Sensory severity predicts insula decoupling in autistic participants during motion stimulation

Sensory processing is a common target in autism spectrum disorder (ASD) research, yet the latent structure of sensory experience is disputed. Researchers frequently explore the presence of "subtypes" to categorize sensory heterogeneity, but such discrete models can fail to capture the intrinsic geometry of phenotypic data. In this study, we aim to characterize heterogeneous sensory profiles in ASD and explore if the same characterization can describe neurobiological function. First, we apply unsupervised spectral manifold dimensionality reduction to item-level Sensory Profile data from a large cohort of autistic participants (n=223) to compare categorical subtyping against continuous models. The behavioral results reveal unstable and irreproducible subtyping solutions; instead, sensory processing differences are best characterized as a continuous, non-linear manifold of sensory severity. To determine the neurobiological relevance of this sensory gradient, we employed voxel-wise linear mixed-effects modeling of insula-seeded functional connectivity (n=63). We demonstrate that sensory severity predicts a significant decoupling between the insula and sensorimotor cortices during externally driven stimulation involving motion stimuli, but not during resting state. This finding supports the interpretation that sensory-related neural hypoconnectivity is context-dependent and not reflective of intrinsic traits. Further, we identify a significant sex-by-sensory gradient interaction, indicating heightened sensitivity of connectivity patterns to sensory severity in autistic males. These findings indicate that sensory atypicality in ASD points toward a continuous regulatory manifold linked to disrupted social-sensory integration.