Repeated exposure to a consistent change in target motion allows the smooth pursuit eye movement system to learn rapidly and emit an eye movement that predicts the time and direction of a change in target motion [1]. To understand how the neural signals for learning are represented and modified as they progress through the pursuit circuit, we recorded from the smooth eye movement region of the frontal eye field (FEFSEM). The FEFSEM is a motor cortex for pursuit that processes the visual motion input from area MT to drive downstream structures. Pursuit neurons in the FEFSEM do not respond in a stereotyped manner to pursuit. Rather, the coding of pursuit is distributed so that each neuron is most active during specific epochs of the movement [2]. This raises the possibility that a neuron is selectively involved in learning at its preferred times, or the times after the onset of target motion when the neuron is most active.

We recorded from 55 single units in the macaque FEFSEM to explore how these neurons alter their activity during pursuit learning. The change in target direction always occurred at 250 ms after the onset of target motion, but the direction change itself was chosen anew each day to be the cardinal direction closest to the neuron's preferred direction. Learning-related changes in firing rate varied widely across our neural population. We found that the magnitude of neural learning could be predicted by the normalized size of the neuron's response in normal pre-learning pursuit, at the time of the instructive stimulus 250 ms after the onset of target motion. Differences in neural learning could not be explained by other factors, such as the day to day variability in the learned behavior.

Based on our results for when the instructive stimulus occurred at 250 ms, we postulated that FEFSEM neurons tend to express learning when the instructive stimulus occurs around the time of the neuron's maximum response. As a direct test, we exposed 11 neurons to an additional learning experiment which differed from the first learning experiment only in the timing of the instructive stimulus, which was chosen to be 150 ms for some neurons and 350 ms for others. For 8 of the 11 neurons, neural learning was larger when the instructive stimulus occurred at a time more preferred by the neuron. Taken together, these findings indicate that one of the advantages of a temporally distributed coding for pursuit in the FEF may be to allow the system to selectively modify portions of the movement.

Acknowledgements
This work was supported by the Howard Hughes Medical Institute and NIH grant MH77970.