Deficits in information processing and cognition are among the most robust findings in schizophrenia patients. Previous efforts to translate group-level deficits into clinically relevant and individualized information have, however, been non-successful, which is possibly explained by biologically different disease subgroups. We applied machine learning algorithms on measures of electrophysiology and cognition to identify potential subgroups of schizophrenia. Next, we explored subgroup differences regarding treatment response. Sixty-six antipsychotic-naive first-episode schizophrenia patients and sixty-five healthy controls underwent extensive electrophysiological and neurocognitive test batteries. Patients were assessed on the Positive and Negative Syndrome Scale (PANSS) before and after 6 weeks of monotherapy with the relatively selective D2 receptor antagonist, amisulpride (280.3±159 mg per day). A reduced principal component space based on 19 electrophysiological variables and 26 cognitive variables was used as input for a Gaussian mixture model to identify subgroups of patients. With support vector machines, we explored the relation between PANSS subscores and the identified subgroups. We identified two statistically distinct subgroups of patients. We found no significant baseline psychopathological differences between these subgroups, but the effect of treatment in the groups was predicted with an accuracy of 74.3% (P=0.003). In conclusion, electrophysiology and cognition data may be used to classify subgroups of schizophrenia patients. The two distinct subgroups, which we identified, were psychopathologically inseparable before treatment, yet their response to dopaminergic blockade was predicted with significant accuracy. This proof of principle encourages further endeavors to apply data-driven, multivariate and multimodal models to facilitate progress from symptom-based psychiatry toward individualized treatment regimens.

Research findings implicate cerebral glutamate in the pathophysiology of schizophrenia, including genetic studies reporting associations with glutamatergic neurotransmission. The extent to which aberrant glutamate levels can be explained by genetic factors is unknown, and if glutamate can serve as a marker of genetic susceptibility for schizophrenia remains to be established. We investigated the heritability of cerebral glutamate levels and whether a potential association with schizophrenia spectrum disorders could be explained by genetic factors. Twenty-three monozygotic (MZ) and 20 dizygotic (DZ) proband pairs con-or discordant for schizophrenia spectrum disorders, along with healthy control pairs (MZ = 28, DZ = 18) were recruited via the National Danish Twin Register and the Psychiatric Central Register (17 additional twins were scanned without their siblings). Glutamate levels in the left thalamus and the anterior cingulate cortex (ACC) were measured using 1[H]-magnetic resonance spectroscopy at 3 Tesla and analyzed by structural equation modeling. Glutamate levels in the left thalamus were heritable and positively correlated with liability for schizophrenia spectrum disorders (phenotypic correlation, 0.16, [0.02-0.29]; p = 0.010). The correlation was explained by common genes influencing both the levels of glutamate and liability for schizophrenia spectrum disorders. In the ACC, glutamate and glx levels were heritable, but not correlated to disease liability. Increases in thalamic glutamate levels found in schizophrenia spectrum disorders are explained by genetic influences related to the disease, and as such the measure could be a potential marker of genetic susceptibility, useful in early detection and stratification of patients with psychosis.