Differential Protein Mobility of the γ-Aminobutyric Acid, Type A, Receptor α and β Subunit Channel-lining Segments*

The γ-aminobutyric acid, type A (GABAA), receptor ion channel is lined by the second membrane-spanning (M2) segments from each of five homologous subunits that assemble to form the receptor. Gating presumably involves movement of the M2 segments. We assayed protein mobility near the M2 segment extracellular ends by measuring the ability of engineered cysteines to form disulfide bonds and high affinity Zn2+-binding sites. Disulfide bonds formed in α1β1E270Cγ2 but not in α1N275Cβ1γ2 or α1β1γ2K285C. Diazepam potentiation and Zn2+ inhibition demonstrated that expressed receptors contained a γ subunit. Therefore, the disulfide bond in α1β1E270Cγ2 formed between non-adjacent subunits. In the homologous acetylcholine receptor 4-Å resolution structure, the distance between α carbon atoms of 20′ aligned positions in non-adjacent subunits is ∼19 Å. Because disulfide trapping involves covalent bond formation, it indicates the extent of movement but does not provide an indication of the energetics of protein deformation. Pairs of cysteines can form high affinity Zn2+-binding sites whose affinity depends on the energetics of forming a bidentate-binding site. The Zn2+ inhibition IC50 for α1β1E270Cγ2 was 34 nm. In contrast, it was greater than 100 μm in α1N275Cβ1γ2 and α1β1γ2K285C receptors. The high Zn2+ affinity in α1β1E270Cγ2 implies that this region in the β subunit has a high protein mobility with a low energy barrier to translational motions that bring the positions into close proximity. The differential mobility of the extracellular ends of the β and α M2 segments may have important implications for GABA-induced conformational changes during channel gating.

GABA A 1 receptors are allosteric proteins that mediate fast inhibitory neurotransmission in the central nervous system (1)(2)(3). They are members of the Cys-loop receptor ion channel gene superfamily that includes glycine, serotonin type 3 (5-HT 3 ), and nicotinic acetylcholine (ACh) receptors (4 -6). GABA A receptors are formed by five homologous subunits as-sembled around a central channel. Most endogenous receptors contain two ␣, two ␤, and one ␥ subunit arranged in a clockwise orientation ␣␤␣␤␥ when observed from the extracellular end of the channel (7,8). However, expression of just ␣ and ␤ subunits also results in functional receptors with the favored stoichiometry being two ␣ and three ␤ in the order ␣␤␣␤␤ (9 -11). Each subunit has an ϳ200-amino acid, extracellular, N-terminal, ligand-binding domain and a C-terminal, channel-forming domain with four membrane-spanning segments (M1, M2, M3, and M4).
The channel is principally lined by the five ␣-helical M2 segments (12,13). An index numbering system facilitates comparisons between M2 segments of superfamily members (14). The 0Ј position is defined as the positively charged residue located near the cytoplasmic end of the channel, GABA A ␤ 1 R250. The 20Ј position, GABA A ␤ 1 E270, is aligned with the acetylcholine receptor extracellular ring of charge (15) and is predicted by amino acid sequence analysis to be the extracellular end of M2 (16). Experimental evidence indicates that M2 extends two helical turns beyond the 20Ј position (17). The 4-Å resolution structure of the homologous Torpedo ACh receptor confirms this and demonstrates that the 20Ј position lies at the level of the extracellular membrane surface (13). In the 4-Å resolution cryo-EM structure the narrowest region of the closed channel, inferred to be the gate, is near the midpoint, between the 9Ј and 14Ј positions (13). Cysteine accessibility studies in the 5-HT 3 receptor are consistent with this, although similar studies in the ACh receptor concluded that the gate was at the channel's cytoplasmic end (18,19). Evidence from ACh, GABA A , and 5-HT 3 receptors indicates that the structure of the cytoplasmic end of the channel is relatively fixed and rigid (20 -22). This would be consistent with evidence that the size and charge selectivity filters, and the major determinants of single channel conductance are located at the cytoplasmic end of the channel (12,15,(23)(24)(25). In contrast, the extracellular end of the channel undergoes conformational motion due to both thermal protein motion and agonist-induced gating (17,20,21,26). In the cryo-EMderived structure, the extracellular ends of the M2 domains are loosely packed, suggesting that these domains might possess a high degree of flexibility/mobility (13). Consistent with this, substituted cysteine accessibility method studies of the GABA A receptor ␤ 1 subunit M2 domain concluded that the M2 segment extracellular halves were loosely packed and/or highly mobile (21).
We previously used disulfide trapping experiments in ␣␤ receptors to probe thermal protein motion and proximity relationships between M2 segment, channel-lining residues in different subunits (26). The ability for a pair of cysteines to form a disulfide bond depends on the presence of an oxidizing environment and on the collision frequency. The collision frequency depends on the average separation distance of the sulfhydryls, their relative orientation in the protein, and the flexibility/ mobility of the protein in the region of the Cys residues. We used copper phenanthroline (Cu:phen) to create an oxidizing environment. Cu:phen catalyzes the formation of reactive oxygen species, such as superoxide and hydroxyl radicals, from molecular oxygen (27). We showed that at the 20Ј level disulfide bonds formed between Cys substituted for the ␤20Ј but not between Cys substituted for the ␣20Ј. In order for a disulfide bond to form, the Cys ␣ carbons must come to within 5.6 Å of one another (28). Assuming that the 4-Å resolution ACh receptor structure is a good model for the GABA A receptor structure, the average distance between the 20Ј ␣ carbons of residues in adjacent and non-adjacent subunits is 12 and 19 Å, respectively (13). Because there are three ␤ subunits in the ␣␤ receptors used in our previous work, we could not distinguish whether the disulfide bond was forming between Cys substituted in adjacent or in non-adjacent positions. Thus, the extent of the thermal motion could not be determined (26). To resolve this issue the current experiments have been performed in ␣␤␥ receptors where the two ␤ subunits are not adjacent.
The extent of the movements that would be required to explain the disulfide bond formation in our original studies highlights a potential limitation of disulfide trapping experiments. Because disulfide bonds are covalent they may trap relatively rare conformational states of the protein. In the aspartate chemotaxis receptor, a protein of known crystal structure, thermal protein movement allowed disulfide bond formation between pairs of engineered Cys whose ␣ carbons were separated in the crystal structure by 15 Å (28). Thus, disulfide trapping may provide insight into the extent of thermal motion, but it does not necessarily measure the average separation distances. To address this issue, in the present work we have measured the Zn 2ϩ binding affinity of receptors containing pairs of engineered Cys. Pairs of Cys can form bidentate, high affinity Zn 2ϩ -binding sites if they are positioned appropriately. The Zn 2ϩ affinity of these sites will depend on the orientation of the Cys sulfur atoms, their average separation distance, and the energy needed to distort the average protein structure to bring the Cys into position to bind the Zn 2ϩ ion. The Zn 2ϩ affinity will lie between the picomolar range, the affinity of Zn 2ϩ for peptides containing four Cys Zn 2ϩ fingerbinding protein sequences (29), and the 10 -1000 M range, the Zn 2ϩ affinity of single Cys (10,30,31). In crystal structures of high affinity Zn 2ϩ -binding sites the Cys ␣ carbons are separated by about 5-7 Å. The non-covalent nature of this interaction provides a better estimate of the average separation and/or the energy required to distort the average conformation to the structure necessary for high affinity binding.
Here we report that in ␣␤␥ receptors disulfide bonds form when all five subunits contain 20Ј engineered Cys residues, but for the single Cys mutants they only form when the engineered Cys is in the ␤ subunit, not when it is in ␣ or ␥. Consistent with this, a high affinity Zn 2ϩ -binding site is only formed when the engineered Cys is in the ␤ subunit. These results provide insights into the extent and asymmetric nature of the thermal motion near the extracellular ends of the GABA A receptor channel-lining M2 segments and have implications for the channel gating process.
Reagents-A 100 mM stock solution of GABA (Sigma) in water was aliquoted and stored at Ϫ20°C. 1 M stock solutions of dithiothreitol (DTT; Sigma) and o-phenanthroline (Sigma) were made in nominally calcium-free frog Ringer's solution (CFFR: 115 mM NaCl, 2.5 mM KCl, 1.8 mM MgCl 2 , and 10 mM HEPES, pH 7.5) and Me 2 SO, respectively, aliquoted, and stored for not more than 1 month at Ϫ20°C. A stock solution of 100 mM CuSO 4 was made in water. CuSO 4 and o-phenanthroline were mixed in CFFR directly before use to a final concentration of 100 M CuSO 4 and 400 M o-phenanthroline, expressed as 100:400 M Cu:phen. A 100 mM stock solution of N-ethylmaleimide (NEM, Sigma) was made in CFFR directly before use. A 10 mM diazepam stock solution was made in Me 2 SO and stored at Ϫ20°C. A 100 mM ZnCl 2 (Sigma) stock solution was made in water with 10 mM HCl, to prevent the precipitation of Zn 2ϩ (OH Ϫ ). Tricine (Sigma) was diluted directly into 1ϫ buffer at a concentration on 10 mM. Stock solutions of 500 mM N-(2-acetamido)iminodiacetic acid (Sigma), pH 7.3, and 100 mM diethylenetriaminepentaacetic acid (Sigma), pH 7.3, were made in water.
Electrophysiology-Two-electrode voltage clamp recordings were conducted at room temperature in a 250-l chamber continuously perfused at 5-6 ml/min with CFFR solution. For the Zn 2ϩ dose-response curves, CFFR was replaced by a buffer containing 100 mM NaCl, 2.8 mM KCl, 0.3 mM BaCl 2 , and 5 mM HEPES, pH 7.3. Currents were recorded from oocytes using two-electrode voltage clamp recording at a holding potential of Ϫ60 mV. The ground electrode was connected to the bath via a 3 M KCl/Agar bridge. Glass microelectrodes filled with 3 M KCl had a resistance of Ͻ2 M⍀. Data were acquired and analyzed using a TEV-200 amplifier (Dagan Instruments, Minneapolis, MN), a Digidata 1322A data interface (Axon Instruments, Union City, CA), and pClamp 8 software (Axon Instruments). Currents were elicited by applications of GABA separated by at least 5 min of CFFR wash to allow complete recovery from desensitization. Currents were judged to be stable if the variation between consecutive GABA pulses was Ͻ5%.
Diazepam-induced Current Enhancement-The GABA dose-response relationship was determined on each cell. A GABA EC 5 concentration was applied in the absence and presence of 1 M diazepam. To prevent disulfide bond formation receptors were kept reduced by the application of 10 mM DTT at the beginning of the experiment and between pulses of reagents. The amount of potentiation by diazepam was calculated by the equation, % potentiation ϭ (I Dzpm /I) ϫ 100, where I Dzpm and I are the GABA-induced currents with and without diazepam, respectively. The diazepam potentiation is presented as the mean Ϯ S.E.
NEM Inhibition of Zn 2ϩ Binding-After a control pulse of 30 ⌴ GABA, 5 M Zn 2ϩ was applied for 1 min, followed by co-application of GABA and Zn 2ϩ . To eliminate Zn 2ϩ binding to the engineered cysteines, receptors were treated with 100 M NEM for 5 min. We then reapplied GABA, first alone and then in the presence of Zn 2ϩ . (In some cases the effect of NEM on Zn 2ϩ -induced inhibition was tested on separate cells. Because this did not alter the outcome, the results of all experiments were combined.) Receptors were kept reduced by the application of 10 mM DTT at the beginning of the experiments and between pulses of reagents. Prior to every pulse of GABA or GABA plus Zn 2ϩ , receptors were reduced with DTT (10 mM, 5-10 min) and either washed (GABA pulses) or treated with Zn 2ϩ (GABA plus Zn 2ϩ pulses) for 1 min. The degree of inhibition by Zn 2ϩ both before and after NEM treatment was determined by the equation, % inhibition ϭ [1 Ϫ (I Zn /I)] ϫ 100, where I Zn and I are the GABA-induced currents with and without Zn 2ϩ , respectively. The % inhibition is given as mean Ϯ S.E.
Disulfide Bond-induced Inhibition-Once GABA-induced control currents had stabilized, 10 mM DTT was applied for 10 min. GABA was reapplied directly following DTT to determine the extent of potentiation produced by reduction of spontaneously formed disulfide bonds. The cells were treated with 100:400 M Cu:phen for 3-6 min, and then two or more GABA test pulses were applied. A second application of DTT was followed by a final pulse of GABA, to assure the reversibility of the Cu:phen effects.
Intersubunit disulfide bond formation was completely spontaneous and did not require catalysis by Cu:phen. Cu:phen only increased the rate of disulfide bond formation. Therefore, the extent of inhibition attributed to the presence of disulfide bonds was calculated by the equation, % inhibition ϭ [(I DTT Ϫ I)/I DTT ] ϫ 100, where I DTT is the GABA-induced current following application of DTT, and I is the current prior to DTT treatment. A saturating concentration of GABA was used for all experiments, except where specified. The range of GABA concentrations was between 0.3 and 10 mM. The maximal currents and % effect for each reagent are presented as the mean Ϯ S.E.
Reoxidation Rates-A control pulse of a low concentration of GABA (below the GABA EC 50 ) was followed by a 10-min application of 10 mM DTT. This led to an increase in the size of the currents. Pulses of GABA were then applied at 5-to 10-min intervals to monitor the return of the currents to their unreduced levels. The peak currents were fit with the equation, I t ϭ (I 0 Ϫ I ϱ )exp(Ϫt/) ϩ I ϱ , where I t is the current at time t, I 0 is the initial current, I ϱ is the final current, and is the time at which 63% of the total current decay occurred.
Zn 2ϩ Dose-response Curves-A control pulse of GABA was followed by coapplications of GABA with increasing concentrations of Zn 2ϩ . Prior to every pulse of GABA or GABA plus Zn 2ϩ , receptors were reduced with DTT (10 mM, 5-10 min) and either washed or treated with Zn 2ϩ for 1 min. Currents were normalized to the initial, control current, and fit with the Hill equation, where I is the current, I max is the control current, IC 50 is the Zn 2ϩ concentration that produces half-maximal inhibition, Zn is the Zn 2ϩ concentration, and n is the Hill coefficient. Fits were performed in Prism 3.02 (GraphPad Software, San Diego, CA).
To avoid artifacts caused by potential submicromolar levels of heavy metal contamination, Zn 2ϩ dose-response curves were performed in the presence of heavy metal chelators as described (38,39). 10  Statistics-All statistical analyses were performed in Prism 3.02 using a one-way analysis of variance followed by the Newman-Keuls multiple comparison test.

A ␥ Subunit Is Present in Functional Receptors-In ␣␤␥
receptors ␤ subunits are found only in non-adjacent positions, whereas in ␣␤ receptors there is a pair of ␤ subunits in adjacent positions (see the introduction). Therefore, if a disulfide bond formed between ␤ subunits, we could only know that this bond occurred between non-adjacent subunits if we also knew that a ␥ subunit was present in the receptor. We tested for the presence of a ␥ subunit using two approaches, diazepam potentiation and Zn 2ϩ inhibition. In ␣␤␥ receptors, 1 M diazepam is reported to potentiate currents induced by an EC 5 concentration of GABA by more than 100%, whereas ␣␤ receptors are unaffected by diazepam (40,41). Zn 2ϩ also enables us to distinguish between ␣␤ and ␣␤␥ receptors, because ␣␤ receptors have a Zn 2ϩ IC 50 of ϳ0.5 M, whereas ␣ 1 ␤ 1 ␥ 2 receptors are insensitive to Zn 2ϩ (42,10). The high affinity Zn 2ϩ -binding site is formed in ␣␤ receptors by the ␤ 1 His-267 (M2 17Ј) in the adjacent ␤ subunits. In ␣␤␥ receptors there are no adjacent ␤ subunits and therefore no high affinity Zn 2ϩ -binding site.
We tested the effects of diazepam on two populations of the mutant ␣␤20ЈC␥ receptors. Half of the cells were injected with a 1:1:1 and half with a 1:1:10 molar ratio of ␣, ␤, and ␥ mRNA. If the cells injected with a 1:1:1 ratio of mRNA expressed the ␥ subunit in all cell surface receptors, then there should be no increase in the amount of diazepam potentiation in the cells injected with 1:1:10 compared with 1:1:1. To assure that the presence of spontaneously formed disulfide bonds did not interfere with the effects of diazepam and Zn 2ϩ , the reducing (D) mRNA was tested before and after treatment with 100 ⌴ NEM (5 min). Zn 2ϩ inhibition was nearly abolished by NEM in ␣20ЈC␤20ЈC␥20ЈC, but not ␣20ЈC␤20ЈC, proving that a ␥ subunit was present in receptors injected with ␥20ЈC mRNA. Bars above traces indicate application of reagent. Leak currents have been subtracted. Current is not shown during application of Zn 2ϩ alone or NEM. Holding potential: Ϫ60 mV. agent DTT was applied for several minutes before the application of all reagents. As seen in Fig. 1 (A and B), the degree of diazepam-induced potentiation in the two populations of receptors was not significantly different (1:1:1, 179 Ϯ 70% (n ϭ 3); 1:1:10, 108 Ϯ 18% (n ϭ 3)). Therefore, in our hands, the majority of the GABA-induced current from oocytes injected with an equimolar ratio of the three subunits arises from receptors containing a ␥ subunit. Similar results were obtained with other Cys mutant receptors used in this study.
To further support the conclusion that the functional cell surface receptors used in this study contained a ␥ subunit, we examined the extent of inhibition by Zn 2ϩ . Two different mutants were used for the experiments with Zn 2ϩ : the double Cys mutant, ␣20ЈC␤20ЈC, and the triple Cys mutant, ␣20ЈC␤20ЈC␥20ЈC. In both cases we injected equimolar amounts of mRNA for each subunit. 5 M Zn 2ϩ should inhibit more than 50% of the current in the ␣20ЈC␤20ЈC mutant while having no effect on the current of the ␣20ЈC␤20ЈC␥20ЈC mutant. Surprisingly, the two mutants showed similar amounts of Zn 2ϩ -induced inhibition: 85 Ϯ 4% (n ϭ 2) and 69 Ϯ 8% (n ϭ 5) for ␣20ЈC␤20ЈC and ␣20ЈC␤20ЈC␥20ЈC, respectively ( Fig. 1, C  and D). Because the engineered cysteines in these mutants could potentially bind Zn 2ϩ , we retested the effects of Zn 2ϩ after exposing the reduced receptors to the alkylating agent NEM (100 M, 5 min). Alkylation should abolish the ability of cysteine to bind heavy metals. NEM diminished currents in both mutants. However, following alkylation with NEM, 5 M Zn 2ϩ inhibited the remaining currents in ␣20ЈC␤20ЈC and ␣20ЈC␤20ЈC␥20ЈC by 60 Ϯ 2% (n ϭ 3) and 6 Ϯ 2% (n ϭ 4), respectively. Therefore, we infer that double mutant ␣20ЈC␤20ЈC cells retained a high affinity for Zn 2ϩ following alkylation, because the native high affinity, Zn 2ϩ -binding site composed of ␤ 1 His-267 (17Ј) in adjacent ␤ subunits (42, 10) are insensitive to NEM. In contrast, in the triple mutant, ␣20ЈC␤20ЈC␥20ЈC, after NEM alkylation Zn 2ϩ no longer inhibited significantly, because in the presence of the ␥ subunit there are not adjacent ␤ subunits to form a high affinity Zn 2ϩbinding site. The overall conclusion from the experiments with Zn 2ϩ and diazepam is that, when injected with a 1:1:1 ratio of ␣, ␤, and ␥ mRNA, the large majority of cell surface receptors contain a ␥ subunit. The important implication of this finding for the experiments described below is that two ␣ subunits are not in adjacent positions nor are the two ␤ subunits.
We next determined whether the oxidizing agent Cu:phen, which catalyzes disulfide bond formation, could reverse the effects of DTT in ␣20ЈC␤20ЈC␥20ЈC. A 3-min application of 100:400 M Cu:phen decreased the ␣20ЈC␤20ЈC␥20ЈC currents by 89 Ϯ 12% (n ϭ 3). This was not significantly different than their initial levels. In contrast, a similar Cu:phen application had no effect on wild-type currents (n ϭ 3) (Fig. 2, A and B). A subsequent DTT application restored the mutant receptor currents to within 5 Ϯ 2% (n ϭ 3) of the levels produced by the first DTT application.
To ensure that the effect of DTT was not due to chelation of contaminating heavy metals in the buffer, we tested whether the metal chelator EGTA could also potentiate currents in ␣20ЈC␤20ЈC␥20ЈC. DTT potentiated currents by 981 Ϯ 162%, whereas, when applied for several minutes, 1 mM EGTA only potentiated currents by 35 Ϯ 16% (n ϭ 3). Therefore, we conclude that ␣20ЈC␤20ЈC␥20ЈC formed one or two spontaneous disulfide bonds, which could be reduced by DTT and reformed by Cu:phen. Disulfide Bond Formation in Receptors with Single 20Ј Cys Mutant Subunits-Further experiments were aimed at gaining insight into the subunits involved in disulfide bond formation in ␣20ЈC␤20ЈC␥20ЈC. We first examined mutant receptors containing a Cys in only one subunit: ␣20ЈC␤␥, ␣␤20ЈC␥, and ␣␤␥20ЈC. Disulfide bonds formed spontaneously in ␣␤20ЈC␥. The initial I max before DTT application, 1078 Ϯ 214 nA (n ϭ 6), was smaller than in the wild-type receptors, and after reduction with DTT I max increased to 3251 Ϯ 395 nA (n ϭ 6) similar FIG. 2. Disulfide bonds formed in ␣20C␤20C␥20C and ␣␤20C␥, but not wild-type, receptors. The effects of DTT (10 mM, 10 min) and Cu:phen (100:400 M, 3-6 min) were tested on the maximal GABA-induced current of wild-type (A), ␣20ЈC␤20ЈC␥20ЈC (B), and ␣␤20ЈC␥ (C) receptors. The initial currents of both mutants were smaller than that of wild-type receptors. DTT increased the mutant currents, and subsequent Cu:phen application reversed the effects of DTT. Subsequent DTT application increased currents to an extent comparable to the initial DTT application. Neither reagent significantly altered currents in wild-type receptors. Bars above traces indicate application of reagent. Leak currents have been subtracted. Current is not shown during application of DTT and Cu:phen. Holding potential: Ϫ60 mV.
to wild-type currents (Figs. 2C and 3A). Furthermore, application of 100:400 M Cu:phen returned currents to within 3 Ϯ 23% of the initial untreated levels (n ϭ 3), and a second DTT application increased currents to within 33 Ϯ 5% (n ϭ 2) of those after the initial DTT application.
In contrast to ␣␤20ЈC␥, in ␣20ЈC␤␥ the initial I max was 3422 Ϯ 664 nA (n ϭ 4) comparable to that of wild-type receptors. Application of DTT (3784 Ϯ 813 nA, n ϭ 4) or Cu:phen (2655 Ϯ 761 nA, n ϭ 3) did not significantly alter GABA I max from the initial I max (Fig. 3A). As expected, given that there is only one ␥ subunit per receptor, there was no evidence for disulfide bond formation in ␣␤␥20ЈC receptors. The initial I max was 2779 Ϯ 572 nA (n ϭ 4). Currents were unaffected by application of either DTT (2791 Ϯ 617 nA, n ϭ 4) or Cu:phen (2317 Ϯ 640 nA, n ϭ 3) (Fig. 3A). From the experiments on receptors containing a single mutant subunit we conclude that at the 20Ј position intersubunit disulfide bonds formed between ␤ subunits, but not between ␣ subunits.
Disulfide Bond Formation in Receptors with Two Subunits Containing 20Ј Cys Mutants-We tested mutants containing Cys in two different subunits for their ability to form disulfide bonds. The initial I max of ␣20ЈC␤␥20ЈC, ␣20ЈC␤20ЈC␥, and ␣␤20ЈC␥20ЈC were 1781 Ϯ 209 nA (n ϭ 10), 1545 Ϯ 215 nA (n ϭ 11), and 567 Ϯ 106 (n ϭ 11) respectively. All were significantly less than the wild-type receptor I max . Reduction with DTT increased the currents to levels similar to those of wild type bringing the currents to 3414 Ϯ 265 nA, 3907 Ϯ 195 nA, and 2846 Ϯ 245 nA for ␣20ЈC␤␥20ЈC, ␣20ЈC␤20ЈC␥, and ␣␤20ЈC␥20ЈC, respectively (Fig. 3A). Cu:phen reversed the effects of DTT, and a second DTT application duplicated the effects of the first DTT application (data not shown). We conclude that disulfide bonds formed in all three double mutants.
The extent to which disulfide bond formation at the 20Ј level inhibits GABA-induced currents can be quantified for each mutant using the equation % inhibition ϭ [(I DTT Ϫ I)/I DTT ] ϫ 100, where I and I DTT represent the GABA currents before and after DTT, respectively. Because the currents of untreated receptors (spontaneously oxidized) were of the same magnitude as those of receptors treated with Cu:phen, we used the initial currents in our calculation. The mutants containing a disulfide bond fall into three significantly different groups based on the extent of inhibition (Fig. 3B). Group 1 contains the mutant ␣20ЈC␤␥20ЈC with 49 Ϯ 4% inhibition. Group 2 contains ␣␤20ЈC␥ and ␣20ЈC␤20ЈC␥ with 68 Ϯ 6% and 61 Ϯ 5% inhibition, respectively. Group 3 contains ␣␤20ЈC␥20ЈC and ␣20ЈC␤20ЈC␥20ЈC with 81 Ϯ 3% and 89 Ϯ 2% inhibition, respectively. From the results above we infer that 1) the two ␤ subunits in a receptor can form a disulfide bond with one another (Fig. 3B, bar #3); 2) the disulfide bond in ␣20ЈC␤␥20ЈC must be between an ␣ and a ␥ subunit (Fig. 3B, bar #6), because a disulfide bond does not form in either of the single mutants ␣20ЈC␤␥ or ␣␤␥20ЈC (Fig. 3B, bars #2 and #4); and 3) some portion of the disulfide bonds found in ␣␤20ЈC␥20ЈC must be between a ␤ and a ␥ subunit, because there is more inhibition in the double mutant than the single ␤ mutant (Fig. 3B, compare bars #3 and #7).
The Rate of Disulfide Bond Formation Is Fastest in ␣20ЈC␤20ЈC␥-To learn more about the relative proximity and mobility of the different subunits around the channel, we measured the rates of spontaneous disulfide bond formation in ␣20ЈC␤20ЈC␥, ␣20ЈC␤␥20ЈC, ␣␤20ЈC␥20ЈC, and ␣␤20ЈC␥. As shown for ␣␤20ЈC␥20ЈC in Fig. 4A, a test pulse of GABA was followed by a 10-min application of 10 mM DTT. Pulses of GABA were then applied every 5-10 min, depending on the mutant. Over the course of several minutes the currents returned to their initial values, indicating the spontaneous reformation of the disulfide bonds. The peak currents of all the pulses were fit with the single exponential equation (Fig. 4B). The values for the mutants were as follows (n Ն 4): ␣20ЈC␤20ЈC␥, 3 Ϯ 1 min; ␣20ЈC␤␥20ЈC, 23 Ϯ 7 min; ␣␤20ЈC␥20ЈC, 15 Ϯ 2 min; and ␣␤20ЈC␥, 17 Ϯ 4 min. ␣20ЈC␤20ЈC␥ is the only mutant to have a significantly different from the other mutants. The difference in disulfide bond formation rates between ␣20ЈC␤20ЈC␥ and ␣␤20ЈC␥ implies that at least some of the disulfide bonds in ␣20ЈC␤20ЈC␥ are between Cys in ␣ and ␤ subunits. This implies a higher collision rate between the ␣ and ␤ Cys than between the non-adjacent ␤ Cys in ␣␤20ЈC␥. This faster rate may be due to disulfide bond formation between adjacent ␣ and ␤ subunits.
␣␤20ЈC␥ Forms a High Affinity, Zn 2ϩ -binding Site-The intersubunit disulfide bond that forms between ␤20ЈCys in ␣␤20ЈC␥ receptors indicates that the combined movement of the two non-adjacent ␤ M2 segments was sufficient to traverse the channel diameter. The frequency of this event is unknown, because disulfide bonds can trap the receptor in a rare conformation. To address this issue we sought to determine whether these Cys could form a high affinity Zn 2ϩ -binding site. Due to the significantly lower energy involved in the Zn 2ϩ -Cys interaction compared with a covalent disulfide bond, Zn 2ϩ would be unable to trap the rare conformations that could be trapped with a disulfide bond. These experiments were carried out with reduced receptors to ensure that the Cys were fully available to bind Zn 2ϩ . In addition, the buffer contained heavy metal chelators to remove any trace metals that might compete with Zn 2ϩ for binding to the Cys.
␣20ЈC␤␥ receptors also showed some increased sensitivity to Zn 2ϩ compared with wild-type receptors (n ϭ 4; Fig. 5B). However, because the predicted IC 50 would be greater than 100 M a complete Zn 2ϩ dose-response relationship was not determined. The increase over wild-type sensitivity was abolished if, in addition to adding a Cys at ␣20Ј, the glutamate at the ␤20Ј position was replaced with an asparagine: 100 M Zn 2ϩ inhibited GABA-induced currents in ␣20ЈC␤␥ by 32 Ϯ 4% (n ϭ 4), but only altered the ␣20ЈC␤20ЈN␥ currents by 1 Ϯ 2% (n ϭ 3). Therefore, in ␣20ЈC␤␥, a low affinity, bidentate, Zn 2ϩ -binding site formed at the 20Ј position between the engineered cysteine in the ␣ subunit and the native glutamate in the ␤ subunit, but not solely between ␣20Ј Cys.
An Intersubunit Disulfide Bond Does Not Form in ␣␤17ЈC␥-The 17Ј position is one ␣ helical turn down from the 20Ј position. Because the distance across the channel between 17Ј residues should be shorter than it is between 20Ј residues (13), we tested the ability of ␣␤17ЈC␥ to form a disulfide bond.
DTT and Cu:phen had no effects on maximal GABA-induced currents in ␣␤17ЈC␥ (DTT: ϩ6 Ϯ 2%, n ϭ 3; Cu:phen: Ϫ6 Ϯ 3%, n ϭ 3). Receptors were also unaffected by DTT using an EC 10 concentration of GABA, which is more sensitive to modifications that affect gating (wild-type: ϩ46 Ϯ 1% (n ϭ 2); mutant: ϩ19 Ϯ 6% (n ϭ 3)). Thus, it appears that ␤-␤ disulfide bonds do not form between non-adjacent ␤ subunits at the 17Ј position in ␣␤␥ receptors. DISCUSSION We used disulfide trapping and Zn 2ϩ binding to study the mobility of the GABA A receptor M2 segments in ␣␤␥ receptors. In these receptors there are two ␣, two ␤, and one ␥ subunits. The two ␣ subunits are in non-adjacent positions around the channel axis as are the ␤ subunits (7)(8)(9)(10)(11). Our experiments showed that disulfide bonds formed between Cys residues substituted for ␤ 1 Glu-270 (20Ј) but not between Cys substituted for the aligned ␣ 1 subunit residue ␣ 1 Asn-275 (20Ј). Disulfide bond formation between the engineered ␤ Cys implies that the collision frequency between the engineered ␤20Ј Cys is significantly higher than between the ␣ engineered 20Ј Cys. In the FIG. 5. The engineered cysteines in ␣␤20C␥ form a high affinity, bidentate Zn 2؉ -binding site. A, receptors were exposed to increasing Zn 2ϩ concentrations. Prior to every pulse of GABA or GABA plus Zn 2ϩ , receptors were reduced with DTT (10 mM, 5 min) and either washed or treated with Zn 2ϩ for 1 min. A pulse of 30 M GABA was then applied in the presence of Zn 2ϩ . All experiments were done in the presence of metal chelators (see "Experimental Procedures"). Bars above traces indicate application of reagent. Leak currents have been subtracted. Current is not shown during application of Zn 2ϩ alone or DTT. Holding potential: Ϫ60mV. B, Zn 2ϩ dose-response curves for ␣20ЈC␤␥ (triangles) and ␣␤20ЈC␥ (squares). Data were normalized to the current in the absence of Zn 2ϩ and plotted against the Zn 2ϩ concentration. The IC 50 for ␣␤20ЈC␥ was determined by fitting the doseresponse curve with the Hill equation (line). A fit for ␣20ЈC␤␥ was performed to obtain a partial curve.
ACh receptor 4-Å structure the 20Ј residues have a similar orientation to and distance from the channel axis (13), suggesting that these are not the bases for the disparity between ␣ and ␤. Thus, at this level in the channel the ␤ subunit M2 segments must be more mobile and/or more flexible than the ␣ subunit M2 segments in ␣␤␥ receptors. This implies that the ␤ M2 segments are less tightly packed with the rest of the protein than the ␣ M2 segments (43,21).
In the ACh receptor 4-Å structure the ␣ carbons of nonadjacent 20Ј residues are ϳ19 Å apart (Fig. 6) (13). Because disulfide trapping involves formation of a covalent bond, it does not provide information on the energetics of bringing the 20Ј residues to within ϳ5 Å necessary to form the disulfide bond. High affinity Zn 2ϩ binding involves a non-covalent interaction with pairs of engineered Cys residues. The ␣ carbon separation of the two Cys residues is comparable in a bidentate Zn 2ϩbinding site and in a disulfide bond (29, 44 -47). However, unlike disulfide bonds, the energetics of apposing two Cys residues can be measured through the affinity of the resultant binding site for Zn 2ϩ . The higher the Zn 2ϩ affinity the lower the energy barrier to bringing the two Cys residues close enough to form a bidentate-binding site. Bound Zn 2ϩ ions usually display tetrahedral coordination (44,47,48). The affinity of a site for Zn 2ϩ depends on the number of chelating Cys residues. The Zn 2ϩ affinity of sites containing a single Cys residue is generally in the tens of micromolar to millimolar concentration range (29,31,49), whereas the Zn 2ϩ affinity of proteins containing four Cys residues chelating a Zn 2ϩ ion range from 10 Ϫ18 to 10 Ϫ12 M (38, 50 -52). The Zn 2ϩ affinity for ␤ 1 E20ЈC containing receptors was 34 nM. To achieve this affinity the Zn 2ϩ must be bound by both Cys.
It is difficult to know what the theoretical maximum affinity of two ideally positioned Cys residues is for Zn 2ϩ in part because there are no structural Zn 2ϩ -binding sites with just two Cys ligands. With this number we could calculate the amount of energy lost to protein distortion by Zn 2ϩ binding to the ␤20ЈCys receptors to give the measured affinity of 34 nM. In proteins containing two Cys residues the Zn 2ϩ affinity ranges from nanomolar to micromolar (10,36,38). Thus, the 34 nM affinity that we have measured is toward the higher end of measured affinities for two Cys binding. This implies that there is a relatively small energy barrier to bringing the two ␤ Cys from their 19-Å separation distance in the ACh receptor structure to the optimal separation for Zn 2ϩ binding.
Disulfide Bonds between ␣-␤, ␣-␥, and ␤-␥ 20Ј Cys Mutants-Although the ␣20ЈC␤␥ mutant did not form an intersubunit disulfide bond, the ␣20ЈCys was able to form disulfide bonds with the ␤20ЈCys and with the ␥20ЈCys. We infer the formation of these ␣-␤ and ␣-␥ disulfide bonds, because the extent of inhibition following oxidation was different in the double Cys mutant ␣20ЈC␤20ЈC␥ than in the single Cys mutant ␣␤20ЈC␥ and there was disulfide bond formation in ␣20ЈC␤␥20ЈC (Fig.  3). In both cases the disulfide bonds could form either between adjacent or between non-adjacent subunits. At present we cannot distinguish between these possibilities. If the disulfide bonds form between ␣20ЈCys and a Cys in an adjacent ␤ or ␥ subunit the M2 segments to which the residues are attached would need to both rotate and move ϳ7 Å based on the ACh receptor structure (Fig. 6). Therefore, while the ␣M2 20Ј region, in conjunction with the ␤M2 or ␥M2 regions, is sufficiently flexible to move the 7 Å necessary to disulfide bond with a Cys on an adjacent subunit, the 14 Å required for disulfide bonding with a non-adjacent subunit, i.e. the other ␣ subunit, appears to be too great a distance for the ␣M2 segments to overcome.
Gating and Spontaneous Protein Movement-The mobility of the extracellular ends of M2 that we have detected may be related to the channel gating process. In the 4-Å ACh receptor structure the channel gate is in the region between the 9Ј and 14Ј levels (13). Transduction of agonist binding in the extracellular domain to the gate may proceed through the extracellular end of M2. The strong inhibitory effect of a single disulfide bond at the 20Ј position ranging from 49% inhibition in ␣20ЈC␤␥20ЈC to 81% inhibition for ␣␤20ЈC␥20ЈC demonstrates the importance of this region in channel gating. The motion that we have detected may represent the unsynchronized fluctuation of the extracellular ends of the M2 segments between their closed and open state conformations. Channel opening would require the concerted movement of all five M2 segments away from the channel axis into their open state conformation, an event that rarely occurs in the absence of agonist. The movement of the M2 segments may be similar to the spontaneous conformational fluctuations that the voltage sensing S4 segments undergoes in voltage-dependent K ϩ channels as they sense the membrane potential on the two sides of the membrane (53)(54)(55).
Our observations of asymmetric motion in the ␤ and ␣ subunits raises the question of whether channel gating involves a larger movement of the ␤ subunit M2 segments than of the ␣ M2 segments. The GABA A ␤ subunits form the principle portion of the agonist-binding sites, analogous to the ACh receptor ␣ subunits (7). A greater fraction of the agonist surface area interacts with the principle subunit-binding site (7,56). Whether this causes a greater movement in the ␤ M2 segment is unknown. Differences in the potential coupling of the ␤ and ␣ extracellular domains to the membrane-spanning domains have been observed (57,58). Whether these differences relate to the extent of M2 segment movement during gating is unknown at present. Perhaps consistent with our finding of greater conformational change in the principle subunit, Unwin and colleagues (59), based on differences between the extracellular domain structure of the Torpedo ACh receptor, which is  (13). Native residues have been replaced with cysteines. The van der Waals surfaces are only shown for the cysteines (yellow, sulfur; red, oxygen; blue, nitrogen; and white, carbon). The dotted lines show the ␣ carbon center-to-center distances between adjacent (12 Å) and nonadjacent (19 Å) subunits. To form a disulfide bond or a high affinity Zn 2ϩ -binding site the ␣ carbons must approach to within ϳ5-6 Å.
probably in the closed state, and acetylcholine binding protein (AChBP), which is probably in the activated state, have suggested that agonist binding causes a larger shift in the ACh ␣ subunit structure.
In the region of the gate in the 4-Å structure, the M2 domains from the different subunits appear to make close contact with one another. Specifically, hydrophobic side chains from the 9Ј to the 14Ј positions interact to form "a tight hydrophobic girdle around the pore" (13). This girdle, which should impede the mobility of the individual M2 domains, may explain why, at the more proximal 17Ј position, the ␤M2 ␣-helix is less mobile than at the more distal 20Ј position. The constriction at the central portion of the channel may also help to explain why, in a previous study, no disulfide bonds formed between aligned residues from the 17Ј to the 6Ј positions.
The mobility that we have demonstrated in the 20Ј region with disulfide linkage is also consistent with the results of studies with unnatural amino acids. Using ␣-hydroxy acids in place of amino acids they converted the backbone peptide amide to an ester linkage. They inferred that there was more backbone conformational changes in the extracellular half of the ACh receptor M2 segment (20). Using linear free energy relationship analysis, it has also been shown that the extracellular half of M2 appears to move as a unit in the ACh-induced gating process (2,60).
The range of motion that we have inferred for the ␤ subunit M2 ␣-helices is not without precedent. In a study of protein backbone flexibility in the Escherichia coli D-galactose chemosensory receptor, a protein of known crystal structure, Falke and colleagues found two cysteines that could traverse 15 Å to form a disulfide bond (28,61,62). The time constant for the formation of this disulfide bond in the chemosensory receptor mutant was 3630 s in the presence of 1.5:4.5 mM Cu:phen, more than ten times slower that the formation time constant in ␣␤20ЈC␥, where disulfide formation went to completion within 360 s in the presence of 100:400 M Cu:phen (data not shown). The difference in the formation rates are probably even greater because in the chemosensory receptor experiments the Cu: phen concentration was over 10 times higher and the temperature was 12-15 degrees higher (37°C and 25°C for the chemosensory and GABA A receptors, respectively). Extrapolating from the results of Careaga and Falke (28), the collision rate between the engineered ␤20Ј Cys must be at least 10 3 s Ϫ1 and is probably higher, because their experiments were performed under significantly more oxidizing conditions than ours. In the bacterial mechanosensitive channel MscL state-dependent disulfide bond formation occurs between engineered Cys residues that are separated by Ͼ10 Å in the crystal structure. Disulfide bond formation between the MscL V15C required 1 mM Cu:phen and took about 30 min to go to completion (63). This suggests a lower collision frequency in MscL V15C than we observed for the GABA A receptor ␤20Ј Cys in the present study.
Alternative Interpretations-Although we believe that the above interpretation of our data is most likely, it rests on the structural foundation provided by the 4-Å resolution ACh receptor structure. An alternative interpretation that we believe is unlikely but that we cannot exclude is that our data suggest that the closed-state structure of the GABA A receptor is different than the published ACh receptor structure (13). The ability to form disulfide bonds and a high affinity Zn 2ϩ -binding site only between the ␤20ЈCys may indicate there is a significant structural asymmetry at the 20Ј level such that the average separation of the ␤20Ј residues is smaller than the ␣20Ј residues. CONCLUSION We have shown that, at the 20Ј level in the GABA A receptor M2 segments, a disulfide bond or a high affinity Zn 2ϩ -binding site can form between engineered Cys residues in the ␤ subunits. In contrast, an engineered Cys at the aligned position in the ␣ subunits forms neither. Based on the roughly symmetrical positions of the aligned residues relative to the channel axis in the 4-Å ACh receptor structure, we infer that the extracellular ends of the ␤ M2 segments are more mobile than the ␣ M2 segments. The increased mobility is likely due to looser protein packing around the ␤ M2 segments. In the ACh receptor structure the ␣ carbon atoms of the non-adjacent 20Ј residues are separated by ϳ19 Å. Thus, together the two ␤20ЈCys must move ϳ14 Å to form a disulfide bond or each must move about 7 Å toward the central axis. A similar amount of translational movement would be necessary to bring the two Cys into close proximity to form a Zn 2ϩ -binding site. Given the high affinity with which Zn 2ϩ was bound by the two ␤20ЈCys, we infer that there must be a low energy barrier to this movement. This suggests that there is a relatively flat potential energy surface for the movement of the ␤ M2 segments. These experiments begin to provide information on the dynamic movement of the channel-lining M2 segments and complement the static picture of the channel structure that is obtained from the cryo-EM structure of the homologous ACh receptor.