Spiking and locus coeruleus input drive predictive visuomotor plasticity in V1 layer 2/3 neurons

Abstract

Perception relies on sensory predictions that are continually refined by experience. Prediction errors are neuronal signals thought to instruct plasticity that improves the match between predictions and sensory inputs, minimizing future errors. This central tenet of cortical predictive processing, however, remains untested. Here, we show that activity in two sources of prediction errors - layer 2/3 neurons and locus coeruleus (LC) axons - drives changes in visuomotor integration in mouse V1 consistent with prediction error minimization. Using closed-loop manipulations of layer 2/3 neuron spiking during visuomotor experience, we reveal firing-rate-dependent changes in locomotion-related inputs that increase their opposition with visual inputs. Activation of LC axons facilitated similar activity-dependent reorganization evident in layer 2/3 calcium activity, alongside reduced responses to self-generated visuomotor input. Finally, we demonstrate that visual cortical activity itself can elicit pupil-linked neuromodulation, pointing to a feedback pathway linking cortical output to neuromodulatory plasticity signals. Our results support an activity-dependent, neuromodulation-facilitated cortical mechanism that sharpens visuomotor prediction error computation.