Data-driven parcellation and graph theory analyses to study adolescent mood and anxiety symptoms

Abstract

Abstract Adolescence is a period of rapid brain development when psychiatric symptoms often first emerge. Studying adolescents may therefore facilitate the identification of neural alterations early in the course of psychiatric conditions. Here, we sought to utilize new, high-quality brain parcellations and data-driven graph theory approaches to characterize associations between resting-state networks and the severity of depression, anxiety, and anhedonia symptoms—salient features across psychiatric conditions. As reward circuitry matures considerably during adolescence, we examined both Whole Brain and three task-derived reward networks. Subjects were 87 psychotropic-medication-free adolescents (age = 12–20) with diverse psychiatric conditions (n = 68) and healthy controls (n = 19). All completed diagnostic interviews, dimensional clinical assessments, and 3T resting-state fMRI (10 min/2.3 mm/TR = 1 s). Following high-quality Human Connectome Project-style preprocessing, multimodal surface matching (MSMAll) alignment, and parcellation via the Cole-Anticevic Brain-wide Network Partition, weighted graph theoretical metrics (Strength Centrality = CStr; Eigenvector Centrality = CEig; Local Efficiency = ELoc) were estimated within each network. Associations with symptom severity and clinical status were assessed non-parametrically (two-tailed p FWE < 0.05). Across subjects, depression scores correlated with ventral striatum CStr within the Reward Attainment network, while anticipatory anhedonia correlated with CStr and ELoc in the subgenual anterior cingulate, dorsal anterior cingulate, orbitofrontal cortex, caudate, and ventral striatum across multiple networks. Group differences and associations with anxiety were not detected. Using detailed functional and clinical measures, we found that adolescent depression and anhedonia involve increased influence and communication efficiency in prefrontal and limbic reward areas. Resting-state network properties thus reflect positive valence system anomalies related to discrete reward sub-systems and processing phases early in the course of illness.