Abstract
We present a paradigm for recording anywhere in the brain, from several thousands of individual neurons at a time,in larval zebrafish behaving in virtual reality environments, during navigation and motor learning. The animals areparalyzed, and bilateral recordings from motor neuron axons in the tail provide sufficient information for decodingintended forward swims and turns. These intended actions are converted into motion in a virtual environment,rendering realistic visual feedback in response to fictive locomotion. Simultaneously, a two-photon microscopescans over the brain of transgenic zebrafish expressing a genetically encoded calcium indicator in all neurons. Inthis way, activity in large populations of neurons, that may cover the entire brain, can be monitored during diversebehaviors. Whole-brain activity is monitored as fish exhibit three behaviors analogous to the freely swimmingcounterparts: First, the two dimensional optomotor response; second, darkness avoidance; and third, motorlearning. During the 2D optomotor response, whole-hindbrain recordings reveal functional networks involved inforward swimming, left- and right turns. During lateralized swimming, activity is lateralized, but this organization isreversed in part of the cerebellar cortex. During darkness avoidance, neurons in the habenula and the pretectumrespond to luminosity in distinct spatial receptive fields. During motor learning, many brain areas are active duringdifferent phases of the behavior — the learning and the maintenance periods — with strong neuronal activation inthe cerebellum and the inferior olive, brain structures that are involved in motor learning in mammals. Lesioningthe latter structure leads to a loss of the behavior. Statistical methods, including dimensionality reduction, revealmultiple temporal profiles of neuronal activity, localizing to distinct brain areas, suggestive of a functional networkarchitecture. Such whole-brain recordings during behavior, in combination with computational techniques for theanalysis of these high dimensional data, will generate new insights into circuit function underlying behavior.
