Low-dimensional brain-symptom associations delineate depression phenotypes with distinct connectivity biomarkers and symptom profiles

Abstract

Depression is neurobiologically and clinically heterogeneous. New approaches using resting-state functional MRI (rs-fMRI) functional connectivity (FC) data have modeled the neural basis of depression heterogeneity and revealed unique neural phenotypes. Yet, no studies have identified depression phenotypes from electrophysiological magnetoencephalography (MEG) data although MEG measures human brain dynamics at millisecond precision. We demonstrate here unique depression phenotypes based on MEG-oscillation FC. We collected resting-state MEG, MRI, and clinical symptom data from 263 patients with unipolar depression and 75 healthy controls. We assessed MEG-FC with two oscillatory coupling-mode measures that are fundamental for information processing. To define normative phenotypes, we computed their latent-space low-dimensional brain-symptom associations, and used these components to identify phenotypes using unsupervised machine learning. We identified five stable depression phenotypes that were characterized by unique symptom profiles and distinct spectral patterns. Our results demonstrate new neural underpinnings of depression heterogeneity and reveal unique neural phenotypes with potential personalized diagnostic value.