SU38 - A STUDY OF 42 INFLAMMATORY MARKERS IN 321 CONTROL SUBJECTS AND 887 MAJOR DEPRESSIVE DISORDER CASES: THE ROLE OF BMI AND OTHER CONFOUNDERS, AND THE PREDICTION OF CURRENT DEPRESSIVE EPISODE BY MACHINE LEARNING

• Published in: European neuropsychopharmacology Vol. 29; p. S908
• Main Authors: Powell, Timothy, Gaspar, Helena Alexandra, Chung, Raymond, Keohane, Aoife, Gunasinghe, Cerisse, Uher, Rudolf, GENDEP Study Team, SELCoH Study Team, Collier, David, Wang, Hong, Breen, Gerome
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 2019
• ISSN: 0924-977X; 1873-7862
• DOI: 10.1016/j.euroneuro.2017.08.227

Inflammatory markers, such as circulating cytokines, influence neurotransmitter systems and brain functionality related to psychiatric disease pathology. Previous studies have revealed heightened circulating levels of pro-inflammatory cytokines in the blood of Major Depressive Disorder (MDD) patients, which suggests they may have utility as clinically informative biomarkers to aid in the diagnosis of MDD. However, most previous studies have only focused on a small number of inflammatory markers (e.g. C-reactive protein, interleukin-6) and most have failed to co-vary for potentially important confounding factors. We assessed 42 inflammatory markers in the blood (serum) of 321 control subjects and 887 MDD cases using multiplex electrochemiluminescence methods. We tested whether individual inflammatory marker levels were significantly affected by MDD case/control status, current episode, or current depression severity, co-varying for age, sex, Body Mass Index (BMI), smoking, current antidepressant use, ethnicity, assay batch and study effects. We further used machine learning algorithms to investigate if we could use our data to blindly diagnose MDD patients or discriminate those in a current episode. We used the false discovery rate (q<0.1) to account for multiple testing. We found broad and powerful influences of confounding factors on log-protein levels. Notably, IL-6 levels were very strongly influenced by Body Mass Index (BMI; p=1.37×10–43, variance explained=18%), while Interleukin-16 was the most significant predictor of current depressive episode (p=0.003, variance explained=0.9%, q < 0.1). No single inflammatory marker predicted MDD case/control status when a subject was not in a depressed episode, nor did any predict depression severity. Machine learning results revealed that using inflammatory marker data with clinical confounder information significantly increased the precision of blind diagnoses when patients were in episode. To conclude, a wide panel of inflammatory markers alongside clinical information may aid in predicting the onset of symptoms via a machine learning approach, but no single inflammatory proteins are likely to represent clinically useful biomarkers for MDD diagnosis or prognosis. Our study also highlights the need for confounding factors, particularly BMI, to be considered in all future studies pertaining to inflammatory markers.