Auditory deviance detection beyond the auditory pathway: Hierarchical predictive coding in the hippocampus and primary visual cortex

Abstract

[EN] Under the predictive coding framework, perception is understood as an inferential process in which the brain continuously generates predictions about incoming sensory input and updates its internal models when reality deviates from expectation. Auditory deviance detection provides a tractable paradigm to investigate how the brain encodes prediction errors. While these computations are well characterized within the auditory hierarchy, their extension to memory-related and cross-modal structures remains unclear. This thesis examines auditory predictive processing beyond the auditory pathway, focusing on the hippocampus and the primary visual cortex (V1) as potential nodes in a distributed predictive network. Using extracellular recordings in urethane-anesthetized rats, I measured single- and multi-unit activity and local field potentials (LFPs) from hippocampal subfields (dentate gyrus [DG], cornu ammonis 1 [CA1]) and infragranular layers (IV–VI) of V1. Classical oddball paradigms and no-repetition control sequences were employed to dissociate repetition suppression from genuine prediction error components. Approximately one-fifth of hippocampal neurons responded to auditory stimulation, with a subset (~20%) showing enhanced activity to deviant tones. DG neurons displayed shorter latencies and larger mismatch and prediction error indices than CA1, suggesting a distinct contribution to novelty-related processing. Spiking responses reflected prediction error, while LFPs showed distinct contributions from repetition suppression and prediction error that varied across subfields. Early LFP components were enhanced for randomly presented deviants, whereas later components were more pronounced for predictable deviants. In V1, more than half of infragranular neurons responded to tones despite the absence of visual input, 50% showing larger responses to deviant than standard stimuli. These responses predominantly reflected prediction error rather than adaptation and were amplified under temporally uncertain (random) conditions. Layer V neurons exhibited rapid, phasic responses, whereas layer VI responses were broader and more sustained, consistent with laminar specialization in hierarchical predictive coding. Together, these results demonstrate that auditory prediction errors are encoded in both hippocampal and visual circuits, extending predictive processing beyond classical sensory boundaries, positioning both regions as integral components of a distributed predictive coding architecture that underlies hierarchical inference across the brain.