Supramodal and modality-specific neural information supports multi-feature prediction errors across cortical levels

Predictive coding posits that the brain actively anticipates inputs from different senses, generating prediction errors when incoming information deviates from internal expectations. While much research has focused on prediction errors elicited by violations of single sensory features, natural environments frequently present more complex events deviating across multiple stimulus dimensions and sensory modalities. In this study, we employed a hierarchical oddball paradigm (n=30) manipulating auditory and somatosensory stimuli to violate one or two sensory features while high-density EEG was recorded. Temporal decoding revealed that while both single- and double-deviants evoked sustained supramodal activation patterns, double-deviants uniquely elicited a supramodal response starting at 100 ms after the oddball. Effective connectivity analyses identified shared interhemispheric interactions between inferior frontal gyri across modalities, as well as distinct modality-specific connectivity within early and associative sensory cortices. Our findings demonstrate that multi-feature prediction errors recruit both rapid supramodal integration mechanisms and hierarchically organized modality-specific pathways. These results advance our understanding of how the brain flexibly integrates multiple sensory expectation violations across different levels of cortical processing, providing new insights into the neural architecture supporting predictive perception. Author summaryThe brain constantly generates predictions about incoming sensory information. While many studies focus on simple violations of individual features, real-life events often involve simultaneous deviations across multiple sensory attributes. Our goal was to examine supramodal or modality-specific aspects of multi-feature prediction errors. Considering that the cortex needs to converge individual predictions from multiple pathways for multi-feature prediction, we hypothesised that multi-feature prediction errors rely on a mid-latency supramodal process in the inferior frontal cortex. In a high-density EEG study using a nested oddball paradigm, we examined neural responses to somatosensory and auditory stimulus deviations in one or two dimensions. Using temporal decoding, we revealed an early supramodal short-lived cortical process when double-deviants are detected. We also applied Parametric Empirical Bayes Modelling to show that multi-feature prediction errors not only rely on a common interhemispheric inhibition between inferior frontal gyri but also on various supramodal and modality-specific changes in effective connectivity across associative and modality-specific cortices.