Mechanisms of efficient coding of dynamic visual motion signals for pursuit

摘要

Abstract Current theories of how the brain estimates stimuli from sensory populations are based on non-adapting responses to single-parameter, constant stimuli over long time windows – highly unnatural conditions. To understand sensory circuits we need to determine how dynamic, multi-dimensional stimuli are decoded for rapid behaviors by adapting neurons. Progress may depend less on our ability to record ever-larger samples of cortical activity, and more on our ability to relate the activity we observe to the brain's read out. In short, we need a proxy for the answer that we are trying to model from our recordings – a precise behavioral response. We propose to exploit the close connection between cortical visual motion representation and pursuit eye movements in monkeys to study two significant problems: (1) how cortico-pontine projections transform the distributed place code for visual signals in cortex to a form better suited to driving motor areas and (2) the how the brain forms stable sensory estimates with adapting neurons. Our focus is how activity in area MT is transformed by downstream projections to the dorsolateral pontine nucleus (DLPN). DLPN transmits estimates of retinal motion to the flocculus to initiate and maintain pursuit, although other pathways also contribute. The pursuit system is an excellent model for studying sensory decoding because little noise is added in downstream motor processing- - the eye movement is a faithful rendering of the brain's estimate of target motion. Although the eye pursues correctly, we showed that MT neurons do not maintain a fixed relationship between firing rate and retinal motion. Our first aim is to determine if the MT-DLPN projection reduces noise, filters out talk irrelevant signals, and alters the coordinate from direction-speed to the H-V axes of the extra-ocular muscles. We propose to record from MT and DLPN in behaving monkeys, using tetrodes to record groups of nearby neurons and eye coils to monitor eye movements very accurately. Our second aim is to determine how the brain recovers veridical stimulus estimates from an adapting sensory population. Adaptation is ubiquitous in the brain, often driven by rapid changes in natural stimuli. In the previous grant period, we showed that MT neurons adapt their gain to the direction variance of a dynamic motion stimulus, making good use of a limited response bandwidth. Gain adaptation increases bit rates but it also creates ambiguity because the mapping between motion direction and firing rate is not fixed. In our second aim, we will investigate whether a downstream area needs information about the stimulus variance to properly estimate motion direction. Pursuit behavior shows that the brain solves this problem, but gain adaptation seems to foil our current decoding models. We will use information-based methods applied to MT and DLPN data to determine how to form a successful read-out. Our proposed work will create more realistic theories of sensory coding and knowledge of the brain's mechanisms for implementing them.