I. Introduction

Top Next References

Since the finding by Johnson and Ascher (1987) demonstrating that glycine enhances electrophysiological responses mediated by N-methyl-D-aspartate (NMDA)^b-sensitive glutamatergic receptors, considerable interest has been devoted to this topic (for reviews, see Dingledine et al., 1990; Thomson, 1990; Huettner, 1991; Carter, 1992; Kemp and Leeson, 1993; Leeson, 1993; Wood, 1995). Although Johnson and Ascher (1987) were the first to connect the strychnine-insensitive glycine recognition site with the NMDA receptor, the strychnine-insensitive effects of glycine had been recognized earlier, indicating the diversity of presumed glycine receptors (table 1). Kishimoto et al. (1981) reported that the saturation isotherm of Na⁺-independent [³H]glycine binding in the medulla oblongata and spinal cord (but not cerebral cortex) showed biphasic features. In this assay, D-serine was surprisingly >40 times more potent as an inhibitor of [³H]glycine binding in the cortex than was L-serine, and strychnine was inactive, indicating that these sites were not related to the inhibitory glycine receptors forming Cl channels (Young and Snyder, 1974). Later, it was reported that the distribution of [³H]glycine and [³H]strychnine binding is not colocalized, but rather complementary, being strongest in the forebrain and pons/spinal cord, respectively (Bristow et al., 1986). Johnson and Ascher (1987) then published their observations, based on patch-clamp experiments in primary cultures of mouse cortical neurons, that the magnitude of NMDA responses was dependent on the speed of perfusion of agonist solution, i.e., slower flow resulted in larger responses. Subsequently, by using conditioned medium or adding exogenous glycine, it was clarified that the enhancement of NMDA responses was the result of glycine released from cultured cells. These authors reported similar augmentation when serine or alanine was used, although the differences between isomers were not established. Soon after, at the Society for Neuroscience meeting in 1987, it was reported that glycine enhanced 3',5'-cyclic guanidine monophosphate (cGMP) levels in the cerebellum in vivo after intracisternal application (Danysz et al., 1987, 1989c). This finding was soon confirmed in the same experimental model (Wood et al., 1989), and both studies indicated that the glycine site of NMDA receptors is not always saturated in vivo (see Section III.C.7. for a more detailed discussion). At the same Neuroscience Society Meeting, glycine was also shown to potentiate a slow excitatory postsynaptic potential (EPSP) between hippocampal neurons in culture (Forsythe et al., 1987), and the first glycine_B antagonist, kynurenic acid, was introduced (Kessler et al., 1987, 1989b); this is of particular interest considering that kynurenic acid is an endogenous substance (Stone et al., 1987).

Another important milestone was the report by Kleckner and Dingledine (1988) indicating that glycine is in fact probably obligatory for the activation of NMDA receptors. In turn, the term "coagonist" was proposed. This finding has very important implications, because it indicates that maximal inhibition of the glycine site should result in total inhibition of NMDA receptor activity, which in fact has been confirmed in several studies (see Section III.A.). Therefore, this opened new avenues for drug development targeted at the modulation of NMDA receptors.

One of the concerns regarding glycine have been its dual role in the central nervous system (CNS). It activates both inhibitory strychnine-sensitive receptors and the NMDA receptor-coupled site. Therefore, a selective endogenous ligand for the latter site has been sought. Hashimoto et al. (1993c) were the first to show that D-serine is present in the brain at levels that might affect the glycine_B recognition site (for review, see Hashimoto and Oka, 1997). Since that report, at least two candidates have been proposed to be endogenous agonists for the glycine site of NMDA receptors.

One of the major problems in accepting the dynamic role of glycine or D-serine in the regulation of NMDA receptor function was the apparent lack of sufficiently robust regulatory mechanisms (uptake, release, and metabolism). Also, considering the micromolar concentrations of glycine present in the extracellular fluid (ECF) (Matsui et al., 1995), some authors strongly suggested that these glycine sites must be saturated under physiological conditions. However, using coexpression of glycine transporter (GLYT)1 and NMDA receptors in oocytes, it was recently shown that very efficient buffering of local glycine concentration exists, leading to a >100-fold concentration gradient (Supplisson and Bergman, 1997). This strongly implies that the glycine concentration might be well below its K_d value for the glycine site at NMDA receptors if this transporter is expressed sufficiently at the right location in vivo.

There is no uniformly accepted term for the "NMDA receptor-coupled, strychnine-insensitive, glycine modulatory site," which is both confusing and highly impractical. For purely pragmatic reasons, we abbreviate this as the "glycine_B site," in contrast to the inhibitory, strychnine-sensitive, "glycine_A site."

	II. N-Methyl-D-Aspartate Receptors

Top Previous Next References

NMDA-sensitive ionotropic glutamate receptors probably consist of tetrameric, heteromeric, subunit assemblies that have different physiological and pharmacological properties and are differentially distributed throughout the CNS (Seeburg, 1993; Hollmann and Heinemann, 1994; McBain and Mayer, 1994; Danysz et al., 1995a; Parsons et al., 1998b). The exact subunit stoichiometry of these subunit assemblies is still a matter of debate. Although previous data were consistent with pentameric -amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) and NMDA receptors (Wenthold et al., 1992; Brose et al., 1993, Wu and Chang, 1994; Ferrer Montiel and Montal, 1996; Sutcliffe et al., 1996; Premkumar and Auerbach, 1997), recent findings indicate that both AMPA receptors (Rosenmund et al., 1998) and NMDA receptors (Laube et al., 1998) are in fact tetrameric assemblies. However, the debate is not yet over, because two studies using similar single-channel recording techniques came to very different conclusions regarding the number of NR1 subunits, with Behe et al. (1995) claiming two copies and Premkumar and Auerbach (1997) claiming three.

These receptors are positively modulated by glycine, which, as mentioned, binds to a specific, strychnine-insensitive, glycine_B site (see Section II.D.), by polyamines (spermine and spermidine), by histamine, and, under some conditions, by cations (fig. 1). NMDA receptors are coupled to high conductance cationic channels permeable to K⁺, Na⁺, and Ca²⁺ (McBain and Mayer, 1994).

View larger version (62K): [in this window] [in a new window]   Fig. 1.   Model showing that NMDA receptors are probably heteromeric assemblies of four subunits. Each subunit has four hydrophobic regions, although only three form membrane-spanning domains (TM1, TM2, and TM4). TM2 makes a hairpin bend within the membrane and forms the channel pore; the "TM" terminology is therefore inappropriate. Functional NMDA receptor complexes are formed by combinations of NR1 and NR2 subunits, which contain the glycine and glutamate recognition sites, respectively. Alternative splicing at three exons, one in the amino-terminal domain (N1) and two in the carboxyl-terminal domain (C1 and C2), generates eight isoforms for the NR1 subfamily. The NR2 subfamily consists of four individual subunits, NR2A to NR2D. Competitive antagonists such as APV probably bind to one site, which may be distinct from the agonist recognition sites but isosterically coupled in such a way as to allow competitive interactions. Glycine is a coagonist at the glycineB site and prevents Ca2+-independent receptor desensitization. MRZ 2/576 is an example of a glycineB antagonist. Polyamines such as spermine and spermidine are positive modulators but also block the channel at higher concentrations. All heteromeric and homomeric NMDA receptor subtype complexes are permeant to Ca2+, Na+, and K+. The open NMDA channel is blocked by Mg2+ and uncompetitive NMDA receptor antagonists, such as memantine and (+)MK-801, in a voltage-dependent manner, although the speed and voltage dependence of this effect depend on the antagonist affinity and the subunit composition. Ifenprodil is a selective antagonist for NR2B-containing receptors. Zn2+ is a potent, voltage-independent antagonist at NR2A-containing receptors. In addition, most NMDA receptors are influenced by Zn2+ ions in a voltage-dependent manner, as well as by oxidation/reduction and pH.

The NMDA channel is blocked in a use- and voltage-dependent manner by Mg²⁺ (fig. 1). This means that NMDA receptors are activated only after depolarization of the postsynaptic membrane by, for example, AMPA receptor activation, which relieves the voltagedependent blockade by Mg²⁺. This biophysical property and their high Ca²⁺ permeability render NMDA receptors inherently suitable for their role in mediating synaptic plasticity, such as that underlying learning processes and development (Collingridge and Singer, 1990; Danysz et al., 1995b). Similar to Mg²⁺, uncompetitive NMDA receptor antagonists such as ketamine, dextromethorphan, memantine, phencyclidine (PCP), and (+)MK-801 [(+)5-methyl-10,11-dihydro-5H-dibenzocyclohepten-5,10-imine maleate] block the NMDA channel in the open state, although the blocking kinetics and voltage dependence of this effect depend on the antagonist (Rogawski, 1993; Parsons et al., 1995).

To date, two major subunit families, designated NR1 and NR2, have been cloned. It is generally accepted that functional receptors in the mammalian CNS are only formed by combination of NR1 and NR2 subunits, which express the glycine and glutamate recognition sites, respectively (Kuryatov et al., 1994; Grimwood et al., 1995b; Wafford et al., 1995; Hirai et al., 1996; Williams et al., 1996; Laube et al., 1997; Anson et al., 1998) (figs. 1 and 2).

View larger version (29K): [in this window] [in a new window]   Fig. 2.   Schematic presentation of the NMDA receptor subunit family. The tree indicates the degree of homology. For clarity, a NR1 receptor subunit splice variant having all alternatively spliced exons, i.e., N1 (exon 5, 21 amino acids), C1 (exon 21, 37 amino acids), and C2 (exon 22, 38 amino acids), is shown (Zukin and Bennett, 1995). Therefore, the TM and point mutation positions are numbered taking into account the presence of the N1 exon, i.e., positions are shifted by 21 amino acids, compared with original publications. Amino acid residues crucial for the binding of glutamate at NR2A or NR2B subunits and that of glycine at NR1 subunits are shown. The marks (closed and open circles) indicate residues at which point mutations resulted in at least 100-fold decreases in the affinity of glycine or glutamate. Glutamate binding sites also exist in the NR2C and NR2D subunits, but their exact locations have not been defined and are therefore not indicated in the scheme.

The NR2 subfamily consists of four individual subunits, i.e., NR2A to NR2D (Nakanishi et al., 1990, 1992; Kutsuwada et al., 1992; Monyer et al., 1992; Hollmann and Heinemann, 1994; McBain and Mayer, 1994; Danysz et al., 1995a; Parsons et al., 1998b). Various heteromeric NMDA receptor channels formed by combinations of NR1 and NR2 subunits are known to differ in gating properties, magnesium sensitivity, and pharmacological profile (Sucher et al., 1996; Parsons et al., 1998b). The heteromeric assembly of NR1 and NR2C subunits, for instance, has much lower sensitivity to Mg²⁺ but increased sensitivity to glycine and very restricted distribution in the brain. In situ hybridization has revealed overlapping but different expression profiles for NR2 messenger ribonucleic acid (mRNA). For example, NR2A mRNA is distributed ubiquitously (like NR1), with the highest densities occuring in hippocampal regions, and NR2B is expressed predominantly in forebrain but not in cerebellum, where NR2C predominates; NR2D is localized mainly in the brainstem (Moriyoshi et al., 1991; Monyer et al., 1992; Nakanishi, 1992; McBain and Mayer, 1994). NMDA receptors cloned from murine CNS have a different terminology, compared with those from rats (Kutsuwada et al., 1992); 1 remains the designation for the mouse equivalent of NR1, and 1 to 4 represent NR2A to -2D subunits, respectively.

In addition to NR1 and NR2, the NR3A subunit has recently been discovered. This receptor subunit (previously termed chi-1, or NMDAR-L) is a relatively recently identified member of a new class in the ionotropic glutamate receptor family that attenuates NMDA receptor currents when coexpressed with NR1/NR2 subunits in Xenopus oocytes but has no effect when tested with non-NMDA receptors or when expressed alone (Ciabarra et al., 1995; Sucher et al., 1995; Das et al., 1998). This subunit has an open reading frame coding for a predicted polypeptide of 1115 amino acids (with a predicted mass of 110 kDa) and shares 23% homology with other NMDA subunits and 27% homology with non-NMDA subunits. Highest levels are present in the spinal cord, brainstem, hypothalamus, thalamus, CA1 field of the hippocampus, and amygdala, and this distribution remains the same throughout life. However, the absolute levels show a strong peak between postnatal days 7 and 14, which then declines to adulthood. This subunit is expressed as a glycosylated protein subunit (135 kDa) with a distribution that parallels that observed for its mRNA, as determined by in situ hybridization (Ciabarra and Sevarino, 1997). Genetic knockout of NR3A in mice results in enhanced NMDA responses and increased dendritic spines in early postnatal cortical neurons, suggesting that NR3A is involved in the development of synaptic elements by modulating NMDA receptor activity (Das et al., 1998).

Before the molecular biological features of NMDA receptors were resolved, it was reported (Sekiguchi et al., 1990) that NMDA receptors expressed from mRNA from guinea pig cerebellum (as opposed to other brain regions) do not show sensitivity to glycine. However, strychnine-insensitive glycine binding is present in the cerebellum (Bristow et al., 1986; Danysz et al., 1989c), and glycine potentiates electrophysiological responses to NMDA in the cerebellum in situ (Netzeband et al., 1990). It should be stressed that in the cerebellum there are mainly NMDA receptors containing NR2C subunits (with NR1), which show strong sensitivity to glycine (Matsui et al., 1995), and the failure of Sekiguchi et al. (1990) to observe further facilitation may have been the result of saturation of the glycine_B site by background levels of glycine.

Because of the location of the glycine recognition site on the NR1 subunit, this review focuses more on the molecular biological features and ontogeny of this subunit. For more information on the molecular biological characteristics of NR2 subunits, readers are referred to previous overviews (Seeburg, 1993; Hollmann and Heinemann, 1994; McBain and Mayer, 1994).

Alternative splicing generates eight isoforms for the NR1 subfamily (Nakanishi et al., 1992; Durand et al., 1993; Zukin and Bennett, 1995). The variants arise from splicing at three exons; one encodes a 21-amino acid insert in the amino-terminal domain (N1, exon 5), and two encode adjacent sequences of 37 and 38 amino acids in the carboxyl-terminal domain (C1, exon 21, and C2, exon 22, respectively) (figs. 1 and 2). NR1 variants are sometimes denoted by the presence or absence of these three alternatively spliced exons (N1, C1, and C2); NR1₁₁₁ has all three exons, NR1₀₀₀ has none, and NR1₁₀₀ has only the amino-terminal exon (Durand et al., 1993; Zukin and Bennett, 1995). The variants from NR1₀₀₀ to NR1₁₁₁ are alternatively denoted as NR1E, -C, -D, -A, -G, -F, -"H", and -B or NMDAR1-4a, -2a, -3a, -1a, -4b, -2b, -3b, and -1b, respectively, but the more frequently used terminology uses noncapitalized suffixes for the most common splice variants, i.e., NR1a (NR1₀₁₁, NMDAR1-1a, or NR1A) and NR1b (NR1₁₀₀, NMDAR1-4b, or NR1G). This terminology is very confusing, and in our opinion the most logical and useful is that proposed by Zukin and Bennett (1995), i.e., NR1_xxx. The human NR1 gene is composed of 21 exons distributed over a total length of approximately 31 kilobases. Exons 4, 20, and 21 are identical in their amino acid sequences to N1, C1, and C2, respectively, in rats, suggesting that all eight NR1 isoforms found in rats would also be expressed in the human brain (Lebourdelles et al., 1994; Zimmer et al., 1995). In contrast, others reported that the human NR1 sequence diverges from the rodent and murine homologues near the carboxyl terminus (Planellscases et al., 1993). Studies on the function of these splice variants in homomeric receptors expressed in Xenopus oocytes must be viewed with caution, because homomeric NR1 receptors (Durand et al., 1993; Planellscases et al., 1993; Rodriguez Paz et al., 1995) are probably only functionally expressed because of the presence of an endogenous NR2-like protein (XenU1) in these cells (Barnard, 1997; Soloviev and Barnard, 1997).

The highest levels of NR1 mRNA in the adult rat and mouse CNS are in the olfactory bulb, and the lowest levels are expressed in the spinal cord. Intermediate levels were found in frontal cortex, hippocampus, cerebellum, and whole brain (Franklin et al., 1993; Akazawa et al., 1994). Similar findings have been reported with antibodies to NR1 subunits (Petralia et al., 1994; Benke et al., 1995). mRNA for double-splice variants in the C1/C2 regions, such as NR1₀₁₁ (NR1a), show an almost complementary pattern with respect to those lacking both of these inserts, such as NR1₁₀₀ (NR1b); the former are more concentrated in rostral structures such as cortex, caudate, and hippocampus, whereas the latter are principally found in more caudal regions such as thalamus, colliculi, locus coeruleus, and cerebellum (Laurie and Seeburg, 1994b; Luque et al., 1995a; Paupard et al., 1997). Others reported that the predominant splice variants in cortex and hippocampus were those without the N1 insert (also NR1a), whereas in the cerebellum the major variant was NR1₁₀₀ (NR1b), containing N1 (Zhong et al., 1995). In other words, cell-specific patterns for NR1 mRNA lacking N1 inserts parallel those for mRNA containing C1/C2 inserts (NR1₀₁₁, NR1a) and vice versa (NR1₁₀₀, NR1b). In the hippocampus, NR1a mRNA shows high levels in all regions, whereas NR1₁₀₀ is expressed more intensely in CA3 pyramidal neurons (Paupard et al., 1997). mRNA for NR1₀₀₁ and NR1₁₀₁ splice forms is found nearly homogeneously throughout the adult CNS, whereas mRNA with alternative splicing at C1 but not C2 (NR1₀₁₀ and NR1₁₁₀) is scarce, being detected only at very low levels in postnatal cortex and hippocampus (Laurie and Seeburg, 1994b; Paupard et al., 1997). Important from a methodological perspective is the finding that the predominant splice variants in cultured cortical neurons are also those lacking the N1 insert, such as NR1₀₁₁ (NR1a) (Zhong et al., 1994).

In developing rats, NR1 mRNA levels in cortex and hippocampus increased nearly three-fold from postnatal day 3 to day 15 and approximately doubled from day 15 to day 67 (Franklin et al., 1993; Riva et al., 1994; Nowicka and Kaczmarek, 1996). In contrast, cerebellum and brainstem showed no change in NMDAR1 mRNA levels between postnatal days 3 and 15 but levels also doubled from day 15 to day 67 (Franklin et al., 1993). Similar results were reported by a different group, although levels in the hippocampus peaked at postnatal day 10 and declined thereafter (Pujic et al., 1993). In the hippocampus, NR1 mRNAs lacking the N1 insert (such as NR1₀₁₁) dominate at birth and exhibit mature patterns of labeling, with high levels of expression in the CA1 and CA3 regions and the dentate gyrus. In contrast, mRNAs containing this insert (such as NR1₁₀₀) are initially expressed at lower uniform levels but levels increase more in the CA3 region than in the CA1 region or the dentate gyrus in the second and third postnatal weeks (Paupard et al., 1997).

Antisera against the carboxyl- and amino-terminal domains of NR1 receptors revealed similar distributions, which increased strongly in most brain regions until postnatal day 21; the exception was that carboxyl-terminal domain staining decreased in the thalamus, tectum, and brainstem, possibly because of the emergence of carboxyl-terminal splice variants not recognized by the antiserum (Benke et al., 1995; Luo et al., 1996). Interestingly, the amino acid sequences contained within the seven-amino acid, carboxyl-terminal domain of C1 NR1 splice variants and all NR2 subunits may serve to localize NMDA receptors to synaptic domains by interactions with postsynaptic density protein-95 (Ehlers et al., 1995; Kornau et al., 1995). An important finding is that the absolute density of NR1 receptors detected with antisera is close to that found using [³H]MK-801 binding, suggesting that most of the NR1 subunits expressed in the brain exist in an active form (Luo et al., 1996).

NR1 subunit immunostaining in the rat visual cortex is associated with the plasma membrane at early stages of development, before innervation by axons, whereas clustering of receptors at junctions may be promoted by axonal contact (Aoki et al., 1994; Aoki, 1997). At all ages, the prevalence of NR1-immunoreactive profiles was lamina 1 > laminae 4/5 > laminae 6/6B (Aoki et al., 1994). In contrast, others reported that cat and ferret cortical neurons initially show high levels of immunostaining for NR1, which then decline gradually during development, with the notable exception of cortical layers 2/3, where levels of NMDAR1 immunostaining remain high into adulthood (Catalano et al., 1997).

In human embryonic kidney (HEK) 293 cells expressing homomeric NMDAR1 receptors, significant levels of specific binding of the glycine_B antagonists ³H-labeled L-689,560 [4-trans-2-carboxy-5,7-dichloro-4-phenylamino-carbonylamino-1,2,3,4-tetrahydroquinoline] (fig. 3) and MDL-105,519 [(E)-3-(2-phenyl-2-carboxyethenyl)-4,6-dichloro-1H-indole-2-carboxylic acid] (fig. 4) but not of glutamate antagonists or [³H]MK-801 were seen (Grimwood et al., 1995b; Siegel et al., 1996). Similarly, Lynch et al. (1994) reported, in HEK 293 cells transfected with the NR1 subunit, significant binding to the glycine_B site [³H-labeled 5,7-dichlorokynurenic acid (5,7-diCl-KYN)] (fig. 5) but not to the NMDA site [³H-labeled CGP-37849 [DL-(E)-2-amino-4-methyl-5-phosphono-3-pentenoic acid]] or the PCP site ([³H]MK-801).

View larger version (9K): [in this window] [in a new window]   Fig. 3.   2-Carboxytetrahydroquinoline antagonists of the glycineB site.

View larger version (16K): [in this window] [in a new window]   Fig. 4.   2-Carboxyindole (CI) antagonists of the glycineB site.

View larger version (17K): [in this window] [in a new window]   Fig. 5.   Selected kynurenic acid derivatives that are antagonists of the glycineB site.

It should also be noted that NR2 subunits contribute to the affinity of glycine. With homomeric NR1 receptors the glycine K_i is in the range of 1 to 5 µM, but with some heteromeric receptors it is in the nanomolar range (Grimwood et al., 1995b). Interestingly, glycine affinity with wild-type receptors is usually higher, i.e., 100 to 300 nM (Kishimoto et al., 1981; Danysz et al., 1990; Grimwood et al., 1992).

Site-directed mutagenesis of the NR1 subunit at residues corresponding to positions forming the binding site of homologous, bacterial, amino acid-binding proteins indicates conservation of a common amino acid-binding fold from prokaryotic periplasmic proteins to glutamate receptors in the mammalian brain (Kuryatov et al., 1994). Glutamine substitutions at position 387 in NR1 subunits reduce glycine affinity (Kuryatov et al., 1994). Similarly, replacement of cysteines 402 and 418 by alanine largely abolishes the potentiation of glutamate currents by glycine (Laube et al., 1993) (fig. 2). These residues in the amino-terminal domain of NR1 subunits are extracellular before transmembrane domain (TM)1. It should be noted that these mutations in NR1 subunits had little effect on glutamate binding or the affinity of the glycine_B antagonist 7-chlorokynurenic acid (7Cl-KYN) (fig. 5). However, site-directed mutagenesis of the NR1 subunit expressed in Xenopus oocytes revealed that aromatic residues at positions 390, 392, and 466 are crucial determinants of both agonist and antagonist binding, as observed in patch-clamp experiments (Kuryatov et al., 1994). Glutamate efficacy was little affected by these mutations, but inhibition by 7-Cl-KYN was also greatly reduced (Kuryatov et al., 1994). Very similar findings were reported for human NR1 receptors, expressed in Xenopus oocytes, with mutation of residues 481 and 483 (Wafford et al., 1995).

Valine and serine substitutions in NR1 subunits at positions 666 and 669, respectively (in the loop between TM1 and TM4) (figs. 1 and 2), were also found to reduce glycine efficacy (Kuryatov et al., 1994). Others found that alternative mutation of D669 in NR1a to asparagine, alanine, or glutamate had little effect on the potency of glycine but abolished the "glycine-independent" form of spermine stimulation, indicating the importance of this residue for the binding of polyamines and/or the formation of part of the proton sensor (Kashiwagi et al., 1996). Mutation of alanine at position 714 also greatly reduced the apparent affinity for glycine (Wood et al., 1997), as did substitutions of the phenylalanine residues at positions 735 or 736 (Hirai et al., 1996). Interestingly, these mutations did not alter the affinity of 7-Cl-KYN, indicating that this part of the extracellular domain contributes to glycine binding but not antagonist binding. Mutation of D732 to glutamate (D732E), asparagine (D732N), alanine (D732A), or glycine (D732G) not only dramatically reduced the potency of glycine but also changed the sensitivity to other glycine site agonists (and, in some cases, their efficacy) (Williams et al., 1996) (fig. 2). For example, D-serine was a full coagonist at receptors containing NR1(D732N) and NR1(D732A), a partial agonist at receptors containing NR1(D732G), and a competitive antagonist at receptors containing NR1(D732E) (Williams et al., 1996). All of these residues are found in the extracellular M3-M4 loop (figs. 1 and 2).

Homology-based molecular modeling of the glutamate and glycine binding domains indicates that the NR2 and NR1 subunits use similar residues to form their respective agonist binding sites. Therefore, similar mutations of residues within the amino-terminal domain (E387A and K459E) and the loop region between segments TM3 and TM4 (S664G) of NR2B subunits reduced the potency of glutamate >100-fold but had no effect on glycine affinity (Laube et al., 1997). Mutations in NR2A subunits (T671A) dramatically reduced glutamate potency and produced faster deactivation kinetics, without changing channel gating or the affinity for glycine (Anson et al., 1998). Similarly, NR2 subunits containing mutations at NR2A position D731 and NR2B position D732, which correspond to NR1 position D732, did not produce functional receptors when coexpressed with NR1 (Williams et al., 1996). Taken together, these results indicate that the extracellular region before TM1 and the extracellular TM3-TM4 loop form a ligand-binding pocket for glutamate and glycine in NR2 and NR1 subunits, respectively; these findings provide the basis for a refined model for agonist and coagonist binding sites of the NMDA receptor (Hirai et al., 1996).

The absolute stoichiometry of NMDA receptor subunits is not clear. It is widely accepted that NMDA receptors are activated only after the binding of glutamate to at least two NR2 subunits for each receptor (Mayer et al., 1989b; Javitt et al., 1990; Patneau and Mayer, 1990; Curras and Dingledine, 1992; Sather et al., 1992; Wafford et al., 1993; Hirai et al., 1996; Laube et al., 1997), and the same is true for glycine as a coagonist at NR1 subunits (Thedinga et al., 1989; Benveniste et al., 1990a,b; Benveniste and Mayer, 1991; Clements and Westbrook, 1991; Siegel et al., 1996; Williams et al., 1996). This assumption is normally made on the basis of Hill coefficients for steady-state responses, which are considerably greater than unity. Similarly, sigmoidal activation kinetics of NMDA channels in outside-out patches from cultured hippocampal neurons were best fitted by a kinetic model with two glutamate binding sites and two glycine sites, with agonist and coagonist binding being better described by an independent, rather than a sequential, model (Clements and Westbrook, 1991). Taken together, these data led to the hypothesis that the NMDA receptor is at least a tetramer containing four ligand-binding subunits, with a single binding site in each subunit (Clements and Westbrook, 1991). In contrast, others have reported Hill coefficients of less than unity for glycine (Lerma et al., 1990), and detailed analysis of glycine concentration-response curves and kinetics indicated Hill coefficients for glycine of 1.1, with a corrected affinity of 130 nM, consistent with the idea that there is only one glycine binding site (Johnson and Ascher, 1992).

	III. Physiological Role of Glycine

Top Previous Next References

Kleckner and Dingledine (1988) were the first to report that glycine is essential for activation of the NMDA receptors. Using Xenopus oocytes injected with whole-brain mRNA, those authors observed that the response to NMDA vanished when the contamination with glycine was reduced to negligible levels. A similar conclusion regarding the necessity of glycine for the activation of NMDA receptors was later drawn on the basis of patch-clamp studies in neuronal cultures (Mayer et al., 1989a; Henderson et al., 1990; Huettner, 1990; Vornov and Coyle, 1991; Aoshima et al., 1992; Chen et al., 1997). Further evidence that the activation of the glycine_B site is a prerequisite for NMDA receptor activation in vitro or in vivo was obtained by showing that selective antagonists of the glycine_B site completely block the effects of NMDA receptor stimulation (see Section IV.C.). This suggests that residual responses in the nominal absence of glycine are the result of contamination with low background levels of glycine (Benveniste et al., 1990a; Lerma et al., 1990; Kemp and Priestley, 1991; Parsons et al., 1993, 1997; Molnar and Erdo, 1996). Indeed, extrapolation of the lower linear part of glycine concentration-response curves indicates that approximately 20 to 40 nM glycine is found in the extracellular solution, a value that agrees very well with that measured by high pressure liquid chromatography (Benveniste et al., 1990). This provides support for the notion that glycine is an essential coagonist at the NMDA receptor and that responses to NMDA cannot be obtained in the complete absence of glycine (Mayer et al., 1989a,b; Vornov and Coyle, 1991; Aoshima et al., 1992). However, it remains a remote possibility that blockade of the residual peak in the nominal absence of glycine by glycine_B site full antagonists is the result of inverse agonistic effects and that a component of the peak response would still be present in the complete absence of glycine (Kemp et al., 1988b; Mayer et al., 1989a).

The binding of use-dependent NMDA receptor channel blockers, such as [³H]MK-801, under nonequilibrium conditions can be used to investigate certain aspects of receptor function; levels of bound ligand are proportional to the degree of activation (Foster and Wong, 1987; Wong et al., 1988). Using this method, in most cases it was only possible to demonstrate the essential role of glycine by blocking this site with an antagonist, probably because of widespread contamination with glycine. However, in extensively washed membranes it was shown that glutamate and glycine failed to enhance [³H]MK-801 binding when used separately but did produce enhancement when applied together (Ratti et al., 1990). On the other hand, spermine enhanced functional [³H]MK-801 binding in the presence of the glycine_B antagonist 7-Cl-KYN but not when the NMDA site was blocked by 3-(2-carboxypiperazine-4-yl)propyl-1-phosphonic acid (CPP) (Marvizon and Baudry, 1993). It was suggested by the authors of the latter study that glycine is not an absolute requirement for NMDA receptor activation. However, binding is probably not the optimal method to explore such questions, because, for example, [³H]MK-801 has been reported to access the NMDA channel in a closed state via the so-called "lipophilic pathway" (Javitt and Zukin, 1989).

It is also noteworthy that, although glycine concentrations up to 3 mM do not activate NMDA receptors in vitro without NMDA agonists, there are indications that at concentrations above these levels glycine can actually damage neurons (Kleckner and Dingledine, 1988; McNamara and Dingledine, 1990; Wallis et al., 1994; Newell et al., 1997) and induce inward currents in cultured hippocampal neurons (Pace-Asciak et al., 1992), via activation of both glycine_B and NMDA sites. It has also been suggested that, under certain conditions, glycine might achieve such high levels and contribute to neurotoxicity in vivo (Newell et al., 1997).

In vivo activation of glycinergic inhibitory interneurons in the spinal cord by stimulation of 1b afferents elicited a classical, short-latency, glycinergic inhibitory postsynaptic potential followed by an NMDA receptor-mediated EPSP. The EPSP was blocked by ketamine and R(+)HA-966 [R(+)-3-amino-1-hydroxypyrrolidin-2-one] and showed classical voltage dependence. The authors proposed that glycine released at inhibitory interneurons spills over to activate nonsaturated glycine_B sites in vivo. This interpretation is controversial, because 1b afferents also activate excitatory interneurons and NMDA EPSPs are always delayed because of slower activation kinetics (Fern et al., 1996).

Glycine greatly potentiates NMDA receptor-mediated responses by reducing desensitization both in native mammalian neurons and in Xenopus oocytes or HEK 293 cells expressing NMDA receptors (Mayer et al., 1989a; Vornov and Coyle, 1991; Aoshima et al., 1992; Chen et al., 1997) (fig. 6). Glycine-sensitive desensitization is accompanied by a five- to seven-fold decrease in the affinity of the glycine_B site in the presence of agonists for the NMDA site (Lerma et al., 1990; Parsons et al., 1993). With higher concentrations of glycine, the magnitude of this desensitization is decreased but the rate becomes faster. Furthermore, recovery from desensitization after step increases in the concentration of glycine or lower affinity glycine agonists in the continuous presence of NMDA reflects the association kinetics of the agonist concentrations used (Lerma et al., 1990; Parsons et al., 1993). Therefore, desensitization probably occurs rapidly upon binding of both glutamate and glycine, and the apparent rate reflects the balance between slow dissociation from and concentrationdependent reassociation of glycine with the altered receptor. In other words, at higher concentrations of glycine, the forward rate constant for rebinding greatly exceeds the rate of dissociation and the time course of desensitization appears to be faster. The affinities of agonist to induce desensitization are five-fold higher than their respective affinities as agonists at the peak of the response (Chizhmakov et al., 1992).

View larger version (40K): [in this window] [in a new window]   Fig. 6.   Patch-clamp experiments with four different cultured hippocampal neurons, showing that glycineB antagonists have different effects on glycine-sensitive desensitization. NMDA (200 µM) was applied with a rapid concentration clamp (exchange times, 20 msec) for 2.5 sec every 30 sec, in the continuous presence of glycine (0.03 to 1 µM), with low Ca2+ levels (0.2 mM) and a constant membrane potential of 70 mV. Responses at equilibrium have been superimposed for each concentration of glycine or glycineB antagonist tested (continuously present for at least 2 min). Reduction of glycine concentrations reveals glycine-sensitive desensitization. 5,7-diCl-KYN (5,7-DCKA) (0.3 to 3 µM) concentration-dependently reveals glycine-sensitive desensitization in the continuous presence of glycine (1 µM). The higher affinity glycineB antagonist MDL-105,519 (0.1 to 1 µM) reveals less pronounced glycine-sensitive desensitization in the continuous presence of glycine (1 µM). The low intrinsic activity, glycineB site partial agonist R(+)HA-966 (10 to 300 µM) does not reveal glycine-sensitive desensitization in the continuous presence of glycine (1 µM). Note that the current scale bars are different for each antagonist tested.

The opposite has been reported in binding experiments, namely a mutually positive allosteric interaction, with glutamate increasing glycine affinity and glycine increasing glutamate affinity (table 2). Moreover, other electrophysiological studies concluded that desensitization involves structural changes in the channel-lining section of the protein, rather than the glycine or NMDA binding sites, because the induction of desensitization was dependent on channel opening (Zilberter et al., 1991). This seems unlikely, because single-channel recordings and fluctuation analysis show an increase in opening frequency with no change in mean open time or conductance in the presence of glycine and the opposite in the presence of glycine_B antagonists, suggesting that glycine regulates transitions to states that are intermediate between the binding of NMDA receptor agonists and ion-channel gating (Mayer et al., 1989a; Vornov and Coyle, 1991; Parsons et al., 1993). It has also been claimed that aspartate induces desensitization in the absence of glycine, but this interpretation is again complicated by the presence of background levels of glycine in all experiments (Chizhmakov et al., 1992).

A similar form of desensitization is seen in the presence of some glycine_B site full antagonists (Kemp and Priestley, 1991; Parsons et al., 1993), but these vary in their ability to induce glycine-sensitive desensitization. L-689,560, L-701,324 [7-chloro-4-hydroxy-3-(3-phenoxy)phenyl-2(1H)-quinolinone], L-695,902 [7-chloro-4-hydroxy-3-methoxycarbonyl-2(1H)-quinolinone] (fig. 7), and RPR-104,632 [2-(3-bromobenzyl)-6,8-dichloro-3,4-dihydro-2H-1,2,4-benzothiadiazine-1,1-dioxide-3-carboxylic acid] (fig. 5) reveal little or no desensitization but other compounds, such as 7-Cl-KYN and 5,7-diCl-KYN, show 10-fold higher potencies against plateau responses (Molnar and Erdo, 1996; Karcz-Kubicha et al., 1997). Other compounds, such as ACEA 1021 (5-nitro-6,7-dichloro-1,4-dihydro-2,3-quinoxalinedione) (fig. 8), MRZ 2/571 [8-fluoro-4-hydroxy-1-oxo-1,2-dihydropyridazinol[4,5-b]quinoline-5-oxide (choline salt)], and MRZ 2/576 [8-chloro-4-hydroxy-1-oxo-1,2-dihydropyridazinol[4,5-b]quinoline-5-oxide (choline salt)] (fig. 9), induce moderate desensitization. The glycine_B site partial agonist R(+)HA-966 (fig. 10) does not induce desensitization but produces a three-fold allosteric reduction in the affinity of agonists for the glutamate recognition site, which is reflected by slowing of the response onset rise time and acceleration of offset kinetics (Kemp and Priestley, 1991). Taken alone, this result could be considered indicative of a negative reciprocal interaction between the glycine and glutamate recognition sites. However, D-cycloserine and 1-aminocyclopropanecarboxylic acid (ACPC), which are partial agonists with higher intrinsic activity (fig. 10), are also unable to induce desensitization (Karcz-Kubicha et al., 1997) but have intermediate effects on glutamate deactivation kinetics, in line with their relative intrinsic activities (Priestley and Kemp, 1994). Furthermore, the full antagonists L-701,324 and L-695,902 (fig. 7) were recently reported to decrease glutamate affinity in a manner similar to that of R(+)HA-966 (Priestley et al., 1996). The same group reported a reciprocal three-fold interaction between partial agonists at the NMDA recognition site and glycine affinity, with the off-rate of glycine being fastest in the presence of a saturating concentration of the competitive NMDA receptor antagonist cis-2,3-piperidinedicarboxylic acid and progressively slower in the presence of quinolinate, NMDA, and L-glutamate (Priestley and Kemp, 1994). Taken together, these data were interpreted as being more in agreement with binding studies showing reciprocal positive interactions between full agonists at the NMDA and glycine recognition sites, with the extent of the modulation of one site by the other being related to the intrinsic activity of the agonist used, rather than its affinity.

View larger version (19K): [in this window] [in a new window]   Fig. 7.   4-Hydroxyquinolone antagonists of the glycineB site. HNQ, 4-hydroxy-3-nitroquinolin-2(1H)-one.

View larger version (22K): [in this window] [in a new window]   Fig. 8.   Quinoxaline-diones and related compounds that are antagonists of the glycineB site. CNQX, 6-cyano-7-nitroquinoxaline-2,3-dione; HBAD, 3-hydroxy-1H-1-benzazepine-2,5-dione.

View larger version (14K): [in this window] [in a new window]   Fig. 9.   Tricyclic antagonists of the glycineB site.

View larger version (19K): [in this window] [in a new window]   Fig. 10.   Selected agonists and partial agonists of the glycineB site. MSD low effic PA, Merck Sharp and Dohme low efficiency partial agonist.

However, it should be stressed that the comparisons in offset kinetics were all made at steady state and did not address the relative changes in apparent affinities at the peak and plateau of the response. The same is true for binding studies, which all assessed affinities at steady state under highly nonphysiological conditions. Alternatively, the fact that R(+)HA-966 and D-cycloserine do not exhibit glycine-sensitive desensitization (fig. 6) could be the result of both their low potency and allosteric slowing in onset kinetics. It may be that an agonist-induced change in the affinity of the glycine site still occurs and would be seen as very fast desensitization, reflecting the rapid unbinding of this low affinity partial agonist, if it were not masked by the slower rise time of the response. This idea is supported by the fact that desensitization is seen with intermediate concentrations of the higher affinity partial agonist ACPC in the absence of glycine (Karcz-Kubicha et al., 1997). A similar argument can be applied to the apparent lack of desensitization seen with some glycine_B site full antagonists. We previously showed a trend for less desensitization with higher potency full antagonists (Parsons et al., 1997) (fig. 6). In this case, the association of low concentrations of antagonist after agonist-induced dissociation of glycine is much slower than the forward rate constant for reassociation of glycine. Therefore, glycine-sensitive desensitization would be predicted to be revealed with a much slower time course.

Several studies found that the kinetics of NMDA receptor currents in outside-out patches are dominated by a pronounced glycine-insensitive form of desensitization (Sather et al., 1991, 1992; Lester et al., 1993; Tong and Jahr, 1994). This is particularly evident after longer recording periods, indicating the importance of dialysis of intracellular factors in mediating this effect. However, the glycine and glutamate recognition sites are still allosterically coupled, as evidenced by changes in the offset kinetics of glycine in the presence of glutamate (Lester et al., 1993). The appearance of glycine-insensitive desensitization is reduced by intracellular application of 1,2-bis(2-aminophenoxy)ethane-N,N,N',N'-tetraacetic acid (EDTA) or preincubation of neurons with the Ca²⁺ release inhibitor dantrolene, suggesting that this form of desensitization is triggered by a transient increase in intracellular Ca²⁺ levels. The extent of glycine-insensitive desensitization is also reduced by intracellular application of adenosine-5'-O-(3-thio)-triphosphate, the phosphatase inhibitor microcystin, or a peptide inhibitor of calcineurin, implying regulation by the phosphorylation state of the receptor (Tong and Jahr, 1994). Moreover, this form of desensitization seems to be dependent on activation of G proteins (Turecek et al., 1995). This may explain the relatively robust glycinesensitive desensitization seen in the study of Parsons et al. (1993), because single-channel recordings were first made after 20 to 30 min of whole-cell recording with low extracellular Ca²⁺ levels and high intracellular ethylene glycol bis(-aminoethyl ether)-N,N,N',N'-tetraacetic acid, adenosine-5'-triphosphate, and 3',5'-cyclic adenosine monophosphate levels, i.e., under conditions chosen to selectively investigate glycine-sensitive desensitization.

The improved therapeutic profiles of some systemically active, glycine_B site full antagonists could be the result of their abilities to reveal glycine-sensitive desensitization (Parsons et al., 1993). Receptor desensitization may represent a physiological process serving as an endogenous control mechanism to prevent long-term neurotoxic activation of glutamate receptors but allow their transient physiological activation. Interestingly, ischemia increases not only the concentration of extracellular glutamate but also that of glycine and, although this latter effect is less pronounced, it persists for much longer (Globus et al., 1991b). Prolonged repetitive activation of NMDA receptors during ischemia would be effectively reduced at concentrations of glycine_B site full antagonists having less effect on more transient activation during EPSPs, because the time course for glycine-sensitive desensitization (300 to 500 msec) (Mayer et al., 1989a) is somewhat longer than that for NMDA receptor-mediated synaptic events (typically 100 to 200 msec) (Clements et al., 1992). This property may also allow such compounds to differentiate between various forms of NMDA receptor-mediated synaptic plasticity, e.g., to block drug tolerance and dependence and chronic pain states at concentrations having less effect on learning and memory.

Glycine has been traditionally regarded as an inhibitory transmitter in lower CNS regions, where it activates receptors forming chloride channels (Leu et al., 1987). Only during the past 10 years has its function as a positive modulator acting at NMDA receptors been recognized (see Section I.). Meanwhile, high free D-serine (fig. 10) levels, i.e., concentrations of 2 to 300 µM, depending on the brain structure and experimental conditions, have been detected in the mammalian (including human) CNS (for review, see Hashimoto and Oka, 1997). In turn, D-serine has been suggested as a possible endogenous ligand for the glycine_B site. Here the features crucial for dynamic regulation of the concentration of both ligands are discussed in parallel, according to the criteria given in table 3.

It is noteworthy that dynorphin(1-13) was also recently proposed as an endogenous agonist at the glycine_B site, as evidenced by very pronounced increases in the amplitude of NMDA-activated currents in Xenopus oocytes in the presence of low extracellular glycine concentrations (Zhang et al., 1997). Such an effect may exacerbate the well documented spinal toxicity seen with dynorphin peptides after intrathecal administration (Shukla and Lemaire, 1994), although the vasoconstrictive actions of this peptide are probably also important.

To determine the physiological importance of either glycine or D-serine, their extracellular CNS levels should be compared with their affinities. Affinity is often difficult to assess accurately, because of variable levels of contamination with glycine, which have been estimated to be between 20 and 130 nM in binding and patch-clamp studies (Benveniste et al., 1990; Lerma et al., 1990; Johnson and Ascher, 1992; Parsons et al., 1993; Berger, 1995). In the majority of binding studies in brain homogenates, the affinity of glycine is slightly (two- to three-fold) higher than that of D-serine (Kishimoto et al., 1981; Danysz et al., 1990) (table 4). On the other hand, some authors report that, in experiments with recombinant NMDA receptors, D-serine seems to be slightly more potent as an agonist than glycine (Matsui et al., 1995) (but see table 5).

1. Distribution within the central nervous system. In brain homogenates, glycine has been reported to be present at concentrations of 6 to 10 nmol/mg of protein in mice (Saransaari and Oja, 1994), 1 nmol/mg of wet tissue in rats (Katsura et al., 1992), and 27 nmol/mg of protein in humans (Waziri et al., 1993). In the same preparation in rats, D-serine concentrations range from 10 µM in the cerebellum to 400 µM in the cerebral cortex (Nagata et al., 1994b). In adult human brain homogenates, D-serine is present at 66 to 130 µM (Hashimoto et al., 1993c; Nagata et al., 1994b; Hashimoto and Oka, 1997) or 2 to 15 nmol/mg of protein (Chouinard et al., 1993; Waziri et al., 1993). It should be kept in mind that homogenate concentrations indicate very little regarding free extracellular levels, because accumulation in intracellular compartments is likely (for example, see Danysz et al., 1997). In human cerebrospinal fluid (CSF), free glycine levels of 7 to 10 µM have been detected (Ferraro and Hare, 1985).

2. Uptake. Early studies demonstrated the presence, in the cerebral cortex (Hagan et al., 1988), cerebellum (Wilson et al., 1976; Wilkin et al., 1981), and hippocampus (Toth and Lajtha, 1986), of a sodium-dependent glycine uptake system that did not seem to transport D-serine. In the hippocampus, both low affinity (minor Na⁺ dependence) and high affinity (strong Na⁺ dependence) carriers were detected (Fedele and Foster, 1992). Based on lesion studies, it was concluded that this carrier is localized on both neurons and glia (mainly astroglia) (Magnuson et al., 1988). A similar conclusion was derived from studies on cultured cerebellar granule cells and astrocytes (Eberhard and Holtz, 1988) and autoradiographic experiments with [³H]glycine in the cerebellum (Wilkin et al., 1981).

3. Release. In cerebellar cultures, veratridine and high potassium levels induced release of preloaded [³H]glycine from both astrocytes and granule cells, but only the latter was partially Ca²⁺ dependent (Halopainen and Konto, 1989). In primary cultures of striatal neurons, KCl-induced depolarization produced a two-fold increase in glycine release that was partially (75%) Ca²⁺ dependent (Weiss et al., 1989). Also in synaptosomes from human cortex, Ca²⁺-dependent release of preloaded [³H]glycine was observed upon exposure to high potassium levels (15 mM) (Wullner et al., 1993). On the other hand, kainate-evoked release in striatal neurons was apparently not dependent on extracellular Ca²⁺ (Weiss et al., 1989). In hippocampal synaptosomes preloaded with 0.5 µM [³H]glycine, there was also significant enhancement of radioactivity efflux by veratridine and high potassium levels, which was not affected by Ca²⁺ removal (Galli et al., 1993). Because replacement of Na⁺ with Li⁺ attenuated [³H]glycine efflux, it can be concluded that efflux was governed by a glycine carrier working in a reverse mode (Galli et al., 1993). In HEK 293 cells stably expressing GLYT1 and preloaded with [³H]glycine, release was observed that was evoked by, for example, sarcosine, extracellular glycine, and replacement of Na⁺ with Li⁺ (Sakata et al., 1997). The latter finding indicates again that a decrease in the sodium gradient reverses the glycine carrier. Similarly, in hippocampal slices preloaded with radioactive [³H]glycine, potassium-evoked efflux of glycine was not calcium dependent, as concluded by the authors (Saransaari and Oja, 1994), although fig. 5 of that report clearly indicates an inhibitory effect of Ca²⁺ removal in old animals.

4. Source and synthesis. Glycine in the CNS is probably derived from general metabolic pathways (for example, it can be synthesized from L-serine by serine hydroxymethyl transferase) (Pycock and Kerwin, 1981), whereas the precise source of D-serine in the brain is not certain. It could theoretically be derived from (a) ingested food, (b) intestinal bacteria, (c) metabolically stable proteins containing this isomer, or (d) L-serine (Hashimoto and Oka, 1997). It has been suggested that ontogenic changes (e.g., distinct peak periods in various organs) argue against the first three possibilities, although the possibility that the differences observed result not from different rates of synthesis or selective target tissue uptake but from different rates of metabolism (i.e., the activity of D-amino acid oxidase) must be considered (Hashimoto and Oka, 1997). Also arguing against an external source are the findings that serum levels of D-serine are very low, compared with some organs (Hashimoto and Oka, 1997), and the transport of D-amino acids through the blood-brain barrier is rather poor (Olendorf, 1973; Sato et al., 1991). Moreover, it has been suggested that D-serine in the brain does not originate from microorganisms (which contain some D-amino acids) (Man and Bada, 1987), because germ-free and pathogen-free mice have similar brain contents of D-serine (Nagata et al., 1994b) and the distribution of D-serine within the brain is heterogeneous. The third possibility is also unlikely, as argued by Hashimoto and Oka (1997), because D-serine levels are high in the human cortex at embryonic stages, but the contribution of D-amino acid residues to metabolically stable proteins actually increases with age. Another option, namely that D-serine might be synthesized from L-serine, has recently been pursued. It was shown that D-serine is present in brain synaptosomes (Hashimoto et al., 1993c), where it is formed from L-phosphoserine by the actions of phosphoserine phosphatase followed by racemase (Wood et al., 1996). In fact, it was reported that injection of huge doses of L-serine into infant rats produced an increase of D-serine levels (two-fold) that lasted >24 h (Takahashi et al., 1997). The peak of L-serine was higher (four-fold) but of shorter duration (10 h), in accord with the precursor concept. D-Serine injection also produced an increase in L-serine content (but not that of other amino acids, including glycine), suggesting a bidirectional activity of the racemase, which, unfortunately, has not been detected in the brain thus far (Takahashi et al., 1997).

5. Metabolism. Metabolic inactivation of glycine is most probably accomplished through the glycine cleavage system, which is localized in the inner membrane of the mitochondria of astroglial cells but is absent in neurons (Daly et al., 1976; Sato et al., 1991). The reaction of the system is as follows: 2 glycine + H₂O + NAD⁺  serine + CO₂ + NH₂ + NADH + H⁺. The distribution of the glycine cleavage system corresponds to some extent to the distribution of NMDA receptors and is also somewhat inversely related to local glycine levels in the brain (Daly et al., 1976; Sato et al., 1991), indicating its importance in maintaining certain spatially determined basal levels of this amino acid. For example, astrocytes in the telencephalon and cerebellum are strongly labeled with antibody toward this enzyme, and those in spinal cord and medulla oblongata are much more weakly labeled (Daly et al., 1976; Sato et al., 1991).

6. N-Methyl-D-aspartate receptor-coupled recognition site and its distribution. In Na⁺-free medium, at 4°C (to avoid binding to carriers), [³H]glycine labels two types of recognition sites in the CNS. The first type is the the inhibitory glycine_A receptor (chloride channel), which is also labeled by [³H]strychnine (Young and Snyder, 1974; Aprison and Daly, 1978; Danysz et al., 1990) and is abundant mainly in the spinal cord and pons. The second type is the strychnine-insensitive glycine_B site coupled to NMDA receptors (Kishimoto et al., 1981; Galli et al., 1988). This site predominates in the striata oriens and radiatum of the hippocampal CA1 area, the CA3 region, the dentate gyrus, the superficial layer of the cerebral cortex, the striatum, the dorsolateral septum, and the amygdala.

7. Are glycine_B sites saturated in vivo? It has been often claimed that the glycine_B site is saturated under physiological conditions (e.g., see Obrenovitch et al., 1997). However, this conclusion is mainly based on in vitro experiments using brain slices (Fletcher and Lodge, 1988; Kemp et al., 1988a; Taylor et al., 1988; Crawford and Roberts, 1989; Ransom and Deschenes, 1989) and on the observation that the free glycine concentration in ECF in vivo is much higher than the K_d of glycine for the receptor site (see Section III.C.1.). However, the following arguments argue against the assumption that glycine_B sites are saturated.

1.	It should be kept in mind that the traumatic preparation of slices and/or the application of the NMDA agonists leads to release of glutamate and glycine; in other words, slice preparations are not ideally suited to address this question. However, even under such conditions some authors have reported that additional glycine or D-serine enhances NMDA receptor-mediated events (Thomson et al., 1988; Minota et al., 1989).
2.	Estimates of the ECF concentration of any molecule do not necessarily reflect concentrations in the synaptic cleft. If this were the case, NMDA receptors would be continuously saturated by glutamate, because the "resting" ECF concentration of glutamate could be as high as 345 µM, based on in vivo microdialysis (5 µl/min flow) results corrected for in vitro recovery (5%) and a diffusion factor (Benveniste and Hansen, 1991), i.e., 1000 times greater than the Ki of glutamate for NMDA receptors (Watkins et al., 1990). It is, however, likely that the basal level of glutamate was overestimated by Benveniste and Hansen (1991). Using the zero-net flux and mass-transfer methods of in vivo recovery, lower values were obtained by other investigators (Jacobson et al., 1985; Miele et al., 1996), i.e., 3.0 and 4.3 µM, respectively. Even these conservative estimates of resting glutamate levels would be sufficient to strongly activate NMDA receptors. Fortunately, NMDA receptors are clearly not saturated in vivo, because of efficient local buffering mechanisms.
3.	The glycineB antagonist kynurenic acid is endogenous to the brain and under certain conditions may rise to concentrations sufficient to block the glycineB recognition site, according to some authors (Moroni et al., 1988a; Wu et al., 1992). It should be noted that kynurenic acid is not selective for the glycineB site.
4.	Zinc has been shown to decrease the affinity of [3H]glycine binding (Yen et al., 1990). Hence, the affinity of glycine in vitro may be overestimated, compared with in vivo conditions in brain regions where zinc is present (Aniksztejn et al., 1987). However, it should be stressed that Ca2+ and Mg2+ have been reported to increase glycine affinity (Gu and Huang, 1994; Wang and Macdonald, 1995; McBain and Mayer, 1994).

Kynurenic acid is an endogenous molecule found in both rodent and human brain, at concentrations of 5.8 to 36.3 and 50 pmol/g, respectively (Moroni et al., 1988b; Turski et al., 1988, 1989b). It is most likely produced by astroglia (Speciale et al., 1989; Eastman et al., 1994). It is a product of the kynurenine pathway after tryptophan degradation, and its direct precursor is kynurenine (Moroni et al., 1988b; Turski et al., 1988, 1989b). Kynurenic acid is also potentially an endogenous glycine_B antagonist, with a 50% inhibitory concentration (IC₅₀) of 43 µM ([³H]glycine binding) (Danysz et al., 1989b; see also Kessler et al., 1989b). When exogenously applied, it clearly attenuates activation of NMDA receptors (Birch et al., 1988b,c; Velisek et al., 1995). Questions arise regarding whether the brain levels of endogenous kynurenic acid are sufficient to block NMDA receptors under certain conditions and, if they are, why nature would create both agonists and an antagonist for the same site. The kynurenic acid concentration has been reported to be high in prenatal rodents (300 fmol/mg of protein), but it decreases after birth (Beal et al., 1992). Increased levels have also been reported in aged animals (Moroni et al., 1988a; Gramsbergen et al., 1992), associated with various brain insults/diseases, such as septicemia (Heyes and Lackner, 1990), excitotoxic brain lesions (Wu et al., 1992), and kindling (Loscher et al., 1996), and in dystonic hamsters (Richter et al., 1996). Enhanced levels of kynurenic acid in human brain have also been reported in Huntington's disease (Connick et al., 1989; but see Beal et al., 1990) and in Down's syndrome (Baran et al., 1996). Brain levels of kynurenic acid can be efficiently raised to concentrations that interact with NMDA receptors by several pharmacological interventions, for example, by administering the kynurenic acid precursor 4-chlorokynurenine, inhibiting brain efflux with probenecid, or inhibiting kynurenic acid metabolism (see Section IV.C.9.).

	IV. Exogenous Ligands

Top Previous Next References

Numerous high affinity, full antagonists for the glycine_B site were developed shortly after detection of this recognition site, but this was then recognized as being the easy step. It soon became clear that the real challenge was to develop compounds that penetrate the CNS well. Only with such compounds can the suggested usefulness of the glycine_B site for drug targeting be properly verified in animal models. Both the central administration of systemically inactive, full antagonists and the use of partial agonists (which show better blood-brain barrier penetration) often produce confusing results. Although glycine_B site partial agonists and antagonists influence the same site (by definition), in our opinion they should be considered separately. First, the agonist and antagonist glycine_B sites of the NMDA receptor are overlapping but probably not exactly the same, as evidenced by mutation analysis of the receptor protein (see Section II.D.) and by comparison of the structures of these two groups of agents. Second, the outcome of treatment with glycine_B site partial agonists is not predictable on the basis of the intrinsic activity determined in vitro (Karcz-Kubicha et al., 1997).

Of a multitude of amino acids tested in the pioneering study of Johnson and Ascher (1987) with cultured mouse cortical neurons, only alanine and serine were also active, and they were not as effective as glycine. In the same year, glycine and D-serine were reported to have similar potencies in enhancing [³H]TCP and [³H]MK-801 binding in the presence of glutamate, whereas D-alanine was somewhat less potent (Bonhaus et al., 1987; Reynolds et al., 1987; Snell et al., 1987) (fig. 10). Interestingly, the L-isomers of alanine and serine were considerably less potent (Kessler et al., 1987; Reynolds et al., 1987), which was confirmed in electrophysiological studies in Xenopus oocytes (Kleckner and Dingledine, 1988). There are numerous binding and electrophysiological studies supporting the actions of these amino acids at the glycine_B site. Most are listed within this review and confirm that the relative potencies at native receptors show the following rank order: glycine > D-serine > D-alanine L-serine > L-alanine (fig. 10). Much higher (millimolar) concentrations of L-serine, L-cysteine, L-alanine, L-proline, and glycine also evoke inward current responses in cultured hippocampal neurons by acting as agonists at the NMDA recognition site (Pace-Asciak et al., 1992).

One of the first glycine_B site partial agonists described, ACPC (fig. 10), has an intrinsic activity of 80 to 92% (table 5) and a potency of approximately 0.09 to 0.4 µM (Nadler et al., 1988; Marvizon et al., 1989; Monahan et al., 1989a; Watson and Lanthorn, 1990; KarczKubicha et al., 1997). Interestingly, the concentration-response curve for ACPC in patch-clamp experiments shows a biphasic nature, indicating spillover onto other recognition sites at higher concentrations (KarczKubicha et al., 1997). ACPC shows antidepressive, anxiolytic, and neuroprotective activity (see Section VI.) (table 8). These "antagonistic" effects at NMDA receptors are difficult to reconcile with its high intrinsic activity. Interestingly, in cultured neurons subjected to sustained exposure to ACPC, the neuroprotective effect vanishes after 24 h, which has been associated with an increase in the expression of the NR2C subunit, at which it has 150% intrinsic activity (relative to glycine) (Fossom et al., 1995a; Krueger et al., 1997). In contrast, repeated treatment in vivo resulted in diminution of NMDA-induced convulsions and antidepressive activity in the forced-swim test, but anxiolytic and neuroprotective effects remained unchanged (Skolnick et al., 1992; Von Lubitz et al., 1992). These diverse effects strongly suggest that there might be alterations in NMDA receptor composition, as found in vitro.

ACPC seems to penetrate well to the CNS. Initially it was reported that this agent has a brain half-life of <5 min (Rao et al., 1990), which would make interpretation of any in vivo effects extending beyond this time difficult. However, subsequent studies in monkeys and rodents showed that its half-life in blood is in fact 2.5 h and that in the CNS is 6 h; in humans, it is even longer (Cherkofsky, 1995; Maccecchini, 1997). ACPC is currently being developed by Symphony Pharmaceuticals for treatment of stroke and depression (table 8).

Among this group of cyclic analogues, an increase in size decreases affinity and intrinsic activity. In Xenopus oocytes injected with mRNA from rat brain, ACPC was a potent partial agonist with high intrinsic activity at the glycine_B site, 1-aminocyclobutanecarboxylic acid (ACBC) (fig. 10) was a weak partial agonist with low intrinsic activity but also interacted with other sites on the NMDA receptor, and cycloleucine was a very weak, glycine_B site full antagonist (Watson and Lanthorn, 1990; Wong and Kemp, 1991). ACBC has a bigger ring than ACPC, lower intrinsic activity (approximately 16%), and much lower affinity [50% effective concentration (EC₅₀) = 17 to 19 µM], as shown by electrophysiological and binding experiments (Hood et al., 1989b; Watson et al., 1989). Because of its low potency, this compound has not been studied intensively, and it has not been proven that this compound actually penetrates to the CNS. Cycloleucine (fig. 10) has the biggest ring and is a very weak, full antagonist, inhibiting glycine responses at 1 mM concentrations to 36%, with an IC₅₀ of approximately 500 to 600 µM (Snell and Johnson, 1988; Watson and Lanthorn, 1990). However, patch-clamp studies with cultured hippocampal neurons showed higher potency (IC₅₀ = 24 µM with 1 µM glycine), although it should be noted that a component of blockade could not be overcome by higher glycine concentrations, indicating interactions with other sites on the NMDA receptor (Hershkowitz and Rogawski, 1989).

D-Cycloserine (fig. 10) has long been used at high doses as an antibiotic for the treatment of tuberculosis (Mandell and Petri, 1996); more recently, it was discovered to have partial agonistic properties at the glycine_B site, with an intrinsic activity of 40 to 86%, depending on the experimental paradigm (Hood et al., 1989a; Watson et al., 1990; Priestley and Kemp, 1994; Karcz-Kubicha et al., 1997). It has been shown that at lower doses (<20 mg/kg) D-cycloserine shows agonistic effects in vivo but at higher doses an antagonistic action, such as anticonvulsive activity, predominates (Emmett et al., 1991; Peterson and Schwade, 1993; Lanthorn, 1994). Such bell-shaped dose-effect relationships are sometimes falsely interpreted to be "typical" for partial agonists, i.e., agonism at low doses and antagonism at high doses. However, partial agonism actually means that an agent reaches a limited nonmaximal effect at higher doses (intrinsic activity), i.e., antagonizes receptor activation by high concentrations of a full agonist but facilitates low concentrations of a full agonist (Henderson et al., 1990; Karcz-Kubicha et al., 1997). Recent data indicate that the consistent biphasic effects of D-cycloserine observed in vivo may, rather, be related to different affinities and intrinsic activities at NMDA receptor subtypes (see Section V. and table 5). Therefore, D-cycloserine is a partial agonist for the murine equivalents of NR1/NR2A and NR1/NR2B heteromers (38% and 56% intrinsic activity, respectively, compared with 10 µM glycine) but is more effective than glycine at NR1/NR2C receptors (130%) (O'Connor et al., 1996). This effect is accompanied by higher affinity at NR1/NR2C receptors, i.e., NR1/NR2C > NR1/NR2D NR1/NR2B > NR1/NR2A (Krueger et al., 1997). Therefore, it is likely that the biphasic effects seen in vivo are the result of agonistic actions at NR1/NR2C receptors at lower doses and inhibition of NR1/NR2A- and NR1/NR2B-containing receptors at higher doses. This does not explain, however, why complete inhibition of NMDA effects on cGMP production is observed in the cerebellum, where NR1/NR2C receptors predominate (Emmett et al., 1991). The receptor subtype selectivity of D-cycloserine and its differential intrinsic activity could well underlie its promising preclinical profile in some animals models. Enhancement of NMDA receptor-mediated events by D-cycloserine in human and nonhuman subjects is lost after chronic administration (Lanthorn, 1994; Quartermain et al., 1994; Randolph et al., 1994). D-Cycloserine was developed by Searle-Monsanto as a cognitive enhancer for dementia but has been abandoned (table 8).

Racemic HA-966 (fig. 10) was originally developed by Organon in the late 1960s (Bonta et al., 1971), but its antagonistic action at the glycine_B site was recognized years later. In 1978, HA-966 was reported to produce Mg²⁺-like selective antagonism of excitatory amino acid-induced responses in isolated spinal cords of frogs and immature rats in vitro (Evans et al., 1978) and to antagonize amino acid-induced and synaptic excitation of mammalian spinal neurons in vivo (Biscoe et al., 1978). HA-966 antagonism of NMDA-induced responses in rat cortical slices and cultured rat cortical neurons was then found to be sensitive to reversal by glycine (Fletcher and Lodge, 1988; Drejer et al., 1989; Foster and Kemp, 1989; Keith et al., 1989). Systemic administration of HA-966 was then reported to selectively reduce NMDA-induced lesions of the striatum in perinatal rats (Uckele et al., 1989). The first indication that HA-966 might be a partial agonist was the finding that inhibition of [³H]MK-801 binding, NMDA-induced Ca²⁺ influx, and NMDA-induced release of ³H-labeled -aminobutyric acid (GABA) was not complete (Reynolds et al., 1989). Conclusive evidence that both HA-966 and D-cycloserine are partial agonists, with low and high intrinsic activities, respectively, was first provided by patch- and concentration-clamp experiments (Henderson et al., 1990). The in vivo and in vitro activity of HA-966 at the glycine_B site was subsequently found to reside in the R(+)-stereoisomer (Pullan et al., 1990b), whereas the ()-isomer was reported to be a potent -butyrolactone-like sedative (Singh et al., 1990a).

Hence, all old studies using racemic HA-966 should be interpreted with caution, and only R(+)HA-966 (fig. 10) should be used to test effects at the glycine_B site. R(+)HA-966 inhibits NMDA receptors with a potency of 5.6 (K_i) or 28 (IC₅₀) µM and has an intrinsic activity of 10 to 40% (Watson and Lanthorn, 1990; Priestley and Kemp, 1994; Karcz-Kubicha et al., 1997). In contrast to the accepted partial agonistic properties of R(+)HA-966, it was reported that, in the periphery, HA-966 inhibited NMDA responses completely (Campbell et al., 1991). It is not clear whether this was attributable to different properties of NMDA receptors in vivo or the fact that the nonselective racemic mixture was used.

Attempts to improve the potency of partial agonists by 4-substitutions of R(+)HA-966 led to L-687,414 [R-(+)cis--methyl-3-amino-1-hydroxypyrrolid-2-one] (a cis-methyl derivative) (fig. 10) and cis-hydroxy derivatives with increased potency, but there was a rapid loss of activity with longer or trans substitutions (Leeson et al., 1990, 1993b). It was proposed that the axial conformation is preferred, and Merck Sharp and Dohme attempted to develop bicyclic compounds to test this hypothesis. The only compound to result from these attempts was a [3.2.1]bicyclic derivative. Both this compound and L-687,414 are partial agonists with very low intrinsic activities and would be predicted to have effects similar to those of full antagonists. Indeed, L-687,414 is active in DBA/2 mice, with a 50% inhibitory dose (ID₅₀) of 5.2 mg/kg [administered intraperitoneally (i.p.)], but its potency against maximal electroshock (MES)induced convulsions leaves room for improvement [26.1 mg/kg, intravenously (i.v.)] (Foster et al., 1992b; Leeson et al., 1993b). The very strict structure-activity requirements for these partial agonists may be related to differences in allosteric interactions, compared with full antagonists (Kemp and Priestley, 1991). R(+)HA-966 and L-687,414 were under development by Merck Sharp and Dohme but have been abandoned (table 8) (Smith and Meldrum, 1992; Gill et al., 1995).

Although none of these systemically active partial agonists induces receptor desensitization (Henderson et al., 1990; Kemp and Priestley, 1991; Karcz-Kubicha et al., 1997), they have favorable therapeutic profiles in some in vivo models and, in the case of D-cycloserine, also in humans (Lanthorn, 1994; van Berckel et al., 1997; Witkin et al., 1997). This may be, in part, the result of their own intrinsic activities as agonists at the glycine_B site, which would serve to preserve a certain level of NMDA receptor function even at very high concentrations (Priestley and Kemp, 1994; Fossom et al., 1995a; Krueger et al., 1997).

C. Antagonists

1. Kynurenic acid derivatives. Although kynurenic acid has been known for years, Kessler et al. (1987, 1989b) were the first to recognize (using receptor binding assays) that it actually interacts with glycine_B sites. Electrophysiological studies subsequently confirmed these glycine_B antagonistic effects against NMDA responses in infant rat hemisected spinal cord (Birch et al., 1988b; Watson et al., 1988). However, kynurenic acid is very weak and is not selective, because it is also a competitive antagonist at NMDA and AMPA/kainate receptors (Birch et al., 1988c; Stone, 1991).

2. 2-Carboxyindoles. 2-Carboxyindoles such as 2-carboxybenzimidazole (fig. 4) were reported to be active as glycine_B antagonists at the same time as kynurenic acid (Huettner, 1989; Smith et al., 1993) and quinoxaline-2,3-dione derivatives, which led to the assumption that kynurenic acid derivatives exist as the 4-keto tautomer in solution (Leeson et al., 1991a). The 3-substituent seems to have a structure-activity relationship similar to that for the 4-substituent in 2-carboxytetrahydroquinolines and is similarly responsible for hydrogen bonding (Huettner, 1989; Salituro et al., 1990; Gray et al., 1991; Baron et al., 1992; Hood et al., 1992; Rowley et al., 1992; Salituro et al., 1992; Vazquez et al., 1992; Nichols and Yielding, 1993; Rao et al., 1993). Compounds such as SC-49648 [3-(6-chloro-2-carboxyindol-3-yl)ethanoic acid] and MDL-29,951 [3-(4,6-dichloro-2-carboxyindol-3-yl)propionic acid] (fig. 4) are only active after i.c.v. administration (Baron et al., 1992). SC-50132 [3-(6-chloro-2-carboxyindol-3-yl)ethanoic acid ethyl ester] is the 3-ethyl ester analogue of SC-49648 and was found to be a weak, systemically active prodrug for SC-49648 (Rao et al., 1993).

3. 2-Carboxytetrahydroquinolines. The first compounds belonging to this group (fig. 3) were produced by Merck Sharp and Dohme and were developed to investigate the three-dimensional requirements for antagonists at the glycine_B site (Carling et al., 1992; Leeson et al., 1992). The decreased potency of conformationally flexible 2,3-dihydrokynurenic acid derivatives was proposed to be the result of the fact that the pseudoaxial conformation is preferred in solution but the pseudoequatorial conformation is required for potency at the glycine_B site. Introduction of 4-carboxyl groups introduced a second position for stereoisomerization, with trans substitutions being 100-fold more potent than cis. These trans substitutions take on a pseudoaxial conformation in solution and hold the 2-carboxyl group in the optimal pseudoequatorial position. The absolute stereochemistry is 2R/4S, showing similarities to that of agonists and partial agonists at the 2-position.

4. 4-Hydroxy-2-quinolones. Acidity is an important factor for selective high affinity at the glycine_B site. 2-Oxy plus 4-hydroxy substitutions of quinoxalinediones (fig. 7) yield compounds with the desired pK_a of approximately 5 (Leeson et al., 1993a). MDL-104,653 [3-phenyl-4-hydroxy-7-chloro-2(1H)-quinolinone] (fig. 7) is relatively weak as a glycine_B antagonist (K_i = 170 nM against [³H]glycine) but protects against sound-induced clonic seizures in DBA/2 mice after i.p. (ED₅₀ = 2 to 5 mg/kg) or p.o. (ED₅₀ = 6.2 mg/kg) administration. L-698,544 [7-chloro-3-nitro-2(1,3,4H)-quinolinone] is also a relatively weak, nonselective, AMPA/NMDA antagonist in vitro (10 µM) but is systemically active in DBA/2 mice (ED₅₀ = 13 mg/kg, i.p.). These compounds may therefore prove to be a basis for developing good neuroprotective agents (Carling et al., 1993).

5. Quinoxaline-2,3-diones. The first quinoxaline-2,3-dione (fig. 8) derivatives synthesized were structurally related to the kynurenic acid derivatives and were claimed to be selective AMPA receptor antagonists (Drejer and Honore, 1988; Honore et al., 1989). However, it soon became clear that most were not very selective and also blocked NMDA receptors by interacting with the glycine_B site (Harris and Miller, 1989; Kessler et al., 1989a; Sheardown et al., 1989; Verdoorn et al., 1989; Yamada et al., 1989; Patel et al., 1990). They showed different structure-activity requirements for aromatic ring substitution (Leeson et al., 1991b). 6,7-Dinitroquinoxaline-2,3-dione was most potent as an AMPA antagonist, DCQX was nonselective, 5,7-dinitroquinoxaline-2,3-dione (MNQX) was preferential for the glycine_B site, and 6-cyano-7-nitroquinoxaline-2,3-dione was approximately 10-fold selective for the glycine_B site (Kessler et al., 1989a; Sheardown et al., 1989; Verdoorn et al., 1989; Yamada et al., 1989; Watkins et al., 1990).

6. 3-Hydroxy-1H-1-benzazepine-2,5-diones. A -dicarbonyl system is also present in 3-hydroxy-1H-1-benzazepine-2,5-diones such as 8-chloro-3-hydroxy-1H-1-benzazepine-2,5-dione (Kulagowski, 1996), some of which are also active in vivo (Guzikowski et al., 1995, 1996; Jackson et al., 1995). The in vivo potency of these compounds does not depend just on the in vitro affinity at the glycine_B site, as demonstrated by the 7,8- and 6,8-dimethyl derivatives (fig. 8), which both have high-nanomolar affinity in vitro but are effective in the DBA/2 model at 4 to 6 mg/kg (i.p.) (Guzikowski et al., 1996). These compounds contain elements of both the quinoxaline and kynurenic acid structures, but some may act at multiple sites of the NMDA receptor and also at non-NMDA receptors (Swartz et al., 1992). The 7,8-dichloro-6-methyl and 7,8-dichloro-6-ethyl analogues were the most potent 3-hydroxy-1H-1-benzazepine-2,5-diones, with 4.1 and 2.8 nM affinity at NMDA receptors, respectively, and were 100- to 200-fold less potent at AMPA receptors (Guzikowski et al., 1997).

7. Tricyclic glycine_B site antagonists. 2-Aryl-1H-pyrazolo[3,4-c]quinoline-1,4(2H)-diones also have high affinity at the glycine_B receptor. In particular, structure-activity studies identified 7-chloro-3,5-dihydro-2-(4-methoxyphenyl)-1H-pyrazolo[3,4-c]quinoline-1,4(2H)-dione as the most potent of a series of analogues, with an IC₅₀ of 3.3 nM. The fact that these compounds have acidity equivalent to that of carboxylic acids (pK_a typically of 4.0) probably underlies the fact that they are not systemically active (Macleod et al., 1995). However, some tricyclic quinoline-diones, such as SM-18400 [(S)-9-chloro-5-[p-aminomethyl-o-(carboxymethoxy)phenylcarbamoylmethyl]-6,7-dihydro-1H,5H-pyrido[1,2,3-de]quinoxaline-2,3-dione] (fig. 9), are also active in vivo, despite having a much higher pK_a than classical glycine_B antagonists (Nagata et al., 1994a, 1995; Tanaka et al., 1995; Yasuda et al., 1995; Maruoka et al., 1998). The pyridazino[4,5-b]quinoline-diones M244249 and M241247 (fig. 9) and the cyclic diacyl hydrazide M244,646 were reported to be active against global ischemia in gerbils (Patel et al., 1993), although similar compounds were not effective at up to 100 mg/kg (i.p.) in the MES test in our own laboratories.

8. Prodrugs. Many of the compounds described above were essentially inactive in vivo, indicating very poor access to the CNS. Several strategies were followed to develop prodrugs with improved solubility and reduced polarity. Prodrug esters (L-691,470) (fig. 3) and amides of tetrahydroquinoline and kynurenic acid derivatives (fig. 5) seemed to show somewhat increased in vivo potencies, although it should be stressed that even these prodrugs were active only at 30 to 60 mg/kg in the DBA/2 mouse audiogenic seizure model, which is exquisitely sensitive to NMDA receptor antagonists (Carling et al., 1993; Moore et al., 1993a).

9. Modification of endogenous kynurenic acid metabolism. It has been suggested that inhibition of the synthesis of quinolinic acid by kynurenine 3-hydroxylase decreases levels of this NMDA receptor agonist and increases levels of endogenous kynurenic acid by facilitating alternative pathways. This can be accomplished with compounds like FCE 28833A [(R,S)-3,4-dichlorobenzoylalanine; Upjohn-Pharmacia], which, when administered systemically at 400 mg/kg, increases kynurenic acid levels in the brain, as shown in microdialysis experiments (Speciale et al., 1996). However, this treatment produced estimated maximal brain concentrations of only 120 nM (assuming 20% recovery), i.e., much less than the K_i of kynurenic acid for the glycine_B site of the NMDA receptor (15 µM) (Kessler et al., 1989a). Nevertheless, this treatment has been reported to provide a neuroprotective effect against global ischemia in gerbils (Speciale et al., 1996).

A model accounting for glycine receptor binding of trans-4-amido-2-carboxytetrahydroquinolines was proposed, comprising (a) size-limited, hydrophobic binding of substituents on the benzene ring, (b) hydrogen bond acceptance by the 4-substituent, (c) hydrogen bond donation by the 1-amino group, (d) Coulombic attraction of the 2-carboxylate, and (e) a large hydrophobic bulk tolerance region approximately 5 Å above the plane of the tetrahydroquinoline ring system (Carling et al., 1992). The model can also account for the binding of quinoxaline-2,3-diones, quinoxalic acids, 4-hydroxy-2-quinolones, and 2-carboxybenzimidazoles. The model accounts for differences in the structural requirements for substitutions on the benzene ring, with 7-substitutions with halogens having similar effects but substitutions at the 5-position being different for various classes of antagonists, as exemplified for the 4-hydroxy-2-quinolones by L-701,315.

However, the fact that most very high affinity glycine_B antagonists developed to date are essentially inactive after systemic administration indicates that attempts to improve in vivo activity solely by increasing the in vitro potency of glycine_B antagonists with poor pharmacodynamic properties may be the wrong approach. Moreover, the ability of some full antagonists of the glycine_B site to unmask glycine-sensitive NMDA receptor desensitization may underlie their promising therapeutic profiles (see Section VI.) and seems, in part, to be inversely related to affinity.

Binding of 4-hydroxyquinolones and other glycine_B antagonists to plasma protein was recently proposed to limit the brain penetration of higher affinity compounds (Rowley et al., 1997). Thus, whereas affinity at the glycine_B site increases with lipophilicity, up to a maximum at a water/octanol partition coefficient of approximately 3, binding to plasma protein also correlates with the water/octanol partition coefficient. This was supported by the findings that warfarin increased in vivo potency and that brain penetration in an in situ brain perfusion model in rats was good in the absence of plasma protein. This has important implications for the design of novel compounds and suggests that it is necessary to maintain the water/octanol partition coefficient below 2.4 or keep the pK_a high, to obtain good CNS activity (Rowley et al., 1997).

[³H]Glycine (table 4) has long been used as a ligand to label strychnine-sensitive glycine receptors (Young and Snyder, 1974); more recently, investigators observed that, under certain conditions, this compound labels glycine_B sites (Kishimoto et al., 1981). This ligand has been used in several studies, but centrifugation had to be used for bound ligand separation, because of the low ligand affinity. It is also possible to use the filtration method if Mg²⁺ is added to increase glycine affinity (Canton et al., 1992). [³H]Glycine has also been used for receptor autoradiography (Bristow et al., 1986).

D-[³H]Serine (table 4) was suggested to be a more useful ligand because, in contrast to radiolabeled glycine, it clearly allows exclusion of binding to the inhibitory glycine receptor (Danysz et al., 1990). Originally it was used at 20 nM, and incubations were performed in 50 mM Tris acetate, at pH 7.4, for 20 min at 4°C. Specific binding was determined using 1 mM D-serine (approximately 70%). The major problems involved rather high levels of nonspecific binding and low affinity, again necessitating the use of centrifugation methods to terminate incubations. D-Serine has also been used for receptor autoradiography (Schell et al., 1995). It was recently suggested that D-[³H]serine does not label a single population of sites, because 5,7-diCl-KYN failed to completely inhibit binding (Matoba et al., 1997). The profile of the 5,7-diCl-KYN-insensitive component differed from those of all other defined recognition sites. This component was detected at high levels in the cerebral cortex and in the cerebellum (Matoba et al., 1997). These mysterious sites could be important for the in vivo effects of D-serine.

The partial agonist [³H]ACPC (table 4) was used in two investigations to study glycine_B sites, and it seems to label two sites, with different affinities and densities (Monahan et al., 1990b; Popik et al., 1995). [³H]ACPC was used at 10 to 30 nM, and incubations were performed in Tris N-(2-hydroxyethyl)piperazine-N'-(2-ethanesulfonic acid) (10 mM) for 1 h at 25°C, using 1 mM glycine to determine specific binding. Both the filtration and centrifugation termination methods were used. To our knowledge, [³H]ACPC is not commercially available.

[³H]5,7-diCl-KYN (table 4) was introduced by Baron et al. (1991). They used a 10 nM concentration of this radioligand, in 50 mM N-(2-hydroxyethyl)piperazine-N'-(2-ethanesulfonic acid) KOH, and continued the incubation for 15 min at 4°C at pH 8. D-Serine (100 µM) was used to assess specific binding, which was >70%. This assay was originally based on separation by centrifugation (which was not optimal) but was later adapted to filtration methods (Canton et al., 1992; our own experience) but, in this case, inhibition curves are very shallow and results are quite variable. Standard glycine_B antagonists displaced binding to levels below nonspecific levels determined with unlabeled glycine (Parsons et al., 1997; see also Baron et al., 1991; Yoneda et al., 1993). Moreover, there are some indications that this ligand might be unstable, i.e., the equilibrium binding began to decline spontaneously after 90 min of incubation (Wamsley et al., 1994a). [³H]5,7-diCl-KYN can also be used for autoradiography, but high levels of nonspecific binding (40 to 70%) are an obstacle (Wamsley et al., 1994a).

[³H]L-689,560 (fig. 4), with very high affinity, represents an important advancement because it allows incubation termination by filtration (Grimwood et al., 1992). A 1 nM concentration of ligand was used and the incubation was performed in 50 mM Tris acetate buffer at pH 7 for 120 min at 4°C. Glycine was used at 1 mM to determine specific binding, which was >90%. Unfortunately, this ligand is not commercially available.

Of all ligands used to date, [³H]MDL-105,519 (table 4) has the highest affinity and is available from NEN Life Science Products. It is therefore the ligand of choice to label glycine_B sites (Baron et al., 1996; Hofner and Wanner, 1997). It is routinely used in our laboratory for screening purposes and shows very good reliability and ease of handling. In our studies, membranes were suspended and incubated in 50 mM Tris-HCl, pH 8.0, for 45 min at 4°C, with a fixed [³H]MDL-105,519 concentration of 2 nM. Nonspecific binding was defined by the addition of unlabeled glycine at 1 mM (Parsons et al., 1997). This compound has also been used for autoradiography, which showed that B_max and K_d values were heterogeneous in different structures (Chmielewski et al., 1996). B_max was highest in the cortex and hippocampus (20 to 25 fmol/mm²) and lowest in the cerebellum (6 fmol/mm²). The K_D ranged from 7 nM in the cortex to 21 nM in the caudate putamen. For autoradiography, 50 mM Tris acetate at 22°C and pH 7.4 was used for incubation (10 nM ligands, 30-min incubation), and specific activity was determined with 1 mM glycine (Chmielewski et al., 1996).

In an analysis of the data presented in table 4, one of the striking inconsistencies is that different B_max values were obtained by using different radioligands for the glycine_B site. In general, the compounds seem to fall into two categories, i.e., those having B_max values of approximately 3 to 5 or 12 to 15 pmol/mg of protein. This could indicate that [³H]5,7-diCl-KYN and [³H]MDL-105,519 label additional sites.

Glutamate has been reported to enhance [³H]glycine binding in many experiments (e.g., Nguyen et al., 1987) (table 2). In patch-clamp experiments, the off-rate of glycine (which is often slower with higher affinity) is fastest in the presence of partial agonists with low intrinsic activity at the NMDA site and progressively slower in the presence of quinolinate, NMDA (partial agonists), and L-glutamate (full agonists) (Priestley and Kemp, 1994). In contrast, most patch-clamp studies indicate that the more physiologically relevant effect of glutamate involves a decrease in glycine affinity, and this phenomenon finds expression as desensitization (Benveniste et al., 1990a,b).

These interactions in binding experiments are reciprocal, i.e., glycine and D-serine enhance both glutamate affinity and glutamate efficacy to increase [³H]TCP binding (Fadda et al., 1988; Monaghan et al., 1988; McKernan et al., 1989; Stirling et al., 1989). In line with this, electrophoresis experiments in the hemisected spinal cord showed that D-serine increases the affinity of NMDA (Siarey et al., 1990). Again, this effect may be related to intrinsic activity at the glycine_B site, because partial agonists with low intrinsic activity, such as R(+)HA-966, ACBC, and L-687,414 (table 2), actually inhibit NMDA agonist binding. Similar effects on response kinetics were seen in patch-clamp experiments (Priestley and Kemp, 1994). Moreover, the opposite has been observed in binding experiments using radiolabeled C7 competitive NMDA receptor antagonists such as [³H]CPP (Danysz et al., 1989a; Compton et al., 1990; Hood et al., 1990; Kaplita and Ferkany, 1990; Grimwood et al., 1993). In that case, enhancement was associated with lower intrinsic activity (Grimwood et al., 1993). In contrast, the binding of competitive antagonists of the C5 type [e.g., ³H-labeled CGS-19755 [cis-4-(phosphonomethyl)-2-piperidinecarboxylic acid]] is decreased by agonists at the glycine_B site and partial agonists with high intrinsic activity, such as D-cycloserine, but not by partial agonists with low intrinsic activity (table 2).

Glycine_B site full antagonists have been thought not to allosterically affect the NMDA recognition site (table 2). This hypothesis has been confused by the finding that glycine_B site full antagonists such as L-701,324 and L-695,902 also enhance the binding of [³H]CPP and [³H]CGS-19755 (Grimwood et al., 1995a) and decrease glutamate affinity in a manner similar to that of R(+)HA-966 (Priestley et al., 1996). Hence, this allosteric interaction is not directly related to intrinsic activity at the glycine_B recognition site but rather involves other features of the interacting molecules (Grimwood et al., 1995a). In addition, other ligands show interactions with the glycine_B site; this refers particularly to polyamines such as spermine, which have been shown to enhance glycine but not 7-Cl-KYN potency in the functional [³H]MK-801 assay (Sacaan and Johnson, 1989; Ransom and Deschenes, 1990; Benveniste and Mayer, 1993; Marvizon and Baudry, 1994).

Similarly, glycine affinity is modified by divalent cations in a manner not related to direct actions on channel conductance (Peters et al., 1987; Forsythe et al., 1988). As previously mentioned, Zn²⁺ has been shown to decrease the affinity of glycine (Reynolds and Miller, 1988; Yen et al., 1990). This effect could possibly increase the threshold for saturation of the glycine_B site under physiological conditions, because zinc is present in the brain and in some regions, e.g., dentate gyrus of the hippocampus, can be released in an activity-dependent manner (Aniksztejn et al., 1987).

The endogenous polyamines spermine and spermidine have multiple effects on the activity of NMDA receptors. These include an increase in the magnitude of NMDA-induced whole-cell currents seen in the presence of saturating concentrations of glycine, an increase in glycine affinity, a decrease in glutamate affinity, and voltage-dependent inhibition at higher concentrations (Johnson, 1996; Williams, 1994a,b, 1997; Williams et al., 1994).

Glycine-independent stimulation by polyamines requires the presence of NR1 variants that lack an amino-terminal insert, such as NR1₀₁₁ (NR1a). The stimulatory effect is also controlled by NR2 subunits in heteromeric complexes; it is observed at heteromeric NR1a/NR2B receptors but not at heteromeric NR1₀₁₁/NR2A or NR1₀₁₁/NR2C receptors (Williams et al., 1994; Kashiwagi et al., 1997). Glycine-dependent stimulation is mediated via a three-fold increase in glycine affinity (Williams, 1994a; Williams et al., 1994) accompanied by a similar decrease in the rate of development of glycine-sensitive desensitization and a decrease in the rate of dissociation of glycine (Benveniste and Mayer, 1993). This effect probably involves a second binding site, because it is also seen at NR1₀₁₁/NR2A receptors. Argiotoxin 636 produced a similar potentiation of NMDA responses at positive potentials, accompanied by a slowing in the rate of current desensitization and an increase in the affinity for glycine (Donevan and Rogawski, 1996).

Protons inhibit NMDA receptor function through interactions with the NR1 subunit, and both polyamines and the N1 (exon 5) insert potentiate receptor function through relief of the tonic proton inhibition present at physiological pH. A single amino acid (lysine 211) mediates the effects of N1 in the rat brain. This effect, together with the structural similarities between polyamines and the surface loop encoded by N1, suggest that exon 5 may act as a tethered, pH-sensitive, constitutive modulator of NMDA receptor function (Traynelis et al., 1995).

Mg²⁺ also induces a shift of the pH sensitivity of NMDA receptors at the same site as polyamines (Paoletti et al., 1995), and relatively high concentrations (EC₅₀ = 3 mM) potentiate NMDA receptor-mediated responses at positive potentials by increasing the potency of glycine. This effect is associated with a reduction in the magnitude of desensitization and slower desensitization and glycine offset kinetics. Similar effects on kinetics were seen with Ca²⁺, indicating that both cations may enhance NMDA responses at positive potentials via interactions with the glycine site (Gu and Huang, 1994; Wang and Macdonald, 1995). Ketamine has also been reported to increase NMDA receptormediated currents in cultured mouse hippocampal neurons and HEK 293 cells expressing NMDA 1/2 receptors by increasing the affinity of the glycine_B site (Zhang et al., 1994; Wang and Macdonald, 1995), and we have made similar observations with memantine and some novel uncompetitive NMDA receptor antagonists (Parsons et al., 1998a).

Nondesensitizing responses of NR1/NR2B receptors expressed in Xenopus oocytes elicited in the presence of subsaturating concentrations of glycine were converted into desensitizing responses by the addition of ethanol, an effect that was reversed with increasing glycine concentrations. The ability of ethanol to promote glycine-sensitive desensitization further suggests an interaction between glycine and ethanol inhibition of the NMDA receptor (Buller et al., 1995).

	V. N-Methyl-D-Aspartate Receptor Subtypes: Differences in GlycineB Recognition Sites

Top Previous Next References

Heterogeneity of glycine_B sites was first suggested by Monaghan et al. (1988), who showed that NMDA receptors in various brain regions differed in the stimulatory effects of glycine on [³H]glutamate binding, as assessed by autoradiography, i.e., the effect was stronger in the thalamus and cortex than in the striatum, septum, and cerebellum. Similarly, D-cycloserine was approximately 6 times more potent in inhibiting [³H]glycine binding in cortical membranes than in the cerebellum; a similar phenomenon was observed for glycine, but the difference was much smaller (2.5-fold) (O'Shea et al., 1991). In contrast, there was no difference in the potencies of D-serine and 7-Cl-KYN (O'Shea et al., 1991). Also, in comparisons of [³H]glycine and D-[³H]serine binding in the spinal cord and cortex, it was observed that 7Cl-KYN was a very weak inhibitor in the former preparation, whereas HA-966 was equally active in the two preparations (Danysz et al., 1990). In the slice preparation, glycine was potent in reversing the desensitization-related reduction of NMDA-stimulated dopamine release but failed to reverse the diminution of acetylcholine release, possibly indicating the involvement of different NMDA receptor subtypes (Cai et al., 1991).

In fact, it was confirmed that NMDA receptor subtypes do indeed show different pharmacological characteristics of the glycine_B site. This aspect is very important for drug development, because targeting of specific NMDA receptor subtypes might allow better separation of therapeutic and side effects (Watanabe et al., 1993b; Wenzel et al., 1995). Studies in Xenopus oocytes and cell lines expressing NMDA receptor subtypes show substantial differences in the sensitivity to glycine and other ligands at the glycine_B site (table 5). The most consistent feature is the observation that glycine is more potent at NMR2D-containing receptors. Although one binding study indicates relative affinities of NR1/NR2C NR1 (homomeric) = NR1/NR2B = NR1/NR2D > NR1/NR2A (Laurie and Seeburg, 1994a), most electrophysiological studies indicate a rank order of NR1a/NR2D > NR1a/NR2C > NR1a/NR2B > NR1a/NR2A (Kutsuwada et al., 1992; Wafford et al., 1993; Matsui et al., 1995; Priestley et al., 1995; Woodward et al., 1995a; Buller and Monaghan, 1997) (table 5). This difference is reflected in very much slower deactivation kinetics of receptors with higher affinity for glycine (Monyer et al., 1992) and glutamate (P. Spielmanns, unpublished data). Interestingly, D-serine shows far less selectivity (Matsui et al., 1995; Priestley et al., 1995; Krueger et al., 1997). The partial agonists R(+)HA-966 and L-687,414 were three-fold more potent at NR1a/NR2B receptors than at NR1a/NR2A receptors, but their intrinsic activities were similar (Priestley et al., 1995). Differences in the potencies and intrinsic activities of other partial agonists were discussed in Section IV.B. (see also table 5).

The molecular basis for this selectivity has not been fully elucidated but is probably related to allosteric interactions between NR2 subunits and the glycine recognition site on NR1 subunits (Hirai et al., 1996; Laube et al., 1997; Wood et al., 1997). It is also likely that affinity is directly dependent on splice variants of NR1 subunits, but this is difficult to test conclusively because homomeric NR1 receptors do not normally form functional receptors when expressed in mammalian cell lines (Grimwood et al., 1995b; Ishmael et al., 1996). Nonetheless, when expressed as homomeric receptors in Xenopus oocytes, splice variants containing the N1 insert, such as NR1₁₀₀, showed lower affinity for glutamate, NMDA, and 2-amino-5-phosphonovaleric acid (AP5), no glycine-independent potentiation by spermine, histamine, or Mg²⁺, greater potentiation by protein kinase C, and stronger inhibition by (+)MK-801 and ketamine, compared with NR1a receptors (Durand et al., 1992, 1993; Williams, 1994b; Paoletti et al., 1995; Rodriguez Paz et al., 1995; Albrecht et al., 1996). It should be noted that others have attributed apparent differences in sensitivity to channel blockers to different blocking kinetics (Monaghan and Larsen, 1997). Levels of inhibition by Zn²⁺ were similar, but potentiation at lower concentrations was seen only with receptors lacking the amino-terminal insert (Hollmann et al., 1993). Variants differing only in the carboxyl-terminal domain showed little change in agonist affinity or spermine potentiation (Durand et al., 1992, 1993).

More relevant for the present review is the fact that NR1₀₁₁ and NR1₁₀₀ receptors showed nearly identical affinities for glycine and Mg²⁺ (Durand et al., 1992). It is truly unfortunate that other studies have not directly addressed this point of paramount importance, because the glycine recognition site is located on NR1 receptors. This is probably because the splice variants of NR1 receptors could be predicted to have a minor influence on the putative binding pocket for glycine, and hence glycine affinity. However, confirmation of this concept would be helpful. Coexpression studies with common NR2 subunits such as NR2A would overcome problems in the interpretation of such experiments in Xenopus oocytes.

There are surprisingly few data on the NMDA receptor subtype selectivity of glycine_B antagonists for all four NR2 subunits. The exception is the thorough electrophysiological studies by CoCensys, which showed that the IC₅₀ values of ACEA 1011 against NR1A/NR2A, NR1A/NR2B, NR1A/NR2C, and NR1A/NR2D subunit combinations expressed in Xenopus oocytes ranged from 0.4 to 7.8 µM in the presence of 1 µM glycine. This 20-fold variation in sensitivity was the result of a combination of subunit-dependent differences in glycine and antagonist affinities; EC₅₀ values for glycine ranged from 0.09 to 0.7 µM, and K_b values for ACEA 1011 ranged from 0.2 to 0.7 µM (Woodward et al., 1995b). The same group reported similar data for ACEA 1021 (Woodward et al., 1995a).

[³H]5,7-diCl-KYN was reported to bind with similar affinities to all recombinant receptors (K_d ~ 50 to 100 nM) (Laurie and Seeburg, 1994a). However, it should be noted that the rank orders of potency of both glutamate and glycine in that study diverged strongly from those observed in electrophysiological studies. [³H]L-689,560 was also reported to bind to both NR1₀₁₁- and NR1₀₀₀-transfected HEK 293 cells with high affinity. The affinities of glycine_B antagonists to inhibit [³H]L-689,560 binding to NR1₀₀₀-transfected cells were similar to those observed with rat brain membranes, whereas affinities at NR1₀₁₁ receptors were two-fold lower. Affinity values for agonists and partial agonists were 4- to 16-fold lower, indicating that the glycine site of homomeric NR1 receptors is in an antagonist-preferring state (Grimwood et al., 1995b).

Receptor subtype selectivity would allow glycine_B site full antagonists to block NMDA receptor function in a manner similar to that of partial agonists in cells expressing heterologous populations of NMDA receptors. For example, highly selective antagonists for receptors containing NR2A subunits would block responses to a maximum of 50% in cells expressing mostly NR1/NR2A and NR1/NR2B receptors at equal levels. Although some glycine_B antagonists have already been reported to show true NMDA receptor subtype selectivity, perhaps more important for in vivo therapeutic profiles is functional subtype selectivity, which could also be related to differences in the affinity of glycine and/or regional variations in endogenous glycine concentrations. Such effects may underlie the improved TIs of some glycine_B site antagonists. For example, MDL-100,458 and MDL-102,288 are equipotent as glycine_B antagonists in vitro but exhibit strikingly different in vivo profiles for audiogenic seizures in DBA/2 mice and for separation-induced ultrasonic vocalizations in rat pups (a model of anxiolytic activity) (Kehne et al., 1995).

It should be stressed that the selectivity of most of the existing glycine ligands for certain subtypes of NMDA receptors is probably far too small for determination of pharmacological characteristics in vivo. However, our intention was to emphasize, by giving selected examples, potential pharmacological targeting of specific subtypes.

	VI. Therapeutic Aspects for Agents Acting at the GlycineB Site

Top Previous Next References

Although some studies using central injection of glycine_B antagonists are discussed in the present section, they are of limited therapeutic importance. In our opinion, only experiments based on systemic administration yield findings regarding possible therapeutic applications, as exemplified by the study showing paradoxically anticonvulsive effects of the GABA_A receptor antagonist bicuculline after central microinjections into the striatum (Turski et al., 1989a). Exemption from this rule should include therapeutic indications where local administration is desirable, e.g., intrathecal infusion to obtain antinociceptive effects.

The general potential usefulness of NMDA receptor antagonists as therapeutic agents has been discussed elsewhere (Meldrum, 1985; Rogawski, 1993; Leeson and Iversen, 1994; Danysz et al., 1995a; Besnard et al., 1996; Ishimaru and Toru, 1997; Parsons et al., 1998b). In the present review, we therefore decided to focus on specific aspects of glycine or its recognition site on NMDA receptors. It is necessary to stress that the common opinion that NMDA receptors are involved only in learning and pathological changes, because of their voltage-dependent blockade by Mg²⁺, must be put into context. If the situation were so simple, then the development of NMDA receptor antagonists (including glycine_B antagonists) for any application except neuroprotection would seem pointless. It is becoming clear that NMDA receptors also play crucial roles in other forms of plasticity, such as drug dependence and addiction, chronic pain, and CNS development, as well as in normal or disturbed synaptic transmission in some areas of the CNS. In view of the basic aspects discussed above, it is clear that activation of NMDA receptors depends not only on the level of synaptic activity but also on other factors, such as agonist affinity, gating kinetics, and Mg²⁺ sensitivity. In turn, the role of NMDA receptors in various processes depends on the subtype composition and area of the CNS involved.

The crucial questions that have been addressed are as follows:

1.	Is there specific dysfunction in glycine/D-serine modulation of NMDA receptors in various disease states or models thereof?
2.	Are glycineB antagonists different from or better than agents acting at other recognition sites of the NMDA receptor?
3.	Is initial clinical experience with glycineB antagonists promising?

There is a rational theoretical basis for a "better" therapeutic potential of glycine_B antagonists, particularly when neuroprotection is considered. Moderate concentrations of glycine_B antagonists would not "switch off" responses completely but, rather, would negatively modulate NMDA receptors through an increase in receptor desensitization, i.e., by potentiation of a built-in protective mechanism against prolonged overactivation.

In our opinion, there have been two major negative trends in the development of new glycine_B antagonists, which have slowed progress in this area. First, a great deal of effort was put into the development of high affinity agents without much initial concern regarding brain availability; this point has already been discussed. Second, very sensitive convulsion tests are often selected for detection of in vivo activity in the CNS; this favors a positive effect. For example, observation of audiogenic seizures in DBA/2 mice seems to be an exquisitely sensitive model for NMDA receptor antagonists, compared with either MES- or NMDA-induced convulsions (Chapman, 1991; Tricklebank et al., 1994; Bristow et al., 1996b). In fact, on the basis of activity in this model, several glycine_B antagonists were claimed to be systemically active, but they were obviously not sufficiently active for further development (Russi et al., 1992; Moore et al., 1993a; Rowley et al., 1993; see Parsons et al., 1997).

Another major problem with some glycine_B antagonists (e.g., SC-49648, ACEA 1021, and MRZ 2/576) is not brain penetration, which is quite good and fast (1 to 5 min), but rapid transport back out of the brain. This is clearly reflected in brain pharmacokinetics or short duration of functional effects in the CNS, such as anticonvulsive actions or inhibition of cGMP stimulation in the cerebellum (Robinson et al., 1993; Baron et al., 1997; Hesselink et al., 1997; Parsons et al., 1997). This efflux from the brain is most likely associated with the activity of the organic acid transporter in the choroid plexus, because the addition of probenecid considerably prolongs half-lives (Hesselink et al., 1997; Parsons et al., 1997). This may be a less important problem for other compounds such as MDL-105,519 and L-701,324, which have longer half-lives of approximately 3 and 1 h (or >15 h for the -phase), respectively (Bristow et al., 1996b; Baron et al., 1997); in human subjects, half-lives may be even longer (see Section VII.).

Because glycine_B antagonists are generally intended to be CNS-active agents, the brain concentrations reached must compare with in vitro activity at the glycine_B site, to relate therapeutic effects to the proposed mechanism of action and permit adjustment of dosages accordingly. One of the most spectacular examples (from another area of investigation) in which this rule was ignored involves the claim that aniracetam acts as a cognitive enhancer through inhibition of AMPA receptor desensitization, which is seen at 2 mM concentrations in vitro (Ito et al., 1990). In contrast, brain levels reached at cognition-enhancing doses are in the low-nanomolar range (Ichihara et al., 1986).

The best method to study brain concentrations of drugs is microdialysis with in vivo recovery (Benveniste and Huttemeier, 1990). By comparisons of brain and serum levels, ratios can be established, which, in turn, are very helpful for adjusting doses in clinical studies.

Based on studies in animals, there are indications that glycine_B antagonists do not produce many of the side effects often observed with competitive NMDA receptor antagonists and high affinity channel blockers. Early studies suggested better TIs for glycine_B antagonists, but it should be stressed that agents with very poor brain penetration (such as 7-Cl-KYN) were used (Rao et al., 1993). It was reported that 7-Cl-KYN at up to 500 mg/kg did not affect locomotion and was devoid of myorelaxant activity in mice (screen test) (Ginski and Witkin, 1994). The same agent, however, did produce ataxia at low doses (20 nmol) when administered i.c.v. to rats (Danysz and Wroblewski, 1989). It has even been claimed that agents acting at the glycine site are completely devoid of certain side effects (Chiamulera et al., 1990). This is difficult to reconcile with the necessity of glycine for NMDA receptor activation (Kleckner and Dingledine, 1988) and the fact that higher concentrations of glycine_B antagonists produce complete blockade of channel activity. These apparent contradictions can possibly be explained as follows.

1.	The differences between glycineB antagonists and competitive antagonists or channel blockers are not qualitative but rather are quantitative, resulting in better TIs.
2.	Compared with competitive or uncompetitive antagonists, some glycineB antagonists may show better selectivity for NMDA receptor subtypes of "therapeutic" interest versus those mediating side effects.
3.	Subtype-selective glycineB antagonists could produce functional partial antagonism in cells expressing heterologous populations of NMDA receptors.
4.	The induction of desensitization by glycineB antagonists could differentiate among different forms of NMDA receptor activation.

1. Drug discrimination. Drug discrimination is often used to test whether different agents share similar subjective cues, e.g., compared with commonly abused agents such as cocaine or PCP. In the case of NMDA receptor antagonists, this paradigm has been used to address the question of whether agents acting at different sites might have similar side effects and abuse potential.

In rats, R(+)HA-966 (at 30 mg/kg), but not its S() isomer, has been found to produce discriminative behavior in the two-lever operant responding paradigm (Singh et al., 1990b

2. Learning impairment. In animals, most NMDA receptor antagonists produce impairment of learning when given at sufficiently high doses before the association phase but not when administered after this phase or during retrieval (Danysz et al., 1995b). Although this statement is an oversimplification, it is the intention of the present review not to discuss the specificity of such effects but, rather, to assess the risk of learning impairment by glycine_B site antagonists. This is of therapeutic relevance because other types of NMDA receptor antagonists, e.g., ketamine (Malhotra et al., 1996), PCP (Luby et al., 1959), and D-3-(2-carboxypiperazine-4-yl)-1-propenyl-1-phosphonic acid (Rockstroh et al., 1996; Herrling et al., 1997), produce clear learning deficits in humans.

3. Ataxia, myorelaxation, and sedation. Ataxia and myorelaxation seem to be very common side effects observed with NMDA receptor antagonists (Koek and Colpaert, 1990; Murata and Kawasaki, 1993; Carter, 1994). These side effects (measured using rotarod or inclined mesh tests, for example) have often been used in comparison with anticonvulsive potencies to provide TI values. It should be stressed that the absolute TI values are almost meaningless, because they depend on the type of seizures, the strength/dose of convulsive treatment, rotarod speed, etc.; modification of these parameters may result in a wide range of TI values. This also implies that often no sensible comparison of results from different investigations is possible. Instead, reference agents should always be used to draw any conclusions regarding the relative risk of side effects.

4. Neurotoxicity in the retrosplenial/cingulate cortex. Olney et al. (1989, 1991) observed that high doses of competitive or uncompetitive NMDA receptor antagonists produce neuronal vacuolization in the cingulate/retrosplenial cortex in rodents. Some of the neurons containing vacuoles (emerging from mitochondria) may eventually die through necrosis and possibly programmed cell death (Fix et al., 1994). This feature is seen with all known and properly tested uncompetitive and competitive NMDA receptor antagonists.

5. Psychotomimetic side effects. Psychotomimetic effects are often seen in human subjects treated with high doses of either uncompetitive or competitive NMDA receptor antagonists (Luby et al., 1959; Albers et al., 1991; Kristensen et al., 1992; Muir et al., 1994). In rodents, indications of such activity are (among others) locomotor activation, backpedaling, and stereotypic sniffing, circling, and head weaving. Indeed, in rats most uncompetitive NMDA receptor antagonists produce these effects when given at sufficiently high doses (Contreras et al., 1988; Koek and Colpaert, 1990; Hoffman, 1992; Kretschmer et al., 1992; Murata and Kawasaki, 1993; Schmidt and Kretschmer, 1997). However, it has been shown that competitive antagonists also cause locomotor activation when the test is performed after a sufficient time delay or high doses are given (Löscher and Honack, 1993; Furuya and Ogura, 1997).

In 1986 it was reported that a high dose of the competitive NMDA receptor antagonist 2-amino-7-phosphonoheptanoic acid produced anxiolytic activity (Bennett and Amrick, 1986; Stephens et al., 1986). These findings were followed by several publications confirming this action with competitive and uncompetitive NMDA receptor antagonists in the conflict test (Corbett and Dunn, 1993; Plaznik et al., 1994), the social interaction test (Dunn et al., 1989; Corbett and Dunn, 1993), the elevated plus-maze test (Dunn et al., 1989; Corbett and Dunn, 1993), and the test of separation-induced vocalization in rat pups (Kehne et al., 1991), as well as by blockade of the fear-potentiated startle response (Anthony and Nevins, 1993). Similarly, glycine_B antagonists and partial agonists have been suggested to show an encouraging therapeutic anxiolytic profile (Trullas et al., 1989; Corbett and Dunn, 1991, 1993; Kehne et al., 1991, 1995; Anthony and Nevins, 1993; Plaznik et al., 1994).

The partial agonist D-cycloserine (30 mg/kg) exerts anxiolytic activity in the fear-potentiated startle response test in rats (Anthony and Nevins, 1993). Similarly, we recently observed a clear anxiolytic effect of this agent in the elevated plus-maze test in rats (starting at 10 mg/kg) (Karcz-Kubicha et al., 1997). In the same model, ACPC at 300 to 400 mg/kg (but not at 500 mg/kg) was shown by Trullas et al. (1989, 1991) to increase both the percentage of time spent in open arms and the percentage of entries into open arms for mice, which was interpreted as a decrease in anxiety. However, the efficacy of ACPC was significantly lower than that of chlordiazepoxide (Trullas et al., 1989). Anxiolytic properties of ACPC have also been shown in rats in the fear-potentiated startle response test (starting at 200 mg/kg, with complete blockade at 500 mg/kg) (Anthony and Nevins, 1993) and in the conflict test (100 and 200 mg/kg) (Przegalinski et al., 1996). Similarly, positive effects of this agent (12.5 to 200 mg/kg) in the model of separation-induced ultrasonic vocalization in rat pups have been reported (Winslow and Insel, 1991). In mice trained in a dark avoidance task, the step-down latencies measured within 24 h, but not 1 h, were prolonged by posttraining administration of ACPC (400 mg/kg), 7-Cl-KYN (30 mg/kg), or D-cycloserine (1 mg/kg), indicating anxiolytic activity (Faiman et al., 1994). Pretreatment with glycine abolished this effect, confirming the involvement of the glycine_B site (Faiman et al., 1994). In contrast, we failed to observe anxiolytic effects of ACPC (up to 600 mg/kg) in rats using the plus-maze test (Karcz-Kubicha et al., 1997).

Corbett and Dunn (1991) demonstrated anxiolytic-like activity of racemic HA-966 (1 to 3 mg/kg) in the conflict, social interaction, and plus-maze tests in rats. The selective R(+)-enantiomer has also been shown to decrease fear-potentiated startle responses in the same species at doses of 10 and 30 mg/kg (Anthony and Nevins, 1993). In accordance with this result, R(+)HA-966 injected into dorsal periaqueductal gray matter increased both the percentage of time spent in open arms and the percentage of entries into open arms in the elevated plus-maze test in rats (Matheus et al., 1994). In our studies, R(+)HA-966 also produced anxiolytic effects in this model (starting at a dose of 1 mg/kg), but the magnitude of this action was very modest, in comparison with those of diazepam and D-cycloserine (KarczKubicha et al., 1997).

The glycine_B site full antagonist 7-Cl-KYN, given i.p. at a dose of 25 mg/kg or administered directly into the dorsal periaqueductal gray matter, decreased anxiety in the elevated plus-maze test (Trullas et al., 1989; Matheus et al., 1994). Another glycine site antagonist, 5,7-diCl-KYN, when injected i.p. significantly increased social interaction behavior (doses of 30 and 100 mg/kg), open-arm exploration time (100 mg/kg), and conflict responding in rats (Corbett and Dunn, 1993). This compound was also tested, with positive effects for anxiety, in the model of separation-induced ultrasonic vocalization in rat pups (Kehne et al., 1991) and in the Vogel conflict test (5 µg, i.c.v.) (Plaznik et al., 1994). Another glycine_B antagonist, MDL-105,519, has been shown to inhibit ultrasonic vocalization in rat pups (Baron et al., 1997), with an ED₅₀ of 40 mg/kg, which is close to its ED₅₀ for myorelaxant effects (73 mg/kg). In our studies, the specific and high affinity glycine_B antagonist L-701,324 (Bristow et al., 1996b; Priestley et al., 1996) had very weak anxiolytic effects in the elevated plus-maze test (3 to 10 mg/kg) and was ineffective in the Vogel conflict test (0.1 to 10 mg/kg) (Karcz-Kubicha et al., 1997). Similarly, no effect of MRZ 2/576 (or related agents) was seen in rats in the elevated plus-maze test or in the Vogel conflict test, at doses that result in brain levels sufficient to inhibit NMDA receptors (i.e., up to 10 mg/kg) (Karcz-Kubicha et al., 1997; Parsons et al., 1997). Somewhat in line with these findings, Wiley et al. (1995) failed to show consistent anxiolytic activity of ACEA 1021 in the elevated plus-maze test in rats; a significant but modest increase was seen only in open-arm entries at the highest dose of 30 mg/kg. In contrast, anxiolytic activity of L-701,324 (at 2.5 and 5 mg/kg) in the plus-maze test and in the punished drinking test was recently reported (Kotlinska and Liljequist, 1998).

An intriguing question is whether anxiolytic activity is an inverse function of the intrinsic activity of glycine_B site antagonists/partial agonists, i.e., whether activity is simply a consequence of NMDA receptor inhibition. By analyzing the data described above and by directly comparing several agents with different levels of intrinsic activity (from 0 to 92%), we obtained a negative answer to this question (Karcz-Kubicha et al., 1997).

These somewhat inconsistent results could indicate that the antagonists studied have preferences for NMDA receptor subtypes differently involved in fear regulation. In fact, studies with MDL-100,458 and MDL-102,288 indicated that the former was 100 times more potent as an anticonvulsant in DBA/2 mice, whereas the latter was 13 times more potent as an anxiolytic in a separation-induced vocalization model (Kehne et al., 1995). One consistent finding was that, in the majority of cases, the anxiolytic efficacy of glycine_B antagonists is far weaker than that of benzodiazepines. Therefore, although they are scientifically interesting, this class of agents cannot currently be considered to be promising for the development of new anxiolytic agents that are superior to existing treatments.

Many antidepressants seem to have NMDA channel-blocking properties, but their modest affinities in this regard do not allow conclusions to be drawn regarding the relevance of these properties to the therapeutic efficacy of these drugs (Reynolds and Miller, 1988; Leander, 1989; Sills and Loo, 1989; Kitamura et al., 1991). However, there are indications that dysfunctions of glycine_B site regulation could occur in depression. In patients with major depression, decreases in serum glycine concentrations and increases in the serine/glycine ratio were observed (Altamura et al., 1995); another report indicated an increase in plasma serine levels (Maes et al., 1995).

In the frontal and parietal cortex of suicide victims, no difference in the glycine stimulation of [³H]MK-801 binding was observed (Palmer et al., 1994a), but Nowak et al. (1995a) reported that the proportion of high affinity, glycine-displaceable binding of ³H-labeled CGP-39653 [DL-(E)-2-amino-4-propyl-5-phosphono-3-pentenoic acid] was reduced in the frontal cortex. However, other parameters, such as the potency or efficacy of glycine to interact with [³H]5,7-diCl-KYN, [³H]CGP-39653, or [³H]MK-801 binding, were not altered (Nowak et al., 1995a).

In animal models of depression, such as the forced-swim test and stress-induced anhedonia test, NMDA receptor antagonists exert positive effects in most cases (Maj, 1992; Moryl et al., 1993; Papp and Moryl, 1994; Przegalinski et al., 1997). A positive effect of ACPC (400 mg/kg) in the forced-swim test in rats was probably the first indication that glycine_B ligands might have antidepressive potential (Trullas and Skolnick, 1990). An antagonistic effect of ACPC at the glycine_B site was involved in these findings, because the effect was reversed by glycine (Trullas and Skolnick, 1990). This original study was recently replicated, i.e., ACPC (200 to 400 mg/kg) produced a reduction of immobility time in the forced-swim test in rats (Przegalinski et al., 1997). ACPC and D-cycloserine were also studied in a chronic mild stress model of depression (Papp and Moryl, 1996), where a decrease in the consumption of sucrose is used as an index of anhedonia and is sensitive to antagonism by classical antidepressant drugs. A positive effect was seen after chronic (5-week) treatment with ACPC (starting at 100 mg/kg), whereas less consistent (non-dose-dependent) effects were seen with D-cycloserine (significant effect at 10 mg/kg) (Papp and Moryl, 1996).

It is noteworthy that chronic treatment of mice with (+)MK-801 or ACPC produces -adrenoceptor down-regulation (measured as [³H]dihydroalprenolol binding) in cortical membranes (Paul et al., 1992); this effect was previously reported to be the most common change after various antidepressive therapies (Vetulani and Sulser, 1975). Antidepressants also change NMDA receptor function. Chronic antidepressant treatment (or electroconvulsive shock treatment) has been reported to cause a reduction in the potency of glycine to displace [³H]5,7-diCl-KYN binding and a reduction in the proportion of high affinity, glycine-displaceable binding of the competitive NMDA receptor antagonist [³H]CGP-39653 (Nowak et al., 1993; Paul et al., 1993, 1994). It is interesting to note that various treatments, such as classical antidepressants (amitriptyline and desipramine), atypical antidepressants having different mechanisms of action (mianserine and citalopram), and electroconvulsive shocks, all produce the same reduction in glycine potency to inhibit [³H]5,7-diCl-KYN binding. On the other hand, chronic stress (which produces motivational deficits in rats) had the opposite effect, i.e., an increase in the potency of glycine to inhibit [³H]5,7-diCl-KYN binding (Nowak et al., 1995b).

Recently, chronic treatment with huge doses of glycine (500 mg/kg) was shown to attenuate the increase in activity after bulbectomy (another model used to study antidepressive agents) and to attenuate the activity change in response to PCP in this model, resembling results obtained with antidepressants (Redmond et al., 1996). This is somehow in contrast to the antidepressive-like effects of NMDA receptor inhibition described above.

In summary, there are data suggesting that chronic treatment with antidepressants produces a change in glycine interactions with the NMDA receptor (possibly resulting from alterations in the subunit composition), which requires further investigation. The impact of this change on the function of NMDA receptors and the pathomechanism of depression is also unclear. Concerning the therapeutic potential of glycine_B antagonists as antidepressants, only ACPC is being developed in this direction (table 8); ongoing clinical trials of treatments for depression should validate the therapeutic utility of glycine_B modulators in depression. In our opinion, even if ACPC proves to be effective, this finding should not be generalized to other glycine_B antagonists or partial agonists, which all seem to have very different pharmacological profiles.

Because high affinity NMDA channel blockers, such as PCP, mimic both positive and negative symptoms of schizophrenia in humans, it has been suggested that hypofunction of the glutamatergic system might occur in this disease (Carlsson and Carlsson, 1990; Javitt and Zukin, 1991; Halberstadt, 1995; Ishimaru and Toru, 1997). In fact, there are some indications of glutamatergic hypofunction in schizophrenic patients, such as decreased NMDA receptor-stimulated glutamate release in synaptosomal preparations (Sherman et al., 1991; Ishimaru and Toru, 1997). It has been suggested that enhancement of glutamatergic transmission could potentially have beneficial effects (Carlsson and Carlsson, 1990). It would be difficult to consider use of direct NMDA receptor agonists, because of their neurotoxic and convulsive potential, but positive stimulation of the glycine site might be a favorable approach. The latter idea is supported by the findings that some schizophrenic patients exhibit deficiencies in a glycine-synthesizing enzyme (serine hydroxymethyl transferase) and may respond positively to glycine administration (Waziri, 1988). In contrast, other authors did not find differences in glycine concentrations for schizophrenic patients after loading with high doses of L-serine (the direct precursor of glycine), arguing against a deficit in serine hydroxymethyl transferase (Lucca et al., 1993; but see Ishimaru et al., 1994). Nevertheless, in schizophrenia there might be dysfunctions of the glycine/serine balance, because increased glycine levels in the orbitofrontal cortex and reduced serine levels in the putamen have been shown (Kurumaji et al., 1996). The number of strychnine-insensitive [³H]glycine binding sites (B_max) was found to be increased in 6 of 16 cortical areas studied in postmortem brain samples from chronic schizophrenics (Kurumaji et al., 1996). The authors suggested that the increase could possibly be ascribed to compensation for impaired glutamatergic neurotransmission.

In rats, chronic treatment with neuroleptic agents (haloperidol, pimozide, clozapine, and risperidone) resulted in decreases in glycine stimulation of [³H]MK-801 binding (McCoy and Richfield, 1996). The effect was specific, because stimulation by NMDA or spermidine was not affected. Interestingly, based on glycine-sensitive inhibition of NMDA responses in voltage-clamped hippocampal neurons (IC₅₀ = 1.9 µM), it has been suggested that haloperidol is a partial agonist at the glycine_B site (Fletcher and Macdonald, 1993).

In 1986, systemic p.o. treatment with high doses of glycine (approximately 370 mg/kg) was shown to decrease PCP-induced hyperactivity in mice (Toth and Lajtha, 1986), indicating antipsychotic potential. Also, D-serine administered i.c.v. attenuated stereotypic behavior produced by the NMDA channel blockers PCP and (+)MK-801 in rats (Contreras, 1990). Similarly, in this species D-serine and D-alanine injected i.c.v. inhibited stereotypy, hyperactivity, and ataxia induced by PCP (Tanii et al., 1994) but, surprisingly, the former was less potent. D-Cycloserine (3 mg/kg) was also effective in counteracting hypermotility after (+)MK-801 administration (Dall'Olio and Gandolfi, 1993). A systemically active glycine uptake inhibitor, glycyldodecylamide, was found to be very potent in reversing PCP-induced hyperactivity in rodents (Javitt and Frusciante, 1997). The fact that glycine_B site activation attenuates the behavioral effects of NMDA channel blockers is in contrast to in vitro experiments, where the opposite effect is seen as an expression of use dependence (Huettner and Bean, 1988; Reynolds and Miller, 1988). One explanation might be that the glycine_B site is far from saturated in vivo and the apparent competitive reversal of MK-801 effects is the result of recruitment of more receptors by stimulation of the glycine_B site with D-cycloserine. Stimulation of the glycine_B site might also increase the efficacy of neuroleptic agents, e.g., D-cycloserine enhanced ()-sulpiride potency to attenuate apomorphine-induced stereotypy (Dall'Olio and Gandolfi, 1993).

As already mentioned, in humans, glycine has been shown by some authors to be beneficial in the alleviation of negative symptoms or the enhancement of the efficacy of neuroleptic agents (Waziri, 1988; Rosse et al., 1989; Javitt et al., 1994; Herescolevy et al., 1996; Leiderman et al., 1996). In contrast, D-cycloserine occasionally produced psychotic symptoms at high doses (250 to 500 mg) in patients with tuberculosis (Mandell and Petri, 1996) and aggravated positive symptoms in schizophrenic patients (Cascella et al., 1994). Others found that D-cycloserine did not improve symptoms in patients when used as an adjunct to clozapine therapy (Goff et al., 1996). In a study showing that D-cycloserine attenuated negative symptoms, this finding was possibly the result of arousal effects (Goff et al., 1995). In another study, D-cycloserine produced an improvement at low doses (50 mg) but caused worsening at higher doses (250 mg), which probably reflects actions at different NMDA receptor subtypes at which D-cycloserine has different intrinsic activities (Cascella et al., 1994; Goff et al., 1995) (table 5).

In recent years, data opposed to the "classical" hypothesis of glutamatergic deficits in schizophrenia have appeared. It was reported that L-701,324 (at 5 mg/kg) antagonized amphetamine-induced locomotor activation and increases in dopamine turnover in the nucleus accumbens and also attenuated the deficit in the prepulse inhibition of the acoustic startle response induced by isolation in rats (Bristow et al., 1995, 1996a). Interestingly, it did not inhibit apomorphine-induced stereotyped behavior (sniffing and licking/biting), which is believed to result from action in the striatum and is sensitive to the typical neuroleptic haloperidol (Bristow et al., 1996a). On the basis of these findings, L-701,324 was suggested to have features like those of the atypical neuroleptics, without extrapyramidal side effects. In our studies, although this agent did attenuate PCP and amphetamine locomotor stimulation in an apparently specific manner at 10 mg/kg (significant interaction factor in an analysis of variance), it failed to affect the PCP-induced deficit of prepulse inhibition, in contrast to atypical neuroleptics, in rats (Karcz-Kubicha et al., 1998b). Similar results were obtained with MRZ 2/576 at 10 mg/kg (Karcz-Kubicha et al., 1998b) and R(+)HA-966 (Furuya et al., 1998). In contrast, another glycine_B antagonist, MDL-103,371 [(E)-3-[2-(3-aminophenyl)-2-carboxyethanol]-4,6-dichloroindol-2-carboxylic acid], had no effect on prepulse inhibition or dopamine levels by itself but antagonized the (+)MK-801-induced increase in dopamine levels, implicating antipsychotic-like activity (Schmidt et al., 1996).

These clinical and preclinical data indicate that glycine and D-cycloserine might find some use in the alleviation of negative symptoms in schizophrenia. In the case of the latter agent, an important obstacle is the narrow range of effective doses. In our opinion, the concept of the use of glycine_B antagonists as "atypical neuroleptics" is interesting but is not substantiated by sufficient data from animal models using different types of compounds.

Of all epileptic patients (approximately 20 to 50 million people), 75% experience partial seizures; of these patients, approximately 85% suffer from complex partial seizures. Therefore, this type of epilepsy should be a major target for drug development (Rogawski and Porter, 1990; Löscher, 1993).

Although one of the first suggested therapeutic applications of NMDA receptor antagonists was in epilepsy (Czuczwar and Meldrum, 1982; Meldrum, 1985), only a few such agents reached clinical testing for this indication; these agents failed to show sufficient benefits and produced serious side effects (Troupin et al., 1986; Leppik et al., 1988; Sveinbjornsdottir et al., 1993). It should be pointed out that the efficacy of the agents under investigation has often been tested in animal models that are more relevant for global seizures (NMDA, MES, and sound models), whereas clinical testing usually involves patients with complex partial seizures, which are better modeled in animals by kindling experiments (Löscher and Schmidt, 1988). Using that model, Löscher and Hönack (1991) demonstrated the lack of evident anticonvulsive efficacy of NMDA receptor antagonists in the expression of symptoms, accompanied by an exaggeration of side effects indicative of psychotomimetic activity. This observation later found confirmation in a clinical study of D-3-(2-carboxypiperazine-4-yl)-1-propenyl-1-phosphonic acid in patients with complex partial seizures, which was terminated because of several CNS-related side effects (Sveinbjornsdottir et al., 1993). Of course, these failures should not discourage testing of glycine_B antagonists, which might show quite different profiles.

Are there any indications for the specific involvement of glycine, or the glycine_B site, in these seizures? In rats, during tonic seizures evoked by PTZ administration, there was a rapid increase (approximately 40%) in glycine levels measured in the cisterna magna (Halonen et al., 1992). This effect was not specific, however, because other amino acid levels were also increased. Tissue samples collected from epileptogenic loci of 35 epileptic patients showed marked increases in glycine levels (Perry and Hansen, 1981). In addition, increases in glycine concentrations were detected in biopsy brain samples (Hamberger et al., 1991). Analysis of surgically removed vascular malformations (cavernous angiomas causing neurodegeneration and epileptoform activity) revealed high levels of serine and glycine (Von Essen et al., 1996), which were related to overactivation of NMDA receptors. Similarly, using brain microdialysis in humans, it was reported that during seizures glycine levels in epileptic loci rose (eight-fold) more than did levels of glutamate (seven-fold) and serine (four-fold); the basal levels were 0.4, 1.2, and 2.0 µM, respectively (Ronneengstrom et al., 1992; see also Carlson et al., 1992). If nonsaturating glycine levels are assumed before seizure initiation (see discussion above), this could indicate a contribution of glycine to this phenomenon. However, considering the poor temporal resolution of the microdialysis method, it is not clear whether amino acid changes are a causal factor or a direct or indirect consequence of the seizures. Even if the latter is the case, the use of glycine_B antagonists might be useful in preventing the brain damage seen in epilepsy, because this damage is probably related to overactivation of NMDA receptors (Meldrum, 1991).

As expected for agents inhibiting NMDA receptor function, glycine_B antagonists (after i.c.v. administration) inhibit convulsions induced by sound, MES, PTZ, or NMDA (Danysz and Wroblewski, 1989; Chiamulera et al., 1990; Koek and Colpaert, 1990; Singh et al., 1991; Baron et al., 1992; Rowley et al., 1993; De Sarro et al., 1996b; Ilyin et al., 1996). However, because systemically active compounds are obviously more relevant as possible therapeutic agents, primarily these are discussed further. For example, an anticonvulsant effect of MNQX was one of the first reported CNS effects of selective and systemically active glycine_B site full antagonists (Sheardown et al., 1989). To our knowledge, no confirmation of this finding has ever been published (see Bisaga et al., 1993).

In general, the effects of glycine_B site partial agonists are quite unpredictable on the basis of their degree of NMDA receptor inhibition. For example, the partial agonists D-cycloserine and ACPC have high intrinsic activities but are quite effective anticonvulsants, whereas conflicting results have been obtained with R(+)HA-966, despite its lower intrinsic activity (Skolnick et al., 1989; Vartanian and Taylor, 1991; Peterson et al., 1992; Bisaga et al., 1993; Carter, 1994). It should be stated that there are other studies challenging the concept that there is a simple inverse relationship between the intrinsic activity of partial agonists and anticonvulsive effects (Peterson, 1991a,b, 1992; Peterson and Schwade, 1993). Using the MES model, Peterson (1991) reported that in rats glycine (4 g/kg, p.o.) and D-serine (2 g/kg, p.o.) enhanced the anticonvulsive activities of traditional anticonvulsive drugs, apparently via agonistic effects at the glycine_B site, because this action was antagonized by 7-Cl-KYN, which was ineffective when administered alone. Also, D-cycloserine attenuated MES-induced tonic convulsions with an ED₅₀ of 153 mg/kg, acting at the same site (Peterson and Schwade, 1993). Further complicating the situation, it has been shown that glycine potentiates convulsions produced by strychnine (Larson and Beitz, 1988).

Although the low intrinsic activity, partial agonist L-687,414 (fig. 10) did inhibit MES-induced convulsions in rats starting at 10 mg/kg, the dose-response curve was very flat (12-fold dose difference between 20 and 80% protection), and 100% protection was not obtained (Tortella and Hill, 1996). L-687,414 has also been shown to inhibit photically induced myoclonic seizures in primates (baboons) without producing any adverse effects (Smith and Meldrum, 1992).

The glycine_B site full antagonist MDL-104,653 (fig. 7) protected DBA/2 mice against sound-induced clonic seizures, with an ED₅₀ of 1.7 mg/kg (i.p., 45 min) (Chapman et al., 1995), whereas MDL-100,458 (fig. 4) exhibited an ED₅₀ of 20.8 mg/kg (Kehne et al., 1995). Also, with L-701,324 (fig. 7) a potent anticonvulsive effect was seen against NMDA-, PTZ-, MES-, and sound-induced convulsions in mice (ED₅₀ values of 0.96 to 3.4, i.v., and 1.9 to 3.4, p.o.) (Bristow et al., 1996b; see also Parsons et al., 1997). Similarly, several phthalazine-diones (MRZ 2/570, MRZ 2/571, and MRZ 2/576) (fig. 9) were quite effective inhibitors of MES-induced tonic convulsions in mice, in the range of 7 to 15 mg/kg (Parsons et al., 1997).

As stated above, the kindling model is probably much more relevant for screening of potential therapies, compared with simple audiogenic or chemically induced seizure models (Löscher, 1993). Croucher and Bradford (1990, 1991) were the first to describe a positive effect of glycine_B antagonists, using 7-Cl-KYN and R(+)HA-966 administered directly into the amygdala, against kindled seizures. This finding was not confirmed in another study after the i.c.v. route of administration, although a significant effect on seizure development was observed (Namba et al., 1993). Rundfeldt et al. (1994) tested the glycine_B antagonist 7-Cl-KYN and the partial agonists R(+)HA-966 and D-cycloserine, also using i.c.v. administration. All three drugs increased the focal seizure threshold, and none of them induced adverse behavioral effects or motor impairment (measured in the rotarod test) at anticonvulsant doses, indicating the superiority of glycine_B site antagonists and partial agonists, compared with competitive or uncompetitive NMDA receptor antagonists. In this model, i.c.v. injection of D-serine or i.p. administration of D-cycloserine (160 to 320 mg/kg), but not R(+)HA-966 (10 to 40 mg/kg), increased the afterdischarge threshold in amygdala-kindled rats (Löscher et al., 1994; see also Wlaz et al., 1994a).

It was recently reported that L-701,324 alone (up to 10 mg/kg) failed to alter the afterdischarge threshold or seizure severity in kindled rats, despite apparent ataxia in the rotarod test (Ebert et al., 1997). However, it is noteworthy that combination with ifenprodil resulted in anticonvulsive effects without further exaggeration of negative motor effects (Ebert et al., 1997). Similarly, MRZ 2/576 was not effective in this model, up to doses that impair motor behavior (10 mg/kg) (Wlaz and Löscher, 1998). In contrast, systemic administration of another glycine_B antagonist, MDL-104,653 (20 mg/kg), inhibited both the development and expression of kindled seizures in rats (Chapman et al., 1995).

Felbamate was shown to possess clear anticonvulsive activity that was weakened by coadministration of glycine or D-serine and was thus claimed to involve glycine_B site antagonistic effects (McCabe et al., 1993; White et al., 1995). However, felbamate inhibits [³H]5,7-diCl-KYN binding with an IC₅₀ of only 306 µM (Wamsley et al., 1994b), which is probably far above concentrations reached in the brain during therapy. It showed a promising therapeutic profile for the treatment of epilepsy and was introduced to the American market but was withdrawn shortly thereafter because of reported cases of aplastic anemia (McCabe et al., 1993; Burdette and Sackellares, 1994). Also, its mechanism of action is more complicated than initially anticipated; recent electrophysiological studies showed that felbamate is more likely to act as a low affinity NMDA receptor channel blocker than it is to act at the glycine_B site (Subramaniam et al., 1995), and the compound also has Na⁺ channel-blocking properties (Srinivasan et al., 1996).

In summary, at present there is no indication that glycine_B site full antagonists could be useful in monotherapy for the treatment of epilepsy. Still open as a possibility is the combination of these agents with traditional treatments (Czuczwar et al., 1996; Wlaz et al., 1996) or agents acting at other recognition sites of the NMDA receptor (Norris et al., 1992; Wlaz et al., 1994b; Ebert et al., 1997). Moreover, paradoxically, the best preclinical profiles in animal models recorded to date have been observed with high intrinsic activity, partial agonists such as D-cycloserine.