In this observational, longitudinal study we examined glutamate metabolites in the ACC and striatum in a cohort of patients with FEP undergoing antipsychotic treatment according to their normal clinical care. The main finding was that higher levels of glutamate metabolites in the ACC were predictive of non-response to antipsychotic treatment, supporting our primary hypothesis. ACC glutamate metabolite levels at baseline were predictive of improvements across positive, negative and general symptom domains. In contrast, glutamate metabolites in the caudate were not associated with response. We did not detect significant reductions in glutamate metabolites over time, nor associations between early glutamatergic change over 2 weeks and symptom severity at 6 weeks. Our results support the view that ACC glutamate elevations are associated with a poor antipsychotic response. With gains in predictive accuracy, ACC glutamate measures could form one component of a multivariate model to predict antipsychotic response from the first episode of psychosis.
The finding that elevated ACC glutamate metabolites are linked to a subsequently poor antipsychotic response is consistent with our previous prospective study examining FEP [18]. It is also accordant with the findings from most cross-sectional studies [10–16, 42] and meta-analysis [16] showing elevated ACC glutamate metabolites in antipsychotic nonresponsive patient groups. In contrast, two longitudinal studies in FEP have not found an association between elevated ACC glutamate metabolites at baseline and subsequent non-remission [20, 43] and one found the opposite relationship [44]. The latter study [44], examined a small voxel in the pgACC in an antipsychotic naïve cohort who had no history of substance use and reported lower glutamate levels in non-responders to risperidone, although this was not significant when glutamate was scaled to the internal reference of creatine-containing metabolites. One difference between our current and previous [18] studies and the two studies that did not find an association between ACC glutamate metabolites and response in FEP [20, 43] is that we examined a more rostral region of the ACC, corresponding to the perigenual ACC (pgACC). The pgACC is involved in internal mental processes and processing affective information [45, 46]. Glutamate concentrations are higher in rostral than caudal ACC regions [47] and differentially relate with whole brain resting state functional connectivity [48]. Early changes in pgACC connectivity predict antipsychotic response in FEP [29] and pgACC glutamate correlates with ACC control over sensory regions in antipsychotic-responsive but not treatment resistant schizophrenia [49]. Therefore, one possible explanation is that the mechanisms underlying antipsychotic response are localised to within the rostral, pgACC region.
Compared to some earlier research, our current study was designed to better reflect the possible ‘real-world’ conditions under which neuroimaging and other predictive biomarkers of response might be applied soon after first presentation. Prospective studies examining glutamate metabolites in relation to response in FEP have included participants who were antipsychotic naïve or had received minimal (< 2 weeks) antipsychotic exposure [18–20, 43, 44] and most examined a single antipsychotic compound [18–20, 44], whereas in our current study all participants were medicated at the baseline scan and antipsychotics were prescribed as usual by the treating clinical team.
Unlike in the ACC, we did not observe an association between caudate glutamate metabolites and subsequent antipsychotic response. Caudate glutamate metabolites also do not appear to differ in relation to response in cross sectional studies [12, 13, 15, 16]. One study reported higher striatal glutamate longitudinally in non-responders to risperidone, although this group difference was more apparent after risperidone treatment [19]. In meta-analysis, increases in glutamate metabolites in the striatum in schizophrenia are the most consistent finding across illness stages [16, 50], which indicates that they may be less sensitive to clinical variables and thus could provide a trait marker for schizophrenia spectrum disorders [16].
In contrast to some previous studies [18, 24–27], we did not observe decreases in glutamate metabolites during antipsychotic treatment. Although systematic review and meta-analysis has also indicated ACC or striatal glutamate decreases related to antipsychotic treatment [16, 17, 22, 23] this is below significance in several individual studies [19–21, 51–54]. The relatively short observation period and inclusion of previously medicated patients may also explain why we did not observe these effects. The effects of antipsychotics on glutamate are likely to be indirect and mediated through mechanisms downstream of antipsychotic binding at D2/3 and other receptor subtypes. These indirect mechanisms may lead to more subtle changes, and more variability in effect related to differences in antipsychotic pharmacology [55], or other factors such as duration of administration or patient characteristics. Glutamatergic excess associated with antipsychotic non-response is more likely to be effectively reduced by novel therapeutics which directly target the glutamate system.
While our study was designed to provide greater generalisability to normal clinical care of FEP than prior studies, the presence of antipsychotic medication at baseline and lack of standardization of antipsychotic treatment may have added variability and reduced statistical power. The study design, requiring capacity to consent and willingness to participate in multiple MRI scans and clinical assessments, may have excluded more severely unwell patients and this group could also be less likely to respond to antipsychotic treatment. Data acquisition and analysis was harmonized across two sites, but the application of MRI platforms from different manufacturers was associated with site effects and may have reduced statistical power. The 1H-MRS signal reflects the total amount of MR-visible glutamate or Glx in the voxel rather than glutamate neurotransmission specifically. Future studies examining changes in glutamate occurring in response to stimuli with functional 1H-MRS [56, 57] or combining 1H-MRS glutamate measures with functional connectivity [49], may provide greater mechanistic information about the role of glutamate in antipsychotic response and possibly greater predictive accuracy.
Our findings have potential clinical applications. Previous research has shown that when patients fail to respond to a first antipsychotic drug they are unlikely to respond to a second non-clozapine antipsychotic [58], and that the trial of a second antipsychotic rarely results in response, and merely introduces a delay in the use of clozapine, which in turn is associated with worse outcomes [6]. Accurate prediction of non-response during initial antipsychotic treatment could therefore identify a stratum of patients most likely to benefit from clozapine as a second line treatment. Hypothetical cost-effectiveness analysis indicates that predictive models with even modest sensitivity and specificity (over 60%) to indicate clozapine as a second line antipsychotic could achieve improvements in quality of life [8]. Baseline ACC glutamate or Glx was associated with values within this range, achieving up to 79% specificity and 67% sensitivity when the early percentage change in PANSS was included in the model. However, similarly to our previous study [18], antipsychotic responders were identified more accurately (maximum PPV: 82%) than non-responders (maximum NPV: 62%). This could be related to greater clinical or biological heterogeneity within the non-responder group, for example in some cases response may have emerged over a longer observation period. For the potential application of supporting earlier clozapine initiation in non-responders, improving the NPV is of clinical importance to avoid unnecessary clozapine initiation and the associated risks. Gains in accuracy could potentially be achieved by combining glutamate measures with other predictors of response, such as resting state activity [59], brain volumes [9], perfusion [60], striatal dopamine synthesis capacity [61], cognitive performance [62] and clinical or demographic predictors [40, 41, 63].
In summary, this prospective study in FEP found that higher levels of glutamate metabolites in the ACC were predictive of non-response to antipsychotic treatment 6 weeks later. This finding supports the proposition that elevated ACC metabolites may be associated with a poor antipsychotic response. Combining 1H-MRS glutamate measures with other variables associated with antipsychotic response may increase predictive accuracy.