Abstract
The cerebral cortex develops an internal model of the world by learning from sensory experience. The representation of edge orientation in carnivore’s primary visual cortex (V1) offers a great model to study this as it matures with experience after eye opening. In adult V1, orientation preference is similar across layers but changes smoothly across the surface forming a columnar map. Previous research has focused on how experience refines inputs to individual neurons generating selectivity. Yet, how orientation preference becomes organized to form a map remains unclear. Models suggest that recurrent connections could contribute to this process by imposing a structure on experience-driven cortical activity across development. Sensory experience is conveyed by short-latency feedforward inputs, which arrive first to layer 4 and are then relayed to other layers in a columnar fashion. In contrast, recurrent interactions act with longer latencies and are most prominent in layers 2/3. Therefore, the contribution of recurrent connectivity to cortical responses can be inferred by resolving the laminar and temporal structure of evoked responses. Here, we combined electrophysiology and calcium imaging to measure this in visually naïve and experienced animals. In naïve animals, responses in layers 2/3 exhibited modular patterns with high variability and weak selectivity. These patterns emerged with long latency and were not present in layer 4, suggesting they are shaped by recurrent interactions within layers 2/3. With experience, responses became highly reliable, strongly selective, consistent across layers, and emerged with short latency. These results are consistent with the established view that mature responses are supported by feedforward connectivity. Altogether, our data indicate that recurrent interactions shape stimulus-driven cortical activity in naïve animals while feedforward inputs determine mature response properties. Early response patterns shaped by recurrent networks could provide an initial scaffold upon which experience-dependent refinement of feedforward inputs generates organized stimulus representations.