The geometry of appetitive-aversive value representations in medial prefrontal networks

Abstract

The value of rewards and punishments - namely, how good or bad they are perceived - guides approach or avoidance behaviors. Valence refers to the negative or positive "sign" of the state elicited by an event, whereas salience refers to the amount of attention an event attracts, disregarding its valence. While identifying these signals conveys critical information for understanding circuits involved in emotional processing, they are often confounded due to their underlying correlation. Moreover, whereas the study of the neural basis of value coding has been intensively investigated in the appetitive domain, the neural substrates for how aversive values are established for different threat intensities and guide defensive behavior have yet to be discovered. The dorsomedial prefrontal cortex (dmPFC) is a key region in the control of defensive actions, although how different aversive values are encoded at the neuronal level within this region and drive defensive behaviors remains unknown. Here, we developed an instrumental approach/avoidance task in mice that, by matching motivational salience levels elicited by cues predicting rewards or punishments, allows univocally disentangling the presence of either salience, valence, or value coding from brain signals. We performed freely moving large neuronal population calcium imaging in the dmPFC of mice performing our task, conducting appetitive/aversive outcome devaluation/revaluation behavioral tests. We found that, while a similar fraction of single neurons decoded valence and value information, and only a minor fraction decoded salience, value coding was observed at the neuronal population level. Moreover, different value representations of the same valence lay within similar subspaces of the neural state space while values of opposed valence were encoded in orthogonal subspaces, unveiling how the brain stores associative appetitive and aversive information in medial prefrontal networks.