Abnormal gray matter structural covariance networks associated with suicidal ideation in adults with major depressive...

Suicidal ideation (SI) is common in patients with major depressive disorder (MDD) and may represent a clinically meaningful phenotype within MDD rather than merely a marker of more severe depressive symptoms. SI may involve disturbances in affective regulation, cognitive control, reward valuation, self-referential processing, and pain/social-threat processing. Although MDD with SI has been linked to gray matter abnormalities in anterior cingulate, prefrontal, limbic, and cerebellar regions, it remains unclear whether these abnormalities show SI-related morphometric covariance patterns-that is, coordinated interindividual variation in gray matter morphology-at the network level. To address this question, we compared gray matter structural covariance networks between MDD patients with and without SI.

Using data from the REST-meta-MDD project, we included 226 healthy controls and 226 patients with MDD. The MDD cohort comprised an SI group (MDD-SI, n = 113) and a non-SI group (MDD-NSI, n = 113). Depressive symptoms were assessed with the 17-item Hamilton Depression Rating Scale (HAMD-17), and depression severity excluding the suicide item (HAMD-16; HAMD-17 minus item 3) was additionally considered. Source-based morphometry (SBM) was used to construct group-level gray matter structural covariance networks and compare their loading coefficients across groups. Exploratory post hoc associations with SI scores and an exploratory internal machine-learning classification analysis were also examined.

In the analysis including all participants, relative to healthy controls, the MDD-SI group showed reduced expression of the anterior cingulate covariance network and cerebellar covariance network C, whereas the MDD-NSI group showed increased expression of cerebellar covariance network C and frontal covariance network G. Compared with the MDD-NSI group, the MDD-SI group showed reduced expression in the anterior cingulate covariance network, cerebellar covariance network C, frontal covariance network E, and frontal covariance network G. In an exploratory internal classification analysis, the stacking model achieved an AUC of 0.82 in the internal test subset. Because the cohort-level SBM/ICA representation was derived before the training-test split, this value should not be interpreted as fully independent out-of-sample validation.

This study provides network-level evidence that SI in MDD is associated with coordinated gray matter alterations involving anterior cingulate, cerebellar, and frontal covariance networks. The machine-learning findings are exploratory and require validation using scanner-balanced, fully independent cohorts and training-derived imaging representations.