Introduction

The pathogenesis of schizophrenia (SZ) is complex due to the heterogeneity of its neural bases, which also makes it difficult to develop new treatments. In this context, the glutamate (Glu) hypothesis for SZ was proposed in the 1980s [1], and previous studies suggested that N-methyl-D-aspartate receptor (NMDAR) impairment may be associated with dysfunction of parvalbumin-containing γ-aminobutyric acid (GABA) interneurons, resulting in excessive glutamate release [2, 3]. In fact, for example, anti-NMDAR encephalitis is known to cause psychotic symptoms, and low-dose ketamine, an NMDAR antagonist, has been associated with SZ-like symptom manifestation [4]. Further, large genome-wide association studies on SZ noted that several genes associated with this disorder encode proteins involved in glutamatergic neurotransmission [5]. Preclinical research also has shown that disruption of excitatory/inhibitory balance in rodents has been found to lead to impairments in working memory, retention of strategy-switching behaviors, and social play behaviors, which are closely related to the pathogenesis of SZ [6]. Thus, simultaneous measurement of Glu and GABA levels in the brain could lead to a better understanding of the pathogenesis of SZ.

To date, several meta-analyses were conducted based on proton magnetic resonance spectroscopy (1H-MRS) studies that examined the regional levels of glutamatergic/GABAergic metabolites in patients with SZ. With regard to glutamatergic metabolites, Marsman et al. (28 studies: 647 patients with SZ and 608 healthy controls [HC]) was the first to conduct a meta-analysis on glutamatergic metabolite levels reporting that Glu levels in the medial frontal region were decreased in patients with SZ compared with HC [7]. Subsequently, Merritt et al. (59 studies: 1686 patients and 1451 HC) showed elevated levels of Glu in the basal ganglia and elevated levels of Glx (Glu+glutamine [Gln]) in the basal ganglia and medial temporal lobe [8]. In terms of GABA, Egerton et al. (16 studies: 526 patients and 538 HC) found no group differences in the medial prefrontal cortex, parietal/occipital cortex, and striatum between patients with SZ and HC [9]. On the other hand, to the best of our knowledge, six studies focused on 1) clinically specific subpopulations (i.e., high risk for psychosis [HR] [10], first-episode psychosis [FEP] [11], or treatment-resistant schizophrenia [TRS] [12]); 2) ultra-high magnetic field strength [13]; or 3) brain regions (i.e., DLPFC [14] or frontal cortex [15]).

However, Merritt et al. had several limitations. First, there were only a small number of studies in some areas, especially when clinical groups are analyzed separately, which made it impossible to perform the meta-analysis of those regions. Also, given that recent studies have focused on the anterior cingulate cortex (ACC) and midcingulate cortex (MCC), two regions involved in different functions [16], they need to be analyzed separately. Second, since several studies suggested the possibility of antipsychotic effect on metabolite levels [17,18,19,20], subgroup analysis for unmedicated patients is needed. Finally, the work by Merritt et al. did not stratify SZ into TRS or treatment-responsive SZ in terms of antipsychotic treatment responsiveness although it is hypothesized that TRS may have different pathophysiology compared with treatment-responsive SZ.

Therefore, we aimed to conduct an exhaustive meta-analysis using a large dataset including glutamatergic and GABAergic metabolite levels in various regions in patients with different stages and treatment response to the disease. The total number of publications has more than doubled since the aforementioned meta-analysis by Merritt et al., which included studies before 2016, thus more detailed regions and clinical stages can now be analyzed. The main objective of this study was to conduct a case-control meta-analysis of all published articles on glutamatergic and GABAergic measures in individuals with SZ-spectrum disorders. As exploratory analyses, we also sought to assess the impact of age, sex, illness stage, symptom severity, magnetic field strength, metabolite correction method, macromolecule nulling, the status of antipsychotic medication, and treatment response/resistance on these metabolite levels in individuals with SZ-spectrum disorders.

Patients and methods

Protocol registration

The full protocol was uploaded to the International Prospective Register of Systematic Reviews website (CRD42020220603).

Study search

The current meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement [21]. The search was performed with Embase, Medline, PsycINFO, and PubMed using the following search terms: (MRS OR magnetic resonance spectroscopy) AND (schizophrenia OR psychosis OR UHR OR at risk mental state OR ultra-high risk OR clinical high risk OR genetic high risk OR prodrom* OR schizoaffective). The searches were rerun just before the final analyses on 7 November 2020. A hand search was conducted by two authors (S.T. and T.N.). Candidate articles were independently screened by these authors. Discrepancies in study selection were resolved by discussion among the three authors, including the senior author (S.N.).

Inclusion and Exclusion criteria

Full-length, or short articles written in English were included, which met the following criteria: the studies that (1) included SZ, schizoaffective disorder, FEP, or HR group; (2) compared Glx, Glu, Gln, or GABA levels in the brains of the patient group and HC using 1H-MRS; (3) acquired Glu or GABA levels at field strengths of 3T or above and Gln levels at 4T or above [22]; and (4) included sufficient data to be able to obtain standardized mean differences (SMDs) between groups. The following articles were excluded: (1) studies reporting on a sample completely overlapping with the other studies, or (2) studies examining the metabolite levels with chemical shift imaging.

Risk of Bias

The risk of bias assessment tool for Nonrandomized Studies [23] was employed for the following factors: participant selection, confounding variables, measurement of exposure, blinding of outcome assessment, incomplete outcome data, and selective outcome reporting.

Primary outcomes

The co-primary outcomes were Glx, Glu, Gln, and GABA levels regardless of the method of metabolite correction. Based on previous meta-analyses [8, 9], the regions of interest (ROIs) were chosen as follows: (1) basal ganglia; (2) hippocampus; (3) ACC; (4) MCC; (5) medial frontal cortex (MFC); (6) posterior cingulate cortex (PCC); (7) dorsolateral prefrontal cortex (DLPFC); (8) thalamus; (9) frontal cortex; (10) frontal white matter (WM); (11) temporal cortex; (12) temporal WM; (13) occipital cortex; (14) occipital WM; (15) occipito-parietal cortex; (16) parietal cortex; (17) parietal WM; and (18) cerebellum. The classification method of ROI is described in Supplementary Material 1 and Supplementary Fig. 1. Data extraction of the variables was detailed in Supplementary Material 2.

Statistical analyses

All the continuous primary outcomes were compared between the patients and HC using SMDs. The formulation of the SMD was Hedge’s g, which adjusted for small sample bias. The magnitude of the SMD was interpreted as small for SMD = 0.2; medium for SMD = 0.5; and large for SMD = 0.8. The calculation of SMD and the meta-analysis with a random-effect model was conducted with the “metafor” package in R (version 4.0.2). We conducted the main meta-analyses between the whole group (HR + FEP + SZ) or SZ group and HC for each metabolite in each ROI that included three cases or more. Subgroup analyses were performed for metabolite levels based on medication status (unmedicated [i.e., antipsychotic-free] patients [FEP + SZ]), illness stage (HR or FEP or patient [PT] group which includes FEP + SZ), treatment resistance (TRS or PT group except TRS cases), macromolecule nulling (GABA with or without macromolecules), metabolite correction method (creatine [Cr] scaling, CSF correction, or partial volume correction [PVC]), and magnetic field strength (only for 7T study). If there were three or more cases of TRS, we conducted subgroup analyses for PT group excluding TRS cases. The statistical methods of meta-regression, sensitivity analysis, and publication bias were described in Supplementary Material 3.

In the main meta-analysis, a Bonferroni-corrected threshold was applied for statistical significance of P < 0.00076 (=0.05/66) since we tested 66 group differences between the whole group or SZ group and HC. Two-sided 95% confidence intervals (CIs) were used for all subgroup analyses to assess significance, depending on whether the CIs included the null value. The R codes of this study are available from the corresponding author upon reasonable request.

Results

Characteristics of included studies

Out of 2864 initial records, 134 articles were identified, with a total of 7993 patients and 8744 HC (Supplementary Fig. 2). There are 84 studies for Glx [18, 24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93], 68 for Glu [12, 26, 27, 30, 31, 33,34,35, 37,38,39, 45, 46, 48, 49, 51, 54,55,56, 65, 67, 77, 80, 85, 86, 88, 90, 92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131], 11 for Gln [12, 49, 86, 93, 94, 99, 100, 107, 117, 125, 129], and 30 for GABA [32, 34, 41, 42, 49, 52, 57, 64, 69, 77, 79, 80, 83, 86, 91, 93, 96, 104, 107, 116, 118, 126, 129, 130, 132,133,134,135,136]. There are 23 studies on the HR group, 41 on FEP, 77 on SZ, 39 on unmedicated patients, and 5 on TRS. Regarding the ROIs, the number of cases is described in Table 1. The sample sizes of included cases ranged from 5 to 88 for the patient group and 9 to 184 for HC. The average ages ranged from 19.2 to 46.3 for HR group, 19.5 to 30.0 for FEP group, 14.0 to 66.3 for SZ group, and 11.0 to 70.0 for HC. The patient details were described in Supplementary Material 4. Seventeen, 96, 8, and 11 studies were performed at a magnetic field strength of 1.5T, 3T, 4T, and 7T, respectively. Thirty-eight, 1, 22, 38, 32, and 2 studies used Cr scaling, N-acetylaspartate scaling, water scaling, CSF correction, PVC, or other methods for metabolite correction. MRI protocols of the included studies are described in Supplementary Table 1. Out of 134 studies, 92 (68.7%) showed a “low” risk of bias for five items (Supplementary Fig. 3), excluding “selective outcome reporting”, which we judged to be unclear since we could not obtain experimental protocols.

Table 1 Meta-analysis result summary in all brain regions.
Full size table

Meta-analysis and subgroup analysis in a random-effect model

Table 1, Figure 1, and Supplementary Fig. 4 showed the detailed results of each analysis. We could not perform some meta-analyses due to the limited number of studies.

Fig. 1: Summary effect sizes for metabolite group differences in each brain region.
figure 1

Effect size estimates and 95% CIs are presented for meta-analysis results of glutamatergic and GABAergic metabolite levels in each brain region. Negative Hedge’s g values denote lower metabolite concentrations in patients than healthy controls; positive values denote higher metabolite concentrations in patients than healthy controls. Colored circles indicate significant group differences while white circles indicate non-significant group differences in the corresponding meta-analyses. (a) Comparison between the whole group and healthy controls. (b) Comparison between the schizophrenia group and healthy controls. Abbreviations: ACC anterior cingulate cortex, CI confidence interval, DLPFC dorsolateral prefrontal cortex, GABA γ-aminobutyric acid, Gln glutamine, Glu glutamate, Glx combined glutamate and glutamine signal, MCC midcingulate cortex, MFC medial frontal cortex, WM white matter.

Full size image

Basal ganglia

Meta-analyses

Glx in the basal ganglia were higher in the whole group than in HC, while no difference was found in Glu and GABA levels between the groups. There were no significant differences in glutamatergic or GABAergic metabolite levels between the SZ group and HC.

Subgroup analyses

There were higher Glx concentrations of the basal ganglia in the FEP group, PT group, and unmedicated group in comparison with HC. Individuals with HR and PT group had higher Glu levels in the basal ganglia compared with HC while Glu levels did not differ between FEP or unmedicated group and HC. TRS group did not show significant differences in Glx and Glu levels. We could not perform subgroup analyses for Glx in HR group and GABA, and meta-analysis of Gln.

Hippocampus

In the hippocampus, no significant differences were found in Glx or Glu levels between the whole group, SZ, PT, FEP, or HR groups, and the HC group. Unmedicated patients had higher Glx levels than HC, while we could not examine the group difference in Glu levels between them. We could not perform meta-analyses of Gln and GABA levels, or a subgroup analysis for TRS.

ACC/MCC/MFC

In the ACC, MCC, and MFC, there were no significant differences in glutamatergic or GABAergic metabolite levels between the whole group or SZ and HC groups. Subgroup analyses showed that GABA levels in the MCC were decreased in the FEP group, unmedicated group, and PT group compared with HC. Patients with TRS had higher Glx and Glu levels of the MCC than HC. In addition, we found a significant decrease in Glu levels in the PT group excluding patients with TRS than HC. The detailed results of subgroup analyses were shown in Supplementary Material 5.

DLPFC

In the DLPFC, there were no significant results in Glx, Glu, or GABA levels between the whole group or SZ group and the HC group. Unmedicated patients had elevated Glx levels compared to HC, while no differences were found in other subgroup analyses of Glx levels. We could not perform subgroup analyses for Glu and GABA levels or meta-analyses for Gln levels.

Occipital cortex

In the occipital cortex, there were no significant differences in Glx, Glu, and GABA levels between the whole group or SZ group and the HC group. PT group had lower GABA levels than HC. There was insufficient data for other subgroup analyses and meta-analysis of Gln.

Temporal WM

In the temporal WM, no significant differences were found in Glx, Glu, and GABA levels between the whole group or SZ group and the HC group. Glx levels were higher in PT group than HC. There were insufficient data for meta-analyses of Glu, Gln, and GABA in the whole and SZ groups. Also, we could not perform other subgroup analyses of Glx.

Thalamus

In the thalamus, there was no significance in Glx and Glu levels between the whole, SZ, PT, or FEP groups, and HC. Individuals with HR had decreased Glx and Glu levels compared with HC. Gln levels did not significantly differ between the whole group and HC group while higher Gln levels were found in the PT group than HC. We could not perform other subgroup analyses of Gln levels. Also, data were insufficient for a meta-analyses of GABA levels and subgroup analyses for Glx and Glu levels in the TRS or unmedicated groups.

Other regions

In the frontal cortex, frontal WM, occipito-parietal cortex, temporal cortex, and cerebellum, there were no significant differences in the main meta-analyses and subgroup analyses. We could not perform meta-analyses for the PCC, parietal cortex, parietal WM, and occipital WM.

The results of subgroup analyses based on the metabolite correction method were shown in Supplementary Fig. 5 and Supplementary Material 6. Subgroup analyses for 7 T 1H-MRS studies did not show any significant result in each ROI between the whole group and HC. Also, in subgroup analyses for macromolecule nulling, MCC GABA with macromolecules did not differ between the whole group and HC. We could not perform subgroup analysis of GABA without macromolecules due to the limited number of studies. Results of meta-regression, sensitivity analysis, and publication bias were shown in Supplementary Fig. 6, Supplementary Fig. 7, and Supplementary Material 7.

Discussion

Utilizing the largest dataset to date, we performed a meta-analysis to compare levels of Glx, Glu, Gln, and GABA in a wide range of brain regions between individuals with SZ-spectrum disorders at different clinical stages and HC considering clinico-demographic factors that may potentially correlate with these levels. Our main findings are five-fold. First, in the whole group, Glx levels in the basal ganglia were higher than HC with small effect sizes. Glu levels in the basal ganglia, Gln levels in the thalamus, and Glx levels in the temporal WM were elevated in comparison with HC whereas GABA levels in the MCC and occipital cortex were lower than HC, with small effect sizes, which did not survive after the correction for multiple comparisons. Second, in the SZ group, there were increases in Glx levels in the basal ganglia with a small effect size, which did not remain after the correction for multiple comparisons. Third, in the HR group, Glu levels were increased in the basal ganglia with a large effect size whereas Glx and Glu levels of the thalamus were lower in the HR group than HC with small effect sizes. Fourth, unmedicated patients had decreased GABA levels in the MCC and elevated Glx levels in the basal ganglia, hippocampus, and DLPFC with small effect sizes. Fifth, the TRS group had elevated Glx and Glu levels in the MCC with medium effect sizes, and MCC Glu levels were decreased in the PT group except TRS with a very small effect size. The strengths of our paper are as follows. We included more than twice as many papers as most recently published meta-analysis papers [8], which allowed us to examine metabolite abnormalities with distinguishing detailed regions of the brain. Also, we could perform subgroup analyses based on clinical stages and treatment responses that were previously unknown due to the small number of published papers. Since this meta-analysis included more studies at high magnetic field strengths (especially the results of 7 T 1H-MRS) than the previous meta-analysis, we could examine group differences in concentrations of both glutamatergic neutometabolite and GABA measured with ultra-high field magnetic strengths.

Through every illness stage, Glx levels in the basal ganglia were consistently increased, although the results in the SZ group did not remain after the correction for multiple comparisons and we could not examine the HR group. This suggests elevated glutamatergic metabolite levels in the basal ganglia could be considered as one of the traits related to the pathophysiology of SZ-spectrum disorders. These results are consistent with the data suggesting increased glutamatergic activity due to reduced NMDAR function in ketamine animal models of early SZ [21, 137, 138]. A preclinical model suggested that glutamatergic overactivity in the hippocampus promotes subcortical dopamine overrelease via multisynaptic glutamatergic projections into the striatum [139]. Indeed, in healthy adults, striatal Glu levels were positively associated with dopamine synthesis capacity [140], which was reported to be increased in patients with SZ [141]. Thus, glutamatergic abnormalities in the basal ganglia may affect striatal dopaminergic signaling and contribute to the positive symptoms of SZ. With regard to the changes in glutamatergic metabolite concentrations over the illness stage, the HR group showed the largest effect size of Glu while the effect sizes were reduced in the PT group. Moreover, no significant differences were found in Glu levels between the FEP or SZ group and HC group. Similarly, for Glx levels, the HR group may have contributed to the increased Glx levels in the whole group, although we could not perform the subgroup analysis of the HR group since there were few studies. This is suggested by the results that the effect sizes of basal ganglia Glx in FEP and SZ group were smaller than the SMDs of two HR studies (SMD = 0.44 in de la Fuente–Sandoval 2011, SMD = 0.84 in de la Fuente–Sandoval 2015) [52, 65]. Also, unmedicated patients have increased Glx levels compared with HC. Meta-regression analyses showed that SMDs of Glu levels in the basal ganglia were negatively associated with age, duration of illness, and proportion of medicated patients. These findings support that abnormally high Glu and Glx levels of the basal ganglia likely exist in patients with SZ even before the onset of psychosis [52, 65] and antipsychotics normalize those levels [32, 142]. In line with these results, animal studies reported that administration of several antipsychotics inhibited glutamate increase in the prefrontal cortex induced by NMDA antagonists including PCP or MK-801 [143,144,145]. Although there are some hypotheses that the blockade of neurotransmitter receptors such as 5-HT2A receptors and α1 adrenoceptors by antipsychotics may inhibit the PCP-induced increases in Glu release [143], the mechanisms of antipsychotic effects remain unclear. On the other hand, a recent meta-analysis did not show any significant difference in the Glx levels of the basal ganglia pre- and post-antipsychotic treatment in SZ [146]. However, only three studies examined the effect of antipsychotics on the Glx levels, which clearly warrants further investigations.

In the hippocampus, unmedicated patients had higher Glx levels compared with HC, while no differences were found in other groups. These results suggest that antipsychotics may normalize elevated Glx levels of the hippocampus in SZ, although Kraglujac et al. previously reported no change in hippocampal Glx levels with 6-week antipsychotic treatment. Since they examined short-term effects of antipsychotic treatment, further longitudinal studies are warranted to examine the long-term effect of antipsychotic treatment on the glutamatergic system in the hippocampus of SZ. Similar to the elevated Glx in the basal ganglia, the glutamate hypothesis may explain elevated Glx levels in the hippocampus of SZ. Ketamine administration may induce the hypofunction of parvalbumin interneurons, resulting in an elevation of Glx levels in the hippocampus of HC [147]. Thus, imbalance of excitation-inhibition in the hippocampus may be associated with pathophysiology of SZ. Unfortunately, we could not perform meta-analyses of hippocampal GABA levels, since there was only one study that compared hippocampal GABA levels between patients with SZ and HC [126], which emphasizes the need for further studies.

The present study revealed new results by dividing the MFC into the ACC and MCC. GABA levels in the MCC were lower in the FEP, unmedicated, and PT group while those levels did not differ in the ACC in comparison with HC. The previous postmortem study noted that the distribution of Glu and GABA receptors was different between the ACC and MCC [148]. Also, several studies in HC suggested that these regions may be involved in different functions and inhibit one another [149]. A meta-analysis reported that cognitive tasks activated the MCC, while emotionally-valenced tasks activated the ACC [16]. Thus, it is speculated that these two regions may differently contribute to the manifestation of symptoms of SZ. On the other hand, there were no differences in glutamatergic metabolites and GABA levels in the MFC between the whole group and HC. These results were consistent with previous meta-analyses that showed no differences in the glutamatergic metabolite and GABA levels in the MFC between patients and HC [8, 9, 11], which might be caused by lack of discrimination between the ACC and MCC. Future studies should consider detailed voxel locations within the MFC (i.e., the ACC or MCC) and analyze these two regions separately.

Lower levels of the MCC GABA were in line with postmortem studies in SZ consistently reporting a reduction in the GABA-synthesizing enzyme, GAD67 mRNA, and protein [9, 150,151,152]. In addition, animal models suggested that hypofunction of GABAergic neurons leads to the disinhibition of glutamatergic pyramidal neurons, and causes reduced power or synchrony of evoked gamma oscillation, which is thought to be associated with cognitive dysfunction in SZ [153]. Moreover, lower GABA levels were found in both FEP and unmedicated patients though those levels did not differ between HR group and HC. Thus, GABA levels in MCC may decrease with the onset of psychosis. Also, decreased GABA of the occipital cortex were found in the PT group. These GABA reductions may possibly be associated with the pathophysiology of SZ-spectrum disorders.

The present study also examined glutamatergic and GABAergic metabolite levels in patients with TRS. TRS group had higher Glx and Glu levels in the MCC than HC with medium effect sizes. On the other hand, subgroup analysis for the PT group except for patients with TRS showed that MCC Glu levels were significantly decreased compared with HC. In both of the two studies, glutamatergic metabolite levels were more widely distributed in the TRS group than in the non-TRS group, which may be due to the diverse subgroups in the TRS group (i.e. early onset TRS/late onset TRS, clozapine-responsive/-resistant) [31, 82], and some patients with TRS had by far increased glutamatergic metabolite levels. Thus, while patients with SZ generally have lower Glu levels in the MCC, a certain subgroup of TRS may have higher glutamatergic metabolite levels in this region, which may contribute to the treatment resistance to antipsychotics. In addition, Glx and Glu levels of the thalamus were also lower in the HR group than HC. Our findings of decreased glutamatergic metabolite levels in the MCC and thalamus were inconsistent with the previous meta-analysis that did not find any reduction in glutamatergic metabolites in SZ [8]. There are several hypotheses for decreased glutamatergic metabolite levels such as glutamatergic excitotoxicity, antipsychotic effects, or neurodevelopmental dysfunction, which are discussed in Supplementary Material 8. In addition, while Glx and Glu levels in the thalamus were decreased in the HR group than HC, Gln levels in the thalamus were higher in the whole group and PT group. Since no study examined Gln levels in the thalamus of the HR group, it remains unclear whether these opposite results were induced by differences in illness stages or glutamatergic metabolite types. Further longitudinal studies are needed to reveal the change of glutamatergic metabolites over the clinical course.

We performed subgroup analysis from the perspective of the methodological differences of quantifying metabolite levels, such as magnetic field strength, macromolecule nulling, and metabolite correction method. Discussion about these analyses and limitations of our study are described in Supplementary Material 9-11.

Conclusion

This study provides new insights into the dynamics of neurometabolites, indicating that alterations in glutamatergic or GABAergic neurotransmissions are different among detailed brain regions and disease stages in SZ-spectrum disorders. The regional pattern of neurometabolite abnormalities may progress over the course of the illness or respond differently to antipsychotic treatment. This evidence of abnormal neurotransmission has also led to an interest in the therapeutic potential of pharmacological compounds that act on glutamatergic or GABAergic function for this population.