High-Frequency Activity Encodes the Temporal Dynamics of Hierarchical Prediction Errors in Humans: An Electrocorticography Study

Abstract

Prediction error refers to the discrepancy between expected and actual sensory input. Its hierarchical organization has been demonstrated through decomposed brain responses in the local-global paradigm, where temporal regularities are established locally at individual stimulus transitions and globally across multi-tone sequence structures. In macaques and marmosets, local and global prediction-error signals have both been found in high-frequency oscillations (60-150 Hz) recorded with ECoG. In humans, however, these signals have primarily been observed in a lower frequency range (30-100 Hz) using EEG. Here, we recorded human ECoG to achieve higher signal fidelity, enabling examination of neural oscillations above 100 Hz (high-frequency activity, HFA), which are believed to be closely linked to local spiking activity. Eight participants listened to auditory sequences that either followed their local and global regularities (local and global standards) or violated them (local and global deviants). Robust HFA responses were observed for the local deviants, but these responses were reduced when the deviants were expected based on the global regularity, indicating both levels of prediction-error processes contribute to the observed activity. Importantly, these HFA responses could be decomposed into two subcomponents: an early component reflecting local prediction-error signals localized to lateral auditory regions, and a late component reflecting global prediction-error signals prominent in both lateral auditory and frontal cortices. Together, these findings demonstrate that neural oscillations above 100 Hz encode hierarchical prediction errors not only in non-human primates but also in humans.