From a Heterogeneous ASD Phenotype to Quantitatively Distinct Putative ASD Subtypes

Background: Although it is accepted that the autism phenotype is variably expressed across cases and that there are likely underlying subtypes of autism (i.e. the “autisms”), it is not clear how best to identify phenotypic subtypes that are truly distinct.  In addition, “gene first” (e.g. comparing genetic conditions associated with ASD), and “brain first” (e.g. brain function -symptom correlations) approaches have yielded mixed results suggesting that efforts to parse ASD at the phenotypic level  may still be useful, and could then in turn inform genetic and neuroimaging studies.

Objectives: To develop a quantitative method for isolating distinct ASD phenotypes.

Methods: In the present study we compared a large and diverse sample of children and adolescents with ASD (n = 224, identified by ADOS, ADIR, & expert clinical diagnosis; age range = 2 to 18 yrs; IQ range = 40 – 140; 13% female) to a sample of typically developing children matched on age, IQ, & gender, using a multi-method (parent ratings, observational measures) and multi-measure (SCQ subscales, SRS, RBS-R subscales, CCC subscales, ADOS subscales) approach for mapping the core features of autism.  We used a 3 step process to isolate potential phenotypic  subtypes: (1) psychometric analyses identified the subset of measures with superior reliability & validity, (2) multivariate techniques (correlation, regression, & PCA) identified measures of core features with maximum between-measure divergence, (3) radar graphs of standardized scores isolated extreme outlier cases for each measure of divergent core feature (social, communication, RRBs).

Results: There was little evidence that the “core” features of autism were strongly related (e.g. correlations of social-communication function with RRBs across several measures were in the range of 0.2 to 0.4).  PCA analyses also yielded consistent evidence for divergence into the 3 core features (social, communication, RRBs), and also divergence within the core features into multiple sub-factors.   Radar graphs identified cases that were extreme outliers (> 1.5 SDs) on only 1 core feature: 9% on social impairment, 18% on communication impairment, and 11% on RRBs suggesting that up to one-third of the sample may be “extreme phenotypic outliers” within the ASD spectrum.  To investigate the validity of these putative subtypes, we are examining validation of these subgroups using eye-tracking measures of social and nonsocial attention, and fMRI measures of social and nonsocial reward circuitry activation.

Conclusions: Our analyses indicate that there is significant divergence of the 3 core features of autism (social, communication, RRBs) across methods and measures. Further, multi-method, multivariate approaches using psychometrically acceptable measures may be useful to identify “extreme” phenotypic outlier cases within each of the core features.  This approach may be useful in helping to identify specific genetic and neurobiologic factors that are associated with each core feature of autism in a way that previous research using a “one size fits all” phenotypic approach (e.g. ASD versus non-ASD) has not been able to do.