Cortical deviance detection represents a canonical difference signal

Abstract

Context modulates neural processing of sensory stimuli. Neural responses are suppressed to stimuli that are typical in their context and augmented to stimuli that deviate from their context. The latter has been conceptualized as a "prediction error", which can serve to enhance the salience, direct attention, or support learning about behaviorally relevant events. Predictive coding theories posit that prediction errors act to signal the difference between internal predictions and actual sensory input, yet most paradigms simultaneously alter both predictions and input, so cannot test for a true difference signal. Increased neural responses to deviants could, instead, encode generalized surprise or augmented bottom-up signaling. Here we compare neural responses to auditory stimuli across oddball paradigm variants. We found that responses of putative excitatory neurons in primary auditory cortex (A1) to auditory deviants contain frequency change information and a memory trace of contextual information. Interestingly, in a fixed-deviant oddball paradigm where predictions are altered but deviant input remains constant, neural response patterns encoded standard-to-deviant frequency difference. These results support the interpretation that A1 deviance detection can be interpreted as a sensory prediction error that represents the difference between prediction and sensory input, a corollary of the predictive coding framework.