Midbrain Signals for Differential Reinforcement Learning

Project Summary/Abstract Associative learning is a fundamental cornerstone of behavior and cognition, but the underlying neural mechanisms remain poorly understood. In vertebrates, the inferior colliculus (IC) is a layered, midbrain hub important for sound localization, speech perception, and a potential site of plasticity for associative auditory learning. Indeed, classic studies in avians showed that neurons of the external “shell” IC layers have a substantial capacity for plasticity of auditory space maps. Additionally, early mammalian studies suggested that lesions of the analogous shell IC regions can selectively impair learned auditory associations. Interestingly, the first- and second-order targets of mammalian shell IC neurons include higher-order auditory thalamic nuclei and the amygdala, respectively; forebrain regions famously necessary for the acquisition and expression of associative learning. However, whether learning-related activity in upstream shell IC neurons contributes to these functions is less clear, owing to the difficulty of studying behaviorally relevant activity in shell IC neurons via standard techniques. We are addressing this knowledge gap using 2-photon Ca2+ imaging of shell IC neurons in behaving mice. We find that shell IC neuron activity predicts mice's behavioral choice on a trial-by- trial basis, implying substantial learning-related activity regarding the behavioral relevance of sounds. Additionally, many shell IC neurons are strongly active during trial outcomes as mice consume appetitive reinforcers. These results are exciting, as similar outcome activity is central to reinforcement learning models and provides an “instructive signal” for learning-related plasticity in many other brain regions. Given this background, we hypothesize that shell IC neurons are early plasticity loci to associate acoustic stimuli with reinforcing outcomes, thereby promoting learned instrumental responses to behaviorally relevant sounds. To test our hypothesis, we established novel cell-type specific methods to record and manipulate activity in shell IC neurons and/or their downstream targets. We propose applying these approaches as head-fixed mice learn and execute an appetitive, differential reinforcement task. The results will establish how an evolutionarily ancient midbrain circuit supports learned associations.