A Frontal-Sensory Cortex Circuit Encoding Context-Dependent Stimulus Statistics

Abstract

Animals continuously estimate the statistics of sensory events in their environment to guide behavior. Yet, how the brain constructs internal models that represent the probability distributions of stimuli through learning, how these models are updated during rapid contextual shifts, and what brain regions integrate contextual and probabilistic information remains unclear. Here, we devised a behavioral task in which mice used rapidly changing contextual information to estimate the probability of upcoming odors. Recordings from the piriform and orbitofrontal cortices during learning revealed two distinct neuronal subpopulations that differentially encoded odor probabilities. One subpopulation exhibited ramping activity that scaled with the estimated probability of the behaviorally relevant odor and sharply dropped at its onset. In contrast, a second subpopulation became active at odor onset and was modulated by prediction. Notably, probability information emerged in the orbitofrontal cortex before the piriform cortex, and silencing the orbitofrontal cortex abolished the propagation of this information to the piriform cortex, impairing probability learning and retrieval but not context identification. These findings reveal a frontal-sensory circuit that facilitates the learning and representation of context-dependent, behaviorally relevant event probabilities, and suggest that learned state transitions, a core feature of a cognitive map, are propagated to the sensory cortex through learning.