13.2.1. Synaptic Function

Excitatory synaptic transmission in the vertebrate brain relies on the release of L-glutamate from presynaptic terminals that diffuses across the synaptic cleft and binds to postsynaptic AMPARs and NMDARs. Activation of AMPARs is fast and transient, causing brief depolarizations that last no longer than a few milliseconds. NMDARs are not critical for this basal synaptic transmission, but instead they regulate functional and structural plasticity of individual synapses, dendrites, and neurons by allowing activation of specific calcium-dependent signaling cascades. Several unique properties of NMDARs prevent their activation by L-glutamate released during a single synaptic event.

First, NMDARs activate significantly slower than AMPARs and kainate receptors. Glutamate released from a presynaptic terminal following arrival of an action potential is removed efficiently from the synaptic cleft by the actions of glutamate transporters located in the presynaptic terminal and nearby astrocytes [10,11]. Consequently, glutamate is available for receptor binding only briefly during low frequency synaptic transmission. Because NMDARs have relatively a high affinity for glutamate, the millisecond-long neurotransmitter pulses should be able to partially (and slowly) activate NMDARs. However, individual excitatory synaptic inputs received during baseline activity do not result in calcium (Ca²⁺) influx because of a second NMDAR property: its pronounced voltage dependence.

At resting membrane potentials, external magnesium (Mg²⁺) ions enter the NMDAR pore, but unlike the permeant Ca²⁺ ions, they bind tightly and prevent further ion permeation [12,13]. Mg²⁺ ions are present at millimolar concentrations in the external milieu of neurons, while intracellular Mg²⁺ concentrations are in the micromolar range, resulting in a net inward driving force for Mg²⁺ ions at negative membrane potentials. A depolarization of sufficient amplitude and duration is required to dislodge and repel the Mg²⁺ ions from the pore, thereby allowing the flow of permeant ions. As a result, the NMDAR acts as a molecular coincidence detector [14]: efficient activation and ion permeation through the NMDAR requires both a sufficiently strong depolarization and synaptic release of glutamate. This dual input requirement, together with their slow activation and deactivation kinetics allows NMDARs to integrate and decode incoming synaptic activity. The high Ca²⁺ permeability of NMDARs enables them to transduce specific synaptic input patterns into long-lasting alterations in synaptic strength.

13.2.2. Long-Term Potentiation and Depression

The strong depolarization required to remove Mg block from synaptically localized NMDARs can be achieved in several ways. High frequency synaptic inputs may allow the excitatory postsynaptic potentials (EPSPs) generated by AMPAR activation to accumulate and build over time. This phenomenon underlies the paradigm of long-term potentiation (LTP) discovered in 1973 by Bliss and Lomo [15] in which a short burst of high frequency synaptic input (15 Hz for 15 sec or 100 Hz for 3 sec) results in strengthening of excitatory synapses for a prolonged period (hours to days).

The role of NMDARs in LTP induction in the hippocampal CA1 area is well documented [16]. LTP is NMDAR-dependent in many other regions of the brain [17], although NMDAR-independent LTP has also been observed [18,19]. The synaptic strengthening observed during LTP has been attributed to two major mechanisms: (1) phosphorylation of AMPARs, resulting in an increased open probability and (2) enhanced trafficking of AMPARs to the postsynaptic membrane (see Chapter 8 for details).

Long-term depression (LTD), the counterpart of LTP, may be experimentally induced by prolonged low frequency (0.5 to 3 Hz) stimulation of excitatory synapses [20]. Induction of LTD is also NMDAR-dependent in the hippocampal CA1 region and, like LTP induction, requires Ca²⁺ influx through NMDARs [21]. LTD can also be induced by other mechanisms including stimulation of metabotropic glutamate receptors [22]. The mechanism by which NMDA-dependent LTD reduces synaptic strength is to reverse the effects of LTP: dephosphorylation of AMPARs, thus reducing their open probability [23] and removal of AMPARs from the synaptic plasma membrane by endocytosis (see Chapter 8). In these examples of homosynaptic LTP and LTD, the synapse that receives the low/high frequency input is weakened/strengthened. However, synaptic strength can also be altered in either direction if a single synaptic input is coupled with a postsynaptic depolarization, resulting in heterosynaptic LTP, which has been proposed as a model for associative memory. Postsynaptic depolarizations can occur by various mechanisms. In many cases, neuronal dendrites play critical roles in the generation and processing of these plasticity-inducing signals.

13.2.3. Dendritic Function

Dendrites exhibit active conductances, mediated by voltage-gated Na and Ca channels, as well as the NMDAR itself, which allow them to generate a back-propagating action potential (bAP) [24,25]. Dendritic bAPs have a longer duration than axonal spikes and therefore permit the removal of Mg²⁺ block from NMDARs. Any glutamate release occurring during a bAP-induced depolarization may therefore result in Ca²⁺ influx through activated NMDARs and subsequent alterations in synapse strength [26,27]. Because bAPs take time to propagate down distal dendrites, efficient activation of NMDARs by this mechanism requires precise timing of synaptic input relative to bAP generation.

The direction and magnitude of the resulting alteration in synaptic strength depend critically on the temporal relationship between the two processes [28]. The precise timing dependence of presynaptic firing relative to the firing of the postsynaptic neuron, has given rise to a model of long-term synaptic plasticity called spike timing-dependent plasticity (STDP) [29–35]. This promising model extends the LTP–LTD paradigm by proposing that the coupling strength between neurons depends on the degree of correlation of their spiking activities [36].

Thus, STDP implements an important and long ignored aspect of the most influential model for synaptic strengthening formulated in 1949 by Donald Hebb [37]. In Hebb’s words (relevant wording in bold): “When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A’s efficiency, as one of the cells firing B, is increased.” Hebb’s insistence that the presynaptic neuron causes the postsynaptic neuron to fire an action potential appears visionary.

The voltage dependence and ion selectivity of NMDARs provide them with regenerative properties that allow the generation of dendritic action potentials (NMDA spikes) that do not originate from bAPs, but from highly synchronized excitatory synaptic inputs to a region of the dendrite [38–42] (see Section 9.4). The ability to generate NMDA spikes endows dendrites with interesting nonlinear computational capabilities, the impact of which is only now beginning to become clear [43–45].

13.2.4. Privileged Ca²⁺ Ions

The long-lasting effects on synaptic efficacy resulting from activation of NMDARs depend on Ca²⁺ influx into the postsynaptic compartment (see Chapter 9 for a review of NMDAR-mediated Ca signaling in dendritic spines). It is intriguing that both synaptic strengthening and weakening are mediated by Ca²⁺ influx through NMDARs. Several hypotheses have been proposed to explain this phenomenon. The magnitude of the rise in Ca²⁺ concentration may be an important determinant. The direction of change in synaptic strength may also be determined by the temporal properties of the internal Ca concentration changes [46].

Alternatively, there may be a spatial difference in calcium rises associated with by LTP and LTD. NR2B receptors have been proposed to be localized extrasyn-aptically [47] (although this is not a universal finding [48]). Such spatial separation of NR2A and NR2B receptors would allow Ca²⁺ ions to activate distinct signaling complexes [49,50]. An earlier proposal that the NR2 subtype determines the direction of synaptic change, with activation of NR2A- and NR2B-containing NMDARs resulting in LTP and LTD, respectively [51,52], has not held up in later experiments [53,54]. More work is needed to define the molecular mechanism by which spatiotemporal differences in Ca concentrations resulting from activation of synaptic and extrasynaptic NMDARs of varying subtype compositions lead to bidirectional, long-term changes in synaptic efficacy.

Ca²⁺ ions entering the neuron through the NMDAR are privileged, because they are able to act locally on large signal transduction complexes associated with synaptic NMDARs, which consist of calcium-dependent enzymes, second messengers, protein kinases and phosphatases, scaffolding proteins, cytoskeletal elements, GTP binding proteins and their regulators, and adhesion molecules [55–57]. NMDAR-mediated calcium influx activates these complexes, generating signal transduction cascades that produce long-lasting changes in synaptic function and structure [57–63].

The ability of an NMDAR to integrate and decode synaptic inputs and generate or amplify dendritic spikes depends on its kinetics of activation, deactivation, and desensitization. The amount of Ca influx is also largely determined by kinetics, in particular the rate of deactivation. Subunit composition is an important determinant for NMDAR (de-)activation kinetics and intracellular calcium dynamics, as discussed in Chapter 9. The amount of Ca influx resulting from NMDAR activation is further modulated by other external physiological factors including pH, zinc ions, and polyamines that are discussed in Chapters 11 and 12. The next section considers the roles of the two coagonists, glutamate and glycine, in activation of NMDARs.

13.2.5. Coagonists Glutamate and Glycine

NMDARs are unique among ligand-gated ion channels in that their activation requires binding of two coagonists, glycine and L-glutamate [64–66]. Glycine is sometimes cited in the literature as an NMDAR modulator—to set it apart from the agonist L-glutamate, but as explained below, their binding sites are structurally similar and seem to play equivalent roles in receptor activation. Physiologically, however, glycine and glutamate have distinct functions. While L-glutamate is released from specific presynaptic terminals, low concentrations of ambient glycine present at the synapse are thought to be sufficient to allow receptor activation. Interestingly, a recent paper suggests that D-serine released by astrocytes is the endogenous glycine site agonist in certain brain regions, allowing glial cells to actively control synaptic metaplasticity [67].

Because glycine plays a more modulatory role in vivo [68,64], while glutamate is the ‘active’, released neurotransmitter, the glycine and glutamate binding sites on the NMDAR represent two distinct therapeutic targets. Particular efforts have been devoted to the development of partial agonists for the glycine site [69–76] that may act as negative modulators of NMDAR function by competitively displacing the full agonist glycine from its binding site. The pharmacology of NMDARs is described in detail in Chapter 12.

An interesting but poorly studied property of glycine is its ability to partially activate NMDARs in the absence of glutamate. This action has been observed in NMDARs containing NR1 and NR2A subunits expressed in Xenopus laevis oocytes, where 10 μM glycine alone activated the receptor by a few percent [77]. Due to the inhibition of this effect by the competitive glutamate-site antagonist APV, it has been suggested that glycine may act as a partial agonist at the glutamate binding site [77]. It is not clear whether ambient glycine in the CNS also exhibits this property, which would allow a small amount of Ca influx through NMDARs independent of synaptic input during large depolarizations that remove Mg²⁺ block. Also, it is not known whether D-serine, which can be released from astrocytes, has the same ability as glycine to partially activate NMDARs in the absence of glutamate.

The importance of ambient glycine binding to NMDARs for cognitive function is underscored by experiments with transgenic mice carrying mutant alleles of the NR1 subunit [78]. Mice that express NR1 subunits with lowered glycine affinity display a spectrum of cognitive and learning defects including nonhabituating hyperactivity, increased stereotyped behavior, disruptions of nest building activity, and poor performance in the Morris water maze [78], a measure of cued learning. The behavioral phenotypes of these glycine-insensitive mutant mice resemble some of the positive and negative symptoms displayed by schizophrenia patients, consistent with NMDAR hypofunction as one of the leading hypotheses for schizophrenia [79–82]. Section 13.3.7 discusses the (limited) successes of glycine site partial agonists in the clinic. The next two sections describe studies of the behavior of individual NMDAR channels.

13.2.5. Single Channel Behavior

Ion channels are unique among proteins in that their behavior can be studied at the level of a single molecule. This important experimental paradigm was made possible by the advent of the patch clamp technique pioneered by Neher and Sakmann in the mid 1970s [83]. In single channel recordings, individual ion channels are seen to alternate stochastically between two states, open and closed. No measurable ion flux is present in the closed state whereas the open state is characterized by a constant, channel-specific conductance. Activation of an ion channel results from an increase in the probability of being in the open state, not a change in its single channel conductance.

Analysis of single channel behavior provides detailed information about mechanisms of action that cannot be extracted from macroscopic currents measured in whole cell recordings [84]. Early single channel studies suggested that Mg²⁺ ions inhibited NMDARs via an open channel block mechanism, as indicated by the presence of short closings within otherwise stable open periods [12]. The allosteric modulator spermine exerts multiple effects on single channel behavior, increasing opening frequency at low concentrations, while decreasing single channel conductance and mean open time at higher concentrations, suggesting the existence of two binding sites [85]. Single channel analysis of ethanol inhibition of NMDARs (see Chapter 4) indicated that ethanol exerts its effect through an allosteric mechanism by reducing agonist efficacy [86].

Single channel behavior of NMDARs from hippocampal CA1 neurons was studied using very low glutamate concentrations to improve temporal resolution of individual glutamate binding events [87]. Openings resulting from individual receptor activations showed surprising complexity: they consist of a long cluster of bursts of openings. Furthermore, the NMDARs appeared to have different gating modes, occasionally entering periods of very high open probability [87]. These results demonstrated that the slow deactivation kinetics of the NMDAR result from intrinsic gating properties.

Single channel analysis can also provide more insight in how NMDARs function at the synapse. In response to a brief pulse of glutamate, mimicking synaptic release, NMDARs activate slowly over hundreds of milliseconds and continue activating long after all glutamate has been removed from the synaptic cleft, thereby briefly “memorizing” the occurrence of a synaptic input. Single channel analysis of NR1 and NR2A receptors indicates that after a brief pulse of glutamate, receptors enter a high affinity closed state from which either channel opening or agonist unbinding occurs with approximately equal probability [88]. A single synaptic event is therefore expected to only partially activate NMDARs. Consequently, a closely spaced second pulse of agonist is able to further increase the open probability, endowing the NMDAR with an ability to decode synaptic input frequency (Figure 13.1).

FIGURE 13.1

Decoding of synaptic inputs by NMDARs. Because of their high glutamate affinity and relatively slow activation and deactivation kinetics, NMDARs can “memorize” recent low frequency synaptic inputs that alone cannot elicit ion permeation (more...)

Comparison of the single channel behavior of NR2A- and NR2B-containing NMDARs revealed several kinetic differences in these receptor subtypes. NR2A receptors respond faster to brief synaptic-like pulses of glutamate and reach higher open probabilities [89]. It has been proposed that these differences in channel gating kinetics result in preferential opening of NR2A-containing receptors during high frequency synaptic inputs that stimulate LTP. Conversely, NR2B receptors are associated with lower frequency inputs that cause LTD. However, the NR2A/LTP and NR2B/LTD connections were discredited by recent experiments [54].

13.2.6. Subconductance Levels, Permeation, and Gating

One of the more intriguing aspects of NMDAR single channel behavior is that it is not strictly binary. Channels occasionally fail to open or close fully and instead visit intermediate conductance levels. Such subconductances levels (sublevels) have been observed in both native [90] and recombinant glutamate receptors [91–93]. Typically, these sublevels are visited during transitions between closed and fully open states, not as isolated opening events. NR2A- and NR2B-containing receptors display similar short-lived sublevels whose conductance is approximately 80% of the main open state. In contrast, NR2C- and NR2D-containing receptors briefly visit a 50% sub-level and a fully open state with approximately equal probability.

Subconductance levels have been observed in virtually every type of ion channel, although the number of levels, stability, and abundance vary widely. Sublevels have been well-characterized in K channels, which are evolutionarily related to NMDARs [94]. Our laboratory has described sublevels observed during open–closed transitions in the voltage-gated K channel K_v2.1 (drk1),95 and we have proposed a model that attributes sublevels to heteromeric pore conformations visited when one, two, or three of the four subunits move to the open conformation (Figure 13.2). Because channel opening appears to be a strongly cooperative process, the heteromeric pore conformations are predicted to be highly unstable and the associated sub-levels may be very short-lived.

FIGURE 13.2

Subunit-subconductance model for NMDAR gating. The permeation pathway of the NMDAR is formed by two NR1 and two NR2 subunits (A). The conformation of each subunit alters between a “closed” conformation that does not support permeation (more...)

One prediction of the subunit–sublevel hypothesis is that sublevels should be more abundant when a receptor or channel is incompletely activated, which was experimentally confirmed for K_v2.1.95 In a more stringent test of this hypothesis, a K_v2.1 tandem dimer combining two K channel subunits with different activation thresholds was created. Single channel behavior of this tandem dimer at potentials between the two thresholds was dominated by two sublevels, whose kinetics and voltage dependence indicated that they resulted from the activation of one and two subunits [96]. These results support the subunit–subconductance hypothesis, which implies that gating and permeation are strictly coupled. In this model, the selectivity filter is directly responsible for opening and closing the channel [97,98]. A specific mechanism has been proposed for how the selectivity filter may function as the channel gate, in which the filter alternates between two conformations with high and low affinity for the permeant ion [98]. Experimental support for an affinity-switching selectivity filter has recently been shown by NMR experiments using the KcsA K channel [99].

Single channel analysis of AMPARs and NMDARs also provided evidence for a strong coupling between permeation and gating, as well as support for a link between sublevels and partial receptor activation. Two sublevels associated with partially activated GluR3 AMPARs were proposed to result from the activation of two and three subunits [100]. Mutations in the selectivity filter were shown to stabilize sublevels both in K channels [101,103] and NMDARs [102,104]. The ion selectivity of the sub-levels in these mutant NMDARs is different from that of the fully open state [104], as was reported for the Shaker K channel [101], thereby providing a direct causal linkage between subconductance gating and the selectivity filter.

Under bi-ionic (Na⁺/Cs⁺) conditions, gating of these mutant NMDARs becomes strongly asymmetric, with sublevels visited either during openings (external Cs⁺) or during closings (external Na⁺) [104], providing additional evidence for a strong coupling of gating and permeation. Finally, partial agonists were shown to determine the open probability of subconductance levels in the GluR2 AMPAR [105], and a model was suggested which is identical to the subunit–subconductance model (Figure 13.2) proposed by Chapman and VanDongen in 1997 [97,95]. Section 13.3.7 addresses the structural basis of partial agonism in more detail. The next section discusses the structures of NMDARs and their mechanisms of activation.

13.3.1. Domain Structure of Subunits

Ionotropic glutamate receptor subunits are organized into four discrete functional domains (Figure 13.3): an extracellular N-terminal domain (NTD), a ligand binding domain (LBD), a pore forming transmembrane region, and an intracellular C-terminal domain. Hydrophobicity analysis performed after expression cloning of the first iGluR (AMPAR GluR1) [109] identified four hydrophobic segments (M1 through M4), the first three of which are closely spaced and separated by a long linker from the fourth segment.

FIGURE 13.3

Domain structure of the NMDAR. A: Modular nature of linear amino acid sequences of NMDAR subunits: four hydrophobic domains (M1 through M4), two ligand binding domains (S1 and S2), and amino and carboxy terminal domains (NTD, CTD). B: Membrane topology (more...)

The three-plus-one hydrophobicity profile is very similar to previously characterized ligand gated ion channels that include the nicotinic acetylcholine receptor (nAChR) family, GABA_A, glycine, and 5-HT3 receptors. It was therefore initially assumed that iGluRs would have the same membrane topology as these receptors, in which each hydrophobic domain is thought to form an α helix that crosses the membrane once [109,110]. The N-termini of all ligand-gated ion channels contain signal peptides, ensuring their extracellular localization. Therefore, in the membrane topology model for the nAChR super-family, the linker between the third and fourth transmembrane segment is predicted to be cytoplasmic.

The similarities of the nAChR and iGluR hydrophobicity profiles were later proven to be “red herrings.” A functional N-glycosylation site was identified in the M3–M4 linker of the GluR6 kainate receptor [111,112]. Since the enzymes responsible for sugar modification of asparagine residues reside in the lumen of the Golgi, N-glycosylation is a reliable marker of extracellular localization. Initially, a new topology model was proposed with an additional transmembrane segment in the middle of the M3–M4 linker [111,112]. However, functional N-glycosylation was also observed in the M3–M4 linker of kainate receptors from goldfish brain, and modification was not affected by deletion of the M2 segment, prompting the authors to conclude that M2 does not cross the membrane and the M3–M4 linker is extracellular [113].

Similar results were obtained from N-glycosylation studies of the GluR1 AMPAR and NR1 NMDAR subunits, confirming that the M2 segment does not cross the membrane [94,114]. Mutagenesis experiments on the NMDAR NR1 subunit identified residues in the M3–M4 linker critical for binding of glycine site ligands, indicating it must be extracellular [115]. An identical conclusion was reached by exchanging domains between the GluR3 AMPAR and GluR6 kainate receptor [116]. These studies also observed that two glutamate receptor domains display homology to bacterial periplasmic amino acid–binding proteins for which X-ray crystal structures already existed. This homology was first noted by Nakanishi et al. [117] and extended by O’Hara et al. [118] The first domain (S1) immediately precedes M1, while the second (S2) comprises most of the M3–M4 linker [115,116] (Figure 13.3A).

An extracellular location of the M3–M4 segment implies that the first three hydrophobic domains must span the membrane an even number of times. This discrepancy was explained by the significant amino acid sequence homology discovered between the M2 segments of iGluRs and the pore forming P regions of K channels [94], which fold as a reentrant hairpin loop. Figure 13.3 illustrates the current model for the membrane topology of NMDAR subunits [94,114].

The extracellular domains come together to form a ligand binding domain (LBD) consisting of the S1 segment preceding M1 and the S2 segment sandwiched between M3 and M4. The NTD is located N terminal to S1 and exhibits homology with metabotropic glutamate receptor (mGluR) binding domains, suggesting that it forms a ligand binding structure distinct from the agonist binding S1–S2 domain. In the NR2 subunits, the NTD has been proposed to bind zinc ions (NR2A) or polyamines (NR2B), both of which modulate NMDAR function (see Chapters 11 and 12). The ligand for the NTD of the NR1 subunit, if one exists, is currently not known.

13.3.2. Structure of the Ligand Binding Domain

The modular nature of glutamate receptors prompted experiments to isolate the ligand binding module as a separate soluble protein. Early experiments with the GluR4 AMPAR demonstrated that the isolated LBD consisting of the S1 and S2 domains connected by a short peptide linker retained the pharmacology of the full-length receptor, binding antagonists and agonists with normal affinity [119]. This approach enabled the crystallization of the GluR2 AMPAR LBD in complex with kainate [120], illustrated in Figure 13.4A.

FIGURE 13.4

(See color insert following page 212.) Crystal structures of GluR2 and NR2A ligand binding domains. X-ray crystallographic structures are shown for the isolated ligand binding domains (LBDs) of the AMPAR GluR2 in complex with kainate (A) and the NMDAR (more...)

The bilobate structure consists of two globular domains or lobes (D1 and D2), connected by a flexible hinge, and bears a striking similarity to the bacterial periplasmic amino acid–binding proteins. The S1 segment forms lobe 1 (purple) and hinge 1 (blue) that connect to the first transmembrane segment M1 in the intact receptor. S2 forms lobe 2 (green) and hinge 2 (yellow), which crosses back to lobe 1. Two helices (J and K, gray) follow hinge 2, running across the backs of the two lobes and connecting to the M4 segment. The kainate agonist is sandwiched between the two domains and forms hydrogen bonds with both lobes, thereby stabilizing the closed cleft conformation.

GluR2 structures were obtained in the absence of ligand (apo) and in complex with an antagonist (DNQX), the partial agonist kainate, and full agonists AMPA and glutamate [121]. Separation of the two domains was significantly increased in the apo state, and all ligands tested produced some amount of cleft closure. The degree of domain closure increased as follows: apo < DNQX < kainate < AMPA = glutamate. This result led to the suggestion that agonist-induced domain closure activates the channel and that degree of domain closure determines the extent of activation. Although the competitive DNQX antagonist still produces a small amount of domain closure relative to the apo state, it apparently is not sufficient to activate the channel. Section 13.3.7 discusses the structural basis for partial agonism in more detail.

Crystallographic structures for the S1–S2 LBD were obtained for additional members of the iGluR family including the GluR5 and GluR6 kainate receptors [122], GluR0 [123], a prokaryotic glutamate-gated K channel, and the NMDAR subunits NR1 [124] and NR2A [125]. All LBDs have the same basic structure, although certain conformational differences exist between the lobes and hinge regions. Figure 13.4C shows the LBD structure of the NMDAR NR2A subunit in complex with glutamate; this can be compared with the structure of GluR2 in Figure 13.4A. The availability of high-resolution crystal structures greatly facilitates the design of specific point mutations and the interpretations of resulting phenotypes.

13.3.3. Affinity and Efficacy

The interaction of a ligand with its receptor is characterized by two fundamental pharmacodynamic properties: affinity and efficacy. Affinity measures how tightly a ligand binds to a receptor and is characterized by the equilibrium dissociation constant, K_D. Efficacy measures how effectively a ligand, once bound, activates the receptor. Antagonists have no efficacy. The efficacies of partial agonists are lower than those of full agonists binding to the same site. The processes of agonist binding and receptor activation are strongly coupled, with the former initiating the latter.

This coupling complicates the interpretation of phenotypes caused by mutations. A point mutation that causes a shift in the concentration–response curve can do so without affecting binding, because the midpoint of the curve (EC₅₀) is dependent on both the binding equilibrium (affinity) and the activation process (efficacy) [126]. Examples are provided by the previously published mutations in the pore of the nicotinic acetylcholine receptor, many of which altered acetylcholine sensitivity by affecting open state stability [127,128]. Another complication is that a change in efficacy caused by a mutation does not necessarily result in a measurable change in maximum response. When efficacy is very high, even substantial changes in efficacy exert minimal effects on maximum response. Moreover, when a change in maximum response is observed for a mutation, it is often difficult to exclude altered protein folding efficiency, protein stability, and inefficient expression as possible explanations.

To solve this dilemma, we developed an approach that allows the evaluation of affinity and efficacy roles for individual amino acid positions in a receptor. First, the position to be investigated is mutated to cysteine, which ideally should have little effect on receptor pharmacology. Second, concentration–response curves are collected for a full agonist and a partial agonist before and after covalent modification of the introduced cysteine [129]. Due to their small size, MTS (methanethiosulfonate) compounds are very useful for this purpose. These experiments yield values for the EC₅₀ and intrinsic activity (α), a measure of relative efficacy, before and after covalent modification of the cysteine.

The in situ mutagenesis produced by MTS modification does not alter the population of receptors studied because it typically takes less than a minute to complete; therefore, any change in maximum response must result from an alteration in efficacy. Using a partial agonist guarantees that any changes in efficacy caused by MTS modification will reflect a change in intrinsic activity. Based on the values of EC₅₀ and α before and after MTS modification, it is possible to quantitate changes in affinity and efficacy caused by the modification [129].

Using this approach, Kalbaugh et al. assigned affinity and efficacy roles to positions in the LBDs of the NR1 and NR2A NMDAR subunits. In situ mutation of residues in direct contact with bound ligand affected both efficacy and affinity, while positions that stabilize the closed cleft conformation without a direct ligand interaction contributed only to efficacy [129]. These results provide a molecular basis for the tight coupling of agonist binding and receptor activation. The same residues that mediate stabilization of the closed cleft conformation by the bound ligand are also critical for ligand binding to the apo state, resulting in an agonist-bound open cleft conformation.

Crystallization of the isolated and soluble S1 and S2 LBDs for several iGluRs has significantly increased our understanding of the molecular bases of ligand binding and receptor activation mechanisms. The structure and function of the membrane domain (M1 through M4) will be discussed next.

13.3.4. Structure of the Membrane Domain: K Channel as Model

High-resolution structures are available for many LBDs of iGluRs but intact receptors, which are large integral membrane proteins, have thus far resisted crystallization. However, the M1–M3 segment appears to be structurally and functionally related to K channels, for several of which X-ray structures do exist [130–132]. Wo and Oswald (1994) originally hinted at a structural relationship between K channels and iGluRs by suggesting that the kainate receptor M2 segment folds back into the membrane similar to the P regions of K channels [113] (Figure 13.5A). N-glycosylation experiments in the M3–M4 linker by Wood et al. (1995) were prompted by a high degree of amino acid homology between NMDAR M2 segments and K channel P regions [94].

FIGURE 13.5

Relationship of glutamate receptors and K channels. A: Comparison of membrane topologies of ionotropic glutamate receptors, the prokaryotic K channel KcsA, and voltage-gated K channels. A common pore forming motif (M1–M2–M3, TM1-P–TM2, (more...)

Interestingly, the K channel GYGD signature sequence that forms part of the selectivity filter is replaced in NR1 with GIGE (Figure 13.5B). However, transplantation of the selectivity filter sequence TVGYG from K channels into iGluRs failed to transfer K selectivity to these channels, although functional channels were obtained and many pore properties were altered [133]. If the transplanted TVGYG regions fold similarly in iGluRs and K channels, as suggested by the functionality of the chimeras, it is likely that permeating ions are coordinated by backbone carbonyls in both channel families. Because iGluRs do not distinguish between Na⁺ and K⁺ ions and some members conduct divalent Ca²⁺ ions, the atomic distance between backbone carbonyls forming the oxygen cage should be significantly more flexible and dynamic in these nonselective cation channels.

The amino acid homology in the pore-forming regions prompted us to suggest that K channels and iGluRs have a common evolutionary ancestor [94]. This is supported by the identification of a “missing link,” the prokaryotic GluR0 glutamate receptor that has a pore-forming domain with high homology to K channel P regions and exhibits K⁺ selectivity [134]. Using the GluR0 amino acid sequence as a guide, the homology of K channels and iGluRs can be extended to the transmembrane domain following the pore region: M3 in iGluRs and S6 or TM2 in K channels [135] (Figure 13.5B). In K channels, the S6 and TM2 segments undergo conformational changes during channel activation [136–138]. Evidence suggests that the iGluR M3 segment may play a similarly important role during receptor activation.

13.3.5. Role of the M3 Domain

The M3 segment contains a 9-amino acid sequence (SYTANLAAF) that is highly conserved among glutamate receptors; many cysteine substitutions in this region display state-dependent accessibility [135,139]. Altered residue accessibility often indicates a conformational change [140], suggesting that the SYTANLAAF region moves in response to receptor activation; thus, M3 has been proposed to function as a transduction element, coupling ligand-binding to channel gating [135]. A recent study suggests that M3 is the only transmembrane domain contributing to the deepest portion of the pore, supporting a prominent role in gating [141].

The functional importance of the SYTANLAAF motif was originally identified in the GluRδ2 receptor, an orphan receptor with homology to the iGluR family, but no known glutamatergic agonist. A single-point mutation, A8T, was found to cause an inherited neurological defect in mice [142]. Known as lurcher, the phenotype, characterized by ataxia and neurodegeneration, is caused by constitutive activation of δ δ2 receptors, which produces excitotoxicity and apoptosis of cerebellar Purkinje cells. Introduction of the lurcher mutation in GluR1 and GluR6 produced some constitutive activity, increased agonist potency, slower deactivation, and conversion of an antagonist into an agonist [143–145]. The same substitution in NR1 and NR2A did not produce constitutive activity, but exhibited very slow deactivation; interestingly, increased glutamate potency was observed only in the NR1 mutant [143].

Similar to the lurcher phenotype, constitutive activity and/or current potentiation were observed upon thiol modification of substituted cysteines at several positions within the SYTANLAAF region [135,139,146–148]. Of particular interest is the A7C substitution in NR1 and NR2 that appears to lock the channel in a fully activated conformation upon modification [135,139,149]. A7C can be modified only after receptor activation and accessibility exhibits a linear correlation with agonist efficacy [135], strongly suggesting a role for M3 in coupling ligand binding to channel gating. Conservation of the A7C phenotype is observed between NR1 and NR2 and across NR2 subtypes, implying functional conservation of M3 among all NMDARs [139].

Several lines of evidence suggest that the NR1 and NR2 M3 domains may play distinct roles in receptor activation. NR1 and NR2 contribute differently to the M2 loop [150] that forms the inner pore of the channel, and copper coordination of substituted cysteines indicates that their M3 domains may be staggered by a full α helical turn [151]. Consistent with a dimer-of-dimers arrangement, the extracellular vestibule of the homomeric AMPAR exhibits two-fold rotational symmetry, as opposed to the four-fold symmetry of K⁺ channels [152]. Recent work evaluating the voltage dependence of cysteine modification suggests that the NR1 M3 domain contributes mainly to the deep portion of the pore, while the NR2 M3 comprises more of the shallow extracellular vestibule [141].

These subtle structural differences among subunits may be critical for NMDAR activation; for example, kinetic modeling with partial agonists revealed that distinct fast and slow pregating conformational changes are mediated by NR1 and NR2, respectively [153]. Residues displaying state-dependent accessibility were found to cluster together on helical net diagrams, opposite from putative pore-lining residues, suggesting that positions exhibiting lurcher-like phenotypes may be located at dynamic interfaces between transmembrane segments [141]. Thus intersubunit interactions at the transmembrane level may be critical for NMDAR activation.

13.3.6. Dimerization of Ligand Binding Domains

One of the most intriguing findings from iGluR structural studies is the tendency of S1S2 LBDs to crystallize as homomeric dimers, initially observed in a set of GluR2 crystal structures [121]. In vitro, isolated iGluR LBDs dimerize with a dissociation constant in the millimolar range, most likely due to high protein concentrations. This effect may be mimicked in vivo by association of the N terminal domains (NTDs) that occurs at lower protein concentrations [154]. In AMPARs and kainate receptors, the dimer interface is formed exclusively by hydrogen bonds, salt bridges, and hydrophobic interactions between the D1 domains, burying 900 to 1600 Ų of solvent-accessible surface area [121,155]. The D2 domain linking the LBDs to the ion channel portion of the receptor is free to change conformation, suggesting a mechanism by which LBD conformation may be coupled to channel gating.

The functional importance of the LBD dimer interface was initially demonstrated via site-directed mutagenesis, which was used to modulate interactions at the GluR2 interface. Single-point mutations were identified which either attenuated or increased desensitization. Crystallography, ultracentrifugation, and electrophysiological studies established that dimer stability inversely correlates with extent of desensitization [156]. Subsequent studies confirmed and extended this paradigm in AMPARs and kainate receptors [157,158].

Domain–domain (D1-D1) separation associated with GluR2 desensitization has been estimated to be between 12.4 and 16.2 Å at the top of the interface, based on modification of cysteine interface mutants via bifunctional crosslinkers. Identification of a GluR2 mutant stabilized in the desensitized state (S729C) led to a crystal structure depicting a relaxed, destabilized dimer interface [158]. Crystal structures of GluR2 in complex with several positive allosteric modulators revealed that these molecules exert their potentiating effects via interface stabilization. Cyclothiazide, which blocks receptor desensitization, binds near the edge of the interface, while aniracetam stabilizes the open state by binding near the LBD hinge [156,159].

The NR1 subunit repeatedly crystallized in monomeric form until the recent cocrystallization of NR1 and NR2A revealed the existence of a heteromeric LBD dimer interface [125]. In contrast to other iGluRs, the LBD dimer interface of the NMDAR contains both D1 and D2 interactions, providing additional opportunities for intersubunit coupling. Furthermore, a critical tyrosine residue in NR1 (Y535) occupies a site homologous to the GluR2 aniracetam binding site and the sizes of substituted amino acids at this position were found to inversely correlate with deactivation rate. Thus, Y535 appears to function as an endogenous positive allosteric modulator. This key difference in interface stability has been proposed to underlie the slow deactivation of NMDARs required for their role in synaptic transmission [160], suggesting a novel modulatory site for NMDAR-based therapeutics.

The NTDs are clamshell-shaped structures, known to interact with each other and modulate the LBDs, suggesting a possible extension of the interface model [154]. Zn²⁺ binding to the NR2A NTD is allosterically coupled to the glutamate-binding domain. Similar results were obtained for ifenprodil in NR2B [161,162]. An NTD dimer interface may provide the stability for binding of negative modulators to effectively couple to LBDs.

13.3.7. Structural Basis of Partial Agonism

Crystal structures of the GluR2 S1S2 domain, in complex with a range of full and partial agonists, established a strong correlation between agonist efficacy and degree of LBD closure. A similar relationship was observed in the GluR6 kainate receptor [105,122]. These results led to a structural model of iGluR activation, in which agonist binding induces LBD closure by rotating D2 toward D1, separating the linker regions and promoting channel opening. According to this hypothesis, partial agonists induce less domain closure and consequently less linker separation, slowing a subunit-specific pregating conformational change [163].

Engineering steric clashes in the LBD, thus destabilizing the closed cleft state, reduces agonist efficacy and apparent affinity in NR2, GluR2, and GluR6 [164–166], while elimination of agonist–cleft clashes in GluR2 reportedly slows receptor deactivation and increases affinity and efficacy [167]. Stabilization of the NR2 closed cleft state via modulation of an endogenous D1–D2 interaction can also increase open probability, kinetically linking LBD closure and channel gating [168]. Interestingly, crystal structure data obtained from mutagenesis studies has not always followed the cleft closure–agonist efficacy correlation; GluR2 L650T, for example, stabilizes the closed cleft state, increasing kainate-induced efficacy and degree of domain closure, but the AMPA-bound structure revealed both partially and fully closed conformations [167]. A cleft stabilizing GluR6 D1–D2 interaction dramatically increased glutamate sensitivity and slowed deactivation when introduced in GluR2, but did not affect cleft closure. Full agonist-bound GluR6 is almost 6° more closed relative to GluR2 [164].

Crystal structures obtained for the NR1 subunit revealed that full and partial agonists adopt similar degrees of domain closure, indicating that partial agonist action at an NMDAR may follow a different structural paradigm from other iGluRs [169]. Because no partial agonist-bound crystal structure is presently available for the NR2 subunit, it is not yet known which model describes the behavior of the NR2 LBD. Both structural and molecular dynamics studies suggest that the LBD hinge region, particularly the second interdomain β strand, changes conformation according to agonist efficacy [169,170]. Helix F that forms part of the LBD dimer interface has been implicated in coupling agonist efficacy to channel gating in NR2A [163]. NR1-site partial agonism has also been correlated with increased inter-pocket motion, and both glycine and DCS can reportedly move within the pocket without affecting domain closure [170].

Incomplete and unstable cleft closures have been proposed to affect single channel gating similarly [166], suggesting that degree of cleft closure may simply be a physical readout of closed cleft stability. Thus, closed cleft stability, and not simply a physical change in conformation between D1 and D2, may be the principal determinant of agonist efficacy at iGluRs.

Understanding the structural basis of partial agonism is of both scientific and clinical interest, since NMDAR hypofunction has been shown to cause cognition and memory defects in animal models [171,172]. Glutamatergic dysfunction has been implicated in the pathophysiology of schizophrenia [173] and receptor augmentation may prove therapeutically valuable. Although both glycine and D-cycloserine (DCS), a well-tolerated partial glycine site agonist, initially exhibited some efficacy in the treatment of schizophrenia [174–177], a recent multisite double blind randomized trial found both glycine and DCS ineffective for treating negative symptoms and cognitive impairments [178]. DCS has been shown to enhance learning and memory performance in adult and aged rodents [179–182] and monkeys [183], and is undergoing testing as a cognitive enhancer for treating Alzheimer’s disease [184–186], head trauma [187], and fear extinction [188–193]. Reported subtype-specific partial agonism at the NMDAR also opens the possibility of individually targeting NR2 subtypes that exhibit regional and developmental variations in expression [89].

13.3.8. Positive Allosteric Modulators

Few positive modulators, endogenous or otherwise, have been found to act on the extracellular portions of NMDARs. Polyamines such as spermine and spermidine exert both inhibitory and stimulatory effects, depending on concentration and subunit composition (see Chapter 11). Polyamines are generally found within cells and whether they participate in physiological regulation of NMDARs [194] remains unclear. Pregnenolone sulfate (PS), a neurosteroid synthesized de novo in the CNS, also exhibits subunit-specific potentiation and inhibition [195,196]. A recent study suggests that PS is released in an activity-dependent manner and may attain synaptic concentrations in the micromolar range, suggesting a possible endogenous modulatory role [197].

PS potentiates NR2A- and NR2B-containing NMDARs via increases in channel open probability; however, due to a glutamate-induced reduction of receptor affinity for PS, the effect is relatively transient [198]. In contrast, NMDARs containing NR2C or NR2D subunits are inhibited by PS, and all four NR2 subtypes are inhibited by pregnanolone sulfate, a closely related neurosteroid [196]. Planar and bent ring structures appear to favor stimulation and inhibition, respectively, while both necessitate a negatively charged C3 moiety [199]. PS binding has been localized to a steroid modulatory domain, SMD1, comprised of the M4 transmembrane domain and helix J/K located in the S2 segment of the ligand binding domain [200]. Since PS is a charged molecule and potentiation displays no voltage dependence [201], helix J/K is a more likely candidate for direct binding. This region of S2 has recently been shown to form part of the LBD dimer interface, a region identified as a binding site for numerous positive allosteric modulators of AMPARs [159].

PS treatment of neuronal cultures exacerbates NMDA-induced excitotoxicity [199], while intracerebroventricular administration in mice has been shown to cause convulsions, although that effect may be due to inhibitory actions at the GABA_A receptor [202]. Hippocampal PS concentration has been correlated with preserving cognitive functions in aging animals [203] and in vivo infusion of PS enhances neurogenesis in mice [204]. PS has also been shown to increase LTP in the hippocampus [205], possibly providing a mechanism for its role as a cognitive enhancer.

Although the nonselective effects of PS may preclude its clinical use, it may serve as a promising starting point for therapeutic development. Positive allosteric modulators of AMPARS, known as ampakines, improve learning and memory and suppress symptoms of schizophrenia, depression, and ADHD in animal models. Initial studies in humans have also shown positive effects on memory and psychiatric symptoms [206]. Crystal structures of GluR2 in complex with two ampakines, aniracetam and CX614, revealed that both modulators bind within the LBD dimer interface, slowing deactivation and/or desensitization [159]. Ampakines are proposed to enhance cognitive function in a three-fold manner: expansion of cortical networks, facilitation of LTP, and upregulation of BDNF expression [206]. Based on the structural homology among iGluRs, and the role of NMDARs in LTP induction, the development of an NMDAkine is certainly within the realm of possibility.