Ye, S. Y.; Banqueri, M.; Jordan, R.

Computational models propose that prediction errors drive plasticity to improve cancellation between sensory and predictive inputs - a process known as prediction error minimization. In mouse V1, layer 2/3 neurons exhibit spiking responses consistent with prediction errors between visual flow and locomotion-based predictions, but evidence that such spiking drives plasticity consistent with prediction error minimization is limited. To address this, we performed in vivo whole cell recordings in mice running in virtual reality while manipulating spike rates via closed-loop locomotion-coupled current injection. Spiking during visuomotor experience altered locomotion-related inputs in a firing rate-dependent manner. The direction of change depended on each neuron's visual responsiveness, increasing the opposition between locomotion-related and visual inputs. Analyzing a published two-photon calcium imaging dataset, we demonstrate that a similar activity-dependent reorganization of visuomotor inputs is evident at the population level, but only when locus coeruleus axons were optogenetically activated during visuomotor experience. Together, our results provide evidence for refinements of prediction error computation in layer 2/3 that are driven by spiking and facilitated by neuromodulation across active experience.