Ann. Rev. Nellrosci. 1987. 10.' 97-129 Copyright �� 1987 by Annual Reviews Illc. All rights reserved VISUAL MOTION PROCESSING AND SENSORY-MOTOR INTEGRATION FOR SMOOTH PURSUIT EYE MOVEMENTS s. G. E.
Lisberger, J.
Morris, and L. Tychsen Division of Neurobiology and Department of Physiology, University of Cal if ornia at San Francisco, San Francisco, Cal if ornia 94143 INTRODUCTION Much of our motor activity is guided by what we sec, with a final goal of grasping or tracking an obj ect. On ce intent h as been established, the generation of visually guided movement involves processing sensory stim�� uli and transforming them into commands for activation of the relevant musculature. Because eye movements are simpler than other movements in manyways, the oculomotor sys tem provides an ideal opportunity for investigation of the brain mechanisms subserving visually guided move�� ment. The mechanics of the eyeball are straight fo rward (Robinson 196 4), and the eyeball is not sub ject to sudden changes in load. In addition, the study of the eye movements is unen cumbered by the problems of complex kinematics and dyn amics that confound the study of limb motion (e.g. Atkeson & Holle rbach 1985). Much is no w known about the processing of visual inputs and their use in generating smooth pursuit eye movements. The important neural pathways are becoming well known, and a subst antial body of psycho�� physical exp eriments have provided de tailed info rmation about the properties of pursuit. As a result, we kno w the properties of visual inputs that drive pursuit and we can correlate those propert ies with neural path�� ways that may transmit inputs for pursuit. Mo reover, we have identified transfo rmations performed in the b rain to convert visual inputs into motor 97 o I 47-00 6X/87 /030-0097$02.001 by Universidad de Chile on 07/27/11. For personal use only.

98 LISBERGER, MORRIS & TYCHSEN commands. Finally, we can correlat e this inf ormation with the firing prop�� erties and connections of brain cells in structures that are necessary for normal pursuit. It now seems to be within our grasp to integrate anatomical, physio�� logical, and behavioral results and to develop models that (a) include the elements, signals, and transf ormations identified by biol ogical exper�� imentation, and (b) when simulat ed on the computer, reproduce existing data on the performance of smooth pursuit. In most of our review, we focus on experiments that have provi ded the facts on which models of pursuit now can be based. In the final section, we discuss a model of pursuit as a means of synthesiz ing the wealt h of recent data. GENERAL PROPER TIES OF SMOOTH PURSUIT Pursuit eye movements are most highly developed in primates. Pursuit occurs in response to a movi ng visual stimulus, and the resultsmoothing eye rot ation keeps the fo vea pointed at the stimul us. Pursuit is different fr om ot her kinds of smooth ocular fo llowing, in that it allows pri mates to track a small obj ect accurately, even as it moves across a stationary, patterned background (Coll& ewi jn Tamminga 1 9 84, Kowler et al 1 9 78). This ability distinguishes primates fr om species such as cats, rabbits, and fish, all of which make smoot h eye movements when the entire visual sc is moved, and none of which has strong pursuit systems. Pursuit has been investigat ed in two species of primates, humans and macaque monkeys. The performance of the two species is qualita tively similar, although monkeys generally respond more strongly to a given visual stimulus (com pare Lisberger & Westbrook 1 9 85 with Tychsen & Lisberger 198 6a). Normally, a smoothly movismall ng target evokes a combination of smooth and saccadic eye movements, as shown in Figure l A o Although the saccades play an important role in maintaining accurate tracking, we define pursuit as the smooth component of the response. Even early investigators recognized the basic difference between pursuit and saccades (Dodge 1 9 03, Westheimer 1 9 5 4), and our rather conse rvative definition of pursuit is now supported by abundant evidence that the smootandh saccadic parts of the response are generated by separat e neural systems. Smooth pursuit and saccades are affect ed differentially by lesions of specific regions of the cerebral cort ex (Newsome et al 1 9 85a,b) and cerebellum (Zee et al 19 81 , Op tican & Robinson 19 8 0), have different latencies (Rash�� bass 1 96 1 , Robinson 1 9 65, Fuchs 1 9 67a), and respond to different aspects of the visual stimulus (Rashbass 1 9 61). In addition, single unit recordings in behaving animals have identified several classes of cells that modulate their dischar ge only in relation to saccades or to pursuit, but not both. by Universidad de Chile on 07/27/11. For personal use only.

SMOOTH PURSUIT EYE MOVEMENTS 99 A B ��� ��� '" CD ." 0 Xl r-- ���ll i ��� eye position I 300ms I Figure 1 Pursuit eye movements of a monkey in response to sinusoidal target motion at 0.6 Hz, peak-lo-peak amplitude 20 deg (panel A) and step-ramp target motion at 30 degjs (panel B). The eye velocity records were obtained by electronic differentiation of the eye position records. The rapid deflections in eye velocity are associated with saccadic eye move�� ments. In panel B, the upward arrow indicates the initiation of pursuit, which preceded the first saccade. Properties of the Eye Movement Most investigators have found that the onset of smooth target motion evokes pursuit after late ncies of 80 to l3ms. 0 Pursuit latency can be as short as 65 ms for a small target (Lisbe rger & Westbr ook 1 9 85) and depends on a numbe r of visual parameters, including the target's lumin�� ance, size, and initial position in the visual field (Li sberger & Westbrook 1985, Tychsen & Lisbe rger 1986a). Pursuit eye movements are most effective when target speed is relatively slow. For ramp target motion, smooth eye velocity is usually unable to keep up with target mo tion at spe eds in excess of30 deg/s (Robinson 19 65, Fuchs 1967a). Thus, pursuit ensures optimal vision only when the target is moving slowly, since visual acuity starts to decrease when the velocity of retinal image sli p exceeds 3 deg/s (Westheimer & Mc Kee 19).For 75 sine wave target motion, smooth pursuit is excellent when the amplitude and fr equency of target motio n are low (i.e. less than 1 . 0 Hz). When the amplitude or fr equency is increased, smooth eye velocity becomes smaller than and lags increasingly behind target velocity (e.g. Fuchs 19 67b, Col�� lewi jn & Tamminga 1984 , Yasui & Young 1 9 84, Lisberger et al 1 98 1 ). Because the smooth eye movement is no t able to keep up with the target, the tracking is increasingly punctuated with saccades. Althotargetsugh moving that fast cannot be tracked accurately, pursuit eye velocity as high as 1 8 0 deg/s has been reported in humans (Lisberger et aI 198).1 Pursuit is specialize d for tracking slo wly moving targets and is inad- by Universidad de Chile on 07/27/11. For personal use only.

100 LISBERGER, MORRIS & TYCHSEN equate for other fu n ctions, which are sub served by the other classes of eye movements. For example, the long delays and slo w eye speeds achieved by pursuit make it too slo w to stabiliz e the retinal images of stati onary ob jects effectively during head turns. This fu nction is sub served largely by the vestibul o-ocular reflex, which has a late ncy of 1 4 msec (Lis berger 1984) and is accurate for head turns at speeds in excess of 300 deg/s (Keller 1 9 78). Similarly, pursuit moves the eyes too sl owly to be eff ective for scanninvisual g the scene or to point the eyes at station ary eccentric objects. These fu nctions are subserved by saccadic eye mo vements, which have relatively long latencies of about 200 ms, but can execute a shift in the position of gaze that (a ) covers 20 deg in about 40 msec in monkeys and 70 msc c in humans and (b) brings the eye accurately to the desired position. The reader is ref erred to several reviews of the neural subsystems sub�� serving the vestibulo-ocular reflex (Miles & Lisberger 1981, Ito 198or2) saccades (Fuchs et al 1985). Operational Definition of Pursuit THE ADEQUATE STIMULUS FOR PURSUIT Under normal viewinconditionsg outside the laboratory, pursuit is generated only when there is a moving target. In most in dividuals, voluntary attempts to "move the eye fr om left to right as smoothly as possible" produce a staircase of small saccadic eye movements. In the in tersaccadic in tervals, which last about 200 msec each, the eyes are stati onary. These anecdotal observations show that a visual stimulus is necessary to elicit pursuit, but Rashbass ( 1 9wasthe 61) first to show that pursuit is a response to the motion of the target's images across the retina. He used the "step-ramp" target motion shown in Figure 1 B to generate a situation in which target position and target motion called for eye movements in opposite directions. This stimulus evoked pursuit that took the eyes in the direction of target motion at latencies of about 100 ms. Saccadic eye movements had longer laten cies and took the eyes in the direction of target position. OTHER STIMULI FOR PURSUIT? The clear distin ction between the adequate stimuli for saccades and pursuit eye movements has been challeninged recent years. Pol a & Wyatt (1979) have sh own that station ary stimuli can evoke pursuit if the target is artificially stabiliz ed on the retina, slightly eccentric from the fov ea. Ho wever, this kin d of observation should not be taken as evidence that a stati onary eccentric target normally elicits pursuit. Human sub jects have proven to be extremely flexible in the use of their oculomotor systems, and one study (Cushman et a1 19 84) reports that they can learn to make a variety of smooth eye movem ents under conditions by Universidad de Chile on 07/27/11. For personal use only.

SMOOTH PURSUIT EYE MOVEMENTS 1 0 1 of image stabilization. Thus, some caution must be exerciinterpretingnise d the results of experiments that use prol onged periods of im age stabili�� zation. Al though visual inputs provide the main stimulus for pursuit, cognitive factors also play an important role. First, activation of the pursuit system requires that a movin g target be sele cted-targets can be ignored easily (Kowler et al 1 9 8 4). Second, the pursuit system is capable of prediction, so that trackin g is much more accurate when target motion is periodic than when target motion is unpredictable (St ark et al 19 62, Mich&ael Me lvill Jones 1 9 66). Third, humans can make smooth eye movements that anticipate a change in the position or velocity of a target (Kowler & Ste inman 19 79). Alth ough low in velo city (less than 1 . 0 deg/s), these drifts may be related to pursuit eye movements. Finally, humans can take advantage of "glo bal motion" to pursue targets that are perceived but not actually visible, such as the center of a rolling wheel that is marked only by several small lamps attached to its rim (S teinbach 1 9 76). PURSUIT VS THE OPTOKINETIC RESPONSE In our review, we concentrate on the neural subsys tem that generates smooth pursuit of small objects as they move across a structured background. Sm ooth eye movements can also be evoked by large or fufield, ll moving patterns. These eye movements may share some of the neural mechanisms used by pursuit, so we briefly compare the smooth eye mo vements evoked by large and small targets. A horizontally moving, fufield, ll vertically-st ripe d drum generates opto�� kinetic nystagmus with slo w phases in the direction of stimulus motion and quick phases (sac cades) in the opposite direction. The smooth part of the eye movements has two compone nts (Cohen et al 1 9 77, Lisbeet rger aI 1981b ). One, the "slo w component ," is clearly different fr om pursuit. It takes seconreachfuamplitude,to ds to its ll appears be generated by neural networks in the brainstem, and is seen in sp ecies that do not pursue small objects. The other, "f ast component," shows several properties that parallel pursuit. It takes only a few hundred millise conds to reach its fu ll amplitude and is largest in primates, which have excellent pursuit. Lesions of the primary visual cortex (Zee et al 198or 6) the flocculus of the cerebellum (Zee et a1 1 9, 81 Waespe et a1 19 83) cause large deficits in both the fast component of the optokinetic response and pursuit, so we assume that similar brain path ways are used. However, it seems likely that diffe rent affer ent neurons are in volved. Many neurons in the visual cortex respond to the motion of a small target but do not respwell ond to the motion of the fufield ll stimuli that are require d to generate an optokinetic response (Allman et a1 19 85, Tanaka et al 1986). by Universidad de Chile on 07/27/11. For personal use only.