A mega-analysis of functional connectivity and network changes in youth major depression

554 

Abstract Submission 

Nga Yan Tse¹, Aswin Ratheesh², Robin Cash¹, Andrew Zalesky¹

¹Department of Psychiatry, The University of Melbourne, Carlton, Victoria, ²Orygen, Parkville, Victoria

Nga Yan Tse  
Department of Psychiatry, The University of Melbourne
Carlton, Victoria

Aswin Ratheesh  
Orygen
Parkville, Victoria

Robin Cash  
Department of Psychiatry, The University of Melbourne
Carlton, Victoria

Andrew Zalesky, PhD  
Department of Psychiatry, The University of Melbourne
Carlton, Victoria

Major depressive disorder (MDD) has a lifetime prevalence of 11.1-14.6% [1,2] and represents the leading cause of disability due to mental health conditions for young people aged 10-24 years worldwide [3,4]. Functional neuroimaging can delineate the neural substrates of psychiatric, cognitive, and neurological disorders and potentially provide targets for treatment [5-8]. Youth MDD research however lag behind that in adults where existing resting-state functional MRI (rs-fMRI) studies have yielded inconsistent findings [9]. Further, mega-analysis, involving compilation of independent cohort datasets and offering unprecedented advantages of improved generalizability and increased statistical power, has remained unexplored in youth MDD.

Here, we conducted the first mega-analysis of functional connectivity (FC) changes in youth MDD, encompassing 810 raw, unprocessed T1 and rs-fMRI images collated from 7 international sites (n=440 youths with MDD and 370 healthy controls aged between 12-25 years). Standardized fMRIprep pre-processing, quality control, and COMBAT harmonization were completed for all fMRI data. Whole-brain FC (connectomes) were mapped for each individual based on the Schaefer Yeo 7-network 400 functional atlas. Using impartial, whole-brain-based statistical inference termed network-based statistic (NBS) [10], mega-analyses of between-group and symptom severity-related connectivity differences were conducted at the scale of functional connections and canonical networks. All NBS analyses were adjusted for age and sex and corrected for family-wise error at p<.05. Linear support vector machines with ridge regularization and leave-one-site-out cross validation were applied to build predictive models of diagnostic status and symptom severity (as measured by the Montgomery-Asberg Depression Rating Scale). Accuracy was tested on unseen datasets kept out of the model training to ensure robustness to inter-site variability. Supplementary analyses were conducted to ensure findings are not biased by global signal regression, atlas choice, or potential head motion differences.

Our analyses consistently implicated core nodes of the default mode network (DMN) in youth MDD, particularly the rostral/subgenual anterior cingulate, medial prefrontal cortex, posterior cingulate, and precuneus (Fig. 1-2). Altered connectivity of components of the limbic, and dorsal (DAN) and ventral attentional (VAN) networks also tended to emerge, localizing to the orbitofrontal cortex, insula, striatum, and intraparietal sulcus/superior parietal cortex (Fig. 1-2). Strikingly, these regions have been implicated as rich-club nodes in past literature and consistently demonstrated a higher level of hubness in our analyses, supporting extensive earlier studies reporting hub involvement in early psychopathology development.

Critically, individual variation in FC within these networks of regions was significantly associated with depression symptom severity (r=-.58 and r=.65 for hypo- and hyper-connected regions; both p<.001), supporting the clinical importance of these connectivity alterations. Consistently, our machine learning analysis further demonstrated that these FC features provided capacity to classify diagnostic status and depression severity in a generalizable and robust fashion beyond site-specific confounds with good accuracy (averaged cross-validated AUC=73% and r=.63).

Our data-driven, connectome-wide FC and machine learning analyses converge to implicate robust involvement of hub regions within the DMN, DAN, VAN, and limbic network. Adolescence, coinciding with a protracted period of significant plastic changes and psychosocial transitions, represents a unique window of increased vulnerability to altered hub development. This may in turn confer risks for altered network dynamics and discoordination of a myriad of processes centred on the attentional, affective, and introspective systems, and ultimately early emergence of youth MDD.

Neurodevelopmental/ Early Life (eg. ADHD, autism)

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ¹

Early life, Adolescence, Aging

Connectivity (eg. functional, effective, structural) ²

fMRI Connectivity and Network Modeling

DISORDERS

Emotions

FUNCTIONAL MRI

PEDIATRIC

Psychiatric

Psychiatric Disorders

Other - Major depressive disorder; functional connectivity; functional network; youth depression