Comparing network configurations of Schizophrenia and Autism Spectrum Disorders using CovSTATIS

1527 

Abstract Submission 

Ju-Chi Yu¹, colin hawco^1,2, Lindsay Oliver¹, Maria Secara^1,2, Iska Moxon-Emre¹, Fariah Sandhu³, Zara Khan⁴, Peter Szatmari^1,2, Meng-Chuan Lai^1,2, Miklos Argyelan⁵, James Gold⁶, Sunny Tang⁵, George Foussias^1,2, Robert Buchanan⁶, Anil Malhotra⁵, Aristotle Voineskos^1,2, Stephanie Ameis^1,2, Erin Dickie^1,2

¹Centre for Addiction and Mental Health, Toronto, Ontario, ²University of Toronto, Toronto, Ontario, Canada, ³York University, Toronto, Ontario, ⁴McMaster University, Hamilton, Ontario, ⁵Zucker Hillside Hospital, Glen Oaks, NY, ⁶Maryland Psychiatric Research Center, Baltimore, MD

Ju-Chi Yu  
Centre for Addiction and Mental Health
Toronto, Ontario

colin hawco, PhD  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Lindsay Oliver  
Centre for Addiction and Mental Health
Toronto, Ontario

Maria Secara  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Iska Moxon-Emre  
Centre for Addiction and Mental Health
Toronto, Ontario

Fariah Sandhu  
York University
Toronto, Ontario

Zara Khan  
McMaster University
Hamilton, Ontario

Peter Szatmari  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Meng-Chuan Lai  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Miklos Argyelan  
Zucker Hillside Hospital
Glen Oaks, NY

James Gold  
Maryland Psychiatric Research Center
Baltimore, MD

Sunny Tang  
Zucker Hillside Hospital
Glen Oaks, NY

George Foussias  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Robert Buchanan  
Maryland Psychiatric Research Center
Baltimore, MD

Anil Malhotra  
Zucker Hillside Hospital
Glen Oaks, NY

Aristotle Voineskos  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Stephanie Ameis  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Erin Dickie  
Centre for Addiction and Mental Health|University of Toronto
Toronto, Ontario|Toronto, Ontario, Canada

Schizophrenia Spectrum Disorder (SSD) and Autism Spectrum Disorder (autism) are both characterized by social cognitive deficits, of which the severity and impact on everyday life vary substantially within diagnoses, with overlap across diagnoses. To understand such heterogeneity, we aimed to identify group-specific and shared functional network configurations present during a social processing functional magnetic resonance imaging (fMRI) empathic accuracy (EA) task.

We included fMRI data during the EA social cognition task from 67 autism (35 females; mean age = 20.78, age range = [15, 33]), 174 SSD (121 females; mean age = 30.95, age range = [18, 54]), and 170 Healthy Controls (HC; 92 females; mean age = 31.47, age range = [17, 55]). Background functional connectivity of the EA task was extracted from parcellated connectomes (Ji et al., 2019) and analyzed by CovSTATIS (i.e., multi-table multidimensional scaling, or DiSTATIS, for covariance matrices) (Abdi et al., 2012). CovSTATIS is a multivariate method that analyzes multiple correlation matrices by creating a compromise that best captures the shared pattern. This compromise is a linear combination of all matrices, weighted by their similarity to the common pattern. This compromise is then decomposed by the singular value decomposition to extract latent dimensions characterizing prominent EA-associated network configurations. The factor score of each region of interest (ROI) illustrates the network configuration on each dimension with proximity indicating a close association. Mean factor scores of each network are then computed to examine network segregation. Furthermore, CovSTATIS projects individual connectivity matrices to compute partial factor scores, which illustrate each individual's network configuration in the same dimension space. These partial factor scores were averaged based on networks and participant groups to examine group differences in network segregation (See Figure 1 for details). Bootstrap procedures (1000 iterations) were used to estimate the 95% bootstrap confidence intervals of network and group means and were illustrated by ellipses in Figures 1-2. Bootstrap tests were used to examine configuration and group differences with two non-overlapping ellipses indicating a significant difference at α = .05.

From CovSTATIS, we considered all dimensions that contribute <1% of the variance being noise and the others as signal dimensions. Two orthogonal latent dimensions of EA task-associated activity were identified across all groups. The first dimension's configuration (explaining 14.98% of signal dimensions; x-axis in Figure 2A) was characterized by differentiation between language (LAN), and default mode networks versus visual (VIS), and dorsal attention networks (DAN). The second dimension's configuration (explaining 11.97% of the signal dimensions; y-axis in Figure 2A) was characterized by differentiation between auditory, and somatomotor versus frontoparietal networks. On group-wise comparison, HC showed significant differentiation between primary VIS and DAN (bootstrap p < .05; Figure 2B) that was absent in autism and SSD. Autism showed differentiation between LAN and ventral-multimodal network (Figure 2C), while HC and SSD groups did not (bootstrap p < .05). Finally, while HC and autism showed significant differentiation between primary and secondary VIS, SSD did not (bootstrap p > .05; Figure 2D).

In the EA task, prominent configurations that characterize the latent dimensions did not differ significantly across groups. Divergent patterns were found between all groups with specific configurations for autism group (in LAN) and SSD group (in VIS). The associations of these latent dimensions to cognition and clinical outcomes can help identify potential biomarkers for social cognitive deficits in autism and SSD.

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ²

Social Cognition

Connectivity (eg. functional, effective, structural) ¹

Multivariate Approaches

Autism

Cognition

Computational Neuroscience

FUNCTIONAL MRI

Informatics

Multivariate

Psychiatric Disorders

Schizophrenia

Other - functional connectivity