Proteomic Analysis of Plasma Markers in Patients Maintained on Antipsychotics: Comparison to Patients Off Antipsychotics and Normal Controls

Abstract

Background: Schizophrenia (SZ) and bipolar disorder (BD) share many features: overlap in mood and psychotic symptoms, common genetic predisposition, treatment with antipsychotics (APs), and similar metabolic comorbidities. The pathophysiology of both is still not well defined, and no biomarkers can be used clinically for diagnosis and management. This study aimed to assess the plasma proteomics profile of patients with SZ and BD maintained on APs compared to those who had been off APs for 6 months and to healthy controls (HCs).

Methods: We analyzed the data using functional enrichment, random forest modeling to identify potential biomarkers, and multivariate regression for the associations with metabolic abnormalities.

Results: We identified several proteins known to play roles in the differentiation of the nervous system like NTRK2, CNTN1, ROBO2, and PLXNC1, which were downregulated in AP-free SZ and BD patients but were "normalized" in those on APs. Other proteins (like NCAM1 and TNFRSF17) were "normal" in AP-free patients but downregulated in patients on APs, suggesting that these changes are related to medication's effects. We found significant enrichment of proteins involved in neuronal plasticity, mainly in SZ patients on APs. Most of the proteins associated with metabolic abnormalities were more related to APs use than having SZ or BD. The biomarkers identification showed specific and sensitive results for schizophrenia, where two proteins (PRL and MRC2) produced adequate results.

Conclusions: Our results confirmed the utility of blood samples to identify protein signatures and mechanisms involved in the pathophysiology and treatment of SZ and BD.

Keywords: antipsychotics; biomarkers; bipolar disorder; metabolic syndrome; proteomics; schizophrenia.

Figure 1

Circulating proteome profiles in SZ and BD patients (+/− AP). (A) Number of significantly regulated plasma proteins (post-hoc FDR < 0.05) in SZ and BD patients (do not receive APs for at least 6 months) and patients taking APs. (B) Heatmap is showing standardized log₂ protein level ratios for most significant proteins (post-hoc FDR < 0.01) of SZ and BD patients (+/-AP) compared to HC. Four major clusters I-IV revealed using complete Euclidean distance clustering are labeled. (C) Distributions of protein level ratios of significantly changing proteins (ANCOVA P < 0.01) across comparisons of conditions and treatment to HCs. (D) Statistically over-represented IPA disease annotation terms within significantly down- or upregulated groups of proteins in SZ and BD patients (+/− AP). Statistical over-representation of terms is expressed as -log₁₀ P-value derived from a Fisher's exact test.

Figure 2

ROC curves showing the diagnostic performance in discriminating SZ and BD patients from HC. Classification between MD patients and HC was performed using 2, 5, or 10 proteins. Proteins for classification were selected from a list of the most significantly regulated proteins identified in this study, followed by training a random forest prediction model to choose the most robust protein classifiers. ROC curves and classification model performance metrics using selected protein classifiers, as shown in the plot, were established using 10-fold cross-validation.

Figure 3

Clinical variables and their associations with differentially abundant proteins. (A) Boxplot indicating the distribution of BMI, waist circumference, systolic, and diastolic blood pressure across conditions. Indicated are ANCOVA P-values and post-hoc P-values for group comparisons correcting for gender and age. (B) Heatmap shows the significance of association for differentially abundant proteins with clinical variables. The plot shows the significance of differential abundances comparing BD and SZ to HC, the significance of these proteins for the association with a clinical variable, and the significance of interaction with the MD group (BD and SZ combined). (C) Boxplots for proteins were found to have a significant interaction term with a clinical variable.