DISEASE PROGRESSION MODELING OF BRAIN MACROSTRUCTURE IN TREATMENT-RESISTANT DEPRESSION

Background

Progressive abnormalities in brain structures have consistently been reported in depression. However, the distinct trajectories of progressive structural abnormalities in TRD remain unknown. Moreover, treatment-resistant depression (TRD) is suggested to have a heterogeneous pathophysiology. The Subtype and Stage Inference (SuStaIn) algorithm, an unsupervised machine learning technique, has shown premise in distinguishing disease biotypes with different progression trajectories.

Aims & Objectives

This study aimed to identify patterns of structural abnormalities in TRD with SuStaIn and associate each of them with clinical characteristics.

Method

The study received approval from the ethical committee at Keio University School of Medicine. All participants gave written informed consent. We included 129 adult patients with TRD (45.0±12.5 years old, 59 females [46%]) and 93 healthy controls (HC) (46.2±17.7 years old, 37 females [40%]).

Participants underwent magnetic resonance imaging (MRI) scans with a Siemens Prisma 3T MRI scanner. Cortical thickness and gray matter volume from 13 cortical regions and bilateral hippocampus reported by mega-analysis studies (Schmaal et al., 2017; Schmaal et al., 2016) were calculated using FreeSurfer 6.0 and then z-scored using HC data.We applied the SuStaIn algorithm with 10-fold cross-validation using pySuStaIn. The optimal number of subtypes was determined through cross validation information criterion and test set log likelihood across folds. Furthermore, we explored the relationship between SuStaIn output (subtype and stage) and clinico-demographic variables.

Results

The SuStaIn algorithm identified two subtypes with distinct progression patterns. Subtype 1 (medial frontal cortex (mFC) type) (n=65) exhibited mild cortical thinning in the bilateral medial orbitofrontal cortex (mOFC) followed by the rostral anterior cingulate cortex (rACC). Subtype 2 (hippocampus and posterior cingulate cortex (Hip &PCC) type) (n=21) showed a marked reduction in the bilateral hippocampal volume followed by mild cortical thinning in the left PCC.

Hip &PCC type had an earlier stage than mFC type (p<0.001). No significant group difference in scores of the Montgomery-Å sberg Depression Rating Scale (p=0.73), onset age (p=0.33), or duration of untreated depression (p=0.79) was observed. There was no correlation between stage and these clinical variables in the whole group. In Hip &PCC type, stage correlated with visuospatial constructional ability assessed by the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) (r=-0.57, p=0.007).

Discussion & Conclusion

Previous studies suggest that abnormalities in the mOFC and Hip are the starting points for progressive structural changes in depression. Our results indicate that they may separately emerge in the two distinct subtypes. Additionally, both subtypes revealed progressive structural abnormalities in the regions in the hippocampal-medial prefrontal cortex network, a brain network crucial for depression.

Hip &PCC type had an earlier stage than mFC type, suggesting less severe and more localized neurodegeneration in Hip &PCC type than mFC type. Only in Hip &PCC type, stage progression was linked to visuospatial dysfunction. Given the association between reduced Hip volume and cognitive impairment in MDD, the trajectories of structural changes might provide insights into the heterogeneity of cognitive impairment in TRD.