Two types of locus coeruleus norepinephrine neurons drive reinforcement learning

To make decisions in a dynamic world, animals adjust their behavior in response to environmental feedback. To understand how this flexibility is achieved, we seek to understand how the nervous system learns from choice outcomes. Theories proposed that norepinephrine (NE) neurons in the locus coeruleus (LC) modulate learning from reinforcement. We tested this hypothesis by measuring activity from identified mouse LC-NE neurons during a behavioral task requiring ongoing learning from reward prediction errors (RPE). Mice were trained to make choices to lick leftward or rightward for reward with changing probabilities. Because reward probabilities were unknown to the animals, they used reward history to make future choices. We made electrophysiological recordings from identified LC-NE neurons by “tagging” channelrhodopsin-2-expressing LC-NE neurons in Dbh-Cre mice. We found two types of biophysically distinct LC-NE neurons featuring different action potential shapes, dorsal-ventral distribution, and excitability. These two types of neurons also responded differently to decision outcomes. Type I neurons were excited by positive RPE while type II neurons were excited by lack of reward. To test the hypothesis that LC-NE neurons modulate learning from decision outcomes, we silenced their activity using the inhibitory opsin GtACR2. The inactivation caused increased probability of switching choices on the trials following no reward, which can be captured by modeling inhibition as a negative shift of RPE. Our data indicate biophysically and anatomically distinct modules in LC and reveal a function for LC-NE neurons in modulating learning from experience.