Circuit mechanisms of cortical predictive learning

To perform sensory guided behaviors, animals need to distinguish self-generated and externally generated sensory inputs. Predictive processing theories propose that the brain does this by learning to predict sensations caused by self-motion. The key signals thought to drive this learning are prediction errors: differences between predicted and actual sensory input. My previous work shows that neurons in the primary visual cortex (V1) compute visuomotor prediction errors, and that prediction errors activate the locus coeruleus, a brain-wide neuromodulatory system. We will now investigate the circuit and neuromodulatory mechanisms underlying the learning of sensory predictions, using V1 as a model. I hypothesize that input to V1 from higher order cortical areas undergoes plasticity during self-generated sensory feedback. This plasticity should be driven by prediction errors in V1 activity, modulated by locus coeruleus output, and improve detection of externally generated visual flow during self-motion. We will test this hypothesis using innovative methods, including a multimodal virtual reality system and a novel object detection task, combined with in vivo whole cell recordings, two-photon imaging, and optogenetics. The specific aims are to (1) investigate how prediction errors are communicated between the locus coeruleus and the cortex, (2) decipher the mechanisms of predictive plasticity within the V1 circuit, and (3) assess the behavioral relevance of this plasticity. The knowledge gained will have a fundamental impact on our mechanistic understanding of predictive learning in the cortex and the role of neuromodulation in this process, which will have significance for 1) understanding conditions in which the processing of self-generated sensory feedback is thought to be disrupted (e.g., neurodevelopmental conditions and psychosis), and 2) development of AI and brain-machine interfaces that deal with self-generated sensor feedback (e.g., prostheses).