Tachykinin 1 neurons in the lateral habenula signal negative reward prediction error

Evaluating outcomes to accurately predict which actions lead to reward is essential for survival. Discrepancies between expected and realized outcomes, termed reward prediction errors (RPEs), serve as teaching signals to update subsequent predictions and promote adaptive behavior.^1,2,3,4,5 Neural correlates of RPEs have been identified in several brain regions,^{6,7,8,9,10,11,12} including the lateral habenula (LHb), which contains a subpopulation of neurons encoding negative RPE (nRPE): they are excited by worse-than-expected outcomes and inhibited by better-than-expected outcomes.^{13,14,15,16,17,18} LHb activity shapes firing in dopaminergic neurons and plays a well-established role in reward learning and decision-making.^14,19,20,21 However, the LHb engages in many behaviors,^22,23,24 and it remains unclear whether specific cell types mediate its diverse functions. Little is known about the transcriptomic identity of nRPE-encoding neurons, which limits the use of genetically-targeted tools to understand how these signals contribute to outcome valuation. Using cell-type-specific recording in mice performing reward-guided tasks, we demonstrate that LHb neurons expressing tachykinin 1 (Tac1) are selectively tuned to nRPE. LHb^Tac1 activity is sensitive to changes in both the expected and realized value of rewards, and scales with the magnitude of the difference. LHb^Tac1 neurons show little modulation to other task-related events and are only weakly driven by aversive stimuli. Together, these data demonstrate that Tac1 marks a subpopulation of LHb neurons that encodes valence-biased prediction errors, preferentially responding to worse-than-expected outcomes in appetitive contexts. Our results provide insight into cell-type-specific contributions of habenular neurons in nRPE signaling and enable more targeted manipulations to understand their role in reward-guided behavior.