Multiplexed representations of uncertainty by mouse pulvinar-prefrontal projections during goal-directed behaviors

Abstract

Processing sensory information to generate decisions and action is a central component of learned, goal-directed behavior. Even during ongoing sensory-motor processing, our sensory landscape is fltered through our prior expectations and ongoing goals. This active perceptual process hinges on a distributed network of cortical and subcortical areas. The pulvinar, or homologous rodent lateral posterior (LP) nucleus, is a higher-order visual thalamic nucleus that bridges many of these subcortical and cortical structures. In particular, LP/pulvinar interactions with the prefrontal cortices such as the anterior cingulate cortex (ACC) have been implicated in regulating attentional processes. However, the anatomical inputs integrated and precise information carried by this projection during decision-making and action-selection has never been clarifed. We address this gap by leveraging genetic tools available in mouse models to examine the role of LP-ACC inputs directly with projectionspecifc anatomical mapping, axonal calcium imaging with two-photon microscopy in animals viewing visual stimuli passively or performing a decision task, and optogenetic manipulations. We fnd that LP-ACC axons integrate inputs from a vast network of subcortical and cortical structures that are implicated in attention, visuomotor functions, and spatial cognition. During passive viewing, activity of the LP-ACC projection is dominated by global arousal states while visual information is poorly represented. During a two-alternative graded random dot motion direction discrimination task, LP-ACC activity in individual axons and the axonal population represents multiple task variables. The activity of single axons ranges from the coding of stimulus coherence and direction in the random dot stimuli to the signaling of diferent task epochs in individual trials. At the population level, we fnd highly structured representations of task variables: LP-ACC activity jointly represents direction and coherence of visual stimuli in a low-dimensional geometric manifold that facilitates visual decoding. Furthermore, LP-ACC axons dynamically represent the outcome and uncertainty of previous trials, and integrate past and current trial uncertainty throughout the task. These physiological responses infuence trial-by-trial behavior, which can be disrupted by optogenetic perturbation of specifc trial epochs. Our fndings demonstrate that the LP contributes to attention and decision-making by providing a read-out of ongoing uncertainty, integrated over time with behavioral history, to adaptively tune neuronal responses and guide goal-directed behavior on a trial-to-trial basis.