Metabolic profiling of CSF: evidence that early intervention may impact on disease progression and outcome in schizophrenia

Abstract

Background: The identification of schizophrenia biomarkers is a crucial step towards improving current diagnosis, developing new presymptomatic treatments, identifying high-risk individuals and disease subgroups, and assessing the efficacy of preventative interventions at a rate that is not currently possible.

Methods and findings: (1)H nuclear magnetic resonance spectroscopy in conjunction with computerized pattern recognition analysis were employed to investigate metabolic profiles of a total of 152 cerebrospinal fluid (CSF) samples from drug-naïve or minimally treated patients with first-onset paranoid schizophrenia (referred to as "schizophrenia" in the following text) and healthy controls. Partial least square discriminant analysis showed a highly significant separation of patients with first-onset schizophrenia away from healthy controls. Short-term treatment with antipsychotic medication resulted in a normalization of the disease signature in over half the patients, well before overt clinical improvement. No normalization was observed in patients in which treatment had not been initiated at first presentation, providing the first molecular evidence for the importance of early intervention for psychotic disorders. Furthermore, the alterations identified in drug-naïve patients could be validated in a test sample set achieving a sensitivity and specificity of 82% and 85%, respectively.

Conclusions: Our findings suggest brain-specific alterations in glucoregulatory processes in the CSF of drug-naïve patients with first-onset schizophrenia, implying that these abnormalities are intrinsic to the disease, rather than a side effect of antipsychotic medication. Short-term treatment with atypical antipsychotic medication resulted in a normalization of the CSF disease signature in half the patients well before a clinical improvement would be expected. Furthermore, our results suggest that the initiation of antipsychotic treatment during a first psychotic episode may influence treatment response and/or outcome.

Figure 1. Metabonomic Analysis of CSF Samples from Drug-Naïve Patients with Schizophrenia

(A) Partial ¹H NMR spectrum of a CSF sample from a representative drug-naïve patient with first-onset schizophrenia (red) and a matched control (black) illustrate a characteristic pH-dependent shift in the β-CH₂ and γ-CH₂ resonances of glutamine. The prominent signals at ~3.7 and 1.2 ppm correspond to ethanol, a contaminant from skin disinfection prior to lumbar puncture. These signals were removed from statistical analysis. (B) PLS-DA scores plot showing a differentiation of drug-naïve patients with schizophrenia from demographically matched controls as determined by the ¹H NMR CSF spectra. (C) PLS-DA loadings plot showing major contributing variables towards the separation in the PLS-DA scores plots.

Figure 2. Effects of “Typical” and “Atypical” Medication on CSF Metabolic Profiles in Patients with First-Onset Schizophrenia

(A) Spectra from 28 CSF samples from patients with first-onset schizophrenia minimally treated (<9 d, see text for details) with either typical (n = 6, blue diamonds) or atypical (n = 22, green circles) antipsychotic medications were compared to first onset, drug naïve patients (red triangles) and healthy volunteers (black squares) using PLS-DA. The PLS-DA scores plots show that atypical antipsychotic drug treatment resulted in a shift of approximately 50% of patients with schizophrenia towards the cluster of healthy controls. (B) The same PLS-DA scores plot as (A) except that only minimally treated patients (from both drug groups) with more than one psychotic episode prior to antipsychotic treatment are shown. Note that none of these patients shifted towards the healthy control cluster.

Figure 3. Validation and Prediction of Schizophrenia Group Membership Using a PLS Model

A PLS model was constructed using the OSC-filtered data from 37 drug-naïve patients with first-onset schizophrenia from the first cohort (red points) and 50 randomly selected healthy volunteers (blue points) (the “training set”). The scores plot (A) and the loadings plot (B) indicate the key resonances contributing to the separation: lactate, glucose, glutamine, and citrate. This model was then used to predict “group membership” (i.e., schizophrenia or control) in a test set of 17 drug-naïve patients (second cohort) with first-onset schizophrenia and the remaining 20 healthy volunteers which had not been used in the construction of the model. Predictions are made using a Y-predicted scatter plot with an a priori cut-off of 0.5 for class membership (C).