Date / Location: 18 March / II-028

Decisions are often guided by detecting subtle signals in a dynamic sensory environment. Although the brain must track such decision-relevant signals, how they are represented and transformed by neural activity across the sensorimotor axis remains poorly understood. Here, we recorded neural activity with Neuropixels probes across dozens of brain regions while mice performed a visual change-detection task. We trained mice to detect a sustained increase in temporal-frequency (TF) of a drifting grating stimulus, whose speed fluctuates stochastically around the mean of 1Hz. The task requires mice to remain stationary while continuously monitoring the grating with noisy speed which could increase at any moment, thereby allowing us to study the processing of dynamically changing task-relevant sensory evidence (i.e. TF) in the absence of overt movement and prior to the reporting of choice (lick). We find that even transient fluctuations (50 ms) in TF recruit activity in 10-20% neurons across a large number of distributed brain regions in the absence of choice and other movements. Beyond the visual system, we find such representations in posterior parietal cortex, premotor cortex, higher-order thalamus, midbrain, cerebellum and basal ganglia. Strikingly, only brainstem nuclei driving orofacial movements appear to be devoid of such sensory evidence representations. Interestingly, momentary increases in TF caused transient responses in neurons in visual areas (dLGN, V1 and superior colliculus), but more sustained responses in downstream areas previously associated with sensorimotor learning, working memory and motor planning (e.g. frontal-premotor cortex, basal ganglia, cerebellum). These sustained responses allowed for integration of multiple samples of stimulus speed, which could provide a robust neural substrate for better detecting sustained changes in noisy sensory evidence. These findings highlight how sensory evidence is transformed by distributed circuits making it available for computations in the entire sensorimotor axis to guide decisions.