Introduction: glutamatergic transmission and schizophrenia

Though monoaminergic theories of the treatment of schizophrenia have dominated research and drug development for decades, there is increasing interest in non-monoaminergic strategies. Glutamatergic mechanisms are of special pertinence in view of the following: (1) their reciprocal interactions with monoaminergic networks (Schmidt and Kretschmer 1997; Marino and Conn 2002; Trudeau 2004); (2) their crucial role in the control of cognition, mood and motor function, which are disturbed in schizophrenia (Schmidt and Kretschmer 1997; Danysz and Parsons 1998; Paul and Skolnick 2003); (3) evidence that a dysfunction of glutamatergic transmission is implicated in psychotic states (Meador-Woodruff and Healy 2000; Breese et al. 2002; Pralong et al. 2002; Schiffer 2002; Konradi and Heckers 2003); and (4) the rich palette of ionotropic and metabotropic glutamatergic receptors available for therapeutic intervention.

The present article is principally devoted to NMDA receptors or, more precisely, their GlycineB co-agonist site—also referred to as the “NR1 subunit glycine binding site.” This site can be distinguished from GlycineA receptors which mediate a major mode of inhibitory transmission in the CNS. Detailed reviews of the significance of NMDA receptors to the etiology of schizophrenia have appeared in recent years (Marino and Conn 2002; Millan 2002; Heresco-Levy 2003; Van Berckel 2003). The present paper emphasizes recent findings supporting the notion that a functional deficit at NMDA receptors participates in the induction of psychotic states. It also focuses upon novel concepts for clinical compensation of this hypoactivity. For an understanding of such issues, it is indispensable to outline the complex nature of NMDA receptors and their regulation by endogenous ligands.

NMDA receptors: structure and modulatory sites

CNS-localized NMDA receptors are heteromers comprised of various assemblies of (probably two) NR1 subunits together with two or three NR2 subunits (Fig. 1) (Danysz and Parsons 1998; Dingledineet al. 1999; Yamakura and Shimoh 1999; Cull-Candy et al. 2001; Madden 2002; Millan 2002). At least four classes of NR2 subunit are known: A, B, C and D. The contribution of specific NR2 subunits to NMDA receptors is important in determining their functional profiles, desensitisation kinetics, modulation and both the affinity and efficacy of agonists at GlycineB sites (Vicini et al. 1998; Cull-Candy et al. 2001; Sheinin et al. 2001; Madden 2002; Liu et al. 2004). Certain NMDA receptors contain several types of NR2 subunits or various isoforms of NR1 subunit. Both NR1 and NR2A subunits are distributed throughout the CNS, being concentrated in the hippocampus, thalamus, frontal cortex and other structures implicated in psychotic states and their control (Danysz and Parsons 1998; Goebel and Poosch 1999; Cull-Candy et al. 2001; Millan 2002). These regions are also rich in NR2B subunits, whereas, despite their presence in forebrain structures, NR2C and NR2D subunits are preferentially found in the cerebellum and brainstem/spinal cord, respectively (Goebel and Poosch 1999; Yamakura and Shimoh 1999; Cull-Candy et al. 2001). Integration of developmentally regulated NR3 subunits into NMDA (NR1/NR2) receptors attenuates their activity, while construction of NR3/NR1 subunits only yield glycine-sensitive—but glutamate-refractory—receptors of low Ca2+ permeability (Chatterton et al. 2002). Though such sites may be relevant to abnormal processes in the developing schizophrenic brain (Deutsch et al. 2001; Millan 2002; Lipska 2004), it is not clear whether they exist in the adult. Further, NR3/NR1 heteromers are resistant to psychotomimetic open channel blockers (OCBs) (see below) questioning their relevance to the genesis of psychotic states in adults.

Fig. 1
figure 1

Schematic illustration of central NMDA receptors bearing co-agonist glutamate and GlycineB sites. OCB Open channel blocker, P phosphorylation site(s). Numerous modulatory sites on NR2 subunits responsive to, for example, haloperidol and polyamines, are not indicated for the sake of clarity

Full size image

NMDA and GlycineB recognition sites are located in homologous regions of NR2 and NR1 subunits, respectively (Fig. 1) (Yamakura and Shimoh 1999; Cull-Candy et al. 2001; Moretti et al. 2004). Despite their physical separation, glutamate and GlycineB binding sites functionally interact (Danysz and Parsons 1998). For example, glycine enhances the affinity and efficacy of glutamate, an action contributing to its ability to delay desensitisation and to increase the duration and frequency of the open state of the channel (Dingledine et al. 1999). Irrespective of subunit composition, all (NR1/NR2) species of NMDA receptor reveal voltage-dependent blockade by Mg2+ which binds to (multiple) sites in the ion channel, restricting the flux of Ca2+ (Dingledine et al. 1999; Cull-Candy et al. 2001). Neuronal depolarisation relieves the depolarisation block of NMDA receptor-coupled ion channels. A further common feature of NMDA receptors is a binding site recognized by OCBs, such as the pro-psychotic agents, ketamine and phencyclidine (PCP) (Dingledine et al. 1999).

NMDA receptors are primarily neuronal but may also be found on astrocytes (Nedergaard et al. 2002; Bezzi et al. 2004). They bear a variety of modulatory sites accessible to intracellular and extracellular mediators such as polyamines, protons, ifenprodil, zinc, glutathione, neurosteroids, ATP and even the antipsychotic haloperidol (Brimecomb et al. 1998; Dingledine et al. 1999; Jang et al. 2004; Kloda et al. 2004; Turecek et al. 2004). Despite their localization on NR2 subunits, modulatory sites can modify the functional status of GlycineB receptors on NR1 subunits (Yamakura and Shimoh 1999; Cull-Candy et al. 2001). Functional characteristics of NMDA and GlycineB binding sites can also be modified upon phosphorylation of NR1 or NR2 subunits by protein kinases (Dingledine et al. 1999; Yamakura and Shimoh 1999; Cull-Candy et al. 2001). NMDA receptors interact with diverse postsynaptic proteins incorporated into a “postsynaptic density” (Fig. 1). These proteins regulate the clustering of NMDA receptors in the plasma membrane, modify channel activity and influence their interaction with phosphorylating kinases (Husi et al. 2000; Madden 2002; Iwamoto et al. 2004).

NMDA receptors: multiple endogenous ligands—agonists and antagonists

Perhaps the most surprising feature of NMDA receptors is their responsiveness to a diversity of endogenous ligands, several of which behave as antagonists (Figs. 2, 3).

Fig. 2
figure 2

Generation and synaptic clearance of major endogenous agonists (glutamate, aspartate, glycine, d-serine) at NMDA and GlycineB recognition sites on NMDA receptors. GlyT Glycine transporter, EAAT excitatory amino acid transporter, asc alanine–serine–cysteine transporter, SNAT specific neutral amino acid transporter, vGluT vesicular glutamate transporter, CAC citric acid cycle, D-AAO d-amino acid oxidase, α-KG α-keto-glutarate. As discussed in the text, the precise role of multiple glial and neuronal transporters of d-serine remains unclear. Neuronal SNAT 1 also clears glycine. Mechanisms of glutamine efflux are unclear but likely involve reversal of several classes of transporter. Note that the elements illustrated are not necessarily all enriched and co-localized throughout the CNS. For example, the concentration of d-AAO is high in the cerebellum yet low in the forebrain, whereas the GlyT-2 transporter is principally found in the hindbrain and spinal cord

Full size image
Fig. 3
figure 3

Generation and synaptic clearance of various endogenous ligands at NMDA and GlycineB recognition sites on NMDA receptors. 5-HT Serotonin, NAAG N-acetyl-aspartate-glutamate, NAA N-acetyl-aspartate, GCP glutamate carboxypeptidase, KAT kynurenate amino transferase; K-3H kynurenate-3-hydroxylase. Quinolate possesses agonist properties at GlycineB and NMDA sites. At high concentrations, kynurenate antagonizes AMPA and kainate receptors. It is also a potent antagonist at α7 nicotinic receptors. Note alternative pathways for transformation of tryptophan into 5-HT/melatonin. Mechanisms for release and clearance of kynurenate are unknown. Several subclasses of inhibitory (Group II and III) metabotropic mGluR receptor are localized on glutamatergic terminals. At high concentrations, the κ-opioid agonist, dynorphin, interacts with NMDA receptors, for example, in hippocampus (Caudle and Dubner 1998; Wollemann and Benyhe 2004), but it is unclear whether its inhibitory effects are relevant to psychotic states. Further, in schizophrenic patients, conflicting data have been documented concerning levels of dynorphin (Heikkilä et al. 1990), and linkage studies have not clearly related the dynorphin gene to schizophrenia (Ventriglia et al. 2002)

Full size image

Glycine

Though some glycine may be available to NMDA receptors following “spillover” from glycinergic neurones (Ahmadi et al. 2003), the majority is derived from glial cells (Millan 2002; Miller 2004). Therein, glycine is generated from l-serine via the reversible enzyme, serine hydroxymethyltransferase. Non-released (and recaptured) glycine may be converted back into l-serine, or catabolised (in mitochondria) into inactive metabolites via the poorly characterised multi-enzyme “glycine cleavage system” (Sakata et al. 2001; Millan 2002; Ichinohe et al. 2004). Liberation of glycine from astrocytes is primarily non-vesicular and Ca2+-independent, presumably effected via reversal of glycine transporters upon changes in extracellular levels of glycine and alterations in ion flux (Gadea and Lopez-Colome 2001). Indeed, glial cells possess a high density of glycine-1 transporters (GlyT-1), of which the cerebral distribution tracks the localization of synapses bearing NMDA receptors (Gadea and Lopez-Colome 2001; Chen et al. 2003). These transporters (of which three isoforms have been identified) rapidly take up glycine in a Na+- and Cl−-dependent fashion and are principally responsible for clearing glycine from the synaptic cleft. In addition, some glycine may be taken up (and released) by small neutral amino acid transporters (SNAT), System A: SNAT 5 on astrocytes and SNAT 1 on neurones (Javitt 2002; MacKenzie and Erickson 2004; Cubelos et al. 2005). A little may also be removed by spatially remote GlyT-2 transporters on glycinergic neurones. Despite the original assumption that glycine fully occupies GlycineB sites, GlyT-1 and SNAT transporters are efficient in maintaining locally low and non-saturating levels of glycine (Danysz and Parsons 1998; Haradahira et al. 2003). Accordingly, GlycineB agonists and drugs blocking GlyT-1—glycine reuptake inhibitors (GRIs)—increase activity at NMDA receptors, providing a basis for potential antipsychotic properties.

d-Serine

l-Serine can be transformed into d-serine by d-serine racemase (Wolosker et al. 1999; Miller 2004; Xia et al. 2004). Though present in the liver and kidneys, d-serine racemase is enriched in the hippocampus, cortex and other cerebral structures which possess high levels of NMDA receptors (Schell et al. 1995; Wolosker et al. 1999; De Miranda et al. 2000). d-Serine is a high efficacy agonist at GlycineB sites and, dependent upon subunit composition, may be more efficacious than glycine itself (Danysz and Parsons 1998; Wolosker et al. 1999; Mothet et al. 2000). Like d-serine racemase, d-serine is predominantly found in astrocytes enveloping glutamatergic terminals in forebrain regions (Wolosker et al. 1999; Mothet et al. 2000; Miller 2004; Xia et al. 2004). In fact, levels of d-serine appear to be inversely correlated with those of the glial enzyme, d-amino acid oxidase (d-AAO), which cleaves (deaminates) d-serine into hydroxypyruvate (Wolosker et al. 1999; Urai et al. 2002; Miller 2004). The localization of d-AAO in glial cells implies that they participate in elimination of d-serine from the synaptic cleft. Though the identity of these glial d-serine transporters (which may be Na+-dependent) is unclear, they are certainly different from GlyT-1 transporters (Ribeiro et al. 2002). Neurones can also take up d-serine (and l-serine) via a Na+-independent, alanine–serine–cysteine (asc) transporter termed asc-1 found in pyramidal cells of the cortex and hippocampus: it is localized postsynaptically on soma and dendrites, as well as presynaptically on terminals (Nakauchi et al. 2000; Helboe et al. 2003; Matsuo et al. 2004). A further (alanine-insensitive) d-serine transporter has been reported in rat brain, though its nature remains unclear (Javitt et al. 2002; Helboe et al. 2003). Reversal of the glial d-serine transporter may lead to liberation of d-serine (Ribeiro et al. 2002; Miller 2004), but release is principally accomplished (in contrast to glycine) in a vesicular and Ca2+-dependent fashion (Mothet et al. 2000, in press; Cook et al. 2002; Bezzi et al. 2004; Parpura et al. 2004).

Glutamate

In neurones, glutamate is loaded by vesicular glutamate transporters (vGluT) into vesicles, then released onto postsynaptic NMDA receptors (Danysz and Parsons 1998; Madden 2002; Trudeau 2004). Several classes of Na+-dependent, excitatory amino acid transporter (EAAT) have been identified. They are localized on postsynaptic neurones, on presynaptic glutamatergic terminals and, most importantly, on glial cells (Danbolt 2001; Nedergaard et al. 2002). The principal circuit for regeneration of glutamate is, then, its recapture by astrocytes and subsequent conversion by glutamine synthase into glutamine, which is also generated from glutamate derived from the citric acid cycle (Marcaggi and Attwell 2004). Glutamine is liberated from astrocytes by multiple Na+-dependent and Na+-independent mechanisms, including reversal of the transporters “ASCT2” and a System N subtype of SNAT (Fig. 3) (Deitmer et al. 2003; Dolinska et al. 2004; Kanamori and Ross 2004). The latter differs to the “A” class of SNAT on neurones (SNAT 1) which takes up glutamine in a Na+-dependent fashion (Kanamori and Ross 2004; MacKenzie and Erickson 2004). Completing the cycle, glutamine is transformed by glutaminase into glutamate in neurones. An additional pool of glutamate is provided by glial cells, partly via reversal of EAAT1 (GLAST) and/or EAAT2 (GLT-1). Glutamate is also liberated from astrocytes via gap junction hemichannels (Ye et al. 2003). In addition, astrocytes possess vGluT1 and vGluT2, permitting its release by Ca2+-dependent and exocytotic mechanisms (Nedergaard et al. 2002; Bezzi et al. 2004; Montana et al. 2004; Parpura et al. 2004).

N-Acetyl-aspartate-glutamate

A final source of extracellular glutamate is provided by extracellular cleavage of N-acetyl-aspartate-glutamate (NAAG) (Neale et al. 2000; Barinka et al. 2004) via two forms of glutamate carboxypeptidase (GCP) II and III localized on plasma membranes of astrocytes (Berger et al. 1999; Speno et al. 1999; Bacich et al. 2002; Bzdega et al. 2004; Vieira and Devlin 2004). GCP II/III simultaneously generate a further agonist, aspartate, which is also derived from glutamatergic—as well as GABAergic—terminals (Gundersen et al. 2004). Extracellular aspartate is, in fact, generated from NAAG via N-acetyl-asparate (Fig. 5), a weak agonist at the NMDA recognition site and an agonist at excitatory metabotropic receptors (Rubin et al. 1995; Yan et al. 2003). Transformation of NAAG into glutamate, N-aspartyl-aspartate and asparate shifts the balance to postsynaptic excitation since NAAG has low intrinsic activity at NMDA receptors sites and behaves as an antagonist relative to glycine and d-serine, for example, in the CA1 region of the hippocampus (Grunze et al. 1996; Coyle 1997; Bergeron et al. 2005). Moreover, NAAG is an agonist at presynaptic metabotropic (mGluR)3 receptors inhibitory to glutamate release (Neale et al. 2000; Garrido Sanabria et al. 2004; Olszewski et al. 2004). Thus, a disequilibrium in conversion of NAAG may modify activity at NMDA receptors—and this occurs in psychotic states. Astrocytes do not only cleave NAAG: they also take it up via a proton-coupled di/tripeptide transporter termed “PEPT2” (Fujita et al. 2004; Terada and Inui 2004).

Kynurenate

Kynurenate behaves as an antagonist at the glutamate recognition site and, with greater potency, at GlycineB sites (Schwarcz and Pellicciari 2002). The ultimate source of kynurenate is tryptophan, and the glial pathway, which results in formation of kynurenate, is an alternative to neuronal generation of serotonin and melatonin (Stone and Darlington 2002). Interestingly, kynurenine also behaves as a weak antagonist at AMPA and kainate sites (Stone and Darlington 2002) and possesses antagonist properties at α7-nicotinic receptors (Alkondon et al. 2004). Despite these additional actions, blockade of GlycineB and NMDA sites by kynurenate likely contributes to its influence upon mood, monoaminergic transmission and motor function, including its perturbation of sensory auditory gating, an effect common to many pro-psychotic agents (Erhardt and Engberg 2002; Stone and Darlington 2002; Erhardt et al. 2004). Currently, little is known concerning glial release and recapture of kynurenate. Interestingly, l-kynurenine can be transformed into quinolinate, a weak agonist at GlycineB and NMDA receptors—which possesses neurotoxic properties (Schwarcz and Pellicciari 2002; Stone and Darlington 2002). Thus, alternative enzymatic conversion of l-kynurenine can alter the balance between endogenous agonists and antagonists at NMDA receptors.

Evidence that a deficit in transmission at NMDA sites is involved in schizophrenia

Pro-psychotic actions of open channel blockers

The OCBs, PCP and ketamine, trigger re-emergence of symptoms in remitted schizophrenia patients and elicit hallucinations and other psychotomimetic effects in normal subjects (Steinpreis 1996; Adler et al. 1999; Millan 2002; Coyle and Tsai 2004). Though their effects are not identical to deficits seen in schizophrenia, they more closely resemble psychotic disorders than those of monoaminergic psychostimulants such as amphetamine, notably as regards cognitive disruption and the induction of thought disorders and negative symptoms (Lahti et al. 2001; Abel et al. 2003; Van Berckel 2003; Morgan et al. 2004). Their actions are variably attenuated by antipsychotics such as haloperidol and clozapine (Malhotra et al. 1997; Lahti et al. 2001; Oranje et al. 2002, Van Berckel 2003). Neuronal mechanisms underlying the effects of OCBs are many, of which the following should be briefly evoked:

  1. 1)

    Sensitisation of subcortical and cortical dopaminergic pathways (Kegeles et al. 2000, 2002; Balla et al. 2003; Laruelle et al. 2003)

  2. 2)

    Activation of mesolimbic serotonergic pathways and consequent recruitment of corticolimbic populations of 5-HT2A receptors, effected independents of NMDA receptors (Millan et al. 1999; Aghajanian and Marek 2000)

  3. 3)

    Disruption of striatothalamic filtering of sensory input into the cortex (Carlsson et al. 2001)

  4. 4)

    A generalized perturbation of hippocampal function (Tamminga et al. 2003)

  5. 5)

    Disinhibition of cortico-cortical glutamatergic loops (Moghaddam and Jackson 2003)

  6. 6)

    A generalized disruption of cortical activity via desynchronization and reduced efficiency of neural transmission (Jackson et al. 2004)

  7. 7)

    Excessive cholinergic transmission in cortex (Farber 2003)

  8. 8)

    Excitotoxic damage elicited via non-NMDA receptors (Deutsch et al. 2001; Lewis and Levitt 2002; Farber 2003)

One common mechanism underlying these changes may be interruption of a NMDA receptor-mediated, tonic excitation of GABAergic interneurones inhibitory to projection neurones (Carlsson et al. 2001; Farber 2003; Schiffer et al. 2003; Shi and Zhang 2003; de Lima et al. 2004).

Underpinning a role of NMDA receptors in the psychotomimetic effects of PCP and ketamine in man and rodents, they are at least partially mimicked by antagonists at the NMDA recognition site (Lowe et al. 1994; Muir and Lees 1995; Bakshi et al. 1999; Dyker et al. 1999). However, it would be unwise to automatically attribute the full complement of pro-psychotic properties of OCBs to interruption of transmission at NMDA receptors. Thus, selective blockade of NR2B or NR2A subunits alone may not elicit psychosis (Higgins et al. 2003; Spooren et al. 2004). Further, GlycineB antagonists do not mimic behavioural effects of OCBs in rodents and, at least at modest doses, do not appear to be psychotomimetic in man (Danysz and Parsons 1998; Lees et al. 2001; Beardsley et al. 2002; Millan 2002). It is also worth pointing out that memantine, an “atypical,” low affinity OCB with distinctively rapid kinetics, does not elicit psychotic symptoms in man at doses exerting clinically relevant benefit in Alzheimer’s disease (Reisberg et al. 2003). Finally, direct interactions of PCP and ketamine with sites other than NMDA receptors (including monoaminergic receptors and transporters, sigma binding sites and ion channels) contribute to their functional and, possibly, pro-psychotic properties (see point 2 above) (Steinpreis 1996; Millan et al. 1999; Kapur and Seeman 2002; Millan 2002; Yu et al. 2002; Van Berckel 2003).

Psychotic-like phenotype of mice possessing genetically modified NMDA receptors

In rodents, PCP, ketamine and other more selective OCBs elicit bizarre behaviours, hyperactivity and cognitive deficits, certain of which are attenuated by antipsychotic agents (Steinpreis 1996; Schmidt and Kretschmer 1997; Millan 2002). Notably, mice in which the functional status of NMDA receptors has been genetically modified show phenotypes bearing comparison to the effects of OCBs. First, “compound” heterozygotic mice bred from two lines possessing point mutations in the NR1 subunit showed a sustained (antipsychotic-resistant) hyperactivity. They also revealed a suppression of hippocampal long-term potentiation which could be rescued by d-serine (Ballard et al. 2002). Second, mice in which the NR1 subunit was substantially “knocked-down” (by 95%) showed motor hyperactivity and stereotypies which were specifically abrogated by clozapine (Mohn et al. 1999). Dopaminergic transmission was not accelerated, but these mice also manifest deficits in sexual and social behaviour, as well as compromised sensory filtering (prepulse inhibition) underpinning a psychosis-like phenotype (Mohn et al. 1999; Duncan et al. 2004b, Miyamoto et al. 2004). Third, homozygotic mice in which the NR2A subunit was deleted revealed learning deficits and a hyperlocomotion which was attenuated by antipsychotics. This increase in locomotor behaviour was possibly related to an overactivity of ascending dopaminergic and serotonergic pathways (Miyamoto et al. 2001). However, this interpretation must be made cautiously since elimination of activity at both NR2A and NR2B receptors is requisite for the induction of hyperlocomotion and a psychotic-like phenotype (Kadotani et al. 1996; Higgins et al. 2003; Spooren et al. 2004). Finally, though of less obvious pertinence to schizophrenia, mice overexpressing NR2B receptors revealed enhanced hippocampal long-term potentiation and improved cognitive performance (Tang et al. 1999).

Alterations in NMDA receptors in schizophrenia

Neurochemical studies of glutamatergic neurotransmission in psychotic patients support the hypothesis that the functional activity of NMDA receptors is perturbed. However, observations have proven surprisingly inconsistent (Table 1) (Millan 2002; Schiffer 2002; Van Berckel 2003; Coyle and Tsai 2004). Reasons underlying disparate data include the following: (1) the influence of treatment with antipsychotics or other drugs—very few studies have controlled for this; (2) disease status at the time of measurement; (3) contrasting findings for measures of mRNA encoding NMDA receptor subunits as compared to radioligand binding studies; (4) age; and (5) differences between cerebral structures. Decreases in NMDA receptor density may reflect reduced function. On the other hand, increases in NMDA receptor density have also been construed as “compensating” (upregulation) for reduced stimulation by endogenous ligands. Though there are data supporting the latter notion, one must be cautious in adopting such interpretations. Indeed, changes in levels of mRNA encoding NMDA receptor subunits are difficult to interpret in the absence of information on levels of the corresponding protein.

Table 1 Summary of major alterations in expression (mRNA levels) of NMDA receptor subunits in schizophrenia
Full size table

Notwithstanding the above caveats, schizophrenia is accompanied by a broad pattern of alterations in NMDA receptor-related subunits in the thalamus, a structure dysfunctional in schizophrenia (Clinton and Meador-Woodruff 2004a, b). Thus, robust decreases in levels of mRNA encoding NR1 subunits have been reported, together with reduced binding of the GlycineB radioligand, [3H]MDL105,519 (Ibrahim et al. 2000; Meador-Woodruff et al. 2003; though see Popken and Leggio 2002). The diminished expression of NR1 subunits may be specific to the exon 22 containing isoform, a region of the NR1 subunit which interacts with the postsynaptic density; correspondingly, expression of genes encoding several of these proteins was also modified in the thalamus of schizophrenic patients (Clinton et al. 2003; Meador-Woodruff et al. 2003; Clinton and Meador-Woodruff 2004a, b). Though NR2A and NR2D subunits were unaffected, NR2B and NR2C subunits were diminished, changes paralleled by a reduction in binding of the NR2B subunit radioligand, [3H]ifenprodil; in contrast, binding to NMDA recognition sites and to ion channels was unaffected. Consistent with enhanced glutamate clearance, expression of glial EAAT1/2 was elevated (Ibrahim et al. 2000, but see Clinton and Meador-Woodruff 2004a). Completing the picture of alterations in glutamatergic function, expression of vGluT2 and glutamatinase increased, though these changes suggest enhanced glutamate availability (Smith and Haroutunian 2001a, b; Meador-Woodruff et al. 2003).

The thalamus provides a major afferent pathway to the cortex. This projection appears to be overactive in schizophrenia reflecting compromised filtering of sensory information (Carlsson et al. 2001; Clinton and Meador-Woodruff 2004a, b). In fact, findings in the cortex are more variable than for the thalamus. Decreases in NR1 subunits were found in entorhinal and temporal cortex, yet inconsistent decreases, a lack of change or even increases were seen in subterritories of frontal cortex (Meador-Woodruff and Healy 2000; Millan 2002; Van Berckel 2003). Interestingly, in the study of Humphries et al. (1996), the reduced level of mRNA encoding NR1 subunits in temporal cortex was correlated with cognitive decline. Underpinning a relationship of alterations in NR1 subunit expression to cognitive status, reductions have likewise been seen in the frontal and occipital cortex of Alzheimer’s patients (Dracheva et al. 2001; Hynd et al. 2004). As compared to NR1 and other subunits, a relative increase of NR2A subunits was seen in frontal and occipital cortex (mRNA), of NR2B subunits in temporal cortex (binding) and of NR2D subunits (mRNA) in prefrontal cortex (Akbarian et al. 1996; Grimwood et al. 1999; Dracheva et al. 2001; Woo et al. 2004). Despite these changes, binding of radioligands to GlycineB sites was increased in several cortical areas; further, no consistent pattern of changes has been seen with radioligands at recognition sites for glutamate (Ishimaru et al. 1994; Grimwood et al. 1999; Millan 2002; Zavitsanou et al. 2002; Van Berckel 2003). Thus, while cortical NMDA receptors are affected in schizophrenia, there is no clear evidence for a reduction in their activity.

Nevertheless, in the hippocampus, which is dysfunctional in schizophrenia (Medoff et al. 2001), decreases in NR1 subunit mRNA were asymmetrically localized to the left half of the brain; this laterality resembles other functional deficits characterising schizophrenia (Gao et al. 2000; Law and Deakin 2001; Crow 2004). A relative increase in NR2B mRNA but no change in NR2A subunits was noted (Harrison et al. 2003). Paralleling decreases in NR1 subunits, a reduction in mRNA encoding vGluT1 was seen: this suggests a reduction in glutamate loading into synapses and in glutamate release (Harrison et al. 2003).

In the striatum, neither studies of NMDA receptor subunit expression nor of radioligand binding have revealed marked changes (Meador-Woodruff and Healy 2000; Millan 2002; van Berckel 2003). Cortical glutamatergic pathways provide a major input to the striatum. They also (together with the bed nucleus of the stria terminalis, subthalamic nucleus and lateral dorsal tegmentum) send glutamatergic afferents to the substantia nigra and ventrotegmental area (Meltzer et al. 1997; Stefensen et al. 1998; Georges and Aston-Jones 2002; Laruelle et al. 2003; Sesack et al. 2003). These glutamatergic inputs, which act via both NMDA and non-NMDA (AMPA) receptors (Mathé et al. 1998; Adell and Artigas 2004), target dopaminergic perikarya and GABAergic neurones (Fig. 4) (Carlsson et al. 2001; Chen et al. 2001; Sesack et al. 2003; Takahata and Moghaddam 2003). Thus, an interesting finding was an increase in levels of NR1 subunits in the substantia nigra in schizophrenia (Mueller et al. 2004), coinciding with the argument that excessive activity at NMDA receptors on subcortical dopaminergic cell bodies may contribute to their hypersensitivity/hyperactivity in schizophrenia. Hence, antagonist properties at these populations might be favourable to its management (Fig. 4) (Carlsson et al. 2001; Millan 2002; Moghaddam 2003).

Fig. 4
figure 4

Overview of the NMDA receptor hypoactivity hypothesis of schizophrenia. PCP Phencyclidine, VTA ventrotegmental area, SNPC substantia nigra pars compacta, NACC nucleus accumbens, GRI glycine reuptake inhibitor. The organization of pathways is simplified—structures other than the cortex provide glutamatergic input to the striatothalamic filter and to dopaminergic neurons. Reduced activity at NMDA receptors is provoked by genetic, developmental and/or environmental factors (mimicked by PCP). Their hypoactivity leads to a reduction in GABAergic inhibitory tone onto glutamatergic neurons in the thalamus and cortex. Hyperactivity of the latter is implicated in the induction of psychotic states, possibly reflecting an indirect influence (not shown) upon subcortical mesolimbic dopaminergic pathways. The sensitivity and activity of mesolimbic dopaminergic pathways may also be enhanced upon a reduction in activity at NMDA receptors leading to a relief of inhibitory GABAergic tone. Accordingly to this configuration, antipsychotic strategies other than D2/D3 receptor blockade may be effective, e.g., recruitment of GlycineB receptors, blockade of certain classes of non-NMDA receptor and a reduction in cortical glutamate release (by lamotrigine or metabotropic agonists)

Full size image

Thus, alterations in NR2 subunits in schizophrenic brains vary in an isoform and structure-dependent fashion and require further characterisation. However, observations in the thalamus, hippocampus and cortex coincide with the notion of reduced function at NR1 subunits bearing GlycineB sites. Despite the need for confirmatory data on levels of protein rather than mRNA, this contention is underpinned by studies of their endogenous ligands.

Alterations in endogenous ligands of NMDA receptors in schizophrenia

In certain investigations, evidence for reduced levels of glutamate (and aspartate) was reported in the cortex and hippocampus of schizophrenics, and, in one study, the magnitude of this decrease correlated with the intensity of positive symptoms (Fig. 5) (Tsai et al. 1995; Faustman et al. 1999; see Millan 2002; Théberge et al. 2003). Interestingly, synaptosomal liberation of glutamate was reduced in schizophrenic brain (Sherman et al. 1991). An additional mechanistic basis for a reduction in extracellular glutamate levels was recently provided by an elegant study of Matute et al. (2005). They showed that levels of both the mRNA and the protein for glial EAAT1 are elevated in the frontal cortex of schizophrenics and also that these transporters display a marked (fourfold) increase in functional activity. This change appears to be specific inasmuch as levels of glial EAAT2 are reduced (Ohnuma et al. 2000). In the hippocampus, lower levels of glutamate may, on the other hand, be attributable to reduced numbers of glutamatergic terminals or to a decreased density of glial vGluT1 (Harrison et al. 2003). However, several studies have not found evidence for reduced glutamate levels in schizophrenia (see Millan 2002; Van der Heijden et al. 2004). Further, studies of the glutamate–glutamine cycle have revealed a complex pattern of changes not invariably consistent with reduced generation of glutamate. Thus, glutamine levels have been found (by the same group) to be either increased or decreased in cingulate cortex (Théberge et al. 2002, 2003), whereas both glutamine and glutamate were elevated in prodromic adolescents (Tibbo et al. 2004). Levels of glutamine synthase were reduced (consistent with elevated glutamate levels). On the other hand, higher levels of glutamate dehydrogenase suggest more rapid cleavage of glutamate to α-keto-glutarate, whereas elevated levels of glutamate acid decarboxylase indicate more rapid conversion to GABA (Gluck et al. 2002; Burbaeva et al. 2003). From these findings, it cannot be asserted with confidence that there is a generalized reduction in glutamate availability to cortical NMDA receptors in schizophrenia. Moreover, in the thalamus, increases in levels of glutamine, glutaminase, vGluT2 and EAAT1/2 might reflect greater rather than lesser availability of glutamate (Théberge et al. 2002, 2003; Meador-Woodruff et al. 2003). In any case, there is an interpretational challenge inasmuch as the relationship between changes in glutamate levels and specific populations of glutamatergic receptor—NMDA or other—remains unknown. Accordingly, changes in levels of other endogenous ligands may afford more direct information concerning the functional status of NMDA receptors.

Fig. 5
figure 5

Alterations in levels of endogenous ligands for NMDA receptors in schizophrenia. NAAG N-Acetyl-aspartate-glutamate, NAA N-acetyl-aspartate, GCP glutamate carboxypeptidase, KYN kynurenate, Gln glutamine, FCX frontal cortex, Hipp hippocampus. ↓ = Decrease, ↑ = increase and – = no significant change. For details of glutamate–glutamine cycling, see text

Full size image

As pointed out above, NAAG is cleaved by an astrocyte-localized GCP II/III into glutamate and aspartate. Tsai et al. (1995) reported higher levels of NAAG in frontal cortex and hippocampus, together with a decrease in the enzymatic activity of GCP II, that is, a shift in equilibrium from higher efficacy ligands (glutamate/aspartate) to a low efficacy ligand (NAAG) at NMDA receptors. However, contrary to expectations, levels of mRNA for GCP II were increased in the CA3 region of the hippocampus—though preliminary data indicate a reduction of mRNA for GLP II in frontal cortex (Hakak et al. 2001; Ghose et al. 2004).

Though it has long been accepted that glycine–serine metabolism is perturbed in schizophrenia, data showing reduced availability of glycine and d-serine to central NMDA receptors have proven difficult to obtain (Kumashiro et al. 1995; Millan 2002; Hashimoto et al. 2003; Sumiyoshi et al. 2004). However, Hashimoto et al. (2003) recently found that concentrations of d-serine in plasma are markedly reduced in psychotic patients. Levels of l-serine and of total serine were actually higher, an observation corroborated by Sumiyoshi et al. (2004). These (and other) authors also reported a decrease in circulating levels of glycine which was correlated with the severity of negative symptoms (Ermilov et al. 2004). These data likely reflect decreases in central availability of d-serine and glycine, but it would obviously be desirable to reproduce such findings at the cerebral level. In this light, it is of note that elevations in levels of the endogenous antagonist, kynurenate, were observed in the cortex and cerebrospinal fluid (Erhardt et al. 2001; Schwarcz et al. 2001).

Finally, the positive modulator of NMDA receptors, glutathione, interacts with an allosteric site—probably the same one as zinc—and decreases in its levels in schizophrenia were interpreted as contributing to reduced activity at these sites (Do et al. 2000).

To summarize (Fig. 5), these findings are globally in line with reduced stimulation of NMDA receptors in schizophrenia. However, data for glutamate are ambivalent. Further, the extent to which changes in plasma and cerebrospinal fluid levels of ligands for GlycineB and NMDA recognition sites reflect changes at the synaptic level remains unclear since local concentrations are tightly controlled by neuronal and astrocytic mechanisms of uptake and degradation (Figs. 2, 3).

Susceptibility genes for proteins which interact with NMDA receptors

Despite the high heritability of schizophrenia, the existence of multiple susceptibility genes of modest effect has hindered their identification (Collier 2003; Fukumaki and Shibata 2003; Harrison and Owen 2003; Elkin et al. 2004). Indeed, with the possible exception of NR2B subunits (Ohtsuki et al. 2001; Di Maria et al. 2004), no associations between genes expressing NMDA receptor subunits and schizophrenia have been found—despite positive reports for mGluR receptors and AMPA subunits (Schiffer 2002; Williams et al. 2002; Fukumaki and Shibata 2003). Nevertheless, polymorphisms in the promoter regions of NR1, NR2A and NR2B subunits may be associated with reduced NMDA receptor function and increased risk of schizophrenia (Miyatake et al. 2002; Itokawa et al. 2003; Lipsky and Goldman 2003). Further, several studies have pinpointed polymorphisms in genes which interact with NMDA receptors (Schiffer 2002; Harrison and Owen 2003).

Thus, linkage has been found between chromosome 8p and schizophrenia, and several markers in the neuroregulin (NRL1) gene located in this region comprise a haplotype associated with increased risk for schizophrenia (Stefansson et al. 2002, 2004; Williams et al. 2003; Yang et al. 2003). Further, “subtle” changes in expression patterns of three NRL1 isoforms were seen in schizophrenic brains (Hashimoto et al. 2004; Law et al. 2004). NRL1 is localized to vesicles in neuronal terminals containing glutamate. Following release, the NRL1 “ectodomain” binds to the receptor ErbB4 which co-localizes with NMDA receptors within the common postsynaptic density (Garcia et al. 2000; Dracheva et al. 2001; Moghaddam 2003; Stefansson et al. 2004). Accordingly, NRL1 enhances expression of NMDA receptors and increases their activity by promoting tyrosine kinase-mediated phosphorylation (Buonnano and Fischbach 2001). Support for functional interrelationships between NRL1 and NMDA receptors relevant to schizophrenia was acquired in mice with mutant NRL1 genes. These animals displayed a reduced density of NMDA receptors and a motor hyperactivity reduced by clozapine (Gerlai et al. 2000; Stefansson et al. 2002, 2004). Further, mice lacking the gene for ErbB4 displayed cognitive deficits (Golub et al. 2004). In another region of the 8p chromosome close to the NRL1 gene, a further polymorphism associated with schizophrenia was found by Gerber et al. (2003). The gene encodes the catalytic unit of calcineurin which controls both activity at NMDA receptors and synaptic plasticity (Krupp et al. 2002; Hedou and Mansuy 2003). By analogy to NRL1, mice lacking calcineurin display behavioural abnormalities resembling schizophrenia (Miyakawa et al. 2003). Dysbindin may also be located in the postsynaptic density of NMDA receptors, and linkage to schizophrenia was reported in several studies (Vaillend et al. 1999; Schiffer 2002; Straub et al. 2002; Schwab et al. 2003). Interestingly, its levels are reduced in the cortex of schizophrenics (Weickert et al. 2004).

An association of the gene encoding d-AAO with schizophrenia was demonstrated by Chumakov et al. (2002), together with a polymorphism for a primate-specific gene, G72, which interacts with, and possible activates, d-AAO. These findings were recently corroborated by Korostishevsky et al. (2004). Collectively, the data are consistent with an overactivity of d-AAO in schizophrenia and, correspondingly, lower levels of d-serine.

Finally, the gene encoding GCP II was detected near a translocation breakpoint region related to increased risk for schizophrenia (Semple et al. 2001).

The above findings have been enthusiastically embraced by many commentators. However, it remains to be shown that such polymorphisms (alone or collectively) are associated with changes in the functional status of NMDA receptors likely to precipitate psychotic states.

Antipsychotic properties of drugs increasing activity at NMDA receptors

Agonists, partial agonists, GRIs: interpretational challenges

Numerous, chemically diverse antagonists at GlycineB sites have been synthetized, including the kynurenate analogue, 5,7-dichlorokynurenic acid, and the selective agent, L701,324 (Fig. 6). In contrast, reflecting limited scope for modification of the structure of glycine, it has proven difficult to design novel agonists at GlycineB sites (Bräuner-Osborne et al. 2000; Millan 2002). Despite extensive metabolism both peripherally and centrally, and its poor penetration of the blood–brain barrier, systemic administration of glycine increases brain levels of glycine in rodents and man (D’Souza et al. 2000; Javitt et al. 2004a). d-Serine is also highly metabolised but shows superior penetration into the CNS, allowing for the use of lower doses (Hashimoto and Chiba 2004). A further advantage vs glycine is low affinity for GlycineA receptors, though nephrotoxicity limits its utility in rats. As regards synthetic ligands, all selective GlycineB agonists possess lower efficacy than glycine and d-serine at GlycineB sites, for example, the cyclic agents, S18841 and d-cycloserine (DCS) (Fig. 6) (Danysz and Parsons 1998; Cordi et al. 1999; Millan 2002).

Fig. 6
figure 6

Chemical structures of synthetic ligands at GlycineB receptors and glycine transporters. NFPS N-[3-(4′-Fluorophenyl)-3-(4′-phenylphenoxy)propyl]sarcosine, ORG 24598 R-(−)-N-methyl-N-[3-[(4-trifluoromethyl)phenoxy]-3-phenyl-propyl]glycine, SSR 504,734 (2-chloro-N-[(S)-phenyl[(2S)]-piperidin-2-yl]methyl)-3-trifluoromethyl benzamide. NFPS is also known as ALX-5407

Full size image

Knowledge of the precise degree of “resting” GlycineB receptor stimulation by endogenous ligands is critical for interpretation of the effects of exogenous ligands. Indeed, DCS and other partial agonists can, in principle, either activate or block GlycineB sites. Unfortunately, few studies have attempted to resolve this issue in showing that their actions are either blocked (if due to agonist properties) by selective antagonists or mimicked (if due to antagonist properties) (Millan 2002). Further compounding interpretation of the effects of partial agonists, their actions are less pronounced at high vs low doses (Danysz and Parsons 1998). It is difficult to attribute biphasic dose–response curves to partial agonist properties since agonist effects should, on the contrary, be apparent at high low doses. Alternative explanations include the following: (1) high potency agonist actions at NMDA receptor subtypes differing to subtypes blocked at higher concentrations; (2) high dose interactions with “allosteric” sites or postsynaptic proteins negatively coupled to NMDA receptors; and (3) induction of NMDA receptor internalisation at high concentrations (Nong et al. 2003). Finally, DCS may exert actions independently of NMDA receptors (Rouaud and Billard 2003).

The amino acid and GRI, sarcosine, have been clinically evaluated in schizophrenia (see below; Tsai et al. 2004a, b), and several novel GRIs have been described, including ORG 24598 (Harsing et al. 2003), NFPS (Kinney et al. 2003) and SSR 504,374 (Depoortere et al. 2004) (Fig. 6). The major difficulty in interpreting their actions is a lack of knowledge concerning the significance of glycine vs d-serine at GlycineB sites implicated in psychotic states.

Actions of GlycineB receptor ligands and GRIs in models of antipsychotic properties

Glycine displays little effect in classical models of antipsychotic activity, such as blockade of the actions of hallucinogens and of psychostimulants (Table 2) (Javitt et al. 1997; Javitt 2002; Millan 2002). The awkward question arises of whether this poor activity suggests a lack of clinical antipsychotic properties. Alternatively, such models may be inappropriate to studies of GlycineB receptor ligands lacking, in contrast to conventional antipsychotics, antagonist properties at dopaminergic and serotonergic receptors. The latter position is underpinned by positive effects of glycine in two other models of schizophrenia. The first is neonatal lesions of the ventral hippocampus in rats. This developmental model is characterised by “psychosis-like” behaviours in adults and reduced release of glutamate in the hippocampus and frontal cortex (Schroeder et al. 1999; Lipska 2004). The second is chronic treatment with PCP which enhances the responsiveness of cortical and subcortical dopaminergic pathways to amphetamine. This phenomenon is seen both in rodents and in man and resembles psychotic states (Breier et al. 1997; Kegeles et al. 2000; Balla et al. 2001a, b; Javitt et al. 2004a). In the former model, glycine blocked increases in locomotor behaviour elicited by novelty and amphetamine (Kato et al. 2001). Further, it normalized the disruption of sensory motor gating—decreased prepulse inhibition—displayed by these animals (Le Pen et al. 2003). In the model of chronic PCP administration, long-term administration of glycine attenuated the enhanced ability of amphetamine to provoke central release of dopamine (Javitt et al. 2004a). The effects of glycine in these procedures presumably reflect its ability to normalize sustained deficits in activity at NMDA receptors. However, observations that glycine reduces the locomotor hyperactivity and subcortical dopamine release evoked by acute administration of OCBs are less easy to explain (Toth and Lajtha 1986; Javitt et al. 1997, 1999, 2000; Millan et al. 1999; Millan 2002).

Table 2 Actions of GlycineB agonists and glycine reuptake inhibitors (GRIs) in experimental models of antipsychotic properties
Full size table

Central administration of d-serine likewise blocked acute motor actions of PCP, an action expressed stereospecifically inasmuch as l-serine was ineffective. Further, the action of d-serine was prevented by 7-chlorokynurenate (Tanii et al. 1994). d-Serine also prevented disruption of learning by PCP (Campbell et al. 1999). When administered during the vulnerable postnatal period, PCP profoundly disrupts synaptogenesis leading to cognitive deficits in adult rats: chronic treatment with d-serine reversed the compromised spatial working memory shown by rats exposed to PCP (Andersen and Pouzet 2004). This neurodevelopmental model resembles neonatal hippocampal lesions (vide supra) and may relate to the disruption of declarative memory seen in psychotic patients (Perry et al. 2000).

Though DCS does not itself enhance prepulse inhibition in adult rats, it reversed its disruption by microinjection of GlycineB antagonists into the nucleus accumbens (Kretschmer and Koch 1998; Geyer et al. 2001). This suggests, in line with a broad pattern of procognitive properties in rodents (Andersen et al. 2002; Jones et al. 2004; Stouffer et al. 2004), that partial agonists might improve deficits in cognitive-attentional function shown by psychotic patients. Unfortunately, actions of partial agonists have not as yet been described in protocols of chronic PCP administration and neonatal hippocampal lesions. By analogy to glycine and d-serine, DCS reduces the induction of hyperlocomotion and limbic release of dopamine by PCP. However, it remains to be proven that such effects reflect agonist actions at GlycineB sites since they were mimicked by GlycineB antagonists (Millan 2002) which actually show antipsychotic actions in certain models (see Millan et al. 2000). Though at first sight paradoxical, these findings are compatible with a model depicted in Fig. 4 which permits both a direct excitatory and an indirect (GABA-mediated) inhibitory influence of NMDA receptors on dopaminergic perikarya (Carlsson et al. 2001; Moghaddam 2003). By analogy to glycine, DCS exerts little influence upon the actions of amphetamine and hallucinogens (Przegalinski et al. 1999; Javitt 2002; see Millan 2002). Finally, of direct relevance to adjunctive use in man, co-administration of DCS with other partial agonists enhances actions of haloperidol against both amphetamine and PCP in rodents (Millan et al. 2000, unpublished observations).

There is a striking concordance in the actions of GRIs as compared to those of glycine in several experimental models (Table 2). Thus, NFPS mimicked glycine in preventing the sensitisation of amphetamine-induced dopamine release elicited by chronic administration of PCP (Javitt et al. 2004a). In a separate study, a further GRI, ORG 24598, mimicked glycine in preventing sensory motor (gating) deficits in rats sustaining neonatal lesions of the hippocampus (Le Pen 2003). Further, several GRIs abrogated acute PCP-induced hyperlocomotion with potencies correlating to their affinities at GlyT-1 sites (Javitt and Frusciante 1997; Javitt et al. 1999; Harsing et al. 2003). GRIs also reversed PCP-elicited changes in EEG power spectra in conscious rats (Harsing et al. 2003) and elicited cerebral patterns of c-fos expression similar to those seen with clozapine (Kinney et al. 2003). By contrast, GRIs show little activity in psychostimulant (amphetamine) models of antipsychotic activity, mimicking the weak activity of glycine. Indicative of improved cognitive function, GRIs enhanced hippocampal long-term potentiation and enhanced prepulse inhibition (Kinney et al. 2003). Finally, in line with these findings, in heterozygous mice with genetically inactivated GlyT-1 transporters, mnesic performance was improved and sensory motor gating was less perturbed by exposure to amphetamine (Tsai et al. 2004b).

Glycine and DCS were recently shown to suppress the vacuous oral movements provoked by long-term administration of haloperidol to rats (Shoham et al. 2004). This effect may reflect agonist actions at NMDA receptor sites incorporating NR2A subunits (Blanchet et al. 1999). Though surprising (Schmidt and Kretschmer 1997; Kretschmer 1998; Andreassen et al. 2003), this finding is paralleled by clinical findings outlined below.

To summarize, GlycineB agonists and GRIs present a coherent pattern of data overall consistent with antipsychotic properties in rodent models. However, such actions are generally seen alone. By contrast, as discussed below, their effects alone in schizophrenia have not been evaluated, rather their facilitatory influence upon the actions of antipsychotic drugs. In distinction to agonists, data indicating antipsychotic effects of the partial agonist, DCS, in rodents are not compelling, and it is unclear whether its potentiation of the actions of conventional antipsychotics reflects an increase in activity at NMDA sites.

Actions of GlycineB receptor ligands and GRIs in schizophrenic patients

Doses of glycine of up to 0.8 g/kg/day (generally 40–60 g) safely achieve increases in cerebrospinal fluid levels of glycine, and about a dozen studies of nearly 200 patients have described the effects of glycine as an adjunct to antipsychotic treatment (Heresco-Levy 2000, 2003; Millan 2002; Heresco-Levy and Javitt 2004; Tuominen et al. 2005). Several trials (usually over 6 weeks) were placebo-controlled and double blind, and glycine was shown not to influence serum levels of antipsychotics (Leiderman et al. 1996; Javitt et al. 2001). As summarized in Table 3 and discussed in the above citations, addition of glycine to haloperidol and other conventional neuroleptics achieves a dose-dependent decrease in primary negative symptoms and more modest but significant improvements of cognitive and positive symptoms. Notably, these effects were seen in otherwise treatment-resistant patients, and the effectiveness of glycine was inversely proportional to pretreatment levels of glycine in serum. Moreover, there was no exacerbation of the extrapyramidal side effects of antipsychotics, rather a tendency for improvement (Rosse et al. 1989; Heresco-Levy et al. 1999). Importantly, similar findings were documented for the newer antipsychotics, olanzapine and risperidone: adjunctive glycine (0.8 mg/kg/day) improved negative and, less markedly, cognitive and positive symptoms in treatment-refractory subjects, while also ameliorating tardive dyskinesia (Heresco-Levy et al. 2004). Intriguingly, glycine is not generally effective in patients receiving clozapine (Potkin et al. 1999; Evins et al. 2000; Millan 2002; Heresco-Levy 2003; though see Heresco-Levy and Javitt 2004). One simple explanation would be that clozapine is uniquely effective against negative symptoms (“ceiling effect”). However, a more likely explanation is that clozapine itself enhances activity at NMDA receptor sites (see below).

Table 3 Antipsychotic actions of drugs activating GlycineB sites in association with haloperidol and other antipsychotic agents
Full size table

Underpinning the above findings, in a 6-week study, association of d-serine (30 mg/kg/day, ca. 2 g in total) with conventional neuroleptics improved negative and, less markedly, positive and cognitive symptoms, without worsening side effects (Tsai et al. 1998). Likewise, by analogy to glycine, d-serine did not enhance the efficacy of clozapine (Tsai et al. 1999).

Mimicking its less robust effects (alone) than glycine in experimental models, the partial agonist, DCS, shows less marked improvements when given in association with antipsychotic agents to patients. Dose–response studies in subjects receiving conventional antipsychotics have shown that low doses (ca. 30 mg/day) are ineffective, high doses aggravate positive symptoms and only an intermediate dose of ca. 50 mg/day provides a reduction in negative symptoms in otherwise treatment-resistant patients without a significant improvement in positive or cognitive symptoms (Goff et al. 1999; Van Berckel 2003; Javitt 2002; see Millan 2002; Heresco-Levy 2003). As with glycine treatment, low levels of serum glycine were predictive of a good response to DCS (Heresco-Levy et al. 1998). Comparable findings of a modest improvement in negative symptoms were obtained with patients on risperidone or olanzapine (Evins et al. 2002; Heresco-Levy et al. 2002). DCS actually worsened negative symptoms when given in association with clozapine (Goff et al. 1996; Goff et al. 1999; Javitt 2002). Recently, in a retrospective analysis of a 5-year period of parallel investigations, Heresco-Levy and Javitt (2004) underpinned this impression of less robust effects of DCS as compared to those of glycine. In a meta-analysis, Tuominen et al. (2005) came to a similar conclusion, and Duncan et al. (2004a) recently obtained negative results with adjunctive DCS in patients showing mainly negative symptoms. The lack of a clear improvement in cognitive symptoms with DCS is, perhaps, surprising in view of the procognitive actions of partial agonists in rodents. On the other hand, it may be relevant that the influence of DCS upon mnemonic function in patients with dementia was not sufficiently robust to justify pursuing clinical trials (Johannesen and Myhrer 2002; Jones et al. 2004).

Despite essentially anecdotal reports of pro and/or antipsychotic effects of very high (tuberculostatic) doses of DCS (up to 3,000 mg/day) in non-psychotic subjects (see Millan 2002; Heresco-Levy 2003), no controlled trials of its effects alone in schizophrenia have been undertaken. Similarly, effects of glycine and d-serine alone remain to be elucidated. Results of such studies would be important for several reasons: (1) according to the NMDA receptor hypoactivity hypothesis, they should be effective alone; (2) most studies of antipsychotic actions in rodents have been undertaken with GlycineB agonists alone; and (3) should GlycineB agonists display clinically relevant antipsychotic effect alone, this would vastly improve perspectives for the development of novel therapeutic agents of this class (see below).

By analogy to GlycineB receptor ligands, potential antipsychotic actions of GRIs alone remain to be examined. However, sarcosine exerted clinical effects remarkably similar to those of glycine and d-serine in a 6-week double-blind study of patients under treatment with conventional antipsychotics including, principally, risperidone (Tsai et al. 2004a). That is, sarcosine, which was well tolerated, achieved improvement in negative and, less markedly, positive and cognitive symptoms. Though sarcosine dehydrogenase can demethylate sarcosine to glycine, it is unlikely to be acting as a “prodrug” since active doses of sarcosine (2 g/day) are far lower than those of glycine (40–80 g) needed for efficacy.

Influence of antipsychotics upon NMDA receptors: possible role in their actions

Administration (generally chronic) of antipsychotics elicits a complex, drug, time, receptor-subunit and tissue-dependent influence upon glutamatergic pathways and NMDA receptors (Bardgett et al. 1993; Fitzgerald et al. 1995; Ossowska et al. 1999; Millan 2002; Gemperle et al. 2003; Heresco-Levy 2003; Tarazi et al. 2003). Interesting differences have emerged between clozapine and haloperidol. These agents preferentially enhance glutamate levels and activity at NMDA receptors in cortex vs striatum, respectively (Yamamoto et al. 1994; Arvanov et al. 1997; Hayashi et al. 1999; Rodriguez and Pickel 1999).

Clozapine-induced increases in NMDA receptor activity in cortex involve several mechanisms depicted in Fig. 7. Clozapine reduces uptake of glutamate in cortex by decreasing glial expression of EAAT1 and neuronal expression of EAAT3 (See and Lynch 1996; Chen and Yang 2002; Millan 2002; Melone et al. 2003; Schmidt et al. 2003). Extracellular levels of glutamate may also be raised by inhibition of GABAergic interneurones and, upon chronic administration, by a diminution in the activity of GCP II (Squires and Saederup 1998; Michel and Trudeau 2000; Flores and Coyle 2003). Interestingly, patients receiving clozapine—and olanzapine—show higher plasma levels of glutamate (Evins et al. 1997; Goff et al. 2002), and a recent imaging study suggested that clozapine increases occupation of thalamic NMDA receptors by glutamate (Bressan et al. 2003). Elevations in glutamate levels will directly recruit NMDA receptors and indirectly enhance their activity via activation of AMPA receptors, though clozapine does not affect AMPA sites per se (Gemperle et al. 2003). AMPA receptor recruitment of astrocytes will also induce release of d-serine. In parallel, clozapine may enhance glycine levels by inhibiting SNAT 1 sites for neuronal uptake of glycine (Javitt et al. 2004b; Schwieler et al. 2004) and GlyT-1 sites for glial reuptake of glycine (Williams et al. 2004). Direct agonist actions of clozapine at GlycineB sites were speculated to attenuate actions of kynurenate, but there is no direct evidence for this (Schwieler and Erhardt 2003; Schwieler et al. 2004). Independently of glutamate and glycine, clozapine enhances the functional activity of NMDA receptors via their phosphorylation by protein kinase A (possibly dopamine and D1 receptor mediated) (Leveque et al. 2000; Chen and Yang 2002; Tseng and O’Donnell 2004), protein kinase C and calmodulin II (Hayashi et al. 1999; Seamans et al. 2000; Jardemark et al. 2003; Ninan et al. 2003a; Gonzalez and Robinson 2004; Naudon et al. 2004). Finally, N-desmethylclozapine, a major metabolite of clozapine, is an agonist at M1 receptors which allosterically facilitate activity at NMDA receptors (Sur et al. 2003; Weiner et al. 2004). The preferential enhancement by clozapine of NMDA receptor-mediated transmission in frontal cortex is associated with enhanced long-term potentiation and may be related to its beneficial influence upon negative symptoms (Gemperle et al. 2003; Serretti et al. 2004). It may also explain the inability of GlycineB agonists to improve clinical effects of clozapine.

Fig. 7
figure 7

Overview of multiple mechanisms implicated in the facilitatory influence of clozapine at NMDA receptors in cortex. DA Dopamine, GCP glutamate carboxypeptidase, AC adenylyl cyclase, EAAT excitatory amino acid transporter, GlyT glycine transporter, Musc muscarinic receptor, PKA protein kinase A, PKC protein kinase C, CaMII calmodulin II. It is unclear how clozapine recruits PKC and CaMII. Its influence upon EAAT2 and GCPII is only seen upon long-term administration in vivo. The suggestion that clozapine may directly (and allosterically) engage GlycineB site remains speculative. Note that (1) many of these potential mechanisms would benefit from confirmation; (2) they have been documented under a variety of conditions using a diversity of techniques; and (3) mechanisms shown are expressed in the cortex or other specific brain regions but not necessarily throughout the CNS

Full size image

Haloperidol more markedly increases striatal levels of glutamate than clozapine (Bardgett et al. 1993; Yamamoto et al. 1994; see Millan 2002). Its actions are exerted by several mechanisms including blockade of D2 and, perhaps, D4 receptors inhibitory to glutamate release (Berger et al. 2001; Cepeda et al. 2001; Rivera et al. 2002; Centonze et al. 2004; Gan et al. 2004) and reduced striatal expression of EAAT2 (De Souza et al. 1999; Schneider et al. 1998; Schmidt et al. 2003). Haloperidol may also enhance NMDA receptor function by inducing striatal expression of NR1 subunits (Fitzgerald et al. 1995). In vitro, it enhances activity at NMDA receptors by increasing protein kinase A-mediated phosphorylation of NR1 subunits: this action may reflect both blockade of D2 receptors (inhibitory to protein kinase A) and increased DA release leading to activation of co-localized D1 sites (facilitatory to protein kinase A) (Leveque et al. 2000; Liu et al. 2004). Though haloperidol interacts with NR2B and GlyT-1 sites, its potency is probably too low to be of relevance in vivo (Ilyin et al. 1996; Brimecombe et al. 1998; Lee and Rajakumar 2003; Williams et al. 2004, Yanahashi et al. 2004). Sustained reinforcement of glutamatergic transmission in the striatum may be excitotoxic (Leveque et al. 2000; Millan 2002) and contribute to the long-term onset of tardive dyskinesia with haloperidol. Nevertheless, long-term administration of glycine attenuated extrapyramidal motor effect of haloperidol in man and in rodents countering this possibility (Heresco-Levy 2003; Shoham et al. 2004).

Certain studies suggest that the influence of antipsychotics upon NMDA receptors is even more complex and may also involve inhibitory effects (Levine et al. 2003; Ninan et al. 2003b). Nevertheless, the general pattern of data clearly supports the above-described facilitatory influence of clozapine in the cortex and of haloperidol in striatum. In future work, it will be important to further analyse the functional consequences of their actions in additional brain regions and at specific constellations of NMDA receptor subunits. It will also be interesting to characterise the influence of other, mechanistically novel, antipsychotics upon activity at NMDA receptors.

Open questions and future perspectives

Multiple targets for antipsychotic modulation of activity at NMDA receptors

The above discussion highlights many potential strategies for modulating the functional status of NMDA receptors. First, it may be possible to develop more effective direct agonists at GlycineB sites. In this regard, efforts should be made to target specific subpopulations of NMDA receptors implicated in the induction of psychotic states, for example, those on GABAergic interneurones inhibitory to cortical glutamatergic pathways and to dopaminergic cell bodies. This may be feasible if such populations incorporate specific isoforms of NR1 subunits or specific assemblies of NR2 subunits modulating the ligand-binding profile of GlycineB sites on NR1 subunits. Supporting this possibility, several drugs have been developed which interact with discrete classes of NR2 subunit (Danysz and Parsons 1998; Dingledine et al. 1999; Yamakura and Shimoh 1999; Madden 2002, Feng et al. 2004). Second, alternative targets on NMDA receptors would be modulatory sites recognizing, for example, neurosteroids, polyamines or glutathione (Dingledine et al. 1999). Third, it may ultimately be possible to alter the function of NMDA sites via actions at NRL1, dysbindin or other postsynaptic proteins. Fourth, a novel possibility for indirect modulation of NMDA sites would be to target receptors which control the release of d-serine, glycine and glutamate from astrocytes (Bezzi and Volterra 2001). Fifth, drugs which affect reuptake, synthesis and/or degradation of d-serine, kynurenate and NAAG are an intriguing possibility. However, several questions remain. For example, inasmuch as d-serine and glycine are interconverted in glial cells, modification of the availability of one will inevitably affect the other. There would be little point in enhancing levels of d-serine if this indirectly results in a compensatory reduction in levels of glycine. Further, modification of the availability of kynurenate may indirectly influence serotonergic and cholinergic transmission with uncertain consequences for psychotic states (Stone and Darlington 2002; Alkondon et al. 2004).

Which endogenous ligands control activity at NMDA receptors?

A major question is whether GlycineB sites are tonically saturated. There is now a consensus that they are not—and functional actions of glycine, d-serine and GRIs in animals and man bear testimony to this (Danysz and Parsons 1998; Javitt 2002; Millan 2002; Haradahira et al. 2003). However, many therapeutically relevant uncertainties remain, notably, the degree of occupation of specific populations: in defined brain regions, in psychotic patients exposed or not to antipsychotics and in normal subjects exposed to stress or illicit drugs. Irrespective of the overall degree of occupation of GlycineB sites, a fundamental and related question is the participation of various endogenous ligands. For example, what is the relative importance of glycine as compared to d-serine in psychotic patients, and what is the contribution of these endogenous agonists as compared to the antagonist kynurenate? This issue is of far more than academic interest. Indeed, it underpins all current efforts to develop novel drugs for treating schizophrenia via GlycineB sites. For example, GRIs and/or d-AAO inhibitors will only display antipsychotic activity if glycine and/or d-serine, respectively, are genuinely ligands of NMDA receptors involved in the induction and control of psychotic states. Analogous arguments apply to agents which reduce the synthesis of kynurenate.

Significance of NMDA receptors in relation to AMPA and metabotropic receptors

The significance of glutamatergic mechanisms to schizophrenia and its treatment is not limited to NMDA receptors and extends to their AMPA, kainate and metabotropic counterparts. The interrelationship between these sites is beyond the scope of the present review. However, two intriguing aspects of contrasting implications should be briefly mentioned. First, AMPA and kainate receptors rapidly enhance activity at NMDA receptors by neuronal depolarisation, which relieves their Mg2+ block. In addition, AMPA receptors enhance release of d-serine from astrocytes, thereby indirectly enhancing activity of NMDA sites in a slower and more sustained fashion (Schell et al. 1995; Nedergaard et al. 2002). Whether such actions are relevant to the proposed use of AMPAkines for the improvement of cognitive dysfunction in schizophrenia would be of interest to determine (Johnson et al. 1999; Goff et al. 2001; Marenco et al. 2002). Second, the paradox of why PCP and other OCBs act psychotomimetically yet enhance glutamate release may be explicable by glutamatergic loops separated by intervening GABAergic interneurones (see above) (Fig. 4). PCP may then mimic schizophrenia by blocking NMDA input onto GABAergic neurones. This leads to an increase in downstream glutamate release onto AMPA and other classes of glutamatergic receptor mediating pro-psychotic effects. One implication of this hypothesis is that drugs reducing glutamate release, such as lamotrigine (Anand et al. 2000; Hosak and Libiger 2002; Tiihonen et al. 2003) or presynaptic metabotropic receptor agonists (Moghaddam 2002; Schoepp and Marek 2002; Winter et al. 2003), may be useful antipsychotic agents—though they might exacerbate hypoactivity at certain populations of NMDA sites. Alternatively, antagonists at AMPA receptors may be of interest as antipsychotic agents (Mathé et al. 1998; Johnson et al. 1999; Sebban et al. 2002; Takahata and Moghaddam 2003), though this notion is diametrically opposed to the above-mentioned use of AMPAkines as cognitive enhancers.

Clinical efficacy of NMDA receptor modulators alone

There is currently no evidence that enhancing activity at GlycineB sites is itself sufficient for antipsychotic activity. Results of clinical trials addressing this issue will be critical since a lack of therapeutic efficacy alone implies the need for adjunctive utilization of GlycineB agonists, GRIs and other classes of agent. This is hard to envisage on a broad scale and would complicate development of novel agents. One alternative strategy would be to combine within a single molecule (direct or indirect) modulatory activity at GlycineB sites and D2/D3 dopamine receptor antagonism. Such a “multitarget” approach would fit well with the multifactorial origins of schizophrenia. For both selective and multitarget agents at GlycineB sites, it would be desirable to focus on the influence upon cognitive symptoms in view of their importance to the overall outcome of treatment and the major role of NMDA receptors in mnemonic processes (Kane et al. 2003).

Confirmation of the role of GlycineB sites in the actions of GlycineB agonists and GRIs

The convergent effects of glycine, d-serine and GRIs in experimental models of antipsychotic activity and in psychotic patients support the notion of a common mode of action: enhanced activity at GlycineB sites on NMDA receptors. Nevertheless, no formal proof is available from clinical investigations, and there have been few rigorous tests of this assumption in rodent studies. In principle, their actions should be prevented by selective GlycineB receptor antagonists. Further, assuming actions at common sites, in the presence of GlycineB agonists at doses sufficient to saturate GlycineB sites, GRIs should exert no further effect. Reciprocally, assuming that glycine is a critical endogenous ligand, in the presence of GRIs, direct GlycineB agonists should exert no further actions. Such studies remain to be performed. This is important since it is difficult to exclude additional central actions of these agents. For example, a dose of glycine sufficient to evoke substantial elevations in its levels in the brain is likely to influence cerebral concentrations of interrelated modulators—including d-serine. Thus, in future work, it will be necessary to more rigorously underpin the GlycineB hypothesis of antipsychotic activity with appropriate pharmacological controls.

Conclusions

In conclusion, there is a compelling body of experimental and clinical data implicating NMDA receptors in the pathogenesis and, potentially, treatment of schizophrenia. Further, several mechanisms are available for countering the hypoactivity of NMDA receptors which is thought to participate in psychotic states. However, two fundamental questions remain. First, which is the most appropriate therapeutic strategy (for example, direct agonists, modulators of glycine reuptake and/or modulators of d-serine availability)? Second, will drugs, which selectively modulate activity at GlycineB sites, be therapeutically effective alone? If so, this would transform the landscape of drug discovery in schizophrenia. On the other hand, they may only be useful as adjunctive agents. Rather than selective drugs, multitarget drugs interacting with GlycineB receptors as well as D2/D3 receptors (or other complementary sites) may represent novel and improved antipsychotics. This remains to be seen. In any case, in view of evidence that AMPA, kainate and metabotropic receptors are also involved in the etiology of schizophrenia, one may be reasonably optimistic as regards the future utility of glutamatergic strategies for the control of this devastating disorder.