DNA methylation signature aberration as potential biomarkers in treatment-resistant schizophrenia: Constructing a methylation risk score using a machine learning method

Abstract

Treatment-resistant schizophrenia (TRS) is defined as a non-response to at least two trials of antipsychotic medication with an adequate dose and duration. We aimed to evaluate the discriminant abilities of DNA methylation probes and methylation risk score between treatment-resistant schizophrenia and non-treatment-resistant schizophrenia. This study recruited 96 schizophrenia patients (TRS and non-TRS) and 56 healthy controls (HC). Participants were divided into a discovery set and a validation set. In the discovery set, we conducted genome-wide methylation analysis (human MethylationEPIC 850K BeadChip) on the subject's blood DNA and discriminated significant methylation signatures, then verified these methylation signatures in the validation set. Based on genome-wide scans of TRS versus non-TRS, thirteen differentially methylated probes were identified at FDR <0.05 and >20% differences in DNA methylation β-values. Next, we selected six probes within gene coding regions (LOC404266, LOXL2, CERK, CHMP7, and SLC17A9) to conduct verification in the validation set using quantitative methylation-specific PCR (qMSP). These six methylation probes showed satisfactory discrimination between TRS patients and non-TRS patients, with an AUC ranging from 0.83 to 0.92, accuracy ranging from 77.8% to 87.3%, sensitivity ranging from 80% to 90%, and specificity ranging from 65.6% to 85%. This methylation risk score model showed satisfactory discrimination between TRS patients and non-TRS patients, with an accuracy of 88.3%. These findings support that methylation signatures may be used as an indicator of TRS vulnerability and provide a model for the clinical use of methylation to identify TRS.

Keywords: DNA methylation; Epigenetics; Methylation risk score; Schizophrenia; Treatment-resistant.