Abstract
Spatially accurate saccade planning and perceptual localization require the integration of retinal and extraretinal signals. The way that retinal and extraretinal signals are combined determines the specific computation that is performed. For vector addition, retinal and extraretinal signals are spatially correlated, while for vector subtraction, the signals are anti-correlated. To characterize the interaction of retinal and extraretinal signals, we trained 3 macaque monkeys to perform a delayed visually guided saccade task in which both target eccentricity and initial eye position were varied. We recorded from 70 neurons in frontal eye field. We first tested cells with a delayed memory saccade task to determine the receptive/movement field. Then, using the visually-guided saccade task, we determined the eye position sensitivity and retinal sensitivity for neuronal responses during the visual and pre-saccadic task epochs. These sensitivities give an estimate of the gain field and the retinal receptive field of the neurons respectively. We found that the initial response to the saccade target was modulated by initial eye position in 51/70 (73%) neurons. In 47/70 (67%), the retinal and eye position signals were spatially anti-correlated and thus satisfied the constraint appropriate for vector subtraction, while 4/70 (6%) satisfied the vector addition constraint. Presaccadic activity was modulated by initial eye position in 42/70 (60%) neurons. Spatially anti-correlated retinal and extraretinal signals were found in 30/70 (43%) neurons, while correlated signals were found in 12/70 (17%) neurons. Across the population, the manner in which retinal and extraretinal signals are combined in FEF strongly favors vector subtraction rather than addition. The results are consistent with a model in which vector subtraction is computed from the responses of a network of cells with retinal receptive fields that are modulated by eye position in a gainfield-like manner.