Decoding the neural basis of sensory phenotypes in autism

BACKGROUND: Differences in sensory processing are a defining characteristic of autism, affecting up to 87% of autistic individuals. These differences cause widespread perceptual changes that can negatively impact cognition, development, and daily functioning. Our research identified five sensory processing 'phenotypes' with varied behavioural presentations; however, their neural basis remains unclear. This study aims to ground these sensory phenotypes in unique patterns of functional connectivity. METHODS: We analyzed data from 146 autistic participants from the Province of Ontario Neurodevelopmental Network. We classified participants into five sensory phenotypes using k-means clustering of scores from the Short Sensory Profile. We then computed a connectivity matrix from 200 cortical and 32 subcortical regions and calculated graph-theoretic measures (betweenness centrality, strength, local efficiency, and clustering coefficient) to assess information exchange between these regions. We then trained machine learning models to use these measures to classify between all pairs of sensory phenotypes. RESULTS: Our sample was clustered into five sensory phenotypes. The machine learning models distinguished seven of the ten total pairs of sensory phenotypes using graph-theoretic measures (p < 0.005). Information exchange within and between the somatomotor network, orbitofrontal cortex, posterior parietal cortex, prefrontal cortex and subcortical areas was predictive of sensory phenotype. CONCLUSIONS: Sensory phenotypes in autism correspond to differences in functional connectivity across cortical, subcortical, and network levels. These findings support the view that variability in sensory processing is reflected in measurable neural patterns and motivate continued work to refine models of sensory processing, with the goal of better understanding and capturing the heterogeneity implicit in autism.