Because glutamate is ubiquitous in the brain as the primary excitatory neurotransmitter, a model positing generalized abnormalities of glutamatergic activity would be unlikely to account for the clinical characteristics of schizophrenia with any degree of specificity. However, specificity has been observed with pharmacological challenges, where antagonism of the glutamatergic N-methyl-d-aspartic acid (NMDA) receptor complex has produced behavioral and cognitive deficits in normal subjects that closely mimic schizophrenia (1), and in therapeutic trials, in which agents that enhance NMDA receptor activity have selectively improved symptoms in schizophrenia patients (2). In addition, postmortem studies have identified abnormalities of glutamate receptor density and subunit composition in the prefrontal cortex, thalamus, and temporal lobe (3–5), areas that exhibit impaired activation during performance of cognitive tasks in schizophrenia (6, 7). These findings suggest that glutamatergic dysregulation may occur in regionally specific subpopulations of glutamatergic receptors and so support the potential value of a glutamatergic model for guiding research into the pathophysiology and treatment of schizophrenia.

Although the relationship is speculative, glutamatergic receptor dysfunction could also play a role in neuroarchitectural abnormalities that have been described in schizophrenia, such as aberrant neuronal migration (8, 9) or reduced synaptic connections (10), because of the role of glutamatergic receptors in regulating neuronal migration, neurite outgrowth, synaptogenesis, and the "pruning" of supernumerary neurons by apoptosis (11–14). Neuronal excitotoxicity mediated by glutamatergic receptors has also been proposed as a consequence of dysregulated glutamatergic transmission in schizophrenia (15), but evidence for neurodegeneration from glutamate toxicity in the brain in schizophrenia remains poorly established. Because an extensive and functionally diverse range of glutamate receptor subtypes are genetically encoded and can interact with environmental stressors during brain development, the model of glutamatergic dysfunction may account for the interplay of genetic and environmental risk factors identified in schizophrenia. Furthermore, a proposed dysfunction of glutamatergic neuronal systems is not inconsistent with the dopamine hypothesis of schizophrenia, since reciprocal synaptic relationships between forebrain dopaminergic projections and glutamatergic systems have been well described (16), and dysregulation of one system by illness or pharmacological interventions would be expected to alter neurotransmission in the other.

This review will briefly summarize glutamate receptor physiology and evaluate the evidence for glutamatergic dysfunction in schizophrenia, focusing on postmortem findings, pharmacologic models, and clinical trials examining the effects of glutamatergic agents on the symptomatic manifestations of schizophrenia.

Glutamate and the structurally related acidic amino acid aspartate activate two families of receptors: ionotropic receptors, which gate cation channels, and metabotropic receptors, which are coupled to G-proteins that affect intracellular metabolic processes (17). The ionotropic receptors are designated by the potent glutamate analogues that selectively activate them: the kainate receptor, the α-amino-3-hydroxy-5-methyl-isoxazole-4-proprionic acid (AMPA) receptor, and the NMDA receptor. Electrophysiologic effects of all three ionotropic receptor families are mediated by the opening of cation channels permeable to Na⁺ and, in a subtype-specific fashion, to Ca²⁺, thereby depolarizing or "exciting" the neuron. AMPA and kainate receptors play the primary role in mediating fast excitatory postsynaptic potentials responsible for excitatory neurotransmission.

The NMDA receptor serves a different role. At resting membrane potential, its channel is blocked by Mg²⁺. Upon depolarization caused by activation of the kainate and/or AMPA receptors, the Mg²⁺ block is removed, permitting glutamate to open the NMDA channel. The channel is permeable not only to Na⁺ but also to Ca²⁺, an important intracellular signaling ion that activates nitric oxide synthetase, among other enzymes (18). The NMDA receptor has a number of modulatory sites that affect its activity. Within the channel, there is a binding site for the dissociative anesthetics such as phencyclidine (PCP, "angel dust") and ketamine, which serve as noncompetitive antagonists (19). There is also a strychnine-insensitive binding site for the co-agonist glycine, which must be occupied in order for glutamate to open the ion channel (20). This site on the NMDA receptor is distinct from the strychnine-sensitive site associated with the inhibitory glycine receptor in the brainstem and spinal cord. Electrophysiologic studies indicate that the glycine modulatory site is not fully occupied under normal conditions (21–23).

Because the NMDA receptor is recruited only during periods of substantial neuronal depolarization, it appears to serve the purpose of a "coincidence" detector. In this way, the NMDA receptor plays a critical role in a major form of use-dependent synaptic plasticity known as long-term potentiation. In long-term potentiation, a brief period of high-intensity excitatory synaptic activity, which markedly depolarizes the neurons and recruits NMDA receptors, results in a subsequent persistent increase in synaptic efficacy. Long-term potentiation has been linked to memory formation (24).

Molecular cloning has revealed a family of eight genes encoding the metabotropic receptors (19). Four genes encode the peptides (GluR1–GluR4) that form the AMPA receptor. The molecular diversity of AMPA receptors is further enhanced by several splice variants, in which mRNA is constructed from different exons (coding regions of genes), resulting in a number of physiologically distinct receptor channels from the same gene, and from posttranscriptional processing of mRNA. Three genes encode the family of kainate receptors (GluR5–GluR7), whereas two additional genes encode polypeptides (KA1 and KA2) that alter the pharmacologic features of the kainate receptors. This rich receptor diversity in theory can account for significant differences in relative activity and potential toxicity of glutamate receptors and might permit selective pharmacologic manipulations of excitatory neurotransmission.

The glutamatergic ionotropic receptors are formed by the aggregation of four or five subunits, which transverse the cell membrane and form the receptor-ion channel complex. The subunits differ in important pharmacodynamic properties, including affinity for glutamate, threshold for channel opening, and permeability to Ca²⁺ ions. For example, the GluR₂ subunit reduces calcium permeability of the AMPA-gated channels; this effect is dependent on the posttranscriptional editing of the GluR₂ mRNA. AMPA receptors that lack the GluR₂ subunit produce channels permeable to Ca²⁺ influx, which promotes excitotoxic neuronal degeneration (25). Subunit composition may also differ according to the functional role of the receptor. For example, hippocampal inhibitory neurons preferentially express NMDA receptor subunits 2C and 2D (NR_2C and NR_2D), which are more sensitive to activation by glutamate, owing to a lower threshold for Mg²⁺ blockade, and are more sensitive to antagonists (25, 26). Subunit composition may also be modified by exposure to drugs such as alcohol, nicotine, and antipsychotics (27, 28).

Kim et al. (29) reported reduced concentrations of glutamate in the CSF of patients with schizophrenia and first proposed that decreased glutamatergic activity may be an etiologic factor in the disorder. This finding was replicated by some (30, 31) but not all subsequent studies (32–34). Tsai et al. (35) examined eight regions in postmortem brains and found lower concentrations of glutamate and aspartate in the prefrontal cortex and a lower concentration of glutamate in the hippocampus of patients with schizophrenia than in comparison subjects.In addition, the concentration of N-acetyl-aspartyl glutamate (NAAG), an acidic dipeptide that acts as an antagonist at NMDA receptors (36), was increased in the hippocampus, and the activity of glutamate carboxypeptidase II (GCP II), the enzyme that cleaves NAAG to produce glutamate and N-acetyl aspartate (NAA), was selectively reduced in the frontal cortex, temporal cortex, and hippocampus in the schizophrenia brains. It is noteworthy that magnetic resonance spectroscopic studies in schizophrenic subjects have demonstrated significant reduction of NAA levels in these same regions (37). However, the factors that regulate brain NAA levels are complex (38). While these findings suggest diminished activity at glutamatergic receptors in relevant brain regions, it remains uncertain whether this represents a primary vulnerability factor or a compensatory response to a more proximal defect.

Ligand binding studies in postmortem brains from individuals with schizophrenia have revealed consistent increases in kainate receptors in the prefrontal cortex (39, 40) and decreased AMPA and kainate receptor binding in the hippocampus (41, 42) without consistent abnormalities in NMDA receptor density (3, 4). Immunocytochemical analyses have confirmed a decrease in AMPA receptors in the medial temporal lobe, although reductions in the hippocampus were not found in one study (3). Whereas ligands binding to the cation channel of the NMDA receptor complex ("PCP binding site") have not demonstrated consistent alterations in density (3), Ishimaru et al. (43, 44) reported increased binding to the glycine site of the NMDA receptor throughout the primary sensory cortex and related association fields. In addition, the binding of [³H]-d-aspartate, which labels the transporters that remove synaptic glutamate, was increased in the frontal cortex and decreased in the striatum (39, 45).

The cloning of receptor subunits has facilitated the measurement of glutamate receptor expression in the brain. Most consistent has been the finding of a lower level of mRNA encoding AMPA receptor subunits in the hippocampus and parahippocampus of schizophrenia brains than in the brains of comparison subjects (3, 46, 47). In addition, Akbarian et al. (48) found a higher proportion (1% versus 0.1%) of unedited GluR₂ mRNA in the prefrontal cortex of patients with schizophrenia and Alzheimer’s disease than in normal subjects, suggesting a higher level of permeability of AMPA receptors to calcium, which could increase the potential for neurotoxicity. Although less studied, the measurement of mRNA encoding kainate receptor subunits has demonstrated a similar pattern of lower density in the hippocampus and parahippocampus (3). Although ligand binding studies have generally failed to find altered NMDA receptor density in the brain in schizophrenia, two studies have found evidence of altered subunit composition of NMDA receptors. Akbarian et al. (48) found a relatively higher level of the NR_2D subunit in the prefrontal cortex of schizophrenia subjects than in normal subjects and neuroleptic-treated comparison patients, suggesting increased potential responsiveness to glutamate. In contrast, Gao and colleagues (4) recently reported a relative decrease of the NR₁ subunit in the hippocampus of schizophrenia patients. The investigators demonstrated that this finding was unlikely to represent a medication effect since treatment with haloperidol for 6 months produced no effect on NMDA receptor subunit composition in the rat hippocampus (4). NMDA receptors lacking an NR₁ subunit are nonfunctional; the relative lack of the NR₁ subunit in the brain in schizophrenia suggests less than normal pharmacodynamic responsiveness of NMDA receptors in the hippocampus. Mohn and colleagues (49) used recombinant DNA technology to develop transgenic mice expressing only 5% of the normal levels of NR₁; the NMDA receptor-deficient mice exhibited hyperactivity, stereotypies, and social isolation. The hyperactivity and stereotypies were ameliorated by treatment with haloperidol and clozapine, but only clozapine corrected the impairments in social behaviors. Finally, Ibrahim and colleagues (5) recently reported lower levels of mRNA expression for subunits composing NMDA, AMPA, and kainate receptors in the thalamus of schizophrenia patients, and lower levels of binding to the polyamine and glycine binding sites of thalamic NMDA receptors. Differences between schizophrenic and comparison subjects were most prominent in nuclei with reciprocal projections to limbic regions.

It has long been recognized that PCP produces a syndrome in normal individuals that closely resembles schizophrenia (50, 51) and exacerbates symptoms in patients with chronic schizophrenia (50, 52). At subanesthetic doses, PCP binds to a site within the ion channel of the NMDA receptor that blocks the influx of cations, thereby acting as a noncompetitive antagonist (19). After reports linking PCP to protracted psychosis, abuse, and neurotoxicity (for review see reference 53), PCP was abandoned as an anesthetic agent in humans. Ketamine, another cyclohexylamine anesthetic that has approximately a 10-to-50-fold lower affinity for the NMDA receptor, continues to be used as an anesthetic in children. It is interesting to note that psychotic reactions associated with exposure to ketamine are reported to occur less frequently in children than in adults, suggesting the similar age dependence in vulnerability to psychoses associated with NMDA antagonists and onset of schizophrenia.

When infused intravenously to normal subjects, ketamine produces an amotivational state characterized by blunted affect, withdrawal, and psychomotor retardation (54). Psychotic symptoms typically take the form of suspiciousness, disorganization, and visual or auditory illusions. Psychotomimetic and perceptual effects of PCP are diminished under conditions of sensory deprivation, suggesting that processing of sensory information, rather than perception, is disrupted (55). Dissociative symptoms are also prominent. Although dissociative symptoms are not typically associated with schizophrenia, depersonalization may be an important early feature of the schizophrenia prodrome (53). Finally, ketamine produces the characteristic cognitive deficits of schizophrenia, including impaired performance on the Wisconsin Card Sorting Test and on verbal declarative memory, delayed word recall, and verbal fluency tests, without evidence of global impairment on the Mini-Mental State Examination (54, 56, 57).

When administered to patients with schizophrenia who were stabilized with conventional neuroleptics, ketamine produces delusions, hallucinations, and thought disorder, consistent with the patient’s typical pattern of psychotic relapse (58, 59). Cognitive functioning, particularly recall and recognition memory, are further impaired. It is noteworthy that treatment with clozapine but not with haloperidol attenuated ketamine’s exacerbation of clinical symptoms (58, 59).

Jentsch and Roth (19) argued that repeated administration of the NMDA receptor antagonists provides a more valid model of schizophrenia than acute administration. Whereas psychotic symptoms resulting from single-dose infusions of ketamine in normal subjects tend to be mild and somewhat inconsistent, prolonged exposure in PCP abusers is associated with severe, persistent psychotic symptoms more typical of schizophrenia (19, 51). However, it is debated whether the experience of chronic abusers is a valid model for PCP effects on the normal brain (60). Acute administration of ketamine in normal subjects increased perfusion in the prefrontal cortex and anterior cingulate (61–64) and decreased hippocampal perfusion (62), whereas chronic PCP abusers displayed classical "hypofrontality" (65, 66). Compared with single-dose administration, chronic treatment with PCP produced in monkeys more perseveration and fewer nonspecific cognitive deficits and caused memory deficits that persisted after PCP was discontinued (67). These memory deficits were prevented by clozapine treatment.

In rodents, acute administration of NMDA receptor antagonists markedly increases the release of dopamine and glutamate in the prefrontal cortex and subcortical structures (68–70). Moghaddam et al. (71) demonstrated in rats that ketamine-induced augmentation of dopamine release in the prefrontal cortex was associated with impaired performance on a memory task sensitive to prefrontal cortical function; these alterations could be ameliorated by treatment with an AMPA/kainic acid receptor antagonist. Using single cell recordings from dopamine neurons of the ventral tegmental area in rats, Svensson et al. (72) demonstrated that NMDA antagonists increase the rate but decrease the variability of neuronal firing, thereby impairing the signal-to-noise ratio. Burst firing was increased in the ventral tegmental area dopamine neurons that projected to limbic regions but was decreased in dopamine neurons that projected to the prefrontal cortex, indicating regional specificity of effects. By using positron emission tomography to monitor the displacement of [¹¹C]raclopride binding in the striatum after acute administration of ketamine in normal volunteers, several groups have demonstrated increased dopamine release of a magnitude comparable to the effects of amphetamine; furthermore, raclopride displacement correlated with severity of psychotic symptoms (73–75). Microdialysis in monkeys further revealed that increased striatal dopamine release was accompanied by increased reuptake, resulting in increased turnover but unchanged extracellular dopamine concentrations (76).

Whereas the acute administration of NMDA receptor antagonists enhances dopamine turnover in the prefrontal cortex, subchronic administration is associated with decreased dopamine turnover in the frontal cortex (67, 77), reflecting potentially persistent, compensatory effects. Jentsch et al. (77) found a reduction of approximately 75% in prefrontal dopamine utilization, as reflected by the ratio of 3,4-dihydroxyphenylacetic acid (DOPAC) to dopamine in brain tissue after daily administration of 10 mg/kg of PCP in rats for 7 days. Jentsch et al. (78) also demonstrated a 40% reduction in extracellular dopamine by in vivo microdialysis in conscious rats after administration of 5 mg/kg of PCP twice daily for 7 days. In contrast, Lindefors et al. (79) reported that daily administration of 25 mg/kg of ketamine for 7 days increased prefrontal dopamine concentrations without altering concentrations of dopamine metabolites. The explanation for the conflicting results obtained with subchronic ketamine versus phencyclidine administration is not clear but may reflect the shorter half-life of ketamine (19). It is interesting to note that chronic administration of NMDA receptor antagonists results in decreased expression of the dopamine D₁ receptor mRNA in the prefrontal cortex of rats and monkeys (19, 80, 81). The D₁ receptor has been shown to be critical for working memory function (82).

Revisions of the dopamine hypothesis for schizophrenia have posited diminished dopaminergic activity in the prefrontal cortex and a reciprocal dopaminergic hyperactivity in the mesolimbic pathways (83). Consistent with this model, chronic PCP administration also increases subcortical dopamine release, particularly in the nucleus accumbens (68, 84). Increased mesolimbic dopaminergic activity associated with long-term administration of PCP produces sensitization to the behavioral effects of NMDA receptor antagonists such as PCP, ketamine, and MK801 (85–87), dopamine agonists (88, 89), and stress (89). Chronic administration of PCP also leads to increased mesolimbic dopamine response to haloperidol (89). Together, these findings emphasize the reciprocal modulation of glutamate and dopamine neuronal systems and are consistent with the "sensitization model" of schizophrenia (90), which may account for the progressive course of the illness and the vulnerability to stress of individuals with the illness.

A characteristic feature of schizophrenia is the inability to adapt to an auditory stimulus that is preceded by a low-level warning tone, a response that is known as prepulse inhibition and is believed to reflect a defect in attentional "filtering" of nonnovel stimuli (91). The atypical antipsychotic drugs clozapine, olanzapine, remoxipride, and quetiapine have all been found to reverse ketamine-induced deficits in prepulse inhibition (92–94). Haloperidol and selective antagonists at D₁, D₂ and serotonin₂ (5-HT₂) receptors did not correct the deficit in sensory gating caused by NMDA receptor antagonists (92, 95, 96), whereas the α₁-adrenergic antagonist prazosin blocked the disruptive effects of PCP (97). Chlorpromazine also blocks the effects of ketamine on prepulse inhibition, possibly by means of mediation by its potent antagonism of α₁-adrenergic receptors (96). Both clozapine and olanzapine attenuated the ketamine-induced increase in cortical metabolic activation measured by [¹⁴C]-2-deoxyglucose in rats, an effect that was not achieved with either haloperidol or risperidone and that required a higher dose of olanzapine (10 mg/kg) than would be expected to produce maximal D₂ and 5-HT₂ blockade (98, 99). Corbett et al. (100) reported that olanzapine (0.25 mg/kg) and clozapine (2.5 mg/kg) but not haloperidol or risperidone reversed PCP-induced social withdrawal in rats. In another model, Olney (15) performed a series of experiments comparing the efficacy of antipsychotic drugs in preventing neuronal degeneration induced by NMDA receptor antagonists in the posterior cingulate and retrosplenial cortex in rats. Olanzapine, clozapine, and fluperlapine strongly prevented the neurotoxicity, whereas haloperidol and thioridazine displayed intermediate effectiveness (101–103). Neuroprotective effects have also been observed with muscarinic receptor antagonists, benzodiazepines, σ receptor ligands, and α₂-adrenergic receptor agonists, thereby making the clinical implications of activity in this complex model unclear (104).

Antipsychotic drugs can affect glutamatergic neurotransmission by modulating release of glutamate, by interacting with glutamate receptors, or by altering the density or subunit composition of glutamate receptors. Recent research has demonstrated that antipsychotic drugs acting through the D₂ receptor promote the phosphorylation of the NR₁ subunit of the NMDA receptor, thereby enhancing its function and consequent gene expression (105). Thus, dopamine-glutamate interactions occur intraneuronally as well as intrasynaptically (106). Free glutamate concentrations in the striatum measured by in vivo dialysis were increased by as much as fivefold by chronic administration of haloperidol or fluphenazine but were unaffected by clozapine (107–109). The augmentation of glutamate release in the striatum by conventional antipsychotic drugs appears to be mediated by D₂ inhibitory axoaxonic synapses on glutamatergic corticostriatal terminals (109), although long-term haloperidol treatment has also been shown to decrease expression of the glial glutamate transporter GLT-1 in the rat striatum (110). The elevation of excitatory amino acid concentrations may have important clinical consequences, as indicated by significant correlations between ratings of tardive dyskinesia and CSF concentrations of aspartate and glutamate in neuroleptic-treated patients (111, 112). In addition, perforated synapses in the caudate, which have been associated with haloperidol-induced extrapyramidal side effects, have been shown to occur in glutamatergic synapses and to be mediated by NMDA receptors (113).

Growing evidence suggests that the effects of certain atypical antipsychotics on NMDA receptors may differentiate these agents from conventional antipsychotics. Lidsky et al. (114) measured haloperidol and clozapine displacement of [³H]MK801 binding in rat striatal and cortical membranes and found that haloperidol did not significantly interact with NMDA receptors at clinically relevant concentrations but that clozapine displaced the ligand from the NMDA receptor at therapeutic levels. Using intracellular recordings and a voltage clamp, Arvanov et al. (115) found that clozapine but not haloperidol produced an enhancement of NMDA-receptor-mediated neurotransmission. Both the selective 5-HT_2A antagonist M100907 and clozapine prevented PCP-induced blockade of NMDA receptors, as measured by depolarization of rat medial cortical pyramidal neurons (116), whereas selective D₂ blockers had no effect. Clozapine and several conventional agents have also been reported to act as partial agonists at the glycine modulatory site of the NMDA receptor, increasing neuronal depolarization at low concentrations and inhibiting depolarization at high concentrations (117, 118). Consistent with this interpretation, an increase in extracellular glycine attenuated the potentiation by haloperidol of NMDA-receptor-evoked response (119). In contrast, long-term administration of antipsychotics may result in desensitization of the glycine modulatory site of the NMDA receptor, as evidenced by a reduction in strychnine-insensitive glycine binding associated with both clozapine and conventional agents (120). These findings suggest that the glutamatergic effects of antipsychotics are importantly concentration-dependent and that, depending on their relative dose-response curves, different agents may act either as agonists or antagonists at therapeutic concentrations.

Several investigators have found that chronic antipsychotic treatment alters the expression of mRNA encoding glutamate receptor subunits, which varies depending on the drug type, the subunit, and the brain region (27, 121–123). In general, conventional antipsychotics increased the amount of mRNA encoding NMDA receptor subunits (NR₁ and NR₂) in the striatum, whereas clozapine treatment produced no change (122). This difference may reflect differential liability for extrapyramidal side effects. Conventional and atypical antipsychotics also differed in their effects on certain AMPA receptor subunits (GluR₂ and GluR₄), whereas GluR₁ was increased and GluR₃ decreased by both haloperidol and clozapine (123). Dissimilarities also were found for kainic acid receptors, with only clozapine reported to elevate expression of mRNA encoding GluR₆, GluR₇, and KA2 (123).

Evidence for hypoactivity of NMDA receptors in schizophrenia has led to therapeutic trials with agents that indirectly activate the receptor. Direct agonists at the NMDA receptor have not been studied because of the risk that excessive stimulation may cause excitotoxic damage to neurons (124). A more promising target is the glycine modulatory site on the NMDA receptor. Early trials of glycine administered orally at doses of 5–15 g/day produced inconsistent results, probably because glycine poorly crosses the blood-brain barrier (125–128). More recently, Javitt, Heresco-Levy, and colleagues (129–131) performed a series of placebo-controlled crossover trials with high doses of glycine (30–60 g/day) added to antipsychotic drugs and have demonstrated selective improvement in negative symptoms. In a 6-week trial, glycine also significantly improved subjects’ ratings on the cognitive subscale of the Positive and Negative Syndrome Scale (131). Javitt and colleagues (23) demonstrated that glycine inhibits PCP-induced stimulation of subcortical dopamine release in a dose-related fashion in rats. Glycine transport inhibitors were also found to block PCP-induced behavioral hyperactivity (23), and they may represent a potential therapeutic approach. In another therapeutic approach with a full agonist at the glycine modulatory site, Tsai et al. (132) added d-serine to ongoing antipsychotic medication at a daily dose of 30 mg/kg for 8 weeks and reported significant improvements, compared to effects of placebo, in negative symptoms, psychosis, and cognitive function as measured by the cognitive subscale of the Positive and Negative Syndrome Scale and performance on the Wisconsin Card Sorting Test.

In a related approach, several groups have administered d-cycloserine, an antitubercular drug that acts as a relatively selective partial agonist at the glycine modulatory site over a narrow range of concentrations (133). Compared to glycine, d-cycloserine produces approximately 60% activation of the NMDA receptor, thus acting as an agonist in the presence of low concentrations of glycine (and related endogenous agonists) and as an antagonist in the presence of high concentrations of glycine. In an initial placebo-controlled, partly blinded, dose-finding study of d-cycloserine added to conventional neuroleptics, Goff et al. (134) found an inverted U-shaped dose response with significant reductions in negative symptoms and improvement in performance on a test of working memory at a d-cycloserine dose of 50 mg/day. Van Berkel et al. (135) administered d-cycloserine in a small, open trial to medication-free schizophrenia patients and observed selective improvement of negative symptoms at a d-cycloserine dose of 100 mg/day. In an 8-week, fixed-dose, placebo-controlled, parallel-group trial involving 46 patients who met criteria for the deficit syndrome of schizophrenia (126), 50 mg/day of d-cycloserine significantly improved negative symptoms when added to conventional antipsychotics, but it did not improve performance on a cognitive battery (136). It is noteworthy that a full response was not achieved until weeks 4–6. Rosse et al. (137) found no improvement in negative symptoms when 15 mg/day or 30 mg/day of d-cycloserine was added to molindone.

Since clozapine (arguably) produces substantial therapeutic effects on negative symptoms in patients who respond poorly to typical neuroleptics and since its effects on glutamatergic systems differ from those of conventional agents, it was of interest to determine whether the addition of d-cycloserine would have further ameliorative effects in clozapine responders. Two separate trials of 50 mg/day of d-cycloserine added to clozapine resulted in worsening of negative symptoms (138, 139). In contrast, controlled trials in which the full agonists glycine and d-serine were added to clozapine produced no change in negative symptoms or cognitive function (140–142). One possible explanation for these findings is that clozapine may exert its effects on negative symptoms partly by increasing occupancy of the glycine modulatory site on the NMDA receptor, thereby transforming the partial agonist d-cycloserine into an antagonist and precluding additional therapeutic effects with the exogenous full agonists glycine and d-serine.

A final but quite preliminary area of investigation involves the study of drugs acting at the AMPA receptor that have recently become available for clinical trials. This family of drugs, known as ampakines, act as positive modulators of the AMPA receptor complex. CX516, the first drug of this class to be studied, has been shown to increase the peak and duration of glutamate-induced AMPA receptor-gated inward currents (143). In rats, ampakines increased hippocampal neuronal activity in response to stimulation of glutamatergic afferents and enhanced long-term potentiation (144, 145). These findings suggest that ampakines, by potentiating AMPA receptor-induced depolarization, indirectly enhance NMDA receptor function. In behavioral models to test learning in rats, ampakines improved acquisition and retention in the radial arm maze, water maze, and olfactory cue tasks (146). CX516 also synergistically blocked methamphetamine-induced rearing behavior in rats when it was added to clozapine and to conventional antipsychotic agents, an effect believed to predict antipsychotic efficacy (147).

CX516 was added to clozapine in a placebo-controlled, 4-week, escalating-dose trial involving six patients with schizophrenia and in a placebo-controlled, fixed-dose, parallel-group design with an additional 13 patients; the combination was well tolerated without significant adverse effects (148). Combined results from the two trials (N=19) revealed a consistent pattern of improvement in performance on tests of attention, memory, and distractibility. Comparisons between groups demonstrated moderate-to-large effect sizes favoring CX516 over placebo for most cognitive tests, but inferential tests of statistical significance were not performed due to the small number of subjects. Although the ampakines show promise as a treatment for cognitive deficits of schizophrenia, these preliminary data require replication in larger groups of patients.

Multiple lines of evidence have linked abnormalities in glutamatergic receptor expression, subunit composition, and function in schizophrenia. Similarities between behavioral effects of NMDA receptor antagonists and the clinical symptoms of schizophrenia have focused attention on treatment trials targeting a putative hypoactivity of a subpopulation of NMDA receptors. However, currently available antipsychotic drugs alter glutamatergic activity in multiple ways by enhancing release of glutamate in the striatum, directly interacting with NMDA receptors, altering glutamate receptor density, and changing the subunit composition of glutamate receptors. Many of these effects are regionally selective and vary among the antipsychotic drugs, with important differences emerging between atypical and conventional drugs. Clinical trials in which NMDA receptor activity was enhanced by agents acting at the glycine modulatory site have demonstrated decreases in negative symptoms and variable improvements in cognitive function. Electrophysiologic and neurochemical evidence suggests that clozapine, aside from its interactions with aminergic receptors, may also be acting through the NMDA receptor in affecting negative symptoms. Although the findings are preliminary, recent work with an ampakine indicates that positive modulation of the AMPA receptor may also provide another glutamatergic approach to treat cognitive deficits in schizophrenia. Thus, drugs that modulate glutamatergic neurotransmission hold promise for novel treatments for schizophrenia, especially for the cognitive impairments and negative symptoms associated with the disorder.

Received June 2, 2000; revisions received Nov. 29, 2000, and Feb. 27, 2001; accepted March 12, 2001. From the Schizophrenia Program, Massachusetts General Hospital, Boston; and the Consolidated Department of Psychiatry, Harvard Medical School. Address reprint requests to Dr. Coyle, Consolidated Department of Psychiatry, Harvard Medical School, 115 Mill St., Belmont, MA 02478; joseph_coyle@hms.harvard.edu (e-mail). Supported in part by NIMH grants MH-51290 and MH-606450 (to Dr. Coyle) and MH-54245 and MH-57708 (to Dr. Goff).