Probabilistic Perceptual Decisions During Navigation Are Driven by Small Subpopulation of Neurons in a Single Cortical Column of Primary Somatosensory Cortex

Perceptually driven behavioral choices are thought to develop gradually from sensation to perception in the somatosensory cortex to guide decision-making in higher order cortical areas. Primary somatosensory cortex (wS1) of rodents related to their mystacial whiskers has been a model system to study this information flow. However, the role of wS1 in this process is often debated based on controversial results of loss-of-function behavioral experiments that often require prolonged training and movement restraints. Here, to elucidate the role of wS1 in decision-making, we developed an ethological whisker-guided virtual reality (VR) paradigm that closely mimics natural navigation in underground burrows. Untrained mice of either sex navigate left and right turns at high speed by sensing VR walls with just a pair of their C2 whiskers. Inactivating layer 4 of C2 barrel results in loss of ability to produce turns contralateral to the lesion. Using probabilistic model of collision avoidance in the presence of noise and uncertainties we hypothesize that wS1 is involved in a feedback control loop that requires continuous updates and predictions to infer the optimal path for collision avoidance.<b>Significance Statement</b> Perceptual decisions driven by sensing salient changes in the environment are thought to develop from sensation in primary cortex (S1) to decisions in pre-motor cortical areas. However, the role of S1 in this process is debated based on controversial results of loss-of-function behavioral experiments that often require prolonged training and movement restraints. Here, by utilizing an ethological whisker-guided virtual reality, we show that perceptual decisions causally depend on small subpopulation of neurons in layer 4 of a single cortical barrel. Whisker-guided navigation requires continuous updates and predictions of relative positions of the body and obstacles to infer the optimal path for collision avoidance. These complex computations are likely to rely on nested feedback loops that directly involve wS1 hence making it indispensable.