An exploratory data-driven stratification of depression based on functional network connectivity and symptom profiles...

The marked clinical and biological heterogeneity of major depressive disorder (MDD) suggests that it may be more appropriately conceptualized as a syndrome with multiple biologically relevant patterns. However, traditional stratification approaches based on either clinical symptomatology or isolated biological markers, including neuroimaging data, often fail to capture the complex brain-behavior relationships underlying MDD.

In this study, we applied regularized canonical correlation analysis to capture multivariate covariation between functional network connectivity (FNC) and symptom profiles. Based on the resulting low-dimensional representation, we performed clustering to explore potential stratification patterns in a cohort of 51 patients with MDD and examined whether these patterns were associated with variation in response to intermittent theta burst stimulation. Transcriptional signatures related to pattern-specific FNC abnormalities were further explored using partial least squares regression.

Our study found two stratification patterns in MDD. After false discovery rate correction, only increased connectivity between the ventral default mode network (vDMN) and cerebellar network in MDD2 relative to healthy controls remained significant. In exploratory uncorrected analyses, MDD1 was characterized by increased connectivity between the vDMN and salience network, as well as between the dorsal DMN (dDMN) and orbitofrontal cortex, together with greater depressive severity and more prominent appetite disturbance, whereas MDD2 showed broader connectivity increases and more prominent sleep disturbance. Imaging transcriptomic analysis provided complementary molecular contextualization. Treatment-related connectivity changes were mainly observed in MDD2, while patients within both stratification patterns showed symptom improvement, with appetite disturbance improving in MDD1 and sleep disturbance improving in MDD2. Finally, prediction analyses suggested potential prognostic relevance of baseline FNC features, particularly the connectivity between the dDMN and right executive control network.

By integrating neuroimaging, clinical symptomatology, and transcriptomic data, this study offers a preliminary, data-driven framework for exploring heterogeneity in MDD and its potential relevance to treatment sensitivity. These results may inform future efforts to refine biologically grounded stratification strategies for neuromodulation-based treatment prediction.

The clinical trial number is ChiCTR2100054793, and the registration date is December 27, 2021.