Language-related functional neuroimaging biomarkers in autism infants and toddlers with differing language outcome and developmental trajectory

Background:  Early language development is central for parsing phenotypic heterogeneity and predicting later outcome in ASD. However, little evidence exists testing whether atypical neural mechanisms in ASD infants and toddlers underlie such heterogeneity in early language development.

Objectives: To test early behavioral trajectories and neural response to speech in infant/toddler ASD subgroups differentiated on the basis of early childhood language outcome.

Methods:  Eighty-four age- and sex-matched 11-47 month olds (60 ASD; 24 TD; mean age = 26.4 months old, SD = 8.7) were scanned with fMRI at UCSD during sleep using the Story Language Paradigm (Redcay & Courchesne, 2008). Development was characterized longitudinally on the Mullen, ADOS, and Vineland over a period of 8-51 months. ASD was split into 3 language outcome subgroups (‘poor’, ‘good’, and ‘optimal’) on the basis of Mullen receptive (RL) and expressive language (EL) T-scores at about 1 year after scanning (ASDPoor: n=24, T<40; ASDGood: n=24, 40≤T<50; ASDOptimal: n=12, T≥50). fMRI data were analyzed with standard pipelines (e.g., AFNI, SPM) with head-motion measures as covariates in all analyses and correction for multiple comparisons at FDR q<0.05.

Results:  Group-by-development interactions were observed for EL (F(3,78) = 13.57, p = 3.3 x 10^-7) and RL (F(3,78) = 5.76, p = 0.001).  ASDGood, ASDOptimal, and TD toddlers improved with age, while ASDPoor toddlers declined or failed to sustain developmentally appropriate progress. Similar group-by-development interactions were observed across the visual reception subscale of the Mullen and Vineland communication, socialization, and adaptive behavior subscales, but not the ADOS. ASDGood, ASDOptimal and TD toddlers also showed robust bilateral speech-related activation of language-related perisylvian temporo-parietal areas, while ASDPoor toddlers showed little to no activation even at the most liberal uncorrected thresholds (e.g., p<0.05 uncorrected). Group comparisons revealed significantly less activation in ASDPoor compared to all other groups in bilateral language-related temporal cortices, as defined by an independent ROI from a meta-analysis of 553 studies on ‘language’ from neurosynth.org (LH:  F(3,79) = 4.49, p = 0.006; RH: F(3,79) = 2.84, p= 0.04) and with whole-brain analyses. ASDGood and ASDOptimal toddlers also showed increased speech-related superior temporal cortex connectivity with frontal cortex, amygdala, and ventral striatum compared to TD.  Finally, a support vector machine classifier made ASDGood vs. ASDPoor predictions solely based on speech-related superior temporal cortex activation patterns with ~70% accuracy (p=0.005) and generalized at similar levels when making predictions for ASDOptimal and TD.  

Conclusions:  Heterogeneity in early ASD language development is manifest as subgroups with differing early neural organization for language. Hypoactive language-related temporal cortex is specific to ASD toddlers who exhibit very poor early language development and outcome. In contrast, these same language-related temporal cortex regions remain functionally intact and are over-connected with frontal and subcortical areas in ASD toddlers with good or optimal language outcome and improving development. Such differences may be indicative of non-overlapping etiologies (e.g., genetic, molecular/cellular pathways) and/or distinct early developmental processes (e.g., experience-dependent change). This work is also important for further developments in the discovery of early neural biomarkers for early language prognosis in ASD.