Structural basis of AMPA receptor inhibition by 4-BCCA

Abstract

Background and Purpose:

AMPA receptors, which shape excitatory postsynaptic currents and are directly involved in overactivation of synaptic function during seizures, represent a well-accepted target for antiepileptic drugs (AEDs). Trans-4-butylcyclohexane carboxylic acid (4-BCCA) has emerged as a new promising AED in multiple in vitro and in vivo seizure models but the mechanism of its action remained unknown. The purpose of this study is to characterize structure and dynamics of 4-BCCA interaction with AMPA receptors.

Experimental Approach:

We studied the molecular mechanism of AMPA receptor inhibition by 4-BCCA using a combination of X-ray crystallography, mutagenesis, electrophysiology and molecular dynamics simulations.

Key Results:

We identified 4-BCCA binding sites in the transmembrane domain (TMD) of AMPA receptor, at the lateral portals formed by transmembrane segments M1-M4. At this binding site, 4-BCCA is very dynamic, assumes multiple poses and can enter the ion channel pore.

Conclusion and Implications:

4-BCCA represents a low-affinity inhibitor of AMPA receptors that acts at the TMD sites distinct from noncompetitive inhibitors, such as AED perampanel, and ion channel blockers. Future studies might examine a possibility of synergistic use of these inhibitors in treatment of epilepsy and a wide range of neurological disorders and gliomas.

1. INTRODUCTION

AMPA receptors, a subtype of ionotropic glutamate receptors, represent a well-accepted target for AEDs (Rogawski, 2011). They play a pivotal role in conducting postsynaptic currents and are thus essential for the over activation of synaptic function during seizures. The first approved drug treatment for epilepsy based upon AMPA receptor inhibition was perampanel (Fycompa; PMP), a noncompetitive AMPA receptor antagonist that is used as an adjunctive treatment for partial-onset and primary generalised tonic-clonic seizures (Frampton, 2015). The binding site for PMP on AMPA receptors has been well characterised, providing insight to how it inhibits receptor function (Chang et al., 2016; Narangoda, Sakipov & Kurnikova, 2019; Yelshanskaya, Singh, Sampson, Narangoda, Kurnikova & Sobolevsky, 2016; Yuan, Shi, Srinivasan, Ptak, Oswald & Nowak, 2019). However, PMP has dose-dependent behavioural side-effects, limiting its use in some patients (Rugg-Gunn, 2014).

Further interest in AMPA receptors as a molecular target in epilepsy treatment arose following the identification that these receptors are inhibited by a component of the medium chain triglyceride (MCT) diet that is used in the treatment of patients with drug resistant epilepsy (Neal et al., 2009). Although this diet was considered to function through ketone generation for seizure control (Augustin et al., 2018a; Bough & Rho, 2007), recent studies have shown a direct anti-seizure effect of decanoic acid (or capric acid; Figure 1) provided in this diet through the non-competitive inhibition of AMPA receptors (Chang et al., 2016; Chang et al., 2012; Chang et al., 2015; Wlaz et al., 2015). Interestingly, decanoic acid was suggested to bind to AMPA receptors at a site distinct to that of PMP (Chang et al., 2016). This was confirmed when combinatory treatment using both PMP and decanoic acid was found to directly inhibit AMPA receptors in a synergistic manner, and to synergistically inhibit seizure-like activity in rodent and human brain slice experiments (Augustin et al., 2018b). These data strongly suggest a direct binding site for decanoic acid and PMP on AMPA receptors, with inhibiting at both sites reducing seizure-like activity.

Figure 1.

Chemical structures of decanoic acid (DA) and trans-4-butylcyclohexane carboxylic acid (4-BCCA).

Due to the restrictions with dietary interventions, a search for drug-based approaches to reproduce the molecular mechanism of the MCT diet was pursued. Novel compound screening employed a range of structures related to both decanoic acid and the widely used AED, valproate, included assessment of their ability to inhibit seizure-associate changes in phosphoinositide turnover (Chang et al., 2012; Chang, Walker & Williams, 2014), and direct inhibition of AMPA receptors (Chang et al., 2016), with effectiveness in multiple in vitro and in vivo seizure models (Chang et al., 2012; Chang, Terbach, Plant, Chen, Walker & Williams, 2013; Chang et al., 2015; Warren, Walker & Williams, 2018). The most promising of these compounds (trans-4-butylcyclohexane carboxylic acid; 4-BCCA; Figure 1) showed strong activity in multiple in vitro and in vivo seizure models associated with distinct seizure types (Chang et al., 2012; Chang, Terbach, Plant, Chen, Walker & Williams, 2013; Chang et al., 2015; Warren, Walker & Williams, 2018) (Barker-Haliski, 2019): the mouse psychomotor seizure model reflecting partial or focal epilepsy syndromes (6Hz model: ED₅₀ 81 mg/kg); the mouse corneal-kindled model reflecting pharmacoresistant epilepsy with chronic seizures that exhibits behavioral and neuropathological changes of epilepsy (CDK model: ED₅₀ 44 mg/kg); the rat maximal electroshock model for generalized tonic clonic seizures (MES model: ED₅₀ ~100 mg/kg) and the mouse subcutaneous metrazol seizure threshold test model of a clonic, forebrain seizure (scMES model: ED₅₀ ~150 mg/kg). In all of these models, 4-BCCA showed enhanced potency over valproate (Chang et al., 2015). Importantly, this compound lacked the teratogenic-associated effect of histone deacetylase (HDAC) inhibition (Chang et al., 2015), widely shown for valproate (Gottlicher et al., 2001; Gurvich, Tsygankova, Meinkoth & Klein, 2004) that causes a range of detrimental developmental and neurological outcomes (Alsdorf & Wyszynski, 2005). These data provide a strong rationale for the continued analysis and development of 4-BCCA as a replacement AED for valproate. A clear understanding of the mechanism of AMPA receptor inhibition and synergistic inhibition with PMP will improve our understanding of the important new AED.

2. METHODS

2.1. Constructs and expression

In our experiments, we used rat GluA2i (flip) (NP_058957) subunit (also known as GluRBi or GluR2i) (Hollmann, O’Shea-Greenfield, Rogers & Heinemann, 1989; Keinänen et al., 1990), which has an amino acid sequence 99.7% identical overall and 100% identical in the transmembrane region (where 4-BCCA binding sites are located) to human GluA2i. Rat GluA2i was introduced with its native signal peptide into the pEG BacMam vector for expression in baculovirus-transduced HEK293 GnTI^– cells (Goehring et al., 2014). The construct, which we used for crystallization and structure determination, GluA2_Del, was described previously (Yelshanskaya, Singh, Sampson, Narangoda, Kurnikova & Sobolevsky, 2016) and included additional modifications. For purification and fluorescence detection purposes, coding sequences for a thrombin cleavage site (GLVPRG), an enhanced green fluorescent protein (eGFP) (Cormack, Valdivia & Falkow, 1996) and a Strep-tag (WSHPQFEK) were introduced at the carboxyl terminus. To improve crystallization behaviour of GluA2_Del, 36 residues were removed from the carboxyl-terminus, 6 residues (L378, T379, L381, P382, S383, and G384; numbering according to the mature polypeptide sequence) were deleted from the ATD-LBD polypeptide linker, three of four predicted N-linked glycosylation sites were knocked-out (N235E, N385D and N392Q), the 22 residue-long M1-M2 linker was replaced with the 3-residue DTD linker, the R586Q mutation at the Q/R site (Sommer, Kohler, Sprengel & Seeburg, 1991), which makes GluA2 receptors calcium permeable (Burnashev, Monyer, Seeburg & Sakmann, 1992; Hume, Dingledine & Heinemann, 1991), was introduced to stabilize the tetrameric state of the receptor and the C589A mutation was introduced to reduce non-specific disulfide bond formation (Sobolevsky, Rosconi & Gouaux, 2009).

2.2. Purification

HEK293 GnTI^– cells were harvested 60–96 hours after addition of BacMam P2 virus and collected by a low-speed centrifugation (6000 rpm, 10 min). Cells were disrupted using a Misonix Sonicator (18 × 15 s, power level 7) in a buffer containing 150 mM NaCl, 20 mM Tris-HCl (pH 8.0), 0.8 μM aprotinin, 2 μg/ml leupeptin, 2 μM pepstatin A and 1 mM phenylmethylsulfonyl fluoride (25 ml buffer/1L HEK293 culture). The homogenate was clarified using a Sorval centrifuge (8000 rpm, 15 min) and crude membranes were collected by ultracentrifugation (Ti45 rotor, 40000 rpm, 1 hour). The membranes were mechanically homogenized by pipetting and subsequently solubilized for 2 hours in a buffer containing 150 mM NaCl, 20 mM Tris-HCl (pH 8.0) and 20 mM C₁₂M (n-dodecyl-β-D-maltopyranoside; 0.25 g of C₁₂M / 1 g membranes). Insoluble material was removed by ultracentrifugation (Ti45 rotor, 40000 rpm, 40 min) and Strep-Tactin resin (0.5–1.0 ml per liter of cells) was added to the supernatant. After binding for 3–18 hours, the protein was eluted with the buffer containing 150 mM NaCl, 20 mM Tris-HCl (pH 8.0), 1 mM C₁₂M, and 2.5 mM desthiobiotin. Following concentration, the protein was first treated with EndoH (1:1 mass ratio of EndoH to receptor, 24 hours at 4°C), then digested with thrombin (1:200 mass ratio of thrombin to receptor, 1 hour at 22°C) and loaded onto a size exclusion chromatography (SEC) Superose 6 column equilibrated in a buffer composed of 150 mM NaCl, 20 mM Tris-HCl (pH 8.0), 1 mM n-undecyl-β-D-thiomaltopyranoside (C₁₁Thio) and 0.01 mg/ml lipid – 3:1:1 POPC:POPE:POPG (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine: 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine, and 1-palmitoyl-2-oleoyl-sn-glycero-3-[phospho-rac-(1-glycerol)]). Peak fractions were pooled and concentrated to ~2 mg/ml for protein crystallization. All steps were performed at 4°C unless otherwise noted.

2.3. Crystallization and cryoprotection

The best crystals of the GluA2_Del construct grew at 4°C in the hanging drop configuration. Prior to crystallization experiments, 2 mM 4-BCCA was added to the GluA2_Del protein, which was then subjected to ultracentrifugation (Ti100 rotor, 40000 rpm, 40 min, 4°C). The protein solution was then mixed with a crystallization buffer at receptor to crystallization buffer ratio of 2:1. The best crystals grew in two conditions, when crystallization buffer was composed of 8–11% (w/v) PEG 8,000, 0.2 M magnesium acetate and 0.1 M sodium cacodylate (pH 6.3–6.7) or 11–14% (w/v) PEG 6,000, 0.1 M ammonium phosphate and 0.1 M TRIS (pH 7.9–8.0). Crystals appeared after 5–7 days, grew as bars and reached the maximal size (~100 μm × ~100 μm × ~300 μm) in 30–60 days. Cryoprotection was carried out at 4°C by serial transfer into buffers containing increasing concentrations of glycerol, up to the maximum concentration of 20% (v/v), and then plunged into liquid nitrogen.

2.4. Structure determination and analysis

X-ray diffraction data collected at the Advanced Photon Source (beamlines 24-ID-C/E) were indexed, integrated and scaled using XDS (Kabsch, 2010) or HKL2000 (Otwinowski & Minor, 1997) (Table 1). The structure of GluA2 in complex with 4-BCCA (GluA2_4-BCCA) was solved by molecular replacement using Phaser (McCoy, 2007) and the apo state GluA2_Del structure (PDB ID: 5L1B) (Yelshanskaya, Singh, Sampson, Narangoda, Kurnikova & Sobolevsky, 2016) as a search probe. All structures were iteratively built in Coot (Emsley & Cowtan, 2004) and refined using Phenix (Adams et al., 2002). The four subunits in the tetrameric receptor structures contain residues 10 to 817, N-linked carbohydrates at N355 and molecules of 4-BCCA bound in the TMD. Regions with poor or missing electron density – residues in the M1-M2 loop (R545 to G572), in the non-helical region of M2 (A589 to R594) and in the carboxyl-terminus (K817 to G832) – were presumably disordered and therefore excluded from the final models. Structures were superposed using the CCP4 suite program Superpose (Krissinel & Henrick, 2004). Structural illustrations were made using the PyMOL Molecular Graphics System (Schrodinger, LLC) (DeLano, 2002).

Table 1.

Data collection and refinement statistics.

	GluA2Del-4BCCA
PDB accession code	6XSR
Data Collection
Beamline	APS-24ID-C
Space group	P212121
Cell dimensions
a, b, c, (Å)	92.8, 110.4, 600.1
α, β, γ (⁰)	90, 90, 90
Wavelength (Å)	0.92
Resolution (Å)	48.91–4.25
Completeness (%) *	99.53 (99.77)
Redundancy *	9.82 (7.53)
|I/σ| *	9.35 (1.45)
Rmeas (%)*	8.34 (103)
CC 1/2 *	99.7 (65.7)
Refinement
Resolution (Å)	4.25
Completeness (%) *	99.53 (99.77)
Number of reflections *	44507 (4352)
R work /R free *	0.2560 (0.3629)/0.2924 (0.3633)
Number of atoms
Total	3799
Ligand	82
B-factor (Å2)
Protein	257.5
Ligand	269.7
RMS deviations
Bond length (Å)	0.003
Bond angles (⁰)	0.50
Ramachandran
Favored (%)	92.0
Allowed (%)	7.3
Disallowed (%)	0.7

^*Highest resolution shell in parentheses.

5% of reflections were used for calculation of R_free.

2.5. Electrophysiology

DNA encoding wild type or mutant GluA2 was introduced into a plasmid for expression in eukaryotic cells that was engineered to produce green fluorescent protein via a downstream internal ribosome entry site (Yelshanskaya, Singh, Sampson, Narangoda, Kurnikova & Sobolevsky, 2016). All mutations were made in wild type GluA2 background. Human embryonic kidney HEK293 cells grown on glass coverslips in 35-mm dishes were transiently transfected with 1–5 μg of plasmid DNA using Lipofectamine 2000 Reagent (Invitrogen). Recordings were made 24 to 72 hours after transfection at room temperature. Currents from whole cells or outside-out patches, typically held at a −60 mV potential, were recorded using Axopatch 200B amplifier (Molecular Devices, LLC), filtered at 5 kHz and digitized at 10 kHz using low-noise data acquisition system Digidata 1440A and pCLAMP software (Molecular Devices, LLC). The external solution contained (in mM): 140 NaCl, 2.4 KCl, 4 CaCl₂, 4 MgCl₂, 10 HEPES (pH 7.3) and 10 glucose; 7 mM NaCl was added to the extracellular activating solution containing 3 mM L-glutamate (Glu). The internal solution contained (in mM): 150 CsF, 10 NaCl, 10 EGTA, 20 HEPES (pH 7.3). Rapid solution exchange was achieved with a two-barrel theta glass pipette controlled by a piezoelectric translator. Typical 10–90% rise times were 200–300 μs, as measured from junction potentials at the open tip of the patch pipette after recordings. Data analysis was performed using the computer program Origin 9.1.0 (OriginLab Corp.).

2.6. Molecular Modelling

The TMD and the TMD-LBD linkers (residues 506–544, 573–632, 782–816) of the receptor were used as the starting structure for all simulations. Residues 545–572 were modelled in ModLoop webserver (Fiser & Sali, 2003) and all other missing atoms were added in tleap program in AMBER16 MD simulation software package (Case et al., 2016). The receptor was embedded in a pre-equilibrated POPC lipid membrane and solvated in TIP3P water and neutralizing K⁺ ions. Each simulated system consisted approximately of 98×10³ atoms, including 648 protein residues, either two or four 4-BCCA ligands, 240 POPC lipids, and approximately 18,600 waters and neutralizing ions. The topology and parameter files for charged and uncharged 4-BCCA were created using Antechamber package (Wang, Wang, Kollman & Case, 2001) and General AMBER force field (GAFF) (Wang, Wolf, Caldwell, Kollman & Case, 2004). All MD simulations were carried out using pmemd.cuda in AMBER16 (Case et al., 2016), using ff14SB force field (Maier, Martinez, Kasavajhala, Wickstrom, Hauser & Simmerling, 2015) in combination with GAFF (Wang, Wolf, Caldwell, Kollman & Case, 2004) and Lipid14 (Dickson et al., 2014) force fields. The simulation time step was 2 fs and all covalent bonds to hydrogen atoms were constrained via SHAKE (Dickson et al., 2014). Electrostatic interaction calculations were performed using Particle Mesh Ewald (PME) method (Darden, York & Pedersen, 1993). Periodic boundary conditions were applied in all directions with a cutoff radius of 10 Å. The Langevin thermostat was used to control temperature and the Berendsen barostat with anisotropic pressure scaling was used to maintain pressure. All equilibration and production simulations were carried out at 1 atm pressure and 300K temperature.

Each system was minimized for 50,000 steps prior to MD simulations. The system was heated to 100K at the constant volume and brought up to 300K at the constant pressure. Heating was carried out in five 50 ps steps, with protein Cα atoms and ligand heavy atoms restrained using a harmonic potential with the force constant of 20 kcal mol⁻¹Å⁻². The system was then equilibrated at a constant pressure for 12 ns during which the restraints on protein Cα atoms and ligands were gradually decreased from 20 to 0.1 kcal mol⁻¹Å⁻². During the final 5 ns of this step, only protein Cα atoms and ring carbons of 4-BCCA were restrained. Then all restraints on ligands and protein were removed and the final equilibration step lasted 20 ns. Production simulations were carried out for 200 ns.

Post-processing of trajectories was done using cpptraj (Roe & Cheatham III, 2013) and VMD (Humphrey, Dalke & Schulten, 1996). Hydrogen bond analysis was carried out in cpptraj (Roe & Cheatham III, 2013) using the distance cut-off of 3.6 Å and angle cut-off of 135°. Protein-ligand contacts analysis and ligand density calculations were performed in VMD (Humphrey, Dalke & Schulten, 1996). Ligand RMSD was computed for heavy atoms of the ligand with reference to their initial positions, with each frame of the trajectory aligned using the coordinates of protein Cα atoms for residues 602 to 626 in the M3 helix.

2.7. Data and statistical analysis

The data and statistical analysis comply with the recommendations of the British Journal of Pharmacology on experimental design and analysis in pharmacology (Curtis et al., 2015). The results are presented as mean ± SEM. Statistical analysis was undertaken only for studies where each group size was at least n = 5. We used t-Test for comparisons, with differences considered statistically significant when P < 0.05.

2.8. Materials

Trans-4-Butylcyclohexanecarboxylic Acid (4-BCCA) was provided by Tokyo Chemical Industry UK Ltd, B1136.

3. RESULTS

3.1. Structure

To structurally characterize the interaction of 4-BCCA with AMPA receptors, we used a modified rat GluA2 subunit construct (GluA2_Del) that displays functional properties similar to wild type receptors and was used previously to solve structures of AMPA receptor in the apo state and in complex with noncompetitive inhibitors (Yelshanskaya, Singh, Sampson, Narangoda, Kurnikova & Sobolevsky, 2016). Crystals of GluA2_Del grew readily in the presence of 4-BCCA, belonged to the P2₁2₁2₁ space group with one tetramer per asymmetric unit and diffracted to 3.7–4.25 Å resolution. We solved the structure of GluA2 in complex with 4-BCCA (GluA2_4-BCCA) by molecular replacement using the apo state structure (PDB ID: 5L1B) as an initial search probe. The resulting electron density maps were of sufficient quality to build a molecular model of the entire receptor, excluding sections of the intracellular linkers connecting the transmembrane domain (TMD) segments M1 to M2 and M2 to M3, which were not visible in electron density maps, presumably due to their dynamic nature. The final model showed reliable crystallographic statistics and stereochemistry (Table 1).

GluA2_4-BCCA structure has a typical AMPA receptor domain arrangement and the overall Y shape (Sobolevsky, Rosconi & Gouaux, 2009) (Figure 2a). A large extracellular domain (ECD) comprises the amino-terminal domain (ATD) layer that sits atop of the Y and is necessary for receptor assembly, trafficking and functional regulation (Ayalon & Stern-Bach, 2001; Traynelis et al., 2010) and the ligand binding domain (LBD) layer in the middle of Y, which harbours binding sites for ligands that activate, modulate or antagonize the receptor (Gouaux, 2004; Stern-Bach, Bettler, Hartley, Sheppard, O’Hara & Heinemann, 1994). Below the ATD and LBD layers is the transmembrane domain (TMD) (Figure 2b) that forms a cation-selective ion channel (Wollmuth & Sobolevsky, 2004). Each LBD is comprised of two polypeptide stretches, S1 and S2, which form a clamshell structure, with ligand binding occurring in the middle, between the LDB upper (D1) and lower (D2) lobes. The TMD includes three transmembrane helices (M1, M3 and M4) and a re-entrant intracellular loop M2 between helices M1 and M3. Each LBD is tethered to the TMD through flexible polypeptide linkers S1-M1, M3-S2, and S2-M4. These linkers communicate conformational changes in the LBD induced by ligand binding to the TMD (Twomey & Sobolevsky, 2018).

Figure 2.

Structure of AMPA receptor in complex with 4-BCCA. (a) GluA2_4-BCCA structure viewed parallel to the membrane and perpendicular to the overall two-fold axis of molecular symmetry. Each subunit is in different colour. Inner and outer sides of membrane are indicated by parallel grey bars. 4-BCCA molecules are shown in space-filling representation. (b) Transmembrane domain viewed extracellularly, along the overall two-fold axis of molecular symmetry. (c) Close-up view of 4-BCCA binding site in the side portal between subunits C and B. 4-BCCA molecule is shown in ball-and-stick representation. Green mesh shows positive electron density for 4-BCCA in the Fo-Fc omit map contoured at 3σ.

The Fo-Fc maps generated for about 30 complete crystallographic datasets collected from crystals of GluA2_Del grown in the presence of 4-BCCA, with the maximum diffraction resolution between 3.7 and 4.25 Å, consistently showed positive densities in the TMD side portals located at the intersubunit interfaces and formed mainly by the M1 and M3 segments. These densities have not been previously observed in the apo-state, agonist-bound or noncompetitive inhibitor-bound structures (Yelshanskaya, Li & Sobolevsky, 2014; Yelshanskaya, Singh, Sampson, Narangoda, Kurnikova & Sobolevsky, 2016). The strength of these densities in GluA2_4-BCCA varied between the datasets as well as between intersubunit interfaces, being on average stronger for one diagonal pair of interfaces, A-D and B-C, and weaker for another pair, A-B and C-D (Figure 2b–c), consistent with different structural and functional roles of the two pairs of diagonal subunits (Dong & Zhou, 2011; Kazi, Dai, Sweeney, Zhou & Wollmuth, 2014; Twomey, Yelshanskaya, Grassucci, Frank & Sobolevsky, 2017). The size and shape of these densities matched those of 4-BCCA molecules but the precise orientation of the 4-BCCA molecules fitting these densities was rather ambiguous due to relatively low resolution. The corresponding putative binding sites of 4-BCCA are surrounded by mostly hydrophobic residues of M1 (L521, W526 and I529), M2 (F584, M585) and M3 (T609, I612, I613, Y616 and T617) of one subunit and M3 (F607, L610, I611 and S614) of the neighbouring subunit (Figure 2c).

3.2. MD simulations of 4-BCCA interaction with wild type AMPA receptors

To verify putative 4-BCCA binding sites identified by X-ray crystallography and to better understand the chemical nature of 4-BCCA interactions with AMPA receptor, we performed molecular dynamics (MD) simulations. First, we constructed a model of the GluA2 receptor transmembrane domain based on the crystal structure (see Methods). The protein model was embedded into a POPC lipid bilayer surrounded by water and ions (Figure 3a). The entire model system was equilibrated using the previously developed protocol (Narangoda, Sakipov & Kurnikova, 2019). Parameters for 4-BCCA model molecule were specifically developed for this study. In addition to the crystal structure with two 4-BCCA molecules at A-D and B-C subunit interfaces, two complexes were created by inserting 4-BCCA into all four equivalent binding sites of the protein in each of the two conformations in the crystal structure. Each system was simulated in both protonated and deprotonated states of the carboxylate group of 4-BCCA, resulting in six simulations of the wild type protein containing a total of twenty 4-BCCA ligands. On average, 4-BCCA molecules remained in close proximity to their initial positions but were mobile and exhibited significant reorientation during simulations, illustrated by representative positions of 4-BCCA from all simulations (Figure 3b) and combined density of the ligand averaged over all simulations (Figure 3c). As expected, deprotonated 4-BCCA entered the channel pore more readily than protonated 4-BCCA (Figure 3b). Three deprotonated ligands escaped the binding site and moved towards the pre-M1 and TMD-LBD linkers. Nevertheless, the region of the highest ligand density in simulations clearly coincides with the 4-BCCA position resolved by X-ray crystallography (Figure 3c). Consistent with the crystal structure (Figure 2c), the majority of simulated ligands formed contacts with residues of M1 (W526, I529), M2 (F584), and M3 (I612, I613, Y616, T617) of one subunit and M3 (F607, L610, I611, S614) of the neighbouring subunit (Figure 3c).

Figure 3.

MD simulations of AMPA receptor in complex with 4-BCCA. (a) Simulated GluA2_4-BCCA system in POPC lipid membrane and water. Protein subunits are shown in different colours. 4-BCCA molecules are shown in space-filling representation. Some lipids and waters are removed for clarity. (b) Representative positions of protonated (purple) and deprotonated (green) 4-BCCA from all simulations overlapped with the two crystal structure poses (yellow). Deprotonated ligands that did not remain in the binding region are not shown. (c) Top: density for 4-BCCA averaged over all simulations (green mesh) with several representative ligand orientations (yellow sticks). Bottom: the region with the highest ligand density in simulations representing the most likely orientation of 4-BCCA (mesh) overlapped with crystal pose (black sticks). Red and yellow regions of density represent polar/charged and hydrophobic regions of 4-BCCA, respectively. (d) Average lipid density at the binding site from four GluA2 simulations in the absence 4-BCCA (orange mesh) overlapped with ligand density from all GluA2_4-BCCA simulations (green mesh). Density plots were created using VolMap tool of VMD with an isovalue of 0.05 based on the positions of ligand/lipid heavy atoms. The lipid molecules shown in colored ball and stick representation are taken from a single simulation for illustrative purposes.

Interestingly, the average position of 4-BCCA in MD simulations is immediately adjacent to the preferred positions of lipid molecules in the vicinity of the 4-BCCA binding site. In MD simulations, residues W526 and F607, and sometimes I529 and F584, are in contact with membrane lipids (Supplementary Table 1). In the absence of 4-BCCA (or when 4-BCCA shifts away from its average binding position), lipids tend to form contacts with additional residues of the M3 helices that form the binding pocket (Supplementary Table 1). In simulations of AMPA receptors with noncompetitive inhibitors PMP, GYKI, and CP, which bind to a different site in the extracellular collar of the receptor, lipids interacted with M585 in addition to the residues mentioned above. Figure 3d shows the overlap in the positions of ligand density in 4-BCCA simulations and membrane lipids in a simulation of GluA2 in the absence of 4-BCCA. We hypothesize that the mechanism of AMPA receptor inhibition by 4-BCCA may involve destabilization of the protein surface by 4-BCCA outcompeting some of the annular lipids.

3.3. Probing of 4-BCCA binding sites by mutagenesis and electrophysiology

To functionally verify the putative binding sites, we mutated several residues in close proximity to the presumed 4-BCCA location and used the whole-cell patch-clamp technique to record currents from wild type and mutant AMPA receptors expressed in HEK 293 cells (Figure 4). Wild type GluA2 receptors were weakly inhibited by 4-BCCA showing only 40 ± 4 % current inhibition in response to 2 mM 4-BCCA application (Figure 4a). About half of the GluA2 mutants did not show any measurable current (Figure 4b), consistent with the location of the putative 4-BCCA binding sites right next to the ion channel activation gate formed by T617, A621, T625 and M629 at the M3 bundle crossing (Sobolevsky, Rosconi & Gouaux, 2009; Twomey & Sobolevsky, 2018; Twomey, Yelshanskaya, Grassucci, Frank & Sobolevsky, 2017). Indeed, many previous studies showed that mutations in this region of iGluR often result in leaky or non-functional receptors or receptors with severely altered kinetics (Beck, Wollmuth, Seeburg, Sakmann & Kuner, 1999; Chang & Kuo, 2008; Jones, VanDongen & VanDongen, 2002; Kashiwagi et al., 2002; Kohda, Wang & Yuzaki, 2000; Sobolevsky, Prodromou, Yelshansky & Wollmuth, 2007; Sobolevsky, Yelshansky & Wollmuth, 2003).

Figure 4.

Mutagenesis at the 4-BCCA binding site. (a) Examples of whole-cell currents recorded in the continuous presence of 30 μM cyclothiazide at −60 mV membrane potential from HEK293 cells expressing GluA2_WT, GluA2_M585A or GluA2_S614L in response to 1 s applications (black bars) of 3 mM Glu alone (blue) or co-applications of 3 mM Glu and 2 mM 4-BCCA (red). (b) Extent of current inhibition by 4-BCCA measured for wild type and mutant GluA2 (mean ± SEM), with individual measurements shown as the open circles. Symbols # indicate that the corresponding mutants did not show measurable currents and * statistically different extent of inhibition for the mutant compared to wild type GluA2 (t-Test, P < 0.05, n = 5–8).

For those mutants that did show currents in response to glutamate application, we observed three types of effects: (1) no change in 4-BCCA inhibition (e.g., L610F and S614Q, Figure 4b), (2) reduced inhibition (e.g., M585A, Figure 4a) and (3) increased inhibition (e.g., S614L, Figure 4a). To better understand possible reasons for such effects, we performed MD simulations of the mutant receptors and compared the results with those for the wild type receptors.

3.4. MD simulations of 4-BCCA interaction with mutant AMPA receptors

We have performed MD simulations of the mutants L610F, S614L, S614N, S614Q, M585A, and M585Q in complex with four protonated 4-BCCA molecules. In these simulations of mutant AMPA receptors, 4-BCCA exhibited two distinct changes in the behaviour compared to wild type receptors. In S614N and S614Q mutants, the 4-BCCA ligands interacted strongly with the mutated residues as most of them formed a hydrogen bond with the side chain amide group (Figure 5a and Table 2). In both S614N and S614Q mutant receptors, one of the ligands had entered the ion channel (Figure 5b). Formation of a hydrogen bond between the side chain of S614 and 4-BCCA was also observed in simulations of wild type GluA2. However, the protonated ligands in simulations of wild type receptors tended to remain at their binding sites. The lack of the hydrogen bond in the S614L mutant resulted in reduced interaction between 4-BCCA and leucine at position 614 and fewer contacts of 4-BCCA with I611 and I613 as well (Figure 5c and Table 2). In contrast to the S614N and S614Q mutants, the 4-BCCA ligands in S614L receptors tended to shift away from the ion channel pore toward the membrane lipids. Phenylalanine substitution in the L610F receptors also resulted in reduced interactions between 4-BCCA and residues S614, I611, and I613 (Figure 5d and Table 2). Although the ligands in L610F and S614L mutant receptors did not enter the ion channel pore, they interacted more extensively with the adjacent to the selectivity filter residues F584 and M585 (Figure 5d). In M585W and M585A receptors, the 4-BCCA ligands did not interact significantly with the mutated residues although one ligand in the M585W receptor had entered the ion channel pore. Overall, those molecules of 4-BCCA that had entered the pore also interacted with the residues at positions 614 and/or 610 of multiple subunits, as illustrated in Figure 5b. We also observed changes in the selectivity filter stability. For example, much lower flexibility of the selectivity filter and the entire M2 segment was observed in simulations of the S614N mutant compared to wild type receptors, signifying an improvement of the overall protein stability.

Figure 5.

MD simulations of AMPA receptor mutants in complex with 4-BCCA. The panels illustrate 4-BCCA modes of binding to GluA2 mutants, with mutated residues (red labels) and residues that interact with 4-BCCA shown. Hydrogen bonds between 4-BCCA and protein residues are shown as red dashed lines. (a) 4-BCCA at the binding site in S614N mutant. (b) 4-BCCA that entered the channel pore in S614N mutant. (c) 4-BCCA at the binding site in S614L mutant. Note that 4-BCCA does not interact with the mutated residue. (d) 4-BCCA at the binding site in L610F mutant.

Table 2.

4-BCCA interactions with key residues.

Protein	Ligand	Contact Frequency (%)
		F584	Residue 585	Residue 610 adjacent subunit	Residue 610 adjacent subunit	I611 adjacent subunit	I613	I613 adjacent subunit	I613 adjacent subunit	I613 opposite subunit	Residue 614 adjacent subunit	Residue 614 adjacent subunit
WT	A	-	-	97.60	-	67.20	73.80	-	-	-	97.80	-
	B	64.00	-	82.40	-	79.60	87.40	-	-	-	-	-
	C	-	-	54.80	-	76.20	-	-	-	-	60.60	-
	D	77.20	57.60	56.20	-	90.00	86.20	50.60	-	-	99.60	-
L610F	A	92.40	89.60	97.00	-	-	79.80	-	-	-	-	-
	B	-	-	100.00	-	84.40	99.00	-	-	-	97.80	-
	C	59.60	-	67.20	-	85.60	88.80	-	-	-	-	-
	D	51.40	-	-	-	-	-	-	-	-	-	-
S614L	A	-	-	80.22	-	95.60	53.63	-	-	-	-	-
	B	61.32	-	70.11	-	76.48	-	-	-	-	60.22	-
	C	94.07	86.15	-	-	-	-	-	-	-	-	-
	D	66.59	79.12	100.00	-	-	79.34	59.56	-	-	85.05	-
S614Q	A	86.17	-	87.07	-	66.21	75.28	-	-	-	82.54	-
	B*	-	-	74.60	-	-	64.40	-	-	-	86.39	74.38
	C	-	54.88	90.48	-	74.60	76.19	-	-	-	99.77	-
	D	57.60	-	71.43	-	82.77	75.06	-	-	-	78.68	-
S614N	A	92.40	-	73.80	-	81.60	87.60	-	-	-	-	-
	B	-	-	72.00	-	97.20	77.40	-	-	-	96.60	-
	C*	-	-	79.40	59.00	-	85.20	-	65.00	80.20	86.00	97.60
	D	-	-	80.00	-	87.20	79.80	-	-	-	74.20	-
M585W	A	89.00	-	82.20	-	84.80	95.60	-	-	-	-	-
	B	-	-	81.40	-	90.00	87.80	-	-	-	77.60	-
	C	54.20	-	64.40	-	67.20	70.60	-	-	-	95.40	-
	D*	-	-	-	93.20	-	-	-	89.60	74.40	-	65.00
M585A	A	65.60	-	97.80	-	84.60	86.40	-	-	-	70.80	-
	B	-	-	82.60	-	92.00	92.80	-	-	-	74.20	-
	C	65.80	-	78.60	-	84.00	73.80	-	-	-	75.20	-
	D	97.20	-	77.40	-	88.60	88.60	-	-	-	-	-

Contact frequencies between 4-BCCA and key protein residues in all mutant simulations and an example wild type simulation. Contacts that lasted for less than 50% of the simulation are not shown. Ligands that entered the channel pore are marked with asterisks (*).

The reduced current inhibition observed for the M585A mutant (Figure 4) could be due to reduced hydrophobicity of the side portal that leads to the putative 4-BCCA binding site, which might be necessary for the inhibitor entry. On the contrary, the weakly reduced inhibition observed for the M585W mutant could be due to the bulky indole ring of tryptophan physically obstructing the access of 4-BCCA to its binding site but this effect, if present, is insignificant. MD results do not contradict the experimental results because the simulations started with 4-BCCA already at the binding site. In M585A and M585W, ligands did not interact with the mutated residue. In M585W, one ligand entered the pore. In case of S614 mutations, different substitutions produced different effects, suggesting that the size, hydrophobicity and possibly the exact chemical character of the substituent side chain affects 4-BCCA inhibition. Thus, substitution of the hydroxymethyl group of serine with the amide group of asparagine in the S614N mutant resulted in significantly reduced current inhibition (Figure 4b), possibly due to the altered chemical microenvironment that is important for 4-BCCA binding. Our MD simulations also showed that the selectivity filter is less flexible in the S614N mutant compared to the S614Q mutant and wild type protein, probably reflecting the altered protein dynamics in the presence of 4-BCCA.

Interestingly, in case of the S614Q mutant, an increase in the size of the substituent side chain, which also has the amide group, almost entirely eliminated the reduction of inhibition observed for the S614N mutant, indicating an importance of the side chain size. On the other hand, substitution of the hydroxymethyl group of serine with the bulkier isobutyl group of leucine in the S614L mutant resulted in increased current inhibition, suggesting the importance of local hydrophobicity for 4-BCCA binding. In MD simulations of the S614L mutant, the loss of hydrogen bond between residue 614 and the ligand also resulted in reduced interactions between 4-BCCA and residues I611 and I613. No ligands entered the channel pore but 4-BCCA interacted with the adjacent to the selectivity filter residues F584 and M585 more extensively. Simulations of L610F, which showed no significant change in current inhibition (Figure 4b), gave similar results. At the same time, the majority of ligands in simulations of S614L moved away from the pore and interacted with lipids slightly more compared to other mutants. Overall, the results of our MD simulations and electrophysiological experiments on mutant and wild type receptors support the location of 4-BCCA binding sites identified crystallographically and provide clues about the mechanism of AMPA receptor inhibition by 4-BCCA.

4. DISCUSSION AND CONCLUSIONS

4.1. Unique 4-BCCA binding site and mechanism of AMPA receptor regulation

Using X-ray crystallography, we identified binding sites of 4-BCCA in the TMD of GluA2 AMPA receptor. These sites are located in the intramembranous side portals between transmembrane segments M1-M4, which connect the interior of the lipid bilayer to the ion channel pore, and are mostly contributed by residues in M1 and M3 (Figure 2). The 4-BCCA molecules are highly dynamic at their binding sites, assume different poses and occasionally enter the ion channel pore or the extracellular collar cavities between the pre-M1 and M3 segments (Figure 3). The dynamic character of 4-BCCA molecules bound to AMPA receptor and the approximate location of their binding sites were supported by mutagenesis combined with electrophysiology (Figure 4) and by MD simulations of mutant receptors (Figure 5). The dynamics of 4-BCCA is consistent with its low affinity towards GluA2 (Figure 4), which has been shown to be ~5 fold lower than towards GluA3 (Chang et al., 2015), and is a likely reason for the lack of markers indicating possible detrimental developmental and neurological side effects (Chang et al., 2015) that are typical for AEDs (Coyle, Clough, Cooper & Mohanraj, 2014; Rugg-Gunn, 2014; Steinhoff et al., 2014).

The location of 4-BCCA binding sites suggests possible molecular mechanisms of AMPA receptor inhibition. We hypothesize that 4-BCCA inhibits AMPA receptor by either making its presence at the ion channel pore and directly interfering with the flow of permeant ions (Figure 3b), or rigidifying the dynamics of the M3 segments, kinking of which is necessary for the channel to open (Twomey & Sobolevsky, 2018; Twomey, Yelshanskaya, Grassucci, Frank & Sobolevsky, 2017), or destabilizing the protein surface through competition with the surrounding membrane lipids (Figure 3d). It remains unclear how 4-BCCA molecules reach their binding sites. Being hydrophobic, 4-BCCA molecules can enter the ion channel side portals by going through the membrane but reaching the same sites by going through the pore cannot be excluded either, which would be consistent with 4-BCCA molecules entering the pore in MD simulations (Figure 3) and inhibiting currents in the presence of Glu in electrophysiological experiments (Figure 4). The identified 4-BCCA binding sites are likely locations for the whole cohort of the MCT ketogenic diet-associated medium chain fatty acids and their branched derivatives that share the inhibitory mechanism and represent promising AEDs (Chang et al., 2016; Chang et al., 2015). iGluRs are also modulated by polyunsaturated fatty acids, like docosahexanoic acid (DHA) and arachidonic acid (Miller, Sarantis, Traynelis & Attwell, 1992; Wilding, Chai & Huettner, 1998). Mutations of residues M589 and S618 in GluK2, homologous to residues M585 and S614 in GluA2 that affect 4-BCCA binding (Figure 4), produced strong effects on regulation of kainate receptors by DHA (Wilding, Chen & Huettner, 2010; Wilding, Fulling, Zhou & Huettner, 2008), suggesting that some of the effects of polyunsaturated fatty acids on iGluRs might be due to their binding to the 4-BCCA sites. Future studies will be required to examine this possibility in detail.

4.2. Synergy with other TMD regulators of AMPA receptor and clinical relevance

Interestingly, 4-BCCA binding sites are located in close proximity to the extracellular collar binding sites of noncompetitive inhibitors, including AED PMP (Figure 6). While our MD simulations suggest a possibility that such close proximity might result in 4-BCCA molecules occasionally approaching noncompetitive inhibitor binding sites, most of the time they stay away from these sites. There is therefore a strong possibility of simultaneous binding of 4-BCCA and PMP to AMPA receptor, which explains the synergy between these molecules observed in rodent and human brain slice experiments (Augustin et al., 2018b).

Figure 6.

Binding sites of 4-BCCA, PMP and NASPM in the TMD of AMPA receptor. The TMD is viewed parallel to the membrane, with the front and back subunits (B and D) removed for clarity. Molecules of 4-BCCA, PMP and NASPM are in stick representation (yellow).

The dynamic character of 4-BCCA revealed by MD simulations also suggests that this molecule can periodically enter the ion channel pore (Figure 3). In calcium-permeable AMPA receptors (which have glutamine at the Q/R-site, like the receptors in the present study), the ion channel pore is a binding site for blockers, which include endogenous polyamines, such as spermine and spermidine (Bowie & Mayer, 1995; Donevan & Rogawski, 1995; Kamboj, Swanson & Cull-Candy, 1995; Koh, Burnashev & Jonas, 1995), as well as exogenous polyamine- or acylpolyamine-containing toxins, such as philanthotoxin-433 from the wasp Philanthus triangulum (Eldefrawi et al., 1988), argiotoxin-636 from the spider Argiope lobata (Antonov, Grishin, Magazanik, Shupliakov, Vesselkin & Volkova, 1987; Grishin, Volkova, Arsen’ev, Reshetova & Onoprienko, 1986), Joro spider toxin JSTX-3 from Nephilia clavata (Aramaki, Yasuhara, Shimazaki, Kawai & Nakajima, 1987) and its synthetic analogue 1-naphthyl acetyl spermine (NASPM) (Koike, Iino & Ozawa, 1997; Takazawa, Yamazaki, Kanai, Ishida, Kato & Yamauchi, 1996; Twomey, Yelshanskaya, Vassilevski & Sobolevsky, 2018) (Figure 6). Since excessive activity of calcium-permeable AMPA receptors results in increased neuronal oxidative stress, which leads to cell damage and degeneration (Carriedo, Sensi, Yin & Weiss, 2000; Carriedo, Yin, Sensi & Weiss, 1998; Carriedo, Yin & Weiss, 1996; Weiss, 2011), calcium-permeable AMPA receptors have been linked to neurological disorders, including amyotrophic lateral sclerosis (Carriedo, Sensi, Yin & Weiss, 2000; Selvaraj et al., 2018; Tateno et al., 2004; Van Damme, Braeken, Callewaert, Robberecht & Van Den Bosch, 2005), Alzheimer’s (Whitehead, Regan, Whitcomb & Cho, 2017) and Parkinson’s (Kobylecki, Cenci, Crossman & Ravenscroft, 2010) diseases, ischemia-induced neuronal cell death (Liu et al., 2004; Talos et al., 2006; Yin, Hsu, Yu, Rao, Sorkin & Weiss, 2012) and epilepsy (Lippman-Bell, Zhou, Sun, Feske & Jensen, 2016; Rajasekaran, Todorovic & Kapur, 2012; Rogawski, 2013). Calcium-permeable AMPA receptors have also been associated with increased sensitivity to pain and hyperalgesia (Gangadharan et al., 2011; Park et al., 2009; Sullivan, Farrant & Cull-Candy, 2017), addiction (Mameli, Bellone, Brown & Luscher, 2011; Schmidt et al., 2015) and development of glaucoma (Cueva Vargas et al., 2015). In addition, calcium-permeable AMPA receptors have been recently shown to cause an increase in glioma cell intracellular calcium and brain tumour progression (Venkataramani et al., 2019; Venkatesh et al., 2019), while their elevated expression during development contributed to abnormal CNS development and fragile-X syndrome (Achuta et al., 2018).

All three classes of molecules binding at the TMD of AMPA receptor (Figure 6) are therefore promising candidates for the development of therapeutics targeting a broad spectrum of neurological disorders and brain tumours. Due to the distinct binding sites for all three compounds, they are likely to simultaneously bind to AMPA receptor, resulting in synergetic inhibition of the receptor, as it was demonstrated for DA and PMP (Augustin et al., 2018b). Thus, co-administration of representatives of these three classes of molecules might provide a future strategy for the treatment of neurological disorders and gliomas.