Daily rewiring of a neural circuit generates a predictive model of environmental light
Ongoing sensations are compared to internal, experience-based, reference models; mismatch between reality and expectation can signal opportunity or danger, and can shape behavior. The nature of internal reference models is largely unknown. We describe a model that enables moment-to-moment luminance evaluation in flies. Abrupt shifts to lighting conditions inconsistent with the subjective time-of-day trigger locomotion, whereas shifts to appropriate conditions induce quiescence. The time-of-day prediction is generated by a slowly shifting activity balance between opposing neuronal populations, LNvs and DN1as. The two populations undergo structural changes in axon length that accord with, and are required for, conveying time-of-day information. Each day, in each population, the circadian clock directs cellular remodeling such that the maximum axonal length in one population coincides with the minimum in the other; preventing remodeling prevents transitioning between opposing internal states. We propose that a dynamic predictive model resides in the shifting connectivities of the LNv- DN1a circuit.