Analysis of Temporal Dynamics of Brain Functional Connectivity in Autism

Background: There is increasing evidence that autism spectrum disorder (ASD) is associated with disruptions in the excitatory/inhibitory balance of neural activity leading to abnormal functional connectivity of the brain, thereby causing inefficient information processing. Examination of oscillatory resting-state MEG activity within specific frequency bands affords insight into these brain abnormalities. Amongst these frequency bands, recent studies have demonstrated delta-band (1–4 Hz) connectivity alterations in ASD, with the connectivity being computed over large time intervals. However, analysis of resting-state brain connectivity dynamics in short-time windows may provide a greater insight into the differential effect of ASD pathology.

Objectives: The aim of this work is to investigate difference in temporal connectivity dynamics of the resting-state MEG source-space signal, between ASD and typically developing control (TDC), with connectivity computed in short-time windows in the delta band. Connectivity is characterized using network measures of modularity, clustering coefficient, long/short-range connectivity ratio and connectivity entropy.

Methods: Resting-state MEG eyes-closed data was collected on 27 ASD (aged 6.4 – 13.6 years) and 27 TDC (aged 6.1 – 13.9 years) age-matched children (p>0.8) using a 306-channel Elekta machine. The data was band-pass filtered to the delta band and mapped into the frequency domain using Fourier transforms. Sensor-to-source space mapping was done using VESTAL to obtain densely sampled source time-courses. 100-sec source signals, obtained after artifact removal, were segmented into 10-sec courses with 5-sec overlap, yielding 19 segments. Functional connectivity at each segment was quantified using time-frequency synchronization likelihood (SL) yielding a symmetric matrix. Network measures of modularity, clustering coefficient, long/short-range connectivity ratio and Von Neumann matrix entropy were computed for these 19 matrices per subject. Group differences were analyzed via t-tests at each of 19 time instants.

Results: The temporal connectivity dynamics showed differences between ASD and TDCs. Specifically, increased modularity in ASD versus TDC was observed in over 70% of time instances (p<0.05). In contrast, the clustering coefficient was consistently higher in TDC with high significance in one third of time instances (p<0.05). The ratio of long- over short-range connections was higher in TDC, with significance in over half of the time instances (p<0.05). The connectivity matrix entropy had a greater differentiating power compared to traditional global network measures, with ASD displaying higher entropy in 100% of time instances (p<0.001).

Conclusions: This study of temporal connectivity dynamics of resting-state MEG data demonstrated increased network modularity and decreased clustering coefficient in ASD. This is indicative of ASD being inconsistent with the small-world network structure. The decrease in long/short-range connectivity ratio in ASD demonstrates that ASD suffers from an imbalance in the proportion of long- and short-range connections, with short-range connectivity dominating. Finally, the increased matrix entropy in ASD may be indicative of a disorganized brain network, which perhaps leads to less regulation of neural activity. This study highlights the need for quantifying the temporally changing connectivity patterns of MEG data, so that the subtle changes induced by pathology can be appropriately characterized.