Cerebellum-machine interface to understand cerebellar roles in motor learning

A cerebellum-machine interface (CMI) was developed to understand cerebellar roles in motor control and learning by testing direct causality between single unit cerebellar Purkinje cell activity and motor learning. The CMI converts Purkinje cell simple spike firing rate into pulse-width modulation signals that drive a single joint robot arm. The CMI has no adaptive capability, thus any change observed in robot arm motion can be attributed directly to Purkinje cell firing activity. We employed a vestibuloocular reflex (VOR) adaptation paradigm in goldfish as the test for motor learning. Changes in eye muscle characteristics, or equivalently adding visual motion stimulus during head rotation require oculomotor commands to be calibrated by minimizing image slip across the retina, which acts as a control error signal. This calibration, known as VOR motor learning, has been postulated to depend on synaptic plasticity in the vestibulo-cerebellum, where Purkinje cells are believed to encode motor commands for eye movement. If this is true, then vestibulo-cerebellar Purkinje cells may similarly learn to correct error in a motor command signal when the oculomotor system is replaced with a different system, such as a robot arm. Desired motion and control error signal of the robot arm were presented as head rotation and retinal slip, respectively, to the fish VOR model. The control error of the robot arm was shown to decrease gradually, but not monotonically, and in many cases only in one direction. This evidence is the first direct demonstration that activity of a single Purkinje cell is capable of implementing adaptive motor control. The results also suggest that a single Purkinje cell can be responsible, at least in part, for directional selective VOR motor learning previously reported in goldfish [6] and monkeys [3].