State-Dependent Reductions in Brain Network Modularity and Behavioral Inflexibility in Childhood Autism Spectrum Disorder

Background: Behavioral inflexibility, the inability to shift to a different thought or action following a change in a situation, is a hallmark of autism spectrum disorders (ASD). Emerging evidence in healthy adults suggests that the dynamic nature of the brain’s large-scale functional network architecture enables adaptive goal-oriented behaviors. With this principle in mind, we hypothesized that behavioral inflexibility in ASD is associated with an aberrant adaptation of brain networks during cognitive states requiring this ability. Graph theory metrics, such as modularity, may be particularly well-suited to capture these atypical patterns of whole-brain communication.

Objectives: Use a multi-state approach to test for state-dependent brain network modularity reductions in ASD children.

Methods: Seventeen children with ASD and age-, IQ-, gender-matched typically developing (TD) children performed three functional magnetic resonance imaging (fMRI) tasks. The first task consisted of passive fixation (i.e., resting-state). The second task required monitoring a central visual stream of shapes and responding to a target shape. The third task utilized visual stimuli that were identical to those used in the second but imposed behavioral flexibility demands through an alternative set of instructions. This manipulation allowed testing whether differences in brain network modularity between ASD and TD children are specific to a cognitive state requiring behavioral flexibility (Task 3) versus a perceptually invariant cognitive state not requiring behavioral flexibility (Task 2) or a resting-state (Task 1). For each participant, a whole-brain functional connectivity (FC) matrix was generated for each state using the regions-of-interest created by Power et al. (2011) with the effects of participant motion, physiological noise, and task structure regressed. FC matrices were decomposed into distinct modules by implementing the Newman–Girvan quality function Q. This function was optimized using the Louvain algorithm, as implemented in the brain connectivity toolbox (https://sites.google.com/site/bctnet/), and iterated 1000 times. Independent sample t-tests were applied to test for differences in modularity (i.e., Q) between ASD and TD children for each state. In addition, average ASD and TD FC matrices were entered into network visualization software to compare how network topologies, represented as spring graphs, reconfigured between states.

Results: Brain network modularity was reduced in ASD children relative to TD children [t(33) = 1.70, p<0.05] in the cognitive state requiring behaviorally flexibility (Task 3), but not in the cognitive state requiring no behavioral flexibility (Task 2) or the resting-state (Task1). In line with this finding, spring graphs revealed a striking pattern of cross-module signaling specific to Task 3 in the ASD group. 

Conclusions: Our results suggest that behavioral inflexibility in ASD may be associated with reductions in brain network modularity (i.e., less segregation of networks). However, this work has broader implications for ASD brain connectivity research, which often assumes the resting-state captures intrinsic brain network abnormalities. In contrast, our findings highlight the importance of understanding the influence of specific cognitive states on brain networks in ASD.