Abstract
Delta selective compound 2 (DS2; 4-chloro-N-[2-(2-thienyl)imidazo[1,2-a]pyridin-3-yl]benzamide) is one of the most widely used tools to study selective actions mediated by δ-subunit–containing GABAA receptors. DS2 was discovered over 10 years ago, but despite great efforts, the precise molecular site of action has remained elusive. Using a combination of computational modeling, site-directed mutagenesis, and cell-based pharmacological assays, we probed three potential binding sites for DS2 and analogs at α4β1δ receptors: an α4(+)δ(−) interface site in the extracellular domain (ECD), equivalent to the diazepam binding site in αβγ2 receptors, and two sites in the transmembrane domain (TMD) - one in the α4(+)β1(−) and one in the α4(−)β1(+) interface, with the α4(−)β1(+) site corresponding to the binding site for etomidate and a recently disclosed low-affinity binding site for diazepam. We show that mutations in the ECD site did not abrogate DS2 modulation. However, mutations in the TMD α4(+)β1(−) interface, either α4(S303L) of the α4(+) side or β1(I289Q) of the β1(−) side, convincingly disrupted the positive allosteric modulation by DS2. This was consistently demonstrated both in an assay measuring membrane potential changes and by whole-cell patch-clamp electrophysiology and rationalized by docking studies. Importantly, general sensitivity to modulators was not compromised in the mutated receptors. This study sheds important light on the long-sought molecular recognition site for DS2, refutes the misconception that the selectivity of DS2 for δ-containing receptors is caused by a direct interaction with the δ-subunit, and instead points toward a functional selectivity of DS2 and its analogs via a surprisingly well conserved binding pocket in the TMD.
SIGNIFICANCE STATEMENT δ-Containing GABAA receptors represent potential drug targets for the treatment of several neurological conditions with aberrant tonic inhibition, yet no drugs are currently in clinical use. With the identification of the molecular determinants responsible for positive modulation by the known compound delta selective compound 2, the ground is laid for design of ligands that selectively target δ-containing GABAA receptor subtypes, for better understanding of tonic inhibition, and ultimately, for rational development of novel drugs.
Introduction
Inhibition in the brain is primarily mediated by GABA acting through GABA receptors, with the ionotropic GABA type A receptors (GABAAR) being responsible for fast inhibition. Thus, GABAARs play an essential role in transmitting inhibitory signaling in the brain. Structurally speaking, GABAARs belong to the Cys-loop receptor family of pentameric receptor complexes and are composed from a repertoire of 19 different subunits in mammals, with the most commonly expressed in the central nervous system being α1-6, β1-3, γ1-3, and δ (Olsen and Sieghart, 2009). The subunit stoichiometry of the archetypical GABAA receptor is 2α, 2β, and a third subunit, most typically a γ- or a δ-subunit, but other stoichiometries have also been reported (Olsen and Sieghart, 2009). Studies on the subunit arrangement of the most abundantly expressed synaptic subtype, α1β2γ2, and a number of other γ-containing subtypes, show that the subunits are arranged as γ-β-α-β-α in a counterclockwise fashion around the central ion channel (Tretter et al., 1997; Baumann et al., 2002). Although it is generally accepted that the δ-subunit in its cognate receptors simply replaces the γ-subunit with respect to arrangement (Barrera et al., 2008), this is still not unequivocally established (Baur et al., 2009; Kaur et al., 2009; Wagoner and Czajkowski, 2010; Patel et al., 2014). Irrespectively, the orthosteric binding sites are located at the β(+)α(−) interfaces in the extracellular domain (ECD), and a number of allosteric binding sites have also been identified in the subunit interfaces of both the ECD and TMD (Olsen, 2018). These include, for example, the benzodiazepine site in the ECD α(+)γ(−) interface, responsible for mediating the anxiolytic and sleep-inducing effect of the benzodiazepines, including diazepam (Valium), widely used in the clinic (Sigel and Steinmann, 2012; Simeone et al., 2019).
The δ-containing GABAARs are located primarily at extrasynaptic sites where they mediate tonic (persistent) inhibition (Mody, 2001; Farrant and Nusser, 2005), hence controlling neuronal excitability (Belelli et al., 2009). Tonic inhibition is involved in various physiologic responses and pathophysiological conditions (Lee and Maguire, 2014), underlining a continued interest in targeting these receptors in conditions like insomnia (Wafford and Ebert, 2006), ischemic stroke (Clarkson et al., 2010; Lie et al., 2019), some forms of epilepsy (Cope et al., 2009), and peripheral immunomodulation (Yocum et al., 2017; Neumann et al., 2019). However, compared with the synaptic γ-containing receptors, pronounced insight into the physiologic and pathophysiological role of δ-containing receptors is still limited by the low number of potent and selective compounds.
One highly used model compound with selectivity for δ-containing receptors is the positive allosteric modulator (PAM) delta selective compound 2 (DS2; 4-chloro-N-[2-(2-thienyl)imidazo[1,2-a]pyridin-3-yl]benzamide) (Wafford et al., 2009). DS2 is extensively used as a tool compound to confirm the presence of δ-receptor–mediated tonic currents both in vitro and in vivo (Wongsamitkul et al., 2016; Falk-Petersen et al., 2017; Zhang et al., 2017; Dalby et al., 2020). DS2 was identified in a screening campaign and reported as a δ-selective PAM at α4β3δ GABAARs, showing no or limited effects at α4β3γ2 and α1β3γ2 receptors (Wafford et al., 2009). This selectivity was confirmed in thalamic relay neurons, where only extrasynaptic tonic currents were enhanced (Wafford et al., 2009), and using δ−/− mice (Jensen et al., 2013). DS2 displays limited brain permeability (Jensen et al., 2013) but was, nonetheless, shown to improve recovery after stroke in mice, plausibly by dampening peripheral immune activation (Neumann et al., 2019). Recently, a methoxy analog of DS2, termed DS2OMe, was identified and confirmed to have potential as a positron emission tomography tracer for visualization of δ-containing receptors in brains of larger mammals, such as pig (L’Estrade et al., 2019).
In 2018, the first cryogenic electron microscopy (cryo-EM) structure of a human GABAAR pentamer α1β2γ2 was published (Zhu et al., 2018). Afterward, a structure of the α1β3γ2 receptor was solved in complex with diazepam, revealing both the known high-affinity diazepam binding site in the α(+)γ(−) interface in the ECD and a novel low-affinity binding site located in the α(−)β(+) interface of the TMD (Masiulis et al., 2019).
Based on the notion that binding pockets evolved through nature are often highly conserved, combined with the structural similarities between DS2 and the benzodiazepine site ligand zolpidem (Rostrup et al., 2021), we hypothesized that similar pockets are present in δ-containing subtypes and that either of them could represent the long-sought-after DS2 site. We here report the identification of two residues, α4(S303) and β1(I289), within the predicted α4(+)β1(−) TMD interface of α4β1δ receptors as necessary for DS2 modulation. These findings are supported by docking of DS2 analogs into the identified binding pocket.
Materials and Methods
General
The study is exploratory by nature and follows the guidelines detailed in Michel et al. (2020). Data collection were in some cases defined by some preset standards, as detailed under each experimental section.
Chemicals and Materials
The compounds DS2, (4-chloro-N-[2-(2-thienyl)imidazo[1,2-a]pyridin-3-yl]benzamide); AA29504, ([2-amino-4-(2,4,6-trimethylbenzylamino)-phenyl]-carbamic acid ethyl ester); etomidate, ((R)-1-(1- phenylethyl)-1H-imidazole-5-carboxylic acid ethyl ester); picrotoxin; and GABA were obtained from Tocris Bioscience (Bristol, UK). DS2OMe (4-methoxy-N-[2-(thiopen-2-yl)imidazole[1,2-a]pyridine-3-yl]benbamide) was synthesized in-house as described previously (Yakoub et al., 2018). The purity test was done by High Performance Liquid Chromatography (HPLC), and the combustion analysis calculated for C19H15N3O2S was 350.09 and was found to be 350.09. Reverse-phase High Performance Liquid Chromatography (HPLC) is as follows: Retention time (MeCN/H2O, 1:1) = 6.75 minutes, purity > 99%. DMEM with GlutaMAX-I, FBS, penicillin-streptomycin, hygromycin B, trypsin-EDTA, Dulbecco’s phosphate-buffered saline, and Hank’s balanced salt solution were purchased from Life Technologies (Paisley, UK). DMSO, HEPES, MgCl2, CaCl2, polyd-lysine, and MgATP were purchased from Sigma-Aldrich (St. Louis, MO). The fluorometric imaging plate reader membrane potential (FMP) Blue dye was purchased from Molecular Devices (Crawley, UK), and Polyfect transfection reagent was from Qiagen (West Sussex, UK). Stocks of DS2 and DS2OMe were prepared at 1 mM and 10 mM concentrations in DMSO with final DMSO concentration < 0.1%. Because of the moderate solubility and 4× concentrations used in the FMP assay, the buffer was preheated to 37°C in a water bath before addition of compound and preparation of serial dilutions. Only stocks with final concentrations below 12 μM were used for further dilutions. Furthermore, higher concentrations were prepared separately.
Cells and Transfections
A human embryonic kidney (HEK) 293 Flp-In cell line stably expressing the human δ-GABAAR subunit (Falk-Petersen et al., 2017) was used for transfection, with human α- and β-subunits to express recombinant wild-type (WT) and mutant GABAARs, using transfection ratios optimized as described (Falk-Petersen et al., 2017). Cells were maintained in DMEM containing GlutaMAX-I supplemented with 10% FBS and 1% penicillin-streptomycin and kept in an incubator at 37°C and a humidity of 5% CO2. In total, 200 μl/ml hygromycin B was added to the media as positive selection. Transfection was performed using Polyfect (Qiagen) following the manufacturer’s instructions, except for using half the volume of transfection reagent for each transfection. The α- and β-subunits were cotransfected in a 1:1 ratio for FMP experiments and, for patch-clamp experiments, additionally cotransfected with GFP in a 0.5:1:1 ratio (0.8:1.6:1.6 μg in 6-cm culture dishes) to visualize transfected cells.
Plasmids and Mutant Constructs
The plasmids used for transfection to transiently express GABAA receptors have been described previously (Falk-Petersen et al., 2017). The WT human α4- and β1-subunits were subcloned into the pUNIV vector (Addgene, Cambridge, MA) and the human δ-subunit into the pcDNA5/FRT vector (Invitrogen, Paisley, UK) using the δ-construct described previously (Falk-Petersen et al., 2017). Plasmids carrying single and double mutations were generated and sequence-verified by GenScript (Piscataway, NJ). The numbering of the mutants refers to the sequences with the signal peptide included.
Generation of Stable Cell Lines
Mutations introduced into the δ-subunit were established as stable HEK293 Flp-In cell lines (Invitrogen), generating a stable cell line for each mutant. The stable cell lines were generated using the pcDNA/FRT/V5-His TOPO TA Expression kit (Invitrogen) performed according to the manufacturer’s protocol and as described previously (Falk-Petersen et al., 2017), except for using 25 μl Polyfect and 4 μg DNA for transfection in a 10-cm culture dish.
Fluorometric Imaging Plate Reader Membrane Potential Assay
The FMP assay was performed exactly as described previously (Falk-Petersen et al., 2017). In brief, 48 hours before the assay, cells were transfected. At 16–20 hours later, cells were plated into clear-bottomed poly(d-lysine)–coated black 96-well plates in a number of 50,000 cells per well, suspended in cell media, and placed in an incubator at 37°C with a humidity of 5% CO2 until performing the assay. At 44–48 hours post-transfection, the medium was removed, and cells were washed in assay buffer (100 μl/well) and incubated in 100 μl/well 0.5 mg/ml FMP Blue dye freshly dissolved in assay buffer (Hank’s balanced salt solution containing 20 mM HEPES adjusted to pH 7.4 and supplemented with 2 mM CaCl2 and 0.5 mM MgCl2) for 30 minutes and shielded from light in an incubator at 37°C and a humidity of 5% CO2. Ligand solutions were prepared in 4× assay buffer and added to a ligand plate, which was placed in a FLEXstation3 plate reader (Molecular Devices, Crawley, UK) that was preheated to 37°C for temperature equilibration for 10–15 minutes. After transferring the cell plate to the reader, the fluorescence was measured at baseline and after ligand addition by detecting emission at 560 nm caused by excitation at 530 nm.
FMP Experimental Design and Data Analysis
For FMP experiments, some preset formats were used for assay design and data analysis. Compound-induced signals were reported as changes in relative fluorescence units (ΔRFU), with the signal given as the difference between the average of the baseline signal (∼30-second recording) subtracted the peak response (or minimum response for decreases in baseline). All raw traces were manually inspected for obvious artifacts after compound addition. For high concentrations of DS2 (1–20 µM), we regularly observed negative RFU values below the buffer responses that in certain cases were excluded (see below). This phenomenon was independent of receptor subtype, as it was observed for both δ-HEK and mock cells. The phenomenon was less pronounced for DS2OMe, which is why this compound was preferred in some substudies. To circumvent this problem, we set up the following exclusion criteria: negative ΔRFU or decreased ΔRFU values for high concentration (>1 μM) of DS2 and DS2OMe compared with the ΔRFU for a lower concentration in the same experiment (indicative of precipitation). Additionally, curve fittings resulting in ambiguous EC50 values and R2 values lower than 0.80 due to very small responses were omitted from analyses, resulting in the following number of excluded experiments (excluded/total number) using either DS2 or DS2OMe at the given receptor subtypes: α4β1δ, 6/18; α4(F133A), 2/7; α4(F133L), 4/7; α4(R135A), 6/10; α4(R135H), 4/7; α4(G191A), 4/8; α4(G191E), 2/7; α4(G191L), 4/7; δ(E71L), 0/5; δ(F90A), 0/4; δ(H204A), 1/5; δ(S155Q), 1/5; δ(A73N), 1/5; α4(S303L), 0/3; α4(L302Y), 1/5; α4(L302Y,S303L), 2/5; β1(I289Q), 1/8; β1(S290F), 0/4. For mutants, experiments were generally performed in three to five independent experiments with technical triplicates, which was decided prior to execution based on the level of variation observed in previous work. For technical reasons, a few experiments had to be conducted at n = 6 to 7 (Supplemental Table 5; Table 1). WT data were performed in 8–11 independent experiments, as they served as controls across experiments.
Experimental data are shown in scatter plots with 95% confidence intervals with n values given in the figure legends. Curves were normalized to GABA to allow side-by-side representation and depicted as representative data (means ± S.D.). Mean EC50 values and pEC50 values with 95% confidence intervals are collected in tables along with statistical values. Concentration-response curves were fitted using nonlinear regression, with log-transformed concentrations as x-values, using the four-parameter concentration-response equation, to determine the EC50 value and Hill slope (nH). The “bottom” and ‘”top” denote the upper and lower nonconstrained plateau of the curve, respectively. The calculated EC50 values were log-transformed to obtain mean pEC50 values. Statistical analysis of mutated receptors was performed on the pEC50 values using the two-sided Welch’s t test compared with WT, correcting for multiple comparison using the original FDR method of Benjamini and Hochberg with a discovery rate of 0.05. Both adjusted and unadjusted P values are reported. Data analysis and statistics were performed in GraphPad Prism (version 8.4.3; GraphPad, San Diego, CA).
Whole-Cell Patch-Clamp Electrophysiology
Whole-cell patch-clamp experiments were performed on δ-HEK cells transiently coexpressing WT or mutant α- and β-subunits and GFP as described previously (Falk-Petersen et al., 2020). In short, the transfected cells were transferred to 35-mm Petri dishes (100,000–200,000 cells) the day prior to performing the experiment. On the day of experiment, cell media were exchanged for ABSS [containing the following (in mM): NaCl 140, KCl 3.5, Na2HPO4 1.25, MgSO4 2, CaCl2 2, glucose 10, and HEPES 10; pH 7.35] at room temperature (20–24°C) before placing at the stage of an Axiovert 10 microscope (Zeiss, Germany). Viewing the cells at 200× magnification and visualizing cells containing green fluorescent protein with UV light from an HBO 50 lamp (Zeiss, Germany), the cells were approached with micropipettes of 1.2–3.3 MΩ resistance manufactured from 1.5-mm OD glass (World Precision Instruments, Sarasota, FL) on a microelectrode puller, model PP-830 (Narishige, Tokyo, Japan). The micropipettes contained an intracellular solution composed of the following (in mM): KCl 140, MgCl2 1, CaCl2 1, EGTA 10, MgATP 2, and HEPES 10; pH 7.3.
Recordings were made from cells in the whole-cell configuration using the standard patch-clamp technique in voltage mode and an EPC-9 amplifier (HEKA, Lambrecht, Germany). The clamping potential was −60 mv, and series resistance was 80% compensated. Whole-cell currents were recorded using Pulse and PulseFit software (version 8.80; HEKA). Ligand solutions, prepared in ABSS, were applied using two VC3-8xP pressurized application systems feeding into a 16-barreled perfusion pipette (ALA Scientific Instruments Inc., Farmingdale, NY) ending approximately 100 μm from the recorded cell. PAMs were tested using coapplication with a concentration of GABA corresponding to GABA EC10–35 at the respective receptor subtype. Preapplication was not used for DS2 and DS2OMe, as results from preliminary experiments showed no difference in the size of the peak current with and without preapplication of the PAM. PAMs and GABA were coapplied for 10–30 seconds until the peak current was reached. Agonists were applied for 5 seconds. Between compound applications, compound-free ABSS was applied from one of the barrels to quickly remove the compounds from the cell, and cells were allowed to recover for 1 minute before the next ligand application.
Patch-Clamp Data Analysis and Statistics
As for FMP, some preset formats regarding assay design and data analysis were used. All currents were normalized to the maximum GABA current and given as %I/Imax. All currents are reported as normalized mean currents with 95% confidence interval. Based on previous experience, currents from at least five different cells from at least two transfections were used. In a few cases, up to 16 cells were used for technical reasons (see Supplemental Fig. 4). All n values are given in the figure legends. Data sets with GABA controls (0.1–0.5 μM) deviating from GABA EC10 to EC35 were excluded from the analysis.
Statistical analysis was applied to test whether the PAMs potentiated the GABA control response using two-sided Welch’s t test as for FMP data. Analysis of currents was performed using Pulse and PulseFit (HEKA), and current traces were visualized using IgorPro (version 6.2.2.2; Wavemetrics, Lake Oswego, OR). Collected data and statistical analysis were performed using GraphPad Prism (version 8.4.3).
Homology Model for the Extracellular Domain Binding Site
The homology model of the ECD α4(+)β1(−) interface has been described previously (Rostrup et al., 2021). The model was used to identify residues for the mutational study based on the docking of DS2 into the model described previously (Rostrup et al., 2021).
Homology Model for the Transmembrane Domain Binding Site
The homology model of the transmembrane part of the α4β1 interface was constructed with Modeler 9.24 (Webb and Sali, 2016) using the α1β3 interface from the α1β3γ2L crystal structure (Protein Data Bank code: 6HUP) (Masiulis et al., 2019) as template. Model and template sequences of the transmembrane helices and the connecting loops making up the subunit interface were obtained and aligned in UniProt (http://www.uniprot.org/) (UniProt Consortium, 2019): sequence IDsα4 P48169, β1 P18505, α1 P114867, and β3 P28472. To adhere as much as possible to the very closely related template structure, the “very fast” keyword was used to output the initial model that is only subjected to a brief optimization, thus retaining the copied coordinates for all conserved residue positions. This procedure was selected based on the high sequence similarities and assumed structural conservation combined with the fact that the binding site residues are optimized relative to the ligand in following computational steps.
Induced-Fit Docking of DS2 into the Transmembrane α4(+)β1(−) Site and In Silico Mutagenesis
The homology models were prepared for docking with the Protein Preparation Wizard [Schrödinger Release 2020-2; Schrödinger, LLC, New York, NY (Sastry et al., 2013)] using default settings. The chemical structure of DS2 was downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov) (Kim et al., 2019; chemical ID: 979718), and the analogs DS2OMe (L’Estrade et al., 2019) and Br-DS2OPh (Rostrup et al., 2021) were built from DS2 in MarvinSketch 20.15.0 (ChemAxon; http://www.chemaxon.com). All three ligands were prepared for docking with default settings in LigPrep (Schrödinger Release 2020-2) and used for induced-fit docking in the model of the transmembrane α4β1 interface with the standard protocol. The binding site center was defined by Ser303 and Ala324 from α4 plus Pro253 and Ile289 from β1. The ligand length was set to ≦14 Å, and XP precision was used in the redocking step, while all other settings were default. The best-scoring docking poses according to the IFD score were selected for each compound as the most likely binding mode. In silico mutagenesis was performed with the built-in protein mutagenesis wizard in PyMOL (The PyMOL Molecular Graphics System, version 2.0; Schrödinger, LLC.) and the backbone dependent rotamer library selecting the most probable rotamer with the fewest steric clashes with surrounding residues. For the α4S303L and β1I289Q mutations the first (one of four; 47.4%) and second (2 of 16, 14.6%) most likely rotamers were selected, respectively.
Results
To identify central residues for the activity of DS2 at δ-containing GABAARs, we systematically investigated three potential binding pockets: one in the ECD (the α4(+)δ(−) interface) and two in the TMD (the α4(+)β1(−) and the β1(+)α4(−) interfaces). Although δ-containing TMD interfaces are present in the receptor complex, we focused solely on the α-β interfaces because of the confirmed existence of binding sites at these interfaces (Ernst et al., 2005; Puthenkalam et al., 2016; Laverty et al., 2019) and our focus on benzodiazepine binding sites as potential binding sites for DS2 due to the structural resemblance between DS2 and zolpidem (Rostrup et al., 2021). Key interacting residues were identified using homology models and pharmacologically characterized in well established HEK cell–based assays using the α4β1δ receptor as a model receptor, which has been carefully characterized in our hands (Falk-Petersen et al., 2017, 2020; Dalby et al., 2020).
Investigation of the ECD α(+)δ(−) Interface as the Site of Modulation by DS2
First, using the homology model published in Rostrup et al. (2021), we studied the pocket located in the C-loop of the α(+)δ(−) interface in the ECD of α4β1δ receptors (Fig. 1A). From our previous docking into the model, we identified three potential key residues on the α4(+) side of the interface that could either interact directly with DS2 or were placed centrally within the binding pocket: α4(F133), α4(R135), and α4(G191) (Fig. 1A). Additionally, five residues on the complementary δ(−) interface were identified: δ(Ε71), δ(Α73), δ(F90), δ(S155), and δ(Η204). The selected residues were mutated with the principle of removing potential interactions (mutation into alanine) and/or gradually decreasing the space in the binding pocket (mutation into various amino acid residues), thus expecting a reduced modulation by DS2 compared with WT. This resulted in seven different α4-subunit mutants and five different δ-subunit mutants (Supplemental Tables 1 and 2). Each of the mutated subunits were expressed in HEK cell lines to form α4β1δ receptors and tested in the FMP assay as single mutants. Whereas α4-mutants were simply cotransfected with WT β1 into WT stable δ-HEK293 Flp-In cells, each of the δ-mutants were established as stable HEK293 Flp-In cell lines transfected with WT α4β1 to transiently express α4β1δ. In general, extending the utility of this expression system from WT to mutated δ-containing receptors is highly reliable and suitable for controlling expression and reliably studying these in-some-instances cumbersome receptor subtypes (Karim et al., 2012).
All seven α4-mutant receptors were found to express functionally active receptors and to respond to GABA with 2 to 3 times the potency observed for WT (Fig. 1, B and C; Supplemental Fig. 1; Supplemental Table 1). The expression levels of α4(F133Α/L)β1δ and α4(G191E)β1δ appeared lower as compared with WT, as the maximal ΔRFU values were consistently reduced in all experiments (Fig. 1B). To characterize the sensitivity to DS2, it was applied together with a GABA EC20 concentration, calculated for each mutant (Supplemental Table 2; Table 1). Among the seven different α4-mutants, α4(F133A)β1δ and α4(G191E)β1δ showed no apparent or only small modulation by DS2, whereas the potency of DS2 at the other α4-subunit mutants was either unchanged [α4(F133L)β1δ, α4(R135A), and α4(G191A/L)β1δ] or slightly increased [α4(R135H)β1δ] compared with WT (Fig. 1, D–F; Supplemental Fig. 1; Supplemental Table 2; Table 1). Interestingly, only a single of the introduced mutations at α4(F133) and α4(G191) showed changed responses, which could not readily be explained.
As we and others have previously observed methodological limitations in the FMP assay (Wafford et al., 2009; Falk-Petersen et al., 2017), we suspected that the apparent lack of modulation could be due to sensitivity limitations. Thus, to follow up, the two mutants, α4(F133A) and α4(G191E), were tested using whole-cell patch-clamp electrophysiology. At both mutated receptors, DS2 modulated the GABA EC20-induced currents in a concentration-dependent manner similar to WT or with even higher efficacy (Fig. 1, G and H). Each of the five δ-subunit mutations were also tested in the FMP assay. These were all functional and displayed unchanged responsiveness to DS2 compared with WT (Supplemental Fig. 2; Supplemental Tables 3 and 4).
Altogether, we conclude that the C-loop pocket in the ECD α4(+)δ(−) interface is not the site responsible for the PAM effect of DS2.
Identification of the TMD α4(+)β1(−) Interface as the Site of Modulation by DS2
Next, we looked into two pockets in the TMD αβ interfaces (specifically involving TM2) as potential recognition sites for DS2 based on the hypothesis that diazepam and DS2 exhibit analogous binding sites in the TMD. Mutations in α4β1δ TMD pockets were suggested based on the cryo-EM structure of the human GABAAR α1β3γ2L (Protein Data Bank code: 6HUP) in combination with a sequence alignment due to the high (>90%) local sequence identity of the subunits within the TMD region of interest. The first pocket is located in the β(+)α(−) interface in a site equivalent to the recently identified low-affinity binding site for diazepam (Laverty et al., 2019).
The mutations, β1(S290F) on the β1(+) side and α4(L302Y) on the α4(−) side, were initially probed because of an apparent central positioning of the residues in the pocket and orientation toward diazepam in the cryo-EM structure (Fig. 2A). As a similar pocket is present at the reverse subunit interface, we also included the corresponding mutations in the α4(+)β1(−) interface, α4(S303L) on the α4(+) side and β1(I289Q) on the β1(−) side. However, as this pocket appears noticeably smaller than the β1(+)α4(−) pocket, these were mutated into more flexible and less bulky residues. Additionally, to probe both proposed pockets simultaneously, we included the double-mutant receptors α4(L302Y,S303L)β1δ and α4β1(I289Q,S290F)δ. The introduced mutations were expected to revert hydrophilicity/hydrophobicity and introduce steric hindrance and thus would be anticipated to decrease or altogether abolish the effect of DS2.
First, we show that all the single-mutant receptors were GABA-responsive and thus functional in the FMP assay (Fig. 2B). The two β-mutants β1(S290F) and β1(I289Q) displayed 6.8 and 8.1 times increased GABA potencies, respectively, and the α4(S303L)β1δ mutant displayed 2.9 times increased potency compared with WT (Fig. 2C; Supplemental Table 5). Since the receptors were functional, we continued with the studies.
In the modulation experiments, we switched to DS2OMe, an analog of DS2 (L’Estrade et al., 2019) with the same pharmacological profile, because of both solubility issues with DS2 (described in the Materials and Methods section) and the general sensitivity limitations observed in the FMP assay on the ECD mutants. First, we examined the modulation of GABA EC20 at mutations introduced in the β(+)α(−) interface, equal to the low-affinity diazepam binding site in the γ-containing receptor (Fig. 2D). These mutations did not affect the modulation by DS2OMe, as both the α4β1(S290F)δ and α4(L302Y)β1δ mutant receptors had DS2OMe potencies similar to WT, although a small, significant increase in efficacy for the α4(L302Y)β1δ mutant compared with WT was observed (**P = 0.0063, two-tailed Welch’s t test, response of 3 μM DS2OMe) (Fig. 2, D and F; Table 2).
By contrast, when turning to the alternative α4(+)β1(−) interface, we observed significant decreases in responsiveness to modulation by DS2OMe. The α4(S303L)β1δ receptor lacked responsiveness to modulation by DS2OMe, and the β-mutant receptor α4β1(I289Q)δ had a statistically significant 3.2 times reduction of the potency of DS2OMe compared with the WT receptor (Fig. 2, E and F; Table 2). Additionally, as expected from the individual mutations, the double-mutant receptor α4(L302Y,S303L)β1δ was not modulated by DS2OMe (Supplemental Fig. 3; Supplemental Table 6).
To verify the FMP results, we performed whole-cell patch-clamp electrophysiology recordings. Convincingly, we found no or very limited DS2 modulation of the GABA currents in the α4(S303L)β1δ and α4β1(I289Q)δ receptor [only 10 μM modulated the α4β1(I289Q)δ receptor by significantly increasing the GABA control current to 54.8% of the GABA Imax] (**P = 0.0063, two-tailed Welch’s t test, adjusted, n = 5 to 6) (Fig. 2, G and H). Further, we included the double-mutant receptor α4(S303L)β1(I289Q)δ, which was even less modulated by 10 μM DS2, amounting to 44% of the GABA Imax (*P = 0.016, two-tailed Welch’s t test, adjusted, n =6–9) (Fig. 2, G and H; Supplemental Tables 3 and 6). DS2OMe showed no modulation of the GABA response in either α4β1(I289Q)δ or α4(S303L)β1δ receptors or the double-mutant α4(S303L)β1(I289Q)δ receptor (Supplemental Fig. 4).
Together, these results strongly advocate for the identified TMD α(+)β(−) interface site as the site responsible for the modulatory action of DS2.
Known GABAAR PAMs Show Unchanged Modulation at DS2-Insensitive Mutant Receptors
To confirm that the mutant receptors with altered DS2 sensitivity were not overall compromised in their general PAM responsiveness, we tested etomidate (Hill-Venning et al., 1997) and AA29504 (Hoestgaard-Jensen et al., 2010; Olander et al., 2018) at both WT and the single mutants α4(S303L)β1δ and α4β1(I289Q)δ. In the FMP assay, both compounds showed intact positive modulation of both mutants compared with WT. Potencies (pEC50) of etomidate were at WT α4β1δ determined to 5.11 (EC50 7.8 μM), and for the α4(S303L)β1δ and α4β1(I289Q)δ mutants to 5.21 and 5.03 (EC50 6.2 μM and 9.3 μM), respectively (Fig. 2, I and J; Table 3) (NS, two-tailed Welch’s t test, n =3 to 4). Further, AA29504 showed similar potentiation at the mutants and WT receptors (Supplemental Fig. 5; Supplemental Table 7), indicating that it does not mediate its effect through the same site as DS2, correlating with a proposed binding site for AA29504 in the TMD β(+)α(−) interface (Olander et al., 2018).
Induced-Fit Docking of DS2 and DS2OMe Corroborates Mutational Results
Guided by the mutational data confirming the molecular recognition site mediating the effect of DS2 and DS2OMe in the transmembrane part of the α4(+)β1(−)-subunit interface, we constructed a model of the modulator-receptor binding mode. Based on the structure of the desensitized α1β3γ2L receptor bound to GABA and diazepam (Masiulis et al., 2019), we constructed a homology model of the α4β1-subunit interface into which DS2 and DS2OMe were fitted using induced-fit docking. Allowing residue side chains in the “empty” homology model to adapt to the modulators, we obtained very similar binding modes for the docked compounds (Fig. 3; Supplemental Fig. 5). The core scaffold binds with the amide carbonyl of DS2 showing a potential hydrogen bond to the hydroxy group of Ser303 in α4. Using a backbone-dependent rotamer library, we observe that the α4(S303L) mutation removes the hydrogen bond and sterically blocks the binding site, providing an explanation for the observed lack of potentiation on this mutant (Fig. 3B). Ile289 in β1 lines 4-fluorophenyl of DS2, contributing to the binding through substantial van der Waal contacts, and the I289Q mutation has a steric clash with DS2 (Fig. 3B). As for the α4(S303L) mutant, these effects provide a possible explanation for the observed abolishment of potentiation at all but the highest concentration of DS2 and DS2OMe in our patch-clamp experiments, and the obtained binding mode thus concurs with the experimental results. Our previously published analogs of DS2 show that there should be room for much larger substituents than the methoxy of DS2OMe as well as a bromo atom in the 5-position on the imidazol[1,2-a]pyridine scaffold (Rostrup et al., 2021). Thus, we provide further proof of concept for the predicted binding site by docking the recently published analog, Br-DS2OPh (Rostrup et al., 2021). This confirmed that the OPh substituent can fit the binding site in the homology model with only a minor shift in the binding mode (Supplemental Fig. 5).
Discussion
From our experiments using systematic structural iterations and experimental validation of known and proposed binding sites, we here present the elusive DS2 interaction site as a distinct site encompassing α4S303 and β1I289 residues in the TMD α4(+)β1(−) interface of α4β1δ receptors. This novel site is similar in nature both to the low-affinity diazepam binding site identified in the cryo-EM structure of the α1β3γ2 receptor (Laverty et al., 2019) and the site for general anesthetics (e.g., etomidate) (Li et al., 2006) and the proposed binding site for AA29504 (Olander et al., 2018). Notably, the residues in the new DS2 site are located on the alternative intersubunit interface α4(+)β1(−), explaining why our mutations do not affect etomidate or AA29504 PAM activity. Indeed, it has been reported that pockets exist in all the TMD intersubunit interfaces (Sieghart et al., 2012; Forman and Miller, 2016; Iorio et al., 2020) and that several known allosteric modulators can bind in these pockets (Olsen, 2018). This shows how different GABAA receptors subtypes have evolved to include different functionally relevant allosteric sites.
The α4β1δ subtype was selected as model receptor in this study because of previous success with this for detailed and reliable molecular pharmacology examination (Falk-Petersen et al., 2017, 2020; Dalby et al., 2020). The expression of the δ-subunit relies on an in-house–generated stable δ-HEK cell line and subsequent transfection with α- and β-subunits of choice, including mutated subunits, to generate functional αβδ receptors that can be evaluated via measurements of membrane potential changes by fluorescence in the medium-throughput FMP assay (Falk-Petersen et al., 2017). The combination of α4- and β1-subunits was selected because α4 is often encountered together with δ in native receptors (Lee and Maguire, 2014) and because β1 conveniently does not lead to the formation of homomeric receptors in this system (Falk-Petersen et al., 2017). Indeed, using this setup, we here demonstrate the measurement of both reliable GABA responses and various PAM effects at α4β1δ receptors. We also report the successful generation of several stable δ-mutant cell lines, thus underlining this expression system as a versatile methodological tool for studying δ-GABAA molecular pharmacology in HEK cells with low variability. In cases of low-expressing receptors, as seen with some of the mutants examined here, we observed some discrepancies in the data obtained between FMP and patch-clamp electrophysiology. In this case, the FMP assay, which measures overall changes in membrane potential, appears to have some limitations as a result of lower overall sensitivity, especially in relation to efficacy of low-expressing mutants.
In our path to identifying the DS2 TMD interface binding site, we first examined one of the usual suspects, the ECD intersubunit α4(+)δ(−) interface, or the C-loop pocket (Jensen et al., 2013; Masiulis et al., 2019), as the site responsible for the PAM effect of DS2. Although the existence of this pocket has previously been debated (Wafford et al., 2009; Jensen et al., 2013; Ahring et al., 2016), we included it because of an observed structural resemblance between DS2 and the benzodiazepine binding site ligand zolpidem, which could suggest a potentially shared benzodiazepine-like binding site. We can now refute this hypothesis, also corroborated by our recent structure-activity relationship study of DS2 analogs targeting this site (Rostrup et al., 2021). Incidentally, one of these analogs (Br-DS2OPh), designed to bind in the ECD α(+)-δ(−) interface, fits well into the identified DS2 TMD pocket in the α4(+)β1(−) interface, showing the importance of experimental validation of binding site hypotheses based on molecular modeling. This further identifies Br-DS2OPh as a useful DS2 analog for future studies.
From our data, it is evident that the δ-subunit is not directly involved in the modulation by DS2, questioning what determines the δ-selective profile of the compounds. This is in accordance with previous data by Yakoub et al. (2018), who found that DS2 is capable of modulating receptors (in particular binary α6β3 receptors) that do not contain a δ-subunit. It is plausible that this is a matter of functional selectivity, similar to that observed for the superagonist 4,5,6,7-Tetrahydroisoxazolo[5,4-c]pyridin-3-ol hydrochloride (THIP) (gaboxadol), in which case binding in a highly conserved α-β interface gives rise to 10 times higher potency at the δ-containing receptors compared with both γ-containing and binary αβ receptors (Stórustovu and Ebert, 2006; Mortensen et al., 2010; Falk-Petersen et al., 2017). Also, PAMs (such as neurosteroids) have been found to display functional selectively at δ-containing receptors (Stell et al., 2003; Ahring et al., 2016), supposedly because GABA itself is only a partial agonist, leaving room for further activation (Dalby et al., 2020). Now, having a homology model of the confirmed DS2 binding site, the next step is to use this for structure-based drug design of DS2-related analogs or a radiolabeled analog for further validation of the binding site. Already, we have shown that classic medicinal chemistry approaches can improve both potency and selectivity of DS2 (Rostrup et al., 2021) and potentially brain permeability (L’Estrade et al., 2019). However, in line with already reported PAM effects, DS2’s functional activity is most effective at δ-containing subtypes (Jensen et al., 2013; Ahring et al., 2016; Yakoub et al., 2018) as a result of yet unknown factors. Ultimately, a cryo-EM structure in complex with one of the DS2 analogs would map the binding pocket including additional molecular interaction points and discern potential differences among subtypes.
In conclusion, our identification of the long-sought-after DS2 interaction site in the α4β1δ receptor may promote new insights into this highly important drug target class of δ-containing receptors suffering a general lack of selective tool compounds. Novel δ-selective analogs will aid to improve our understanding of the physiologic and pathophysiological role of δ-containing receptors. Such compounds may therefore potentially serve as leads for future rational drug development to treat the vast majority of neurologic disorders with dysregulated tonic inhibition as well as targeting conditions involving δ-containing GABAA receptors in the periphery, such as inflammation and immune disorders.
Acknowledgments
We would like to thank Dr. Uffe Kristiansen for intellectual input and scientific guidance with the patch-clamp electrophysiology studies and Durita Poulsen for technical assistance.
Authorship Contributions
Participated in research design: Falk-Petersen, Rostrup, Harpsøe, Gloriam, Frølund, Wellendorph.
Conducted experiments: Falk-Petersen, Rostrup, Löffler, Buchleithner, Harpsøe.
Contributed new reagents or analytic tools: Rostrup, Frølund.
Performed data analysis: Falk-Petersen, Rostrup, Löffler, Buchleithner, Harpsøe, Wellendorph.
Wrote or contributed to the writing of the manuscript: Falk-Petersen, Harpsøe, Wellendorph.
Footnotes
- Received February 18, 2021.
- Accepted April 22, 2021.
This work was financially supported by the Lundbeck Foundation [Grant R230-2016-2562] (to C.B.F-P.) and [Grant R277-2018-260] (to P.W.) and the Drug Research Academy (C.B.F-P. and F.R.). F.R was financially supported by a 2018 Lundbeck Foundation pregraduate scholar stipend in pharmaceutical neuroscience.
The authors declare no conflict of interest.
Part of this work was described in the Ph.D. thesis by C.B.F.-P.: Falk-Petersen CB (2020) Pharmacological insight into GABAA receptors with focus on β1-containing extrasynaptic subtypes Ph.D. thesis, the Faculty of Health and Medical Sciences, University of Copenhagen, Denmark.
↵This article has supplemental material available at molpharm.aspetjournals.org.
Abbreviations
- Br-DS2OPh
- N-(6-bromo-2-(thiophen-2-yl)imidazo[1,2-a]pyridin-3-yl)-4-phenoxybenzamide
- cryo-EM
- cryogenic electron microscopy
- DS2
- delta selective compound 2
- DS2OMe
- (4-methoxy-N-[2-(thiopen-2-yl)imidazole[1,2-a]pyridine-3-yl]benbamide)
- ECD
- extracellular domain
- FMP
- fluorometric imaging plate reader membrane potential
- GABAAR
- GABA type A receptor
- HEK
- human embryonic kidney
- PAM
- positive allosteric modulator
- RFU
- relative fluorescence unit
- TMD
- transmembrane domain
- WT
- wild type
- Copyright © 2021 by The American Society for Pharmacology and Experimental Therapeutics