30108
Individual Differences in Intrinsic Brain Networks Predict Symptom Severity Variation in Autism Spectrum Disorders (ASD)
The neurobiology of heterogeneous neurodevelopmental disorders such as autism spectrum disorders (ASD) is still unknown. Specifically, the link between altered neurodevelopment and ASD symptomatology has not yet been identified. As the phenotypic expression of ASD is highly heterogeneous, findings across multiple studies are unreliable and often fail to replicate.
Recent work has suggested that heterogeneity in ASD is likely related to subject-specific differences in brain structure and function in this population. A roadblock to current knowledge of the condition is thus the high degree of individual variation in symptom-related neurobiology of ASD.
Objectives:
In this study, we hypothesized that individual differences in intrinsic brain networks were important features that could predict individual variation in ASD symptom severity. We developed a novel subject-level distance-based method to investigate if subject-specific features of functional organisation in the brain based could accurately predict individual differences in ASD symptom severity. To ensure that findings were reproducible and generalizable, we repeated analyses in independent singleton and monozygotic cohorts for validation.
Methods:
Task-free functional magnetic resonance imaging (fMRI) data was acquired from multiple imaging centres in singleton ASD cohorts matched to controls on age, sex, IQ and image acquisition site (ASD: n=100, age=11.43 years, IQ=110.58, 84 males; controls: n=100, age=11.43 years, IQ=110.70, 84 males). We extracted intrinsic brain network components using projective non-negative matrix factorization, an unsupervised machine learning method for network decomposition. Within each matched case-control pair, the intrapair Euclidean distance in network component strength was used to predict within-pair differences in severity of social dysfunction (Social Responsiveness Scale). To ensure reproducibility of findings, analysis was repeated in an independent locally recruited monozygotic twin sample concordant or discordant for ASD (n=12; age range, 5 to 18 years).
Results:
Across all paired subjects from a large multi-cohort dataset, within-pair differences in strength of a subnetwork was robustly associated with individual differences in social impairment severity (Figure 1A; T=2.206, p=0.0301). Specifically, as individual differences in subnetwork strength increased, differences in symptom severity between case-control pairs became more extreme. Subject-level variation in the subnetwork robustly predicted individual differences in symptom severity, such that ASD subjects demonstrated weaker subnetworks and more severe symptoms (Figure 2). The subnetwork comprised of hubs of the salience network (SN) and the occipital-temporal face perception network. Remarkably, the same subnetwork was reproducible in the independent twin cohort as a predictor of individual variation in social dysfunction severity (Figure 1B; T=-4, p=0.016, R2=0.75). Further validation analyses were performed across different experimental parameters.
Conclusions:
We identified an intrinsic brain subnetwork of salience attribution and face perception underlying social dysfunction in ASD. Importantly, results were replicated across multiple cohorts, singleton and twin populations, and experimental parameters. Because monozygotic twins share identical genes and environmental influences, non-shared environmental exposures as risk factors are implicated in twin differences in the expression of ASD within-pairs.
The robustness of findings provide a critical step forward in the reliable identification of candidate brain biomarkers in ASD, and provide new insights into individual differences in the neurobiology and symptom expression of the condition.