Abstract
The mammalian ionotropic glutamate receptor family encodes 18 gene products that coassemble to form ligand-gated ion channels containing an agonist recognition site, a transmembrane ion permeation pathway, and gating elements that couple agonist-induced conformational changes to the opening or closing of the permeation pore. Glutamate receptors mediate fast excitatory synaptic transmission in the central nervous system and are localized on neuronal and non-neuronal cells. These receptors regulate a broad spectrum of processes in the brain, spinal cord, retina, and peripheral nervous system. Glutamate receptors are postulated to play important roles in numerous neurological diseases and have attracted intense scrutiny. The description of glutamate receptor structure, including its transmembrane elements, reveals a complex assembly of multiple semiautonomous extracellular domains linked to a pore-forming element with striking resemblance to an inverted potassium channel. In this review we discuss International Union of Basic and Clinical Pharmacology glutamate receptor nomenclature, structure, assembly, accessory subunits, interacting proteins, gene expression and translation, post-translational modifications, agonist and antagonist pharmacology, allosteric modulation, mechanisms of gating and permeation, roles in normal physiological function, as well as the potential therapeutic use of pharmacological agents acting at glutamate receptors.
I. Introduction and Nomenclature
The past decade has revealed both breathtaking advances in our understanding of structure and function and a growing sophistication at virtually all levels of experimental design. The structure of a membrane-spanning tetrameric glutamate receptor has been described, revealing unprecedented features of channel structure together with long-awaited details on the pore-forming elements and the channel gate, subunit arrangement, and the nature of linkers connecting multiple semiautonomous domains that comprise the extracellular portion of the receptor. These compelling data have set the stage for a predictably explosive increase in work on all aspects of function and hold the promise of catalyzing timely breakthroughs in therapeutic strategies.
Assembling this review was an exciting yet daunting task. A staggering volume of literature has been published over the last 11 years, the period this review most seeks to summarize. We have focused primarily on the pharmacology of glutamate receptors, the structural basis of receptor function as it relates to neuronal function and neurological disease, and the regulation of receptor function by phosphorylation. We only touch upon the anatomical distribution of glutamate receptors, their role in behavior and cognition, their role in nervous system development, and the means by which the myriad of proteins that bind to glutamate receptors regulate receptor trafficking. We focus on mammalian receptors, with an emphasis on their relation to potential therapies now under development. In selecting the necessarily limited number of references used to illustrate advances, we have sought to recognize principle, precedent, perspective, and (importantly) to acknowledge the full spectrum of talented individuals and productive laboratories engaged in this field. We regret that space does not allow a complete listing of relevant work related to each point made; many fine articles simply could not be cited.
After the first report appeared in December 1989 of the cloning of a glutamate receptor subunit (Hollmann et al., 1989), the early 1990s witnessed a flurry of activity, resulting in reports of more than a dozen glutamate receptor clones in various species within the subsequent 6 months. As might be expected, the nomenclature was uncoordinated, with species-or laboratory-specific names for the same transcript being promoted in the literature. This situation has resolved slowly. An excellent history of glutamate receptor cloning and nomenclature has appeared (Lodge, 2009). Glutamate receptor nomenclature has recently undergone a needed and systematic revision, the International Union of Basic and Clinical Pharmacology name replacing the common names (Collingridge et al., 2009) (see http://www.iuphar-db.org/LGICNomenclature.jsp). Table 1 summarizes the nomenclature used throughout this review for both genes and gene products.
II. Structure
A. Subunit Organization and Quaternary Structure
Ionotropic glutamate receptors are integral membrane proteins composed of four large subunits (>900 residues) that form a central ion channel pore. Sequence similarity among all known glutamate receptor subunits, including the AMPA,1 kainate, NMDA, and δ receptors, suggests they share a similar architecture (Table 2). Glutamate receptor subunits are modular structures that contain four discrete semiautonomous domains: the extracellular amino-terminal domain (ATD), the extracellular ligand-binding domain (LBD), the transmembrane domain (TMD), and an intracellular carboxyl-terminal domain (CTD) (Fig. 1A). Apart from the CTD and the M4 segment, each of the individual domains exhibits low sequence homology to bacterial proteins with known structures and, in some instances, a related function (O'Hara et al., 1993; Wo and Oswald, 1995; Wood et al., 1995; Paas, 1998; Kuner et al., 2003). Detailed crystallographic structures have been described for a membrane-spanning tetrameric glutamate receptor (Sobolevsky et al., 2009) as well as the isolated ATDs and LBDs in complex with various agonists, antagonists, and modulators (discussed in section VI). These data, along with functional and biochemical experiments, have begun to define the relationship between receptor structure and function.
The first views of the quaternary glutamate receptor structure were provided by single particle images of recombinant and native AMPA receptors obtained by electron microscopy (Safferling et al., 2001; Tichelaar et al., 2004; Nakagawa et al., 2005, 2006; Midgett and Madden, 2008). Although these images show the receptors at lower resolution (∼40–20 Å), some structural features could be extracted. For example, an internal 2-fold rotational symmetry was observed for some of these receptor structures (Tichelaar et al., 2004; Midgett and Madden, 2008), consistent with indications that glutamate receptors assemble as a dimer of dimers. This proposed 2-fold rotational symmetry for glutamate receptors is in contrast to the symmetry observed in structures of other ion channels, such as tetrameric K+-channels and the pentameric nicotinic acetylcholine receptor, in which the quaternary subunit arrangement leads to rotational symmetries that correlate with subunit-number (MacKinnon, 2003; Miyazawa et al., 2003; Sobolevsky et al., 2004; Wollmuth and Sobolevsky, 2004).
Crystallographic studies have provided the first detailed structure of a membrane-spanning glutamate receptor (3.6 Å) (Fig. 1B). This structure of an antagonist-bound tetrameric rat GluA2 demonstrates that the receptor has an overall 2-fold symmetry perpendicular to the membrane plane; the extracellular ATDs and LBDs are organized as dimers of dimers, and the ion channel domain exhibits a 4-fold symmetry (Sobolevsky et al., 2009). This subunit arrangement relates one ATD dimer to another and one LBD dimer to the second, and half of the pore-forming TMDs to the other half. The symmetry mismatch between the ATDs and LBDs arises because the receptor contains two conformationally distinct subunits, which can be denoted A/C and B/D subunits (Fig. 1, C and D). Consequently, the A/C subunits will couple differently to the ion channel gate than will the B/D subunits, which may have important implications for the function of the glutamate receptors.
In the tetrameric structure (Sobolevsky et al., 2009), the ATD forms two distinct types of subunit-subunit contacts. The most extensive contact is formed between A/B and C/D subunits, and this contact is identical to that observed between subunits in the structure of the isolated GluA2 and GluK2 ATD dimer (Clayton et al., 2009; Jin et al., 2009; Kumar et al., 2009). The other contact is located on the 2-fold symmetry axis and is formed between B and D subunits of the A/B and C/D dimers (Fig. 1C). In addition, at the level of the LBD, two distinct types of subunit-subunit contacts are formed. The LBDs are arranged as A/D and B/C dimers with contacts between the A and C subunits (Fig. 1C). The domain swapping or subunit crossover causes a different subunit arrangement at the levels of the ATD and the LBD. As predicted by topology studies (Hollmann et al., 1994; Bennett and Dingledine, 1995) and the homology to the tetrameric K+-channels (Wo and Oswald, 1995; Wood et al., 1995; Kuner et al., 2003), the glutamate receptor TMD consists of three transmembrane helices (M1, M3, and M4) and a membrane re-entrant loop (M2) (Sobolevsky et al., 2009). In addition, the subunits have a short helix (pre-M1) that is oriented parallel to the membrane. M1, M2, and M3 form a structure that closely resembles that of an inverted K+-channel pore, and M4 primarily makes contacts with the TMD of an adjacent subunit.
The observation that subunits with the same polypeptide sequence adopt two distinct conformations in the tetrameric receptor complex is without precedent in an ion channel (Sobolevsky et al., 2009). The subunit crossover between the ATD and LBD levels of the tetramer (Fig. 1D) is primarily mediated by the ATD-S1 amino acid linkers that connect the ATD with the LBD. The ATD-S1 linkers of the A/C subunits adopt a compact conformation, whereas the ATD-S1 linkers of the B/D subunits have an extended conformation. This structural role of the ATD-S1 linker is intriguing, because previous studies have implicated this segment in the control of the open probability of NMDA receptors (Gielen et al., 2009; Yuan et al., 2009a). The symmetry mismatch between the LBD and the TMD levels also is mediated primarily by the linkers connecting the two domains (S1-M1, M3-S2, and S2-M4 linkers). Also here, the linkers adopt two different conformations corresponding to the A/C subunits and the B/D subunits. The involvement of the TMD-LBD linkers in the function of glutamate receptors has been extensively studied (Krupp et al., 1998; Villarroel et al., 1998; Sobolevsky et al., 2002a,b; Watanabe et al., 2002; Yelshansky et al., 2004; Balannik et al., 2005; Schmid et al., 2007), and the tetrameric structure provides an excellent opportunity to interpret these and other results in a structural context. Whereas tetrameric kainate receptors appear to have the same extracellular architecture as AMPA receptors (Das et al., 2010), it remains to be shown how well the tetrameric AMPA receptor structure corresponds to structures for NMDA receptors.
B. Subunit Stoichiometry
The glutamate receptors assemble as tetrameric complexes of subunits (Laube et al., 1998; Mano and Teichberg, 1998; Rosenmund et al., 1998; Greene, 2001; Matsuda et al., 2005; Nakagawa et al., 2005; Sobolevsky et al., 2009), and functional receptors are formed exclusively by assembly of subunits within the same functional receptor class (Partin et al., 1993; Kuusinen et al., 1999; Leuschner and Hoch, 1999; Ayalon and Stern-Bach, 2001; Ayalon et al., 2005). Glutamate receptors are grouped into four distinct classes based on pharmacology and structural homology, including the AMPA receptors (GluA1–GluA4), the kainate receptors (GluK1–GluK5), the NMDA receptors (GluN1, GluN2A–GluN2D, GluN3A, and GluN3B), and the δ receptors (GluD1 and GluD2). The AMPA receptor subunits GluA1 to GluA4 can form both homo- and heteromers. The kainate receptor subunits GluK1 to GluK3 also form both homo- and heteromers, but GluK4 and GluK5 form functional receptors only when coexpressed with GluK1 to GluK3. The δ receptors GluD1 and GluD2 are capable of forming homomeric receptors yet seem incapable of forming heteromers with AMPA, kainate, and NMDA receptor subunits, both in native cells and in heterologous expression systems (Partin et al., 1993, 1995; Mayat et al., 1995; Zuo et al., 1997; Kohda et al., 2000; Ikeno et al., 2001; Naur et al., 2007). In addition, GluD1 and GluD2 seem incapable of forming receptors that can be activated by any known agonists (see section V.A). Whether GluD1 and GluD2 can form heteromeric receptors is unresolved.
Functional NMDA receptors require assembly of two GluN1 subunits together with either two GluN2 subunits or a combination of GluN2 and GluN3 subunits (Monyer et al., 1992; Schorge and Colquhoun, 2003; Ulbrich and Isacoff, 2007, 2008). NMDA receptors further require simultaneous binding of both glutamate and glycine for activation (Johnson and Ascher, 1987; Kleckner and Dingledine, 1988; Lerma et al., 1990). The GluN1 and GluN3 subunits provide the glycine binding sites (Furukawa and Gouaux, 2003; Furukawa et al., 2005; Yao et al., 2008), and the GluN2 subunits form the glutamate binding sites (Furukawa et al., 2005). The GluN1 subunit expressed alone in Xenopus laevis oocytes responded weakly to coapplication of glutamate and glycine (Moriyoshi et al., 1991; Nakanishi et al., 1992; Yamazaki et al., 1992). These responses have been proposed to arise because X. laevis oocytes express low levels of endogenous NMDA receptor subunits (XenGluN1 and XenGluN2) that under some circumstances functionally assemble with GluN1 (Green et al., 2002; Schmidt et al., 2006, 2009; Schmidt and Hollmann, 2008, 2009), which can complicate studies on NMDA receptors using the X. laevis expression system. No responses are observed from GluN1 expressed alone in mammalian cells.
GluN1 also can combine with two different GluN2 subunits to form triheteromeric receptors. Numerous studies support the formation of GluN1/GluN2A/GluN2B, GluN1/GluN2A/GluN2C, GluN1/GluN2B/GluN2D, GluN1/GluN2A/GluN2D receptors in different brain regions and in specific neuronal subpopulations (Chazot et al., 1994; Sheng et al., 1994; Chazot and Stephenson, 1997; Luo et al., 1997; Sundström et al., 1997; Dunah et al., 1998a; Cathala et al., 2000; Green and Gibb, 2001; Piña-Crespo and Gibb, 2002; Brickley et al., 2003; Dunah and Standaert, 2003; Fu et al., 2005; Jones and Gibb, 2005; Lu et al., 2006; Brothwell et al., 2008). Few studies have addressed the functional implications of the presence of two different GluN2 subunits in the NMDA receptor complex (Brimecombe et al., 1997; Cheffings and Colquhoun, 2000; Hatton and Paoletti, 2005).
The GluN3 subunits bind glycine and do not form functional receptors alone (Chatterton et al., 2002; Yao and Mayer, 2006). When coexpressed with GluN1 in X. laevis oocytes, GluN1/GluN3 receptors can form receptors that are activated by glycine alone (Chatterton et al., 2002), but these excitatory glycine receptors have not yet been observed in GluN3-expressing neurons (Matsuda et al., 2003). At present, surface expression of glycine-activated GluN1/GluN3A or GluN1/GluN3B receptors in HEK293 cells is unresolved, but GluN1/GluN3A/GluN3B shows some functional expression (Smothers and Woodward, 2007). When GluN3 is coexpressed with GluN1 and GluN2 in X. laevis oocytes, NMDA- and glutamate-activated current amplitudes are reduced compared with current from GluN1/GluN2, suggesting that either triheteromeric GluN1/GluN2/GluN3 receptors form that have a lower conductance, or GluN3 expression reduces trafficking or assembly of GluN1/GluN2 (Das et al., 1998; Perez-Otano et al., 2001; Ulbrich and Isacoff, 2007, 2008). Triheteromeric GluN1/GluN2/GluN3 receptors presumably form in cortical neurons based on the observation of single-channel currents with properties that could not be attributed to either GluN1/GluN2 or GluN1/GluN3 receptors (Sasaki et al., 2002). The subunit stoichiometry and surface expression of GluN3-containing NMDA receptors and the physiological relevance of triheteromeric GluN1/GluN2/GluN3 receptors are not fully resolved.
C. Receptor Assembly and Trafficking
AMPA receptors assemble as dimers of dimers with ATD interactions presumably mediating the initial dimer formation. Subsequent tetramerization (i.e., assembly of two subunit dimers) occurs through interactions of the LBDs and the TMDs (Ayalon and Stern-Bach, 2001; Mansour et al., 2001; Ayalon et al., 2005). Receptor assembly occurs in the endoplasmic reticulum (ER), where quality control mechanisms ensure correct subunit folding and assembly. Data suggests that conformational changes associated with the normal function of glutamate receptors, such as ligand binding, activation, and desensitization, take place in the ER lumen, and these conformational changes may influence trafficking (Greger et al., 2002; Fleck et al., 2003; Grunwald and Kaplan, 2003; Mah et al., 2005; Valluru et al., 2005; Greger et al., 2006; Priel et al., 2006; Penn et al., 2008). Consequently, glutamate receptors may require ligands or “chemical chaperones” for efficient folding and export from the ER. This is evident when the conformational changes associated with the normal function are modified by mutagenesis. Nondesensitizing GluA2(L483Y) mutants exit from the ER inefficiently, whereas GluA2 (N754D), which has increased desensitization, exits efficiently from the ER (Greger et al., 2006). Block of desensitization has been shown to similarly influence kainate receptor trafficking (Priel et al., 2006; Nayeem et al., 2009). The mechanisms are unclear, but block of desensitization could interfere with association and/or dissociation of chaperones and/or transport proteins, with potential candidates being TARPs or CNIHs that are thought to be auxiliary subunits (see section II.H).
Data suggest that the ATD plays a crucial role in receptor oligomerization and perhaps trafficking (Kuusinen et al., 1999; Leuschner and Hoch, 1999; Ayalon and Stern-Bach, 2001; Ayalon et al., 2005; Qiu et al., 2009). The interaction between the ATDs is sufficient to allow isolated ATDs to form stable dimers in solution (Clayton et al., 2009; Jin et al., 2009; Kumar et al., 2009). A key role of the AMPA receptor ATD may be to direct assembly of the tetrameric receptor and to prevent kainate or NMDA receptor subunits from entering the tetramer (Kuusinen et al., 1999; Leuschner and Hoch, 1999; Ayalon et al., 2005), and some segments of the AMPA receptor ATD have been implicated in subtype-specific assembly (Leuschner and Hoch, 1999; Ayalon et al., 2005). In addition, AMPA receptor subunit stoichiometry is controlled by RNA editing, which precedes mRNA splicing and protein synthesis at two sites that modulate function: the RG site within the GluA2 to GluA4 LBD, and the QRN site at tip of the reentrant pore loop. Editing switches the codon at the QRN site from Gln to Arg in a majority of GluA2 RNA. These sites are located within subunit interfaces and are thought to affect receptor assembly by favoring heterodimerization over homodimerization, which partly explains why GluA2-containing AMPA receptors are mostly heteromers (Mansour et al., 2001; Greger et al., 2002, 2003, 2006). In addition, GluA2 subunits edited at the QRN site have increased dwell time in the ER compared with other AMPA receptor subunits, thereby increasing their availability for assembly with other subunits (Greger et al., 2002).
Three models have been suggested for assembly of NMDA receptors. The first model suggests that GluN1-GluN1 and GluN2-GluN2 homodimers initially form and subsequently coassemble to form the tetrameric receptor (Meddows et al., 2001; Schorge and Colquhoun, 2003; Papadakis et al., 2004; Qiu et al., 2005). The second model proposes that a GluN1-GluN1 homodimer forms a correctly folded stable complex to which two GluN2 monomers are added sequentially to form the NMDA receptor tetramer (Atlason et al., 2007). The third model suggests initial GluN1-GluN2 heterodimer formation and subsequent tetramerization (Schüler et al., 2008). At present, there are insufficient data to distinguish between the different models. However, the two conformationally distinct subunits with two types of subunit-subunit contacts observed in the GluA2 AMPA receptor structure (Sobolevsky et al., 2009) might provide the structural framework needed to design experiments to resolve this issue. Similar to AMPA receptors, the ATD is thought to mediate initial dimer formation of NMDA receptor subunits (Meddows et al., 2001; Papadakis et al., 2004).
D. The Extracellular Ligand Binding Domain
The LBD is highly conserved within the different glutamate classes (Table 2) and is formed by two extracellular stretches of amino acids historically referred to as S1 and S2 (Stern-Bach et al., 1994) (Fig. 1A). The structures of excised S1 and S2 amino acid sequences joined by an artificial polypeptide linker to form the LBDs have been described both with agonist and antagonist bound. All LBD structures adopt a clamshell-like conformation, where the polypeptide segment S1, located on the amino-terminal side of membrane helix M1, forms most of one half of the clamshell (D1), and the segment S2 between the M3 and M4 membrane helices forms most of the opposite half of the clamshell (D2) (Fig. 1A). The agonist binding pocket is located within the cleft between these two lobes. Several lines of experimental work have validated that the agonist-binding site in the soluble LBDs used for crystallization faithfully resembles the binding sites in intact receptors (Armstrong and Gouaux, 2000; Furukawa and Gouaux, 2003; Du et al., 2005; Gonzalez et al., 2008; Sobolevsky et al., 2009). In addition, comparison of UV absorption spectra that probe the molecular configuration of the AMPA receptor antagonist CNQX bound to either the isolated GluA2 LBD or the full-length GluA2 suggests that the structure of the soluble LBD resembles that within the full-length receptor (Deming et al., 2003).
The initial step in glutamate receptor activation is binding of the agonist to the LBD. Glycine, d-serine, aspartate, and glutamate analogs are agonists and uniformly contain moieties that correspond to the α-amino and α-carboxyl groups. The regions of the binding pocket that form atomic interactions with the α-carboxyl and the α-amino groups are similar in all LBD structures and are composed primarily of residues from D1 (Fig. 2; see also section V.A). Crystallography studies together with homology modeling of the AMPA receptor subunits GluA1 to GluA4 show that residues that directly interact with α-carboxyl and α-amino groups of glutamate, AMPA, and kainate are conserved (Armstrong et al., 1998; Armstrong and Gouaux, 2000; Bjerrum et al., 2003, Pentikäinen et al., 2003; Gill et al., 2008) (Fig. 2). Greater variation is observed for the binding mode of the γ-positioned groups among AMPA receptor agonists, with a variety of atomic contacts being made for different agonists. Residues lining the agonist binding cavities of kainate receptor subunits are not fully conserved, providing opportunities for the development of subunit-selective agonists (Mayer, 2005) (Fig. 2; discussed in section V). Residues that interact with agonists in the NMDA receptor GluN2 subunits are fully conserved (Anson et al., 1998; Laube et al., 2004; Chen et al., 2005; Hansen et al., 2005a; Kinarsky et al., 2005; Erreger et al., 2007). As expected from sequence alignments, the agonist binding pockets of GluN1 and GluN3 are similar to those of GluA2 and GluN2A, but several key differences suggest how these subunits discriminate between glutamate and glycine (discussed in section V). Glutamate receptor activation involves a conformational change of the LBD upon binding of the agonist. Direct structural evidence for this idea arose from comparison of GluA2 LBD structures with and without agonist bound, as well as structures with bound competitive antagonists (Armstrong and Gouaux, 2000). In the antagonist-bound and the unbound apo structures, D1 and D2 are separated and adopt a more open conformation than in the agonist-bound structure, where D1 and D2 adopt a closed conformation (see also section VII.B for more detail). This mechanism is likely to be conserved in all glutamate receptor subunits, because all agonist-bound LBDs examined so far adopt conformations that are closed to different degrees relative to the apo structure.
Agonist-induced cleft closure within the LBD dimer, arranged with 2-fold symmetry in a back-to-back fashion, is an early conformational event that triggers the subsequent transition of the ion channel domain into an open state (see section VII). The intersubunit D1-D1 contacts formed across the dimer interface create both monovalent and divalent ion binding sites as well as sites for drug-like allosteric modulators (see section VI). In brief, upon agonist binding, the D2 lobes move and probably trigger rearrangement of the short segments that link the ion channel-containing TMD to the LBD, which drive rearrangement of M3 and subsequent channel opening (Erreger et al., 2004; Mayer, 2006; Hansen et al., 2007) (Fig. 3 discussed in section VII). The movement of D1 and D2 relative to each other results in instability at the TMD and at the LBD dimer interface. Stability can be restored by LBD reopening, which is the first step in the process of agonist dissociation, and we assume that it must be preceded by channel closure (or a change in subconductance state). Alternatively, the reduced stability of the interactions at the LBD dimer interface upon agonist binding can lead to a rearrangement of the dimer interface, allowing the receptor to enter a desensitized state (Sun et al., 2002; Jin et al., 2003., 2005; Horning and Mayer, 2004; Armstrong et al., 2006; Weston et al., 2006b) (Fig. 3; see section VII).
Alternative splicing of the AMPA receptor subunits generates two isoforms of the LBD termed flip and flop (Sommer et al., 1990), which control desensitization and deactivation as well as sensitivity to allosteric modulators (Mosbacher et al., 1994; Partin et al., 1994, 1995). The growing list of structures for LBDs from all subfamilies in complex with different agonists provides a firm basis for understanding agonist selectivity. For several of these ligands, structural studies in combination with site-directed mutagenesis and homology modeling have provided the structural determinants within the binding pocket that guide subunit selectivity (see section V).
E. The Extracellular Amino-Terminal Domain
Beginning at the extracellular ATD, all glutamate receptors contain a short signal peptide (14–33 residues) that targets the protein to the membrane and is removed by proteolysis after membrane insertion. Subsequent to the signal sequence, the first ∼400 to 450 residues in all glutamate receptor subunits (except bacterial GluR0, which lacks the ATD) fold into a semiautonomous domain. Glutamate receptor ATDs have sequence homology and are structurally similar to the LBD of the metabotropic glutamate receptor mGluR1a and a group of soluble bacterial periplasmatic amino acid binding proteins, such as the leucine/isoleucine/valine binding protein (O'Hara et al., 1993; Paas et al., 1996; Paas, 1998; Masuko et al., 1999a; Paoletti et al., 2000; Clayton et al., 2009; Jin et al., 2009; Karakas et al., 2009; Kumar et al., 2009). However, the similarity is confounded by numerous structural differences, such as the different locations of disulfide bonds, as well as inserts and deletions. Nonetheless, the similarity between the glutamate receptor ATD and these proteins suggests that the function of the ATD could be to bind endogenous ligands, perhaps within a putative pocket located between the lobes. Numerous mutant subunits have been created that lack the entire ATD (Fayyazuddin et al., 2000; Pasternack et al., 2002; Horning and Mayer, 2004; Matsuda et al., 2005; Rachline et al., 2005; Gielen et al., 2009; Yuan et al., 2009a), and these truncated subunits seem to assemble into receptors that are functionally similar to wild-type receptors. The nonessential nature of the ATD for the core function of the glutamate receptors is consistent with a regulatory role for this domain. Truncations of the ATD have been found to influence open probability, deactivation, desensitization, and regulation of subunit-specific assembly (Kuusinen et al., 1999; Leuschner and Hoch, 1999; Ayalon and Stern-Bach, 2001; Meddows et al., 2001; Ayalon et al., 2005; Gielen et al., 2009; Yuan et al., 2009a). The ATD also harbors binding sites for divalent cations, such as Zn2+, and subunit-selective negative allosteric modulators, such as the phenylethanolamine ifenprodil (see sections V and VI). In addition, the ATD may contain binding sites for extracellular proteins, such as N-cadherin (Saglietti et al., 2007) and neuronal pentraxins (NARP and NP1) for AMPA receptors (O'Brien et al., 1999; Sia et al., 2007) the ephrin receptor for NMDA receptors (Dalva et al., 2000; Takasu et al., 2002); cerebellin1 precursor protein for GluD2 receptors (Matsuda et al., 2010; Uemura et al., 2010; see also Uemura and Mishina, 2008; Kakegawa et al., 2009).
The glutamate receptors are glycosylated during their passage through the endoplasmic reticulum and Golgi. The consensus sites for N-linked glycosylation primarily are located in the ATD, but a few are located in the LBD (Hollmann et al., 1994; Standley and Baudry, 2000). It is not clear how many of these consensus sites are glycosylated, but cell-specific differences in glycosylation of the glutamate receptor subunits might contribute to the differences in ligand affinities, trafficking, and molecular weights observed between different native receptors and those expressed in heterologous systems (Chazot et al., 1995; Sydow et al., 1996; Everts et al., 1997; Standley et al., 1998; Standley and Baudry, 2000; Clayton et al., 2009: Kumar et al., 2009). Although the effects of glycosylation on glutamate receptor function have not been studied in detail, glycosylation can affect desensitization and maximal currents of AMPA and kainate receptors (Hollmann et al., 1994; Everts et al., 1997). In addition, the lectin concanavalin A (con A) inhibits desensitization of kainate receptors in a manner that involves association of con A with the N-linked oligosaccharides (Partin et al., 1993; Everts et al., 1997, 1999) (see section VI).
Like the glutamate receptor LBDs, the GluN2B ATD is a clamshell-like structure, roughly composed of two halves (R1 and R2) tethered together by loops (Karakas et al., 2009). The N terminus is located at the top of R1, and the linker to the LBD is located at the bottom of R2. Overall, the GluN2B ATD structure resembles the ligand binding domain of the metabotropic glutamate receptor mGluR1 (Kunishima et al., 2000), although the position of R1 is 50° twisted relative to R2 in GluN2B ATD compared with mGluR1. The cleft between R1 and R2 can be divided into three sites: 1) the hydrophilic pocket at the outer end of the cleft, which contains polar residues involved in Zn2+ binding; 2) the hydrophobic pocket deep inside the cleft, which contains residues that seem to affect ifenprodil binding; and 3) the ion-binding site that accommodates Na+ and Cl− ions with unknown physiological relevance. Binding of ifenprodil to GluN2B and Zn2+ to GluN2A or GluN2B has been proposed to stabilize a closed-cleft conformation of the ATD (see section VI; Karakas et al., 2009), although structural data in support of the hypothesized intracleft binding site is lacking. Nevertheless, the proposed cleft-closure has been speculated to lead to separation of the two R2 lobes in the ATD dimer (Gielen et al., 2008, 2009).
In contrast to NMDA receptors, no ions or small molecules are known to bind to the AMPA or kainate receptor ATD. Crystal structures of the GluA2 and the GluK2 ATDs show that these AMPA and kainate receptor ATDs adopt an overall structure similar to that of the ATD from the NMDA receptor subunit GluN2B, but the twist between R1 and R2 in GluN2B ATD was less pronounced in GluA2 and GluK2 ATDs (Clayton et al., 2009; Jin et al., 2009; Kumar et al., 2009). Unlike GluN2B, the isolated GluA2 and the GluK2 ATDs form dimers in solution and in the crystal lattice. Likewise, the ATDs of GluA1 and GluA4 also form dimers in solution (Kuusinen et al., 1999; Wells et al., 2001b; Jin et al., 2009).
Comparison of the R1 and R2 lobes of GluA2 and GluK2 ATDs with the corresponding domains of mGluR1 shows that the GluA2 and GluK2 ATDs adopt a conformation that is intermediate between the canonical open-cleft and closed-cleft states of mGluR1. In addition, there are extensive interactions between the two ATD subunits of the dimer for both GluA2 and GluK2 that involve multiple R1-R1 and R2-R2 domain contacts (Clayton et al., 2009; Jin et al., 2009; Kumar et al., 2009). The extensive interactions between the R2 lobes are mostly hydrophobic contacts situated in a large patch that is buried after dimerization of the ATD. The residues in this hydrophobic patch are conserved or conservatively substituted between AMPA and kainate receptors. In NMDA receptors, the sequence conservation is lower at the R2-R2 interface, consistent with the idea that binding of modulators to the NMDA receptor ATD could stabilize ATD cleft closure and separation at the R2-R2 interface (Gielen et al., 2008, 2009). A separation at the R2-R2 interface in the non-NMDA receptor ATD dimer would expose the large hydrophobic patch on the R2 lobe to the solvent, which would be energetically unfavorable. The “weak” R2-R2 interface in the NMDA receptor could better allow closure of the R1-R2 clamshell and separation at the R2-R2 interface, thereby triggering allosteric modulation of the ion channel. Solution of dimeric forms of the ATD will help clarify these ideas.
F. The Transmembrane Domain
In all glutamate receptors, the LBD is connected to the conserved TMD through three short linkers (Fig. 1A). The transmembrane helices M1, M3, and M4 from each of the four subunits contribute to formation of the core of the ion channel and have a small but significant sequence homology with the inverted ion channel domain of K+ channels (Wo and Oswald, 1995; Kuner et al., 2003). This similarity is further highlighted by the bacterial glutamate receptor, GluR0, which shares strong functional and structural homology with the mammalian glutamate receptors and is a potassium-selective channel with inverted topology compared with the mammalian glutamate receptors (Chen et al., 1999a). The permeation properties of GluA2-containing AMPA receptors and GluK1 and GluK2 kainate receptors are modified post-transcriptionally by RNA editing at the Gln codon that resides at the apex of the re-entrant M2 loop (QRN site). The glutamine within the QRN site is converted to arginine by adenosine deaminase (Sommer et al., 1991; Bass, 2002). For GluA2, the overwhelming majority of RNA is edited. AMPA or kainate receptors that contain the unedited form of GluA2 (Q) have high permeability to Ca2+ and are insensitive to extracellular and intracellular polyamine channel blockers, whereas AMPA receptors containing the edited form of GluA2 (R) have low Ca2+ permeability and are insensitive to polyamine channel blockers (see section VIII.C). It is noteworthy that the extended region of the M2 loop in the new GluA2 structure that encompasses the QRN site is disordered. It is unclear whether this reflects crystallization conditions or a native conformation, which might have significant functional consequences for ion permeation and block.
The structure of the antagonist-bound tetrameric rat GluA2 shows that the four subunits arrange their TMDs in a 4-fold axis of symmetry with the core of the ion channel (M1–M3), strikingly similar to K+ channels (Sobolevsky et al., 2009) (Fig. 1C). The M2 loop lines the inner cavity of the pore, whereas the M3 helices line the outer cavity, with positions at the apex tightly opposed, presumably forming the gate that occludes the flux of ions in the closed state (see sections VII and VIII). The M1 helix is positioned on the exterior of M2 and M3. It is noteworthy that the M4 segment from one subunit is associated with the ion channel core (M1-M3) of an adjacent subunit. In addition, the linker region preceding M1 (pre-M1) makes a short helix that is oriented parallel to the plane of the membrane, making contacts with carboxyl- and amino-terminal ends of transmembrane helices M3 and M4, respectively. The pre-M1s from the four subunits resemble a cuff around the external surface of the ion channel pore that could be an important determinant for channel gating (see section VII).
G. The Intracellular Carboxyl-Terminal Domain and Protein Binding Partners
The CTD is the most diverse domain in terms of amino acid sequence (Table 2), varying in sequence and in length among the glutamate receptor subunits (Figs. 5⇓–7). It shows no sequence homology to any known proteins but encodes short docking motifs for intracellular binding proteins. No structural details exist for this domain except for part of the GluN1 CTD with bound Ca2+/calmodulin (Ataman et al., 2007). The CTD is thought to influence membrane targeting, stabilization, post-translational modifications (see section IV), and targeting for degradation. For some glutamate receptor subunits (e.g., GluN1, GluN2A), deletion of this domain does not abolish function but does alter regulation (Köhr and Seeburg, 1996; Ehlers et al., 1998; Krupp et al., 1998; Vissel et al., 2001), because the CTDs contain different phosphorylation sites (see section IV) and binding sites for intracellular proteins important for regulation of membrane trafficking and receptor function. Several ER retention signals reside in alternatively spliced exons of GluN1, as well as in GluN2B (Horak and Wenthold, 2009). It is noteworthy that there is also a short span of sequence immediately C-terminal to M4 in GluN2 that also participates in trafficking (Hawkins et al., 2004).
Virtually all members of the glutamate receptor family bind to a variety of intracellular proteins, which fall into several classes. Tables 3 and 4 contain noncomprehensive lists that summarize some of the better known interactions between glutamate receptor C-terminal and PDZ, cytoskeletal, scaffolding, adaptor, anchoring, structural, signaling, and other proteins. In addition to these interactions, several glutamate receptor subunits bind directly to signaling proteins, including GluA1 and cGMP-dependent protein kinase II (Serulle et al., 2007), GluA4 and PKC (Correia et al., 2003), multiple NMDA receptor subunits and Ca2+/calmodulin-dependent protein kinase (CamK) II (Gardoni et al., 1998; Strack and Colbran, 1998; Leonard et al.,, 1999, 2002), as well as tyrosine phosphatase and GluD2 (Hironaka et al., 2000). These interactions allow local signaling to proceed, providing the possibility of spatial and temporal specificity to receptor regulation. Additional localization of signaling molecules can be mediated by adjacent proteins, and the glutamate receptors are embedded into a rich complex of signaling molecules that are localized by a myriad of adaptor and scaffolding proteins within the post synaptic density (Husi et al., 2000). Further enhancing the complexity among different subunits, alternative RNA splicing of several AMPA and kainate receptor subunits as well as the NMDA receptor subunit GluN1 causes variation in the CTD that also will affect binding sites for intracellular proteins.
H. Transmembrane α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic Acid Receptor Regulatory Proteins and other Auxiliary Subunits
A confound in the study of AMPA receptor biophysical properties has been the occasional lack of congruence between the properties of recombinant receptors expressed in heterologous systems and those of native receptors studied in isolated tissue. This mismatch suggests that heterologously expressed receptors lack a modulatory component that can influence essential properties. The discovery of the interaction between AMPA receptor subunits and the transmembrane AMPA receptor regulatory proteins (TARPs) has solved many of these discrepancies. TARPs are integral membrane proteins with four transmembrane domains (Letts et al., 1998; Hashimoto et al., 1999; Chen et al., 2000; Tomita et al., 2003; Coombs and Cull-Candy, 2009) that selectively interact with AMPA receptors early in synthesis and trafficking and direct proper expression and localization of the receptor at the cell surface (Hashimoto et al., 1999; Chen et al., 2000; Schnell et al., 2002; Tomita et al., 2004, 2005a; Vandenberghe et al., 2005). TARPs are present in the majority of AMPA receptor complexes in the brain, suggesting that TARPs are auxiliary subunits for native AMPA receptors (Fukata et al., 2005; Nakagawa et al., 2005, 2006; Vandenberghe et al., 2005). It has been suggested that two or four TARPs can associate with the AMPA receptor tetramer, depending on availability (Vandenberghe et al., 2005; Milstein et al., 2007; Shi et al., 2009). The interaction sites between TARPs and AMPA receptors involve intracellular, transmembrane, and extracellular regions of both proteins (Tomita et al., 2005a, 2007; Bedoukian et al., 2006; Milstein and Nicoll, 2009; Sager et al., 2009).
The classic members of TARPs, γ-2, γ-3, γ-4, and γ-8, interact with all four AMPA receptor subunits. The prototypical TARP (γ-2 or stargazin) originally was identified in the cerebellum as a protein essential for delivery of AMPA receptors to the plasma membrane (Letts et al., 1998; Chen et al., 2000). The functional properties of AMPA receptors associated with the TARP subtypes γ-2, γ-3, γ-4, and γ-8 are different from those of AMPA receptors devoid of auxiliary subunits. In addition to their trafficking capabilities, TARPs increase single channel conductance, increase open probability, increase the activation rate, slow the deactivation time course, and reduce desensitization (Yamazaki et al., 2004; Priel et al., 2005; Tomita et al., 2005a; Turetsky et al., 2005; Zhang et al., 2006b; Kato et al., 2007; Soto et al., 2007, 2009). Prolonged activation of AMPA receptors triggers a form of desensitization that results from dissociation of the TARP, potentially providing a novel mechanism for receptor tuning (Morimoto-Tomita et al., 2009). Finally, γ-2 reduces GluA2-lacking AMPA receptor affinity for polyamine block, resulting in receptors with weak inward-rectification (Soto et al., 2007).
The TARP γ-5 increases glutamate potency and glutamate-evoked peak currents, reduces steady-state currents, and accelerates the time course of deactivation and desensitization only in GluA2-containing receptors, but this modulation does not involve regulation of GluA2 surface expression (Kato et al., 2008). The TARP γ-7 shares many of these properties but is not selective for receptors comprising GluA2-containing subunits, enhancing peak currents in channels containing GluA1 or GluA2 (Kato et al., 2007). In contrast, other studies show that γ-5 interacts with all AMPA receptor subunits and modifies their behavior (Soto et al., 2009).
Proteins with homology to TARP (STG-1 and STG-2) and with similar functional roles have been discovered in Caenorhabditis elegans, Apis mellifera, and Drosophila melanogaster (Walker et al., 2006a; Wang et al., 2008). However, an important difference between TARPs and STG-1 and STG-2 is the obligatory requirement of an additional transmembrane auxiliary subunit (SOL-1) that is structurally unrelated to TARPs and interacts directly with invertebrate AMPA receptor subunits (GLR-1 and GLR-2) to slow and reduce the extent of receptor desensitization (Zheng et al., 2006; Walker et al., 2006a, b).
Another distinct class of transmembrane proteins has been shown to assemble with and regulate AMPA receptors (Schwenk et al., 2009). These proteins, CNIH-2 and CNIH-3, are members of the mammalian CNIH family and are homologous to the cornichon proteins from flies and yeast (Roth et al., 1995; Bökel et al., 2006; Castro et al., 2007). CNIH proteins are necessary for the export of a number of proteins from the endoplasmic reticulum, including the epidermal growth factor receptor ligands. The role of the CNIH proteins in AMPA receptor regulation is not yet fully understood.
Accessory proteins for NMDA and kainate receptors have also recently been described. Neto1 possesses a single transmembrane domain containing two complement C1r/C1s, Uegf, Bmp1 domains and is a component of the NMDA receptor complex (Ng et al., 2009). Neto1 interacts with an extracellular domain of GluN2 as well as through an intracellular interaction with PSD95. Loss of Neto1 in transgenic mice preferentially results in a loss of synaptic GluN2A expression, with only a modest impact on GluN2B expression, which leads to impaired hippocampal LTP and hippocampal-dependent learning and memory (Ng et al., 2009). A second C1r/C1s, Uegf, Bmp1 domain-containing protein, Neto2, interacts with GluK2 to increase peak amplitude and open probability and to slow the decay time course of both GluK2 recombinant receptors and kainate receptor-mediated mEPSCs in cerebellar granule cells (Zhang et al., 2009c). In recombinant expression systems, Neto2 had no impact on GluK2 surface expression.
III. Regulation of Transcription and Translation
The level of expressed glutamate receptors reflects a balance of transcription, translation, mRNA level, protein stability, receptor assembly, and presentation at the cell surface, all of which are integrated through numerous environmental stimuli. Therefore, the particular subunits that each neuron chooses to express are strong determinants of synaptic phenotype, and this is the rationale for understanding how the genetic cis elements and trans factors regulate gene transcription in neural cells. Over the past decade steady work toward understanding the control of ionotropic glutamate expression in neuronal and non-neuronal cells has occurred, roughly doubling both the number of subunits studied and the identification of promoter elements controlling expression in neuronal cells. Furthermore, how chromatin remodeling affects glutamate receptor expression in both neurons and non-neuronal cells has been identified after, for example, status epilepticus or transient ischemia. Most studies have employed a combination of protein-DNA binding assays with functional analysis of native and mutant promoter constructs driving a reporter gene, overexpression of candidate transcription factors in cultured cells or in vivo, occupancy of cis elements by transcription factors in vivo using chromatin immunoprecipitation (ChIP) assays, and the use of real-time quantitative polymerase chain reaction experiments. The use of ChIP assays and real-time quantitative polymerase chain reaction on endogenous gene transcripts (and also on exogenously expressed constructs) have been particularly fruitful in helping to advance our understanding of how an acute stimulus causes a change in transcript level dependent on candidate promoter elements and trans-acting factors. Studies such as these have begun to tie neuronal activity, energy metabolism, and glutamate receptor expression together more coherently. Furthermore, an appreciation for the role of epigenetic modifications and chromatin remodeling at glutamate receptor promoters is also emerging and holds promise for new understanding of the neurobiology of glutamate receptors. Despite this progress, vexing questions still remain regarding the mechanisms that control cell-specific and developmental expression for glutamate receptor subunits.
Glutamate receptor genes have a number of features in common, such as multiple transcriptional start sites in a TATAA-less promoter with high GC content. The 5′UTR ranges between 200 bp (Gria1) to over 1200 bp (Grin2a) and, in the case of Gria4, Grik5, Grin2a, Grin2b, and Grin2c, is formed from multiple exons. Finally, one or more Sp1 elements reside near the major transcriptional start site of all genes studied, several glutamate receptor promoters contain NFκB, CRE, AP1/2, Tbr-1, NRF-1, and RE1/NRSE sites, and gene expression of many is responsive to neuronal activity. The schematic organization of the promoter regions is presented in Fig. 4. It also should be noted that NMDA, AMPA, and kainate receptors are key mediators of signal transduction events that convert environmental stimuli into genetic changes through regulation of gene transcription and epigenetic chromatin remodeling in neural cells, an area of emerging interest (Carrasco and Hidalgo, 2006; Wang et al., 2007; Cohen and Greenberg, 2008; Lubin et al., 2008).
A. α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic Acid Receptors
1. Gria1.
The rat Gria1 gene has been evaluated for transcription initiation and regulation of promoter function by transfection of constructs into primary mixed neuronal cultures. Gria1 promoter activity was stronger in neurons, neuronal specificity being primarily dependent on sequences lying within the regions −1395 to −743 and −253 to −48. Although five CRE sites were able to bind recombinant CRE binding-bZIP proteins, conditions under which these CRE sites come into play in native neurons are unknown. GluA1 receptors (like all AMPA receptor subunits) are expressed by neurons and glial cells in vivo and in vitro (Gallo and Ghiani, 2000), but the density of functional receptors is much lower in astrocytes than neurons. In oligodendrocyte progenitor O-2A cells, the transcriptional rate of GluA1 is increased by platelet-derived growth factor and basic fibroblast growth factor (Chew et al., 1997). Regulation of Gria1 transcription occurs via acid sphingomyelinase and NFκB sites in the Gria1 promoter (Borges and Dingledine, 2001), which was found to both account for the elevation of GluA1 by tumor necrosis factor-α (Yu et al., 2002) and contribute to the sensitization by tumor necrosis factor-α of NT2-N cells to kainate-induced cell death.
2. Gria2.
The relative use of 5′ transcriptional start sites in Gria2 is different in cortex and cerebellum, longer transcripts being more dominant in cerebellum (Myers et al., 1998). Analysis of Gria2 promoter constructs in cultured forebrain neurons and glia revealed that the Gria2 promoter was 30-fold neuronal selective. In this study, Sp1 and NRF-1 positive regulatory elements, nestled at the 5′ end of a 141-bp transcription initiation region, and an RE1/NRSE proximal promoter silencer element were important for neuron-selective expression. The RE1/NRSE repressed expression 2- to 3-fold in non-neuronal cells compared with neuronal cells (Myers et al., 1998). Suppression of GluA2 expression in glia occurs by occupancy of the Gria2 RE1/NRSE element by the REST/NRSF repressor, which in turn recruits histone deacetylase (HDAC) complexes to the Gria2 promoter, resulting in chromatin remodeling and decreased expression (Huang et al., 1999). In neurons, the Gria2 promoter is associated with acetylated H3 and H4 histones, whereas in C6 glioma cells, there is little to no association with acetylated histones, consistent with active and inactive gene expression, respectively (Huang et al., 1999). After induction of status epilepticus by pilocarpine, the acetylation of histone H4 bound to the Gria2 promoter was reduced before GluA2 expression became down-regulated in rat hippocampal CA3 neurons. Seizure-induced GluA2 mRNA down-regulation was reversed by the HDAC inhibitor trichostatin A (Huang et al., 2002a). Likewise, Calderone et al. (2003) showed that global ischemia triggers expression of the repressor REST and reduces GluA2 expression in CA1 neurons destined to die. Moreover, kainate reduces activity of the neuronal Gria2 promoter in a manner consistent with REST occupancy of the RE1 element, recruitment of HDAC to the promoter, and reduced histone acetylation (Jia et al., 2006). It is noteworthy that a preconditioning sublethal ischemic episode can prevent subsequent ischemia-induced down-regulation of GluA2 in CA1 neurons (Tanaka et al., 2002) by preventing an increase in REST expression in these same neurons (Calderone et al., 2003).
TTX reduces GluA2 expression in visual cortical neurons, suggesting that expression is linked to neuronal activity (Wong-Riley and Jacobs 2002; Bai and Wong-Riley 2003). The transcription factor NRF-1 binds to the Gria2 promoter (Dhar et al., 2009), confirming earlier studies identifying the NRF-1 element as a critical feature for neuronal Gria2 transcription (Myers et al., 1998). NRF-1 is a nuclear transcription factor important for regulating multiple cytochrome oxidase (COX) genes. Reduction of NRF-1 with small hairpin RNA prevented the depolarization-stimulated increase of GluA2 expression, whereas overexpression of NRF-1 restored GluA2 expression in the face of TTX treatment. Changes in GluA1, GluA3, and GluA4 expression were not observed, rendering NRF-1 control specific to GluA2 (Dhar et al., 2009). Thus, neuronal activity is tightly coupled at the molecular level to GluA2 expression by a process that involves NRF-1.
3. Gria3 and Gria4.
The human GRIA3 gene is on the long arm of the X chromosome and is subject to X inactivation through methylation (Gécz et al., 1999). The Gria3 transcript contains a TC repeat in the 5′-UTR that is polymorphic and is also present in rodent transcripts. The 5′-UTR and ATG codon are contained in a large 1102-bp exon 1 and are conserved in rodent and human sequences. Neither an RE1 repressor element nor an NRF-1 site was identified.
GluA4 is widely expressed in brain; however, its abundance is less than GluA1–3 (Petralia and Wenthold, 1992). In transfected mixed cortical cultures, Gria4 promoter constructs drove luciferase expression predominantly in neurons, indicating a 6- to 12-fold neuronal preference (Borges et al., 2003). Deletion of the Gria4 transcriptional initiation region decreased luciferase activity in neurons, but increased activity in C6 cells, suggesting that neuronal regulatory elements reside in this region. Sp1, Ikaros, and basic helix-loop-helix binding element sites are conserved in rat, mouse, and human genes within ±150 bp of transcription initiation sites; however, specific evaluation of these elements requires further investigation. A distal region of Gria4, −4427 to −4885, is in a long interspersed element sequence that has been suggested to recruit chromatin remodeling enzymes to the Gria4 gene (Borges et al., 2003).
B. Kainate Receptors Grik1 to Grik5
Regulatory cis elements have been computationally predicted in promoter regions of human GRIK1 and GRIK2 genes but not functionally evaluated (Barbon and Barlati, 2000; Barbon et al., 2001). However, GRIK2 was identified as a novel epigenetic target in gastric cancer as a potential tumor suppressor gene (Wu et al., 2010). There are no reports on the functional evaluation of transcriptional regulation of the Grik3 and Grik4 genes.
Initial studies of the rat Grik5 gene identified a negative regulatory sequence in the first intron that binds nuclear orphan receptors such as chicken ovalbumin upstream promoter transcription factor I in both neural and non-neural cells (Huang and Gallo, 1997; Chew et al., 1999). Transgenic mouse lines carrying 4 kb of the 5′-flanking sequence showed lacZ reporter expression predominantly in the nervous system. Reporter assays in central glial (CG-4) and non-neural cells indicated that a 1200-bp 5′-flanking region could sustain neural cell-specific promoter activity (Chew et al., 2001). Sp1 binding suggests a functional role for Sp1 in initiator-mediated activation of Grik5 transcription that involves transcription factor II D-mediated basal activity (Chew et al., 2001). Removal of two putative AP2 sequences reduced promoter activity in both neural and non-neural cells, suggesting that these sites are also important for basal transcription. Furthermore, a 77-bp sequence termed the kainate cell-specific enhancer region, involved in cell-specific expression, includes a functional Sp1 site that when placed downstream of the Grik5 promoter, silenced reporter expression in NIH3T3 fibroblasts and attenuated activity in CG-4 cells. These studies show that elements contributing to tissue-specific expression are contained within the first exon (Chew et al., 2001).
C. N-Methyl-d-aspartate Receptors
1. Grin1.
Rat, human, and chicken Grin1 promoter regions have been cloned and characterized (Bai and Kusiak, 1993, 1995; Zimmer et al., 1995; Moreno-González et al., 2008). Transcription of the Grin1 gene is controlled by both positive and negative regulatory elements. A consensus RE1/NRSE silencer in exon1 contributes to neuronal-specific expression (Bai et al., 1998; Okamoto et al., 1999). Ablation of the RE1/NRSE site elevated GluN1 expression in non-neural cell lines and undifferentiated P19 cells (Bai et al., 1998, 2003; Okamoto et al., 1999, 2002). Likewise, during differentiation of P19 cells, REST/NRSF is down-regulated, resulting in de-repression of the Grin1 promoter (Okamoto et al., 1999). De-repression of the Grin1 promoter by absence of REST/NRSF occurs before subsequent expression of positive acting trans factors required for full Grin1 promoter activity (Bai et al., 2003). A 27-bp GC-rich region (GC-box) proximal to the transcription start sites has been identified that controls induction of the Grin1 gene upon differentiation of P19 cells, and this site is recognized by Sp1 and myc-associated zinc finger protein transcription factors (Okamoto et al., 2002). These sites previously were known to respond to Sp1, -3, and -4 transcription factors (Bai and Kusiak, 1995, 1997; Bai et al., 1998; Liu et al., 2001) and interact with an element further 5′ in the promoter (−520/−529) that is recognized by myocyte enhancer factor 2C (Krainc et al., 1998). Studies with Grin1 promoter constructs in PC12 cells suggest that NGF uses both the Ras/extracellular signal-regulated kinase (ERK) and phosphatidylinositol 3-kinase pathways to up-regulate Grin1 promoter activity through Sp1 (Liu et al., 2001). Activation of serum glucocorticoid kinase 1, a downstream target of phosphatidylinositol 3-kinase, increases Grin1 promoter activity in PC12 cells and hippocampal neurons in an NFκB-independent manner (Tai et al., 2009). This finding is consistent with a previous report that the Sp-related factors regulate Grin1 promoter activity through occupancy of a putative NFκB consensus element ∼3 kb upstream of the GC box in neurons and in cell lines (Liu et al., 2004a).
AP1 protein complexes containing ΔFosB bind the rat Grin1 promoter at the AP1 consensus element, and AP1 binding is up-regulated after electroconvulsive shocks. Furthermore, an increase in Fos-like immunoreactivity was observed in the same cortical neurons that showed an increase in GluN1 immunoreactivity. Accordingly, up-regulation of GluN1 did not occur after seizures in fosB(−/−) mice (Hiroi et al., 1998).
GluN1 expression might be coupled to energy metabolism based on evidence that both the Grin1 and mitochondrial COX genes are under control of the NRF-1 transcription factor via binding elements in their respective proximal promoter regions (Dhar et al., 2008, Dhar and Wong-Riley, 2009). Grin1 and Grin2b, but not Grin2a or Grin3a, are positively regulated by the NRF-1 transcription factor through NRF-1 promoter elements. Furthermore, control of Grin1 and Grin2b by NRF-1 was activity-dependent. KCl up-regulated and TTX down-regulated expression in cultured rat cortical neurons, and NRF-1 itself is up-regulated at both protein and mRNA levels by depolarization (Yang et al., 2006; Dhar and Wong-Riley, 2009). Thus, NRF-1 is an essential transcription factor in the coregulation of Grin1, Grin2b, Gria2, and COX genes, coupling coordinated expression of glutamate receptors and energy metabolism at the transcriptional level (Wong-Riley et al., 1998a,b; Dhar et al., 2008, 2009; Dhar and Wong-Riley, 2009).
A nonpalindromic T-box element in the Grin1 promoter is likely to be recognized and regulated by Tbr-1/CASK protein complexes in vivo because GluN1 expression was reduced in Tbr-1(−/−) (T-brain-1) mice by ∼ 50% (Wang et al., 2004b). Tbr-1 is a neuron-specific T-box factor (Hsueh et al., 2000) that may play a role in neurogenesis and induction of GluN1. GluN1 expression is subject to control by hypoxia-inducible factors that function under stress conditions, especially during hypoxia (Yeh et al., 2008). Based on GluN1 up-regulation after lipopolysaccharide injection into the prefrontal cortex and in cultured neurons, the predicted cis hypoxia response elements were localized within the Grin1 promoter. However, 1 h of ischemia produced by middle cerebral artery occlusion decreased GluN1 (Gascón et al., 2005), possibly as a result of activation of the RE1 silencer (Fig. 4).
2. Grin2a.
Regulation of the rat Grin2a promoter was explored with a series of 3′- and 5′-truncated constructs in primary neurons, primary glia, and non-neuronal cell lines (Desai et al., 2002; Richter et al., 2002; Liu et al., 2003). The core promoter resides in exon 1, prefers neurons but also requires downstream sequences for full activity, and does not use a consensus TATA box. On the basis of overexpression studies and gel shift assays in stable cell lines, three GC-boxes (A, B, and C) seem to regulate Sp1 and Sp4 but not Sp3 transactivation (Liu et al., 2003).
Two regions of the mouse Grin2a promoter, from −9.2kb/−210 or −1253/−210, were able to confer nervous system expression of a transgene reporter. Based on primary cultures prepared from a −9.2kb/−210 Grin2a luciferase mouse, there was ∼700-fold selective expression in neuronal enriched cultures compared with glial cultures. Two RE1/NRSE-like sequences that contain key mismatches in the consensus sequences were identified at −989 and −427 and do not seem to act as silencers of the Grin2a gene. Thus, neuronal specificity for GluN2A expression seems to result from transcriptional activation selectively in neurons rather than by non-neuronal silencing. Furthermore, three regions were identified (−1253/−1079, −486/−447, 8 kb 5′ of −1253) that are important for maximal neuron-selective expression. Sequences between −9.2 kb and −1253 bp contribute to the maturational increase of Grin2a expression in cultured neurons and elements residing between −1253 and −1180 bp are crucial for this up-regulation (Desai et al., 2002).
Two NFκB sites were identified in the Grin2a promoter, which, when removed by mutation, resulted in loss of modification of transactivation by constitutively active SGK (SGK-S422D) that activates NFκB (Tai et al., 2009). Furthermore, the transactivation of a Grin2a construct was sensitive to the NFκB inhibitor peptide SN50 (Lin et al., 1995). A putative CRE element variant found in numerous promoters was identified at −1195 in mouse and −1215 in rat and raises interest in activity-dependent elevation of Grin2a in vivo. The putative CRE site resides in a region important for positive neuronal expression in both rat and mouse promoters (Desai et al., 2002; Richter et al., 2002; Liu et al., 2003).
3. Grin2b.
Initial promoter analysis using transgene constructs in mice (Sasner and Buonanno, 1996) revealed that the proximal promoter region and exon 1 (−550/+255 relative to the 5′-most transcription site) were sufficient to restrict tissue specificity to brain. However, inclusion of intron 1 and exon 2 in the transgene (−550/+1627) were required both to restrict expression to brain and to recapitulate the proper developmental profile of GluN2B expression in cerebellar granule cells. The presence of an RE1/NRSE-like element at the end of exon 1 was not responsible for conferring neural-selective expression in the mouse transgenes (Sasner and Buonanno, 1996). Of several putative RE1/NRSE elements in the more distal Grin2b promoter, the −2029/−2049 NRSE element bound REST/NRSF and repressed expression of Grin2b reporter constructs transfected into cultured neurons. Moreover, ethanol treatment of cortical cultures reduced REST/NRSF expression, resulting in GluN2B derepression (Qiang et al., 2005). Analysis of the Grin2b promoter identified Sp1 and CRE elements (Klein et al., 1998), and the CRE site was later confirmed to bind to phospho-CREB in a gel-shift assay. Mutation of the CRE motif in the Grin2b promoter region significantly decreased promoter activity in transfected cortical cells and also abolished ethanol-induced increase in promoter activity (Rani et al., 2005). Likewise, an AP-1 site was active in cultured neurons and responsive to ethanol treatment (Qiang and Ticku, 2005). Furthermore, it was found that long-term ethanol exposure promoted demethylation of CpG islands in the Grin2b promoter region that could result in up-regulation of the gene in mouse cortical neurons (Ravindran and Ticku, 2005).
Two nonpalindromic T-box elements in the Grin2b promoter were identified that are conserved across rat, human, and mouse (Wang et al., 2004b), and these elements are recognized by the Tbr-1 protein (Wang et al., 2004b). Functional and mutational studies in rat hippocampal cultures showed that overexpression of Tbr-1 alone and in combination with CASK elevated Grin2b promoter-driven luciferase activity by up to 120-fold dependent on each T-box element with the upstream T-element dominant. Recognition of the Grin2b T-box elements by a Tbr-1–CASK complex in vivo was demonstrated by ChIP analysis using rat hippocampal cultured neurons (Wang et al., 2004a). In Tbr-1–null mice, GluN2B expression was decreased up to 60%, and GluN2B expression was down-regulated in brain regions where Tbr-1 immunoreactivity was lost in the mutant mice. These two T-box elements reside within the minimal transgene construct sufficient for neuron-specific expression of the Grin2b gene (Sasner and Buonanno, 1996). A point mutation in CASK that disrupts CASK–Tbr-1–CASK-interacting nucleosome assembly protein complexes down-regulates Grin2b promoter activity (Huang and Hsueh, 2009). CASK interacts with transcription factor Tbr-1 and CASK-interacting nucleosome assembly protein–cell division autoantigen-1–differentially expressed nucleolar transforming growth factor-β1 target in the nuclei of neurons, which may remodel the chromatin structure flanking Tbr-1 binding sites (Hsueh et al., 2000; Wang et al., 2004a).
As shown for the Grin2a promoter, activation of serum glucocorticoid kinase 1 pathway elevates Grin2b gene expression in hippocampal neurons and Neuro2A cells in an NFκB-dependent manner. The site of NFκB binding in the Grin2b promoter was not specifically identified but was proposed to reside between −1480 and −2020 bp from the rat transcription start site (Tai et al., 2009).
Stimulation of cortical cultures with bicuculline elevated GluN2B expression in a transcription- and calcineurin-dependent manner (Qiu and Ghosh, 2008) and revealed association of the Grin2b promoter with CREST, brahma-related gene 1, CRE binding protein, and HDAC-1. Bicuculline stimulation increased CRE binding protein, decreased HDAC1, and increased the association of the Grin2b promoter with acetylated histones. The increase in GluN2B expression also required NMDA receptor activation and was shown to depend on CREST in vivo because a bicuculline-induced increase in GluN2B expression was absent in CREST-null neurons. These findings suggest that the activity-dependent increase in GluN2B expression involves a switch from a repressor to activator complex and requires CREST function that may involve CRE and Sp1 sites in the promoter in a manner similar to regulation of the immediate early gene c-fos.
Evidence for the coupling of GluN2B expression to energy metabolism has also been described through a series of electrophoretic mobility shift assay, supershift, and ChIP assays and promoter mutations (Yang et al., 2006; Dhar et al., 2009). GluN2B was shown to be regulated in cultured neurons by NRF-1 transcription factor via an NRF-1 element in the proximal promoter region. GluN1, GluN2B, and NRF-1 transcripts are up-regulated by KCl and down-regulated by TTX in cultured primary neurons. Thus, NRF-1 coordinates the coregulation of Grin1, Grin2b, and COX genes (Dhar et al., 2008; Dhar and Wong-Riley, 2009).
4. Grin2c, Grin2d, Grin3a, and Grin3b.
Elements within Grin2c exon 1 and intron 1 in transgenic mouse lines selectively drive expression of β-galactosidase in cerebellar granule cells (Suchanek et al., 1995, 1997). This region contains a consensus RE1/NRSE silencer, but the role of this element has not been fully evaluated. Furthermore, Sp1 and COUP-TF consensus elements (Nagasawa et al., 1996; Pieri et al., 1999) bound Sp1 and fushi tarazu factor 1/COUP-TF protein in gel shift assays, but mutations did not modify promoter function in a transfected neuronal cell line (Pieri et al., 1999). The COUP-TF site seems to require other elements for function. Coexpression of steroidogenic factor-1 elevated Grin2c promoter activity modestly in a neuronal cell line dependent on a promoter region centered −250 from transcription + 1 site (Pieri et al., 1999); however, direct expression in neurons was not examined.
The developmental up-regulation of GluN2C in the adult cerebellum upon innervation of mossy fibers onto granule cells has been reported to be due to neuregulin-β (Ozaki et al., 1997). Neuregulin-β potently up-regulated GluN2C with no change in GluN2B expression in cultured mouse cerebellar slices; up-regulation was sensitive to block by TTX or the NMDA antagonist 2-amino-5-phosphonopentanoate, suggesting that synaptically activated NMDA receptors are involved (Ozaki et al., 1997). Activity regulates GluN2B expression in cerebellar granule cells (Vallano et al., 1996), and both GluN2B and GluN2C in organotypic cultures (Audinat et al., 1994), although the stimulatory effect of neuregulin-β on GluN2C was not recapitulated in cultures of dissociated granule cells (Rieff et al., 1999). For granule cells cultured in low (5 mM) KCl, BDNF up-regulated GluN2C mRNA via the tyrosine kinase receptor ERK1/2 cascade, whereas under 25 mM KCl, depolarization stimulated Ca2+ entry through voltage-sensitive Ca2+ channels and activated Ca2+/calmodulin-dependent calcineurin phosphatase, which opposed GluN2C mRNA up-regulation (Suzuki et al., 2005). However, the depolarization-induced Ca2+ increases simultaneously up-regulated BDNF mRNA via CaMK. Thus, convergent mechanisms of the BDNF and Ca2+ signaling cascades are important for GluN2C induction in granule cells during development (Suzuki et al., 2005). NMDA receptor activation was shown to coordinate both the up-regulation of GluN2C and the down-regulation of GluN2B mRNA, including a switch of GluN2 subunit associated with cell surface NMDA receptors in cultured mouse granule cells (Iijima et al., 2008). Although much has been learned about the signal transduction pathways leading to receptor subunit changes in granule cells, the promoter control elements responsible for transcription subunit switching remain to be identified.
In the human GRIN2D gene, the 3′-UTR contains four half-palindromic estrogen responsive elements within a 0.2-kb region that are highly preserved in the rat, suggesting that the GluN2D subunit may be up-regulated in vivo via neuroendocrine control. In ovariectomized rats, up-regulation of GluN2D mRNA in the hypothalamus upon 17β-estradiol treatment was observed (Watanabe et al., 1999), and the Grin2d half-palindromic estrogen responsive elements, placed in a 5′ or 3′-UTR position in a chloramphenicol acetyltransferase promoter construct, were responsive to estrogen and thyroid hormone exposure in an orientation- and hormone receptor-dependent manner (Watanabe et al., 1999; Vasudevan et al., 2002).
The amino acid sequences and expression profile for Grin3a and Grin3b have been reported (Ciabarra et al., 1995; Sucher et al., 1995; Andersson et al., 2001; Nishi et al., 2001; Chatterton et al., 2002; Eriksson et al., 2002; Matsuda et al., 2002; Bendel et al., 2005). Studies describing control of transcription with respect to cis regulatory elements and trans factors have not been reported.
D. Translational Control of Glutamate Receptors
The translation of mRNA to protein is regulated by mechanisms that control 5′ capping, 3′ polyadenylation, splicing, RNA editing, mRNA transport, stability, and initiation and elongation (VanDongen and VanDongen, 2004; Coyle, 2009). The 5′-UTR of most glutamate receptor mRNAs is unusually long. These long 5′-UTRs often exhibit stretches of high GC content and sometimes contain multiple out-of-frame AUG codons that could act as decoys for scanning ribosomes, reducing or preventing translation initiation at the true glutamate receptor AUG (Myers et al., 1999; VanDongen and VanDongen, 2004). Translational suppression has been inferred for GluN1 mRNA natively expressed in PC12 cells because no GluN1 protein can be detected despite a moderately high mRNA level (Sucher et al., 1993). On the basis of that study (Sucher et al., 1993), it was proposed that translation of GluN1 message may be suppressed, perhaps by an unidentified motif in the 5′- or 3′-UTR. Two pools of GluN1 mRNA with different translational activities have been identified in neonate and adult brain, further implicating potential translation control mechanisms at work in neurons (Awobuluyi et al., 2003). Whether GluN1 translation rate was determined by alternative 3′UTRs was not explored. Visual deprivation in juvenile mice reduced the GluN2A/GluN2B ratio in the deprived cortex with the finding that translation of GluN2B is probably a major regulatory mechanism (Chen and Bear, 2007).
The efficiency of translation of another subunit, GluN2A, also depends upon features of the 5′-UTR. Here, in both in vitro translation in rabbit reticulocyte lysate and the X. laevis oocyte expression system, removal of most of the 282 bases of the 5′-UTR from the cDNA increased GluN1/GluN2A protein expression by more than 100-fold. Mutation of three of the five upstream AUG codons modestly increased translation; however, disruption of a proposed GC-rich stem-loop structure 170 bases upstream of the AUG increased GluN2A translation by 40-fold (Wood et al., 1996). Translation suppression of GluN2A transcripts in vivo is one interpretation of the finding that high GluN2A mRNA levels were measured in the inferior and superior colliculus and striatum of adult rats (Goebel and Poosch, 1999), whereas immunocytochemical studies showed staining for the GluN2A protein to be light in these areas (Petralia et al., 1994).
The translation efficiencies of several GluA2 5′-UTR transcripts in rabbit reticulocyte lysates, X. laevis oocytes, and primary cultured neurons has been investigated (Myers et al., 2004). Transcripts containing long 5′ leaders were translated poorly compared with those with shorter leaders, and short transcripts were preferentially associated with polyribosomes in vivo. Suppression of GluA2 translation was dominated by a 34- to 42-nucleotide imperfect GU repeat sequence in the 5′-UTR predicted to form a secondary structure. It is noteworthy that the GU repeat domain is polymorphic in man and is included in a subset of rat and human GluA2 transcripts based on the site of transcription initiation (Myers et al., 1998, 2004). GluA2 translation was not modified significantly by deletion of any or all of the five upstream AUG codons. An interpretation of both the GluN2A and GluA2 studies is that a scanning ribosome encounters the proposed stem-loop and stalls because of an inability to “melt” the stem-loop structure; alternatively, a ribosome may encounter a blocking protein bound at the stem-loop motif. Either case could result in dissociation of the ribosome from the mRNA.
GluA2 transcripts are processed to form either a short or a long 3′-UTR giving rise to two pools of GluA2 mRNAs of 4 and 6 kb in length in brain. In the hippocampus, long 3′-UTR GluA2 transcripts are retained primarily in translationally dormant complexes of ribosome-free messenger ribonucleoprotein, whereas GluA2 transcripts bearing the short 3′UTR are associated mostly with actively translating ribosomes (Irier et al., 2009b). After pilocarpine-induced status epilepticus, selective translational derepression of GluA2 mRNA mediated by the long 3′-UTR transcripts was observed, suggesting that the long 3′-UTR of GluA2 mRNA alone is sufficient to suppress translation and that an activity-dependent regulatory signaling mechanism exists that differentially targets GluA2 transcripts with alternative 3′-UTRs (Irier et al., 2009b). Differential effects of antibiotics that target translational initiation and elongation suggest that the long 3′-UTR suppresses GluA2 at the initiation step, implying a loop-back mechanism (Irier et al., 2009a). The mechanism of translation is not known but could involve binding of a cytoplasmic polyadenylation element binding protein (CPEB). Among the CPEB proteins, CPEB3 is expressed specifically in neurons (Theis et al., 2003) and seems to bind to GluA2 long 3′-UTR. RNA interference knockdown of CPEB3 mRNA induces GluA2 protein expression in cultured hippocampal neurons (Huang et al., 2006). CPEB protein regulates translation initiation (Richter and Sonenberg, 2005), facilitates targeting of mRNAs to dendrites (Huang et al., 2003), and has been implicated in control of GluN1 translation (Wells et al., 2001a). A related report identified a deletion allele in the Grik4 gene 3′-UTR that was negatively associated with bipolar disorder, and it was proposed, on the basis of expression data, that RNA secondary structure modified mRNA stability to enhance protein expression (Pickard et al., 2008).
The localization of translation machinery near postsynaptic sites (Steward and Levy, 1982; Steward and Reeves, 1988) and differential distribution of mRNAs to dendrites, including those for GluA1, GluA2, and GluN1, have been investigated (Steward and Schuman, 2001; Schratt et al., 2004; Grooms et al., 2006). It is becoming clear that these dendritic mRNAs may form a pool poised for translation to modify neuronal plasticity. Other studies have demonstrated that protein synthesis in dendrites is critical for long-term potentiation (LTP) and long-term depression (LTD) (Kang and Schuman, 1996; Huber et al., 2000; Tang and Schuman, 2002; Bradshaw et al., 2003; Cracco et al., 2005; Mameli et al., 2007). The induction of protein synthesis is, not unexpectedly, dependent upon NMDA receptor activation (Scheetz et al., 2000; Huang et al., 2002b; Gong et al., 2006; Tran et al., 2007).
IV. Post-Translational Regulation
A. α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic Acid and Kainate Receptor Phosphorylation
Considerable advances have been made to identify dynamically regulated post-translational phosphorylation sites on the C-terminal domains of most of the glutamate receptor subunits and to understand the functions of these modifications. Phosphorylation has been shown to regulate glutamate receptor trafficking from the ER, insertion into the plasma membrane, endocytosis, synaptic localization, and binding to other proteins (Malinow and Malenka, 2002). In a few cases, phosphorylation seems to regulate the relative frequency of opening for different subconductance levels of the channels in a manner independent of trafficking. However, more work is needed to understand the mechanisms by which phosphorylation can control glutamate receptor function. In some cases, the strategic combination of specific phosphoprotein antibodies, site-directed mutagenesis, and chromophore-tagged receptors has made it possible to associate the functional consequences of kinase activation with phosphorylation of specific amino acid residues. However, the specific functional roles of many phosphorylation sites have simply not yet been explored, and in nearly all cases, the molecular mechanisms by which phosphorylation changes receptor function (membrane insertion, open probability, etc.) are unknown. These topics represent excellent opportunities for future progress.
The C-terminal domains are shown for each of the subunits in Figs. 5⇓–7, with residues that have been shown to be post-translationally modified in living cells in red; residues identified only in cell-free kinase assays are omitted. Authors have variably included the signal sequence in the numbering schemes and studied various C-terminal splice variants. Here, sites will be referred to by residue position in the literature. To minimize confusion with numbering schemes, we simply include the surrounding sequences in the text together with the most commonly used residue number. This number diverges from the numbering in Figs. 5 to 7 when authors indexed the predicted signal peptide cleavage site as 1.
The GluA1 C terminus has multiple phosphorylation sites, including four for PKC, one for PKA, and one for CAMKII. Phosphorylation of the two membrane-proximal PKC sites (SRS816ES818KR) enhances interaction between the actin-binding 4.1N protein and the GluA1 C-terminal domain, which facilitates insertion of this subunit into the plasma membrane (Boehm et al., 2006; Lin et al., 2009), a mechanism involved in long-term potentiation. Phosphorylation of TST840LPR by PKC has been suggested to influence synaptic transmission in an age-dependent fashion (Lee et al., 2007b). Two other GluA1 phosphorylation sites control functional properties of AMPA receptor channels. Phosphorylation of RNS845GA by PKA (Roche et al., 1996) increases the open probability of homomeric GluA1 channels studied in outside-out patches (Banke et al., 2000), which has been proposed to reflect a change in the equilibrium of GluA1 with an inactive state, perhaps relating to phosphodependent binding of intracellular regulatory proteins. PKA phosphorylation additionally drives GluA1 subunits into synaptic membranes (Esteban et al., 2003; Man et al., 2007). Phosphorylation of QQS831IN by CAMKII or PKC (Barria et al., 1997; Mammen et al., 1997) increases single channel conductance (Derkach et al., 1999; Oh and Derkach, 2005). In vivo evidence from transgenic animals suggests that GluA1 phosphorylation is critical for synaptic plasticity (Lee et al., 2000, 2003; Whitlock et al., 2006; Tsui and Malenka, 2006). Considerably more work will be required before we achieve adequate understanding of the relative contribution of phosphorylation-linked changes in trafficking and channel function to changes in synaptic strength (Song and Huganir, 2002; Derkach et al., 2007; Kessel and Malinow., 2009) (see sections IX.E and IX.F).
GluA2 splice variants create short and long C termini (Fig. 5), phosphorylation of which influences receptor trafficking, synaptic plasticity, and several receptor-protein interactions. The long tail has a phosphorylation site at VMT874PE that is a Jun kinase target (Thomas et al., 2008). Dephosphorylation at this site is activity-dependent and promotes reinsertion of internalized GluA2 back into the plasma membrane. Targeted disruption of the PKC recognition sequence around IES880VK eliminated LTD in mouse cerebellum (Steinberg et al., 2006), confirming a major role for this modification in synaptic plasticity. The activity-dependent phosphorylation of GluA2-short by PKC on IES880VK weakens its binding to GRIP but improves binding to protein interacting with C kinase 1, which slows recycling of GluA2-containing AMPA receptors back to the plasma membrane after internalization (Matsuda et al., 1999; Chung et al., 2000; Seidenman et al., 2003; Lin and Huganir, 2007; Park et al., 2009). The situation is complex, however, because States et al. (2008) identified a large population of synaptic GluA2 receptors bearing phospho-Ser880, which presumably had secured synaptic anchors other than GRIP. Finally, phosphorylation of the nearby NVY876GI by src family kinases also seems to weaken association with GRIP (Hayashi and Huganir, 2004).
Three phosphorylation sites have been identified by mass spectrometry (Munton et al., 2007; Ballif et al., 2008; Trinidad et al., 2008) on the GluA3 C-tail but have not yet been studied functionally. The TES891VK site near the terminus is surrounded by sequences homologous with those of the other AMPA receptor subunits and may be a PKC target.
GluA4 has alternate C-tails, with the short tail terminated by a PKC target at TES899IK (Esteban et al., 2003). The long GluA4 C-tail harbors a combined PKC/PKA target at RLS842IT (Carvalho et al., 1999; Gomes et al., 2007) and may bind directly to some PKC isoforms (Correia et al., 2003). Phosphorylation is enhanced by synaptic activity and promotes surface expression of GluA4 by disrupting its association with α-actinin-1 (Nuriya et al., 2005). Similar to GluA2, the Jun kinase site on GluA4 (VLT855PD) is phosphorylated at rest but rapidly dephosphorylated within minutes of synaptic activity and presumably functions, as in GluA2-long, to enhance surface expression of GluA4 (Thomas et al., 2008). The functional consequence of GluA4 phosphorylation by c-Jun NH2-terminal kinase has not yet been examined.
Most studies have suggested a trafficking function for phosphorylation of specific residues in AMPA receptor C termini, and a similar pattern seems to exist for kainate receptor phosphorylation. Serine residues KKS879RT and GKS885SF of GluK1 are phosphorylated by PKC, resulting in internalization (Rivera et al., 2007). These sites, previously identified by Hirbec et al. (2003) on the basis of in vitro kinase assays, may be involved in autoregulation by kainate receptor activation (Rivera et al., 2007). The long C-tail of GluK2 is phosphorylated by PKA on serine residues KFS825FC and RMS837LK, which potentiates receptor activation in whole-cell patch studies (Kornreich et al., 2007), apparently through an increase in receptor open probability (Traynelis and Wahl, 1997). There are no reported modification sites in the C-tail of GluK3 or GluK5, whereas GluK4 has four phosphorylation sites identified by mass spectrometry.
B. N-Methyl-d-aspartate and δ Receptor Phosphorylation
Given the function of NMDA receptors in synaptic plasticity (see section IX), a wealth of studies exist describing the consequences of modification of specific residues on its C termini. GluN1 has four different C-terminal tails created by alternative splicing (Fig. 6), only the longest of which has been shown to be influenced by phosphorylation. Disrupting a ring of tyrosine residues adjacent to the M4 domain by site-directed mutagenesis of IAY837KR on GluN1 and LFY842WK on GluN2A prevented use-dependent desensitization of GluN1/GluN2A receptors (Vissel et al., 2001). It is noteworthy that tyrosine phosphorylation of GluN2A but not GluN1 was detected with phosphotyrosine antibodies (Lau and Huganir, 1995). This same ring of tyrosines is present in all NMDA receptor subunits as well as all AMPA and kainate receptor subunits except GluK4 and GluK5 (Figs. 5 to 7), but whether each of these tyrosines is subject to phosphorylation by src family kinases with resulting functional consequences has yet to be explored. PKC targets serine residues ASS890FK and RRS896SK on GluN1, and PKA targets the adjacent RSS897KD. Within minutes of PKC activation, phosphorylation of ASS890FK disrupts surface clusters of NMDA receptors (Tingley et al., 1997). The dual PKC-PKA phosphorylation of RRS896S897K, on the other hand, promotes exit of the subunit from the endoplasmic reticulum and transit to the surface membrane (Scott et al., 2001, 2003). The phosphorylation of GluN1 ASS890FK and RRS896SK is achieved by PKCγ and PKCα, respectively (Sanchez-Perez and Felipo, 2005), potentially contributing to the selective modulation of NMDA receptor function and/or intracellular localization. Prolonged synaptic activation during status epilepticus initially causes dephosphorylation of ASS890FK, then hyperphosphorylation that develops over hours and subsides within a day (Niimura et al., 2005). The PKA site RSS897KD behaves similarly, but the adjacent PKC site RRS896SK seems untouched by status epilepticus, demonstrating a remarkable specificity.
The GluN2A subunit also occurs with alternative C-tails (Fig. 6) (444 and 626 amino acids), and each has phosphorylation sites that can modify function. The consequence of phosphorylation of LFY842WK, which is present on both C-tails, was described above. Both C-tails also harbor MRS1232PF; phosphorylation of this serine by CDK5 is associated with increased NMDA-evoked currents (Li et al., 2001), and excitotoxic death of hippocampal CA1 pyramidal cells (Wang et al., 2003). Krupp et al. (2002) demonstrated that glycine-independent desensitization of GluN1/GluN2A receptors could be eliminated by mutation of either of two serine residues proximal to the fourth transmembrane domain, LRS900AK and RGS932LI, but the kinases acting on these serines have not been identified. Nine additional serines or tyrosines have been reported by mass spectroscopy as phosphorylated on both long and short C-tails, but phosphorylation of exclusive residues near the end of the longer C-tail regulates NMDA receptor modulation. For example, src kinase weakens high-affinity zinc inhibition of recombinant GluN1/GluN2A receptors, an effect that is eliminated by mutation of DPY1387KH (Zheng et al., 1998; but see Xiong et al., 1999); this same tyrosine was shown to account for approximately 30% of the src-induced phosphorylation of GluN2A (Yang and Leonard, 2001). Src also phosphorylated HSY1292DN (Yang and Leonard, 2001), but this had no effect on zinc inhibition (Zheng et al., 1998). By contrast, mutation of QVY1267QQ blocked the src effect on zinc inhibition, but this residue does not seem to be directly phosphorylated by src (Yang and Leonard, 2001). These results, considered together, raise the possibility that the electrophysiological consequences of mutating Tyr1267 are due to an allosteric effect on receptor function.
Insulin potentiates the activation of GluN2A-containing NMDA receptors, an effect traced to two serines phosphorylated by PKC, QHS1291YD, and SIS1312LK (Jones and Leonard, 2005). It is noteworthy that Ser1291 is immediately adjacent to the src target Tyr1292, yielding high potential for cross-talk between these kinases; one wonders, for example, whether the lack of effect of Tyr1292 on zinc inhibition (Zheng et al., 1998) was due to obstructive phosphorylation of the adjacent Ser1291 by PKC.
The GluN2B C-tail (Fig. 7) contains a prominent src target very near the C terminus on HVY1472EK, which may regulate endocytosis in some conditions (Cheung and Gurd, 2001; Nakazawa et al., 2001; Snyder et al., 2005). Phosphorylation of this tyrosine is eliminated in a fyn knockout (Abe et al., 2005) and is increased in hippocampal CA1 during LTP (Nakazawa et al., 2001) as well as in a chronic neuropathic pain model (Abe et al., 2005). Although regulation of this tyrosine by fyn kinase is well understood, the immediate functional consequence of phosphorylation has not been studied. This tyrosine is also phosphorylated in an activity-dependent manner by the transmembrane tyrosine kinase EphrinB2 (Nateri et al., 2007), which is activated by upstream ERK pathways. Therefore, multiple kinases target Tyr1472 in an activity-dependent manner.
Another major modification of GluN2B in the postsynaptic density fraction of the forebrain occurs by phosphorylation of QHS1303YD by CaMKII (Omkumar et al., 1996). The CaMKII system provides a mechanism by which GluN2B-containing NMDA receptors can be modified rapidly upon Ca2+ influx associated with NMDA receptor activation and in this sense may represent the prototype of a kinase activated by its target (Bayer et al., 2001), but the functional consequences of this modification have not been directly addressed. Finally, the C-terminal PDZ domain of GluN2B contains a site, IES1480DV, that, when phosphorylated by casein kinase II, disrupts the interaction of GluN2B with PSD-95 and SAP102, thereby decreasing surface expression of this receptor (Chung et al., 2004). This represents another activity-dependent phosphorylation that regulates trafficking of GluN2B in the plasma membrane. The association between active CaMKII and GluN2B seems to be required for LTP, although the phosphorylation target has not been identified (Barria and Malinow, 2005).
PKC and PKA both phosphorylate GluN2C RIS1230SL (Fig. 7) near the extreme carboxyl terminus but exert no apparent effect on surface expression of GluN2C or on its interaction with PDZ family proteins on the adjacent PDZ domain. However, a phosphomimetic GluN2C(S1230E) mutant exhibited faster activation and inactivation kinetics in outside-out patches (Chen et al., 2006). Thus, unlike other glutamate receptor subunits, phosphorylation of this serine near the PDZ domain does not affect trafficking but instead alters channel properties. A site in the C-terminal domain of GluN2C, HAS1096LP, is a unique (among the glutamate receptors) target for PKB/Akt; interestingly, phosphorylation of Ser1096 creates a binding site for 14-3-3 (Chen and Roche, 2009), which apparently assists in the trafficking of GluN2C to the surface membrane. Phosphorylation of Ser1096 is enhanced by insulin-like growth factor-1 stimulation or NMDA receptor activation, providing a link between hormonal state and NMDA receptor function. Three phosphorylated serines in GluN2D have been identified by mass spectroscopy (Fig. 7) but have not yet been studied functionally. Likewise, a phosphotyrosine antibody labels GluN2D immunoprecipitated from rat thalamus (Dunah et al., 1998b), but the targeted tyrosines have not been identified.
No modifications of GluN3A, GluN3B, or GluD1 have been described yet, but GluD2 has four phosphoserines, one of which (TLS945AK) was shown to be a PKC target (Fig. 7) (Kondo et al., 2005). The GluD2 subunit is expressed mainly by cerebellar Purkinje cells, and both it and PKC seem to be essential for long-term depression at the parallel fiber-Purkinje cell synapse (Kashiwabuchi et al., 1995; Kondo et al., 2005). However, LTD could be rescued in a GluD2-null mouse by a GluD2(S945A) transgene, demonstrating that phosphorylation of Ser945 plays no role in LTD (Nakagami et al., 2008). Indeed, a transgene lacking the transmembrane domains could rescue LTD (Kakegawa et al., 2007b), suggesting that GluD2 might function as a scaffolding protein rather than an ionotropic receptor.
C. Other Post-Translational Modifications of Glutamate Receptors
Additional post-translational modifications can affect glutamate receptor localization or activity. All ionotropic glutamate receptor subunits seem to be glycosylated (see section II), which seems to be involved in proper folding of the subunit during synthesis (Everts et al., 1997; Mah et al., 2005; Nanao et al., 2005; Gill et al., 2009). In addition, multiple glutamate receptor subunits undergo dynamic regulation by palmitoylation (Figs. 5⇑–7). The AMPA receptors have two known palmitoylation sites in the membrane domain 2 and C-tail. All AMPA receptors have a conserved cysteine residue proximal to M4 that undergoes palmitoylation (GluA1-EFC811YK, GluA2-C836, GluA3-C841, GluA4-C817) (Hayashi et al., 2005). In GluA1 and GluA2, palmitoylation of the C-tail residue reduces insertion rate (Hayashi et al., 2005) and regulates phosphorylation of the two serines on the C-terminal tail by PKC (Lin et al., 2009). Thus, the interplay between palmitoylation and phosphorylation of residues on this membrane-proximal region of GluA1 influences membrane insertion and thus synaptic availability of this subunit. All AMPA receptors have another conserved cysteine residue just downstream of the QRN site (GluA1-QGC585DI, GluA2-C610, GluA3-C615, GluA4-C611) that, when palmitoylated, increases AMPA receptor surface expression (Hayashi et al., 2005). The homomeric kainate receptor GluK2 undergoes palmitoylation of the cysteines SFC858SA and LKC871QR, both of which have no apparent effect on basal receptor function. In addition to palmitoylation, GluK2 has been shown to be SUMOylated at lysine IVK896TE, a process that facilitates endocytosis after kainate receptor activation (Martin et al., 2007). GluK3 has been shown to be SUMOylated, although the site has not been identified (Wilkinson et al., 2008).
Like AMPA and kainate receptors, the GluN2A and GluN2B subunits also undergo palmitoylation at their C termini (Hayashi et al., 2009). Both of these subunits have two separate clusters of cysteine residues that are palmitoylated with distinct consequences on receptor expression and internalization. The first cluster of palmitoylated cysteine residues within GluN2A (RFC848FT, GVC853SD, YSC870IH) and the homologous cysteines in GluN2B (Fig. 7) increase nearby tyrosine phosphorylation of the C-tails by src family kinases, regulating surface expression and receptor internalization. Tyrosine phosphorylation of the C-tails eliminated receptor interaction with activator protein-2, decreasing clathrin-mediated endocytosis. Palmitoylation of the second cluster of cysteine residues within GluN2A (THC1214RSC1217LS, FKC1236DAC1239LR) and the homologous residues in GluN2B plus EAC1245KK leads to accumulation of the NMDA receptors in the Golgi apparatus and decreased surface expression (Hayashi et al., 2009). Thus, palmitoylation of GluN2A and GluN2B provides a dual mechanism by which post-translational modification controls NMDA receptor surface expression. It is unknown if GluN2C or GluN2D subunits undergo palmitoylation, but both have cysteine residues homologous to GluN2A and GluN2B within their C-terminal tails, leading to the possibility that palmitoylation also regulates their surface expression. Studies so far suggest that the GluN1 subunit does not undergo palmitoylation (Hayashi et al., 2005).
Both GluN1 and GluN2 subunits undergo S-nitrosylation on cysteine thiol groups by endogenous nitric oxide (NO) and exogenously applied S-nitrosothiols. The GluN1 subunit has two cysteine groups within the M3–M4 extracellular linker, QKC744DL and QEC798DS, and the GluN2A subunit has three cysteine groups within the ATD, HVC87DL, ASC320YG, and SDC399EP, that can be S-nitrosylated (Choi et al., 2000). The S-nitrosylated cysteine residues within the GluN1 and GluN2 subunits mainly fit the S-nitrosylation consensus motif of a cysteine residue preceded by an acidic or basic residue and followed by an acidic residue (Stamler et al., 1997). S-Nitrosylation of any of these cysteine residues within GluN1/GluN2A leads to moderate NMDA receptor inhibition, decreasing the response to agonist-evoked currents by approximately 20%, but a majority of the inhibition is due to Cys399 on the GluN2A subunit ATD (Choi et al., 2000). Inhibition of GluN2A current amplitude through S-nitrosylation of Cys399 is due to decreased channel opening, which may be caused by the increased affinity of the receptor for Zn2+ and glutamate, leading to receptor desensitization (Paoletti et al., 2000; Zheng et al., 2001; Lipton et al., 2002). The GluN2A subunit is sensitized to S-nitrosylation when Cys744 and Cys798 of the GluN1 subunit are S-nitrosylated (Takahashi et al., 2007). It is not known whether the remaining GluN2 subunits are S-nitrosylated or inhibited by this modification. The ability of NO to regulate NMDA receptors may provide a feedback mechanism to prevent excessive receptor activity, because NMDA receptor-catalyzed transmembrane currents increase the local Ca2+concentration, which activates neuronal nitric-oxide synthase through their mutual association with PSD-95 (Sattler et al., 1999; Rameau et al., 2003).
NMDA receptors are regulated by the extracellular redox state; sulfhydryl reducing agents such as dithiothreitol and dihydrolipoic acid potentiate NMDA-evoked currents through the formation of free thiol groups, whereas oxidizing agents such as 5,5′-dithio-bis(2-nitrobenzoic acid) and oxidized glutathione inhibit currents through the formation of disulfide bonds (Aizenman et al., 1989; Sucher and Lipton, 1991; Köhr et al., 1994; Choi and Lipton, 2000). Both GluN1 and GluN2 subunits are responsible for redox modulation. Disulfide bonds formed between two pairs of cysteine residues within GluN1 (SVC79ED and RGC308VG within the ATD and Cys744 and Cys798 in the LBD) cause the intermediate and slow components of redox modulation for all GluN2-containing NMDA receptors (Sullivan et al., 1994; Choi et al., 2001). A disulfide bond has been proposed to form in the GluN2A subunit between Cys87 and Cys320 of the GluN2A ATD and has been shown to form between Cys86 and Cys321 in the crystal structure of GluN2B ATD (Karakas and Furukawa, 2009). This disulfide bond has been suggested to mediate a fast-component of redox modulation, although this may also reflect modification of the Zn2+ binding site (Köhr et al., 1994; Choi et al., 2001). Reducing agents can chelate zinc, and transient potentiation that occurs only in the presence of agents such as dithiothreitol reflects in part the relief of zinc inhibition through zinc chelation (Paoletti et al., 1997). In addition, reduction of cysteine residues within GluN1/GluN2A NMDA receptors also reduces high-affinity voltage-independent Zn2+ inhibition of NMDA receptors (Choi et al., 2001).
D. Proteolysis of Glutamate Receptors
A number of proteases can cleave glutamate receptors. Principal among these are serine proteases, which can act on GluN1, GluN2A, and GluN2B. Tissue plasminogen activator can cleave GluN1 at ALR260YA, whereas plasmin and thrombin cleave at multiple sites within GluN1, presumably with functional consequences (Gingrich et al., 2000; Fernández-Monreal et al., 2004; Samson et al., 2008). A specific plasmin cleavage site (EAK317AS) has been described in GluN2A, proteolysis of which leads to removal of the ATD and the high-affinity Zn2+ binding site contained therein (Yuan et al., 2009b). Thrombin cleaves the analogous residue (Lys318) in GluN2B (Leung et al., 2007). In addition, the calcium-dependent nonlysosomal protease family calpains can cleave the C termini of GluA1, GluN2A, GluN2B, and GluN2C, resulting in receptor degradation and reduced synaptic activity (Bi et al., 1998a,b; Guttmann et al., 2001, 2002; Rong et al., 2001; Simpkins et al., 2003; Araújo et al., 2005). It is noteworthy that specific proteolytic cleavage has been proposed to occur in pathological situations such as ischemia (Yuen et al., 2007), blood-brain barrier breakdown (Yuan et al., 2009b), and status epilepticus (Araújo et al., 2005). In addition, matrix metalloproteinase-7 has been shown to cleave the LBDs of the GluN1 and GluN2A subunits, decreasing the NMDA receptor-mediated calcium influx and increasing the ratio of AMPA to NMDA receptors in cortical slices (Szklarczyk et al., 2008). Among the AMPA receptors, GluA3 has been reported to serve as a substrate for proteolysis by gamma secretase and granzyme B (Gahring et al., 2001; Meyer et al., 2002); glycosylation of ISN388DS protects GluA3 from cleavage by granzyme B during breakdown of the blood-brain barrier. GluA1 can be cleaved at SHD865FP through activation of calpain and caspase8-like activity (Bi et al., 1996; Meyer et al., 2002). AMPA receptor proteolysis may be common in neuropathological conditions (Bi et al., 1998a; Chan et al., 1999).
V. Agonist and Antagonist Pharmacology
Amino acid numbering in AMPA and kainate receptor subunits has historically been for the mature protein without the signal peptide, whereas amino acid numbering of NMDA and GluD2 receptor subunits has started with the initiating methionine as 1. To simplify comparison with the published literature, we will maintain this informal convention here.
A. α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic Acid, Kainate, and δ Receptor Agonists
Glutamate activates all AMPA and kainate receptors by binding within the cleft between domains D1 and D2 of the LBD to induce domain closure. Neither NMDA nor l-aspartate can activate the non-NMDA receptors, and d-aspartate acts as a low-affinity competitive antagonist for native AMPA receptors (Gong et al., 2005). In the structure of the glutamate-bound GluA2 LBD (Fig. 8A) (Armstrong and Gouaux, 2000), the α-amino group of glutamate forms a tetrahedral network of interactions with the backbone carbonyl oxygen of Pro478, the side chain hydroxyl of Thr480, and the carboxylate group of Glu705. The α-carboxyl group of glutamate forms a bidentate interaction with Arg485 and receives hydrogen bonds from the backbone amide nitrogens of Thr480 and Ser654. The glutamate γ-carboxyl group interacts with the hydroxyl group and backbone amide nitrogen of Thr655 (Fig. 8A), whereas the isoxazole hydroxyl group of AMPA interacts with the amide nitrogen of Thr655 via a water molecule. Furthermore, the side chain of Tyr450 forms an electron-dense ring structure above the glutamate α- and β-carbon atoms resembling a lid above the agonist binding pocket (Fig. 8A). Structures of the kainate receptor subunits GluK1 and GluK2 LBDs show similar atomic contacts for the α-carboxyl group, α-amino group, and γ-carboxyl groups of glutamate, although residues lining the GluK1 agonist binding pocket are smaller than GluK2 (Mayer, 2005; Nanao et al., 2005; Naur et al., 2005; Mayer et al., 2006) (Fig. 8B). There are several noteworthy differences among the kainate receptor LBDs, such as the loss of a hydrogen bond to the agonist α-amino group at GluK2 Ala487, which is equivalent to Thr480 in GluA2 and Thr503 in GluK1. This may contribute to the higher glutamate EC50 for GluK2 compared with GluK1 (Mayer, 2005). A central feature of AMPA and kainate receptor agonist binding is closure of the cleft in which the agonist binds, a conformational change mediated by movement of D2 relative to D1 within the bilobed LBD (Armstrong and Gouaux, 2000; Mayer, 2005) (see sections II.D, VII.B).
In addition to glutamate, a number of naturally occurring molecules, such as ibotenic acid and willardiine, plus an array of AMPA, ibotenic acid, and willardiine analogs, activate AMPA and kainate receptors (Table 5). It has been difficult to identify naturally occurring or synthetic agonists that discriminate well between all AMPA and kainate receptors. AMPA acts as a partial agonist at some kainate receptor subunit combinations (Herb et al., 1992; Swanson et al., 1996; Schiffer et al., 1997), and kainate can induce very rapid desensitization of neuronal AMPA receptors (Patneau et al., 1993). Nevertheless, there are some examples of kainate receptor-selective agonists, such as (2S,4R)-4-methylglutamic acid (SYM2081) and perhaps ATPA, a tert-butyl analog of AMPA (see Tables 5 and 6). Crystal structures of the GluK1 and GluK2 LBDs have revealed differences between kainate and AMPA receptors that partly explain these kainate receptor-selective actions. The agonist-binding cavities of GluK1 and GluK2 are 40 and 16% larger, respectively, than GluA2, allowing GluK1 and GluK2 to accommodate larger ligands (Mayer, 2005; Naur et al., 2005). Indeed, steric occlusion between the 4-methyl group of SYM2081 and GluA2 Leu650 contributes to its selectivity for GluK1 and GluK2, both of which have a smaller valine in the corresponding positions (GluK1, Val670; GluK2, Val654) (Armstrong et al., 1998, 2003; Mayer, 2005). Likewise, the isoprenyl group of kainate shows reduced steric occlusion in GluK1 and GluK2 compared with GluA2 because of the smaller valine residues. Steric occlusion also may explain selectivity of ATPA for GluK1, because its bulky tert-butyl group interacts with Leu650, Thr686, and Met708 in GluA2, which are replaced by the smaller amino acids Val670, Ser706, and Ser726, respectively, in GluK1 (Lunn et al., 2003). The availability of crystallographic data for multiple kainate receptor subunits emphasizes how useful structural information can be across each subunit family. Thus, there remains the need for new crystallographic data for additional members of each glutamate receptor subtype, as opposed to the reliance on homology modeling and molecular dynamics simulations. Although useful, molecular dynamics simulations of homology models carry significant caveats, including uncertainty associated with representation of amino acid insertions, placement of new ligands, approximations of force fields for membrane-spanning elements, and computational limitations in simulating movement of large multisubunit protein assemblies. At the same time, crystals are rarely formed on demand, low resolution can create ambiguities in protein threading and details of ligand binding pose, and dynamic aspects ranging from ligand interactions to domain coupling are not revealed by a static X-ray structure. Early in a project, in the absence of a three-dimensional structure, a homology model can offer unique insights, whereas in late phases of a project, possession of a crystal structure can benefit significantly from dynamic refinement and trajectory analysis. The approaches are complementary both in terms of a project's evolution and the information content provided by the two structural perspectives.
A few agonists show useful selectivity between the GluA1/GluA2 and GluA3/GluA4 subunits (Table 5). Br-HIBO, an analog of ibotenic acid, preferentially activates GluA1 and GluA2 versus GluA3 and GluA4 receptors (Coquelle et al., 2000) through involvement of water-mediated hydrogen bonding to Tyr702 in GluA1 and GluA2, which is Phe in GluA3 and GluA4. Thus, ordered water molecules within the agonist binding site interact with the ligand to influence specificities among GluA subunits (Banke et al., 2001; Hogner et al., 2002; Pentikäinen et al., 2003; Frandsen et al., 2005). Cl-HIBO was synthesized after molecular modeling predicted that the exchange of bromine for chlorine would improve selectivity (Bjerrum et al., 2003). Cl-HIBO activates GluA1 and GluA2 with 275- to 1600-fold selectivity over GluA3 or GluA4, respectively. The agonist 2-benzyl-tetrazol-AMPA shows 40-fold selectivity for GluA4 over GluA1 (Jensen et al., 2007). Crystal structures of 2-benzyl-tetrazol-AMPA bound to the GluA2 LBD reveal that the benzyl group occupies a novel cavity opened up by movement of Met708 in GluA2, and the selectivity of 2-benzyl-tetrazol-AMPA is due to residues adjacent to this cavity (Val690 and Ala691), which are conserved in GluA2 to GluA4 but correspond to Met686 and Ile687 in GluA1 (Vogensen et al., 2007).
Although some agonists discriminate between GluA1/GluA2 and GluA3/GluA4 subunits, it remains to be shown whether agonists that act selectively at an individual GluA subunit can be developed. The amino acid sequences for the LBDs of the four GluA subunits are 80% identical (Table 2). Ligands with agonist activity at AMPA receptors contain a chemical moiety equivalent to the α-amino and α-carboxyl groups of glutamate, and the binding of this moiety is conserved for the agonists crystallized thus far. Moreover, crystal structures and homology models show the residues in direct contact with agonists such as glutamate, AMPA, and kainate are fully conserved across all GluA subunits. Crystallization of LBDs bound to additional agonists, however, could allow identification of novel binding modes or differences in agonist binding that could be exploited through medicinal chemistry efforts to achieve greater subunit selectivity. Alternatively, molecular modeling, which can account for important motions within the receptor, could allow for analysis of binding to subunits that have not been crystallized and has been used successfully, for example, to predict the activity of Cl-HIBO.
The development of useful subunit-selective compounds is further complicated by the formation of heteromeric receptors in native tissue, which contain two different GluA subunits, and by the association of AMPA receptors with interacting partners. Association of the TARP auxiliary subunits with AMPA receptors (see section II.H) increases efficacy and affinity for a range of AMPA receptor agonists, including kainate (Turetsky et al., 2005; Kott et al., 2007). Because agonist pharmacology has been studied extensively in recombinant systems without coexpression of TARPs, the caveat exists as to whether agonist properties (Table 5) will remain the same in the presence of TARPs.
Kainate has a 2-carboxypyrrolidine-3-acetic acid backbone, and analogs containing this backbone, known as kainoids (Sonnenberg et al., 1996; Hodgson et al., 2005; Sagot et al., 2008; Bunch and Krogsgaard-Larsen, 2009), include domoic acid (Hampson et al., 1992; Alt et al., 2004) and acromelic acid (Kwak et al., 1992; Smith and McIlhinney, 1992). Agonist potency and efficacy are subunit-specific, because kainate and domoic acid are potent agonists of GluK1 and GluK2 but show low potency at GluK3 receptors (Table 6; Jane et al., 2009). Agonists acting preferentially at kainate receptors over AMPA receptors include the glutamate analogs SYM2081 (Zhou et al., 1997), dysiherbaine (Sakai et al., 1997), and neodysiherbaine (Sakai et al., 2001a). SYM2081 has similar potencies for GluK1 and GluK2 and causes pronounced desensitization (Jane et al., 2009). Dysiherbaine has nanomolar affinity for GluK1 and GluK2 and micromolar affinity for GluK5 (Sakai et al., 2001b; Swanson et al., 2002; Sanders et al., 2005). Because of its high affinity for GluK1, dysiherbaine promotes a desensitized state of the receptor that persists for at least 20 to 45 min after removal. This unique activity of dysiherbaine was used to block GluK1 subunits in GluK1/GluK5 diheteromeric receptors, which revealed that glutamate evokes a desensitizing response from the remaining GluK5 subunits (Swanson et al., 2002). This finding suggests that kainate receptors undergo subunit-specific gating similar to AMPA receptors (Jin et al., 2003).
Multiple agonists act selectively at GluK1 over the other kainate receptor subunits (Table 6). The neodysiherbaine analogs 8-deoxy-neodysiherbaine, 9-deoxy-neodysiherbaine, and MSVIII-19 show nanomolar affinity for GluK1 and >1000-fold selectivity over GluK2, GluK3, and GluK5, with 8- and 9-deoxy-neodysiherbaine acting as a partial and full agonists, respectively (Lash et al., 2008). MSVIII-19 was originally reported as a GluK1 antagonist (Sanders et al., 2005), but crystallographic studies revealed that it induces full domain closure of the GluK1 LBD, prompting further functional studies that showed it to be an agonist of extremely low efficacy (Frydenvang et al., 2009). (2S,4R,6E)-2-Amino-4-carboxy-7-(2-naphthyl)hept-6-enoic acid (LY339434), ATPA, (S)-2-amino-3-(3-hydroxy-7,8-dihydro-6H-cyclohepta[d]isoxazol-4-yl) propionic acid, (S)-5-iodowillardiine, and (4R)-isopentyl glutamate are potent agonists at homomeric GluK1 receptors but have little to no activity at GluK2 receptors (Clarke et al., 1997; Jane et al., 1997; Zhou et al., 1997; Small et al., 1998; Brehm et al., 2003; Bunch et al., 2009). GluK1 selectivity arises from the larger GluK1 binding cavity, which relieves steric occlusion of the bulky tert-butyl group of ATPA and halogen atom of (S)-5-iodowillardiine (Mayer, 2005). Likewise, steric occlusion explains why AMPA binds GluK1 but not GluK2, because the isoxazole ring cannot make key contacts with D2 (Mayer, 2005). Mutagenesis studies have confirmed the importance of cavity size, and the exchange of Ser706 in GluK1 to the larger Asn690 in GluK2 predictably alters potency of AMPA, iodowillardiine, and ATPA toward GluK1 (Swanson et al., 1997a, 1998; Nielsen et al., 2003).
The LBD of the GluD2 (δ2) family of glutamate receptor subunits binds d-serine, sharing some but not all features of d-serine binding to GluN1 (Fig. 8F) (Naur et al., 2007). However, GluD2 receptor function remains poorly understood. Transgenic experiments show that insertion of a mutant GluD2 into GluD2(−/−) mice can rescue these mice from neurological deficits even when the inserted GluD2 has a mutation in the ion channel pore that either disrupts Ca2+ permeability (Kakegawa et al., 2007a) or abolishes current flow through the ion channel (Kakegawa et al., 2007b). In addition, GluD2 can induce presynaptic terminal differentiation even without its LBD, which contains the d-serine binding site (Kuroyanagi et al., 2009; Torashima et al., 2009). These data suggest that GluD2 does not influence cerebellar function through actions as a ligand-gated ion channel. No data has yet shown functional ionic currents in wild-type GluD2 receptors. However, a mutation in M3, GluD2(A654T), within the highly conserved SYTANLAAF gating motif causes spontaneously active receptors (Zuo et al., 1997; Kohda et al., 2000), and these receptors are inhibited by the binding of d-serine, perhaps through destabilization of the dimer interface and desensitization (Naur et al., 2007; Hansen et al., 2009). It remains to be determined whether d-serine has functional effects on neuronal GluD2.
B. N-Methyl-d-aspartate Receptor Agonists
NMDA receptors are unique among the glutamate receptor family in that the simultaneous binding of glycine to GluN1 and glutamate to GluN2 is required for activation (Kleckner and Dingledine, 1988). Crystal structures of the bilobed GluN1 LBD show that glycine and related agonists (Table 7) bind within the cleft between the D1 and D2 domains. The α-carboxyl group of glycine forms hydrogen bonds within the binding pocket with Arg522, Thr518, and Ser688, whereas the amino group of glycine interacts with the carbonyl oxygen of Pro516, the hydroxyl group of Thr518, and the carboxylate oxygen of Asp732 (Fig. 8D) (Furukawa and Gouaux, 2003). Trp731 within the GluN1 binding pocket has been proposed to hinder glutamate binding because of steric clash between Trp731 and the glutamate γ-carboxyl group. In GluN2A, the smaller side chain of Tyr730 is in van der Waals contact with the γ-carboxylate of glutamate (Fig. 8C). GluN1 has Val689 corresponding to GluN2A Thr690, which leads to loss of a hydrogen bond donor that stabilizes the glutamate γ-carboxyl group in GluN2A (Fig. 8, C and D). Comparison between crystallographic structures of the GluN1 subunit bound to full and partial agonists indicates that the degree of closure of the LBDs' D1 and D2 domains is not correlated with relative agonist efficacy, as has been demonstrated with the GluA2 LBD. Partial agonists 1-aminocyclobutane-1-carboxylic acid and 1-aminocyclopropane-1-carboxylic acid (Priestley et al., 1995) induce a similar degree of domain closure as glycine, differing by less than 0.5° (Inanobe et al., 2005).
In addition to glycine, the d- and l-isomers of serine and alanine are agonists at the GluN1 subunit (Pullan et al., 1987; McBain et al., 1989) (Table 7). d-Serine is more potent than l-serine and may be the primary ligand for GluN1 in regions such as the supraoptic nucleus (e.g., Panatier et al., 2006). d-Serine is synthesized from l-serine by serine racemase in both astrocytes (Wolosker et al., 1999) and neurons (Mustafa et al., 2004; Miya et al., 2008; Wolosker et al., 2008). It is noteworthy that serine racemase is regulated by NMDA receptor activity, mGluR5 activation, nitrosylation, divalent cations, and nucleotides (Shoji et al., 2006; Baumgart and Rodríguez-Crespo, 2008; Balan et al., 2009; Mustafa et al., 2009), and deletion of serine racemase alters glutamatergic synaptic transmission and produces behavioral phenotypes (Basu et al., 2009).
Cyclic and halogenated analogs of glycine, including d-cycloserine, act as GluN1 partial agonists (Hood et al., 1989; Priestley and Kemp, 1994; Sheinin et al., 2001; Dravid et al., 2010) (Table 7). Although d-cycloserine is a partial agonist of GluN2A-, GluN2B-, and GluN2D-containing NMDA receptors, the responses of GluN2C-containing NMDA receptors are greater in d-cycloserine than those evoked by glycine (Sheinin et al., 2001; Dravid et al., 2010). This raises the possibility that potentiation of GluN2C-containing NMDA receptors could underlie the positive effects of d-cycloserine on cognition, fear extinction, and motor dysfunction (Kalia et al., 2008; Norberg et al., 2008) through action on GluN2C-expressing neurons (Monyer et al., 1994). It is noteworthy that the identity of the GluN2 subunit within the NMDA receptor determines the potencies of GluN1 agonists, which are least potent (highest EC50) for GluN1/GluN2A and most potent (lowest EC50) for GluN1/GluN2D (Kuryatov et al., 1994; Wafford et al., 1995; Furukawa and Gouaux, 2003; Chen et al., 2008) (Table 7).
Glutamate binding to GluN2A involves interactions of the agonist α-carboxylate group with Arg518 and the agonist γ-carboxylate group with Tyr730 within the binding pocket, together with an interdomain hydrogen bond formed between Tyr730 and Glu413 (Fig. 8C; Furukawa et al., 2005). The crystal structure of the agonist-bound GluN1-GluN2A LBD heterodimer suggests a mechanism for selectivity for NMDA over AMPA receptors (Furukawa et al., 2005). The GluN2A subunit has Asp731 within the binding pocket, whereas the GluA2, GluK1, and GluK2 receptor subunits have a glutamate residue at the corresponding position that interacts with the agonist amino group (Fig. 8, A–C). Because the aspartate residue within the GluN2A subunit is a methylene group shorter than the glutamate residue found in GluA and GluK subunits, it cannot interact with the agonist α-amino group of glutamate, which instead forms water-mediated hydrogen bonds to GluN2A Glu413 and Tyr761. Surprisingly, the charge-conserving substitution GluN2A(D731E) and GluN2B(D732E) renders the receptor nonfunctional (Williams et al., 1996; Laube et al., 2004; Chen et al., 2005), perhaps a result of interference with the water-mediated interactions at the α-amino group of glutamate and/or a disruption of the binding pocket. The reduced side-chain length of Asp731 also creates space for NMDA to fit within the GluN2A binding pocket. Modeling of NMDA into the GluN2A LBD suggests that the N-methyl group of NMDA is accommodated in the binding pocket by displacement of the water molecule that binds the α-amino group of glutamate (Furukawa et al., 2005). Other studies using mutagenesis and homology modeling of GluN2A and GluN2B LBDs have suggested that NMDA cannot bind to GluA subunits because of steric clash between the N-methyl group of NMDA and Met708 in GluA2, which is conserved among all AMPA receptors (Laube et al., 2004; Chen et al., 2005).
GluN2 endogenous agonists include glutamate, d- and l-aspartate (