The Primate Oculomotor System Plans Saccades to Objects not Points M. H. Phillips, S. C. Steenrod, M. E. Goldberg Although most laboratory studies of eye movements use point targets, in the real world primates make precise saccades to extended objects1-3, 5. In this study we asked how the primate oculomotor system interacts with the visual system to generate saccades to objects: Does the visual system extract a single point from the object, and input this to the oculomotor system? Or does the oculomotor system itself analyze the object, and generate a saccade by treating the object as a collection of points which are integrated in some way? We investigated this question using saccadic adaptation4 in the Rhesus monkey. When saccades to a given point are adapted using an intrasaccadic target step, saccades to adjacent points are also adapted, albeit less so as the distance between the probed point and the adapted location increases. The area of the visual field in which adaptation occurs automatically is the adaptation field6. If input to the oculomotor system is a simple point, then saccades to a spatially extended stimulus lying partially in and partially out of the adaptation field should be identical. If the oculomotor system represents the spatially extended object as a collection of points, on the other hand, then we would expect saccades to the two targets to be different, reflecting the different gain states associated with the constituent points. We found different adaptation to points and spatially extended stimuli. Subjects were 2 rhesus macaques. The data from one monkey are shown; the data from the other confirmed the main experimental finding. In the main experimental condition, focal adaptation, we adapted saccades to a small visual stimulus at 20-25deg eccentricity. We then determined the shape of the adaptation field using probe targets chosen from locations arrayed orthogonally to the axis of adaptation (fig. 1). On a small percentage of trials the monkey made a saccade to a long, thin bar oriented orthogonally to the axis of adaptation, whose center lay at the adapted point, and whose flanks extended into less adapted points. The bar disappeared during the saccade, but never reappeared, so the monkey had no feedback. If the input to the oculomotor system was simply a point corresponding to the center of the bar, to which it subsequently executed a saccade, then we would expect no difference between the amplitude of saccades to the adapted point and the bar. If, however, the oculomotor system took into account locations all across the bar, and averaged the adaptation of those points, we would expect less adaptation transfer to bar saccades. We found the latter prediction to hold true (figs. 1-3). Moreover, the amount of adaptation to the bar corresponded closely to the average amount of adaptation exhibited by probe stimuli located along the spatial extent of the bar (fig. 4). The different adaptation we found could be due to factors unrelated to shape, such as luminance. To rule this out, we adapted saccades simultaneously to multiple point-like stimuli which were arrayed orthogonally to the axis of adaptation, so that every part of the bar lay at a maximally adapted point. In this condition, generalized adaptation, saccades to the bar were nearly as adapted as saccades to the point stimuli (fig. 3). We conclude that the oculomotor system generates a saccade to a target based on some kind of integration over the spatial extent of the object. In order to gain insight into the nature of the spatial integration the oculomotor system does, we performed a simple linear regression on bar adaptation as a function of mean probe adaptation. Where the slope is close to 1, we have evidence that the oculomotor system simply averages gain states across the spatial extent of the bar to generate the bar saccade; a non-zero y-intercept can be interpreted as the effect that factors unrelated to shape, such as brightness, have on adaptation transfer to bar saccades. In the focal adaptation condition, we found a very strong correlation between average probe adaptation and bar adaptation in the monkey (fig. 4a), with a slope close to 1. The y-intercept was significantly different from 0, suggesting that non-shape-related factors (e.g. total luminance) combined with averaging to further reduce transfer. The results in the generalized adaptation condition (fig. 4b) were less clear, but were consistent with the hypothesis as the slope was not significantly different than 1 (95% confidence bounds). We conclude that we have found evidence that the primate oculomotor system generates a saccade to an object based on its shape, not on a simple point given to it by the visual system.

References:

1.

He, P.Y. and E. Kowler, Saccadic localization of eccentric forms. J Opt Soc Am A, 1991. 8(2): p. 440-9. 2. Kowler, E. and E. Blaser, The accuracy and precision of saccades to small and large targets. Vision Res, 1995. 35(12): p. 1741-54. 3. McGowan, J.W., E. Kowler, A. Sharma, and C. Chubb, Saccadic localization of random dot targets. Vision Res, 1998. 38(6): p. 895-909. 4. McLaughlin, S., Parametric adjustment in saccadic eye movements. Percept. Psychophys, 1967. 2: p. 359-362. 5. Melcher, D. and E. Kowler, Shapes, surfaces and saccades. Vision Res, 1999. 39(17): p. 2929-46. 6. Straube, A., A.F. Fuchs, S. Usher, and F.R. Robinson, Characteristics of saccadic gain adaptation in rhesus macaques. J Neurophysiol, 1997. 77(2): p. 874-95.

1) 1) a)

Stimuli

Results—illustrative session

b)

Mean saccade endpoint

Vertical eye position (deg)

point-like bar stimuli stimulus

(point-like stimuli shown on interleaved trials)

15

before saccadic adaptation

10

after saccadic adaptation

saccades to bar

5 0 -5

Initial fixation point

-10 -10

0 10 20 Horizontal eye position (deg)

Figure 1: Task stimuli (a), and results form an illustrative session (b). Filled symbols in (b) indicate stimulus locations where visual feedback was given after the saccade. Targets denoted by the central filled stimuli were adapted backwards; veridical feedback was provided to the targets at the corners. These ‘anchor’ targets were used in order to increase the curvature of the adaptation field. Figure 2: Same session in figure 1, showing individual probe (solid red, blue dots) and bar (open circles) saccade endpoints, with smoothed (lowess) probe endpoint average. Figure 3: Mean adaptation transfer to bar in focal and generalized adaptation conditions.

Figure 4: Correlation between mean probe adaptation and adaptation to the bar stimulus, in focal (a) and generalized (b) adaptation conditions.