Identification of Novel Specific Allosteric Modulators of the Glycine Receptor Using Phage Display*

The glycine receptor (GlyR) is a member of the Cys-loop superfamily of ligand-gated ion channels and the major mediator of inhibitory neurotransmission in the spinal cord and brainstem. Many allosteric modulators affect the functioning of members of this superfamily, with some such as benzodiazepines showing great specificity and others such as zinc, alcohols, and volatile anesthetics acting on multiple members. To date, no potent and efficacious allosteric modulator acting specifically at the GlyR has been identified, hindering both experimental characterization of the receptor and development of GlyR-related therapeutics. We used phage display to identify novel peptides that specifically modulate GlyR function. Peptide D12-116 markedly enhanced GlyR currents at low micromolar concentrations but had no effects on the closely related γ-aminobutyric acid type A receptors. This approach can readily be adapted for use with other channels that currently lack specific allosteric modulators.

The glycine receptor (GlyR) 2 is a member of the Cys-loop superfamily of ligand-gated ion channels, including also the ␥-aminobutyric acid type A (GABA A ) and serotonin-3 receptors. They share a number of structural features, including ligand-binding sites in the extracellular N-terminal domain and a transmembrane domain consisting of four segments, with a large intracellular loop connecting segments 3 and 4. Individual channels consist of five subunits co-localized with the transmembrane domain 2 segment of each subunit lining the anionconducting pore (1,2). Two classes of GlyR subunits have been identified: the ␣ subunits, of which there are four subtypes, and a single ␤ subunit (3). Most native GlyRs in adult animals consist of heteromeric ␣1␤ subunits, although homomeric ␣2 receptors are the predominant form found prenatally (4). GlyRs constitute the major inhibitory neurotransmitter receptor system in the brainstem and spinal cord (5), where they are thought to play a role in the modulation of pain signals and in the effects of volatile anesthetics (1). Some GlyR mutations result in the startle disorder hyperekplexia. GlyRs are also found throughout the brain, including the thalamus, hippocampus, and nucleus accumbens, where they were recently shown to be involved in the reinforcing properties of ethanol (6). The GlyR is only one of multiple ion channels and receptors thought to play a role in pain perception and alcohol and volatile anesthetic effects and in determining the state of neuronal excitability. The isolation of the role of the GlyR is hindered by the fact that, to date, no potent and efficacious allosteric modulator acting specifically at the GlyR has been identified. Phage display involves the expression of a random library of peptides on the coat proteins of bacteriophage. This method has long been used to identify peptides that can bind with high affinity to selected targets and to aid in identifying binding motifs (7). We combined phage display technology with standard electrophysiological testing to identify peptides that allosterically modulate GlyR function without affecting two closely related GABA receptors.

EXPERIMENTAL PROCEDURES
Materials-All chemicals were obtained from Sigma. Xenopus laevis was purchased from Xenopus Express (Homosassa, FL).
Human Embryonic Kidney (HEK) Cell Culture and Expression of GlyRs-HEK 239 cells obtained from American Type Culture Collection were grown according to standard procedures (8). Cells were cultured at 37°C in a 5% CO 2 atmosphere in Dulbecco's modified Eagle's medium with L-glutamine, sodium pyruvate, and 10% fetal bovine serum (Invitrogen). Cell lines were split every 5 days with trypsin/EDTA in Hanks' balanced salt solution (Invitrogen) for 25 cycles, after which new aliquots of early passage cells were started. Cells were transfected with 4 g of GlyR ␣1 subunit cDNA using PolyFect reagent (Qiagen). Control cells (untransfected HEK cells) were exposed to PolyFect and then split with no cDNA exposure. All cells were incubated for at least 48 h before use in panning.
Phage Display-On panning day 1, a plate of control HEK cells was washed three times with 0.01 M phosphate-buffered saline (PBS) containing 8.2 mM NaPO 4 , 1.5 mM KH 2 PO 4 , 137 mM NaCl, and 2.7 mM KCl with 1.5% bovine serum albumin (BSA) and 0.1% Tween (PBS/BSAϩT). Next, an aliquot containing 2 ϫ 10 11 phage from the D12 phage library (New England Biolabs) was diluted in 1 ml of PBS/BSAϩT. Phage were then applied to blank (control) HEK 293 cells and rocked gently at room temperature for 30 min. Phage that did not bind in this negative selection procedure were removed from the plate with a pipette, applied to the plate of GlyR-expressing cells, and rocked gently at room temperature for 60 min. Non-binding phage were discarded, and the plate was washed five times with PBS/BSAϩT. At this time, positive selection was complete, and the only step remaining was the isolation of phage. Elution of the bound phage was performed by lowering the pH using 0.2 M glycine HCl (10 M HCl buffered to pH 2.2 with glycine) plus 1 mg/ml BSA and rocking at room temperature for 10 min. Eluate was removed and neutralized with 150 l of 1 M Tris-HCl (pH 9.0). Titering was performed, and the remainder of the eluate was added to the 20 ml of Escherichia coli at A 600 in LB broth.
After 4.5 h of incubation, the culture was transferred to a 50-ml Falcon centrifuge tube and spun at 10,000 rpm for 10 min at 4°C. The supernatant was transferred to a fresh tube and respun. The upper 80% of the supernatant was again transferred to a fresh tube, and a one-sixth volume of polyethylene glycol/NaCl (20% (w/v) polyethylene glycol 8000 and 2.5 M NaCl) was added. Phage were allowed to precipitate overnight at 4°C.
On panning day 2, polyethylene glycol precipitates were spun at 10,000 rpm for 15 min at 4°C. The supernatant was decanted, and the precipitate was spun again. Residual supernatant was removed using a pipette. The pellet was resuspended in 1 ml of Tris-buffered saline (50 mM Tris-HCl (pH 7.5) and 150 mM NaCl), transferred to a 1.7-ml microcentrifuge tube, and spun at 10,000 rpm for 5 min at 4°C. In a fresh microcentrifuge tube, the suspended phage was reprecipitated with polyethylene glycol/NaCl on ice for 60 min. After spinning at 10,000 rpm for 10 min at 4°C, the supernatant was discarded. The pellet was resuspended in Tris-buffered saline and spun again for 1 min, and the supernatant was then transferred to a fresh tube and stored at 4°C. The amplified phage were titered. For successive panning rounds, the amplified phage were diluted in PBS/BSAϩT so that the input concentration was always 2 ϫ 10 11 virions. At the end of five rounds of panning, individual plaques from the most recent titer plates were isolated and incubated overnight in LB broth at 37°C with agitation. The overnight culture was then purified using the S.N.A.P. MiniPrep kit (Invitrogen), and phage DNA was sequenced inhouse using a Ϫ96gIII sequencing primer (New England Biolabs). Individual peptide sequences were sent to Peptide 2.0 Inc. (Chantilly, VA) for synthesis.
Oocyte Isolation and cDNA Nuclear Injection-Oocytes were surgically removed from X. laevis housed at 19°C on a 12h light/dark cycle. Stage V and VI oocytes were selected and placed in isolation medium containing 108 mM NaCl, 1 mM EDTA, 2 mM KCl, and 10 mM HEPES, and forceps were used to manually remove the thecal and epithelial layers. The oocyte follicular layer was removed using a 10-min exposure to 0.5 mg/ml Sigma type 1A collagenase in buffer containing 83 mM NaCl, 2 mM MgCl 2 , and 5 mM HEPES. A 30-nl sample of the glycine or GABA receptor subunit cDNAs (1.5 ng/30 nl) in a modified pBK-CMV vector (9) was injected into the animal poles of oocytes by the "blind" method of Colman (10) using a digital micropipette (10 -15-m tip size) attached to a microdispenser. Oocytes were stored in 96-well plates in the dark at 19°C in modified Barth's saline (88 mM NaCl, 1 mM KCl, 2.4 mM NaHCO 3 , 10 mM HEPES, 0.82 mM MgSO 4 ⅐7H 2 O, 0.33 mM Ca(NO 3 ) 2 , and 0.91 mM CaCl 2 at pH 7.5) plus 2 mM sodium pyruvate, 0.5 mM theophylline, 10 units/ml penicillin, 10 mg/liter streptomycin, and 50 mg/liter gentamycin and sterilized by passage through a 0.22-m filter. Oocytes expressed channels within 24 -96 h depending on the receptor expressed, and all electrophysiological measurements were made within 5 days of cDNA injection.
Oocyte Electrophysiological Recording-Oocytes were impaled in the animal poles with two high resistance (0.5-10 megohms) glass electrodes filled with 3 M KCl. Using a Warner Instruments OC-725C oocyte clamp, oocytes were voltageclamped at Ϫ70 mV while modified Barth's saline was perfused over them at a rate of 2 ml/min using a Masterflex peristaltic pump (Cole Parmer Instrument Co., Vernon Hills, IL) through 18-gauge polyethylene tubing. All drug solutions were prepared in modified Barth's saline. Drug applications (5-60 s) were followed by 5-15-min washout periods as appropriate.
Statistics-Current peak responses were measured from chart recorder tracings. Peptide effects were calculated as percent changes compared with the effects produced by glycine or GABA in the absence of peptide. In all cases, oocyte data were obtained from at least two different frogs. The percent enhancement or inhibition values obtained in the presence of peptide were compared by either one-or two-way analysis of variance, as indicated, to determine statistical significance, with a criterion of p Ͻ 0.05 required.

RESULTS
We used a commercially available phage display library, expressing more than two billion unique sequences, to isolate peptides that bind to GlyR ␣1 expressed in HEK 293 cells. Homomeric GlyR ␣1 was chosen as the target rather than heteromeric GlyR ␣1␤ because of its invariant stoichiometry: heteromeric ␣1␤ receptors contain ␣␣, ␣␤, and ␤␣ intersubunit interfaces. In addition, we wanted to avoid the possible complication that some receptors in our population were ␣1 homomers while others were ␣1␤ heteromers. A modified phage display protocol (11) involved first applying phage to untransfected HEK 293 cells in the negative selection step to remove those phage that bound to various endogenous targets that were not of interest. Phage that did not bind in this negative selection procedure were then applied to HEK 293 cells that expressed the GlyR ␣1 subunit, and this constituted the positive selection portion of the procedure. After five rounds of panning, 35 individual colonies were sequenced to identify the peptides inserted in the phage pIII protein that resulted in phage binding (Table 1). Three sequences were harvested from multiple colonies, implying selective enrichment of our phage pool. No clear homology was seen among the collected sequences. Because of the size of the extracellular domain of the GlyR and thus a large number of potential binding sites, this diversity of peptide sequences was not unexpected.
Peptides corresponding to 10 sequences were synthesized for characterization using two-electrode voltage-clamp electrophysiology. A maximally effective concentration of glycine (10 mM) was first applied to Xenopus oocytes to determine their levels of expression of the GlyR 2 days after GlyR ␣1 subunit cDNA injection. A low concentration of glycine eliciting a response equal to ϳ10% of the maximal current (EC 10 , where EC is the effective concentration) was identified in each oocyte and applied several times to ensure stability of responses. The oocyte was then incubated with a peptide at 30 M for 30 s before co-application of peptide with EC 10 glycine. If peptides failed to show an effect at 30 M, they were not tested further. This did not eliminate the possibility that effects might be seen at higher concentrations; however, these peptides would not be as interesting as those that were more potent. None of the peptides tested thus far had any effects in the absence of glycine, implying that they neither function as agonists at the orthosteric glycine-binding site nor affect the functioning of any other naturally occurring oocyte proteins that might influence the holding current. Some peptides, such as D12-106, inhibited the effects of glycine (Fig. 1a), whereas others, such as D12-105, acted as positive modulators (Fig. 1b). The 10 peptides tested thus far exhibited a variety of effects on GlyR function, ranging from 65% inhibition to 124% potentiation of the effects of EC 10 glycine (Fig. 1c).
Although our peptides were chosen based on their abilities to bind to the GlyR, there was a chance that they might also affect structurally similar channels. The Cys-loop receptor superfamily possesses considerable structural and sequence homology among its members (12). Within the superfamily, the GlyR is most similar to the other anion-conducting channels. Of these, the GABA 1 subunit is the most similar to the GlyR ␣1 subunit, with ϳ60% amino acid sequence identity (13), followed by other members of the GABA A group (14). For specificity testing, we compared peptide effects on the GlyR ␣1 subunit with those on homomeric GABA 1 and heteromeric ␣1␤2␥2S receptors. Although many types of GABA A receptors exist, those composed of ␣1, ␤2, and ␥2S subunits are most widely expressed in the adult nervous system. As with the GlyR experiments, oocytes expressing these channels were voltage-clamped at Ϫ70 mV, and the EC 10 GABA was established. Fig. 2 (a and b) show sample tracings of the effects of peptides D12-106 and D12-133 on EC 10 GABA responses generated by ␣1␤2␥2S receptors. D12-106 produced minor inhibition of GABA A receptor function, far smaller than its effects on the GlyR, whereas D12-133 was almost completely without effect. None of the five peptides tested on GABA A receptors produced marked effects (Fig. 2c). Furthermore, none applied alone affected holding currents and thus did not function as agonists at the GABA-binding site.
Peptide D12-116 produced a marked enhancement of GlyR function while not affecting GABA A ␣1␤2␥2S receptors. Fig. 3a shows a sample tracing of the current-enhancing effects of a 30 M concentration of peptide D12-116 on GlyR ␣1. In contrast, this concentration of peptide had no effects when co-applied with an EC 10 of GABA on ␣1␤2␥2S receptors (Fig. 3b) or had no effect on the holding currents of oocytes expressing GlyR ␣1 when it was applied alone for 30 s. However D12-106 inhibited the effects of glycine when it was co-applied with EC 10 glycine. b, sample tracing of the enhancing effects of a 30 M concentration of peptide D12-105 on GlyR function. D12-105 also did not affect the holding currents of oocytes but potentiated GlyR ␣1 function when it was co-applied with EC 10 glycine. c, percent changes in EC 10 glycine responses produced by 30 M concentrations of D12 peptides preapplied for 30 s before also being co-applied with EC 10 glycine. These peptides exhibited varying degrees of potentiation or inhibition at the GlyR, but none acted as agonists in directly activating the receptor. Data are expressed as the mean Ϯ S.E. of four to eight oocytes obtained from at least two frogs.

Specific Peptide Modulators of the Glycine Receptor
homomeric GABA 1 receptors (Fig. 3c). Neither GABA receptor showed any effects of 30 M D12-116, as summarized in Fig.  3d. A 100 M concentration of D12-116 was also without effect on both GABA receptors, implying that this result is due to true specificity at the GlyR rather than to potential differences in peptide potency among Cys-loop receptors. Because this peptide does not affect the channels most structurally similar to the GlyR, it is unlikely to have effects on even more distantly related targets.
A concentration-response curve was next performed using peptide D12-116 (Fig. 4a). When applied with EC 10 glycine, the threshold concentration of D12-116 eliciting an enhancing effect was ϳ3 M, with 40 M D12-116 producing a maximal effect; higher concentrations were less efficacious. This could imply that at higher concentrations, this peptide forms dimers with lower efficacy or that multiple binding sites with different affinities for this peptide exist on the GlyR. We then showed that 30 M D12-116 acted to left-shift the glycine concentration-response curve without any effect at saturating glycine concentrations (Fig. 4b). This result is in concordance with previous studies showing that other allosteric modulators of the GlyR, such as alcohols, volatile anesthetics, and low concentrations of zinc, also act by left-shifting glycine concentration-response curves (15)(16)(17).

DISCUSSION
One of the most pharmacologically relevant of the proposed roles of the GlyR in vivo is the one it may play in alcohol-induced behaviors. Many cell-surface proteins, such as voltageand ligand-gated ion channels, enzymes, and transporters, have been implicated as possible mediators of the effects of ethanol, but the relative importance of each putative target to the effects of alcohol in vivo remains poorly understood (18). A major impediment to the rational identification of compounds to treat alcohol abuse is this lack of understanding of which of these many possible molecular targets mediate which specific in vivo effects of alcohol, such as reinforcement, ataxia, or the D12-106 had no effect on the holding currents of oocytes expressing the GABA A GlyR when it was applied alone for 30 s and minimally inhibited the effects of GABA when it was co-applied with EC 10 GABA. b, sample tracing showing a lack of effect of a 30 M concentration of peptide D12-133 on GABA A ␣1␤2␥2S receptor function. D12-133 also did not affect the holding currents of oocytes in the absence of GABA and had no effect when it was co-applied with EC 10 GABA. c, percent changes in EC 10 GABA responses produced by 30 M concentrations of D12 peptides pre-applied for 30 s before also being co-applied with EC 10 GABA to ␣1␤2␥2S receptors. None of these acted as agonists in directly activating the receptor and had minimal to no effects in the presence of EC 10 GABA. Data are expressed as the mean Ϯ S.E. of two to six oocytes obtained from at least two frogs. FIGURE 3. Demonstration of peptide D12-116 specificity as a GlyR modulator. a, peptide D12-116 produced significant enhancement of GlyR ␣1 function when co-applied with EC 10 glycine for 30 s following a 30-s preincubation with peptide alone. Complete peptide washout was demonstrated by the return to the base-line glycine response 7 min after stopping peptide perfusion. b, sample tracing showing that peptide D12-116 had no effect on EC 10 GABA responses in oocytes expressing ␣1␤2␥2S subunits. c, sample tracing showing that peptide D12-116 had no effect on EC 10 GABA responses in oocytes expressing the GABA 1 receptor. d, bar graph showing the effect of a 30 M concentration of D12-116 on the GlyR and two related GABA receptors, all tested using EC 10 agonist. Peptide D12-116 robustly potentiated the glycine response in GlyR but had no effect on the GABA 1 and GABA ␣1␤2␥2S receptors. A one-way analysis of variance showed a significant difference in peptide responses among receptors (F(2,16) ϭ 17.9, p Ͻ 0.001). Data are expressed as the mean Ϯ S.E. of four to eight oocytes obtained from at least two frogs. development of tolerance. Enhancement of inhibitory GlyR function by ethanol is consistent with some of the behavioral consequences of its administration (6,19). However, because ethanol clearly affects multiple biochemical targets in addition to the GlyR, it has proved difficult to determine conclusively the roles that individual putative targets play in the various behavioral effects of this agent. There would thus be great utility in identifying compounds that can act as either ethanol mimics or ethanol antagonists but at only one putative alcohol target. For example, were we to administer our D12-116 peptide intrathecally to rats and observe ataxia, it would suggest that ataxia produced by ethanol is also mediated by enhancement of GlyR function. Although studies using knock-out and knock-in mice (20) have provided some successes, these methods, due to their nature, necessarily involve a departure from the wild-type tar-get phenotype. An alternative approach is the highly specific pharmacological manipulation of individual targets. Allosteric modulation offers the opportunity for subtle manipulations of a target rather than complete removal or blockade. Our identification of a GlyR-specific enhancing peptide allows one to emulate the effects of ethanol at just this one site. Thus, the identification of agents that modulate or mimic ethanol or volatile anesthetic actions at targets with high specificity could allow for the determination of the contributions of those sites to the effects of these compounds without requiring one to alter wild-type base-line GlyR function. A better understanding of the relative importance of individual targets could also lead to the development of more targeted therapeutics. This rational approach to drug discovery contrasts markedly with the serendipitous approach to the identification of drugs currently used in the treatment of alcoholism.
There are many instances of small molecules acting as specific allosteric modulators of proteins (21,22). For example, benzodiazepines bind at nanomolar concentrations to GABA A receptors containing specific subunits. There they can act as agonists, inverse agonists, or antagonists at a site distinct from the GABA-binding site to modulate GABA A receptor-mediated currents, but they exert no effects if applied on their own (23). Allosteric modulators at G-protein-coupled receptors have also been identified, such as calcimimetics and calcilytics, which shift calcium concentration-response curves at the calcium-sensing receptor to the left and right, respectively (24). In addition, a number of small molecules, peptides, and modified peptides are also known to have substantial effects on neurophysiology, such as enkephalins and conotoxins. Peptides that modulate the function of putative ethanol targets have also been identified and shown to alter ethanol-induced effects (25,26). For example, the octapeptide NAPVSIPQ antagonizes ethanol inhibition of L1-mediated cell adhesion and is protective against ethanol-induced embryotoxicity (25).
Previous work used phage display to identify novel peptides that demonstrated high affinities for their intended targets and that acted orthosterically as either agonists at the thrombin receptor (27) or antagonists at the urokinase receptor (28). In addition, Li et al. (29) used a peptide library to identify a peptidergic antagonist of the N-methyl-D-aspartate receptor. In recent years, there has been increasing interest in the development of novel therapeutics that act at allosteric sites rather than as orthosteric agonists or antagonists binding at the site used by the endogenous ligand. We reasoned that commercially available phage display technology could be used to identify specific allosteric modulators acting to either enhance or inhibit GlyR function, i.e. compounds that left-or right-shift glycine concentration-response curves while having no effects on their own. The peptides we identified acted as allosteric agonists or inverse agonists because some enhanced the effects of glycine (e.g. D12-116), whereas others produced inhibition (e.g. D12-106).
Peptides have several characteristics that make them excellent candidates as experimental and therapeutic agents. Along with the potential for high target specificity, the small sizes of peptides allow for greater tissue penetration. Peptides have lower manufacturing costs than most proteins and anti- bodies and show reduced interactions with the immune system (30). However, peptides have traditionally been discounted as viable drugs due to their low bioavailabilities and general inability to cross the blood-brain barrier (31). In recent years, this view has begun to change, as new technologies have arisen to overcome these limitations. Cell-penetrating peptides have long been used to deliver drugs into cellular cytoplasm, but recently, this technology has also been used to target bioactive peptides to brain tissue (32). Nanoparticles, a more recently developed technology, have also been highly successful in delivering systemically administered peptides into brain tissue at pharmacologically relevant concentrations (33). Efforts are also being made to improve the oral delivery of peptides to increase their marketability as therapeutic agents (34).
In our studies, we modified common phage display techniques to identify allosteric modulators of the GlyR. Our approach utilizes the cellular expression of the target protein rather than scaffolding of a purified target. This allows for selection in the native environment of the target and eliminates concerns about misfolding of the target protein. As a result, our peptides have the opportunity to bind only in the extracellular region of the target. Because none of the peptides tested thus far acted as agonists when applied alone, peptides may not be able to bind at the intersubunit orthosteric glycine-binding site (1) to directly activate the GlyR. In the positive selection portion of the phage display screen, we expressed the GlyR in HEK 293 cells in the absence of exogenously added glycine, so we assume that selection occurred against the GlyR in the closed channel state because these receptors do not activate spontaneously. To circumvent possible concerns about cell type-specific binding of our peptides, we used receptors expressed in HEK 293 cells for panning but switched to the GlyR expressed by Xenopus oocytes for the functional tests.
The wide diversity of peptides identified in our panning screen suggests that a more stringent negative control during phage panning might increase peptide specificity. For example, expressing one or both GABA 1 or ␣1␤2␥2S receptor subtypes in the negative selection portion of the panning procedure would be expected to remove those peptides that bind to both GABA and GlyRs, such as D12-106 and D12-124 ( Figs. 1 and 2). Another possibility would be to attempt to identify peptides that bind only to specific GlyR subunits by expressing the GlyR ␣2 and/or ␣3 subunits in the negative selection when searching for ␣1-specific peptides.
In summary, by pairing a high throughput peptide screen with a standard electrophysiological test, we identified and characterized novel specific peptide allosteric modulators of the GlyR. Other approaches to rapidly identify channel modulators have been developed (35); however, our method is easily transferable to other channel targets and has minimal set-up requirements. This approach could be modified for almost any target expressed in an adhering cell system and is particularly useful for channels or receptors that currently lack specific allosteric modulators.