Structure of the M2 Transmembrane Segment of GLIC, a Prokaryotic Cys Loop Receptor Homologue from Gloeobacter violaceus, Probed by Substituted Cysteine Accessibility*

GLIC is a homopentameric proton-gated, prokaryotic homologue of the Cys loop receptor family of neurotransmitter-gated ion channels. Recently, crystal structures of GLIC hypothesized to represent an open channel state were published. To explore the channel structure in functional GLIC channels, we tested the ability of p-chloromercuribenzenesulfonate to react with 30 individual cysteine substitution mutants in and flanking the M2 channel-lining segment in the closed state (pH 7.5) and in a submaximally activated state (pH 5.0). Nine mutants did not tolerate cysteine substitution and were not functional. From positions 10′ to 27′, p-chloromercuribenzenesulfonate significantly modified the currents at pH 7.5 and 5.0 in all mutants except H234C (11′), I235C (12′), V241C (18′), T243C (20′), L245C (22′), and Y250C (27′), which were not functional, except for 12′. Currents for P246C (23′) and K247C (24′) were only significantly altered at pH 5.0. The reaction rates were all >1000 m−1 s−1. The reactive residues were more accessible in the activated than in the resting state. We infer that M2 is tightly associated with the adjacent transmembrane helices at the intracellular end but is more loosely packed from 10′ to the extracellular end than the x-ray structures suggest. We infer that the charge selectivity filter is in the cytoplasmic half of the channel. We also show that below pH 5.0, GLIC desensitizes on a time scale of minutes and infer that the crystal structures may represent a desensitized state.

vated than in the resting state. We infer that M2 is tightly associated with the adjacent transmembrane helices at the intracellular end but is more loosely packed from 10 to the extracellular end than the x-ray structures suggest. We infer that the charge selectivity filter is in the cytoplasmic half of the channel. We also show that below pH 5.0, GLIC desensitizes on a time scale of minutes and infer that the crystal structures may represent a desensitized state.
Fast chemical synaptic signaling in the nervous system is dependent on ligand-gated ion channels. The Cys loop receptor family of neurotransmitter-gated channels encompasses anionic channels that mediate inhibitory signaling (e.g. GABA A and glycine receptors) and cationic channels that mediate excitatory signaling (e.g. nicotinic acetylcholine (nACh) 2 and serotonin type 3 receptors) (1)(2)(3)(4)(5). These channels are pentameric assemblies in which each subunit has an extracellular domain, four transmembrane ␣-helices (M1-M4), and a large cytoplasmic domain between the M3 and M4 helices. The channel pore is formed along the radial axis and is lined by the five M2 segments, one from each subunit (6 -10). Ligand binding in the extracellular domain induces conformational changes in the transmembrane domain that open the gate and allows ions to pass through the channel. The gate is located in the middle of the membrane (11)(12)(13). Using cryo-electron microscopy, Unwin (12) solved the Torpedo marmorata nAChR structure in the absence of agonist at 4 Å resolution. The structure was consistent with the subunit organization and topology proposed by earlier studies (1,2,5,14). To date, a high resolution crystal structure of a full-length eukaryotic Cys loop receptor is not available.
Recently, prokaryotic Cys loop receptor homologues were identified based on predicted structural similarities in the extracellular ligand-binding domain (15). The Gloeobacter violaceus homologue (GLIC, the G. violaceus ligand-gated ion channel) was functionally expressed in eukaryotic cells and shown to be a proton-gated cation channel (16). GLIC lacks some of the conserved features found in the eukaryotic Cys loop receptors, including the large cytoplasmic M3-M4 loop domain and the pair of Cys that form the eponymous extracellular disulfide-link in the Cys loop. GLIC was crystallized at pH 4.6 and 4, and the structures were solved to 2.9 and 3.1 Å resolution, respectively (17,18). The GLIC structure shows substantial similarities to the nAChR structure, indicating that in spite of the low degree of amino acid sequence identity, GLIC is a Cys loop receptor homologue. Thus, GLIC may serve as a model in which to study the mechanisms of channel function and the conformational changes that the channels undergo as they transition between the closed, open, and desensitized states. However, there are some significant differences between the crystal structures of GLIC and the nAChR. 3 The M2 segments are 1-2 helical turns longer in the nAChR than in GLIC. This alters the position of aligned residues in the M2-M3 loop and thus interactions between the extracellular and transmembrane domains. The interface between the extracellular and transmembrane domains is critical for signal transduction from the binding site to the gate. Furthermore, although crystal structures provide a tremendous amount of information, it is unclear whether the GLIC structures correspond to an open or desensitized state of the channel. It was reported that GLIC does not desensitize (16,17). This supported the contention that the crystal structures prepared in low pH solutions repre-sent an open state. However, a recent report suggests that GLIC does desensitize (19). Furthermore, to crystallize GLIC, the protein was extracted from its native lipid membrane environment and crystallized from detergent solution at a pH that should activate the channel. The effects of detergent solubilization on the structure and desensitization rates are unknown for the GLIC channels. Detergent solubilization can have a significant impact on membrane protein structure (20). Thus, functional studies of GLIC channels in a lipid membrane environment can provide important information about the structure of the functional channel.
In this study, we used the substituted cysteine accessibility method (SCAM) (21) to evaluate the chemical reactivity of Cys substituted, one at a time, for residues in the GLIC M2 segment and M2-M3 loop from Glu-221 (Ϫ2Ј) to Tyr-250 (27Ј). We screened each mutant for reactivity with the sulfhydryl-specific mercurial reagent p-chloromercuribenzenesulfonate (pCMBS). We found that pCMBS reacted with almost every functional Cys mutant on the extracellular half of M2, from 10Ј to 27Ј. This suggested that the extracellular half of the M2 helix is loosely packed with the adjacent transmembrane helices. This, combined with the desensitization of GLIC that we observed with prolonged exposure to pH Ͻ 5.0, suggests that the crystal structures may represent a desensitized or nonphysiological state of the channel.

EXPERIMENTAL PROCEDURES
Cysteine Mutagenesis and mRNA Production-The GLIC coding sequence was cloned by PCR from a Gloeobacter violaceus (ATCC 29082) culture and inserted into the pXOON dualfunction vector for protein expression in Xenopus laevis oocytes or mammalian cells (22). The unique endogenous Cys at position 26 was mutated to leucine (C26L) by QuikChange (Agilent, Santa Clara, CA) site-directed mutagenesis to produce a cysteine-free GLIC construct. An index numbering system is used to facilitate comparisons of M2 segment residues among family members (23). In eukaryotic family members, an absolutely conserved cationic residue near the N-terminal end of the M2 helix is numbered 0Ј. Residues toward the N terminus in the M1-M2 loop are numbered Ϫ1Ј, Ϫ2Ј, etc., whereas residues more C-terminal than 0Ј are sequentially numbered 1Ј, 2Ј, etc. In GLIC, a cationic amino acid is not conserved at 0Ј. Therefore, we aligned the residues in GLIC M2 to an absolutely conserved proline at 23Ј. This alignment predicts that GLIC Asn-223 is the 0Ј residue. Thirty individual Cys mutants from the Ϫ2Ј through 27Ј positions (E221C to Y250C), in and flanking the M2 region, were generated by QuikChange mutagenesis in the C26L GLIC background. Plasmid DNA linearized with NheI was used as a template for in vitro mRNA transcription using the mMachine mMessage T7 kit (Applied Biosystems/Ambion, Austin, TX).
Electrophysiology-Oocytes were prepared as described previously and injected with 10 ng of mRNA (24). 1-3 days after mRNA injection, the currents were recorded using two-electrode voltage clamp at a holding potential of Ϫ40 or Ϫ60 mV via a TEV-200 amplifier (Dagan Corp., Minneapolis, MN) and digitized with a Digidata 1440A A-D converter and pClamp 10.2 software (Molecular Devices, Sunnyvale, CA). Agarose-tipped glass microelectrodes were filled with 3 M KCl (ϳ0.3-0.8 M⍀ tip resistance). The ground electrode was connected to the bath via a 3 M KCl-agar bridge. During recording, the oocytes were continuously perfused at 5 ml/min with GLV buffer (100 mM NaCl, 20 mM NaOH, 2.5 mM KCl, 1 mM MgCl 2 , 2 mM CaCl 2 , 5 mM HEPES, 5 mM citric acid, pH adjusted with HCl) at room temperature. All of the reagents were dissolved in GLV buffer.
Complete pH response curves could not be obtained because the oocytes became unstable at pH Ͻ 4. The oocytes tolerated repeated exposures to pH 5.0 or, to a lesser extent, pH 4.5. Proton-induced currents were measured for each mutant at pH 5.0 and 4.5 to determine whether the mutant was functional and to verify that pH 5.0 induced a submaximal current response. For mutants that had current amplitudes similar to those elicited from water-injected oocytes, we tested for the presence of spontaneously formed disulfide bonds between engineered Cys, by applying 10 mM DTT ϩ 2 mM tris(2-carboxyethyl)phosphine (TCEP) for 2 min at pH 7.5 followed by 2 min at pH 5.0.
pCMBS Accessibility-We evaluated the accessibility of each Cys substitution mutant in the resting (pH 7.5) and submaximally activated states (pH 5.0 for all Cys mutants except 12Ј, where we used pH 6.0 because of the increased pH sensitivity of this mutant). Given that the Cys thiol has a pK a of 8.14 in free solution (25), at pH 5.0 Cys thiols are unlikely to be significantly ionized. Therefore, we did not use methanethiosulfonate reagents because they react at a negligible rate with unionized thiols (26,27). Because of the low pH activation of GLIC channels, we used p-chloromercuribenzenesulfonate (pCMBS) as our sulfhydryl reactive reagent because the organomercurial reagents react with both ionized thiolates (-S Ϫ ) and with unionized thiols (-SH), albeit at different rates. pCMBS is an anion in the pH range 5.0 to 7.5. It is 313 times more soluble in water than in n-octanol (28). pCMBS is essentially impermeable through the lipid bilayer (29). Thus, its concentration in lipophilic regions of the protein should be negligible during application. Therefore, pCMBS should not react with buried Cys or should do so at a very slow rate. It should only modify sulfhydryl groups when they are in an aqueous environment. A potential problem with using pCMBS is that it is an anion and thus may have limited access to residues in the region of the charge selectivity filter in this cation-selective channel.
pCMBS reaction with Cys mutants was assayed functionally. We measured the peak current amplitude evoked by low pH test pulses applied before and after pCMBS application. Low pH test pulses were applied for 2 min, with 2-min pH 7.5 washes between pulses to allow recovery from desensitization. Occasionally we saw run-up of the current amplitude induced by sequential low pH test pulses. When the current amplitudes of two consecutive low pH test pulses were within 10%, the response was considered stable, and the effect of pCMBS was tested. Following a 2-min prewash in buffer at the same pH as the pCMBS solution (pH 7.5, 6.0, or 5.0), 200 M pCMBS was applied for 2 min. When pCMBS was applied at low pH, it was washed out for 20 s with buffer at the same pH before returning to the pH 7.5 buffer. Additional 2-min low pH test pulses were recorded at 2-min intervals until the response was stable, as was done before pCMBS treatment. The effect of modification was calculated as the ratio of the average peak current of the two pulses following pCMBS application to the average of the two pulses immediately before application.
pCMBS Reaction Rates-We measured the reaction rate of pCMBS with the reactive Cys mutants. Reaction rates at pH 7.5 were determined by alternating applications of pCMBS at pH 7.5 with 2-min low pH test pulses. Between the pCMBS and low pH test pulses, the oocytes were washed for 2 min with pH 7.5 buffer. The pCMBS exposure from each application could be controlled by varying the pCMBS concentration (0. 2-200 M) or the time of application (10 -120 s). As expected, similar rates were obtained whether duration or concentration of pCMBS was varied. Five or more sequential pCMBS applications and test pulses were recorded. The test currents following pCMBS application were normalized to the initial pH 5.0 current and plotted against cumulative pCMBS exposure (in units of M⅐s).
Because GLIC is a homopentamer, there are five potential reactive sites in each mutant channel. The observed functional effect may require modification of only one Cys or may require reaction at several Cys before a functional effect occurs. The reaction rate data were fit to a single exponential function, representing a single bimolecular reaction, or to a kinetic model representing two consecutive reactions.
Because pH 5.0 produced a stable current that did not appear to desensitize, reactions at pH 5.0 were observed in real time. pH 5.0 buffer alone was applied for 2 min, until a steady-state current was reached, and then pCMBS was added to the buffer at a concentration that would allow the reaction to proceed to completion during a 5-10-min period. Changes in the recorded current during pCMBS application at pH 5.0 were attributed to sulfhydryl modification, and the recorded curve was fit to single or sequential reaction kinetics. Optical Absorption and Stopped Flow Kinetic Measurements-To identify spectral differences between pCMBS and the mercury-mercaptide product, UV-visible absorption spectra of pCMBS were measured before and after reaction with L-cysteine on an Agilent 8453 UV-visible spectrophotometer (Agilent Technologies, Santa Clara, CA) over the pH range 3.0 to 12.0. The GLV buffer used for electrophysiological recordings had a strong optical absorption at wavelengths shorter than ϳ240 nm. Therefore, reagents were prepared in phosphate acetate-buffered saline (150 mM NaCl, 5 mM NaCH 3 COO-3H 2 O, 2.5 mM Na 2 HPO 4 , 2.5 mM NaH 2 PO 4 -H 2 O, pH adjusted with HCl or NaOH) to minimize the interference of buffer.
We determined that the optical absorbance at 236 nm changed during reaction of pCMBS with Cys in phosphate acetate-buffered saline over the pH range from 3.0 to 12.0 (data not shown). The reaction between pCMBS and cysteine was followed at 236 nm on a *180 stopped flow spectrometer equipped with a mercury-xenon lamp and a photomultiplier tube detector (Applied Photophysics, Surrey, UK). The reactions were carried out with a 10ϫ excess of cysteine to give pseudo first order kinetics. The recorded curves were fit to a single exponential function, and second order rate constants were calculated. The reaction rate was measured over the pH range from 3.0 to 6.5.
Data Analysis and Statistics-All of the experiments were performed in at least three oocytes from at least two separate frogs. The data are presented as the means Ϯ S.E. Statistical significance was determined by one-way analysis of variance with Tukey's post hoc test using GraphPad Prism version 5.02 for Windows (GraphPad Software, San Diego, CA). Curve fitting to determine pCMBS reaction rates was performed with SigmaPlot 11 software (Systat Software Inc., San Jose, CA) or pClamp 10.2 software (Molecular Devices). Curve fitting for stopped flow kinetics experiments was performed with Prism version 5.02 (GraphPad Software).

RESULTS
GLIC Expression and Electrophysiology-Increasing the buffer proton concentration by decreasing the buffer pH to values under 6.5 elicited currents in Xenopus oocytes expressing wild type GLIC. For buffer pH Ն 5.0, the pH-activated current response was similar to previously published results (16,18). However, in contrast to previously published statements that GLIC does not desensitize, the wild type GLIC currents desensitized with prolonged (minutes) exposure to pH Ͻ 5.0 (Fig. 1A). The rate of desensitization was faster at pH 4.0 than at 4.5 (Fig. 1A). The desensitization induced at pH 4.0 at a holding potential of Ϫ50 mV was fitted with an exponential function with a mean time constant of 27.9 Ϯ 2.9 s (n ϭ 5). A recent report also noted that when expressed in HEK293 cells, macroscopic GLIC currents showed desensitization (19).
We tested the effect of a 2-min application of 200 M pCMBS on wild type GLIC. pCMBS application significantly inhibited the subsequent pH 5.0 test currents (Fig. 1B). We inferred that pCMBS reacted with the sole endogenous cysteine in wild type GLIC, Cys-26. We mutated Cys-26 to leucine (C26L) to produce a Cys-free construct that was insensitive to pCMBS (Fig. 1D). Cys-26 lies in extracellular domain in the middle of the ␤1 strand. It faces into the ␤ sandwich. Cys-26 is in close proximity to the ␤8-␤9 loop and residues in the ␤2 and ␤9 strands (Fig. 2, A and B). There are several possible mechanisms by which pCMBS modification of Cys-26 might inhibit GLIC function. The ␤1-␤2 loop plays an important role in signal transduction between the extracellular and transmembrane domains (30). Modification of Cys-26 might disrupt this transduction pathway, or it might disrupt the ␤ sandwich. At present we cannot distinguish between these possibilities.
We could not measure a complete pH response curve to determine the proton EC 50 (the pH that gives rise to 50% of the maximal current) because the oocytes did not tolerate prolonged (minutes) exposure to pH Ͻ 4. The Cys-free GLIC mutant had a pH response that was shifted to lower pH compared with wild type. Like the wild type, Cys-free GLIC did not appear to desensitize during a 2-min pH 5.0 pulse but did desensitize with exposure to pH Ͻ 5.0 (Fig. 1C). All of the sub-sequent Cys substitutions were made in the Cys-free background construct.
Effects of Cysteine Mutagenesis-For each of the 30 individual Cys mutants, we measured the currents induced by low pH test pulses to determine the effect of the mutation on proton-induced channel activation (Fig. 3). The currents induced by pH 5.0 application were not significantly different from those recorded from water-injected oocytes for nine of the 30 cysteine mutants: Ϫ1Ј (A222C), 6Ј (S229C), 11Ј (H234C), 13Ј (A236C), 15Ј (N238C), 18Ј (V241C), 20Ј (T243C), 22Ј (L245C), and 27Ј (Y250C) (Figs. 2C and 3). Because each mutant channel possesses five engineered Cys, we sought to determine whether the lack of function was due to spontaneous disulfide bond formation between engineered Cys. We applied a solution containing the reducing agents DTT (10 mM) and TCEP (2 mM) at both pH 7.5 and 5.0, but this did not rescue channel function in any of the nonconducting mutants (data not shown). Either disulfide bonds were not responsible for the lack of function of these mutants or steric factors limited reducing agent access to the disulfide bonds so they could not be reduced or they reformed very rapidly after reduction.
The 8Ј (L231C), 9Ј (I232C), and 10Ј (A233C) mutants exhibited significant changes in the pH-dependent gating process. The 8Ј (L231C) mutant was constitutively conductive at pH 7.5. Most oocytes injected with L231C mRNA were dead within 2 days. These oocytes displayed irregular patterns of pigmentation on their surface (31) or released their cytoplasmic contents into the storage medium. Electrophysiological recordings from 8Ј-expressing oocytes that appeared healthy on visual inspection Ͼ36 h after mRNA injection showed holding currents in excess of 6 A when clamped at Ϫ60 mV. We recorded from oocytes injected with the 8Ј L231C mutant 12-24 h after injection, when we expected lower expression levels. In these oocytes, holding currents ranged from 1 to 5 A. The holding current was not significantly reduced by application of pH 9.0 buffer or 100 M QX-222 at pH 7.5. Surprisingly, the holding currents were significantly reduced by application of pH 5.0 or 4.5 buffer (data not shown).
Following pH 5.0 exposure, the currents in oocytes expressing the 9Ј (I232C) mutant closed very slowly and did not return to the initial base-line current even after several minutes of pH 7.5 wash. The pH response of 12Ј (I235C) was shifted to higher pH such that the pH 5.0 current amplitudes were of similar magnitude to those at pH 4.5. For this mutant, pH 6.0 was used to elicit a submaximally open state in subsequent experiments.
Effects of pCMBS Modification-We evaluated the ability of pCMBS to react with each Cys mutant in the closed (pH 7.5) and submaximally open (pH 5.0 or 6.0) states (Fig. 4). Covalent modification by pCMBS could have multiple effects on an individual mutant; the subsequent current amplitudes could be altered, and/or the channel gating kinetics or pH dependence could change. We infer that pCMBS reacted with Cys mutants whose subsequent currents or gating kinetics were significantly altered following pCMBS application.
For the mutants ranging from 13Ј to 27Ј, a 2-min application of 200 M pCMBS at pH 7.5 significantly altered the subsequent pH 5.0 test pulse current amplitudes for 13Ј (A236C), 14Ј (F237C), 15Ј (N238C), 16Ј (I239C), 17Ј (L240C), 19Ј (E242C), 21Ј (N244C), 25Ј (T248C), and 26Ј (P249C) (Fig. 4A). For the 13Ј (A236C) and 15Ј (N238C) mutants, prior to pCMBS application, the low pH test pulses did not elicit currents that were significantly different from water-injected oocytes. However, pCMBS modification significantly increased the subsequent currents elicited by low pH test pulses for both mutants (Fig.  2C, purple residues). Thus, we infer that the Cys substitutions at 13Ј (A236C) and 15Ј (N238C) inhibited channel function, but pCMBS modification of the engineered Cys restored channel function. This implies that for these two mutants, the lack of function prior to pCMBS modification was not due to disulfide bond formation between the engineered Cys. For the four other nonfunctional mutants between 13Ј and 27Ј, 18Ј (V241C), 20Ј (T243C), 22Ј (L245C), and 27Ј (Y250C), Residues for which the Cys mutants were nonfunctional are shown on the middle subunit in space-filling format. Cyan, residues for which no function was elicited. Purple, 13Ј and 15Ј residues that were functional following pCMBS modification. To visualize the channel-lining residues, the front two subunits have been removed. The remaining three subunits are color-coded to distinguish them in this and in all subsequent panels in this figure. The middle subunit is colored light purple, with the M2 and M2-M3 loop regions that were mutated to Cys in this study in blue. The two flanking subunits are colored light green, except for the region studied, which is dark green. Note that the Cys mutants were present in all subunits but are only shown in one to simplify the figure. D, close-up view of the extracellular end of M2 to illustrate the position of the pCMBS reactive residues in space-filling format. Orange, pCMBS modification potentiated subsequent currents. Yellow, pCMBS modification inhibited subsequent currents. Red, 10Ј residue where pCMBS modified gating kinetics. Cyan, nonfunctional mutants. Blue, Cys mutants not affected by pCMBS. E, view of the M2 segments looking into the channel. All of the residues are shown in space-filling format. The extracellular domain of the subunit on the right is shown in ribbon format to clarify the extent of that domain in the other subunits. Note that based on the crystal structure, many of the pCMBS-reactive residues are buried in the protein interior. F, view of the lipid bilayer facing side of the same three subunits seen in E. This view results from a 180°rotation around the vertical axis of E. The color scheme is the same as in D. Note that only one of the pCMBS reactive residues, Pro-249 (26Ј), has significant accessibility from the external surface of the protein. G, same view as in F, except that the M1, M3, and M4 segments have been removed from the light purple subunit to reveal the "buried" pCMBS-reactive residues on the non-channel-facing side of the M2 segment.
there were no changes following pCMBS application at either pH 7.5 or 5.0.
All of the mutants that reacted with pCMBS at pH 7.5 also reacted with 200 M pCMBS applied at pH 5.0 (Fig. 4B). In addition, although pCMBS application at pH 7.5 did not have a significant effect on the 23Ј (P246C) and 24Ј (K247C) mutants, application of 200 M pCMBS at pH 5.0 significantly altered the subsequent pH 5.0 test currents (Fig. 4B). Covalent modification of a Cys may potentiate or inhibit subsequent currents. At the 11 mutants between 13Ј and 27Ј where pCMBS modification significantly altered the currents, the currents were potentiated at 13Ј (A236C), 14Ј (F237C), 15Ј (N238C), 17Ј (L240C), and 25Ј (T248C) and were inhibited at 16Ј (I239C), 19Ј (E242C), 21Ј (N244C), 23Ј (P246C), 24Ј (K247C), and 26Ј (P249C) (Figs. 2, D-G, and 4). It is difficult to infer the structural or mechanistic basis for the functional effects of modification, and we have not pursued such studies (Fig. 2D). Note that the pCMBS-reactive residues on the back side of M2 are not accessible from the lipid bilayer in the crystal structure (Fig. 2, compare F and G; M1, M3, and M4 removed in Fig. 2G to visualize the location of the M2 residues buried in the subunit interior).
From Ϫ2Ј to 12Ј, the only pCMBS-reactive residue was A233C (10Ј) where pCMBS application caused a change in gat-ing kinetics (Fig. 5). Application of pCMBS at pH 7.5 slowed down subsequent channel deactivation after the pH was changed from 5.0 to 7.5 and increased the amplitude of the holding current at pH 7.5. Proton-induced currents were increased by 23 Ϯ 7% following pCMBS application (n ϭ 4), but this change was not statistically significant (Fig. 5A). In contrast, following application of pCMBS at pH 5.0, the current remained stable with little reduction in the holding current after the bath pH was returned to pH 7.5 over a 10-min period (Fig. 5B). The Cys mutants at Ϫ1Ј (A222C), 6Ј (S229C), and 11Ј (H234C) were not functional (Figs. 2C and 3), and application of the reducing agents, DTT/TCEP, or pCMBS had no effects on these mutants (Fig. 4 and data not shown).
Thus, of the 30 Cys mutants tested in and flanking the M2 segment, 7 Cys mutants resulted in nonfunctional channels that were not altered by either reduction or pCMBS modification. pCMBS modified 12 of the 30 Cys mutants. All of the pCMBS reactive residues were more extracellular than 10Ј. For all of the residues more cytoplasmic than 10Ј, pCMBS application had no functional effect.
Second Order Rate Constants for Cysteine Modification-Measuring the pCMBS reaction rate with the engineered Cys mutants provides information about the relative reactivity of different residues and of the reactivity of the same residue in different functional states, i.e. pH 7.5 (resting) versus 5.0 (submaximally activated). For GLIC channels, one complication in measuring the reaction rate is that each channel contains five engineered Cys residues. We do not know whether the functional effects occur with modification of the first Cys or require modification of two or more of the subunits. For several positions, the initiation of the functional response appeared to be delayed. This suggested that one or more functionally "silent" reactions occurred before the onset of the functional effect of pCMBS modification. It is important to note that we could only detect a "silent" modification when the second order rate constants for each modification, silent or functional, were similar. If the silent modification occurred much faster than the second "functional" reaction, we would not have observed a delay in the functional response. If the functional modification occurred much faster than the silent reaction, we would have observed a functional change limited by the rate of the silent reaction. Thus, we fit the current response data as a function of the cumulative concentration ϫ time of pCMBS application, with both a single reaction model and a sequential reaction model.
We measured the second order reaction rate constants for pCMBS with the reactive Cys mutants at pH 7.5 and 5.0 (Figs. 6 and 7). At pH 7.5, the reactions with 16Ј, 19Ј, 21Ј, and 26Ј were consistent with the functional effects arising from modification of a single subunit. In contrast, the reactions with the 13Ј, 14Ј, 15Ј, 17Ј, and 25Ј mutants indicated that modification at more than one subunit was required to produce the effect seen with the 2-min application of 200 M pCMBS. We could not fit the reaction with 25Ј to either kinetic model. At pH 7.5, reaction rates with all positions except 10Ј were Ͼ1000 M Ϫ1 s Ϫ1 (Fig. 7). For 10Ј, we could not quantify the changes in closing kinetics following modifica- White bars indicate currents that were not significantly different from waterinjected oocytes. *, L231C (8Ј) was constitutively conductive at pH 7.5. Note that the x axis has a log scale.
tion to determine the reaction rate. A second 2-min application of 200 M pCMBS further slowed channel closing. We infer that the 10Ј reaction did not go to completion with a 24,000 M⅐s pCMBS exposure and that the second order rate constant at 10Ј at pH 7.5 was Ͻ100 M Ϫ1 s Ϫ1 .
At pH 5.0, the reaction rate data at 14Ј, 19Ј, 23Ј, and 24Ј were best fit by a single exponential function consistent with the functional effects arising from modification of a single subunit (Figs. 6 and 7). For the 10Ј, 13Ј, 15Ј, 16Ј, 17Ј, and 26Ј mutants, the reaction rate data were best fit with a sequential reaction model. The reaction at 21Ј was best fit by a double exponential function. At 25Ј, the reaction at 5.0 was not well fit by either single or sequential reaction kinetic models. The reaction rates were all Ͼ1000 M Ϫ1 s Ϫ1 . It should be noted that pH 5.0 induces submaximal activation of GLIC channels. Thus, the reaction rates at pH 5.0 represent a combination of reaction rates with channels in the open, closed, and, presumably, desensitized states. The proportion of channels in each state is unknown. The rates with the fully open state may be significantly different.
pH Dependence of Second Order Rate Constants for pCMBS Modification of Cysteine-Comparison of pCMBS reaction rates with a given Cys mutant at different pH levels is complicated by the fact that 1) the ratio of thiols (-SH) to thiolates (-S Ϫ ) varies with the solution pH, and 2) pCMBS reacts at different rates with thiols and thiolates. To normalize for these effects, we measured the reactivity of pCMBS with Cys in free solution as a function of pH. We measured the reaction of pCMBS with Cys using a spectrophotometer equipped with a stopped flow, rapid mixing apparatus. We fit the recorded Black bars indicate mutants in which the average ratio of the proton-induced currents following pCMBS exposure to those before exposure was significantly different from the average ratio in C26L GLIC by one-way analysis of variance with Tukey's post test. Application of pCMBS to oocytes expressing A233C (10Ј) did not significantly alter the magnitude of the proton-induced current, but the response was altered as shown in Fig. 5. The effect of pCMBS application at pH 7.5 or 5.0 was not determined in oocytes expressing L231C (8Ј) or I232C (9Ј), and the effect of pCMBS application at pH 5.0 was not determined in oocytes expressing wild type GLIC. Note that the x axis has a log scale. absorption traces (data not shown) to single exponential functions to calculate the second order rate constants for the reaction between pCMBS and free cysteine in solutions buffered at different pH values (Fig. 8). At pH Ͼ 6.5, the reaction proceeded too rapidly to be measured accurately using the available apparatus.
To account for the difference in the pCMBS reaction rate with thiols and thiolates, we used Scheme 1, where X is pCMBS,  . Second order rate constants for pCMBS modification of reactive residues. We could not fit the data for modification at 25Ј (T248C) to either kinetic model. Note that the x axis has a log scale. FIGURE 8. The pH dependence of the second order rate constant for reaction between pCMBS and free cysteine. The experimental data points (dots) are the measured second order rate constants. Each data point is the average of at least eight reactions. The error bars are smaller than the data points. The reaction proceeded too rapidly to measure at pH Ͼ 6.5. The curve is a 1/Y 2 -weighted nonlinear regression of the data with Equation 2 using GraphPad Prism version 5.02 for Windows. Because we expect the error in our data to be relative to the magnitude, this weighting is applicable. Note that the y axis has a log scale. k 1 and k 2 are the second order rate constants, and K a is the acid dissociation constant of the sulfhydryl group. In this reaction scheme, the observed second order rate constant, k cys , is given by Equation 2.
Using pK a ϭ 8.14 for Cys (25) and fitting the data to Equation 2, we determined that for the reaction between Cys and pCMBS, k 1 ϭ 3.5 ϫ 10 5 M Ϫ1 s Ϫ1 and k 2 ϭ 1.2 ϫ 10 9 M Ϫ1 s Ϫ1 . This gives values for k obs of 1.2 ϫ 10 6 M Ϫ1 s Ϫ1 at pH 5.0 and 2.2 ϫ 10 8 M Ϫ1 s Ϫ1 at pH 7.5. Thus, the intrinsic reactivity of pCMBS with Cys was 183 times greater at pH 7.5 than at pH 5.0. This is consistent with the differences in second order rate constants measured for other mercurial reagents with thiol and thiolate forms of other sulfhydryl compounds (32).
To account for this 183-fold increase in the intrinsic reactivity of pCMBS at pH 7.5 compared with pH 5.0, we normalized the measured reaction rates for pCMBS with the engineered Cys residues by their intrinsic reaction rates (i.e. for the pH 5.0 rate constants, normalized reactivity ϭ k measured /k cys, pH 5.0 ). With this correction, it is apparent that at every reactive Cys residue, pCMBS reacted faster at pH 5.0 than at pH 7.5 (Fig. 9).

DISCUSSION
Although the crystal structures of GLIC provide valuable new information, little is known about how the crystal structures compare with the structure of a functional channel in a lipid bilayer membrane. Multiple experimental approaches are necessary to gain a clearer view of structure/function relationship in functional GLIC channels. Previous reports stated that wild type GLIC did not desensitize (16,18). However, at pH Ͻ 5.0, we show that the GLIC channels desensitized on a time scale of minutes and the rate of desensitization increased with decreasing pH (Fig. 1A). A recent report also noted that GLIC expressed in HEK293 cells desensitizes at pH Ͻ 5.0 albeit with a rate significantly faster than what we observed (19). This may relate to differences in the lipid composition of oocyte and HEK293 cell membranes. The crystals for both of the published GLIC structures were grown at pH Ͻ 5.0 (17,18). Under these conditions, our data indicate that the GLIC channels would be in a desensitized state on the time scale needed for crystallization. Although we cannot rule out the possibility that the rate of desensitization is different in detergent-solubilized GLIC, we conclude that the GLIC crystal structures may represent a desensitized rather than an open state of the channel.
Consistent with the conclusion that the GLIC structures represent a nonconducting state of the channel, the structures in the absence and presence of propofol are identical (33). Propo-fol inhibits GLIC channels. Thus, it is likely to bind to a nonconducting state with greater affinity than to an open state. The fact that its binding produced no change in the GLIC structure suggests that the structure may be a nonconducting state. Furthermore, the location of the propofol-binding site in GLIC in the interior of the subunit is inconsistent with photoaffinity labeling and mutagenesis studies in GABA A receptors that show the propofol-binding site to be in the intersubunit interface between M3 and M1 (34 -36). Propofol is rather promiscuous in its protein binding and has also been crystallized bound to other proteins including albumin and apoferritin (37,38). Thus, the propofol-bound GLIC crystal structure may not provide insight into propofol binding to pharmacologically relevant targets, such as the GABA A receptor. In addition, it was recently suggested that the ELIC structure, another prokaryotic Cys loop homologue, does not represent a resting closed state but rather a nonconducting, nonfunctional state (19). Thus, studies trying to elucidate the conformational changes occurring during Cys loop receptor gating by comparing the ELIC and GLIC structures as representative of the closed and open states may not be fruitful (17,18).
Although Cys accessibility studies of ion channels lack the high resolution of x-ray crystallography, they have the advantage that they probe the protein structure in functional channels. We determined the ability of pCMBS to react with Cys substituted for residues in and flanking the GLIC M2 segment to assess solvent accessibility of the amino acid side chains. Seven of the Cys substitutions (Ϫ1Ј, 6Ј, 11Ј, 18Ј, 20Ј, 22Ј, and 27Ј) resulted in nonfunctional channels (Figs. 2C and 3). Neither reduction with DTT/TCEP nor application of pCMBS restored the function of these channels. These positions are distributed throughout the region investigated; some face into the channel, and others face into the protein interior. We are unable to detect a pattern that might explain why these Cys substitutions were not tolerated.
For the functional Cys mutants, we inferred that pCMBS covalently modified an engineered Cys when either the current amplitude or kinetics of the low pH test current were irrevers-SCHEME 1 FIGURE 9. Normalized reactivity of cysteine mutants to pCMBS. The second order rate constants reported in Fig. 7 have been normalized to the second order rate constant for the reaction between pCMBS and free cysteine at pH 7.5 or pH 5.0. Note that the x axis has a log scale.
ibly altered following pCMBS application. pCMBS modified every functional Cys mutant from 10Ј to 26Ј except 12Ј (Figs. 2  and 4). For all of the positions except 10Ј, the second order reaction rate of pCMBS was greater than 1000 M Ϫ1 s Ϫ1 (Fig. 7).
There were several unexpected aspects of our results. M2 ends at ϳ22Ј in the GLIC crystal structure, so we expected an ␣-helical pattern in either the chemical accessibility data or in the reaction rates from 10Ј to 22Ј (Fig. 9). In fact, given the tightness of protein packing of this region in the crystal structures, we expected that pCMBS would only have access to channel-lining residues (Fig. 2, F and G). Previous SCAM studies of the M2 segment of eukaryotic Cys loop receptors have shown an ␣-helical pattern of sulfhydryl accessibility or reaction rates in the extracellular half of M2 (9, 11, 21, 39 -42). Surprisingly, in GLIC M2, we observed comparable reaction rates for pCMBS (Figs. 7 and 9) with Cys substituted for channel-lining and non-channel-lining residues in both the resting and submaximally activated state (Fig. 2, E-G). The failure to observe a difference in the reaction rates is unlikely to be due to a modified pK a value of buried Cys residues. The pK a of residues within a protein may vary significantly from the pK a of the corresponding amino acid in bulk solution (10). In the AChR M2 segment, Cymes et al. (10) showed that the pK a values of lysines substituted for channel-lining residues were close to the pK a of lysine in free solution. In contrast, the pK a values of lysines substituted for buried residues were altered significantly to favor the uncharged state. This suggests that the Cys pK a for buried residues might be significantly greater than the pK a of 8.14 in bulk solution (25). This would have reduced the measured reaction rate for buried residues by increasing the percentage of the time spent in the less reactive thiol form. Thus, we conclude that in functional GLIC channels embedded in a lipid bilayer membrane, the top half of the M2 segment is less tightly packed than suggested by the crystal structure. This region of GLIC M2 may be more loosely packed than in the eukaryotic Cys loop receptors that have been similarly studied in the past.
Based on their reactivity, we infer that ostensibly buried Cys more extracellular than 12Ј are at least transiently exposed to the water-accessible protein surface. Perhaps thermal protein motion results in the formation of aqueous crevices into the interior of the transmembrane domain. These crevices must be large enough to accommodate pCMBS, which would fit into a right cylinder 6 Å in diameter and 10 Å in length. We previously suggested that such crevices form in the GABA A receptor during channel gating (41,43). We conclude that when GLIC is embedded in a lipid membrane, the extracellular half of the GLIC M2 segment is significantly more mobile than suggested by the crystal structures. A corollary to this is that for M2 to be that mobile, the protein packing in the transmembrane regions must be significantly looser than that seen in the crystal structures.
Consistent with the idea of loose packing, for several of the GLIC Cys mutants, the reaction rate data are most consistent with a model where pCMBS modification of more than one subunit was necessary to cause the observed functional effects (Fig. 8). The fact that pCMBS could modify one or more M2 Cys with no functional effect suggests that the extracellular half of the M2 segment of GLIC has sufficient flexibility to accommodate the pCMBS-modified engineered Cys.
In contrast to what we observed with the upper part of M2, the cytoplasmic half of M2 was tightly packed against the adjacent membrane-spanning segments. pCMBS application did not have a significant effect on any of the Cys mutant from Ϫ2Ј to 9Ј (Fig. 4). For the channel-lining residues in this region, it is possible that pCMBS is too large to enter this region of the channel. Alternatively, this region may form the charge selectivity filter that might exclude the negatively charged pCMBS from this region. For the buried residues, their lack of reactivity suggests that the aqueous crevices do not extend this deep into the membrane. This is consistent with previous SCAM studies of this region in eukaryotic channels (41).
Differences in the pCMBS reaction rates with Cys in the resting and activated states of the channel provide evidence for structural changes associated with channel gating. A complication for this comparison in GLIC is that the channel is activated by low pH: the intrinsic reactivity of pCMBS with thiols depends on the ionization state of the thiol. We measured that pCMBS reacts 183 times faster with the ionized thiolate than with the unionized thiol. The relative accessibility of each mutant at pH 7.5 and 5.0 was inferred by normalizing the rate constants measured for the mutant to the rate constant for reaction with free Cys in solution at the same pH (Fig. 9). This normalization accounts for the intrinsic difference in reaction rate at pH 7.5 and pH 5.0. With this correction, we found that pCMBS reacted faster with all of the engineered Cys in the submaximally activated state than in the resting state (Fig. 9). The increase in reaction rates with activation suggests a dramatic increase in mutant accessibility when the channels are submaximally activated. It should be noted that at pH 5.0, the channels are submaximally activated; the channels will be in a variety of states including closed, open, and perhaps desensitized. Thus, the reaction rates with Cys in the open channel state may actually be faster than the rates we have measured at pH 5.0. We infer that the transmembrane helices are more loosely packed, more mobile, or more flexible during channel gating than at rest. This is in contrast to the tightly packed structures and gating model proposed based on the ELIC and GLIC crystal structures (17,18).
In the GLIC crystal structures, M2 is tightly packed against the other transmembrane segments (17,18). It is important to note that the GLIC crystal structures were produced from detergent-solubilized protein. The effect of detergent solubilization on membrane protein structure is unknown but can be significant (20). In the Torpedo nAChR cryo-electron microscopy structure, which was derived from protein in its native lipid membrane, the transmembrane segments are more loosely packed (12,17,18). A recently published NMR structure of the transmembrane domains of the nAChR suggests that the transmembrane segments are tightly packed (44). However, the NMR structure was produced in detergent solution from a truncated construct that was missing the N-terminal extracellular domain and the large cytoplasmic loop between M3 and M4. The effects of these modifications on protein structure are unknown. Furthermore, the construct was monomeric in detergent solution, raising questions about its relevance to the native protein structure. Indications of loose packing consistent with the cryo-electron microscopy structure of the Tor-pedo nAChR have previously been observed in SCAM studies of the nAChR and the GABA A receptor (40 -42). Our current data imply that when embedded in a lipid membrane, the GLIC channel has a structure that is also more consistent with the loose packing observed in the Torpedo nAChR structure.
GLIC is a cation-selective channel (16). Our data suggest that the primary determinants of charge selectivity in the GLIC channel are at the cytoplasmic end of the channel. The high rates of reaction that we observed with residues in the extracellular half of M2 and in the M2M3 loop suggest that pCMBS, which is an anion in the pH ranges utilized in these experiments, had free access to these regions in the GLIC channel. Thus, the extracellular vestibule does not contribute significantly to the charge selectivity filter in GLIC. We did not detect pCMBS modification at any residue below 8Ј at pH 7.5 or 5.0. Access to residues below this region may be blocked by the charge selectivity filter or by steric factors. Our results are consistent with chemical accessibility studies in the nAChR and the GABA A receptor, which inferred that charge selectivity occurred near the cytoplasmic end of the channel (9,21) and with the location of an binding site for extracellularly applied Zn 2ϩ at the M2 17Ј ␤His in the anion-selective GABA A channel (45). The location of the charge selectivity filter near the cytoplasmic end of the channel is also consistent with experiments showing that swapping three residues in and flanking the cytoplasmic end of M2 flips the charge selectivity of eukaryotic Cys loop receptor channels (46 -50). It is notable that the GLIC channel lacks the 0Ј Arg or Lys found in virtually all eukaryotic family members. Clearly cation selectivity can be obtained without this residue, which has been inferred to be deprotonated in nAChR channels (10). In contrast to the charge selectivity filter, the determinants of single channel conductance are more widely distributed along the length of the pathway that ions traverse through these proteins. Residues influencing conductance have been identified in the extracellular vestibule, the extracellular end of M2, the cytoplasmic end of M2, and the cytoplasmic MA helices (6,46,(51)(52)(53)(54)(55). These regions may serve minor roles in determining charge selectivity, but our data suggest that residues near the cytoplasmic end of the channel are the primary determinants of charge selectivity in GLIC.
In conclusion, the GLIC channel desensitizes at pH Ͻ 5.0. Cysteine accessibility studies indicate that in contrast to the GLIC x-ray crystal structures, the extracellular half of M2, and the M2-M3 loop are highly accessible to water-soluble, charged sulfhydryl reactive reagents. This implies that the extracellular half of the transmembrane domain is more loosely packed and highly mobile than suggested by the GLIC crystal structures. Access of negatively charged pCMBS to the channel-lining residues in the cytoplasmic half of M2 is limited either by charge or size selectivity filters or both. The cytoplasmic half of M2 is tightly packed, limiting access of pCMBS to residues on the back side of M2. These results suggest that the crystal structures are more likely to represent a desensitized state of the channel than the open state.