Network Structure of the ASD Phenotype Reveals Both Expected and Novel Relationships in a Large Community Sample

Background: Autism Spectrum Disorder (ASD) is characterized by significant heterogeneity within core symptom domains (social communication and repetitive and restricted behaviors). Heterogeneity is often further compounded by comorbid conditions such as anxiety and attention deficits, as well as by impairments in other domains such as executive functioning and language expression and function. Insights into the complex interplay between behavioral domains is crucial to understand mechanisms underlying ASD heterogeneity.

Objectives: To reveal network structure of ASD phenomenology, depicting relationships between social communication, restricted/repetitive behaviors, and other domains of functioning commonly impaired in ASD. To quantify network characteristics that describe connections among these domains in ASD.

Methods: Our sample consisted of 449 participants (91% male, ages: 5-18 years, mean[SD] age: 10.2[3.0] years, mean[SD] Full Scale IQ: 94.7[22.6]) with community diagnoses of ASD. Standardized measures commonly used to assess behavioral functioning of children with ASD were used, including parent-report questionnaires assessing ASD symptoms, executive functioning, adaptive skills, language and communication, attention and hyperactivity/impulsivity, anxiety, and sleep habits, as well as clinician-administered gold-standard diagnostic tools and cognitive assessments. An undirected network representation of measures was constructed, with nodes representing the scales of the measures and edges encoding partial correlations between the scales. The data was processed with a nonparanormal transformation to calculate nonparametric correlations due to non-normal distributions of scales. Network estimation was performed using the graphical LASSO algorithm which yields partial correlations between any two given scales that are controlled for the influence of all the other scales. The extended Bayesian information criterion was used for optimal parameter selection. Three graph indices were computed to quantify topology of the network: betweenness centrality, closeness centrality, and node strength.

Results: The network structure (see Figure 1) indicated only a few strong connections (partial correlation r > 0.1) between the two core ASD domains (social communication and restricted/repetitive behaviors). Connections between these domains were primarily established through measures that assess both domains (e.g. SRS-2, ADI-R). Repetitive behaviors were directly connected to executive functioning through a connection between preference for sameness and the executive skill of shifting (r=0.19). The social anxiety scale, with high centrality indices (z-scores: closeness=1.52, betweenness=2.17), acted as a bridge between other anxiety scales and social functioning. Connections between ADHD symptom domains and core ASD domains were predominantly established through executive functioning or higher-level aspects of language use. Intellectual functioning was primarily linked to language and communication variables, but not to social functioning. Social communication had high centrality (z-scores: closeness=1.69, betweenness=1.58) and high strength (z-score=1.59), reflecting its core role in ASD.

Conclusions: Our results provide insights into the complex interplay among ASD symptoms and co-occurring impairments. Identified links between core symptom domains are consistent with current conceptualization of ASD phenomenology. Novel findings on connections between executive functioning, repetitive behaviors, language, anxiety, and social functioning may define testable hypotheses regarding causal mechanisms of ASD symptomatology.