Dynamic fcMRI Reveals Transient Atypical Connectivity Patterns in Adolescents with Autism Spectrum Disorders

Background: There is ample evidence of atypical static functional connectivity (FC) in autism spectrum disorders (ASDs). However, transient relationships between neural networks are not easily captured by conventional functional connectivity magnetic resonance imaging (fcMRI) methods. Dynamic FC approaches have been used to identify repeated, transient connectivity patterns (“brain states”), revealing spatiotemporal network properties that are not observable in static FC. Several recent studies using this method have found atypical dynamic FC in ASDs (Chen et al., 2017; de Lacy et al., 2017). However, the exact relationship between static and dynamic FC in ASDs remains unclear, as does the characterization of state-specific group differences.

Objectives: We aimed to (i) establish the relationship between static and dynamic FC in typical development (TD) and ASDs; (ii) describe group differences in transient states and compare these to static FC patterns; and (iii) examine temporal stability and flexibility between identified states.

Methods: Resting-state functional MRI data from 62 ASD and 57 TD participants were included, group-matched for motion, age, sex, handedness, and nonverbal cognitive ability. High-order group independent component analysis and dual-regression were used to generate maximally independent, functionally cohesive network regions of interest (ROIs) at the individual level. Static FC was calculated using entire time series. Dynamic FC between ROI pairs was examined using tapered sliding windows, generating a temporally contiguous series of connectivity matrices. For each subject and ROI pairing, the standard deviation of FC (sd-FC) was calculated across all time windows. Windows underwent k-means clustering, generating four analogous brain states in each group.

Results: A one-sample t-test determined that across all ROI pairings, group differences in static FC primarily reflected hyperconnectivity in ASD (t(1080) = 12.06, p <.001); additionally, increased sd-FC in the ASD group predominated (t(1080) = 9.72, p < .001). There was a significant negative relationship between static FC and sd-FC in both the ASD (r = -.34, p <.001) and TD groups (r = -.29, p < .001). However, the strength of this association at the individual level was predicted by diagnosis in a general linear model, being weaker in individuals with ASD (β = .09, p =.04). While static FC group differences did not survive FDR correction, two dynamic states showed widespread significant group differences. In one of these states, relative within-network hypoconnectivity and between-network hyperconnectivity were observed for both default-mode and sensorimotor ROIs. This state was more likely to occur in participants with ASD (χ²= 11.99, p < .001). No significant group differences were observed for time spent per state, mean dwell time, or state-switching frequency.

Conclusions: Higher static FC was associated with reduced variability over time, but this relationship was weaker in ASD participants. Group connectivity differences emerged in dynamic FC states that were not observed in the static FC analysis. Furthermore, findings in one state, which occurred more frequently in ASD participants, were consistent with previous evidence of reduced network integration and differentiation in ASDs. These results highlight the importance of dynamic approaches, and may explain discrepant findings in the static FC literature.