Early Treatment Response in First Episode Psychosis: A 7-Tesla Magnetic Resonance Spectroscopic Study of Glutathione and Glutamate

Early response to antipsychotic medications is one of the most important determinants of later symptomatic and functional outcomes in psychosis. Glutathione and glutamate have emerged as promising therapeutic targets for patients demonstrating inadequate response to dopamine-blocking antipsychotics. Nevertheless, the role of these neurochemicals in the mechanism of early antipsychotic response remains poorly understood. Using a longitudinal design and ultra-high field 7-Tesla magnetic resonance spectroscopy (MRS) protocol in 53 subjects, we report the association between dorsal anterior cingulate cortex glutamate and glutathione, with time to treatment response in drug-naïve (34.6% of the sample) or minimally medicated first episode patients with non-affective psychosis. Time to response was defined as the number of weeks required to reach a 50% reduction in the PANSS-8 scores. Higher glutathione was associated with shorter time to response (F=4.86, P= .017), while higher glutamate was associated with more severe functional impairment (F=5.33, P= .008). There were no significant differences between patients and controls on measures of glutamate or glutathione. For the first time, we have demonstrated an association between higher glutathione and favourable prognosis in FEP. We propose that interventions that increase brain glutathione levels may improve outcomes of early intervention in psychosis.


Introduction
Early treatment response has been identified as one of the most robust predictors of longer-term clinical outcomes in schizophrenia. 1 In particular, lack of early response appears to be strongly indicative of subsequent non-response, 2 failure to achieve remission 3 , and higher rates of treatment discontinuation 4 . Approximately one third of patients with schizophrenia are considered to be treatment resistant 5 , with the majority of these (23-34%) failing to respond appreciably to dopamine-blocking antipsychotic medications from their first episode of psychosis (FEP). 6,7 Nevertheless, the neurochemical mechanism of early response is poorly understood, precluding efforts to prevent or reduce the rates of treatment failure and persistent disability.
The FEP is characterized by a relative state of glutamatergic excess. 8,9 Elevated anterior cingulate cortex (ACC) glutamate has been found to be inversely correlated with striatal dopamine synthesis in patients with FEP 10 . Given that the elevated striatal dopamine synthesis relates to better treatment response 11 in psychosis, the observed glutamatergic excess has been considered to be an index of reduced treatment responsiveness in psychosis 12 .Elevated anterior cingulate cortex (ACC) glutamate has been directly associated with lack of remission in certain severity, with individuals with a higher initial symptom burden being much less likely to achieve remission 38 . As a result, we studied the continuous measure of time to response as the primary clinical outcome of interest, and used the cross-sectional remission criterion as a secondary measure of interest.
The 8 items of the Positive and Negative Syndrome Scale capturing the core symptoms critical in defining remission (PANSS-8 39 ) was administered at baseline, 2 weeks, 4 weeks, and at every clinical encounter thereafter on a 2-4 weekly basis. The PANSS-8 has acceptable internal consistency and comparable sensitivity to early improvement in psychotic symptoms 40 relative to the PANSS-30 41 . The time to achieve a 50% PANSS-8 improvement from baseline 42 , sustained for at least 2 consecutive visits 2 weeks apart, was used as a continuous measure of treatment response. A 50% symptom improvement from baseline roughly equates to a Clinical Global Impression-Schizophrenia (CGI-S 43 ) scale score of "much improved" thus, is clinically meaningful 44 . Relative PANSS8 improvement was calculated as (PANSS8 baseline -PANSS8 endpoint )/(PANSS8 baseline -8) in order to adjust for the minimal possible PANSS8 score 45 .
All patients were observed clinically for a period of at least 6 months, and no patients failed to reach this milestone within this time frame.
We also assessed binary remission status after the first month of treatment (remission or not in remission). Symptomatic remission was allocated based on remission criteria proposed by   39 which categorize remission as achieving scores of mild (3) or less on all PANSS8 items, without any stipulation of a duration criteria, in line with Egerton et al 16,17 .
Finally, social functioning was assessed at baseline using the Social and Occupational Functioning Assessment Scale (SOFAS 46 ).

Medication Adherence
Individuals were treated with long-acting injectable (LAI) medications whenever clinically appropriate. Patients taking LAI's received their injection from a nurse at the PEPP clinic and therefore, it was known if an individual had missed, or was late for their scheduled dose. Assessments of medication adherence were also recorded at each clinical encounter, taking into account information provided by the patient, their family, and/or case manager using a 5point rating scale (ranging from 0 for individuals not taking medication to 4 for those being adherent 75-100% of the time). This measure has been found to correlate with pill counts 47 . We only included subjects who had >75% recorded adherence. of the corpus callosum. The angulation of the voxel was determined to be tangential to the corpus callosum. A total of 32 channel-combined, water-suppressed spectra were acquired using a semi-LASER 1 H-MRS pulse sequence (TR=7500ms, TE=100ms) during each scan session, while participants were at rest and asked to stare at a white cross on a black screen for 4 minutes.
Water suppression was achieved using the VAPOR preparation sequence 48 , and waterunsuppressed spectra were acquired for spectral quantification and line shape deconvolution reference. The 32 spectra were corrected for frequency and phase drifts as described in   49 prior to averaging and lineshape deconvolution using QUECC 50 . Residual water peaks were removed from the averaged spectrum using HSVD 51  to assess the association between glutamate and GSH in patients and healthy controls.
Differences in the magnitude of these correlations were then evaluated using Fisher's r-to-Z transformation (Hypothesis 2).

Correlations Between Metabolite Levels
The association between glutamate and GSH was tested using Pearson correlation coefficients. There was a positive association between levels of ACC glutamate and GSH in both healthy control subjects (r= .91, p< .001), and in patients with FEP (r= .69, p< .001). We then used Fisher's r-to-z transformation to test the significance of difference between the correlations, and found that the correlation between glutamate and GSH was significantly weaker in patients compared to the healthy control subjects (Z= 2.26, p= .023) (see figure 3).

Group differences in GSH and Glutamate
One-way ANOVAs were conducted to evaluate the differences in metabolite levels among patients in remission or non-remission at one month and healthy control subjects. There were no significant difference between groups for glutamate (F(2,50)=.134, p=.875) or GSH (F(2,50)=.712, p=.496) (see Table 2). There were no significant differences between patients (as a single group) and controls on measures of glutamate (t(51)= -.266, p= .791) or GSH (t(51)= -.412, p=.682.
The effects of recreational substance use and types of antipsychotics are presented in the Supplement.

Discussion
This is the first study to use ultra high-field 7T MRS to investigate the role of glutamate and GSH in early treatment response, and the first 7T MRS study on minimally medicated FEP subjects. A previous 7T MRS study included FEP subjects with an average of 55 weeks of antipsychotic exposure 54 , compared to 6 days of median exposure in our sample. A more recent study 23 included FEP subjects with up to 2 years of illness duration, while we recruited all subjects during the acute first episode (mean SOFAS score of 38.1). We report 3 major findings (1) Patients with FEP with higher GSH levels in ACC show a rapid symptom reduction upon starting antipsychotic treatment (2) When compared to healthy controls, GSH levels in patients are dissociated from glutamate levels (3) Glutamate excess predicts the degree of Social and Occupational dysfunction seen at the time of presentation with FEP. Taken together, these results indicate that markers of cortical redox integrity influence the putative glutamatergic toxicity and early treatment response in psychosis.
Neither glutamate nor GSH were associated with binary remission status at one month.
The lack of association is in contrast with the overall results reported by another study 17 , but consistent with the observation reported by 2 out of the 3 sites in that study. These differences can be attributed to methodological variations (the use of 7T spectra, more dorsal voxel placement in our study) as well as notable differences in the clinical samples (the use of injectables and the inclusion of both inpatients and outpatients, and the exclusion of patients with low adherence in our study). Egerton 17 as well as prior observations from our centre 19,20 . A low level of social functioning at FEP is reported to be a robust and independent predictor of later treatment 55 . This finding adds strength to the prevailing notion that glutamatergic excess plays a critical role in shaping the poor outcome trajectory in psychosis.
We found no significant differences in GSH levels between patients and healthy controls. This is not surprising, given that meta-analytic pooling of ACC GSH studies in schizophrenia reveal a small overall effect size 24 , suggesting the possibility of heterogeneity in the GSH levels and thus redox status among patients. Our results suggest that such heterogeneity may map onto antipsychotic responsiveness, resulting in the conflicting findings of GSH levels reported to date in schizophrenia 23 .
We found evidence that despite their significant within-group correlation, when compared to healthy controls, glutamate and GSH levels were less tightly correlated among patients with FEP. A similar dissociation was also reported by Xin  A further limitation is that our spectroscopic analysis was limited to the dorsal ACC and did not include more anterior/ventral portions of the medial prefrontal cortex. We cannot completely rule out the effect of recreational substances on the observed results (see Supplement). One study 65 found that ACC glutamate levels were decreased in individuals who used cannabis regularly, while these results were not replicated in another study 66 . To our knowledge, no studies have investigated the association of GSH with cannabis use and none have examined the effects of cannabis on metabolite levels specifically in a FEP sample. Finally, our patient sample consisted primarily of males, limiting generalization of the results.
A promising implication is that interventions that increase GSH levels early in FEP may have the potential to alter the prognostic trajectory of psychosis (See Supplement -Translational Relevance for further details). A prospective sequential treatment trial 67 in first episode patients has indicated that merely switching antipsychotics may not boost treatment response in early non-responders, and second level treatments such as clozapine may be warranted even before the conventional clinical threshold of Treatment Resistant Schizophrenia (i.e. 2 treatment failures) is met. While early non-response is considered to be an indicator of later non-response and subsequent treatment resistance in schizophrenia 68 , to our knowledge, the association between early non-response in first-episode samples and later sequential treatment failures and the status of conventionally defined TRS is yet to be established. One of the challenges in this regard is the high degree of responsiveness to treatment seen in first-episode patients 69 (also observed in the current study), compared to those with acute exacerbation of chronic schizophrenia 68

Introduction
Early treatment response has been identified as one of the most robust predictors of longer-term clinical outcomes in schizophrenia. 1 In particular, lack of early response appears to be strongly indicative of subsequent non-response, 2 failure to achieve remission 3 , and higher rates of treatment discontinuation 4 . Approximately one third of patients with schizophrenia are considered to be treatment resistant 5 , with the majority of these (23-34%) failing to respond appreciably to dopamine-blocking antipsychotic medications from their first episode of psychosis (FEP). 6,7 Nevertheless, the neurochemical mechanism of early response is poorly understood, precluding efforts to prevent or reduce the rates of treatment failure and persistent disability.
The FEP is characterized by a relative state of glutamatergic excess. 8,9 Elevated anterior cingulate cortex (ACC) glutamate has been found to be inversely correlated with striatal dopamine synthesis in patients with FEP 10 . Given that the elevated striatal dopamine synthesis relates to better treatment response 11 in psychosis, the observed glutamatergic excess has been considered to be an index of reduced treatment responsiveness in psychosis 12  showed no glutamate excess in patients with FEP who did not achieve remission by 1 month 17 .
Nevertheless, relative glutamatergic excess appears to be specific to early stages of illness 8 , and relates to more severe symptoms at presentation 17 , as well as grey matter decline 19 , cognitive 20 and functional 16,17 impairments. The lack of dopamine elevation seen in some patients may explain their lack of response to dopamine blocking medications.
Glutathione (GSH), the brain's most prominent intracellular antioxidant has been suspected to play a key protective role in free-radical-mediated damage to neurons 21 , giving rise to the redox dysregulation hypothesis of schizophrenia 22 . MRS studies have found a small but significant GSH deficit in the ACC in patients with schizophrenia 23 , indicating the presence of subgroups of patients with different redox profiles 24 . The most prominent reduction in GSH seems to occur particularly in patients with persistent residual symptoms, indicating that low levels of GSH may be associated with poor response to antipsychotics 25 . Furthermore, N-acetylcysteine (NAC), a precursor of GSH, appears to increase the rate of symptomatic response when used as an adjunct to antipsychotics 26 .
Glutamate is a precursor of GSH while GSH acts as a neuronal reservoir for glutamate synthesis 27 . As a result, when neuro-glial metabolic integrity is intact, glutamate and GSH levels remain tightly linked in the brain. Glutamatergic excess can result in neurotoxic oxidative stress 28 , while a concomitant elevation of GSH may provide a neuroprotective 'gate-keeping' effect 29 , thus a strong covariance may be a marker of a healthy state. Nevertheless, repeated or prolonged exposure to excess glutamate can deplete GSH levels 30 . Furthermore, the GSHglutamate homeostasis may also be disrupted in patients with schizophrenia due to deficiencies in GSH synthesis 31 , leading to reduced GSH-glutamate covariance in patients with FEP.
In this study, we use ultra-high field 7T MRS for the first time to test the relative contribution of ACC GSH deficiency and glutamatergic excess in predicting early treatment response in FEP. Given the gatekeeper role of GSH in tackling oxidative stress 31 , we expected GSH to be a more critical determinant of early treatment response in FEP. We hypothesized that FEP patients with higher GSH levels would demonstrate faster symptom reduction upon starting antipsychotic treatment (hypothesis 1). As not all patients with FEP will be able to increase GSH in accordance with glutamate levels, we expected a reduction in the strength of correlation between the GSH and glutamate levels in patients compared to healthy controls (hypothesis 2).
Furthermore, in light of the excitotoxic theory of acute schizophrenia 32 , we expected both reduced GSH and increased glutamate levels to predict impaired Social and Occupational Functioning at the onset of illness (hypothesis 3).

Participants
The Healthy control subjects were recruited through the use of posters advertising the opportunity to participate in a neuroimaging study involving tracking outcomes following FEP.
Healthy control subjects had no personal history of mental illness, and no family history of psychotic disorders. Group matching with the FEP cohort for age, sex, and parental education was maintained. Exclusion criteria for both the FEP and healthy control groups involved meeting criteria for a substance use disorder in the past year according to DSM-5 36 criteria (this was based on self-report for controls, and in addition clinical assessment and urine drug screening done at the point of clinical assessment in suspected cases for patients), having a history of a major head injury (leading to a significant period of unconsciousness or seizures), having a significant, uncontrolled medical illness, or having any contraindications to undergoing MRI.

Clinical Measures
While the proportion of FEP patients in remission at any given time appears to be relatively consistent, it is often not the same individuals who remain in remission at each time point 37 . The use of absolute criteria in defining remission is highly dependent on initial illness severity, with individuals with a higher initial symptom burden being much less likely to achieve remission 38  All patients were observed clinically for a period of at least 6 months, and no patients failed to reach this milestone within this time frame.
We also assessed binary remission status after the first month of treatment (remission or not in remission). Symptomatic remission was allocated based on remission criteria proposed by   39 which categorize remission as achieving scores of mild (3) or less on all PANSS8 items, without any stipulation of a duration criteria, in line with Egerton et al 16,17 .
Finally, social functioning was assessed at baseline using the Social and Occupational Functioning Assessment Scale (SOFAS 46 ).

Medication Adherence
Individuals were treated with long-acting injectable (LAI) medications whenever clinically appropriate. Patients taking LAI's received their injection from a nurse at the PEPP clinic and therefore, it was known if an individual had missed, or was late for their scheduled dose. Assessments of medication adherence were also recorded at each clinical encounter, taking into account information provided by the patient, their family, and/or case manager using a 5point rating scale (ranging from 0 for individuals not taking medication to 4 for those being adherent 75-100% of the time). This measure has been found to correlate with pill counts 47 . We only included subjects who had >75% recorded adherence. of the corpus callosum. The angulation of the voxel was determined to be tangential to the corpus callosum. A total of 32 channel-combined, water-suppressed spectra were acquired using a semi-LASER 1 H-MRS pulse sequence (TR=7500ms, TE=100ms) during each scan session, while participants were at rest and asked to stare at a white cross on a black screen for 4 minutes.
Water suppression was achieved using the VAPOR preparation sequence 48 , and waterunsuppressed spectra were acquired for spectral quantification and line shape deconvolution reference. The 32 spectra were corrected for frequency and phase drifts as described in   49 prior to averaging and lineshape deconvolution using QUECC 50 . Residual water peaks were removed from the averaged spectrum using HSVD 51 . Metabolite quantification was acquired using Barstool 52 . Water-subtracted spectra were modelled using the fitMAN, a-priorknowledge based minimization algorithm, and a quantification template including 17 metabolite spectral signatures derived from simulation 52 . Our fitting template included 17 metabolites (alanine, aspartate, choline, creatine, GABA, glucose, glutamate, glutamine, glutathione, glycine, lactate, myo-inositol, N-acetyl aspartate, N-acetyl aspartyl glutamate, phosphorylethanolamine, scyllo-inositol, and taurine). Importantly, at this long echo time, no macromolecules were included in the spectra as their signal had decayed below noise level. Metabolite concentrations were corrected for gray and white matter volumes using the anatomical MRI images and previously described methods 53 . All spectra and spectral fit were inspected visually for quality and Cramer-Rao lower bounds (CRLB) were assessed for each metabolite. The MRS metabolite estimates were not known at the time of clinical outcome characterization. See the Supplement for further details on the MRS methods.

Statistical Analyses
All statistical tests were performed using IBM SPSS Statistics version 24. Differences in demographic and baseline factors between patients and controls were calculated using t-tests for continuous variables, and chi-square analyses for dichotomous variables. A linear regression analysis was used to assess the association between metabolites (glutamate and GSH), and both time to response, and social functioning (Hypotheses 1 and 3). Using ANOVA, we then compared glutamate and GSH measures among patients achieving remission at one month, no remission at one month, and healthy controls. Finally, Pearson correlation coefficients were used to assess the association between glutamate and GSH in patients and healthy controls.
Differences in the magnitude of these correlations were then evaluated using Fisher's r-to-Z transformation (Hypothesis 2).

Patient Characteristics
37 patients completed baseline scanning. Of these, 27 met criteria for a schizophrenia spectrum disorder (SSD: schizophrenia, schizoaffective disorder, or schizophreniform disorder).
Follow-up outcome data were not available for one female patient who was transferred to a different hospital shortly after scanning. In one male patient, time to response was not available due to irregular follow-up however, remission status at one month was obtained. Therefore, the final sample consisted of 26 patients with SSD, with time to response measures available for 25 patients (

Correlations Between Metabolite Levels
The association between glutamate and GSH was tested using Pearson correlation coefficients. There was a positive association between levels of ACC glutamate and GSH in both healthy control subjects (r= .91, p< .001), and in patients with FEP (r= .69, p< .001). We then used Fisher's r-to-z transformation to test the significance of difference between the correlations, and found that the correlation between glutamate and GSH was significantly weaker in patients compared to the healthy control subjects (Z= 2.26, p= .023) (see figure 3).

Group differences in GSH and Glutamate
One-way ANOVAs were conducted to evaluate the differences in metabolite levels among patients in remission or non-remission at one month and healthy control subjects. There were no significant difference between groups for glutamate (F(2,50)=.134, p=.875) or GSH (F(2,50)=.712, p=.496) (see Table 2). There were no significant differences between patients (as a single group) and controls on measures of glutamate (t(51)= -.266, p= .791) or GSH (t(51)= -.412, p=.682.
The effects of recreational substance use and types of antipsychotics are presented in the Supplement.

Discussion
This is the first study to use ultra high-field 7T MRS to investigate the role of glutamate Neither glutamate nor GSH were associated with binary remission status at one month.
The lack of association is in contrast with the overall results reported by another study 17 17 as well as prior observations from our centre 19,20 . A low level of social functioning at FEP is reported to be a robust and independent predictor of later treatment 55 . This finding adds strength to the prevailing notion that glutamatergic excess plays a critical role in shaping the poor outcome trajectory in psychosis.
We found no significant differences in GSH levels between patients and healthy controls. This is not surprising, given that meta-analytic pooling of ACC GSH studies in schizophrenia reveal a small overall effect size 24 63 suggest that serum GSH is affected by typical but not atypical antipsychotics. None of our patients were exposed to A further limitation is that our spectroscopic analysis was limited to the dorsal ACC and did not include more anterior/ventral portions of the medial prefrontal cortex. We cannot completely rule out the effect of recreational substances on the observed results (see Supplement). One study 65 found that ACC glutamate levels were decreased in individuals who used cannabis regularly, while these results were not replicated in another study 66 . To our knowledge, no studies have investigated the association of GSH with cannabis use and none have examined the effects of cannabis on metabolite levels specifically in a FEP sample. Finally, our patient sample consisted primarily of males, limiting generalization of the results.
A promising implication is that interventions that increase GSH levels early in FEP may have the potential to alter the prognostic trajectory of psychosis (See Supplement -Translational Relevance for further details). A prospective sequential treatment trial 67 in first episode patients has indicated that merely switching antipsychotics may not boost treatment response in early non-responders, and second level treatments such as clozapine may be warranted even before the conventional clinical threshold of Treatment Resistant Schizophrenia (i.e. 2 treatment failures) is met. While early non-response is considered to be an indicator of later non-response and subsequent treatment resistance in schizophrenia 68 , to our knowledge, the association between early non-response in first-episode samples and later sequential treatment failures and the status of conventionally defined TRS is yet to be established. One of the challenges in this regard is the high degree of responsiveness to treatment seen in first-episode patients 69 (also observed in the current study), compared to those with acute exacerbation of chronic schizophrenia 68 . In this context, caution is warranted in extrapolating the physiological correlates of early treatment response as indicators of the emergence of categorical treatment resistance at later stages of schizophrenia. Given that GSH levels have a significant impact on the speed of response, we urge further experimental trials that manipulate GSH levels to observe the predicted gain in trajectory of treatment outcomes in FEP.
Preliminary results have demonstrated that NAC, a GSH precursor, may be beneficial in psychotic disorders 70 . NAC has been shown to be efficacious in reducing the symptom burden 71 , especially negative 72 and cognitive symptoms 73 , and has the potential to alleviate treatment resistance in schizophrenia 74 . Our results suggest that treatments such as NAC may be efficacious particularly in patients who demonstrate an early poor response to antipsychotic medication, as they are likely to have a lower ability to synthesize GSH in response to glutamatergic excess. While MRS indices are indirect measures of tissue metabolite concentrations 75 , given the evidence that oral NAC administration in patients with schizophrenia increases GSH content in the ACC 70 , we consider MRS as a viable tool for translational investigations into the redox abnormalities of schizophrenia. More speculatively, we suggest that the association of ACC GSH levels at baseline and eventual clozapine-eligibility would be worth investigating in the future, given the lack of objective predictors of clozapine requirement in schizophrenia.