Biol. Cybern. 73, 83 93 (1995)
Biological
Cybernetics ~(:~Springer- Verlag 1995
The influence of structured visual backgrounds on smooth-pursuit initiation, steady-state pursuit and smooth-pursuit termination Heino Mohrmann, Peter Thier Sektion fi.ir Visuellc Scnsomotorik. Neurologische UniversitStsklinik, Hoppe-Se)lcr-glrasse 3. D-72076 T{ibingen, G e r m a n y Received: 24 October 1994/Accepted in revised form: 16 December 1994
Abstract. Smooth-pursuit eye movements were recorded in two rhesus monkeys in order to compare the influence of structured visual backgrounds on smooth-pursuit initiation, steady-state pursuit and pursuit termination. Different target trajectories were used in order to study smooth-pursuit initiation and termination. The influence of visual backgrounds on pursuit initiation was characterized by recording ocular responses elicited by stepramp target displacements starting from straight ahead. Pursuit termination was characterized by analysing the transition from steady-state smooth-pursuit to fixation when a centripetally directed target ramp was terminated by a small target step in the direction of the ramp as soon as the target had come close to the straightahead position. The quantification of steady-state pursuit was based on ocular responses elicited by either paradigm. In accordance with previous work, we found that the onset of smooth-pursuit eye movements was delayed and initial eye acceleration reduced in the presence of a structured visual background. Likewise, mean eye velocity during steady-state pursuit was reduced by structured visual backgrounds. However, neither the latency nor the time course of smooth-pursuit termination was altered when the homogeneous background was replaced by a structured visual background. The lack of sensitivity of pursuit termination to the presence of visual structured backgrounds supports a previous contention that pursuit termination is mediated by a process which is different from the ones mediating smooth-pursuit initiation and steady-state pursuit. The absence of any noticeable effect of structured backgrounds on pursuit termination suggests that at least the fast component of the optokinetic reflex is suppressed during pursuit termination.
1 Introduction Human and non-human primates make saccadic eye movements in order to move the image of an object of
Correspondence to:
P. Thier
interest into the fovea. Once the object has been foveated, smooth-pursuit eye movements are used to compensate for slow movements of the target relative to the head of the subject, thus keeping the target image on the fovea. Fl'om a computational point of view, the system underlying smooth pursuit may be understood as a control system minimizing retinal error signals such as retinal image slip. Increasingly complex solutions have been suggested which are able to account for most of the observed features of smooth-pursuit eye movements (Young et al. 1968: Robinson et al. 1986: Lisberger el al. 1987). The same control system which is able to track slowly moving targets could, at least in principle, be able to clamp the eyes to a stationary target. In this case, fixation of a stationary target would be a special case of smooth pursuit at zero target velocity. However, the evidence available does not support this possibility. When it slowly moving target comes to a sudden stop, eye velocity starts to decay exponentially with a latency of about lOOms find a time constant of about 901ns {Luebke and Robinson 1988) . If, on the other hand, smooth-pursuit eye movements are initiated by a sudden onset of target motion, one does not find a simple exponential increase of eye velocity. Rather, pursuit initiation is often characterized by high-frequency velocity oscillations ('ringing') of the eyes superimposed on a monotonic increase of eye velocity. Comparable oscillations can also be observed during the ensuing steady-state pursuit (Robinson 1965: Robinson et al. 1986: Goldreich et al. 1992). Oscillations of the fl-equency observed during pursuit initiation and steady-state pursuit are predicted both by the pursuit model suggested by Robinson find later models put forward by lLisberger and coworkers (Krauzlis and Lisberger 1989; Goldreich et al. 1992). They' basically reflect the feedback character of the system. The model's response to a sudden drop of target velocity to zero is characterized by the same oscillations observed during pursuit initiation and steady-state pursuit in monkeys and man (Robinson 1965: Robinson et al. 1986; Goldreich et al. 1992). Their absence during pursuit termination suggests that this part of the smooth pursuit response is not simply a response of the pursuit system to a target velocity step back to zero. On the other
84 hand, it cannot reflect a simple passive relaxation of the oculomotor mechanics either, since the observed time constant of smooth-pursuit termination is much shorter than the time constant derived from the known properties of the oculomotor plant (Robinson 1965). Unlike smooth-pursuit initiation and steady-state pursuit, smooth-pursuit termination has only been studied with visual targets moving in the presence of dark or homogeneous backgrounds. This is a configuration which obviously differs from the one prevailing under natural conditions, where objects of interest usually move through a highly structured visual environment. From previous work we know that structured visual backgrounds impair smooth-pursuit initiation (Keller and Khan 1985; Kimmig et al. 1992; Mohrmann and Thier 1992) as well as steady-state smooth pursuit (Collewijn and Tamminga 1983; Yee et al. 1983; Mohrmann and Thier 1992; Ilg et al. 1993). While the mechanisms of the suppression of smooth-pursuit initiation are not fully understood, it seems likely that they reflect properties of visual motion processing for smooth pursuit. The suppression of steady-state pursuit, on the other hand, is most probably a consequence of an incomplete suppression of the optokinetic reflex (OKR), activated by the pursuit-induced displacement of the background image and directed opposite to the smoothpursuit eye movement (Mohrmann and Thier 1992). If the O K R were fully activated during smooth-pursuit termination, the latter would be shortened if eye movements are carried out in the presence of a structured background. Even if the O K R were only incompletely suppressed, as during steady-state pursuit, we would expect to see a more rapid deceleration of the eyes during pursuit termination against a structured background. In order to test this expectation, we compared smooth-pursuit termination in the presence of a homogeneous background with pursuit termination against structured backgrounds. Smooth-pursuit initiation and steady-state pursuit served as controls, demonstrating the effectiveness of the visual backgrounds used. Our finding that, unlike pursuit initiation or steadystate pursuit, pursuit termination is not affected by the presence of structured backgrounds supports the previous contention that pursuit termination is mediated by a process which is different from that mediating smooth pursuit initiation and steady-state pursuit.
2 Methods
2.1 General procedures Two adult rhesus monkeys (Maccaca mulatta L.), one male and one female, were used in our experiments in full accordance with the current German law regulating animal welfare. The monkeys were trained to foveate a small laser target presented on a tangent screen independent of the target position, its speed or direction of movement by applying the method of Wurtz (1969). This method requires the monkey to release a lever in response to the dimming of the laser target at random intervals (mean
interval: 7 + / - 4 s). Each correct response was rewarded by the release of a drop of water or apple juice, depending on the preference of the individual monkey. The willingness of the animals to work for juice or other kinds of liquids was stimulated by restricting free access to water prior to the experiment. Daily weight controls and supplements of extra fluid, if necessary, guaranteed adequate hydration at any time. Moreover, monkeys had access to ad libitum amounts of water and fresh fruit on weekends, when no experiments were carried out. As soon as the monkeys had acquired the necessary level of performance in the dimming paradigm, a search coil for the recording of eye movements and a head holder for the painless restraint of the head were implanted under general anaesthesia using methods which have been described in detail elsewhere (Judge et al. 1980; Thier et al. 1988). Training was resumed after postoperative recovery, and the experiments were started as soon as the presurgical level of performance was achieved again.
2.2 Stimuli and paradigms The laser spot which served as the target had a diameter of 10' of arc and a luminance 2.5 cd/m z above background. It was back-projected onto the tangent screen from x/y-galvanometers, which were under computer control. The background was either homogeneous or structured. Two types of background patterns, back-projected like the target, were used. The first one was a random dot pattern ('Julesz-pattern' = JP) made up of bright or dark squares (size of individual square: 0.36 ~ x0.36 ~ 1.1~ 1.1 ~ or 3.6~ 3.6~ the second one was a circular grating with a spatial frequency of 1, 0.25 or 0.0625 periods/deg. The mean luminance of a given structured background was adjusted to that of the homogeneous control. Two levels of mean luminance, 0.05 and 0.5 cd/m z, were studied. The contrast between the bright and dark parts of the background was always 0.8. Pursuit initiation and pursuit termination were studied by means of two paradigms. Pursuit initiation was evoked by step-ramp movements of the target, starting from straight ahead (Fig. 1). After a period of stationary fixation of variable duration (1-1.5 s), the target stepped into one of four directions corresponding to up, down, left or right, followed by a ramp-like movement of the target opposite to the direction of the preceding target step at either 5 ~ or 20~ The target step served to reduce the probability and size of catch-up saccades by minimizing the retinal position error at the time of pursuit onset. This was achieved by choosing a step amplitude of 1~ for 5~ ramps and 4 ~ for 20~ ramps. The direction of the ramp, its velocity and duration, the last varying between 0.5 and 1.0 s, were randomized. Pursuit termination was studied using the paradigm introduced by Luebke and Robinson (1988; comp. to Fig. 1B as an example). After a period of stationary fixation in the straight-ahead position, the target was stepped into an excentric location above, below, right or left from straight ahead. From there, the target moved back towards straight ahead at either 5~ or 20~ The amplitude
10
A
pursui initiation// t
?. tO
eye
(/} 0 (3k
~5 This was achieved in a large percentage o f trials by a terminal target step of 3.2 for 2 0 / s r a m p s and 0 . 8 for 5 /s ramps. A daily e x p e r i m e n t a l session consisted of 16 blocks of 160 trials each. Each block was characterized by the structure and mean l u m i n a n c e of the b a c k g r o u n d and by a c o m m o n pursuit p a r a d i g m (initiation versus t e r m i n a tion). Within a given block, direction and velocity were presented in an interleaved r a n d o m fashion. 2.3 Data recordin~ and proce,~'xi~zg
target -5 -0.5
~'-/ ()
--
B
time[s] ./,"'~
target
I / "
0'.5 corrective ~t,,,,," saccade . . . .
"t
--~
//
0-
?, O
-5-
///
pursuit termination
-10
os
1.d
time
[s]
is
zo
Fig. 1A, B Examples of step-ramp sequences used to evoke smoothpursuit eye movements. A Step-ramp sequence preferred for the study of pursuit initiation. The target starts from straight ahead. At time 0, it makes a 4 step to the left. followed by a 20 /s ramp to the right. Note the absence of a catch-up saccade during pursuit initiation. B Pursuit termination. At time 0 (not shown), the target makes a large step to the lefL followed by a 20 /s ramp back towards slraight ahead. At the end of the ramp, a target step of 3.2 further to the right is added. This final target step in the direction of the preceding ramp was added in order to reduce the probability of an early corrective saccade. Note that in this particular example the terminal step was too small to be able to compensate for the overshoot of the eyes over the target. A corrective saccadc (arrow) brings the eye back to to the target.
of the initial target step d e p e n d e d on r a m p velocity. It was r a n d o m l y varied between 8 and 2 2 ' for 2 0 ' / s r a m p s a n d between 0 . 5 a n d 7: for 5'7s ramps. The trial was t e r m i n a t e d by a final target step in the direction of the p r e c e d i n g ramp, ending at r a n d o m i z e d l o c a t i o n s between - 2 a n d + 2" a r o u n d straight a h e a d in o r d e r to prevent the a n i m a l s from a n t i c i p a t i n g the s t o p of the target m o v e m e n t . The final step in target p o s i t i o n served to minimize the o v e r s h o o t , thus reducing the p r o b a b i l i t y of corrective saccades interfering with pursuit t e r m i n a t i o n .
Eye position was s a m p l e d at 200 Hz and filtered digitally using a phase-shift-free, recursive, 2 n d - o r d e r , low-pass tilter ( H a m m i n g 1987) with a c o r n e r frequency of 50 Hz. Eye velocity and acceleration were deriYed from the filtered eye position records by nurnerical differentiation, a p p l y i n g the t w o - p o i n t central-difference a l g o r i t h m suggested by Bahill and M c D o n a l d (1983). The 3-dB c o r n e r frequency c h a r a c t e r i z i n g the frequency content of the eye velocity record yielded by this algorithrn was 7.4 Hz for eye m o v e m e n t s driven by 5 / s target r a m p s and 11.1 Hz for 2 0 / s target ramps. The difference in the c o r n e r fiequencies reflects the fact that o u r estimates of the m a x /mum of the third deriwttive of eye position, an essential p a r a m e t e r in the t w o - p o i n t central-difference a l g o r i t h m , d e p e n d e d on target velocity. Even under o p t i m a l conditions, eye m o v e m e n t records of visual t r a c k i n g are c o n t a m i n a t e d b? occasional small corrective saccades or fast eye movernents, a c c o m p a n y i n g eye blinks (e.g. Collewijn et al. 1985). In o r d e r to quantify s m o o t h - p u r s u i t eye l'novements, their tirne course and d e p e n d e n c e on the visual b a c k g r o u n d , the c o n t r i b u t i o n of these fast eye m o v e m e n t s had to be eliminated first. We basically used an a c c e l e r a t i o n criterion in o r d e r to s e p a r a t e saccades and blinks from s m o o t h - p u r suit eye movements. A saccade or a b l i n k - r e l a t e d eye m o v e m e n t was a s s u m e d when the acceleration exceeded 250 /s x. While s m o o t h pursuit could o c c a s i o n a l l y reach accelerations exceeding this threshold, such segments of pursuit could usually be differentiated easily frorn saccades or b l i n k - r e l a t e d eye m o v e m e n t s by the characteristic acceleration profiles of the latter two, characterized by the close sequence of two (saccades) or three {blink-related eye m o v e m e n t s ) acceleration peaks of a l t e r n a t i n g sign. if a fast eye m o v e m e n t was p i n p o i n t e d in this way, its c o n t r i b u t i o n to eye position was eliminated, and the eye p o s i t i o n between the beginning and the end of the fast eye m o v e m e n t i n t e r p o lated linearly. As first pointed out by R o b i n s o n (1965), the variability between individual pursuit m o v e m e n t s can be very large, possibly o b s c u r i n g essential aspects of the response. We, too, found c o n s i d e r a b l e differences between individual trials, such as differences in the latencies of pursuit initiation or t e r m i n a t i o n . M o r e o v e r , the time course of pursuit was very variable, due to occasional s u d d e n setbacks of eye velocity, which could occur as early as 50 ms and lasting up to 40 ms (f:ig. 2A). Therefore, in o r d e r to eliminate the possibly c o n f o u n d i n g impact of intertrial variability Oll o u r analysis, we decided
86
A 2O
slope difference
target [ k
0
o
I Ii x - / ' '
/
,=,
I[
f
'
10
eye
II
first linear segment
i
,,
'~
[\,.
/",,'~
/
i i t .~,/t ,~ / I l i i
/
v'w",~ I
iJ
',j
~/ 0.1 '
'
'
/
V
"I
'
....
pursuit initiation 0'.2
0.3 '
'
'
o14
'
0.5
time Is]
B
pursuit termination slope difference
0
,
AI ~I
ill I
---
%
--'~ -10
J i J
--eyeV
........start -20
of
pursuit
termination
target 0
0.1
0.2
0.3
0.4
time [s] Fig. 2A, B Single trials of smooth pursuit were quantified by adopting piecewise regression techniques. A Pursuit initiation. The eye velocity record was fitted by a sequence of linear segments whose lengths were determined by a procedure based on the location of the peaks in the curve of differencesof the slopes of forward and backward regressions ('slope difference') calculated for a given sample point. The starting point of the first linear segmentbetween 60 and 300 ms after start of the target ramp served as an estimate of the latency of pursuit onset; its slope was taken as an estimate of initial pursuit acceleration and its length as duration of pursuit initiation. B Pursuit termination. For a given sample of eye velocity, forward and backward regression and the difference of their slopes were calculated ('slope difference').Since eye velocity during pursuit termination declines to zero exponentially, the maximum of the curve of differenceshaving a sign opposite to that of the target velocity corresponds to the onset of pursuit termination. The duration of pursuit termination was characterized by z, the time constant of the exponential function fitting the decay of eye velocity to evaluate only individual trials rather than to quantify average pursuit responses. In order to quantify pursuit initiation in single trials, we devised a piecewise regression algorithm, which allowed us to determine the latency of the responses. The idea underlying this algorithm was to fit a sequence of linear segments of varying slope and duration to the desaccaded eye velocity record (comp. Fig. 2A). Pursuit
initiation was then determined by finding and quantifying the first linear segment which was significantly different from the preceding ones fitting the pre-pursuit baseline. The crucial step in this algorithm is to break down the eye velocity record into segments of appropriate length. We chose the following iterative procedure in order to find a satisfying segmentation. For each given eye velocity sample, two linear regressions were calculated. The first one included the given sample point and the seven preceding ones (backward regression), while the second one included the given sample point and the seven following ones (forward regression). We then calculated a curve of differences of the slopes of the forward and backward regressions which enabled us to adjust the lengths of the segments. This was done by detecting global maxima in the curve of differences, which were then used to redefine segment lengths according to the intervals delimited by two neighbouring global maxima of either sign. The regressions were then recalculated on the basis of the modified segment lengths, and a second-order curve of differences was derived which like the first one served to readjust the segment lengths. This procedure was reiterated until a stable segmentation of the eye velocity record was achieved. Pursuit onset was then determined by finding the first linear segment between 60 and 300 ms after start of the target ramp whose slope was significantly larger than that of the preceding segment and with the same sign as the target velocity. The starting point of this segment led to an estimate of the latency of pursuit onset; its slope was taken as an estimate of initial pursuit acceleration, and its length as duration of pursuit initiation. A simplified version of this algorithm was used for quantifying pursuit termination (Fig. 2B). In order to determine the latency of the onset of pursuit termination, it was sufficient to calculate fixed 40-ms forward and backward regressions for a given sample point and then the curve of differences of their slopes. Since eye velocity during pursuit termination decays to zero exponentially, the m a x i m u m of the curve of differences having a sign opposite to that of the target velocity corresponds to the 9onset of pursuit termination. The location of this maximum relative to the stop of the target ramp gave the latency of the onset of pursuit termination. In order to exclude trials with anticipatory eye movements and trials confounded by decreased attention, only maxima between 30 and 300 ms after the stop of the target ramp were considered. The time course of pursuit termination was characterized by the initial eye deceleration, given by the slope of the linear regression line fitted to the eye velocity record during the first 40 ms of pursuit termination and the time constant ~ of the exponential decay of eye velocity. 2.4
Statistics'
We applied a four-way analysis of variance (ANOVA) with one of the factors being type of background (homogeneous vs. structured) in order to test whether or not pursuit performance was significantly influenced by the presence of a structured visual background. Data
87
obtained with the two different types of background, the JPs and the circular gratings were pooled for this analysis, thereby increasing the probability of demonstrating even subtle effects of the background on smooth pursuit. Pooling seemed to be allowed since a preceding comparison of the influences of the two types of background patterns had shown that they were roughly equipotent, a finding that is consistent with previous work (Collewijn and Tamminga 1983). The other three factors were the velocity of target displacement (5'/s vs 20/s), its orientation {horizontal vs vertical, i.e. after the pooling of up and down and right and left, respectively) and the mean luminance of the background (0.05 vs 0.50 cd/mZ). The dependent variables characterizing pursuit initiation were latency of pursuit onset, initial eye acceleration, mean eye acceleration during the first 100 ms of smooth pursuit as given by the difference of eye velocity 100 ms after and at pursuit onset. Steady-state pursuit was characterized by its gain, defined as mean eye velocity between 200 and 400 ms after pursuit onset divided by target velocity. The dependent variables used to characterize pursuit termination were its latency, the initial eye deceleration and the time constant z. The gain of steadystate pursuit in the termination paradigm was characterized by mean eye velocity in the 50-ms interval preceding onset of pursuit termination. Significant interactions were further analysed by Scheff6 post-hoe comparisons. Effects were considered to be statistically significant if the probability was less than 1%.
m o n k e y E,
target
20
~
o
~o
A
o
0
0.1
0.2
0.3
0.4
time [S]
20
?,
I
target
10
B
3 Results
3.1 Pursuit initiation and steady-state purs'uit In accordance with previous work, we found that a structured background reduced both smooth-pursuit initiation and steady-state pursuit. This is exemplified in Fig. 3A, which compares the first 400 ms of smooth pursuit elicited by a 2 0 / s upward target ramp in front of a homogeneous background with pursuit over a structured visual background, which was a 1 period/deg 7 0 : x 7 0 circular grating. Averaged desaccaded eye movements aligned with respect to eye movement onset obtained from one of the monkeys are shown. As early as 10 to 20 ms after pursuit onset, the two curves started to diverge. The suppression by the structured background increased further over time, reaching a maximal reduction of eye velocity of the order of 20% only in the steady-state part of the response, clearly after 200 ms. Due to the gradually increasing influence of the background effect, the mean eye acceleration for the first 100 ms of pursuit was more reduced than the initial eye acceleration. This difference is worth mentioning since previous studies of pursuit initiation have used the firstmentioned measure, i.e. mean eye acceleration for the first 100 ms. The same dependence of the background-induced suppression on response time seen in this single case is also displayed by the grand averages based on both monkeys, 2 0 / s ramps and all background conditions,
0
.1
0.2
0.3
0.4
time [s] Fig. 3. A Smooth-pursuit initiation in front of a homogeneous background (empty square) and in front of a structured background (square with circular ~lratin~l). The background panern ~ a s a 70 x 70 circular grating of 1 period/deg. Mean background luminance 0 5 cd/m e, contrast of grating m = 0.8. Average curve based on 17 trials (homogeneous background) or 11 trials (structured background) of desaccaded smooth-pursuit eye movements of monkey E, evoked by 20 ,s upward step-ramps. B Grand avcrages of desaccadcd eye movements of both monkeys comparing smooth-pursuit initiation in front of a structured background mm-emptl" squares: 1851 trials, till background patterns and luminances pooled) and a homogeneous background (empty square: 1077 trials, all luminances pooled). Upward, downward, right or left 20 s stcp-ramp movements of the target
which are depicted in Fig. 3B. Again, responses are aligned on eye movement onset. Note that the mean eye velocity in these grand averages is well below the target velocity even for target motion across the homogeneous background. This was mainly due to the generally poorer responses to vertical target motion in both monkeys used. Replacing the homogeneous background by structured patterns reduced pursuit eye velocity increasingly by tip to 8% beyond 200ms after pursuit onset. In contrast to the case shown in Fig. 3A, tile first 70 ms
88
of smooth pursuit, quantified by the slope of the first linear segment fitted to the eye velocity record ( = initial eye acceleration) was not obviously affected by the structured backgrounds. On the other hand, there was a slight (4%) reduction of the mean eye acceleration during the first 100ms of pursuit. A qualitatively similar, albeit quantitatively smaller suppression was found for the smaller (5~ of the two target velocities used (not shown). Since the influences of random dot patterns and circular gratings were not very different, we decided to pool data obtained with these two kinds of backgrounds for the statistical analysis of pursuit performance by a four-way ANOVA with the main factors target velocity (5~ vs 20~ orientation (horizontal vs vertical), luminance (0.05 vs 0.5 cd/m 2) and background (homogeneous vs structured). The results of the statistical analysis are summarized in Table 1, and in addition, bar graphs illustrating the mean background-induced change affecting the various dependent variables are shown in Fig. 4, separately for the lower (A) and the higher (B) target velocity. Average changes plotted in this figure were calculated a s (/)str - - /)hom)//)hom " 100, where Ust r is the mean of a given dependent variable for pursuit in the presence of structured backgrounds and Vhomthe corresponding mean for pursuit in the presence of a homogeneous background. As already suggested by the examples discussed before, in which the background-induced suppression of smooth eye velocity developed gradually, ANOVA yielded a highly significant effect of background type, indicating a suppression of mean steady-state pursuit velocity (VEL) and mean eye acceleration (A100), the reduction of both being 6% (average across all conditions). Unlike what may have been suggested by the
grand averages shown in Fig. 3B, which were based on 20~ target ramps, the reduction of initial eye acceleration was still highly significant, although it was clearly smaller (4%) than the reduction of mean eye acceleration and steady-state pursuit. Figure 4 suggests that this might be due to the contribution of a much larger background-induced suppression of the initial eye acceleration elicited by 5~ target ramps (8% compared with 2% for 20~ ramps), a possibility which will be considered in more detail below. The structured visual background not only affected the amplitude of pursuit initiation and steady-state pursuit but also the latency of pursuit onset. Smooth pursuit in front of a structured background was significantly delayed by 13 ms on average. On the other hand, the duration of pursuit initiation, as quantified by the length of the first linear segment, was on average 70 ms, independent of the presence or absence of a structured background (P = 0.24). Not unexpectedly, we also found main effects of target velocity, orientation and luminance, with pursuit performance being worse for the higher target velocity, for vertical target movement and higher background luminance (see Table 1 for details). The background effects mentioned before were not independent of other factors, as indicated by significant interactions between luminance (LUM) and background (BGR) for the dependent variables latency (LAT) and mean eye acceleration (A100) on the one hand and between target velocity (VLA) and background for the dependent variable steady-state velocity (VEL) on the other. Further analysis of these significant interactions by post-hoc Scheff6 comparisons demonstrated that the delay of pursuit onset was larger for the higher luminance (16 ms), though still significant for the lower background luminance (8 ms).
Table l. Statistical analysis of smooth-pursuit initiation by a four-way A N O V A with the factors velocity (VLA; 5 / s vs 20'/s), orientation (ORI; horizontal vs vertical), m e a n background luminance (LUM; 0.05 vs 0.5 cd/m 2) and background (BGR; homogeneous vs structured). Dependent variables were latency (LAT) of smooth-pursuit onset, initial smooth pursuit acceleration (ACC) as estimated by the slope of the first linear segment, and the mean acceleration during the first 100 ms of smooth pursuit (A100), which in several previous studies on pursuit initiation had been used to characterize eye acceleration during pursuit initiation. The fourth dependent variable, VEL, mean eye velocity between 200 and 400 ms after pursuit onset, was chosen in order to characterize steady-state pursuit in this particular paradigm. Delta gives the percentage difference between the averages of a dependent variable for the two conditions of a given factor. Increases are positive, decreases negative. Effects were considered to be significant if the probability was less than 1% ANOVA: Main effects Independent variable
Dependent variable EAT (s) Means
VLA (1) (2) A (%) ORI (1) (2) A (%) L U M (1) (2) A(%) BGR (1) (2) A(%) Mean square of error
0.146 0.153 + 5 0.150 0.150 0 0.147 0.152 +3 0.143 0.156 +9 0.00177
P level 0.0000
0.9462
0.0000
0.0000
ACC ( / s 2) Means 50 124 + 148 105 69 - 34 89 85 -4 89 85 --4 1789.9
P level 0.000
0.000
0.0000
0.0019
A100 (~/S 2) Means 32 99 + 209 78 53 -- 32 66 64 -3 67 63 --6 802.9
P level 0.000
0.000
0.221
0.0000
VEL ('/s) Means 4.7 17.6 + 283 11.1 11.0 - 1 11.3 10.8 -4 11.4 10.7 --6 9.43
P level 0.000
0.294
0.0000
0.0000
89
v = 5 ~
m o n k e y E,
10 E)3
2oi
m
r
I
0
0~ t--
~>' 10
0
O
-10
o
A
>
"~ -
"-_-~ .g -~ g
=o E'$ 8
A o
>o
target 0
v = 20 ~
10 4
0.1
012
'
I
--0~3
0.4
time [s]
II
tt~ c(o Oa
|
t~
E
R)
o
CL
~
m
-10 i
B
~>" lo
g
g
z-
:.,=
"-
"~o
~--~o
>
E -~ 0
Fig. 4A, B. Purstnt initiation paradigm. Bar plots of average background-induced changes of pursuit latency, initial eye acceleration, mean eye acceleration during the first 100 ms of pursuit and steadystate eye velocity (estimated as mean eye velocity between 200 and 400 ms after pursuit onset) shown separately for the lower (A) and the higher (B) target velocity. Mean change was calculated as (/:sir -- Fhoml//;ho,n" 100, where /~t~ is the mean of the variable for structured backgrounds and t,o., its mean for the homogeneous backgrounds. Significant changes (P < 0.01) are marked by asterisks
Mean eye acceleration was only significantly reduced in the presence of the dark structured backgrounds (11%), whereas bright structured backgrounds led to no change of this variable. In Fig. 4, asterisks identify the dependent variables affected significantly ( P < 0.01, Scheffa comparisons) by the introduction of a structured background for a given target velocity. For both target velocities the pursuit latency was significantly enlarged by structured backgrounds. While there had been a significant main elfect of the factor background on initial eye acceleration, the background influence turned out to be not significant, when considered separately for the two target velocities used. On the other hand, mean eye acceleration for the first 100 ms of pursuit was significantly reduced for both velocities. Finally, only the higher target velocity led to a significant inhibition of steadystate pursuit.
....
target
-
time [s] Fig. 5. A Smooth-pursuit termination in front of a homogeneous background (cmply ,squore) compared with that in fi'ont of a structured background (squm'v ~'itll circuhlr ~lroling). The background panern was a 70 x 7 0 circular grating of 1 perioddeg. Mean background luminance 0.5 cd:m-', contrast of grating m - 0.8. Averages of 11 trials (homogeneous background) and 7 trials (structured backgroundl of dcsaccadcd smooth-pursuit eye nlovemenls of monkey E, evoked by stcp-ramp-stcp sequences with a centripetally directed downward ramp of 2(1 s velocity. B Grand averages of desaccadcd smooth-pursuit eye m o s c m c n t s of both monkeys comparing pursuit tcrmi/qation in front of a h o m o g e n e o u s b a c k g r o t l n d (etnl~t.y squ~lre: 296 trials, all l u m i n a n c e s poolcd} will] that it] front of a structured b a c k g r o m l d (mm-emplt.~quares: 502 trials, all b a c k g r o u n d patterns and l u m i n a n c e s pooled). S l c p - r a m p - s t c p sequences xvith a 20 s centripetally directed ralnp in an upward, d o w n w a r d , right or left direction.
3.2 P m ' s z d l t e v m i n o l i o n
Pursuit termination in our experiments followed the time course described by Luebke and Robinson {1988). After et latency of about 100 ms, eye velocity started to fall back to zero according to a simple exponential profile. As exemplified by Fig. 5A, which shows plots of averaged desaccaded eye movements for one of our monkeys, the
90
only difference introduced by a 70~ 70 ~ 1 period/deg circular grating was a reduction of the steady-state eye velocity, from which the eyes started on their way back to stationary fixation. Neither the latency nor the initial deceleration or the time constant of pursuit termination was influenced by the structured background. This view is further supported by the grand averages shown in Fig. 5B, which are based on both monkeys, 20~ ramps and all directions. Statistical analysis of pursuit termination by a fourway ANOVA with the factors target velocity (20~ vs 5~ orientation (horizontal vs. vertical), luminance (0.05 vs 0.5 cd/m z) and background type (homogeneous vs structured) and the dependent variables steady-state eye velocity before onset of pursuit termination, latency of pursuit termination, its initial deceleration and time constant ~ confirmed the significant reduction of steady-state eye velocity, already found in the first experiment (see Table 2). Neither the latency nor the initial deceleration or the time constant of pursuit termination was significantly affected by a structured visual background. However, all variables characterizing pursuit termination except for latency depended on target velocity, mean background luminance and orientation (except for latency) as shown by significant main effects on every parameter chosen to characterize pursuit termination (see Table 2 for details). The only significant interaction was the one between the factors background type and target velocity for the dependent variable steadystate eye velocity. This interaction indicates that the background influence increased with target velocity. Analysis of this significant interaction by post-hoc Scheff~ comparisons showed that the reduction of steadystate pursuit velocity by the structured back-
grounds was indeed significant (P < 0.01) only for the 20~ target ramps. Figure 6 illustrates the interaction between the factors target velocity and background for the dependent variables latency of pursuit termination, initial eye deceleration, time constant of the decay of eye velocity, and steady-state eye velocity (calculated as mean eye velocity during the last 50 ms before the onset of pursuit termination) by plotting the mean background-induced change affecting the various dependent variables separately for the lower (Fig. 6A) and the higher (Fig. 6B) target velocity used. Average change was calculated as ( U s t r - - D h o m ) / / ) h o m " 100, where Ust r is the mean for pursuit in the presence of structured backgrounds and Vhom the corresponding mean for pursuit in the presence of a homogeneous background. The only significant effect (P < 0.01) seen was a reduction of steady-state velocity when the target velocity was 20~ whereas the steadystate velocity for 5~ target ramps remained unaffected by structured visual backgrounds. Although the time constant was shortened considerably in the case of the lower target velocity, this effect did not achieve significance. 4 Discussion
The major new finding reported here is that pursuit termination is surprisingly little affected by the visual properties of the background on which the target is presented. When the target tracked by smooth-pursuit eye movements comes to a full and unexpected stop, eye velocity drops to zero exponentially with a latency of 100 ms and a time constant of about 85 ms. Neither the
Table 2. Statistical analysis of smooth-pursuit termination by four-way ANOVA with the factors velocity (VLA; 5~ vs 20~'/s), orientation (ORI; horizontal vs vertical), mean background luminance (LUM; 0.05 vs 0.5 cd/m 2) and background (BGR; homogeneous vs structured). Dependent variables were latency (LAT) of the onset of smooth-pursuit termination, the initial eye deceleration as given by the slope of a 40-ms linear segment fitted to the eye velocity record starting at the onset of pursuit termination and the time constant of the exponential e -t/t used to fit the decay of eye velocity to zero. The fourth dependent variable, VEL, mean eye velocity in the 50 ms preceding the onset of pursuit termination was chosen in order to characterize steady-state pursuit in this particular paradigm. Delta gives the percentage difference between the means of a given dependent variable for the two conditions of a given factor. Increases are positive, decreases negative. Effects were considered to be significant if the probability was less than 1% ANOVA: Main effects Independent variable
Dependent variable LAY (s) Means
VLA (1) (2) A (%) ORI (1) (2) A (%) L U M (1) (2) A(%) BGR (1) (2) A(%) Mean square of error
0.112 0.105 - 6 0.108 0.109 + 1 0.112 0.105 --6 0.107 0.110 +3 0.00186
P level 0.0047 0.7031
0.0079 0.1683
DEC (~/s z) Means 63 165 + 162 127 102 --20 111 118 +6 116 113 --3 2419.0
P level 0.000 0.0000
0.0039 0.2001
TAU (ms) Means 75 85 + 13 74 86 + 16 85 75 -12 82 78 --5 1332.9
P level 0.0000 0.0000
0.0000 0.0377
VEL ('/s) Means 4.9 15.0 + 206 10.2 9.3 --9 9.7 9.8 +1 10.0 9.5 --5 12.38
P level 0.000 0.0000
0.4756 0.0024
91
V = 5 ~ 10 e-, to o'}
E
0a
o
ID O.
-10
A ~
o
._= -~
oo
o
8
v = 20 ~ 10t cO O')
oo
]
m "k
-10
B --
. g "~
g,
o
Fig. 6A, B Pursuit termination paradigm. Bar plots of mean background-induced changes of latency of pursuit termination, initial eye deceleration, time constant of the decay of eye velocity and steady-state velocity (calculated as mean cyc velocity in the last 50 ms preceding the onset of pursuit termination) shown separately for the lower {A) and the higher (B) target velocity Average change was calculated as (l?st r -- I?hom)/Phom ~ 100, where l:t ~ is the mean of the variable for s t r u c tured backgrounds and l:~..... its mean for the homogeneous backgrounds
latency of pursuit termination nor its initial deceleration or time constant depends on the structure of the visual background. On the other hand, if a stationary target turns to slow movement, the structured background has a clear impact on the features of the smooth-pursuit eye movement. The eye velocity is reduced and at least under the conditions prevailing in our experiments also the latency of eye movement onset response may be prolonged. Since our observations regarding pursuit initiation and steady-state pursuit correspond at least qualitatively to what has been reported previously (Collewijn and Tamminga 1983; Keller and Khan 1985; Kimmig et al. 1992), we will focus in the following discussion on pursuit termination and touch on pursuit initiation and steady-state pursuit only as relevant for an understanding of the independence of pursuit termination from the background. The suppressive influence of the structured visual backgrounds used in the present study on pursuit initia-
tion was comparatively slnall. One may therefore wonder if the absence of an inttuence on pursuit termination might simply be a consequence of a relatively small overall effectiveness of the visual backgrounds used. It could be that visual backgrounds with only a small influence on pursuit initiation may have been rendered totally, ineffective during pursuit termination due to differences in the prevailing visual conditions. Obviously, the major difference is that the background images are stationary during pursuit initiation, while the images of the same background patterns move according to the eye movements being made during pursuit termination. Although this difference cannot be dismissed, we do not think it is sufficient to explain the absence of a background effect on pursuit termination. The reason is that, while visual conditions are clearly, dissimi][ar for pursuit initiation and termination, they are much more similar regarding steady-state pursuit and pursuit termination, both being characterized by eye-movement-induced slip of the background image. However, only steady-state pursuit was influenced by the introduclion of visual backgrounds, while pursuit termination remained unaffected. Moreover. our data point to even stronger influences of structured visual backgrounds on steady-state pursuit than on pursuit initiation. Anothe, r possible explanation for the absence of background influences on pursuit termination is that comparatively small systematic influences not substantially smaller for pursuit termination than for pursuit initiation might be hidden by the somewhat larger noise for pursuit termination. If this were the case, we would expect, firstly, that the variability of pursuit termination should be much larger than the variability of pursuit initiation and, secondly, to find at least a trend suggesting a difference of some kind between pursuit termination in front of structured backgrounds and the homogeneous control. Actually, a closer look at the results of the statistical analysis showed (Tables 1, 2) that the contribution of unsystematic variance to the overall wuiance of the dependent variables latency and velocity was somewhat larger for pursuit termination than for pursuit initiation. However, as indicated by the virtually identical averages, there was not the slightest trend towards a difference between pursuit termination in front of a structured background as compared with pursuit in front of the homogeneous background. In conclusion, these considerations suggest that the finding of an independence of pursuit termination of the background is valid. It is obviously of interest to ask why the suppressive influence of the structured backgrounds used in the present study was smaller than the one reported by Keller and Khan (1985) and Kimmig et al. (1992). For instance, the former authors reported reductions in initial smoothpursuit velocity as large as 50%, whereas comparable changes in our own study were only' on the order of 20%. There are a ntunber of methodological differences between the three studies which might account for the smaller reduction of initial pursuit eye velocity found here. Among others, there are differences in the structure and size of the backgrounds used, the luminances of background and target, and the algorithms used for
92 the quantification of smooth pursuit. The major and probably least controlled variable, though, is the overall performance level of the monkeys involved in the three studies. That overall performance is a variable influencing the background-induced suppression of pursuit initiation is suggested by ongoing experiments in our laboratory. They show that those monkeys whose pursuit is most reduced by the introduction of structured visual backgrounds are the ones whose pursuit performance in front of a homogeneous background is the worst. Actually, we have seen well-trained monkeys showing a pursuit initiation in front of the most effective structured visual background that is not much different from pursuit initiation in front of a homogeneous background. Another probable factor is background luminance. In our experiments the mean luminance of the structured background was the same as that of the homogeneous control. On the other hand, both Keller and Khan (1985) and Kimmig et al. (1992) used homogeneous backgrounds which were darker than the structured backgrounds presented and also darker than the backgrounds used in our study. However, darker control backgrounds will increase the contrast between target and background. This is relevant since Lisberger and Westbrook (1985) have shown that a larger contrast between target and background facilitates pursuit initiation, thus leading to higher initial eye acceleration. This could also be seen in our own data, which showed relatively poorer pursuit initiation and steady-state pursuit for brighter backgrounds yielding less contrast relative to the target (ANOVA: main effect of luminance, see Table 1 for details). In other words, we hypothesise that the background-induced pursuit impairment in previous work is probably due to two factors, one being the presence of background structure, and the other being decreased contrast between target and surround. In summary, we would argue that the conditions prevailing in our experiments are apt to give more realistic measures of the effect of structured visual backgrounds on pursuit initiation. The lack of an influence of the visual background on pursuit termination is somewhat counter-intuitive. When a target makes a transition from slow motion to rest, one might expect that the eyes would respond as fast as possible in their attempt to keep the target image on the fovea. A structured visual background should be useful to this end. Steady-state pursuit as well as decreasing eye velocity during pursuit termination lead to a coherent displacement of the background image on the retina which is suitable to evoke an optokinetic reflex (OKR) whose direction is opposite to that of the smooth-pursuit eye movement. Actually, much of the reduction in steadystate eye velocity we and others (Collewijn and Tamminga 1983; Yee et al. 1983; Worfolk and Barnes 1992; Ilg et al. 1993) have seen during pursuit in front of a structured background is probably due to insufficient suppression of the OKR. If the suppression of the OKR were incomplete also during pursuit termination, we would expect a faster decline in eye velocity in the presence of a background, thus making the transition to stationary fixation more rapid. Hence, does the absence of a background influence indicate that the suppression
of the OKR during pursuit termination is perfect? When trying to answer this question, we have to bear in mind that the O K R of primates consists of two components, a fast and a slow one. The time constant of the slow component, usually thought to reflect the contribution of a phylogenetically older circuitry dominating the optokinetic response of afoveate mammals such as the rabbit (Collewijn 1969), is of the order of 3 s in the monkey (Robinson 1981). It is therefore too sluggish to respond to changes in eye velocity during pursuit termination. Thus, the lack of any background-induced influence on pursuit termination does not reveal whether the slow component of the OKR is activated during pursuit termination or not. On the other hand, the time it takes for the fast component of the OKR to reach a plateau is of the order of less than 100 ms (Cohen et al. 1977; Robinson 1981) and therefore certainly short enough to modulate pursuit termination. Given the lack of this modulation, we have to conclude that those parts of the OKR circuitry which are responsible for the fast component must be suppressed. There is agreement among workers in the field that the fast O K R component is basically a smooth-pursuit response (Robinson 1981). The fact that we do not find any modulatory contribution of a fast OKR response component therefore fully supports the view, first formulated by Luebke and Robinson (1988), that the smooth-pursuit system is deactivated during pursuit termination. We usually assume that visual reflexes including the O K R help to stabilize the image of a stationary target on our retina, and it is usually implicitly assumed that both the fast and the slow components of the OKR are involved. However, it is worth considering that the deactivation of the fast component of the O K R might prevail beyond pursuit termination. If this were the case, the compensation of any self-induced high-frequency retinal image slip would exclusively depend on vestibular reflexes. A final note relates to the time course and duration of pursuit termination. As first pointed out by Robinson (1965), the decrease of eye velocity during pursuit termination is far too quick to reflect a simple mechanical relaxation. If pursuit termination reflected a simple mechanical response to a velocity step back to zero, eye velocity should decrease with a time constant of about 200 ms and not the measured 85 ms (Robinson 1965). These quantitative considerations therefore suggest that pursuit termination is an active process. If the brain uses an active process in order to make the transition to stationary fixation anyway, why did it then not choose a much more radical strategy to brake the eyes? Although pursuit termination is faster than a mechanical relaxation of the eyeballs, it is still too slow to avoid overshoot of the target. Overshoot could at least principally be avoided if the eyes were braked by a short latency pulse-step force profile of appropriate polarity. However, for reasons which may be inherent to the circuitry available, it may not be possible to generate such saccade-like profiles with the short latencies required. The latency of visually guided saccades, the kind of eye movements which would most closely compare to this hypothetical, visually elicited braking profile,
93
is on the order of 200 ms. Only if attention is removed from the foveal target may shorter latency saccades Eexpress saccades') be found (Fischer and Boch 1983). On the other hand, the latency of pursuit termination, which unlike express saccades is characterized by attention to a foveal target, is on the order of only 100 ms. In other words, thanks to its short latency, pursuit termination may be the strategy leading to the minimum amount of overshoot possible. The consequences of the inevitable overshoot left may be further alleviated by a major feature of pursuit termination demonstrated by the present study. Both the latency and the time course of pursuit termination are independent of the properties of the prevailing visual background. It is tempting to speculate that this constancy may allow the system to pre-programme a saccade, correcting the inevitable overshoot much earlier than a saccade, which would have to be programmed on the basis of the sampled overshoot. Acknowled,clement~s. This work was supported by Deutsche Forschungsgemeinschaft (DFG) grants Th425-1/1 and Zrl:'9 1 to P.T. and a grant by the Graduiertenkolleg Neurobiologie to H. M. We wish to thank Johannes Dichgans. Peter Dicke and Uwe fig for comments on carlier versions of the m a n u s a i p t .
References Bahill AT, McDonald JD (1983) Frequency limitations and optimal step size for the two-point central difference derivative algorithm with applications to human eye movement data. IEEE Trans Bio med Electron BME 30:19l 194 Cohen B, Matsuo V, Raphan T (1977) Quantitative analysis of the velocity characteristics of optokinetic nystagmus and oplokinetic after-nystagmus. J Physiol (Lond) 270:321 344 Collewijn H (1969) Optokinetic eye movements in the rabbit: inputoutput relations. Vision Res 9:117 132 Collewijn H, T a m m i n g a EP (1983) H u m a n smooth and saccadic eye movements during voluntary pursuit of different target motions on different backgrounds. J Physiol (Lond} 351:217 250 Collewijn H, Van der Sleen J, Steinman RM {1985) H u m a n eye movements associated with blinks and prolonged eyelid closure. Vision Res 25:11 27 Fischer B, Boch R (1983l Saccadic eye movements after extremely short reaction times in the monkey. Brain Rcs 260:21 26
Goldreich I), Krauzlis R J, Lisberger SG (1992) [!fleet of changing feedback delay on spontaneous oscillations m smooth pursuit eye movemenls of monkeys. J Neurophysiol 67:625 638 H a m m i n g RW (1997) Digilale Filter. VCH Vcrlagsgcsellschafl, Weinheim llg U J, Brcmmer F, HolEnann K-P (1993) Optokinctic and pursuit system: a case report. Behav Brain Res 57:21 29 Judge SJ, Richmond BJ, Chu FC {1980) Implantation of magnetic search coils Ik)r measurement of eye position: an improved method. Vision Res 20:535 53R Kcllcr EL. Khan NS {19851 Smooth-pursuit initiation in thc presence of a textured background in monkey. Vision Res 26:943 955 Kimmig FIG, Miles FA, Schwarz U (1992) Effects of stationary textured backgrounds on lhc initiation of pursuit eye movements in monkcys..I Ncurophysio168:2147 2164 Krauzlis RJ, Lisbcrger SO (1989) A control systems model of smooth pursuit eyc movements with realistic cmcrgent properties. Neural Comput 1:116 122 kisbcrgcr SG. Wcstbrook LE {1985) Properties of visual inputs that initiate horizontal smooth pursuit eye movements in monkeys. J Neurosci5:1662 1673 Lisbergcr SG, Morris ILl, Tychscn L (1987) Visual motion processing and sensory-motor integration for srnooth pursuit eye movements. Annu Rev Neurosci 10:97 129 Luebke A E, Robinson DA (I 988) Transition dynamics between pursuit and fixation suggest different systems. Vision Res 28:941 946 M o h r m a n n H. Thier P 119921 Smooth-pursuit initiation and steady statc smooth pursuit in lhe presence of a lexlured background: effects of background structure aud training. Eur J Neurosci Suppl 5:278, no. 4221 Robinson DA (1965) The mechanics of h u m a n smooth pursuil eye movements. J Physiol 186:569 591 Robinson DA {1981i Control of eye movemenls, hq: Brooks VB (edl Handbook of physiology', Vol 11, Part 2, The Ncrw~us System. Amcrican Physical Soc., Bcthcsda, pp1275 1320 Robinson DA, Gordon .It,. Gordon SE (1986) A model of the smooth pursuit eye movement system. Biol Cybcrn 55:43 57 Thicr P. Kochlcr W. Buettner UW (19881 Neuronal activity m the dorsolatcral pontinc nucleus of the alert monkey nrodulated by visual stimuli and eye movements. Exp Brain Res 70:496 512 WorhHk R. Barncs GR 11992) Interaction of active and passive slow eye movement systems. Vision Res 90:589 598 Wurtz R H (1969) Visual receptive lields of striate cortex neurons m awake monkeys. J Nctn-ophysiol 32:727 742 Yce RI), l)aniels SA. Jones OW, Baloh RW, Honrubia V (1983} Effects of an oplokinctic background on pursuit eye movements. Invest O p h t h a h n o l V i s Res24:1115 1122 Young LR. Forster JD, Van Houttc N (19681 A revised stochastic sample model for eye tracking movements. 4th Ann NASA-Univ Conf Manual Control, l Jniversity of Michigan, Ann Arbor