Motor Cortex Functional Connectivity Signatures of Autism

Background: Motor impairments experienced by children with autism spectrum disorder (ASD) relate to the communicative/social deficits at the core of their diagnosis and may reflect abnormal connectivity within brain networks underlying motor control and learning. Resting state (rs) functional connectivity (FC) analysis is a potentially powerful tool to estimate brain organization within clinical populations like ASD but also poses challenges for quantitative image analysis, mainly related to the comparison of noisy signals from a large number of sources. Several groups have proposed parceling the brain prior to FC analysis to reduce the dimensionality of the data and to enable rapid calculation of inter-parcel FC signatures for individuals.

Objectives:  Motivated by these potentially scalable methods to investigate brain organization, the aim of this study was to estimate how well FC between subregions of the motor cortex (M1) discriminate individuals with ASD from typically developing (TD) participants.

Methods:  rs-fMRI and anatomical images from the Autism Brain Imaging Data Exchange were used (368 ASD and 412 TD). Included participants were male, 6 to 40 years old and had a mean framewise displacement (between-volume motion) within two standard deviations of the sample mean. Data were adjusted for slice acquisition order and participant motion and normalized to MNI space using unified segmentation (SPM8). Nuisance covariates from white matter and CSF were estimated using CompCor and regressed from the data along with motion parameters, their derivatives, and global mean signal. Data were band-pass filtered (.01-0.1 Hz) and spatially smoothed (6-mm kernel).

The five-region M1 parcellation used to estimate FC signatures for each subject was derived from test-retest rs data from 20 TD adults and reflects the general organization of the motor homunculus. For each subject, correlations between the 10 pairs of mean parcel time courses were computed. Group differences were assessed using a multinomial logistic regression model. Demographic factors and M1 correlations were used as predictors, disease status as the outcome. To account for possible confounds among the many variables, generalized boosted methods were used to estimate model parameters. The spatial correlation between each subject’s normalized data and SPM’s EPI template was also included in the model to account for variability in the consistency of spatial normalization across subjects.

Results:  Preliminary analysis suggests that IQ had a high relative influence in predicting disease status (35%) when all demographic variables and M1 parcel correlations were included. The correlation of the dorsomedial-most (DM) region, normally reserved for lower limb/trunk control, and the posterior lateral (PL) region (near the hand area) had the second highest relative influence in the prediction model (24.6%). The third most influential factor also involved PL FC, but with the dorsolateral (DL)/upper limb region.

Conclusions:  We identified potentially predictive FC signatures of ASD. FC disruptions between the DM/DL regions and the brain outside of M1 have been previously implicated in ASD. Here, we showed that FC between these regions and other parts of M1 (PL) may also be abnormal. These FC differences are consistent with deficits in complex multi-joint coordination associated with ASD.