GABA in homomeric ␤ 3 constructs. The potency of GABA was 2 orders of magnitude lower compared with heteromeric GABA A Rs. Similar to heteromeric GABA A Rs, propofol potentiated submaximal GABA-evoked currents and caused direct activation of ␤ 3 C1 F87Y receptors. Surprisingly, the conservative replacement of Tyr 87 by phenylalanine abolished gating by propofol and caused GABA to have inhibitory effects at high concentrations.
These findings identify structural requirements for the reconstitution of a functional GABA-binding site in ␤ 3 homomeric receptors by transplanting key residues of the ␣ subunit at the heteromeric interface. This approach provides a novel method for developing a better understanding of the structural requirements for gating.

Constructs
The GABA A R constructs were designed based on the published GABA A R ␤ 3 structure, i.e. substituting the ICD for the amino acid sequence SQPARAA (12) and using the human GABA A R ␤ 3 sequence (UniProt accession number P28472). The ELIC WT construct (UniProt accession number P0C7B7) was modified for expression in HEK293 cells, adding a Kozak sequence before the cDNA and using the human 5-HT3A subunit signal peptide as described previously (51).

Mutagenesis of GABA A R ␤ 3 subunit
Genes encoding the human GABA A R ␤ 3 WT, human GABA A R ␤ 3 C1, and Erwinia chrysanthemi ELIC WT were ordered from GeneWiz and cloned into pRK5 and pcDNA3.1 vectors. Single point mutations were performed by overlap extension PCR (52). The QuikChange tool (Agilent) was utilized to design the primers. Multiple template-based sequential PCRs were used to obtain the 5-HT3A signal peptide-ELIC WT chimera (53).
PCR products, mutagenesis reactions, and ligations were verified using agarose gel electrophoresis and DNA sequencing (DNA Sequencing and Services, University of Dundee). The PCR and cloning reagents were bought from Agilent and Thermo Fisher, respectively.
The genes cloned into their respective vectors were used to transform Escherichia coli DH5␣ cells and grow cultures (500 ml of lysogeny broth medium with 50 g/ml carbenicillin) at 37°C overnight. The cells were harvested (6000 ϫ g, 4°C, 20 min) and used for Maxiprep (Qiagen) to obtain a higher yield of the plasmid.

Cell culture and transfection
HEK293 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 100 g/ml penicillin, and 100 units/ml streptomycin at 37°C and 5% CO 2 . Cells were seeded at low density in 35-mm dishes for electrophysiology. Transfections were performed by calcium phosphate precipitation using 1 g of total cDNA per dish as described previously (26). The cDNAs encoding GABA A R ␤ 3 WT and the mutants were cloned into the pRK5 mammalian expression vector. The cDNA encoding ELIC WT was cloned into the pcDNA3.1 vector. The cDNA that encodes enhanced  (30 M) demonstrate that the function of ␤ 3 C1 is impaired, with values indicated by asterisks significantly different from the ␤ 3 -cryst (n ϭ 10, t test, ␤ 3 C1 p Ͻ 0.0001). However, propofol direct activation was partially restored in ␤ 3 C1 F87Y with values significantly different from ␤ 3 C1 (n ϭ 10, p ϭ 0.007, t test). Error bars represent S.D.

Mutations enabling GABA activation of GABA A ␤3 homomers
green fluorescent protein (0.1 g; in pEGFP vector) was included to identify successfully transfected cells using fluorescence microscopy. Cells were washed with medium 16 h after transfection and used for voltage-clamp electrophysiology after 48 -72 h. The tissue culture reagents were obtained from Invitrogen.

Electrophysiology
The whole-cell configuration of the patch-clamp technique was used to record propofol-or GABA-evoked currents from HEK293 cells transiently expressing GABA A R ␤ 3 WT, GABA A R ␤ 3 mutants, and ELIC WT. Recording electrodes were fabricated from borosilicate glass capillaries with resis- Figure 6. Potentiation of GABA-evoked currents by propofol was unaffected in the ␤ 3 mutants. A, an exemplar current evoked by propofol mediated by ␤ 3 -cryst. GABA had no effect, and the current amplitude evoked by propofol is similar to that seen in the absence of GABA. Cells were voltage-clamped at an electrode potential of Ϫ60 mV unless otherwise stated. Currents were evoked by rapid application of GABA or propofol using the three-pipe Perfusion Fast Step system (Warner Instruments) as described previously (26).
The data were recorded using an Axopatch 200B amplifier (Axon Instruments), low pass-filtered at 2 kHz, digitized at 10 kHz using a Digidata 1320 A interface (Molecular Devices), and acquired using pCLAMP8 software (Molecular Devices).

Data analyses
The analyses were carried out using Clampfit 10 (Molecular Devices), Excel 2011 (Microsoft), and Prism 5 (GraphPad). Peak amplitudes were measured using averaged traces from at least three currents. GABA-evoked current amplitudes were expressed as a percentage of the maximum and plotted as a concentration-response relationship. The following logistic (Equation 1) and bell-shaped equations (Equations 2 and 3) were fitted to the data points to determine the Hill slopes (n H ) and the EC 50 . Peak current densities were calculated by normalizing the peak current amplitude to the cell capacitance. The potentiation effect of propofol was calculated using the following formula, where I pot and I GABA represent the potentiated and control peak current amplitudes, respectively. Activation rates were measured using 10 -90% rise time of the GABA-evoked current. Deactivation rate was calculated by fitting a double-exponential function to the decay phase of the GABA-evoked current as follows, where n are time constants and A n represent the proportion of the particular . The best-fit number of exponential terms was determined using an F-test with confidence at the 95% level. Deactivation rates were provided as weighted values using the following equation.

Statistical analyses
Data are presented as mean Ϯ S.D. Differences of three or more groups were compared using one-way ANOVA. Subsequent multiple pairwise comparisons were performed using the Dunnett's or Tukey's correction. Student's t test was used for other pairwise comparisons. In all cases, p Ͻ 0.05 was considered statistically significant. Statistical analyses were performed in Prism 5.

Comparative modeling
The model for GABA A R ␤ 3 C1 was generated in Modeller v9.13 (54) using the GABA A R ␤ 3 structure (Protein Data Bank (PDB) code 4COF) (12) as a template. The proteins share 99% sequence identity according to MUSCLE sequence alignment (55) and thus are suitable for comparative modeling. The best model according to energy, spatial restraints, and stereochemistry was chosen using the Discrete Optimized Protein Energy (DOPE) score (56) and Ramachandran plot (57).

Molecular docking
Molsoft ICM v.3.8-3 (58) was used to perform docking calculations of GABA into the GABA A R ␤ 3 WT structure (PDB code 4COF) and the GABA A R ␤ 3 C1 model. The preparation of the receptor and ligand models involved adding hydrogens, calculating charges at pH 7.0, deleting waters, and treating the receptor as rigid and the ligand as flexible. The whole receptor or potentially important residues of the binding site were selected (principal side, Asp 95 -Leu 99 , Leu 152 -Thr 161 , and Asn 197 -Arg 207 ; complementary side, Asn 41 -Ala 45 , Met 61 -Tyr 66 , Asn 113 -Leu 118 , Leu 125 -Ala 135 , and Ala 174 -Val 178 ), and a box was created around the selection with a 3-Å distance between the residues and the edges. The results were ranked according to the ICM score, which takes into consideration the quality of the complex based on van der Waals interactions and the internal force-field energy of the ligand (58).