Introduction
The N-methyl-D-aspartate receptor (NMDAR) hypo-function model of psychosis originated from pharmacological studies documenting the emergence of positive and negative symptoms as well as cognitive deficits in healthy volunteers exposed acutely to the NMDAR blocker ketamine (1). Also, acute systemic NMDAR blockers in the awake rat lead to an increase in frontal extracellular glutamate (1). Consistent with these findings, a single ketamine infusion in healthy controls (HC) results in an increase in the combined glutamate and glutamine signal (Glx) (2) in medial frontal cortex, as measured with proton magnetic resonance spectroscopy (1H-MRS). This paradoxical increase in glutamate release with NMDAR blockers has been postulated to result from higher sensitivity of NMDAR receptors in GABAergic interneurons than in pyramidal neurons, leading to disinhibition of pyramidal neurons (1).
Schizophrenia studies using single-voxel 1H-MRS suggest glutamate increases in “basal ganglia” (3) though more recent studies at 7T in early antipsychotic-treated psychosis found prefrontal glutamate reductions (4, 5). Furthermore, never-treated schizophrenia patients had no glutamate differences relative to healthy control subjects in dorsal anterior cingulate (6), but increases have been documented in the dorsal striatum in this population (7). Like schizophrenia, bipolar-I is a chronic, neurodevelopmental disorder, often presents with psychotic symptoms, and is commonly treated with antipsychotic medications (8). Several 1H-MRS studies have reported increased glutamatergic measures in bipolar-I (9) [but see (10)]. Hence, the literature on 1H-MRS-measured glutamate in psychotic disorders is far from clear, with region of interest, medication status, chronicity and co-morbidity as possible confounders. To partially address some of these issues, a few direct comparisons between bipolar-I and schizophrenia have been implemented. Two of these used single-voxel 1H-MRS and did not ascertain a history of psychosis in the bipolar-I sample (11, 12). Our previous study used single slice spectroscopic imaging and involved mainly chronic patients, all with history of psychosis (13). However, none of these three studies found glutamatergic differences between the two clinical groups.
Schizophrenia and bipolar-I are disorders that likely involve distributed brain networks and subtle global brain volume reductions (generally greater in schizophrenia than bipolar-I) have been repeatedly documented (14). Three-dimensional proton MR spectroscopic imaging (3D 1H-MRSI) enables measurement in much larger brain regions, thereby reducing the bias of voxel selection intrinsic to single-voxel studies. Using 3D 1H-MRSI with a short echo time (TE) and a voxel-wide approach, we recently reported reduced Glx in the left superior temporal gyrus (STG) in early schizophrenia vs. healthy control subjects as well as more widespread creatine increases in antipsychotic treated schizophrenia (15).
Here we report the use of the same imaging tool to compare Glx and other metabolites of interest in early schizophrenia vs. early bipolar-I disorder with psychotic features. Healthy control subjects (HC) were also examined to assist in interpretation. We include the schizophrenia (N = 36) and HC (N = 29) subjects' data previously reported plus an additional sample of schizophrenia, HC, and bipolar-I subjects. Consistent with our previous findings we expected Glx reductions in STG in schizophrenia (15) but not in bipolar-I. We also hypothesized higher creatine in antipsychotic-treated vs. antipsychotic-naïve schizophrenia (15).
Discussion
To our knowledge this is the first study to compare Glx as well as the other more commonly measured neurometabolites in early schizophrenia and BP-I with psychotic features with a voxel-wise whole brain spectroscopic imaging approach. Several metabolites were higher in Sz subjects compared to BP-I subjects across various regions. Glx was higher in the right middle cingulate, NAA was higher in the left occipital and bilateral frontal areas of cortex, t-Cho was higher in bilateral frontal cortices and myo-inositol was higher in the left frontal regions. These differences were not due to demographic, clinical or spectral quality metrics (see Supplementary Tables 3, 4). The HC group had intermediate values of these metabolites (though not significantly different) relative to the clinical groups, supporting the view that the two early-psychosis groups may have opposite neurometabolic abnormalities: increases in Sz and reductions in BP-I. Additionally we demonstrated effects of antipsychotic treatment in t-Cr in the Sz group, as previously suggested (15). Likewise, the left frontal increases in NAA and t-Cho at least in part appear to be related to antipsychotic treatment. However, the elevated Glx in the right cingulate in Sz appears to be accounted for mainly by the antipsychotic-naïve subjects. Contrary to our hypothesis, Glx in Sz was not lower in the left STG relative to the other groups.
Three 1H-MRS studies have directly compared chronically-ill Sz and BP-I subjects at magnetic fields >1.5T (3T and 4T). Two used single-voxel 1H-MRS and also did not restrict their sample to BP-I with psychosis (11, 12). In a previous study we used single slice spectroscopic imaging and only included BP-I with history of psychosis (13). None of these three studies found significant glutamatergic differences between Sz and BP-I. Atagun et al. (11) did find higher t-Cr and inositol in the left STG in Sz vs. BP-I. We found some age-related differences with higher NAA and inositol in younger Sz (<40) as well as higher NAA in older Sz relative to older BP-I (13). These results are somewhat consistent with the higher NAA and myo-inositol in early Sz relative to BP-I in the current study.
Increased glutamate, glutamine, and Glx (glutamate + glutamine) in the “basal ganglia” has been reported in one schizophrenia meta-analysis (3). This is consistent with the NMDAR hypo-function pharmacological model of psychosis: acute ketamine induces increased extracellular frontal glutamate in rodents and of 1H-MRS glutamatergic measures in healthy humans (1). Though concentrations of glutamate and glutamine measured with 1H-MRS do not directly assesses synaptic glutamate turnover, it has been proposed that the acute increment in extracellular glutamate in rodents measured with micro-dialysis reflects synaptic function (1). In humans, the largest published 1H-MRS ketamine challenge study reported increased Glx in anterior cingulate in a single-voxel (2). Hence, though less specific, Glx may be a more sensitive 1H-MRS index for the paradoxical effect of higher glutamate observed in acute NMDAR inhibition by ketamine. In the present study, the one cluster with increased Glx in Sz involved a more posterior aspect of the right dorsal cingulate. We note, however, that with the whole-brain 1H-MRSI approach we used, we may have been less able to detect group differences in the ventral and anterior cingulate, due to worse spectral quality in these regions (Figure 1). Likewise, the head of the caudate and the anterior thalamus, regions where glutamate increments have been described in antipsychotic-naïve schizophrenia (3), were mostly inaccessible with our approach.
The failure to confirm our previously reported Glx reduction in left STG in early Sz (15) was not due to differences in spectral quality between the original (CRLB = 7.58 ± 1.24) and the added Sz samples (CRLB = 7.07 ± 1.33). Also, the voxel coverage was the same in both groups (16 voxels over left STG). The two samples did differ in the number of antipsychotic-naïve Sz subjects: 57% in the original and 12% in the added samples (Supplementary Table 5). A histogram of z-scores of the examined voxels (Supplementary Figure 14) shows that the data from the current sample had a greater spread of Sz vs. HC differences, with the added subjects having z-scores around zero (i.e., no group effect). However, examination of the histogram z-scores did not support a greater effect in antipsychotic-naïve vs. HC than in treated Sz vs. HC (Supplementary Figure 14).
Results of previous spectroscopic studies examining glutamatergic metabolites are likely to have been affected or limited by antipsychotic medication, chronicity, and the brain region examined. More recent single-voxel studies at 7T in treated early schizophrenia have reported reduced glutamate in anterior cingulate (4, 5), suggesting an effect of antipsychotic medication. Consistently, antipsychotic-naïve schizophrenia patients had increased dorsal striatal glutamate, which normalized with prospective risperidone treatment (7). However, a recent study in antipsychotic-naïve schizophrenia failed to detect any differences in anterior cingulate glutamate (6). Furthermore, two 7T studies of the anterior cingulate in chronically-ill patients found glutamate reductions (26, 27). Hence, the inconsistencies in the results of glutamate studies in schizophrenia may be in part due to differences in stage of illness, medication effects and regions of interest with the single-voxel approach. Clearly, psychotic disorders involve multiple distributed brain networks (14) hence, examining glutamatergic measures with an unbiased voxel-wise approach offers advantages for future studies. Though antipsychotic medications may lower glutamate in some regions (7), glutamatergic dysfunction intrinsic to the illness is supported by the report of several glutamate-related risk-conferring genes in in the largest GWAS study in schizophrenia (28). Finally, in terms of diagnostic specificity our results suggest that glutamatergic abnormalities may differ fundamentally in bipolar-I and schizophrenia, with elevations in schizophrenia, consistent with a primary NMDAR hypofunction in GABA-ergic interneurons, and reductions in bipolar-I, suggestive of a different glutamatergic deficit, perhaps of NMDAR dysfunction in pyramidal neurons. However, the Glx differences were not related to symptom severity, like manic or negative symptoms, which suggests a trait effect of diagnosis.
Higher NAA, t-Cho and myo-inositol in Sz relative to BP-I were unexpected findings in the current study. The majority of the literature on schizophrenia reports lower NAA mainly in “basal ganglia” and frontal lobe (29). However, a recent meta-analysis of cross-sectional studies reported progressive, stage of illness-related NAA reductions: only in hippocampus in high-risk subjects; in frontal and thalamic regions in early schizophrenia; and more widespread reductions (frontal, hippocampal, temporal, thalamic and parietal regions) in chronically-ill patients (30). This is consistent with progressive brain volume reductions from multiple longitudinal MRI studies (31). In BP-I, lower NAA in “basal ganglia” has been reported (29). There is no consistent evidence of alterations in t-Cho or t-Cr in either disorder (29). The findings on myo-inositol have been more sparse, but lower concentrations in the medial frontal region was reported in a meta-analysis in schizophrenia (32). However, the great majority of the analyzed data came from single-voxel studies (29, 32). None of the studies included involved whole brain with voxel-wise analyses.
The literature on antipsychotic medication effects on 1H-MRS measures was recently expanded by a meta-analysis of single-voxel longitudinal schizophrenia studies before and after treatment (33). Frontal Glx was found to be reduced and thalamic NAA increased by medication. In this meta-analysis no changes in inositol were found and t-Cho and t-Cr were not examined. However, one study reported increased medial temporal t-Cr and inositol in treated vs. anti-psychotic-naïve schizophrenia, suggesting a medication effect for these metabolites (34).
Antipsychotics clearly have acute metabolic effects with increased subcortical and reduced cortical metabolism (35, 36). In addition, these agents can induce striatal volume expansion in humans as well as cortical volume reductions as early as 12 weeks after treatment (37). In monkeys, antipsychotics induce global gray matter volume reductions (38) with a 14% glial reduction and 10% increase in neuronal density (39). Hence, because NAA is mainly found in neurons (40) [though also reported in immature oligodendrocytes (41)], it is possible that increased neuronal density causes a relative increase in NAA tissue concentration, which could account for the higher NAA findings in the treated early schizophrenia subgroup of the present study. The consistent more widespread NAA reductions reported later in the illness (30) suggest an effect of chronicity, perhaps related to loss of neuropil without gliosis as described in the postmortem literature (42).
We did confirm increased t-Cr in antipsychotic-treated vs. naïve Sz subgroups consistent with our previous report (15). The higher t-Cr with treatment suggests an effect on energy metabolism. A reduction in the forward rate constant of creatine kinase, the enzyme that converts creatine to phosphocreatine, was reported in antipsychotic-treated schizophrenia (43). Hence, the increase in t-Cr may represent an adaptation to the cortical metabolism-lowering effects of antipsychotics (35, 36), by increasing the total pool of creatine available for energetic transfer. Though alterations in t-Cho have not been consistently reported in single-voxel studies of schizophrenia (29), long-term exposure to antipsychotics in monkeys has been reported to result in increased frontal glial density (44) and reduced parietal glial density (39). Because the t-Cho signal represents cell membrane turnover (40), we speculate that the finding of higher t-Cho only in treated vs. naïve schizophrenia patients may suggest an adaptive glial response to antipsychotics in frontal cortex bilaterally.
The associations between some of the metabolite clusters that differed between Sz and BP-I and symptoms and cognition (for the whole clinical sample) were not predicted and do not survive multiple comparison correction. However, some intriguing patterns emerged. Mainly in bilateral frontal regions, higher t-Cho correlated with worse negative symptoms and more impaired cognition. Not surprisingly, negative symptoms and cognition were negatively correlated (τ = −0.38, p < 0.001). This may suggest that increased frontal glial density (as reflected in higher t-Cho) may result in worsened negative symptoms and cognition in psychosis. However, the similar association between bilateral frontal NAA clusters and negative symptoms is harder to interpret as higher NAA is usually described as an index of improved neuronal viability (40).
This study has several strengths. 3D EPSI allowed examination of most of the brain, including numerous gray and white matter regions known to be affected in schizophrenia and bipolar-I (14). The use of AFNI permitted a voxel-wise approach with false-positive correction, as is standard with other neuroimaging modalities. AFNI also allowed introduction of co-variates at the voxel level, a critical issue in 1H-MRS analyses since the partial volume tissue effects are substantial (23). However, some limitations should be acknowledged. First, the spectral resolution with the current sequence did not allow reliable discrimination of glutamate from glutamine, hence Glx results are presented. Second, though whole brain 1H-MRSI was acquired, greater magnetic field inhomogeneity precluded examination of more ventral regions, such as orbitofrontal and inferior temporal areas. Third, our sample of 21 BP-I is small relative to the other two groups. Finally, as this was a case-control design, mechanistic interpretations are inherently limited.
In summary, this study of early psychosis using voxel-wise examination with 3D 1H-MRSI revealed an increase in glutamate metabolism (Glx) in the right middle cingulate in schizophrenia but not in bipolar-I. This increase did not appear to be related to antipsychotic treatment or due to other common confounders, such as substance use or chronicity. Also, the metabolite levels of the HC group, were intermediate between those of that Sz and BP-I groups. This suggests that Sz and BP-I may have abnormalities in metabolism that alter glutamate in opposite directions. In other regions explored in the current study, NAA, t-Cho, and myo-inositol were also higher in Sz. This differs in particular with the broad schizophrenia literature of reduced NAA in mainly chronically-ill schizophrenia populations examined with single voxels. However, in our sample medication status appears to account statistically for the higher NAA and t-Cho. Likewise, higher t-Cr appears clearly related to antipsychotic treatment. Replication of these findings would support mechanistic postmortem investigations and therapeutic neuromodulation studies of the right cingulate in schizophrenia.