The frontal eye field, laterial intraparietal area, and superior colliculus encode target positions in eye-fixed spatial representations. These target positions must therefore be updated across eye movements, and neurons in these brain structures have shown remapping of activity across saccades. In a topographically-organized structure such as the superior colliculus a target position is represented as a hill of activity during each fixation. It is currently unclear how this hill evolves during remapping, however. We simulated target position remapping using recurrent neural network models in order to examine how this evolution depends on the signals that drive the remapping, the type of eye movement, and the training condition. Eye movements considered were saccades, pursuit movements, and a combination of these. The signals used to drive the remapping in each case were efferent copies of the 3-D eye position (tonic) and velocity (burst) signals that control eye movement. In the saccade paradigm, the 2-D saccade-target retinal position signal in the superior colliculus was also considered. We found that the activity representing target position evolved across saccades as a slightly suppressed moving hill when the saccade-target retinal position signal was not used to drive remapping, and as a strongly suppressed jumping hill when this signal was used. In the pursuit paradigm the pursuit target movements were of constant velocity within each pursuit trial. We considered three pursuit training conditions: one in which both the pursuit duration and velocity were constant across all pursuit trials, one in which the velocity was constant but the duration varied across pursuit trials, and one in which velocity varied but duration was constant across trials. In all cases the hill of activation representing target position evolved as a moving hill during the pursuit movement. Only networks trained in the constant-duration, constant-velocity condition, however, showed significant suppression of this moving hill. In the saccade-plus-pursuit paradigm, in which both pursuit duration and velocity varied across trials, there was some suppression of activity during remapping. When suppression is strong during remapping, a moving hill resembles a jumping hill. The degree of suppression depends on the behavioral condition and the spatiotemporal characteristics of the signals used to drive the remapping.

Support Contributed By: CIHR (Canada). GB is supported by a Marie Curie fellowship (EU) and by CIHR (Canada). JDC holds a Canada Research Chair.