Temporal prediction errors in visual and auditory cortices

Abstract

To form a coherent percept of the environment, the brain needs to bind sensory signals emanating from a common source, but to segregate those from different sources [1]. Temporal correlations and synchrony act as prominent cues for multisensory integration [2-4], but the neural mechanisms by which such cues are identified remain unclear. Predictive coding suggests that the brain iteratively optimizes an internal model of its environment by minimizing the errors between its predictions and the sensory inputs [5,6]. This model enables the brain to predict the temporal evolution of natural audiovisual inputs and their statistical (for example, temporal) relationship. A prediction of this theory is that asynchronous audiovisual signals violating the model's predictions induce an error signal that depends on the directionality of the audiovisual asynchrony. As the visual system generates the dominant temporal predictions for visual leading asynchrony, the delayed auditory inputs are expected to generate a prediction error signal in the auditory system (and vice versa for auditory leading asynchrony). Using functional magnetic resonance imaging (fMRI), we measured participants' brain responses to synchronous, visual leading and auditory leading movies of speech, sinewave speech or music. In line with predictive coding, auditory leading asynchrony elicited a prediction error in visual cortices and visual leading asynchrony in auditory cortices. Our results reveal predictive coding as a generic mechanism to temporally bind signals from multiple senses into a coherent percept.