Barbiturates bind in the GLIC ion channel pore and cause inhibition by stabilizing a closed state

: Barbiturates modulating Despite more than a century of use in clinical practice, the prototypic binding site for this class of drugs within pLGICs is yet to be described. In this study, we present the first X-ray structures of barbiturates bound to GLIC, a cationic prokaryotic pLGIC with excellent structural homology to other relevant channels sensitive to general anesthetics and, as shown here, to barbiturates, at clinically-relevant concentrations. Several derivatives of barbiturates containing anomalous scatterers were synthesized and helped to unambiguously identify a unique barbiturate binding site within the central ion channel pore in a closed conformation. In addition, docking calculations around the observed binding site for all three states of the receptor, including a model of the desensitized state, showed that barbiturates preferentially stabilize the closed state. The identification of this pore binding site sheds light on the mechanism of barbiturates inhibition of cationic pLGICs and allows the rationalisation of several structural and functional features previously observed for barbiturates.

The arrival of the first barbiturates into clinical practice at the beginning of the 20th century caused a revolution in the pharmacology-based treatments of psychiatric and neurological disorders due to their sedative and anxiolytic properties (1). As anticonvulsants, barbiturates were responsible for the first truly effective management regimens for epileptic seizures, whereas in the field of general anesthesia, they were the first injectable agents used for induction. Between 1920 and 1950, barbiturates were the most prominent class of drugs used as sedatives and hypnotics (2). As they are prone to cause respiratory depression, they have now been mostly replaced with the comparatively safer benzodiazepines (3). Despite this, barbiturates still retain several important sedative-hypnotic roles in medical treatment such as for asthmatic and gastrointestinal functional disorders, certain types of epilepsy, violent convulsions, and cerebral hemorrhages. Most importantly, they are still used for the induction of general anesthesia.
Although barbiturates have enjoyed widespread use during the last century, insights into the molecular basis of their action have only arisen during the last couple of decades. Barbiturates are thought to modulate the action of various neural receptors, such as the AMPA/kainate receptors and the P/Q high voltageactivated calcium channels (4,5), as well as members of the pentameric ligand-gated ion channel family, which are major mediators of synaptic transmission (6). It has been shown that barbiturates alter the action of anionic GABA A and glycine receptors (GABA A R 5 and GlyR, respectively), promoting their activation and subsequent polarization of neurons (7)(8)(9), as well as inhibiting cationic ion channels responsible for triggering interneuronal signaling, such as the nicotinic acetylcholine receptors (nAChRs) and 5HT 3 (serotonin) receptors (5HT3Rs) (10 -12). Enhancement of GABA A R signaling is thought to be the primary mechanism responsible for the depressant effects of barbiturates (13), although other targets, such as the nAChRs and their inhibition, might still play an important role in the overall depression of the nervous system.
Pentameric ligand-gated ion channels are either homopentamers or heteropentamers in which subunits are arranged laterally around a central axis of symmetry, thus forming the central pore (14). A single subunit consists of an N-terminal extracellular agonist binding domain, containing ϳ200 amino acids forming several ␤-strands arranged in immunoglobulinlike ␤-sandwich folds, and a transmembrane domain consisting of four ␣-helices, M1-M4. Most of these helices are exposed to the cellular membrane, except for M2, which lines the central ion channel pore. pLGICs can adopt several conformations, such as the closed, open, and desensitized states, which mostly depend on the presence and the length of exposure to agonists (15).
So far, most of the progress on characterizing the binding site of barbiturates within both the GABA A R and the nAChR has been made using binding and photolabeling studies. In the case of cationic nAChRs, such studies have shown that barbiturates bind within the closed or desensitized central ion channel pore, because most of them fully inhibit the binding of channel blockers stabilizing the receptors in these states (16), whereas photoreactive barbiturate derivatives are capable of labeling several residues within the channel pore of receptors in the desensitized state (17). However, detailed structural characterization of the barbiturate binding site within pLGICs is still lacking. Although crystal structures of barbiturate-bound globular proteins (18) have been determined, there is no evidence that these complexes can give us insight into the interactions existing between the ligand and its actual membrane-bound receptor.
To further our knowledge on barbiturate mechanism of action, we used X-ray crystallography to study their binding with the locally closed form of GLIC (Gloeobacter violaceus ligand-gated ion channel), a cationic prokaryotic homologue of pLGICs. GLIC shows remarkable structural homology with the previously solved Torpedo marmorata nAChR structure by electron microscopy (19,20), and shares similar electrophysiological responses to general anesthetics (21). In this study, we describe firstly a potent effect of barbiturates on GLIC, as shown by electrophysiology, and then crystal structures at atomic resolution of GLIC-barbiturate complexes. This reveals a binding site for barbiturates in the closed central ion channel pore of cationic pLGICs, thus providing data on the molecular basis for inhibition by barbiturates. In addition, we built a model of the desensitized form of GLIC based on the only avail-able structure of a representative D-state in the pLGIC family, namely the GABA A receptor ␤3 homopentamer (22), and performed docking calculations of these ligands on the closed, open, and (modeled) desensitized states of GLIC. This allowed us to qualitatively predict the changes in affinity for the observed (pore) binding site in different conformational states of the receptor. The predicted relative affinities suggest a mode of action for barbiturates based upon stabilizing the closed conformation of the receptor.

Results
Chemical Synthesis of Selenocyanobarbital and Thiopental-As membrane proteins are usually prone to limited resolution on X-ray crystallography, we used three barbiturate derivatives harboring atoms that produce specific anomalous signal: the commercially available 5-(2-bromo-ethyl)-5-ethyl-pyrimidine-2,4,6-trione, which we called "bromobarbital," and thiopental and selenocyanobarbital, both chemically synthesized in this study. Selenocyanobarbital 6 was synthesized in a five-step sequence (Fig. 1A). Briefly, dimethyl 2-ethylmalonate 1 was first alkylated in the presence of sodium hydride by 2-(2-bromoethoxy)tetrahydropyran 2. The latter was previously prepared according to a procedure described in the literature, giving the 2,2-disubstituted malonate 3 in a 47% yield (23). Then, the -iodo-barbiturate intermediate 5 was synthesized over three steps: (i) the condensation of urea with malonate 3 in the presence of sodium hydride, followed by (ii) the hydrolysis of the tetrahydropyran ether with p-toluenesulfonic acid in methanol, and then (iii) the iodination of the resulting alcohol afforded the expected iodinated barbiturate in a 47% yield. The reaction of the iodide 5 with potassium selenocyanate in THF furnished the targeted selenocyanobarbital 6 in a 60% yield. Due to stock unavailability, thiopental 7 was synthesized in two steps starting from the commercially available pentobarbital sodium salt (Fig. 1B). The latter was first neutralized with a 1 M HCl solution to release the barbituric acid form, which was then transformed into thiopental 7 in a modest 20% yield, by using Lawesson reagent in refluxing anisole.
Barbiturates Inhibit GLIC at Clinically Relevant Concentrations-To determine whether GLIC is sensitive to modulation by barbiturates, we expressed wild-type GLIC channels in Xenopus oocytes and characterized the effect of pentobarbital as well as the three barbiturate derivatives using two-electrode voltage clamp electrophysiology. None of the tested barbiturates evoked a direct response at GLIC when applied at neutral pH 7. However, in all cases, when the barbiturates were co-applied in the presence of the orthosteric agonist for GLIC (at pH 5.5, Ϸ pEC 10 -20 ), a pronounced inhibition of the protoninduced current was observed (Fig. 2, A-E). The barbituratemediated inhibition was reversible with subsequent applications of pH 5.5 recording solution (in the absence of barbiturate) eliciting undiminished current amplitudes when compared with that obtained prior to drug-application (Fig. 2, B-E). The concentration-inhibition curves for pentobarbital and bromobarbital revealed similar inhibitory potencies at GLIC (IC 50 of 101 Ϯ 4.9 and 177 Ϯ 27.2 M, respectively, n ϭ 5 and 6), with near complete inhibition of the proton response at 1-3 mM ( Fig. 2A). By contrast, the inhibition curve for thiopental revealed an approximate 4-fold increase in potency (IC 50 of 24.4 Ϯ 1.2 M, n ϭ 5) when compared with pentobarbital. Intriguingly, in most recordings at higher concentrations of thiopental (Ն300 M), a rebound current was observed upon washout of the drug that virtually attained the same amplitude as the control proton-activated current (Fig. 2F).
The micromolar to millimolar concentration range for the barbiturates to modulate GLIC channels is similar to that reported for these compounds to impart their effects at eukaryotic pLGICs (24). Moreover, it is notable that independent pharmacological screening studies of the related prokaryotic homolog ELIC (Erwinia chrysanthemi ligand-gated ion channel) revealed insensitivity to modulation by pentobarbital (25). Given the sensitivity of GLIC to pharmacologically relevant concentrations of a range of barbiturate compounds (Fig. 2), we therefore sought to determine whether their mechanism of action utilizes a common binding site on GLIC.
Barbiturates Bind within the Ion Channel Pore of GLIC-To explore the possibility that barbiturates are binding to the channel in an open state, as shown for other anesthetic compounds (26), we collected several diffraction data sets of wild-type GLIC crystals at pH 4 (open form) with varying concentrations of three barbiturate derivatives. However, we were unable to detect any barbiturate binding in the open receptor. This is in agreement with the inhibitory action of barbiturates on GLIC, which suggests that they should stabilize the closed state. To determine whether binding could instead be observed in the closed state of the receptor, we co-crystallized the different bar- biturate compounds with a locally closed mutant of GLIC (27). This mutant has a resting-like conformation in the transmembrane domain where other general anesthetics, such as xenon or bromoform, were recently found to bind, inside the channel pore ( Fig. 3) (28,29).
All three barbiturate derivatives ( Fig. 4A) could be co-crystallized with the locally closed form of GLIC and diffracted up to 3.5-3.0 Å resolution (Table 1). Strikingly, all barbiturate molecules were observed bound within the central ion pore, between the pore-lining Thr-2Ј and Ile-9Ј side chains (Fig. 4). The main interactions responsible for binding are hydrogen bonds formed between the serines at position 6Ј and the carbonyl groups of the barbital ring, although van der Waals forces also play a role, with the long aliphatic tails of the barbiturates being further stabilized by either isoleucines or threonines at levels 9Ј and 2Ј, respectively. No evidence of binding was observed to other sites. The RMSD values of the barbituratebound structures and the apo-protein were small (less than 0.6 Å), with the biggest differences in C␣ position being found at the short intracellular loops between helices M3 and M4, as well as at the top of the flexible extracellular domain, both of which participate in crystallographic contacts. The largest value of RMSD was 0.587 Å for the thiopental bound form of GLIC.
The barbiturates used in the crystallization experiments were a racemic mixture of both enantiomers, therefore dimin- . This site overlaps with the previously described xenon (28) and bromoform (29) binding sites in GLIC. ishing the observability of the aliphatic barbiturate tails. Because of this, their orientation was fitted based mainly on the anomalous peaks of bromine, sulfur, or selenium atoms in the bromobarbital, thiopental, and selenocyanobarbital derivatives, respectively (Fig. 4).
In Bromobarbital, the second carbonyl is positioned between two serine 6Ј side chains at approximately equal distances of 2.9 and 3.0 Å (Fig. 4B), thus indicating the existence of hydrogen bonds between them. The closest serine among those two could also participate in a weaker hydrogen bond with the barbiturate ring nitrogen. This also permits the orientation of one of the other barbiturate carbonyl groups 2.7 Å away from the opposite serine 6Ј and the formation of an additional hydrogen bond (Fig. 4C). The bromoethane tail is held by van der Waals forces in the proximity of isoleucine side chains at the level of the 9Ј ring of M2 residues (Fig. 4B).
In thiopental, the extra methyl group in the isoamyl chain (Fig. 4A) and the more lipophilic nature of the thiol group as a hydrogen bond acceptor impose additional restraints on the possible orientations the ligand can assume in this position. Firstly, hydrogen bonds involving the ring carbonyls are expected to be preferred over those formed by the thiol group, whereas the extra methyl group severely restrains the orientation of the aforementioned ring carbonyls due to clashes between the level 6Ј serines and the methyl group. Consequently, only two hydrogen bonds can form between thiopental and the residues lining the ion channel pore: one between each carbonyl and an adjacent serine. The carbonyls are positioned at distances of 2.8 and 2.7 Å from the closest serines (Fig. 4C). The distances between the rest of the longer aliphatic chain and the threonines at position 2Ј range from 4.7 to 5.2 Å (Fig. 4C). It appears that the long aliphatic chain is less flexible facing downwards, as indicated by the observable density in the 2F o Ϫ F c electron density map. In addition, the anomalous signal of sulfur, located upwards, enabled us to model the thiopental with its aliphatic chain pointing to the bottom of the pore. The different binding mode of thiopental could be related to the specific rebound current observed upon its washout on oocytes when compared with the other barbital derivatives with shorter and simpler aliphatic chains.
Selenocyanobarbital is oriented with its axis of pseudosymmetry parallel to the ion channel, with the side carbonyls of the barbital pyrimidine ring forming hydrogen bonds with serines at level 6Ј (Fig. 4D). Selenocyanobarbital was modeled according to a compromise between the experimental 2F o Ϫ F c density, the selenium anomalous peak, and the propensity to form hydrogen bonds. The observed anomalous peak falls in the middle of the pore axis, possibly because of the 5-fold symmetry of the ion channel rendering barbiturate binding possible in five equivalent different conformations, generated by a rotation of 72°around the C5 pore axis. The observed anomalous signal thus represents an average on the C5 axis of the individual anomalous signal due to the different possible binding poses within the pore (Fig. 4, E and F). In the modeled selenocyanobarbital pose, one of the side carbonyls is positioned between two adjacent serine 6Ј side chains 2.7 and 3.0 Å away, whereas the distance between the other carbonyl and a third serine residue is 3 Å, thus indicating the formation of three hydrogen bonds in total. The pyrimidine ring nitrogen next to the first carbonyl could potentially weakly interact with the adjacent serine. The second carbonyl is facing downward, close to the threonine ring at level 2Ј, the distance from the closest threonine oxygen being 3.6 Å, indicating the presence of a weak hydrogen bond. It could be postulated that the reason the pyrimidine ring of selenocyanobarbital is not found in a sideways orientation similar to that for bromobarbital is because of the length, rigidity, and polar nature of its long terminal tail, which is absent in clinically used barbiturates. The terminal carbon of the ethyl tail is oriented toward the isoleucine 9Ј ring.

Discussion
Previous research, using a photoreactive derivative, has shown the ability of barbiturates to bind to sites within the ion channel pore of the nAChR, a cationic pLGIC (17). In this study, we solved the first crystal structures of barbiturate binding to this class of receptors, exemplified by the locally closed GLIC isoform, and we unequivocally located the barbiturate binding site between residues 2Ј and 9Ј in the ion channel (Fig. 3). The main interactions between barbiturates and the GLIC protein appear to be hydrogen bonds between the barbiturate pyrimidine ring carbonyl groups and the 6Ј serines. We do observe that the aliphatic tails at position 5 on the ring can further stabilize the binding of the ligand by interacting with 9Ј isoleucines through van der Waals forces. Interestingly, the presence of thiopental's branched aliphatic tail appears to invert the orientation of the molecule within the pore when compared with bromobarbital and selenocyanobarbiturate. This could be caused by the additional steric hindrance of the tail arising from the extra methyl group close to the pyrimidine ring, increasing the probability of a steric clash with the surrounding residues and thus favoring the inverted orientation. This agrees with the previously described structural and functional relationships found in barbiturates with the same empirical formula (16). For example, barbiturates containing an unbranched position 5 alkyl tail have higher affinity for the ion channel pore (less steric hindrance); however, if a branch exists on the alkyl tail, higher affinities will be observed if the branching is located farther away from the pyrimidine ring, allowing an upward orientation with the branch closer to the 9Ј isoleucines. The characterization of this binding site presented here thus allows the rationalization of how modifications to barbiturate structure can alter their functional profile.
To predict which state of GLIC is most stabilized by the binding of barbiturates within the ion channel pore, we performed docking calculations similar to the ones described recently (30) on the closed, open, and (modeled) desensitized states of GLIC. Although the structure of this receptor in a desensitized state is currently unknown, the assumption that all pLGICs follow closely related global conformational changes during transitions between different states allowed us to build a putative model of GLIC in the desensitized state, based on the recently published structure of the apparently desensitized GABA A receptor ␤3 homomer (22). Our docking calculations showed that in comparison with the closed pore, in the open and desensitized states, the distances between adjacent 6Ј serines are increased, giving more conformational freedom to barbiturate binding within this region and allowing deeper penetration into the pore, possibly forming additional interactions with the threonine side chains at level 2Ј, and meanwhile losing the stabilizing effect of 9Ј isoleucines (supplemental Figs. 7 and 9). As a result, the predicted dissociation constants of almost all ligands increase dramatically during the opening and subsequent desensitization of the receptor (supplemental Fig. 6). The calculated binding modes for thiopental, selenocyanobarbital, and bromobarbital in the open state of GLIC were almost identical with their axes of pseudosymmetry perpendicular to that of the protein (supplemental Fig. 7). Interestingly, the calculated binding modes of all barbiturates within the closed pore matched the ones observed by crystallographic methods (supplemental Fig. 8), although thiopental was observed to bind with equal binding energies with its tail found both above and below the level 6Ј serines. Although the energetics of the docking are not expected to be exact due to a rather crude model for calculating them, these docking calculations qualitatively suggest that barbiturate binding in the ion channel is stronger when the receptor resides in its closed state (supplemental Table). Ligands binding to this site would therefore tend to stabilize this particular state of the receptor, i.e. a non-activated, shut state of the channel. Binding studies on barbiturates have shown that derivatives with unbranched position 5 tails bind to the nAChR pore preferentially in its closed state, whereas others, such as pentobarbital with its branched methylbutyl tail, prefer to bind to the open state, in which the pore diameter is expanded above 2Ј residues (16). This might indicate the presence of a second potential barbiturate binding site within the pore, positioned higher than the one described here, where an increase in the pore diameter would reduce the steric hindrance arising from the presence of extra methyl groups on the barbiturate aliphatic tails. A similar proposal was suggested following photolabeling (17), which identified a putative barbiturate binding site between 9Ј and 16Ј in the nAChR channel. However, despite this, no barbiturate binding sites have been crystallographically observed within the open-state GLIC ion channel. Indeed, our own efforts to obtain structures of barbiturates bound to wild-type GLIC in the open state (at pH 4) were not successful. It is remarkable that, instead, a tight pentagon of water molecules occupies the 6Ј position in the X-ray structure of open GLIC form, in a manner similar to ice type IX (supplemental Fig. 10) (31). This water network is important for ion permeation through the hydrophobic barrier in the pore. Moreover, in the open structure, six detergent molecules are lodged within the ion channel pore, possibly precluding ligand binding in this region, especially for low affinity ones.
The present study provides direct experimental evidence for the central ion channel pore of cationic pLGICs being an important site for binding barbiturates to this class of receptor. It is clear that this site may also play a larger role in the overall effects of many other general anesthetics because smaller volatile general anesthetics, such as xenon (28) and bromoform (29), have recently been observed to bind to the same site in the locally closed GLIC form (supplemental Fig. 10), whereas propofol has also been predicted to bind in the central ion channel pore within several states of this receptor (32). In contrast, the binding site for propofol and desflurane in the open form of GLIC was located by crystallography to an intrasubunit cavity at the top of the transmembrane domain ␣-helical bundle (26), and this was recently validated by chemical labeling experiments (33). Further studies using mutants of GLIC activated by propofol are currently being pursued in our lab to fully resolve this issue. For barbiturates, because the site described here is found in the closed state of the pore, the inhibitory effects of this class of drugs on cationic pLGICs are readily explained. This site cannot be responsible for the activation of anionic pLGICs, as (i) it blocks the pore and (ii) it has been shown by photoaffinity labeling experiments that in this case the binding site is located at the interface between subunits in the transmembrane domain (34). However, the inhibition potency of this binding site is also likely to be significant for anionic pLGICs, as it has been shown to mediate alcohol inhibition on GABA A receptor (35). The identification of the true binding sites of barbiturates onto GABA A receptors awaits their crystal structure and would be greatly facilitated through the use of molecules such as the ones synthesized for this study thanks to their anomalous signal.

Synthesis of Barbiturate Derivatives
Commercially available reagents were used throughout the synthesis without further purification. Analytical thin layer chromatography was performed on Merck 60F-254 precoated silica (0.2 mm) on glass and was revealed by UV light. 1 H and 13 C NMR spectra were recorded on a Bruker AC 300 at 300 and 75 MHz. The chemical shifts for 1 H NMR were given in ppm downfield from tetramethylsilane (TMS) with the solvent resonance as the internal standard. Infrared (IR) spectra were obtained as neat films on Bruker Vector22 spectrophotometer. High-resolution mass spectroscopy (electrospray ionization) analysis was performed with a time-of-flight mass spectrometer yielding ion mass/charge (m/z) ratios in atomic mass units. Purity of selenocyanobarbital was determined by reverse phase HPLC using a 150 ϫ 2.1-mm (3.5 m) C18-column; the compound was eluted over 20 min with a gradient from 95% acetonitrile/5% (H 2 O ϩ 0.1% HCO2H) to 100% (H 2 O ϩ 0.1% HCO 2 H). The full protocol of the synthesis is provided in the supplemental Methods (see supplemental Figs. S1-S5).

Electrophysiology
Oocyte Preparation-The ovaries were removed from female African Xenopus laevis using procedures approved by the UK Animals (Scientific Procedures) Act 1986. Oocytes were separated from the ovaries by using collagenase type 1 (Worthington) dissolved in a Ca 2ϩ -free OR2 solution containing: 85 mM NaCl, 5 mM HEPES, 1 mM MgCl 2 , pH 7.6, adjusted with 1 M KOH. After ϳ3 h of treatment, defolliculated oocytes were washed three times with OR2, and then stored in a modified Barth's solution containing: 88 mM NaCl, 1 mM KCl, 0.33 mM Ca(NO 3 ) 2 , 0.41 mM CaCl 2 , 0.82 mM MgSO 4 , 2.4 mM NaHCO 3 , 10 mM HEPES, pH 7.6, adjusted with 1 M NaOH. Oocytes were injected into the nucleus with 27.6 nl of GLIC cDNA and then incubated at 17°C in modified Barth's solution.
Two-electrode Voltage Clamp Recording-Oocytes expressing GLIC were used the next day after injection. They were superfused with a solution containing: 100 mM NaCl, 2 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM MES, pH 7.4, adjusted with 1 M NaOH. Currents were recorded using two intracellular electrodes filled with 3 M KCl, with an Axoclamp 2B amplifier, with a Digidata 1322A interface in conjunction with pCLAMP 8 (Molecular Devices). Currents were digitized at 500 Hz and filtered at 50 Hz. Oocytes were voltage-clamped at Ϫ60 mV, and experiments were conducted at room temperature.
Data Analysis-Concentration response data were fitted with the Hill equation using Origin version 6 (OriginLab) software where I and I max represent the current induced by various proton concentrations and the maximal proton-activated current respectively, [A] is the proton concentration, EC 50 is the concentration producing a half-maximal response, and n H is the Hill coefficient. Data points were normalized to provide a maximal current of 1 in control.
To assess the potency of antagonism, inhibition-concentration relationships for pentobarbital, bromobarbital, and thiopental were fitted using where I/I max represents the relative current induced by a pH 5.5 proton concentration in the presence of varying concentrations of barbiturate ([B]). IC 50 is the concentration of barbiturate inducing a half-maximal reduction in the agonist current, and n H is the Hill coefficient.

Crystallization
GLIC was expressed and purified following the previously described protocol (27). It contained a K33C/L246C mutation, which has been shown to induce the formation of a disulfide bridge between loop 2 of the extracellular domain and the M2-M3 loop of the transmembrane domain. This effectively renders the receptor stable in a locally closed state, where the pore is trapped in a closed conformation, whereas the extracellular domain remains in a state similar to that of the open receptor. This so-called locally closed conformation is also observed in single-point mutations of GLIC such as E243P or H235F (27) as well as other mutations in the M2-M3 loop (36).
Crystals were grown by the use of the vapor diffusion method in hanging drops at 18°C. The protein samples were supplemented with bromobarbital, selenocyanobarbital, or thiopental to give final barbiturate concentrations of 5, 10, and 20 mM, respectively. The samples were then mixed at a 1:1 ratio with the reservoir solution (100 mM NaAc, pH4, 400 mM NaSCN, 3% DMSO, 16% glycerol, 12-14.5% PEG 4000). The drops were microseeded with previously obtained crystals of locally closed GLIC. Small parallelepiped-shaped crystals appeared overnight, reaching their maximum dimensions after 1 week. Crystals were collected on cryoloops and immediately flash-frozen and stored in liquid nitrogen.

Data Collection and Processing
The collection of numerous single-crystal datasets took place on beamlines Proxima-I (37) and Proxima-II (38) at Synchrotron Soleil, as well as ID23 (39) and ID29 (40) at the European Synchrotron Radiation Facility. Data collection wavelengths were set at the Kappa peak for each anomalous scatterer present in the ligand of interest. The collected data were integrated with XDS (41) and further scaled with Scala from the CCP4 suite (42). Similarly to the previously described WT and K33C/ L246C GLIC (27), the crystals tested belonged to a C2 space group with one pentamer in the asymmetric unit. To obtain the initial phases of the models, the previously described structure of apo-K33C/L246C GLIC (27) was used as a starting model in Refmac5, followed by ϳ60 -80 cycles of refinement with Buster (43) for each model. NCS restraints were used throughout the refinement process; each cycle of refinement was followed by inspection of the model in the map, and manual adjustments were made using Coot (44). After the initial cycles of refinement, the electron density and anomalous signal maps of the obtained models allowed the identification of ligand binding sites and their subsequent placement within the model structures. Validation of structures was performed by MolProbity (45), and all figures were prepared using Chimera (46).

Modeling and Docking
The model of GLIC in a desensitized state was built by standard homology modeling techniques using as a template the only available structure of a desensitized state in the pLGIC family, the ␤3 homopentameric GABA A receptor (22). We first optimally positioned C␣ coordinates of 7-amino acid-long fragments of the open state structure of GLIC onto the structure of the GABA A receptor. Here optimality refers to finding the positions of the fragments with a minimal local RMSD between the C␣s of the fragments and the C␣s of the ␤3 subunit. At the end of this process, the trace of a model for the desensitized GLIC is obtained, and this model includes only the positions of its C␣ atoms. The full backbone was then rebuilt using a sequence of rigid 4-amino acid-long protein fragments that are concatenated without any degrees of freedom. As described earlier, fragments chosen from a library of representative fragments found in the Protein Data Bank (PDB) are fit to the C␣ of the trace using a gradual build-up method (47). The procedure was recently updated and optimized in a Fortran program, C␣2Full, written by one of us (Patrice Koehl), and is freely available from the author upon request. The C␣ of the reconstructed backbone differed from the C␣ of the trace by less than 0.5 Å. The side chains were then rebuilt using a rotamer library and the mean field optimization method (48), and the whole model was energy-minimized using the CHARMM19 force field.
Coordinates of barbiturate derivatives were generated by Buster (43). Docking of barbiturates within the experimentally observed binding site in the locally closed, open, and modeled desensitized states of GLIC was performed using AutoDock Vina (49) on a Dell PC with an Intel Xeon X5460 quad core processor. The initial conformations of ligands were randomized, and the bonds of the flexible aliphatic barbiturate tails were allowed to freely rotate during the docking calculations. The target structures were the transmembrane domains (residues 193-315 of each subunit) of the locally closed (PDB ID: 3TLV), open (PDB ID: 4HFI), and desensitized model state of GLIC. Before docking, all water and ion atoms, as well as detergent molecules were removed from the structures; all hydrogen atoms were added, partial charges were also included, and nonpolar hydrogens were merged. The docking calculations focused on the experimentally observed barbiturate binding site within the ion pore between residues 2Ј and 9Ј. The search volume was a 16 Å-side cube that included all of the aforementioned residues, the side chain bonds of which were allowed to freely rotate during the calculation. The binding energies calculated by AutoDock Vina were converted to dissociation constants, K D , in molar units using: ⌬G ϭ ϪRT ln(K D ), where R is the gas constant (1.99 ϫ 10 Ϫ3 kcal K Ϫ1 mol Ϫ1 ) and T ϭ 300 K.
Author Contributions-Z. F. and R. R. performed and analyzed the X-ray crystallography experiments. D. L. performed and analyzed the electrophysiology experiments. M. D. and T. S. designed and analyzed the experiments. E. D., S. D. C., and D. J. designed, performed, and analyzed the chemical synthesis of barbiturate derivatives. P. K. designed and wrote the program to perform GLIC D-state modeling. Z. F., R. R., T. S., and M. D. co-wrote the paper. M. D. and T. S. conceived and coordinated the study.