|Home | About | Journals | Submit | Contact Us | Français|
Excitatory neurotransmission plays a key role in epileptogenesis. Correspondingly, AMPA-subtype ionotropic glutamate receptors, which mediate the majority of excitatory neurotransmission and contribute to seizure generation and spread, have emerged as promising targets for epilepsy therapy. The most potent and well-tolerated AMPA receptor inhibitors act via a noncompetitive mechanism, but many of them produce adverse side effects. The design of better drugs is hampered by the lack of a structural understanding of noncompetitive inhibition. Here, we report crystal structures of the rat AMPA-subtype GluA2 receptor in complex with three noncompetitive inhibitors. The inhibitors bind to a novel binding site, completely conserved between rat and human, at the interface between the ion channel and linkers connecting it to the ligand-binding domains. We propose that the inhibitors stabilize the AMPA receptor closed state by acting as wedges between the transmembrane segments, thereby preventing gating rearrangements that are necessary for ion channel opening.
A number of antiepileptic drugs are available for the medical treatment of epilepsies, most of which block voltage-gated sodium or calcium channels, enhance gamma-aminobutyric acid (GABA) function by activation or positive allosteric modulation of GABAA receptors, inhibit GABA aminotransferase, or inhibit GABA reuptake from the synaptic cleft (Löscher et al., 2013; Serrano and Kanner, 2015). However, approximately 30% of all epilepsies have a drug-resistant course and require new treatment options (Steinhoff, 2015). One new direction in antiepileptic drug development is aimed at inhibiting excitatory neurotransmission, which plays a key role in epileptogenesis and seizure spread (Rogawski, 2011). Consequently, AMPA-subtype ionotropic glutamate receptors (iGluRs), which mediate the majority of excitatory neurotransmission, have emerged as a promising new target for epilepsy therapy (De Sarro et al., 2005; Meldrum and Rogawski, 2007). The most potent and well-tolerated inhibitors of AMPA receptors—those with fewer side effects—act via a noncompetitive (negative allosteric) mechanism. The originally discovered noncompetitive AMPA receptor antagonist GYKI 52466 (Donevan and Rogawski, 1993; Tarnawa et al., 1989) became a prototype for the development of more potent and selective 2,3-benzodiazepines (Bleakman et al., 1996; Donevan et al., 1994; Grasso et al., 1999; Ritz et al., 2011; Szénási et al., 2008; Tarnawa and Vize, 1998; Wang et al., 2014; Wang and Niu, 2013), such as GYKI 53655 (GYKI) (Balannik et al., 2005; Donevan et al., 1994), as well as structurally novel noncompetitive antagonists (Pelletier et al., 1996), including the quinazoline-4-one CP 465022 (CP) (Balannik et al., 2005; Lazzaro et al., 2002; Menniti et al., 2000) and the pyridone perampanel (PMP; Eisai) (Bialer et al., 2010; Chen et al., 2014a; Hibi et al., 2012). However, out of hundreds of publically reported compounds (Niu, 2015), PMP is thus far the only one approved for medical use as a safe and effective antiepileptic drug with low incidence of serious adverse effects, particularly at low doses (Patsalos, 2015; Steinhoff, 2015; Steinhoff et al., 2014). Nevertheless, at higher doses, patients taking PMP do experience side effects, including somnolence, dizziness, fatigue, irritability, nausea, headache, and falls, as well as depression and aggression (Coyle et al., 2014; Rugg-Gunn, 2014; Steinhoff et al., 2014), indicating the need for safer and more efficacious drugs.
Gaining a better understanding of how PMP and other compounds elicit their noncompetitive inhibition will aid the development of improved drugs targeting AMPA receptors. Previous studies have described the kinetics, potency, and several amino acid residues involved in interactions of noncompetitive antagonists with AMPA receptors (Balannik et al., 2005; Donevan and Rogawski, 1993, 1998; Lazzaro et al., 2002; Menniti et al., 2000). However, structural information concerning the action of noncompetitive inhibitors remains obscure. To address this knowledge gap, we solved structures of an AMPA-subtype rat GluA2 receptor in complex with several noncompetitive antagonists. Based on our structural data, combined with mutagenesis, electrophysiological recordings, and computational ligand docking, we propose a novel molecular mechanism of AMPA receptor inhibition by noncompetitive antagonists. These results establish a basis for the design of novel therapeutics to treat epilepsy and other disorders related to excitatory neurotransmission.
Attempts to obtain diffraction-quality crystals of previous GluA2 constructs used for structural studies in complex with noncompetitive inhibitors were unsuccessful. Thus, we modified the rat GluA2 AMPA receptor subunit construct (GluA2*) that we used to obtain the structure of agonist-bound receptor (Yelshanskaya et al., 2014) to make a new construct for crystallization. In our new construct, GluA2Del, we introduced the C589A point mutation to reduce non-specific disulfide bond formation (Sobolevsky et al., 2009) and replaced the 22-residue-long M1-M2 linker with the 3-residue aspartate-threonine-aspartate (DTD) linker (Figure S1). GluA2Del yields sufficient amounts of pure, monodisperse protein for crystallization experiments (Figures S2A and S2B). More importantly, GluA2Del exhibits wild-type-like functional behavior (Figures S2C–S2E), atypical of the previous AMPA receptor crystallization constructs that either show altered desensitization properties (Sobolevsky et al., 2009) or reduced current amplitudes (Chen et al., 2014b). More specifically, the maximal amplitude of 3 mM glutamate-induced current (I0), the fraction of non-desensitized receptors (ISS/I0), and the rates of deactivation (τDeact), entry (τDes), and recovery (τRecDes) from desensitization were similar for the wild-type (GluA2WT) and GluA2Del receptors (Figure S2F).
To verify that the GluA2Del construct retains sensitivity to noncompetitive inhibitors, we also tested GluA2WT and GluA2Del receptor-mediated current inhibition by PMP, GYKI, and CP. The extent of current inhibition increased with antagonist concentration (Figure 1A), and the concentration dependencies were similar for GluA2WT and GluA2Del receptors (Figure 1B). Fitting of concentration dependencies with the logistic equation yielded the following values of the half-maximal inhibition concentration (IC50) and the Hill coefficient (nHill) for PMP: IC50 = 0.88 ± 0.06 μM and nHill = 1.08 ± 0.04 for GluA2Del (n = 6) and IC50 = 0.89 ± 0.07 μM and nHill = 1.08 ± 0.05 for GluA2WT (n = 5); for GYKI: IC50 = 18.9 ± 1.8 μM and nHill = 1.29 ± 0.08 for GluA2Del (n = 5) and IC50 = 14.5 ± 1.6 μM and nHill = 1.29 ± 0.08 for GluA2WT (n = 6); and for CP: IC50 = 0.76 ± 0.14 μM and nHill = 1.73 ± 0.21 for GluA2Del (n = 5) and IC50 = 0.76 ± 0.03 μM and nHill = 1.72 ± 0.08 for GluA2WT (n = 6).
Finally, we tested the noncompetitive character of the GluA2WT and GluA2Del receptor-mediated current inhibition (Balannik et al., 2005; Chen et al., 2014a; Donevan and Rogawski, 1993; Lazzaro et al., 2002). Typical of noncompetitive antagonists, inhibition of currents by GYKI, PMP, and CP was voltage independent (Figures 1C and and1D).1D). Additionally, activation of the receptor was not required for current inhibition and recovery from it, as illustrated in Figure 1E with PMP as an example. Indeed, current decline during Glu and PMP co-applications showed progression of the PMP-induced inhibition. Recovery from the PMP-induced inhibition did not require receptor activation, as evidenced by the complete recovery of the current amplitude between consecutive co-applications of Glu and PMP (see the second and third traces in Figure 1E). Pre-incubation with PMP (the fourth and fifth traces in Figure 1E) eliminated the current decline, indicating that the development of the PMP-induced inhibition did not require receptor activation either. Instead, the initial current amplitude in response to Glu application in the continuous presence of PMP (open triangle) was lower than the steady-state current (filled triangle), suggesting that PMP binds to the resting (apo) receptors more efficiently than to activated receptors, similar to what was previously proposed for GYKI and CP (Balannik et al., 2005).
Complexes of GluA2Del and PMP, GYKI, or CP yielded protein crystals that belonged to the P212121 space group, had one GluA2Del tetramer in the asymmetric unit, and diffracted to 3.80–4.37 Å resolution (Table 1). We solved structures of GluA2Del in complex with PMP (GluA2PMP), GYKI (GluA2GYKI), and CP (GluA2CP) by molecular replacement, using the high-resolution structures of isolated amino-terminal domain (ATD; PDB: 3H5V) (Jin et al., 2009) and ligand-binding domain (LBD; PDB: 1FTO) (Armstrong and Gouaux, 2000) and the ion channel from the antagonist-bound full-length receptor structure (PDB: 3KG2) (Sobolevsky et al., 2009) as search probes. The resulting electron density maps were of sufficient clarity to build molecular models of the entire receptor, excluding sections of the intracellular linkers connecting transmembrane domain (TMD) segments M1 to M2 and M2 to M3, which were not visible in electron density maps. These models were refined to good crystallographic statistics and stereochemistry (Table 1).
GluA2PMP, GluA2GYKI, and GluA2CP structures have domain arrangements and the overall Y shape characteristic of the AMPA receptors (Sobolevsky et al., 2009) (Figure 2A). All three structures reveal additional electron densities at four equivalent positions at the interface between the ion channel and linkers connecting TMD to LBDs. Indeed, these densities match the size and shape of PMP, GYKI, and CP molecules (Figures 2B and and3).3). Moreover, the location of these novel noncompetitive inhibitor binding sites is consistent with the previous mutagenesis studies, which proposed binding of GYKI and CP at the interface between the LBD and the ion channel, specifically interacting with S1-M1 and S2-M4 linkers (Balannik et al., 2005).
To further verify the newly identified noncompetitive inhibitor binding sites, we synthesized a heavy atom bromine derivative of GYKI (GYKI-Br) and collected diffraction data at 0.92 Å wavelength from crystals of GluA2Del grown in the presence of GYKI-Br. These data revealed four anomalous difference Br peaks at the locations of the putative electron densities for PMP, GYKI, or CP (Figures 2B and S3), confirming that noncompetitive inhibitors bind at the interface between the ion channel and TMD-LBD linkers. We also observed a strong anomalous signal at the center of the ion channel pore, suggesting a possible binding site for GYKI-Br in the middle of the hydrophobic cavity (Figures S3B and S3C). GYKI-Br can reach this site either through the open ion channel pore or via the hydrophobic pathway through side portals between transmembrane helices (Sobolevsky et al., 2009), similar to small molecules and lipids in voltage-gated sodium channels (Payandeh et al., 2011). No electron density indicative of bound PMP, GYKI, or CP was observed at the center of the ion channel pore in the GluA2PMP, GluA2GYKI, or GluA2CP structures, suggesting that this site is specific to GYKI-Br or has a low affinity/occupancy for PMP, GYKI, or CP. Consistent with these structural findings, our mutagenesis experiments (see below) indicate that the contribution of this site to noncompetitive inhibition is negligible.
We also verified crystallographically identified binding sites and poses for PMP, GYKI, and CP in silico by performing molecular docking analyses (Figures S4A–S4C). The calculated negative energies of binding (−9 to −11 kcal/mol) for all three docked ligands strongly supported favorable binding to each of the four (one per subunit) identified binding sites. Moreover, the crystallographically revealed binding poses (gray in Figures S4A–S4C) scored as the first or second best predictions of 10 to 100 simulated attempts. Our modeling experiments therefore strongly support the structurally determined orientation of noncompetitive inhibitors within their binding pockets.
We also crystallized GluA2Del in the absence of noncompetitive inhibitors and solved the corresponding apo-state structure (GluA2Apo; Figure 4 and Table 1). GluA2Apo has a similar Y shape and closed pore to the noncompetitive-inhibitor-bound structures (Figures 2A, ,4A,4A, ,4E,4E, and and4F).4F). It is also quite similar to a previous “apo” structure of GluA2 (Dürr et al., 2014), showing close individual superposition for the ATD (root-mean-square deviation [RMSD] = 0.72 Å), LBD (RMSD = 0.66 Å), and the ion channel (RMSD = 1.19 Å). However, in the previous structure, the ligand-binding pocket of the LBD contained strong densities in the FO-FC map that were interpreted as bound 2-(N-morpholino) ethanesulfonic acid (MES) molecules. In contrast, our structure does not show any density in the ligand-binding pocket (Figure 4B), indicating that GluA2Apo represents a true apo-state structure. We also observed nearly continuous electron density for the pore-forming M2-M3 region in subunit D (Figure 4C), which makes GluA2Apo the first structure of AMPA receptor with the ion channel resolved in its entirety. Notably, the binding pockets harboring noncompetitive inhibitors in the GluA2PMP, GluA2GYKI, and GluA2CP structures do not contain electron density in the GluA2Apo structure (Figure 4D) and are likely filled with molecules of water.
Three structural elements, namely pre-M1 and extracellular portions of M3 and M4, form each one of the four equivalent noncompetitive inhibitor binding sites symmetrically located at the interface between the ion channel and TMD-LBD linkers (Figures 2B, ,2C,2C, and and4D).4D). Previous studies have placed the binding site of noncompetitive inhibitors at the interface between the subunits (Balannik et al., 2005; Sobolevsky et al., 2009). In contrast, our structures show that each binding site is comprised of residues from a single AMPA receptor subunit, with only one exception, S615 in M3 belonging to the neighboring subunit. The binding pocket is mostly hydrophobic, but along with nonpolar residues (F517 and P520 in the pre-M1 and L620 and F623 in M3), it also includes several polar (S516 in pre-M1, Y616 in M3, S788 and N791 in M4, and S615 in M3 of the adjacent subunit) and one negatively charged residue, D519, in preM1 (Figures 5A and and5B).5B). In all structures reported here, we kept the side chain conformations of residues contributing to noncompetitive inhibitor binding the same as in the original GluA2 structure (Sobolevsky et al., 2009). One exception, the side chain conformation of F623, was different depending on the bound inhibitor and was supported by clear differences in the 2FO-FC and FO-FC maps (Figures S4D and S4E). However, because of the modest resolution of the structures, all the side chain conformations should only be considered approximate. The ten residues contributing to noncompetitive inhibitor binding site in rat GluA2 are 100% conserved in all human AMPA receptor subunits GluA1 to GluA4 (Figure S5). Three of them (P520, Y616, and L620) are conserved across all human iGluR subunits. The remaining seven residues show similarity to AMPA receptors that decreases progressively for human kainate, delta, and NMDA receptor subunits and correlates with stronger noncompetitive inhibition by GYKI and CP of AMPA compared to kainate and NMDA receptors (Balannik et al., 2005; Donevan and Rogawski, 1993; Lazzaro et al., 2002).
To probe the contribution of individual residues to noncompetitive inhibitor binding, we individually mutated residues in the vicinity of the binding site to alanine and studied the effect of these mutations on glutamate-activated current inhibition by GYKI or PMP. Figure 5C illustrates the current inhibition for two alanine mutants, which show higher (D519A) or lower (F623A) affinity to GYKI compared to GluA2WT. Indeed, fitting of the concentration dependencies of current inhibition by GYKI with the logistic equation (Figure 5D) yielded IC50 = 3.80 ± 0.58 μM and nHill = 1.58 ± 0.12 for D519A (n = 7) and IC50 = 153 ± 8 μM and nHill = 1.18 ± 0.07 for F623A (n = 7) versus IC50 = 14.5 ± 1.6 μM and nHill = 1.29 ± 0.08 for GluA2WT (n = 6). The summary of effects of alanine substitutions on glutamate-activated current inhibition by GYKI or PMP is shown in Figure 5E.
Mutations of residues directly involved in noncompetitive inhibitor binding (S516A, F517A, D519A, P520A, S615A, F623A, S788A, and N791A) showed changes in IC50 values compared to wild-type receptors, while those facing away from the binding site (S510A, F515A, L518A, and Y523A) showed no significant difference. In the previous study (Balannik et al., 2005), mutations of two residues outside the noncompetitive inhibitor binding pocket were shown to affect sensitivity to CP (V792) and GYKI (A793). These residues are facing the intersubunit interface between M1 and M4 and their mutations may alter this interface, thus changing the arrangement of the ion channel domains as well as the size and shape of the noncompetitive inhibitor binding pocket accordingly.
In our experiments, alanine substitutions at S615 and F623 had the strongest effects on IC50 (Figure 5E). The two orders of magnitude increase in the IC50 value for these mutations (S615A for GYKI and F623A for PMP) supports direct contribution of S615 and F623 to the main noncompetitive inhibitor binding site. Such a strong effect on inhibition observed for the mutations introduced distal from the ion channel pore also supports our conclusion that the pore binding site identified for GYKI-Br (Figure S3) contributes negligibly, if at all, to noncompetitive inhibition of wild-type receptors by PMP and GYKI. Serine S615, the only residue from the neighboring subunit contributing to noncompetitive inhibitor binding, forms a hydrogen bond with the nitrogen of the pyridinyl ring of PMP (Figure 5A) or aminophenyl group of GYKI (Figure 5B). Phenylalanine F623, which adopts different side chain conformations between structures and forms a “cover” for the binding site, is involved in hydrophobic interactions with the dioxolobenzodiazepin moiety of GYKI or pairs with proline P520 to make a hydrophobic sandwich around the phenyl ring of PMP. Substitutions of two residues, D519 and S788, had no appreciable effect on inhibition by PMP but showed significant changes in IC50 for GYKI. Substitution of D519, which also affected sensitivity to CP (Balannik et al., 2005), increased affinity to GYKI, possibly due to removal of an unfavorable interaction between the carboxyl group of aspartate and the dioxolane group of GYKI. Reduced affinity of the S788A mutant to GYKI is consistent with the previous mutagenesis studies (Balannik et al., 2005) and likely results from the loss of a hydrogen bond between the serine hydroxyl group and the methylcarbamyl group of GYKI.
The results of our mutagenesis studies thus highlight the roles of individual residues in inhibitor binding; while the overall binding pocket is shared, the contribution of certain residues is dependent on the chemical structure of the inhibitor. Moreover, our electrophysiological experiments confirm that the crystallographically identified noncompetitive inhibitor binding sites are physiologically relevant.
Comparison of the GluA2Apo structure with GluA2PMP, GluA2GYKI, and GluA2CP reveals conformational changes in the AMPA receptor associated with noncompetitive inhibitor binding. Interestingly, conformations of the individual ATD, LBD, and TMD, as well as the entire extracellular domain, are similar in the apo and inhibitor-bound structures (Figure S6). However, in the inhibitor-bound structures, the ion channel viewed from the intracellular side undergoes an ~6° clockwise rotation relative to the extracellular domain (Figures 6A and and6B).6B). The overall change in AMPA receptor conformation caused by this rotation is likely the reason why apo and inhibitor-bound proteins produce different crystal forms (Figure S7). However, it is unclear whether the ~6° rotation is a part of the noncompetitive mechanism of inhibition, and this question awaits further resolution using dynamic approaches. On the other hand, there are clear local conformational changes in the TMD due to the inhibitor presence, especially in the relative positioning of pre-M1/M1, M3, and M4, which are pushed apart by the inhibitor molecules wedged in the binding pocket (Figure 6C). As a result, the upper part of the ion channel pore in noncompetitive inhibitor-bound structures is more tightly closed than in the apo structure (Figure 4F).
Based on the observed conformational changes and the location of inhibitor binding sites, we propose a model of noncompetitive antagonism (Figure 6D). The inhibitor molecule binds between pre-M1 and the extracellular portions of M1 and M4, immobilizing the transmembrane segments relative to each other and thus stabilizing the closed state of AMPA receptor. The relative displacement of transmembrane segments and TMD-LBD linkers that normally accompany receptor activation is therefore prevented by the inhibitor, and the ion channel cannot undergo pore opening, an idea reminiscent of conclusions from the previous mutagenesis studies (Balannik et al., 2005). Hence, the noncompetitive inhibitor functions as a wedge that blocks transmission of conformational changes in the iGluR activator domain, LBD, to the effector domain, ion channel. Previously, the “wedge” mechanism of action was proposed for partial allosteric agonism or positive allosteric modulation of pentameric glutamate-gated chloride channels and glycine receptors by ivermectin (Du et al., 2015; Hibbs and Gouaux, 2011). In pentameric channels, the wedge-shaped ivermectin inserts into the TMD interface of two adjacent subunits and stabilizes active conformations that are well described structurally (Althoff et al., 2014; Du et al., 2015; Hibbs and Gouaux, 2011). In contrast, the AMPA receptor noncompetitive inhibitors bind to the site almost entirely enclosed within individual subunits and stabilize the inactive receptor. Structures of AMPA receptors in active conformations are not yet available and understanding conformational selectivity of noncompetitive inhibitors awaits further structural and functional studies.
In summary, we have identified a novel AMPA receptor noncompetitive inhibitor binding site and propose that the inhibitors bound to this site act as wedges preventing conformation changes leading to the ion channel opening. Accordingly, the noncompetitive inhibitor binding pocket represents a new promising target for the design of pharmaceuticals with enhanced safety and efficacy.
Rat GluA2 AMPA receptor subunit was introduced into the pEG BacMam vector and the protein was expressed in suspension of baculovirus-transduced HEK293 GnTI− cells (Goehring et al., 2014). The protein was extracted from cellular membranes using n-dodecyl-β-D-maltopyranoside, treated with EndoH, and purified in two steps: Strep-Tactin affinity chromatography and size-exclusion chromatography (SEC).
The best crystals of GluA2 grew at 4°C in the hanging drop configuration in the presence or absence of 1 mM PMP, 1 mM GYKI, or 1 mM CP. X-ray diffraction data were collected at the synchrotrons and indexed, integrated, and scaled using XDS or HKL2000 (Table 1). The structures were solved by molecular replacement, iteratively built in Coot, and refined using Phenix or Refmac. Pore radii (Figure 4F) were calculated using HOLE. Structural illustrations were made using PyMOL.
DNA encoding wild-type or mutant GluA2 was introduced into a plasmid for expression in eukaryotic cells. HEK293 cells were transiently transfected with the plasmid DNA using Lipofectamine 2000 (Invitrogen). Patch-clamp current recordings were made using Axopatch 200B amplifier and a rapid solution exchange system with a two-barrel theta glass pipette controlled by a piezoelectric translator (Yelshanskaya et al., 2014). Data analysis was performed using Origin.
Docking of GYKI, PMP, and CP into each binding site was performed using the molecular docking program Autodock Vina (Trott and Olson, 2010). Each ligand’s three-dimensional structure was optimized in GAUSSIAN09 (Frisch et al., 2009) using B3LYP density functional theory method with the 6-31G basis set prior to docking. Both rigid docking and docking with rotatable bonds within the ligand molecules were carried out. The protein was either kept rigid or select amino acid side chains were made flexible.
We thank Kei Saotome and Edward C. Twomey for comments on the manuscript. We thank Eisai, Inc. for the generous gift of PMP and Dr. Yael Stern-Bach for GYKI and CP. We also thank the personnel at beamlines 24-ID-C/E of APS and 5.0.1/5.0.2 of ALS. This work was supported by the NIH grant R01 NS083660, the Pew Scholar Award in Biomedical Sciences, the Schaefer Research Scholar Award, the Klingenstein Fellowship Award in the Neurosciences, and the Irma T. Hirschl Career Scientist Award (A.I.S.).
The accession numbers for the atomic coordinates and structure factors reported in this paper are PDB: 5L1B, 5L1E, 5L1F, 5L1G, and 5L1H.
AUTHOR CONTRIBUTIONSM.V.Y. and A.I.S. designed the project. M.V.Y. expressed and purified protein and carried out mutagenesis and electrophysiological recordings. M.V.Y., A.K.S., and J.M.S. performed crystallographic experiments. C.N. and M.K. carried out molecular modeling. M.V.Y., A.K.S., J.M.S., M.K., and A.I.S. contributed to manuscript writing.