Antipsychotics are the first choice of treatment for people with schizophrenia or other related psychotic disorders (see Mental Elf blog by Elwira Lubos, 2017). However, for up to 30% of people with schizophrenia, antipsychotics are not effective and identifying methods for predicting who will respond to treatment remains a major clinical challenge. People with psychosis show substantial biological and clinical heterogeneity, leading to highly variable treatment outcomes and prolonged periods of ineffective treatment. As Dolly Sud (2020) gracefully noted in her Mental Elf blog, within the world of medicine:
how can we best help everyone, when everyone is different?
In this study, the authors ponder whether understanding the neurobiological mechanisms that contribute to a poor antipsychotic response, and identifying biomarkers that can predict response could guide clinical interventions and help inform new treatments that support people with psychosis who have varying responses to antipsychotics.
Proton magnetic resonance spectroscopy (¹H-MRS) is a method used to measure neuro-metabolites or ‘brain chemicals’ linked to schizophrenia (Kraguljac N V. et al., 2012). Previous meta-analyses (e.g. Nakahara et al., 2022; Merritt et al., 2021) suggest that metabolite levels in the brain differ depending on how people respond to antipsychotic treatment. However, studies analysed data from groups, rather than individuals, and measured the differences cross-sectionally at only one point in time. By contrast, this study by King and colleagues (2026) aimed to explore what the profile of 1H-MRS metabolites looked like, in relation to treatment responders and treatment non-responders in schizophrenia using a mega-analysis of individual participant-level data.
Methods
The authors pre-registered the review on PROSPERO and followed PRISMA reporting guidelines (Preferred Reporting Items for Systematic Reviews and Meta-Analyses). A comprehensive strategy was applied to search the Web of Science database for journal articles published up to August 2024, with an updated search in November 2025 for the meta-analysis.
Mega-Analysis: This involved combining original individual patient data from different studies and analysing it collectively, as if it came from one large study (Norman L. & Shaw P. 2024). Separate analyses were conducted for each ¹H-MRS metabolite and brain region due to their distinct biological roles and regional differences. Using linear mixed models, the authors compared antipsychotic non-responders, responders, and healthy controls. Secondary analyses focused on first-episode psychosis (FEP) studies and people who had treatment-resistant schizophrenia. Additional analyses assessed whether medication dose (chlorpromazine equivalents) or symptom severity (PANSS scores) influenced metabolite differences.
Meta-Analyses: Random-effects meta-analysis was used to estimate the overall effect size and the variation between studies.
Results
Using mega-analysis, King and colleagues addressed three key research questions in this study:
1. What did the profile of 1H-MRS metabolites look like for treatment responders and treatment non-responders with schizophrenia.
Non-responders to antipsychotic treatment had higher levels of several metabolites in the medial frontal cortex region of their brain, than those who did respond to treatment. Altered brain metabolites included glutamate, glutamate + glutamine (Glx), n-acetylaspartate (NAA), choline and myo-inositol. This means that biological differences in the medial frontal brain may distinguish treatment responders from non-responders and could help guide future biomarker research. However, the differences were small (effect sizes 0.21 to 0.35), suggesting only modest differences between responders and non-responders.
2. Were baseline metabolites associated with subsequent treatment response?
The authors focused only on studies in which 1H-MRS measures were taken in people experiencing FEP who had minimal exposure to antipsychotic treatment. Elevated medial frontal Glx was already present before substantial antipsychotic exposure in people who later failed to respond to treatment. This means that that glutamatergic abnormalities may precede non-response to treatment. Myo-inositol elevations appeared most pronounced in treatment-resistant schizophrenia, which means that some metabolite abnormalities may be more specific to treatment resistance.
3. Were group differences in metabolites specific to people with treatment-resistant schizophrenia?
People with treatment-resistant schizophrenia showed higher levels of choline and myo-inositol in the medial frontal cortex than people who responded to antipsychotic treatment. This suggests that these metabolites may be more specific markers of treatment resistance.
Conclusions
The review found evidence of altered neurometabolites in people who did not respond to antipsychotic treatment, compared with those who did respond and with healthy controls. These findings:
support a shift in therapeutic strategy for non-responsive patients.
Strengths and limitations
Strengths
This study is the largest meta-analyses of 1H-MRS antipsychotic response studies to date. A key strength is its use of a mega-analysis, which provides a large sample size and individual-level data, increasing precision and allowing identification of hidden patterns.
As the authors analysed individual-level data rather than published summary statistics, they were able to apply consistent inclusion criteria, outcome definitions, and statistical models across cohorts. This reduces some of the heterogeneity that affects conventional meta-analyses. When mega-analyses were previously compared to meta-analyses, mega-analysis showed lower standard errors and narrower confidence intervals (Boedhoe P S W. et al., 2019).
Additionally, the authors used only prospectively reported treatment-response data, as they examined baseline neuro-metabolites in relation to subsequent antipsychotic treatment response. This strengthens the temporal relationship.
Limitations
While the authors used a comprehensive search strategy, they only searched one database (Web of Science). This can increase the risk of missing relevant studies, which can introduce selection bias and reduce the completeness of the evidence base. It also increases the likelihood of publication bias, as different databases cover different journals, regions, and disciplines, so relying on one source may over-represent certain types of research.
As acknowledged by the authors, a key limitation is that the effect sizes for group differences were in the small-to-moderate range, despite showing an association between neuro-metabolite differences in those who responded to antipsychotics and those who did not. The issue with small effect sizes is that the findings might have limited clinical or practical significance, and the real-world benefit for an individual patient may be small.
The study examined treatment response across multiple cohorts, but treatment was not standardised. Participants likely differed in the specific antipsychotic and dose prescribed, as well as in the duration of treatment and adherence. These factors could affect treatment response independently of baseline neuro-metabolite levels.
Although the authors adjusted for key demographic and study-level factors, they did not adjust for potentially important metabolic and lifestyle confounders such as BMI and smoking status. As these factors may influence neuro-metabolite concentrations and differ between treatment-response groups, residual confounding remains possible.
Additionally, the study included a single measurement of neuro-metabolites at baseline only. It is unknown whether metabolite levels changed during treatment, or if repeated measurements could improve prediction.
Implications for practice
The article illustrates that there are differences in some neuro-metabolites between people with schizophrenia who do not respond to antipsychotics, compared to those who respond to treatment. The findings have some important clinical implications.
Stratification by biological profiles
The metabolite differences identified by King and colleagues provide further evidence that treatment-resistant schizophrenia is biologically heterogeneous. Identifying potential biomarkers, such as alterations in brain neuro-metabolites, may help identify biologically meaningful subgroups of people with schizophrenia who are more or less likely to respond to antipsychotic treatment. Approximately one third of patients with schizophrenia meet criteria for treatment resistance (Enache D. et al. 2022), highlighting the need for more personalised approaches to treatment.
The idea of stratifying patients by biological profiles is gaining interest. A recent study by my colleagues and I (Murphy J. et al., 2025) identified latent profiles of inflammation, with one distinct group showing heightened levels of three inflammatory markers. Similarly, Byrne J. et al. (2022) identified and characterised trans-diagnostic inflammatory subgroups across psychiatric disorders. The study found evidence of a novel pattern of inflammatory markers specific to psychiatric disorders, including psychotic disorder, depressive disorder and generalised anxiety disorder (GAD), where participants in the cluster exhibiting higher inflammation were less likely to be in employment, education or training.
Together, these findings support the idea that integrating biological markers, including neuro-metabolite and inflammatory profiles, may help identify subgroups with different treatment trajectories and guide more targeted interventions. However, further validation is needed before these approaches can be translated into clinical practice.
Innovative treatment alternatives
This study by King and colleagues (2026) found small, but consistent alterations in medial frontal brain metabolites associated with non-response to antipsychotic treatment, suggesting that biological differences may contribute to why some people respond to treatment while others do not. These findings support further investigation into biological mechanisms beyond conventional dopaminergic models of schizophrenia. Some of these mechanisms have already been proposed and include altered inflammatory processes (Enache D. et al., 2022), illness chronicity, and structural brain abnormalities (Birur B. et al., 2017).
However, recovery-oriented approaches often extend beyond biological explanations. The patient is a person, not a disease, and understanding sustained functioning, quality of life, and long-term recovery requires attention to individual experiences, as well as neurobiology. For example, Kamitis and colleagues (2022) reported that some people with psychosis and childhood trauma experienced intensified trauma-related flashbacks, thoughts, and physical symptoms while taking antipsychotic medication, leading to issues with adherence. Thus, rather than viewing treatment resistance as a single biological entity, researchers may need to consider multiple interacting mechanisms that contribute to poor treatment response.
Ultimately, improving outcomes for treatment-resistant schizophrenia will likely require approaches that integrate emerging biological insights, such as those identified by King and colleagues, with a person-centred understanding of the psychological and social factors that shape recovery.
Statement of interests
Jennifer Murphy has no conflict of interests to declare.
Editor
Edited by Éimear Foley. ChatGPT assisted with language refinement and formatting during the editorial phase.
Links
Primary paper
Bridget King, Kirsten Borup Bojesen, Charlotte Crisp, Andrea de Bartolomeis,… Alice Egerton et al. (2026) Neurometabolites and antipsychotic response in psychosis: a mega-analysis. JAMA Psychiatry. 2026 Jul 1:e261674. doi:10.1001/jamapsychiatry.2026.1674
Other references
Birur B, Kraguljac NV, Shelton RC, et al. Brain structure, function, and neurochemistry in schizophrenia and bipolar disorder-a systematic review of the magnetic resonance neuroimaging literature. NPJ Schizophr. 2017 Apr 3;3:15.
Boedhoe PSW, Heymans MW, Schmaal L, et al. An empirical comparison of meta- and mega-analysis with data from the ENIGMA Obsessive-Compulsive Disorder Working Group. Frontiers in Neuroinform. 2019;12:102.
Enache D, Nikkheslat N, Fathalla D, et al. Peripheral immune markers and antipsychotic non-response in psychosis. Schizophrenia research, 2021, 230, 1–8.
Kraguljac NV, Reid M, White D, et al. Neurometabolites in schizophrenia and bipolar disorder – a systematic review and meta-analysis. (PDF) Psychiatry Res. 2012 Aug-Sep;203(2-3):111-25.
Lubos E. Antipsychotics for acute treatment of first episode schizophrenia. The Mental Elf. 25 September 2017.
Merritt K, McGuire PK, Egerton A; et al. Association of Age, Antipsychotic Medication, and Symptom Severity in Schizophrenia With Proton Magnetic Resonance Spectroscopy Brain Glutamate Level: A Mega-analysis of Individual Participant-Level Data. JAMA Psychiatry. 2021 Jun 1;78(6):667-681.
Nakahara T, Tsugawa S, Noda Y, et al. Glutamatergic and GABAergic metabolite levels in schizophrenia-spectrum disorders: a meta-analysis of 1H-magnetic resonance spectroscopy studies. Mol Psychiatry. 2022 Jan;27(1):744-757. [PubMed abstract]
Norman L J. & Shaw P. Harnessing mega-analysis in the era of “big data” neuroimaging. Neuropsychopharmacology 2024; 50(1), 332-334.
Sud D. Risperidone and aripiprazole: genotype, metabolism and dosage. The Mental Elf. 11 March 2020.
