[Symposium 28] Neurophysiological studies from a perspective of predictive coding in psychiatry

[SY-28-03] Hierarchical Predictive Coding in Autism Spectrum Disorder

*Zenas C Chao¹ (1. The University of Tokyo (Japan))

キーワード：

Autism spectrum disorder、Predictive coding、Hierarchy、Marmoset model、Human patient

Autism spectrum disorder (ASD) is characterized by atypical sensory experiences, which are often linked to irregularities in predictive coding. Predictive coding theory proposes that the brain constructs hierarchical sensory models through reciprocal interactions of predictions and prediction errors. While irregularities in predictive coding have been proposed to underlie sensory hypersensitivity and cognitive inflexibility in ASD, it remains unclear how these disturbances manifest across different functional hierarchies in the brain. To address this question, we examined a marmoset model of ASD induced by prenatal valproic acid (VPA) treatment. High-density electrocorticography (ECoG) was recorded during an auditory task that engaged two layers of temporal prediction. We then applied a quantitative modeling approach to evaluate the integrity of predictive coding across distinct hierarchies. Our results demonstrate persistent patterns of sensory hypersensitivity and unstable predictions across two cortical hierarchies in VPA-treated animals, accompanied by specific spatio-spectro-temporal neural signatures. Importantly, although imprecise predictions occurred regularly, they manifested in diverse ways, with some neural populations underestimating and others overestimating sensory regularities. This heterogeneity was further reflected in human ASD patients performing a comparable two-level prediction behavioral task. These findings suggest that ASD is not marked by a single deficit in predictive coding, but by diverse and hierarchy-dependent irregularities that may contribute to the wide variability of symptoms observed in patients. For clinicians, this work highlights the possibility of developing multi-level neural biomarkers of predictive coding that could be applied across species. Such biomarkers may help identify subgroups within ASD, link neural irregularities to individual differences in sensory and cognitive symptoms, and eventually guide more targeted interventions.