Language Development and Brain Functional Connectivity in Infants and Toddlers at Risk for ASD

Background: Language delay, which is associated with familial ASD risk, frequently co-occurs with ASD and can be responsive to early ASD interventions. Characterizing brain-behavior relationships for aspects of language during the first two years of life, when core autistic symptoms arise, may therefore elucidate neurodevelopmental mechanisms of ASD.

Objectives: To test whether correlations between language function and ROI-ROI functional connectivity (fc) are enriched within specific brain networks and differ for ages 12 and 24 months.

Methods: Participants: The Infant Brain Imaging Study is a prospective study of infants at high-risk (HR: has ASD-affected sibling) and low-risk (LR: has unaffected sibling) of ASD. Included participants had fcMRI and behavioral data at 12 (n=130; HR+/HR-/LR-=11/72/36) and/or 24 months (n=102; HR+/HR-/LR-=17/59/23). A clinical best estimate ASD diagnosis was assigned at 24 months.
Imaging: Resting state fcMRI data were acquired on identical 3T Tim Trio scanners at 4 sites with up to 3 BOLD runs (130 frames each run with TR=2.5 seconds). Data processing included motion scrubbing at a framewise displacement level of 0.2 mm. One-hundred-fifty frames of clean data were used per subject. Time traces were correlated between 230 functionally-defined regions of interest (ROI) to yield fc values. ROIs were sorted into putative functional networks using the Infomap community detection algorithm on longitudinal fcMRI data from 48 subjects (Fig. 1).
Measures: Receptive and expressive language, distinct language modalities shown to be differentially affected in ASD, were indexed by raw scores on the Mullen Scales of Early Learning.
Brain-behavior analysis: Networks and pairs of networks were identified as significantly enriched for connections whose fc values strongly correlated with concurrent behavior (set at an uncorrected threshold of p<.05). χ2 tests and hypergeometric tests established whether network enrichment exceeded that expected by chance. A 5% false-positive rejection rate was determined by permutation. McNemar tests assessed enrichment differences between 12 and 24 months.

Results: At age 12 months, fc between anterior frontoparietal and dorsal attention networks positively correlated with receptive language (Fig.2a). For expressive language, a positive fc-behavior relationship was observed for visual and temporal default mode networks, whereas a negative correlation was observed for temporal default mode and cingulo-opercular networks (Fig.2b). At age 24 months, fc in default mode, posterior frontoparietal, and somatomotor networks correlated with receptive and expressive language (Fig.2c,d). One network pair, involving somatomotor and posterior frontoparietal networks, showed a positive fc-behavior relationship for both receptive and expressive language. ROIs contributing to network enrichment included regions highlighted in classic adult brain models of language, e.g., superior temporal gyrus and inferior frontal gyrus, as well as motor and cerebellar regions more recently implicated in language function.

Conclusions: Specific network-level relationships were observed for receptive and expressive language, which differed across age during early development. Implicated networks suggest broadly distributed brain involvement in early language development, while ROIs within networks suggest some developmental continuity for regions underlying language function. Future directions include comparing brain-behavior relationships for language in children with and without ASD and investigating whether brain-behavior relationships underlying language development correlate with ASD-related outcomes.