50. IDENTIFICATION OF DEPRESSION-RELATED METABOLIC PROFILE AND ITS ASSOCIATIONS WITH PSYCHIATRIC SYMPTOMS AND POLYGENIC SCORES

• Published in: European neuropsychopharmacology 2023-10, Vol.75, p.S82-S83
• Main Authors: Kurvits, Siim, Võsa, Urmo, Ojalo, Triinu, Esko, Tõnu, Milani, Lili, Haller, Toomas, Lehto, Kelli
• Format: Article
• Language: English
• ISSN: 0924-977X; 1873-7862
• DOI: 10.1016/j.euroneuro.2023.08.156

Although there are no established biomarkers for depression currently in clinical use, biobanks with electronic health record (EHR) linkages now enable conducting large-scale metabolome-wide studies to identify blood metabolites associated with depression. However, due to the substantial comorbidity in the clinical representation of depression, a more granular approach is needed to identify biomarkers associated with specific depressive symptom clusters. The aim of this study is to utilize longitudinal EHR data combined with questionnaire-based self-reported lifestyle and psychosocial factors (LPSF) to discover associations between depression and blood plasma metabolites determined by NMR. Additionally, associations between psychiatric-related polygenic risk scores (PRS) and metabolites are explored. Finally, the prediction power of the metabolites is measured by training and evaluating a machine learning model. We used the Estonian Biobank data source (N=200,000), which is linked to the national EHR databases, covering years 2004-2021. After filtering, 227 blood metabolites measured using the NMR Nightingale panel were used for association testing. Case-control metabolome-wide association study was conducted for the ICD-10 based depression (codes F32 and F33). As the majority of depression cases in Estonian Biobank are female and metabolic profile is influenced by sex the analysis is done separately for females and males. All results given in the abstract are for the female group. LPSF were assessed with a comprehensive Mental Health Online Survey (N=86,000). To identify clusters of associations between metabolites and LPSF (N=100) / PRS (N=36), Hierarchical All-against-All Association Testing (HAllA) was implemented for clustering. Random Forest (RF) binary classification model was trained for predicting ICD-10-based depression using only metabolite measures and age (train test split 90:10). We identified 7303 individuals with depression diagnosed no more than 6 months before or after the blood sample was collected (mean age 47.5 y) and matched 4 controls for every case with no psychiatric diagnoses and same age. We identified 89 metabolites associated with depression (p-value adjusted for multiple testing). The depression diagnosis was associated with cholesterol, lipoproteins, fatty and amino acids measures. We detected 569 statistically significant clusters of LPSF and metabolite measures (FDR 0.2), the top one being alcohol and tobacco usage habits w