Search

Neuroimaging-based brain age is a biomarker that is generated by machine learning (ML) predictions. The brain age gap (BAG) is typically defined as the difference between the predicted brain age and chronological age. Studies have consistently reported a positive BAG in individuals with schizophrenia (SCZ). However, there is little understanding of which specific factors drive the ML-based brain age predictions, leading to limited biological interpretations of the BAG. We gathered data from three publicly available databases - COBRE, MCIC, and UCLA - and an additional dataset (TOPSY) of early-stage schizophrenia (82.5% untreated first-episode sample) and calculated brain age with pre-trained gradient-boosted trees. Then, we applied SHapley Additive Explanations (SHAP) to identify which brain features influence brain age predictions. We investigated the interaction between the SHAP score for each feature and group as a function of the BAG. These analyses identified total gray matter volume (group × SHAP interaction term β = 1.71 [0.53; 3.23]; pcorr

Cortical thinning is a well-known feature in schizophrenia. The considerable variation in the spatial distribution of thickness changes has been used to parse heterogeneity. A 'cortical impoverishment' subgroup with a generalized reduction in thickness has been reported. However, it is unclear if this subgroup is recoverable irrespective of illness stage, and if it relates to the glutamate hypothesis of schizophrenia.We applied hierarchical cluster analysis to cortical thickness data from magnetic resonance imaging scans of three datasets in different stages of psychosis (n = 288; 160 patients; 128 healthy controls) and studied the cognitive and symptom profiles of the observed subgroups. In one of the samples, we also studied the subgroup differences in 7-Tesla magnetic resonance spectroscopy glutamate concentration in the dorsal anterior cingulate cortex.Our consensus-based clustering procedure consistently produced 2 subgroups of participants. Patients accounted for 75%-100% of participants in one subgroup that was characterized by significantly lower cortical thickness. Both subgroups were equally symptomatic in clinically unstable stages, but cortical impoverishment indicated a higher symptom burden in a clinically stable sample and higher glutamate levels in the first-episode sample. There were no subgroup differences in cognitive and functional outcome profiles or antipsychotic exposure across all stages.Cortical thinning does not vary with functioning or cognitive impairment, but it is more prevalent among patients, especially those with glutamate excess in early stages and higher residual symptom burden at later stages, providing an important mechanistic clue to one of the several possible pathways to the illness.

BACKGROUND: Multiple clustering algorithms have been proposed to organize and thereby reduce the clinical, cognitive, neurobiological and functional heterogeneity in the schizophrenia spectrum. Cortical thickness (CT) is a key gene-mediated aspect of brain morphology that may reflect reduced synaptic structure and excessive pruning. It is also an aspect of neurobiology that correlates with the risk of developing schizophrenia. Previous heterogeneity-reducing strategies have included cluster analyses of CT values across the brain (Sugihara et al., 2017). However, cluster analytic approaches have not used regional values associated with the default mode network (DMN). Disruption or overreaction of DMN was found to be associated with psychopathology including schizophrenia (Buckner, Andrews-Hanna, & Schacter, 2008). METHODS: CT was measured with a 3T short bore General Electric MRI machine and the obtained scans were processed with FreeSurfer. Cortical parcellations were obtained for 19 DMN-related regions of interest according to Destrieux, Fischl, Dale and Halgren (2010) and the DMN literature. We carried out agglomerative hierarchical cluster analysis on these CT data obtained from 63 healthy controls and 74 patients with a diagnosis of DSM-IV schizophrenia or schizoaffective disorder. Ward’s method with Euclidean distance was applied and the NbClust function in R was used to determine the optimal number of clusters. Demographic, clinical (PANSS) and cognitive (MCCB) measures were also obtained. RESULTS: Thirteen out of 30 clustering validity indices indicated a 3-cluster solution. The proportion of patients and controls varied significantly across clusters, X2 (2, N = 137) = 12.635, p = 0.002. The first cluster (n1=19) was composed of largely of patients (84%). A large second cluster (n2=92) comprised 50 patients and 42 healthy control participants. In contrast, two-thirds of the third cluster (n3=26) comprised healthy controls. However, there were no significant cluster differences in the distribution of males and females. MANCOVA on all 19 DMN-related regions revealed a main effect of cluster F(38, 230) = 7.467, p