Coherent cortical representations develop after experience via feedforward-recurrent circuit alignment.

Abstract

Sensory cortical areas guide behavior by transforming stimulus-driven inputs into selective responses representing relevant features. A classic example is the representation of edge orientations in the visual cortex ^1-4 , where layer 4 (L4) neurons co-activated by an orientation provide feedforward inputs to specific functional modules in layer 2/3 (L2/3) that share strong recurrent connections ^5-7 . The aligned state of feedforward-recurrent interactions is critical for amplifying selective cortical responses ^8-12 , but how it develops remains unclear. Using simultaneous electrophysiology and calcium imaging in visually naïve animals we find that coactivity of L4 neurons and L2/3 modular responses elicited by oriented gratings lacks the tight relationship to orientation preference found in experienced animals. One factor that could contribute to the lack of functionally specific coactivity is high variability in naïve L4 neuron responses that decreases significantly following experience. But a computational model of feedforward-recurrent interaction suggests that high variability alone is insufficient to explain the naïve state and provides a biological signature of feedforward-recurrent misalignment that we confirm with whole-cell recordings: dynamic changes in orientation tuning of L2/3 subthreshold responses shortly after stimulus onset. In conclusion, we provide diverse evidence for a realignment of feedforward-recurrent interactions following experience that is critical for building reliable sensory representations with interlaminar temporal coherence.