An intersubunit electrostatic interaction in the GABAA receptor facilitates its responses to benzodiazepines

Benzodiazepines are positive allosteric modulators of the GABAA receptor (GABAAR), acting at the α–γ subunit interface to enhance GABAAR function. GABA or benzodiazepine binding induces distinct conformational changes in the GABAAR. The molecular rearrangements in the GABAAR following benzodiazepine binding remain to be fully elucidated. Using two molecular models of the GABAAR, we identified electrostatic interactions between specific amino acids at the α–γ subunit interface that were broken by, or formed after, benzodiazepine binding. Using two-electrode voltage clamp electrophysiology in Xenopus laevis oocytes, we investigated these interactions by substituting one or both amino acids of each potential pair. We found that Lys104 in the α1 subunit forms an electrostatic bond with Asp75 of the γ2 subunit after benzodiazepine binding and that this bond stabilizes the positively modified state of the receptor. Substitution of these two residues to cysteine and subsequent covalent linkage between them increased the receptor's sensitivity to low GABA concentrations and decreased its response to benzodiazepines, producing a GABAAR that resembles a benzodiazepine-bound WT GABAAR. Breaking this bond restored sensitivity to GABA to WT levels and increased the receptor's response to benzodiazepines. The α1 Lys104 and γ2 Asp75 interaction did not play a role in ethanol or neurosteroid modulation of GABAAR, suggesting that different modulators induce different conformational changes in the receptor. These findings may help explain the additive or synergistic effects of modulators acting at the GABAAR.

The ionotropic GABA A receptor (GABA A R) 2 is a pentameric protein belonging to the Cys-loop superfamily family of ligandgated ion channels. Various subunits (␣1-6, ␤1-3, ␥1-3, ␦, ⑀, , and ) combine in multiple combinations to form GABA A Rs. GABA is the predominant inhibitory neurotransmitter in the central nervous system, and its activation of the GABA A R results in anion movement through the integral ion channel pore. Benzodiazepines are used clinically for their sedative, anxiolytic, and anticonvulsant effects. These drugs act at an allosteric site of the GABA A R to positively modulate the channel when activated by an agonist acting at the orthosteric site. Several hypotheses have been suggested to explain the molecular mechanisms of this benzodiazepine enhancement of function, including an increase in the GABA binding affinity of the receptor (1)(2)(3), an increase in GABA efficacy (4,5), and a shift of the receptor toward a "preactivated" state (6).
Different ␣ subunit-containing GABA A receptors account for the various therapeutic indications of benzodiazepines. GABA A Rs containing ␣ 1 subunits are thought to be primarily responsible for the sedative and anticonvulsive effects of benzodiazepines, whereas ␣ 2 -containing GABA A Rs are responsible for their anxiolytic effects (7)(8)(9)(10). The inability of classic benzodiazepines to distinguish between receptors comprising different ␣ subtypes suggests a conserved molecular mechanism of action. Histidine 101 in the ␣ 1,2,3 subunits (103 in ␣ 5 ) plays an important role in benzodiazepine binding with substitution of this residue with arginine rendering receptors less sensitive to benzodiazepines (11). However, the conformational changes in the GABA A R that occur subsequent to benzodiazepine binding are less well understood.
Inter-and intrasubunit electrostatic interactions play important roles in Cys-loop receptor function. For example, electrostatic interactions between residues of adjacent ␣ subunits in the glycine receptor play an important role in its activation (12). Specifically, the aspartate 97 residue is thought to interact with arginine 119 to stabilize the closed state of the glycine receptor, and once this bond is broken after agonist binding, the channel opens. Additionally, electrostatic interactions between aspartic acid 149 and lysine 279 within the same ␣ subunit as well as between aspartic acid 146 and lysine 215 within the same ␤ subunit are implicated in the coupling of GABA binding to the opening of the GABA A R (13,14). Furthermore, glutamic acid 153 and lysine 196 within the same ␤ subunit of the GABA A R may be involved in stabilizing the open state of the receptor (15). Disulfide trapping experiments have led to insights into the conformational changes that benzodiazepines produce in the GABA A R after binding (16); however, thus far an electrostatic interaction has not been identified in the GABA A R that occurs because of this conformational change.
In the current study, we used homology modeling with published structures to produce models of ␣ 1 ␤ 2 ␥ 2 GABA A R. We used these models to identify potential electrostatic interactions occurring before or after the conformational changes produced by benzodiazepine binding, identifying a pair of residues that appear to be interacting in a manner specific for benzodiazepine modulation of the GABA A R.

Molecular modeling identifies possible electrostatic interactions present before and after benzodiazepine binding at the ␣ 1 -␥ 2 subunit interface of the GABA A R
As a starting point for our studies, we used two different models to identify potential electrostatic interactions at the ␣ 1 -␥ 2 subunit interface. The first was based on molecular dynamic modeling performed by Yoluk et al. (17) on the GluCl ligand-gated Cys-loop receptor in the absence of ivermectin (Fig. 1, left). This first model corresponds to the closed, GABAand benzodiazepine-unbound state of the GABA A R in our studies. The second model (Fig. 1, right) is based on the GABAand diazepam-bound GABA A receptor model described by Bergmann et al. (18). Choosing to investigate only charged residues predicted to be 6 Å or less apart, we identified seven interactions that could occur before benzodiazepine binding as well as four interactions that could occur after benzodiazepine binding. Two of these pairs, aspartic acid 56 of the ␣ 1 subunit (␣ 1 Asp 56 ) with arginine 197 of the ␥ 2 subunit (␥ 2 Arg 197 ) and glutamic acid 58 of the ␣ 1 subunit (␣ 1 Glu 58 ) with ␥ 2 Arg 197 were predicted to form electrostatic pairs both before and after diazepam binding. In the present study, we focused on the electrostatic interactions that were predicted to interact closest to the benzodiazepine binding site (Fig. 1, interactions B-F and H-K).

Effects of cysteine substitution on diazepam potentiation of GABA A R function
Diazepam (1 M) enhancement of the effects of a GABA concentration required to produce 5-10% of the maximal response (EC 5-10 ), was tested on a series of cysteine mutants. Cysteine substitution of residues of the ␣ 1 or ␥ 2 subunit predicted to be involved in electrostatic interactions before and/or after diazepam binding resulted in a significant effect of mutation on diazepam potentiation (see Fig. 2 legend for statistics). Replacing ␣ 1 Glu 58 with cysteine (␣ 1 (E58C)) resulted in a significant increase in diazepam potentiation, whereas the ␣ 1 (K104C), ␣ 1 (E137C), ␥ 2 (D75C), and ␥ 2 (R197C) substitutions all resulted in significant decreases in diazepam enhancement (Fig. 2). Six of the other residues substituted with cysteine, ␣ 1 (D56C), ␣ 1 (K105C), ␣ 1 (E165C), ␥ 2 (R97C), ␥ 2 (D120C), and ␥ 2 (R194C), resulted in no significant changes in receptor enhancement by diazepam compared with WT GABA A R. Of the pairs probed, the only hypothesized pair that produced similar changes in diazepam effects upon mutation to cysteine were ␣ 1 (K104C) and ␥ 2 (D75C) (Fig. 1, interaction I). If an electrostatic interaction was occurring between two residues, one would expect similar changes in receptor function if that bond was broken by mutating either residue. For this reason, we focused on the ␣ 1 Lys 104 -␥ 2 Asp 75 pair. Before diazepam binding, ␣ 1 Lys 104 was predicted to be ϳ9 Å from ␥ 2 Asp 75 (Fig. 3, A and B), but after diazepam binding these residues were predicted to move much closer together, to ϳ5 Å apart (Fig. 3C).  (17) and represents the GABA-unbound closed state of the channel. The model on the right is based on the glutamate-bound GluCl crystal structure with a contribution from the ELIC crystal structure (18) and represents the diazepam (in red)-bound receptor. Both models depict the inside of the interface. Labeled interactions represent putative electrostatic interactions of residues 6 Å or less apart that are predicted to occur between residues in the ␣ 1 and ␥ 2 subunits before (A-F) or after (H-K) diazepam binding.

Mechanism of benzodiazepine effects on GABA A receptors
Effects of cysteine substitution on GABA sensitivity at ␣ 1 Lys 104 and ␥ 2 Asp 75 residues GABA concentration-response curves for ␣ 1 (K104C)␤ 2 ␥ 2 , ␣ 1 ␤ 2 ␥ 2 (D75C), and ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors did not significantly differ from those of WT receptors (Fig. 4A). However, one-way ANOVAs revealed that lower GABA concentrations (3 and 10 M) produced greater responses in ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors compared with the single mutants and WT receptors (see Fig. 4 legend for statistics). Despite the model-based hypothesis that the electrostatic interaction between ␣ 1 Lys 104 and ␥ 2 Asp 75 is predicted to occur after diazepam binding, substituting these residues with cysteines could allow a disulfide bond to form spontaneously, which would be able to form between residues at greater distances apart than an electrostatic bond. Therefore, we tested whether the disulfide bond between ␣ 1 (K104C) and ␥ 2 (D75C) had spontaneously occurred. The reducing agent dithiothreitol (DTT) is able to break accessible disulfide bonds. Application of 2 mM DTT to the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptor resulted in an increase in the GABA EC 5 from 3.6 Ϯ 0.4 M before DTT application to 10.5 Ϯ 0.35 M after DTT (Fig. 4B). This is due to the breakage of a single intersubunit disulfide bond as shown in Fig. 5A.
To further probe whether ␣ 1 (K104C) and ␥ 2 (D75C) spontaneously form a disulfide bond in the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptor, we tested propyl methanethiosulfonate (PMTS) for its effects. PMTS is able to covalently bind to free cysteine residues to which it has access. PMTS caused a significant decrease in GABA EC 5 current in the single and double mutant receptors (Fig. 5B, hollow bars with open circles). In the WT and both single cysteine mutant receptors, the effect of PMTS remained unchanged after a prior DTT application (Fig. 5B, hollow bars with triangles). This indicates that in single mutant receptors the cysteine-substituted residues do not form disulfide bonds with endogenous cysteines in GABA A R. Because these single mutant and WT receptors exhibited similar changes in response to PMTS before and after DTT application, we did not test these receptors again 60 min after DTT treatment. For the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptor, a one-way ANOVA revealed a significant effect of PMTS treatment before, 5 min after, and 60 min after DTT treatment (F(2,13) ϭ 108.363, p Ͻ 0.001). Without prior exposure to DTT, application of PMTS resulted in a decrease in current (Fig. 5B, white bar, open circles). However, DTT application before PMTS resulted in an increase in current (solid bar). Waiting 60 min after DTT washout and then applying PMTS resulted in a decrease in current similar to that seen with PMTS application before DTT application. For the double cysteine mutant receptor, the white bar with open circles represents PMTS binding to the single available cysteine residue situated between the ␣ and ␤ subunit interfaces as shown in the illustration on the left in Fig. 5A. When DTT breaks the sole disulfide bond between ␣ and ␥ subunits, PMTS can now bind to up to three free cysteines. Because there was no significant difference between PMTS application before DTT application and 60 min after DTT application, we hypothesize that the disulfide bond breakage produced by DTT is only temporary and that the receptor spontaneously returns to its pre-DTT form within an hour. The reformation of the disulfide bond in the double mutant receptor was also seen experimentally by repeatedly applying the GABA EC 5 to the DTT-treated receptor and observing a gradual increase in current (Fig. 5, C and D). The current produced by a maximally effective concentration of GABA was not changed by applying DTT (data not shown).

Mechanism of benzodiazepine effects on GABA A receptors
Arepeated-measures ANOVA revealed no difference in the concentration-response curve between WT and mutant receptors. However, one-way ANOVAs showed significant effects of mutation at 3 M (F(3,26) ϭ 15.504, p Ͻ 0.001) and 10 M GABA (F(3,26) ϭ 18.163, p Ͻ 0.001) with a Tukey's post hoc test at both concentrations showing a significant increase in response in ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors compared with the other three receptors (***, p Ͻ 0.001). Some symbols are hidden behind other symbols. B, DTT (2 mM; dark symbols and bars) increased the absolute concentration of GABA required to produce an EC 5 response in ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) but not WT receptors. A two-way ANOVA followed by a Tukey's post hoc test revealed a significant effect of DTT treatment on ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors (***, p Ͻ 0.001). Each symbol represents the GABA EC 5 of one oocyte, and each bar represents the mean GABA EC 5 . Error bars represent the S.E.

Mechanism of benzodiazepine effects on GABA A receptors
but produced no changes in responses by WT receptors (Fig. 7,  A and B).

Effects of cysteine substitution on nonbenzodiazepine modulators of the GABA A R
Allosteric modulators of the GABA A R acting at sites other than the benzodiazepine-binding site were next tested to determine the specificity of the electrostatic interactions between ␣ 1 (K104C) and ␥ 2 (D75C). Ethanol and the neurosteroid allopregnanolone produced similar potentiation of the effects of GABA on WT, (Fig. 8). There was no significant effect of DTT treatment on the enhancement of WT or mutant receptors by 200 mM ethanol, 100 nM allopregnanolone (Fig. 8), or 1 M allopregnanolone (data not shown).

Effects of charge reversal of ␣ 1 Lys 104 and ␥ 2 Asp 75 residues on GABA and GABA receptor modulator responses
To test whether reversing the charges of ␣ 1 Lys 104 and ␥ 2 Asp 75 would restore GABA sensitivity, GABA concentrationresponse curves of ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) were compared with those of WT receptors (Fig. 10A). A repeated-measures

Mechanism of benzodiazepine effects on GABA A receptors
ANOVA found a significant difference between WT and ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) concentration-response curves (see Fig. 10A legend for statistics). The average EC 50 value for WT receptors was 86.8 Ϯ 16.5 M, whereas the EC 50 for ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) was increased to 146.3 Ϯ 23.1 M. The charge reversal did not restore to WT levels receptor potentiation by 1 M diazepam or Ro 15-4513 but did restore potentiation by 1 M flunitrazepam and zolpidem (Fig. 10B). Other GABA A receptor modulators (200 mM ethanol and 100 nM allopregnanolone) displayed no changes in potentiation of GABA EC 5-10 after charge reversal compared with WT receptors (data not shown).

Discussion
Signal transduction of ligand-gated ion channels after neurotransmitter binding to its orthosteric site is believed to involve a wave of structural rearrangements (19) in the receptor, and this rearrangement is thought to be separate from the signal transduction pathway produced by allosteric modulators (20). Using molecular modeling to identify potential electrostatic interactions between the ␣ 1 and ␥ 2 subunits, we identified an interaction between ␣ 1 Lys 104 and ␥ 2 Asp 75 that occurs after diazepam binding (Fig. 3). It is likely that these residues interact to stabilize the positively modified state of the receptor and that this interaction is specific for the benzodiazepine signal transduction pathway.
Low concentrations of GABA produced greater responses in ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) than in WT receptors, but this was not seen at higher GABA concentrations (Fig. 4A). This increased response at low GABA concentrations is similar to what one would expect to see in response to coapplication of GABA with a benzodiazepine in WT receptors. Benzodiazepine site agonists increase the effects of low but not higher concentrations of GABA because the ion channel approaches its maximal open probability at saturating GABA concentrations (21).
One would predict that a receptor that is behaving as though a benzodiazepine molecule has already bound would exhibit a decreased response to a coapplication of benzodiazepine with GABA. In the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptor, the two cysteine residues spontaneously formed a disulfide bond. Accordingly, we saw the expected decrease in diazepam, flunitrazepam, and zolpidem potentiation in the double cysteinesubstituted receptors (Fig. 6A, B, and D). After the disulfide bond is broken with DTT, responses to these benzodiazepines increase, suggesting that an electrostatic bond between these residues in WT receptors formed in response to benzodiazepine binding. After DTT application, the response of ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors to flunitrazepam potentiation was rescued to WT levels (Fig. 6B). One reason why potentiation by flunitrazepam, but not diazepam or zolpidem, may be completely rescued following DTT application is that the latter two show weaker modulatory responses after ␣ 1 (K104C)-␥ 2 (D75C) disulfide bond formation than flunitrazepam, i.e.

Mechanism of benzodiazepine effects on GABA A receptors
lower potentiation of GABA responses before DTT application. The conformational rearrangement within the GABA A receptor after benzodiazepine binding most likely depends on the formation of multiple bonds not just ␣ 1 Lys 104 -␥ 2 Asp 75 . Thus, the ␣ 1 Lys 104 -␥ 2 Asp 75 bond may be less important for flunitrazepam potentiation than for diazepam or zolpidem.
Zolpidem is a nonclassical benzodiazepine and at low concentrations is selective for the ␣ 1 subunit-containing GABA A R over those containing other ␣ subunits (22). Disulfide trapping within the ␥ 2 subunit has shown that the conformational change produced by classical benzodiazepines may not be the same as that produced by zolpidem (16). Similarly, there are mutations in the ␣ 1 and ␥ 2 subunits that affect classical but not nonclassical benzodiazepines or vice versa (23)(24)(25)(26). The magnitude of the increase in ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptor potentiation by zolpidem after DTT application was far smaller than the increase seen with the classical benzodiazepines diazepam and flunitrazepam (Fig. 6). We speculate that this may be due to classical and nonclassical benzodiazepines producing overlapping but distinct conformational changes in the GABA A R after binding.
One might hypothesize that the conformational changes produced by benzodiazepines would be different from those produced by inverse agonists such as Ro 15-4513. Indeed, previous studies have shown that this might be the case where disulfide trapping at the ␣-␥ interface of the GABA A R, which affected benzodiazepine potentiation, had no effect on inverse benzodiazepine inhibition (16). Our work supports this hypothesis as the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) mutant GABA A R, which traps the receptor in a "positively modified" state, was not inhibited by Ro 15-4513 as much as WT receptors (Fig. 6C). Once DTT is applied, the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptor is relieved of this positively modified state, and Ro 15-4513 is able to produce inhibition to levels similar to that of WT receptors (Fig. 6C). Ro 15-4513 produced more inhibition in the ␣ 1 ␤ 2 ␥ 2 (D75C) receptor than in the WT, ␣ 1 (K104C)␤ 2 ␥ 2 , and  Each symbol represents the mean from five to six oocytes, and error bars represent the S.E. B, bar graph comparing levels of benzodiazepine enhancement between ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) and WT receptors. The ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) GABA A R was unable to fully rescue responses to WT levels of potentiation by 1 M diazepam and Ro 15-4513 but was able to rescue the responses to 1 M flunitrazepam and zolpidem. Each symbol represents the percent potentiation of the GABA EC 5-10 seen in one oocyte, and each bar represents the mean potentiation observed. Error bars represent the S.E.

Mechanism of benzodiazepine effects on GABA A receptors
␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors. Substituting the ␥ 2 Asp 75 residue with a lysine or alanine residue increased the inhibition by Ro 15-4513 even more so than the cysteine replacement at that residue (data not shown). This decrease suggests that the ␥ 2 Asp 75 residue may be involved in the conformational change produced by Ro 15-4513 as well as a distinct conformational change produced by potentiating benzodiazepines.
After DTT breaks the disulfide bond in ␣ 1 (K104C) ␤ 2 ␥ 2 (D75C) receptors, one might expect the receptor to behave similarly to the ␣ 1 (K104A)␤ 2 ␥ 2 (D75A) receptor. Although the ␣ 1 (K104A)␤ 2 ␥ 2 (D75A) receptor exhibited levels of diazepam and flunitrazepam potentiation similar to WT receptors (Fig.  9B), DTT treatment to ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors did not fully restore levels of diazepam potentiation (Fig. 6A). One possible explanation for this is that DTT treatment, which breaks the disulfide bond by reducing each mutant cysteine, results in two hydrogen-bound cysteine residues that would occupy more volume than alanine residues at those positions, preventing the conformational change produced by diazepam from occurring. Another possibility is that, in the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptor, the spontaneous reformation of a disulfide bond after DTT treatment (Fig. 5, C and D) prevents one from experimentally capturing the maximal amount of enhancement produced by diazepam.
The ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) receptor, bearing two charge-reversing substitutions, displayed a right-shifted GABA concentration-response curve compared with WT receptors (Fig.  10A). Additionally, the ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) receptor did not restore GABA, diazepam, or Ro 15-4513 sensitivity to WT levels (Fig. 10B). This is likely because the ␣ 1 Lys 104 and ␥ 2 Asp 75 residues lie within a pocket of charges and that modifying these residues is preventing other interactions from occurring; i.e. although K104D and D75K substitutions may restore the electrostatic interaction between these residues, there are other charged residues near these sites that may now interact differently with the reversed charge residues compared with the original WT amino acids. Evidently, the ␣ 1 Lys 104 -␥ 2 Asp 75 interaction is not the only interaction that is important for producing the positively modified state of the receptor. If it were, one would see no benzodiazepine potentiation of the receptor after mutating the ␣ 1 Lys 104 and ␥ 2 Asp 75 residues.
One might argue that the data obtained from the alanine substitution experiments at ␣ 1 Lys 104 and ␥ 2 Asp 75 do not fit our overall hypothesis that formation of a bond between these two residues facilitates benzodiazepine effects at the GABA A receptor. Perhaps what is happening is that during the conformational changes produced by benzodiazepine site agonists at WT receptors these two charged residues come close enough together to at least partially neutralize each other's charges. This hypothesis is supported by results obtained using the single alanine substitutions, which retain single charged residues in each pair and display weaker effects of benzodiazepines than those seen in the double alanine mutant (Fig. 9B). A possible explanation may be that the retained charged residue in the single mutants may still be interacting with other nearby charged residues (e.g. ␣ 1 Lys 105 , ␥ 2 Asp 148 , and ␥ 2 Arg 197 ), thus retarding the ability of the receptor to adopt the benzodiazepine-activated conformational state. This would also apply to the single cysteine substitutions, which also display decreased responses to diazepam and flunitrazepam. In scenarios in which the ␣ 1 104 and ␥ 2 75 residues are in close proximity (e.g. cysteines cross-linked), the receptor has already adopted a benzodiazepine positively modified state, and thus adding exogenous benzodiazepine does not have much effect. In cases where these two residues are not initially close together but are capable of moving closer together (e.g. the double alanine substitutions or the double uncross-linked cysteines), a greater effect of applied benzodiazepine will be seen. Lastly, in scenarios in which one or the other of these residues is constrained in its movement (e.g. single substitutions), benzodiazepine effects would be smaller due to the remaining charged residue.
An initial concern was that the receptor mutants were not being incorporated correctly on cell surfaces and that oocytes were expressing primarily ␣ 1 ␤ 2 receptors, not ␣ 1 ␤ 2 ␥ 2 receptors. Previous studies used ZnCl 2 to test for ␥ 2 subunit incorporation as zinc inhibits ␣ 1 ␤ 2 receptors to a greater extent than ␣ 1 ␤ 2 ␥ 2 GABA A Rs (27). However, using this test may not be the most accurate way to test for ␣␤ contamination as even a small fraction of ␣ 1 ␤ 2 receptors present may produce a significant inhibitory effect by zinc (28). Interestingly, the ␣ 1 ␤ 2 receptors display an increase in their GABA-evoked currents after DTT treatment, but ␣ 1 ␤ 2 ␥ 2 receptor currents are unchanged (29). In our study, we saw no change in GABA-evoked currents after DTT treatment in WT and mutant receptors except for ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) receptors for which we actually saw a decrease in GABA-evoked currents (Fig. 5D). This, together with the fact that we injected receptor cDNAs in a 1:1:10 ␣ 1 :␤ 2 :␥ 2 cDNA ratio and still saw a benzodiazepine effect, engenders confidence that the receptors are incorporating WT and mutated ␥ 2 subunits.
One interesting and clinically relevant aspect of this study revolves around the additive and synergistic properties of GABA A R modulators. Benzodiazepines are often coabused with ethanol (30), and the two classes of compounds are thought to act additively or synergistically as central nervous system depressants. Although ethanol is thought to act at the ␣ ϩ -␤ Ϫ interface in ␣␤␦ GABA A Rs, it is not clear whether this is necessarily the case in ␣␤␥ receptors (31). We tested whether mutations that affect both GABA and benzodiazepine responses also produced changes in ethanol responses. In the ␣ 1 (K104C)␤ 2 ␥ 2 (D75C) positively modified receptor, no changes in ethanol potentiation were observed (Fig. 8). Similarly, the charge reversal ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) receptor exhibits ethanol potentiation similar to that of WT receptors (data not shown). These data suggest that the conformational changes in the GABA A R produced by ethanol are experimentally separable from the conformational changes produced by benzodiazepines and that both can occur simultaneously to further enhance receptor function. This provides a possible molecular mechanism for the synergistic/additive effects of benzodiazepines and alcohol.
The neurosteroid allopregnanolone acts as a potent modulator of the GABA A R as well as a direct activator at high concentrations. The binding site for this enhancing action is thought to be within a cavity formed by transmembrane domains 1 and 4 within a single ␣ subunit (32,33). The ␣ 1 (K104C)␤ 2 ␥ 2 (D75C)

Mechanism of benzodiazepine effects on GABA A receptors
positively modified receptor and ␣ 1 (K104D)␤ 2 ␥ 2 (D75K) charge reversal receptor displayed no differences in their sensitivities to the potentiating effects of allopregnanolone compared with WT receptors (Fig. 8). As well as having distinct binding sites, our data suggest that allopregnanolone and benzodiazepines produce distinct conformational changes in the GABA A R.
In summary, our study suggests that an intersubunit electrostatic interaction between ␣ 1 Lys 104 and ␥ 2 Asp 75 occurs after benzodiazepine site agonist binding to help stabilize the GABA A R in a positively modified state. This interaction seems to be more important for classical (nonselective between GABA A R ␣ subunits) benzodiazepines than nonclassical (␣ 1selective) compounds. Additionally, this interaction does not seem to be important for modulators of the GABA A R acting at nonbenzodiazepine sites, suggesting that the ␣ 1 Lys 104 -␥ 2 Asp 75 interaction is specific for benzodiazepine site agents.

Reagents
All chemicals were purchased from Sigma-Aldrich unless otherwise stated below.

Structural modeling
Two homology models of the GABA A R were generated using the Modeler module of Discovery Studio 2016 (Biovia, San Diego, CA) as described previously (34). The first model was of the GABA A R in the benzodiazepine-unbound state. This was built using the GluCl X-ray structure in the absence of ivermectin (17) as a template. This template was produced by starting with the structure of GABA A R with five ivermectin molecules bound (Protein Data Bank code 3RHW), removing the five ivermectin molecules, and then running extensive constrained molecular dynamics simulations using GROMACS 4.5. The resulting model was judged to be in the closed/resting state because the subunits moved closer by 2.0 Å and the pore diameter decreased by 1.2 Å (17). The second homology model illustrated GABA A R after diazepam was bound. This model was based on a GluCl/ELIC X-ray structure that modeled diazepam binding (18). It should be noted that other investigators have proposed a different orientation of diazepam docking at this ␣-␥ interface (35). Because the latter template was built using a novel method of combining coordinates from two X-ray structures, it deserves some comment. The model is based primarily on the glutamate-bound GluCl crystal structure (Protein Data Bank code 3RIF) with a contribution of the ELIC crystal structure (Protein Data Bank code 2VLO) that the authors identified as leading to the best alignment and the best composite structure. Of interest for the present results, in GABA A R ␣ 1 , lysine 104 is in ␤ strand 4, and all coordinates are from GluCl. However, in GABA A R ␥ 2 , aspartic acid 75 is in ␤ strand 2; this residue is conserved in ELIC but not in GluCl. As a result, Bergman et al. (18) used the ELIC structure as a template for residues 75-77.
Because both templates are homopentamers and our goal was to measure intersubunit interactions, we prepared a composite sequence by linking GABA A R ␣ 1 /␤ 2 /␣ 1 /␤ 2 /␥ 2 and aligning the composite with the sequence of the two templates (36).
Then the GABA A R sequences were trimmed to match the length of the template sequences as needed. The two pairs of aligned sequences were submitted to the Modeler module of Discovery Studio 2016.
Both of the resulting homology models were assigned the CHARMm force field in Discovery Studio 2016, minimized, and then subjected to molecular dynamics simulations at 300 K as described previously (34). These two models were analyzed for possible electrostatic interactions using Discovery Studio 2016.

Site-directed mutagenesis
Human cDNAs encoding ␣ 1 , ␤ 2 , and ␥ 2 GABA A R subunits, subcloned into a pBK-CMV vector, were used in this study. Point mutations were introduced in the ␣ 1 and ␥ 2 subunits using a QuikChange site-directed mutagenesis kit (Agilent Technologies, Santa Clara, CA). These mutations were confirmed with dsDNA sequencing.

Harvesting, isolation, and injection of Xenopus laevis oocytes
X. laevis (Nasco, Fort Atkinson, WI) were housed in an Association for Assessment and Accreditation of Laboratory Animal Care International-accredited facility in a room kept at 17°C and under a 12-h light/dark cycle in tanks monitored for water pH and conductivity. Oocytes were surgically removed in accordance with the National Institutes of Health guidelines under a protocol approved by the Institutional Animal Care and Use Committee of the University of Texas at Austin and placed in a hypertonic solution (108 mM NaCl, 1 mM EDTA, 2 mM KCl, and 10 mM HEPES). The thecal and epithelial layers of Stage V and VI oocytes were manually removed using forceps. Isolated oocytes were transferred to a solution (83 mM NaCl, 2 mM KCl, 1 mM MgCl 2 , and 5 mM HEPES) containing 0.5 mg/ml collagenase from Clostridium histolyticum for 10 min to enzymatically remove the follicular layer of the oocytes. The animal poles of oocytes were then injected using a Nanoject II (Drummond Scientific Co., Broomall, PA) with 1.5 ng/30 nl human ␣ 1 , ␤ 2 , and ␥ 2 GABA A R subunit cDNAs in a 1:1:10 ratio. Oocytes were stored singly in 96-well plates containing incubation medium (88 mM NaCl, 1 mM KCl, 2.4 mM NaHCO 3 , 10 mM HEPES, 0.82 mM MgSO 4 ⅐7H 2 O, 0.33 mM Ca(NO 3 ) 2 , 0.91 mM CaCl 2 , 2 mM sodium pyruvate, 0.5 mM theophylline, 10 units/ liter penicillin, and 10 mg/liter streptomycin). The oocytes were kept at room temperature (20°C) and away from light.

Two-electrode voltage clamp electrophysiology
Oocytes expressed GABA A Rs 1-3 days postinjection with cDNA, and all electrophysiological recordings were completed within this time. An oocyte was placed in a 100-l bath containing ND-96 buffer (96 mM NaCl, 2 mM KCl, 1 mM MgCl 2 , 1.8 mM CaCl 2 , and 5 mM HEPES, pH 7.5). The bath was continuously perfused with ND-96 buffer at a rate of 2 ml/min through 18-gauge polyethylene tubing connected to a Masterflex peristaltic pump (Cole Parmer Instruments, Vernon Hills, IL). The tips of two KCl-filled borosilicate glass electrodes, with a resistance of 0.5-10 megaohms, were placed into the animal pole of the oocyte, and it was voltage-clamped at Ϫ80 mV using an OC-725C oocyte clamp (Warner Instruments, Hamden, CT).

Mechanism of benzodiazepine effects on GABA A receptors
Electrophysiological data were collected at a rate of 1 kHz using a digitizer (PowerLab ML866) and LabChart (version 7.4.7) software (both from ADInstruments, Australia).

DTT and H 2 O 2 treatment
DTT and H 2 O 2 were made fresh in ND-96 buffer before each experiment. The GABA EC 5-10 was determined and applied at 3-min intervals until stable responses were obtained. This was repeated after a 2-min DTT (2 mM) application during which the oocyte was unclamped from Ϫ80 mV during the 5-min washout and after a 90-s application of 0.3% H 2 O 2 (oocyte unclamped during the 7-min washout). To measure the effects of DTT and H 2 O 2 on allosteric modulation, the GABA EC 5-10 was determined and ensured to be stable. GABA was then applied in the presence of allosteric modulator as described previously.

PMTS treatment
A 300 mM PMTS (Toronto Research Chemicals, Canada) stock solution in DMSO was stored at Ϫ20°C and diluted to 0.5 mM in ND-96 buffer before each experiment. The GABA EC 5-10 was determined and applied at 3-min intervals until stable responses were observed. Oocytes were then unclamped from Ϫ80 mV and treated with 0.5 mM PMTS for 60 s. After a 2-min wash, oocytes were reclamped to Ϫ80 mV, and the same GABA EC 5-10 was reapplied. Percent changes in current were calculated as ((I GABA after PMTS /I GABA before PMTS ) Ϫ 1) ϫ100. This was repeated after a 2-min treatment with 2 mM DTT, waiting 5 or 60 min after application before applying PMTS.