Signaling of Salience and Prediction Errors by Anterior Insula Neurons

Human fMRI studies suggest that the anterior insula (AIC) encodes salience and deviations from expectations. However, BOLD signals have insufficient resolution to determine how AIC cells encode these variables. To study this question, rats were presented with auditory stimuli that varied from rewarding to aversive, instilling expectations that were later confirmed or challenged. Most AIC neurons showed significant task-related activity, either in the form of motor correlates, responses to appetitive or noxious sounds, in anticipation of reward delivery, or a combination of these factors. There was extensive overlap between neurons excited by the noxious and appetitive stimuli. Yet, stimuli with the highest positive and negative valence excited more AIC neurons than stimuli with intermediate valence, consistent with the signaling of salience. Strikingly, the responses of AIC cells to stimuli that confirmed reward expectations decreased whereas responses to stimuli that challenged reward expectations increased, akin to an unsigned prediction error signal. • We studied the coding properties of AIC neurons in rats. • Many AIC neurons responded to both, rewarding and aversive stimuli. • Highly positive or negative cues excited more AIC neurons than intermediate cues. • AIC firing rate to stimuli that confirmed or challenged expectations dropped vs. rose.