Discovery of a Novel Allosteric Modulator of 5-HT3 Receptors

Background: The 5-HT3 receptors belong to the Cys-loop receptor superfamily. Results: The novel 5-HT3 antagonist PU02 (6-[(1-naphthylmethyl)thio]-9H-purine) is discovered, and its mechanism of action is delineated. Conclusion: PU02 is a potent and selective negative allosteric modulator of 5-HT3 receptors acting through a transmembrane intersubunit site in the receptors. Significance: The study highlights the transmembrane subunit interface in the Cys-loop receptor as a hot spot for allosteric modulation. The ligand-gated ion channels in the Cys-loop receptor superfamily mediate the effects of neurotransmitters acetylcholine, serotonin, GABA, and glycine. Cys-loop receptor signaling is susceptible to modulation by ligands acting through numerous allosteric sites. Here we report the discovery of a novel class of negative allosteric modulators of the 5-HT3 receptors (5-HT3Rs). PU02 (6-[(1-naphthylmethyl)thio]-9H-purine) is a potent and selective antagonist displaying IC50 values of ∼1 μm at 5-HT3Rs and substantially lower activities at other Cys-loop receptors. In an elaborate mutagenesis study of the 5-HT3A receptor guided by a homology model, PU02 is demonstrated to act through a transmembrane intersubunit site situated in the upper three helical turns of TM2 and TM3 in the (+)-subunit and TM1 and TM2 in the (−)-subunit. The Ser248, Leu288, Ile290, Thr294, and Gly306 residues are identified as important molecular determinants of PU02 activity with minor contributions from Ser292 and Val310, and we propose that the naphthalene group of PU02 docks into the hydrophobic cavity formed by these. Interestingly, specific mutations of Ser248, Thr294, and Gly306 convert PU02 into a complex modulator, potentiating and inhibiting 5-HT-evoked signaling through these mutants at low and high concentrations, respectively. The PU02 binding site in the 5-HT3R corresponds to allosteric sites in anionic Cys-loop receptors, which emphasizes the uniform nature of the molecular events underlying signaling through the receptors. Moreover, the dramatic changes in the functional properties of PU02 induced by subtle changes in its binding site bear witness to the delicate structural discrimination between allosteric inhibition and potentiation of Cys-loop receptors.

The Cys-loop receptor is composed of three structural entities: an extracellular domain (ECD) made up by the N-terminal domains, a transmembrane domain (TMD) composed of the four transmembrane ␣-helices TM1 to -4, and an intracellular domain formed by the second intracellular loops in the five subunits in the complex. Considerable insight into the threedimensional architecture of Cys-loop receptors has emerged from electron microscopy images of the Torpedo nAChR (7) and from high resolution x-ray structures of ACh-binding pro-teins (8) and of prokaryotic (9 -12) and invertebrate (13) Cysloop receptor orthologs. These structures and biochemical, biophysical and mutagenesis studies performed over the years have also shed light on the allosteric transitions underlying receptor signal transduction. The signaling is initiated by agonist binding to the orthosteric sites situated at subunit interfaces in the ECD, which subsequently triggers a number of allosteric events leading to the opening of the ion channel in the TMD (5, 14 -16).
The allosteric nature of the Cys-loop receptor complex implies that the signal elicited by orthosteric agonists can be modulated by ligands binding to other receptor regions (5, 14 -16). Numerous positive and negative allosteric modulators (PAMs and NAMs, respectively) of Cys-loop receptors have been identified, and new generations of allosteric modulators of GABA A Rs and nAChRs comprise several potent and selective drugs and highly interesting pharmacological tools (5, 6, 16 -18). In contrast, the allosteric modulators of 5-HT 3 Rs reported to date, including n-alcohols, volatile and intravenous anesthetics, and various antidepressants and antipsychotics, are all characterized by promiscuous pharmacological profiles and by displaying low potencies at the 5-HT 3 Rs (19).
In the present study, the pharmacological properties and mechanism of action of a novel 5-HT 3 R NAM, PU02, are characterized, and its transmembrane intersubunit binding site in the 5-HT 3 A receptor is identified and mapped. Interestingly, several mutations in this site convert PU02 into a complex modulator with a biphasic PAM/NAM profile, which underlines the intricate linkage between the site and channel gating in the 5-HT 3 R.

EXPERIMENTAL PROCEDURES
Materials-Culture media, serum, antibiotics, and buffers for cell culture were obtained from Invitrogen. The compound library and PU01 analogs were obtained from Chembridge Corp. (San Diego, CA). Glycine, GABA, ACh, serotonin, and probenecid were purchased from Sigma, and ondansetron, granisetron, tropisetron, genistein, and epibatidine were obtained from Tocris Cookson (Bristol, UK). The FLIPR TM Membrane Potential Blue (FMP) assay and Fluo-4/AM dyes were purchased from Molecular Devices (Crawley, UK) and Molecular Probes, Inc. (Eugene, OR), respectively. [ 3 H]GR65630 and Opti-Fluor TM were obtained from Perkin-Elmer Life Sciences. The cDNAs encoding for the human 5-HT3A and 5-HT3B subunits were kind gifts from Drs. J. Egebjerg and E. F. Kirkness, respectively, and cDNAs for the human ␣1, ␤2, and ␥2s GABA A R subunits were kindly provided by Dr. P. J. Whiting. The stable h5-HT 3 A-HEK293 cell line used for the FMP assay experiments was a kind gift from Dr. J. Egebjerg, and the stable h5-HT 3 A-and h5-HT 3 AB-HEK293 cell lines used in the Ca 2ϩ /Fluo-4 assay were kind gifts from Dr. C. Rojas (20). The stable cell lines expressing rat ␣3␤4, mouse ␣4␤2, and human ␣7 nAChRs were kind gifts from Drs. Y. Xiao and K. J. Kellar, J. A. Stitzel, and D. Feuerbach, respectively (21)(22)(23). The construction and pharmacological characterization of the stable HEK293 cell lines expressing the human ␣1 GlyR and human 1 GABA A R have been described previously (24,25).
FMP Assay-The screening of the compound library at the h5-HT 3 A-HEK293 cell line and the subsequent functional characterization of the compounds were performed in the FMP assay essentially as described previously (28). Stable h5-HT 3 A-, m␣4␤2-, r␣3␤4-, h␣1-, and h1-HEK293 cell lines or tsA201 cells transiently transfected with 5-HT 3 Rs or ␣1␤2␥2s GABA A R were split into poly-D-lysine-coated black 96-well plates with clear bottoms (BD Biosciences). 16 -24 h later, the medium was aspirated, and the cells were washed with 100 l of Krebs buffer (140 mM NaCl, 4.7 mM KCl, 2.5 mM CaCl 2 , 1.2 mM MgCl 2 , 11 mM HEPES, 10 mM D-glucose, pH 7.4). 50 l of Krebs buffer was added to the wells (in the antagonist experiments, various concentrations of the antagonist were dissolved in the buffer), and then an additional 50 l of Krebs buffer supplemented with the FMP assay dye (1 mg/ml) was added to each well. Then the plate was incubated at 37°C in a humidified 5% CO 2 incubator for 30 min and assayed in a NOVOstar TM plate reader (BMG Labtechnologies, Offenburg, Germany) measuring emission (in fluorescence units) at 560 nm caused by excitation at 530 nm before and up to 1 min after the addition of 33 l of agonist solution. The experiments were performed in duplicate at least three times for each compound at each WT and mutant receptor. EC 70 -EC 90 concentrations of serotonin, epibatidine, glycine, and GABA were used as agonist concentrations for the antagonist experiments at the respective 5-HT 3 Rs, nAChRs, GlyRs, and GABA A Rs. In the preincubation experiments (Fig. 4C), cells were incubated with Krebs buffer or with Krebs buffer supplemented with 30 M PU02 or 10 nM ondansetron for 30 min followed by three washes with Krebs buffer before execution of the assay.
Ca 2ϩ /Fluo-4 Assay-The functional characterization of compounds at the h␣7-GH3 cell line and at the h5-HT 3 A and h5-HT 3 AB cell lines (from Dr. C. Rojas) was performed in the Ca 2ϩ /Fluo-4 assay essentially as described previously (28). Cell lines were split into poly-D-lysine-coated black 96-well plates with clear bottoms. Following a 16 -24-h incubation in the case of the 5-HT 3 R cell lines or a 64 -72-h incubation of the h␣7-GH3 cells, the culture medium was aspirated, and the cells were incubated in 50 l of assay buffer (Hanks' buffered saline solution containing 20 mM HEPES, 1 mM CaCl 2 , 1 mM MgCl 2 , and 2.5 mM probenecid, pH 7.4) supplemented with 6 mM Fluo-4/AM at 37°C for 1 h. Then the buffer was aspirated, the cells were washed once with 100 l of assay buffer, and then 100 l of assay buffer was added to the cells. The assay buffer used for the h␣7-GH3 cells was supplemented with 100 M genistein. The 96-well plate was assayed in a NOVOstar TM microplate reader measuring emission (in fluorescence units) at 520 nm caused by excitation at 485 nm before and up to 60 s after the addition of 33 l of agonist solution in assay buffer. Serotonin was used as agonist for the h5-HT 3 A and h5-HT 3 AB receptors, whereas ACh was used for the ␣7 nAChR.
[ 3 H]GR65630 Binding-The [ 3 H]GR65630 competition binding assay was performed as described previously using membranes from stable h5-HT3A-HEK293 cells (29). Briefly, cells were harvested at 80 -90% confluence and scraped into the assay buffer (Hanks' buffered saline solution containing 20 mM HEPES, 1 mM CaCl 2 , 1 mM MgCl 2 , and 2.5 mM probenecid, pH 7.4), homogenized using a Polytron homogenizer for 10 s, and centrifuged for 20 min at 50,000 ϫ g. The resulting pellet was homogenized in 30 ml of assay buffer and centrifuged again. Then the cell pellet was resuspended in the assay buffer, and the membranes were incubated with 50 pM [ 3 H]GR65630 and various concentrations of the test compounds in a total reaction volume of 800 ml under gentle shaking for 1 h at room temperature. The incubation was terminated by a rapid filtration through Whatman GF/C filters (Whatman, Maidstone, UK) presoaked for 30 min in 0.3% (w/v) polyethyleneimine, followed by three washes with 3 ml of ice-cold isotonic saline solution. Filters were then transferred into vials containing 3 ml of Opti-Fluor scintillation solution, and radioactivity was measured in a liquid scintillation counter. The experiments were performed in duplicate a total of three times. The level of total binding was Ͻ10% of free radioligand for all samples.
Electrophysiology-COS-7 cells expressing the WT or mutant 5-HT 3 A receptors were recorded in the whole cell voltage clamp configuration (30) at room temperature. Cells were superfused with extracellular solution (140 mM NaCl, 11 mM glucose, 10 mM HEPES, 4.7 mM KCl, 0.1 mM CaCl 2 , adjusted to pH 7.4 with NaOH), patch pipettes had a resistance of ϳ2 megaohms when filled with intracellular solution (120 mM KCl, 1.8 mM MgCl 2 , 10 mM EGTA, 10 mM HEPES, adjusted to pH 7.4 with KOH), and recordings were performed as described previously (31).
All data were obtained with an EPC-9 amplifier (HEKA Electronics, Lambrect, Germany) run by a Windows XP personal computer. Experimental conditions and data acquisition were set and obtained using the PULSE software accompanying the amplifier. Data were low pass-filtered and sampled directly to the hard disk, and cells were held at Ϫ60 mV during recordings. Pipettes were pulled from borosilicate glass using a horizontal electrode puller (Zeitz Instrumente, Augsburg, Germany). The cells were rinsed with phosphate-buffered saline and detached from the culture flask by tripleX (0.1% (w/v)) digestion for 2 min at 37°C and seeded on the day of the experiment. Glass coverslips (3.5 mm) precoated with poly-D-lysine (0.005% (w/v)) were placed in Petri dishes, and cells were added at a suitable density.
Coverslips with cultured cells were transferred to a perfusion chamber mounted on the stage of an inverted microscope supplied with Nomarski optics, where transfected cells were identified by green fluorescence. Compounds were dissolved in extracellular solution and applied to the patched cell through a double-barreled application pipette. Application pipettes were fabricated from glass tubes (1.5 mm outer diameter; WPI, Sarasota, FL) and mounted on a piezoelectric device (PZS-100HS; Burleigh Instruments, Quebec, Canada) connected to a piezo-driver (PZ-150M, Burleigh Instruments) driven by TTL pulses from the EPC-9 amplifier. Approximately 1 min after the onset of the gravity flow, a PULSE protocol was initiated, and the current was recorded in 45-s intervals until peak amplitudes were stable. For transfected COS-7 cells, the duration of the recording periods was 6 s, during which the application pipette was switched to the agonist-containing test solution for 500 ms.
Data Analysis-All analysis and curve fitting were performed using Prism (version 5.0d; GraphPad Software, San Diego, CA). Concentration-response curves for agonists and concentration-inhibition curves for antagonists obtained in the FMP and Ca 2ϩ /Fluo-4 assays were constructed based on the difference in the fluorescence units between the maximal fluorescence recording made before and after the addition of agonist obtained for different concentrations of the respective ligands.
Concentration-response data for 5-HT were fitted to a sigmoidal curve with variable slope using nonlinear regression, where X represents the logarithm of the 5-HT concentration, Y is the response, and n H is the Hill slope.
Concentration-inhibition data for the antagonists were fitted to a sigmoidal curve with variable slope using nonlinear regression,

A Novel Allosteric Modulator of 5-HT 3 Receptors
where X represents the logarithm of the antagonist concentration, Y is the response, and n H is the Hill slope.
The EC 50 values for 5-HT and IC 50 values for the antagonists were derived from these equations. The maximal responses (R max values) elicited by 5-HT through the various mutants were derived from the fitted concentration-response curves, and the values were normalized to the R max value obtained for 5-HT at cells expressing the WT receptor on the same plate.
Statistical analysis of pIC 50 and pEC 50 values obtained in the mutagenesis study was performed for mean Ϯ S.E. values in GraphPad Instat (San Diego, CA) by one-way analysis of variance followed by Dunnett's multiple comparisons test (p Ͼ 0.05), using the data for the WT 5-HT 3 A receptor as control.
As for the electrophysiological data, concentration-response data for each cell were normalized to the maximum current for that cell. 5-HT-induced macroscopic currents were extracted from PULSE (HEKA, Lamprecht, Germany) and subsequently analyzed using IgorPro (Wavemetrics, Lake Oswego, OR) software. All electrophysiology data are based on recordings at more than four different cells on different days and transfections.
Determination of Time Constants for the Association and Desensitization Phases of 5-HT Currents-Relevant sections of the rising phase (for association) of selected traces were fitted to the single exponential equation, f(x) ϭ y max ϫ (1 Ϫ exp(Ϫx/)), and sections in the decay phase (for desensitization) of representative traces were fitted to the single exponential equation, Time constant values for WT and mutant receptors obtained in the absence or presence of PU02 were compared for significant difference using paired Student's t test statistical analysis in Prism (95% confidence interval). Time constant values for WT and mutant receptors were compared with each other using one-way analysis of variance in conjunction with Dunnett's multiple comparisons test in Prism.
Amino Acid Sequence Alignments-The coordinates and sequences of the Erwinia chrysanthemi pentameric ligandgated ion channel (ELIC) crystal structure, Protein Data Bank code 2VL0 (9), and the NMR structure of the transmembrane domain of human nAChR ␤2 subunit, Protein Data Bank code 2KSR (32), were downloaded from the Protein Data Bank (33). Initially, we aligned the sequences of ELIC and ␤2 using version 8.99 of T-Coffee (34). Next, we imported the first conformer of the ␤2 structure and the structure of ELIC into PyMOL (35) and overlaid them by superimposing the backbone atoms of those residues in TM1 and TM2, which were identical or highly similar in the T-Coffee alignment. This superimposition was used to anchor the alignment of TM3 on Ile 106 and Val 107 in ␤2 plus Ile 276 and Leu 277 in ELIC to obtain a structurally based alignment of TM1 to -3 in the two structures.
The sequences of human 5-HT3A, ␤2 nAChR, and ␣1 and ␤3 GABA A subunits were retrieved from the Protein Knowledgebase (36) (entries P46098, P17787, P14867, and P28472, respectively) along with the sequences of all other human nAChRs, 5-HT 3 Rs, GlyRs, and GABA A Rs. A sequence alignment of all human Cys-loop receptors was then created using version 8.99 of T-Coffee (34), and we extracted the 5-HT3A, ␤2 nAChR, and ␣1 and ␤3 GABA A R sequences of TM1 to -3, as defined by the helices in the ␤2 nAChR NMR structure. The ELIC sequence of TM1 to -3 was added to this alignment by use of the ␤2 nAChR sequence and the structurally derived alignment described above. The resulting alignment is in agreement with a very recently published alignment including some of the same sequences (13) but which is not based on the ␤2 nAChR NMR structure.
Homology Modeling-A homology model of a dimer interface consisting of TM1 to -3 in h5-HT 3 A was constructed using version 9.9 of Modeller (37) with TM1 to -3 from chain A and B of the ELIC crystal structure as a template and the amino acid sequence alignment of 5-HT3A and ELIC described above. Nand C-terminal ends of the helices were capped with the ACE and CT3 capping groups, respectively, and helical constraints were added to those parts of 5-HT3A where the ␤2 nAChR NMR structure shows a helical structure, whereas ELIC does not: the C-terminal end of TM1 (residues 261-265) and the Nand C-terminal ends of TM3 (residues 303-307 and 325-329). A total of 100 models were constructed using the fastest optimization protocol, and the best scoring model was selected based on the built-in DOPE (discrete optimized protein energy) homology model scoring function (38).

Discovery of a Novel Class of 5-HT 3 R Antagonists-A screen-
ing of a commercial library consisting of 1,680 compounds at a human 5-HT 3 A-HEK293 cell line using the FMP assay identified PU01 (2-amino-6-[(1-naphthylmethyl)thio]-9H-purine) as an antagonist of the receptor (Fig. 1). The structure-activity relationship of this scaffold was elucidated by functional characterization of PU01 and 27 analogs at the receptor. Eight compounds in the series antagonized 5-HT 3 A receptor signaling in a concentration-dependent manner (IC 50 values of 0.49 -12 M), whereas the rest of the analogs were inactive or displayed negligible antagonist activity at the receptor (Fig. 1). The PU02 analog (6-[(1-naphthylmethyl)thio]-9H-purine) ( Fig. 2A) was selected for the subsequent studies.
Electrophysiological Characterization of PU02-The functional characteristics of PU02 at 5-HT 3 A expressed in COS-7 cells were determined by the whole cell patch clamp technique. In agreement with previous studies (26,39), 5-HT elicited currents through the 5-HT 3 A receptor with an EC 50 (Fig. 3A), in agreement with its IC 50 values in the FMP and Ca 2ϩ /Fluo-4 assays ( Fig. 2A). As reflected in the current traces (Fig. 3B), the values obtained for the association and desensitization phases of the 5-HT-mediated response through 5-HT 3 A in the absence or presence of 0.3 M PU02 did not differ significantly (Fig. 3C).
Mechanism of Action of PU02-In contrast to the competitive 5-HT 3 R antagonist tropisetron, PU02 did not inhibit the binding of the orthosteric radioligand [ 3 H]GR65630 to the 5-HT 3 A receptor (IC 50 Ͼ100 M; Fig. 4A). In the FMP assay, the concentration-dependent nature of the PU02 antagonism was demonstrated by determination of 5-HT concentration curves at 5-HT 3 A in the presence of various concentrations of PU02 (Fig. 4B). The impact of pre-exposure of the 5-HT 3 A receptor to 10 nM ondansetron or 30 M PU02 (ϳ30-fold higher con-centrations than the IC 50 values) followed by extensive washing on receptor function was also investigated in this assay (Fig. 4C). Whereas preincubation with ondansetron did not change 5-HT potency markedly, pre-exposure to PU02 gave rise to a ϳ10-fold increased 5-HT EC 50 value, which may be indicative of a slow off-rate of the NAM from the receptor (Fig. 4C).
Identification of the PU02 Binding Site and Construction of a 5-HT 3 A TMD Homology Model-The PU02 binding site in the 5-HT 3 A receptor was identified through functional characterization of 5-HT, ondansetron, and PU02 at WT and 46 mutant receptors transiently expressed in tsA201 cells in the FMP assay (Figs. 5-7 and Table 2). Originally, the search was based on the hypothesis that PU02 could act through a cavity in the TMD, which is a common allosteric site in the Cys-loop receptor family and is the site of action for general anesthetics in GlyRs and GABA A Rs (16). Etomidate, a GABA A R PAM, also acts via an intersubunit site in the ␣1␤3 GABA A R TMD through interactions with Met 263 in TM1 of ␣1 and Asn 290 and Met 311 in TM2 and TM3 of ␤3, respectively (Fig. 8) (40,41). Thus, the first exploratory round of 5-HT 3 A mutants included Ala mutations of the corresponding Phe 255 , Leu 288 , and Val 310 residues (Fig.  8). While transient expression of the F255A and L288A 5-HT 3 A mutants in tsA201 cells did not result in the formation of functional receptors in the FMP assay, 5-HT and ondansetron exhibited EC 50 and IC 50 values at the V310A mutant similar to those at the WT receptor. In contrast, PU02 displayed a 5-fold higher IC 50 value at the mutant than at the WT 5-HT 3 A receptor ( Fig. 5 and Table 2).
The modestly impaired PU02 activity arising from the V310A mutation prompted us to explore the putative contributions of Val 310 and residues in its proximity to PU02 binding. To guide these explorations, a homology model of the human 5-HT 3 A TMD was constructed. Because the assumed binding site area is located in the subunit interface of the TMD, involving residues from the TM1, TM2, and TM3 helices (16,40,41), our homology model comprises a dimer of TM1 to -3, exclud-ing TM4. As a NAM, PU02 is expected to bind to an inactive receptor state, and thus our homology model was constructed based on the crystal structure of the ELIC Cys-loop receptor ortholog from E. chrysanthemi (9), the only high resolution x-ray crystal structure of Cys-loop receptor with a closed ion channel available at the time. The Torpedo nAChR electron microscopy structure was not regarded as a suitable template due to its low resolution and because it is questionable whether it corresponds to a closed ion channel state (7). Despite ELIC being of prokaryotic origin and showing low amino acid sequence identity to eukaryotic Cys-loop receptors, it is clearly structurally homologous to the Torpedo nAChR and other eukaryotic Cys-loop receptors (42), making it a suitable template. To overcome the low sequence identity, a recent NMR structure of the ␤2 nAChR (32) was used as a structural link to establish an unequivocal alignment of the amino acid sequence of 5-HT3A to that of ELIC.
In the model, Val 310 is situated in the third helical turn from the top of TM3, where it lines a predominantly hydrophobic cavity in the subunit interface made up by TM2 and TM3 in the (ϩ)-subunit and TM1 and TM2 in the (Ϫ)-subunit (Fig. 9, A  and B). The candidate residues selected for mutagenesis were predominantly those predicted to shape the observed cavity, but residues in the close vicinity of these were also mutated. The residues were initially substituted with amino acids expected to significantly alter their respective interactions with PU02. Residues identified as important for PU02 activity were subsequently subjected to additional mutations to probe the nature of and the spatial requirements for the putative interactions.
Mapping of the PU02 Binding Site-No significant response was observed in tsA201 cells transfected with 18 of the 46 5-HT 3 A mutants in the FMP assay (5-HT EC 50 Ͼ100 M; Table  2). The reasons for these lacks of responses were not investigated further. Although the maximal responses elicited by 5-HT through some of the 28 functional mutants differed substantially from that in WT 5-HT 3 A-expressing cells, the potencies displayed by 5-HT at the vast majority of the mutants did not, as only I290N and L288R displayed EC 50 values more than

TABLE 1 Functional characteristics of PU02 at 5-HT 3 Rs and other Cys-loop receptors in the FMP assay or in the Ca 2؉ /Fluo-4 assay
The characterization of PU02 was performed at stable cell lines expressing the receptors or at receptors transiently expressed in tsA201 cell lines. The prefixes "h", "m", and "r" are used for human, mouse, and rat receptors, respectively. The IC 50 values for PU02 were determined using EC 70 -90 concentrations of agonists for the respective receptors (determined on the day of the experiment) and are given in M with pIC 50  3-fold different from that of WT 5-HT 3 A (Fig. 5A and Table 2 5B and Table 2). Statistical analysis found 5-HT pEC 50 and ondansetron pIC 50 values displayed by some of the mutants to be significantly different from those at the WT receptor (Table 2). However, these differences are not considered pertinent from a biological perspective. As for PU02, mutations of Phe 244 , Phe 255 , Met 257 , Phe 287 , Thr 302 , and Met 313 in 5-HT 3 A did not alter its antagonistic activity markedly. However, mutations of Ser 292 and Val 310 in the receptor gave rise to subtle increases in PU02 IC 50 , and mutations of Ser 248 , Leu 288 , Ile 290 , Thr 294 , and Gly 306 dramatically altered the functional properties of the antagonist.
Mutation-induced Conversion of PU02 into a Complex Allosteric Modulator-Highly interestingly, PU02 exhibited seemingly biphasic concentration-effect curves at the S248V, T294A, T294L, T294Q, and G306A 5-HT 3 A mutants in the FMP assay, as the responses elicited by EC 80 5-HT were potentiated by low and inhibited by high PU02 concentrations (Fig. 6). This dual action style of modulation of these mutants was further investigated by determination of full 5-HT concentration curves in the presence of five PU02 concentrations in the FMP assay (Fig.  7). The results turned out to be quite complex, as can be seen for the S248V mutant. Whereas low PU02 concentrations leftshifted the 5-HT concentration-response relationship at the mutant, high PU02 concentrations caused a right shift (Fig. 7). Notably, at all concentrations of PU02, the maximal response elicited by 5-HT through the S248V mutant was significantly increased (Fig. 7). The profiles displayed by PU02 at the four other mutants were of similar character (Fig. 7).
Electrophysiological Characterization of PU02 at Selected 5-HT 3 A Receptor Mutants-The detrimental impairment of PU02 activity brought on by introduction of the three mutations S248N, G306I, and L288M/G306S in 5-HT 3 A was verified in electrophysiological recordings at the mutants expressed in COS-7 cells (Fig. 10). Since the 5-HT EC 50 values at these mutants are similar to that at WT 5-HT 3 A, 20 M 5-HT was used as the agonist concentration for these recordings, just as in the electrophysiological characterization of PU02 at the WT receptor (Fig. 3). The ability of PU02 to inhibit 5-HT-evoked currents through all of these three mutants was dramatically impaired compared with WT 5-HT 3 A (Fig. 10, A and B). In contrast, 0.6 nM ondansetron antagonized the 5-HT response through the WT receptor and all three mutants (Figs. 3 and  10A). The association and desensitization time constants for  the mutants did not differ significantly from those of the WT receptor, except for mutant S248N that displayed an increased desensitization time constant (Fig. 10C). The presence of 3 M or 10 M PU02 during the recordings did not significantly affect any of these parameters (data not shown).
Suggested Binding Mode of PU02 to the 5-HT 3 A Receptor-Because the 5-HT 3 A TMD homology model could potentially only include a part of the PU02 binding site, it is not suitable for binding mode prediction by automated molecular docking. However, to insert the experimental results in a structural context, PU02 was manually placed in the hydrophobic cavity in the model (Fig. 9, C and D), taking all mutagenesis data and the shape of the cavity into consideration. In the model, the seven residues found to be important for PU02 activity are located within the first three upper helix turns in the TMD, where they define a cavity in the subunit interface, the (ϩ)subunit contributing with Leu 288 and Ser 292 (TM2) and Gly 306 and Val 310 (TM3) and the (Ϫ)-subunit contributing with Ser 248 (TM1) and Ile 290 and Thr 294 (TM2) (Fig. 9). The Leu 288 and Ile 290 residues shape the binding cavity and comprise a major part of the total van der Waals interactions with PU02. The 13and 20-fold increased PU02 IC 50 values arising from the conservative L288M and I290A mutations, respectively, are in line with loss of van der Waals contacts. Furthermore, the introduction of hydrophilic side chains into the hydrophobic cavity in L288R and I290N would be expected to impair PU02 binding substantially, most likely by repelling the naphthalene moiety of the ligand. The main contribution of Ser 248 and Gly 306 to PU02 binding is to define the necessary size of the cavity. Thus, whereas the lack of effect of the S248A mutation shows that Ser 248 does not form a hydrogen bond to PU02, most substitutions of these two residues introduce a steric clash with the ligand. This is consistent with the detrimental effects on PU02 activity of the larger side chains introduced in S248N, S248Q,  50 values for ondansetron (B) and PU02 (C) at WT and functional mutant 5-HT 3 A receptors transiently expressed in tsA201 cells in the FMP assay. The bars for mutants with mutations in TM1, TM2, or TM3 are shown in green, cyan, and brown, respectively, and the bars for the three double mutants are shown in red. The experiments were performed as described under "Experimental Procedures" using EC 80 5-HT concentrations, and detailed data are given in Table  2. #, IC 50 Ͼ 100 M. ࡗ, complex modulation of PU02. The bar represents the EC 50 for the PAM activity of PU02 at the mutant.   Table 2). As for Thr 294 , its hydroxyl group projects into the cavity, where it may form a hydrogen bond with the 7-nitrogen of the purine moiety in PU02 (Fig. 9C). The complex profile exhibited by PU02 at mutants with Ala, Leu, or Gln residues in this position certainly underlines its importance for the modulation exerted by PU02 (Fig. 6). Finally, the subtle effects of Ser 292 and Val 310 mutations on PU02 activity are consistent with both residues being located in the periphery of the cavity, one helix turn above Leu 288 and below Gly 306 , respectively (Fig. 9C).
The mutations introduced in 5-HT 3 A having no impact on PU02 activity can also be rationalized by the homology model, as the residues either are situated in the periphery of the binding site or seem able to adopt side chain conformations pointing away from the ligand (Fig. 9B). Interestingly, the Phe 309 residue is predicted to project directly into the cavity, where it, together with Leu 288 , defines the bottom of the binding site (Fig. 9B). Although it could not be verified experimentally because substitutions of Phe 309 with several non-aromatic residues yielded non-functional receptors, we propose that Phe 309 may form interactions with the naphthalene moiety of PU02 and thus could be another important molecular determinant of PU02 binding.

DISCUSSION
PU02 is a novel 5-HT 3 R antagonist displaying functional IC 50 values around 1 M at human 5-HT 3 Rs, when measured in fluorescence-based assays or by patch clamp electrophysiology and substantially lower potencies at other Cys-loop receptors ( Figs. 2A and 3D and Table 1). The structure of PU02 differs notably from those of orthosteric 5-HT 3 R ligands, and thus it is hardly surprising that it acts as a NAM, as witnessed by its inability to compete with [ 3 H]GR65630 binding and the complete elimination by 20 M PU02 of responses evoked by 5-HT concentrations as high as 100 M (Fig. 4, A and B). Furthermore, the similar values for the association and desensitization of 5-HT-mediated currents through 5-HT 3 A obtained in the absence and presence of PU02 suggest that the NAM exerts its effects by affecting channel gating efficiency rather than modulating desensitization properties of the receptor (Fig. 3C).
Given the structure of PU02, we found it likely that it would target the TMD of the 5-HT 3 R, analogously to what has been reported for other Cys-loop receptor modulators (12,40,41,(43)(44)(45). The insights into etomidate binding to GABA A Rs (40,41) formed the basis for an elaborate mutagenesis study, in which Ser 248 , Leu 288 , Ile 290 , Thr 294 , and Gly 306 were identified as important molecular determinants of the PU02 activity at 5-HT 3 A, and Ser 292 and Val 310 were identified as minor contributors (Figs. 5 and 9, C and D). The fact that all mutants containing substitutions of these residues displayed 5-HT EC 50 and ondansetron IC 50 values similar to those at the WT receptor demonstrate that basic receptor function has not been compromised, and thus the observed effect of a given mutation on PU02 potency is largely attributable to its impact on the receptor-NAM interaction. Two potential explanations for the effects of these mutations on PU02 activity come to mind: (i) the mutations could allosterically alter the structural architecture of a distant PU02 binding site, or (ii) the mutated residues could contribute directly to PU02 binding. The substantially decreased inhibitory potencies displayed by PU02 at several of these mutants strongly suggest that the introduced mutations directly impair or disrupt PU02 binding to the 5-HT 3 A receptor. To our knowledge, cases of mutations allosterically affecting the activity of a ligand acting through a distant site to the degree observed here without also having marked effects on basic receptor function have not been reported. Moreover, in view of the large spectrum of inhibitory potencies and the overall similar maximal NAM efficacies displayed by PU02 at these mutants, it seems highly unlikely that these residues only are "efficacy modulators" of the PU02 activity at the receptor. For instance, the almost complete elimination of PU02 activity arising from single mutations of Ser 248 , Leu 288 , Ile 290 , and Gly 306 cannot all be ascribed to allosterically mediated changes in a distant binding site. Hence, we conclude that the identified residues line the hydrophobic cavity in the 5-HT 3 A TMD and directly participate in PU02 binding.
As presented under "Results," our suggested binding mode of PU02 to the 5-HT 3 A receptor is highly consistent with the mutagenesis data. The predominantly hydrophobic residues making up the intersubunit cavity match the hydrophobic naphthalene moiety of PU02 proposed to bind here. The thioether spacer enables the purine group of PU02 to protrude out of the TMD and form a hydrogen bond to Thr 294 (Fig. 9, C and D). Considering its location in the extracellular space, the purine group most likely forms additional bonds with residues in the TM2-TM3 linker and/or the ECD loops L2, L7, and L9 known to be in proximity to the TMD during signal transduction (Fig. 9, C and D) (5,14,15). The suggested binding mode of PU02 is supported by observations from the structure-activity relationship study, where the functional properties of analogs PU03, PU08, and TH07 identify the presence of both an aromatic group and a heteroaromatic ring separated by a spacer as important pharmacophore elements of PU02 (Fig. 1).
The seven "PU02-binding residues" identified in 5-HT 3 A are by no means conserved in other Cys-loop receptors, which would explain the inactivity of the NAM at the majority of receptors in this study (Table 1). We will refrain from explaining the activity of PU02 at ␣3␤4 and ␣7 nAChRs, but several other ligands exhibit dual ␣7 and 5-HT 3 activity (4). The seven residues are not even particularly conserved in the other 5-HT 3 R subunits, which is interesting considering the equipo-tent antagonism of PU02 at 5-HT 3 A, 5-HT 3 AB, 5-HT 3 AC, and 5-HT 3 AE receptors ( Fig. 2 and Table 1). Since some of these residues in 5-HT3B actually correspond to mutations of 5-HT3A residues shown to be detrimental to PU02 activity (e.g. L288R and I290N) (Fig. 8), the NAM may not be able to bind to Bϩ/BϪ, Aϩ/BϪ, or BϪ/Aϩ interfaces. Although the subunit arrangement in 5-HT 3 AB has been proposed to be B-B-A-B-A (46), recent studies strongly suggest the presence of at least one Aϩ/AϪ interface in the receptor (47,48), and thus PU02 could be envisioned to inhibit heteromeric 5-HT 3 R signaling through binding to this TMD interface. In support of this, PU02 exhibits

-HT, ondansetron, and PU02 at WT and mutant 5-HT 3 A receptors transiently expressed in tsA201 cells in the FMP assay
The positions of the mutations in the respective mutants are indicated, the terms (ϩ) and (Ϫ) referring to the two subunits forming the intersubunit binding site for PU02. The EC 50 50 values for ondansetron and PU02 were analyzed by one-way analysis of variance followed by Dunnett's multiple comparisons test, where data from mutants were compared with WT data. Asterisks indicate the significance of difference between compared means (*, p Ͻ 0.05; **, p Ͻ 0.01).

Receptor
Position of mutation  negligible inhibitory activity at the 5-HT 3 Rs formed in tsA201 cells co-transfected with 5-HT3A-G306I and WT 5-HT3B cDNAs or with 5-HT3A-S248Q and WT 5-HT3B cDNAs in the FMP assay (IC 50 ϳ30 M and ϳ100 M, respectively), compared with its IC 50 of 0.43 M at WT 5-HT 3 AB-expressing cells (Table 1) (data not shown). Our search for the PU02 binding site took its inspiration from that of etomidate (40,41), and the outcome obviously seems to validate this approach because ␤3-Asn 290 and ␤3-Met 311 in the GABA A R (40,41) correspond to Leu 288 and Val 310 in the 5-HT 3 A receptor. However, Val 310 is located in the periphery of the PU02 site, and ␣1-Met 263 , another important determinant of etomidate binding to the GABA A R, aligns with Phe 255 in 5-HT3A, which is not involved in PU02 binding (Figs. 5 and 9B). Thus, although the binding sites for etomidate and PU02 overlap, PU02 seems to target a site in 5-HT 3 A positioned closer to the extracellular space than etomidate in GABA A Rs. PU02 shares the TMD subunit interface as its site of interaction not only with etomidate but with several other modulators of anionic Cys-loop receptors. Leu 288 and Val 310 in 5-HT 3 A correspond to the Ser 267 and Ala 288 residues in ␣1 GlyR shown to be crucial for the actions of ethanol and general anesthetics (43), and neurosteroids mediate their direct activation of GABA A Rs via binding to an intersubunit site as well (44). Most recently, a crystal structure of the invertebrate glutamate-gated chloride channel GluCl in complex with ivermectin has found this promiscuous allosteric Cys-loop receptor modulator (49,50) to target this interface as well (13). Ivermectin interacts with residues in (Ϫ)-TM1, (ϩ)-TM2, (ϩ)-TM3, and the TM2-TM3 linker in GluCl, several of which correspond to "PU02-binding residues" in 5-HT 3 A (13). Ivermectin does not protrude as deep into the subunit interface as PU02, however, and thus does not interact with residues in the (Ϫ)-TM2 helix (13).
The complex modulation pattern exhibited by PU02 at five 5-HT 3 A mutants in this study (Figs. 6 and 7) appears to be reminiscent of the profiles of anesthetics at GABA A Rs and GlyRs (51,52) and of Zn 2ϩ at nAChRs and GlyRs (53)(54)(55)(56). Biphasic modulation can arise from the existence of distinct high and low affinity sites mediating the potentiation and inhibition of the modulator, respectively, or from the induction of faster receptor desensitization kinetics at higher modulator concentrations (51,54,56). We speculate that binding of PU02 to one or a few sites in the mutant receptor could bring about the potentiation observed at low concentrations, whereas occupation of additional sites in the pentamer at higher concentrations could cause the right shift of the 5-HT concentrationresponse relationship. This is supported by the fact that estimated EC 50 values for the potentiation exerted by PU02 at these mutants are similar to or lower than its IC 50 at WT 5-HT 3 A (Figs. 5C and 6). The Thr 294 residue appears to be a crucial determinant of the nature of PU02 modulation of 5-HT 3 A because all three mutations of this residue in this study induce the complex modulation pattern (Figs. 6 and 7). More gradual transitions of the PU02 effects are observed for Ser 248 and Gly 306 , where the NAM activity at receptors with small residues in these positions (Ser/Ala 248 and Gly 306 ) is converted into complex modulation of mutants with intermediate size residues (Val 248 and Ala 306 ) before PU02 becomes inactive at mutants with bigger residues (Asn/Gln 248 and Ile/Ser 306 ) (Fig.  6). Thus, while PU02 binding affinity may be relatively unaffected by these five mutations, the binding conformation of PU02 in the cavity and/or translation of its binding into effects on channel gating appear to have changed, in agreement with the conclusion that PU02 binding allosterically affects channel activity (Fig. 3C).
The recent crystal structures of Cys-loop receptor orthologs have not only confirmed the presence of an intrasubunit and an intersubunit TMD cavity for each of the subunits in the pentameric complex; they have also underlined the tight connection of these to one another via a pore-facing tunnel (9 -13). Thus, ligand binding to either of these cavities (12,13) or the linking tunnel (57) is likely to induce changes in the entire cavity-tunnel system behind the ion pore. The impact on receptor function of ligand binding to these cavities is often acutely sensitive to small changes, be it in receptor or ligand. For example, the NAM and PAM properties of general anesthetics on nAChRs and GABA A Rs/GlyRs, respectively, have been proposed to arise from different shapes and sizes of the intra-and/or intersubunit cavities in the receptors (12). Furthermore, mutations of Met 253 and Ser 276 in ␣7 nAChR, corresponding to Leu 288 and Val 310 in 5-HT 3 A, convert the PAM ivermectin into a NAM (58). This supports the notion that the intersubunit TMD interface not only accommodates the binding of both PAMs and NAMs but that the functional manifestations of ligand binding here are highly fine tuned. where at least one of the introduced mutations had a significant effect on PU02 activity. Cyan, residues in 5-HT3A where mutations had a weak effect on PU02 activity. Red, residues in 5-HT3A where none of the introduced mutations had an effect on PU02 activity. Gray, residues in 5-HT3A where all introduced mutations resulted in "non-functional" 5-HT 3 A receptors. Underlined, residues in 5-HT3A where introduced mutations were combined into double mutants. Boxed, residues in ␣1 and ␤3 GABA A R subunits reported to be important for etomidate activity.
In conclusion, PU02 is the most potent and selective allosteric modulator of 5-HT 3 Rs reported to date, and to our knowledge, the mapping of the intersubunit binding site for PU02 in the 5-HT 3 A receptor constitutes one of the most detailed delineations of the molecular basis for an allosteric modulator of a Cys-loop receptor. Moreover, it demonstrates that distantly related bacterial receptor orthologs can be used to construct homology models of these receptors that with high accuracy can predict the location of allosteric sites and the residues comprising these. Finally, the mutation-induced transformation of PU02 from a NAM into a more complex modulator at 5-HT 3 A bears witness to the subtle structural discrimination between allosteric inhibition and potentiation of this and other Cys-loop receptors and is suggestive of the potential of identifying 5-HT 3 R PAMs among PU02 analogs.