NEURAL BASIS FOR MOTOR LEARNING IN THE VESTIBULOOCULAR REFLEX OF PRIMATES .3. COMPUTATIONAL AND BEHAVIORAL-ANALYSIS OF THE SITES OF LEARNING

1. We have used a combination of eye movement recordings and computer modeling to study long-term adaptive modification (motor learning) in the vestibuloocular reflex (VOR). The eye movement recordings place constraints on possible sites for motor learning. The computer model abides by these constraints, as well as constraints provided by data in previous papers, to formalize a new hypothesis about the sites of motor learning. The model was designed to reproduce as much of the existing neural and behavioral data as possible. 2. Motor learning was induced in monkeys by fitting them with spectacles that caused the gain of the VOR (eye speed divided by head speed) to increase to values >1.6 or to decrease to values <0.4. We elicited pursuit by providing ramp motion of a small target at 30 degrees/s along the horizontal axis. Changes in the gain of the VOR caused only small and inconsistent changes in the eye acceleration in the first 100 ms after the onset of pursuit and had no effect on the eye velocity during tracking of steady target motion. Electrical stimulation in the flocculus and ventral paraflocculus with single pulses or trains of pulses caused smooth eye movement toward the side of stimulation after latencies of 9-11 ms. Neither the latency, the peak eye velocity, nor the initial eye acceleration varied as a consistent function of the gain of the VOR. 3. The computer model contained nodes that represented position-vestibular-pause cells (PVP-cells) and flocculus target neurons (FTNs) in the vestibular nucleus, and horizontal gaze-velocity Purkinje cells (HGVP-cells) in the cerebellar flocculus and ventral paraflocculus. Node FTN represented only the ''E-c FTNs,'' which show increased firing for eye motion away from the side of recording. The transfer functions in the model included dynamic elements (filters) as well as static elements (summing junctions, gain elements, and time delays). Except for the transfer functions that converted visual motion inputs into commands for smooth eye movement, the model was linear. 4. The performance of the model was determined both by computer simulation and, for the VOR in the dark, by analytic solution of linear equations. For simulation, we adjusted the parameters by hand to match the output of the model to the eye velocity of monkeys and to match the activity of the relevant nodes in the model to the firing of HGVP-cells, FTNs, and PVP-cells when the gain of the VOR was 0.4, 1.0, and 1.6. Analytic solution of the model in the APPENDIX provided equations that expressed the gain of the VOR, the dynamics of the VOR, and the responses of the nodes that represent HGVP-cells and FTNs as functions of the free parameters in the model. These equations demonstrate the stability conditions for the model and provide a complete description of the effects of changes in different parameters on the performance of the model. 5. We first adjusted the parameters of the model to emulate the eye movements and the unit activity recorded in monkeys duringpursuit with the head stationary and during the VOR evoked by rapid head turns in darkness when the gain of the VOR was one. We then used the model to test previous hypotheses about the sites of motor learning in the VOR. With a single site of learning in the strength of vestibular transmission to HGVP-cells, as postulated by Ito, the gain of the VOR was modified, but a constant head velocity caused model eye velocity that showed unstable runaway behavior. With multiple sites of learning that caused parallel changes in the strength of vestibular transmission to FTNs and HGVP-cells, as postulated by Miles and Lisberger, the model produced a stable VOR. Under these assumptions, however, the model could not reproduce either 1) changes in the gain of the VOR as large as those seen in monkeys or 2) the responses of HGVP-cells and FTNs during the VOR. We conclude that both previous hypotheses about the sites of motor learning are incomplete. 6. We used the model to demonstrate a new hypothesis that reproduces the basic features of available data on the eye movements of the VOR and on responses of brain stem and cerebellar neurons before and after learning. The new hypothesis suggests that changes in the vestibular inputs to FTNs in the brain stem are in the correct direction to cause motor learning. Parallel changes in the sustained component of the vestibular inputs to HGVP-cells are in the wrong direction to cause learning but compensate for the changes in the brain stem and maintain a VOR that does not show unstable runaway. Changes in the time course of vestibular inputs to HGVP-cells, implemented here as a change in the balance of their tonic and phasic/tonic vestibular inputs, are in the correct direction to contribute to changes in the gain of the VOR. We suggest that sites of motor learning in the VOR may lie both in the brain stem VOR pathways and in the vestibular inputs to the flocculus and ventral paraflocculus of the cerebellum.