Context-Specific Adaptation of Pursuit Initiation in Humans

Abstract

purpose. To determine if multiple states for the initiation of pursuit, as assessed by acceleration in the “open-loop” period, can be learned and gated by context.

methods. Four normal subjects were studied. A modified step-ramp paradigm for horizontal pursuit was used to induce adaptation. In an increasing paradigm, target velocity doubled 230 msec after onset; in a decreasing paradigm, it was halved. In the first experiment, vertical eye position (±5°) was used as the context cue, and the training paradigm (increasing or decreasing) changed with vertical eye position. In the second experiment, with vertical position constant, when the target was red, training was decreasing, and when green, increasing. The average eye acceleration in the first 100 msec of tracking was the index of open-loop pursuit performance.

results. With vertical position as the cue, pursuit adaptation differed between up and down gaze. In some cases, the direction of adaptation was in exact accord with the training stimuli. In others, acceleration increased or decreased for both up and down gaze but always in correct relative proportion to the training stimuli. In contrast, multiple adaptive states were not induced with color as the cue.

conclusions. Multiple values for the relationship between the average eye acceleration during the initiation of pursuit and target velocity could be learned and gated by context. Vertical position was an effective contextual cue but not target color, implying that useful contextual cues must be similar to those occurring naturally, for example, orbital position with eye muscle weakness.

When primates track a small object moving across their field of view, they use pursuit eye movements to stabilize the image of the object upon the fovea. Because slippage of images across the retina is an important stimulus that drives the pursuit system, which, in turn, generates eye movements to reduce the magnitude of this slip, pursuit is regarded as a visual feedback (“closed-loop”) control system. However, there are inherent delays in the processing of visual information necessary to produce a pursuit command, so that whenever a target of interest changes its speed or direction unpredictably, the pursuit system must operate in an “open-loop” mode for roughly 130 msec, ^{1 2} without the benefit of immediate feedback. Consequently, as is the case for other open-loop ocular motor subsystems such as saccades ^{3 4} and the vestibulo-ocular reflex, ⁵ the open-loop period of pursuit must be accessible to long-term calibration so that the speed of the eyes is brought to that of the target as quickly as possible and kept there without any motor instability or oscillations that would interfere with visual acuity. These considerations predict an adaptive capability for calibrating the open-loop, initial portion of the pursuit tracking response, and indeed such has been demonstrated in both monkeys and humans. ^{6 7 8 9 10 11 12} Typically, the average acceleration of the eye in the first 100 to 130 msec of pursuit tracking was used as a measure of the open-loop response of the pursuit system and was shown to be modifiable in various learning paradigms. Recently, for the vestibular, saccadic, and vergence systems it has been shown that more than one adaptive eye movement response can be learned and gated in or out depending on context. ^{13 14 15 16 17 18 19 20 21} Here, we investigated context-specific adaptation of pursuit, looking at the influence of a nonvisual cue, vertical eye position, and a visual cue, target color. 

Methods

Subjects

Four normal subjects (29–37 years old) participated. They had normal corrected vision, eye motility, and ocular alignment. The head was immobilized with a dental bite bar. Viewing was always binocular and refractive error was corrected with lenses. Subjects were told to maintain their gaze on the visual target. Each subject gave informed consent before the experiments. The research followed the tenets of the Declaration of Helsinki and was approved by the institutional human experimentation committee. 

Experimental Paradigm

The visual target was a white square, subtending a visual angle of 0.2 × 0.2°, and was presented on a video monitor located 60 cm in front of the subject. The room was dark except for the target lights. To induce adaptation of pursuit eye acceleration in the open-loop period, we used a modification of the step-ramp stimulus for pursuit in which there was a double step of velocity (Fig. 1) . In the increasing paradigm, the visual target was first presented for 1.5 seconds at the straight-ahead position (experiment 3) or either 5° up or down (experiments 1 and 2) and then was moved in a step-ramp fashion horizontally, that is, jumped to the right or left and then moved toward the midline at a constant velocity of either 23.3 or 15.6°/sec. In response to the ramp component of the target stimulus, pursuit eye movements began at a latency of about 150 msec. When the target returned to its initial straight-ahead position, approximately 230 msec after the onset of the ramp, its velocity was doubled to 46.7 or 31.1°/sec, respectively. The target then kept moving eccentrically until it reached the lateral extent of the display monitor, 18° from the midline. Similarly, in the decreasing paradigm, the presentation of the stimulus was the same except that the velocity of the target was halved to 11.7 or 7.8°/sec, respectively. 

We tested two types of contextual cues for horizontal pursuit adaptation, the vertical position of the eye in the orbit (experiments 1 and 2) and the color of the target (experiment 3) as shown in Figure 2 . In experiment 1, when the target appeared up 5° in the orbit, it moved horizontally in the decreasing paradigm; when it appeared down 5°, it moved horizontally in the increasing paradigm. In experiment 2, up 5° was associated with the increasing paradigm, and down 5° with the decreasing paradigm. In experiment 3 either the red or green target always appeared at the straight-ahead position, but the red target was associated with the decreasing paradigm, and the green target with the increasing paradigm. In each experiment the direction and velocity of target movement were not predictable. These three experiments were conducted on different days. 

Each experiment consisted of three parts: preadaptation session, training session, and postadaptation session. Control data from the preadaptation state were compared with the immediate postadaptation state using the conventional step-ramp paradigm, in which there was no change in ramp velocity. In total, 24 to 48 trials were presented. The training session, using the double-step of velocity paradigm, lasted for about half an hour. Each training period consisted of a total of 288 trials divided into six subsessions consisting of 48 trials each and separated by a brief pause of about 1 minute. 

Laboratory Apparatus

Horizontal and vertical eye movements of both eye were recorded with an infrared system (Ober 2; Iota AB, Sundsvall, Sweden). A system calibration procedure was performed about every 5 minutes during the experiment, in which the subject was required to fix on targets separated by 5° or 10°. The output signals were sampled by a digital computer at 600 Hz and then stored to a hard disc for later off-line analysis. System noise limited resolution to approximately 0.1°. 

Data Analysis

Pursuit traces were analyzed using a computer-assisted procedure in which individual trials were displayed on a video monitor. Conjugate traces were calculated using the average of the calibrated right eye and left eye horizontal signals. Trials were rejected if eye traces were disturbed by blinks or if saccades occurred within 100 msec after pursuit onset. Position traces were filtered and differentiated with a single-pole analog filter with a cutoff frequency of 15 Hz. The onset of pursuit was taken as the point when smooth eye velocity exceeded 3°/sec and continued to accelerate in the direction of the target ramp. This point was verified by the experimenter for each trial. Thus, though eye velocity during fixation occasionally exceeded the 3°/sec threshold value, these spontaneous fluctuations above the threshold value could be separated easily from the onset of pursuit tracking. Average eye acceleration during pursuit initiation was then taken as the value of smooth eye velocity at 100 msec after the pursuit onset. 

Results

Representative examples of the results from experiments 1 and 2, in which vertical eye position was the contextual cue, are shown for subject 3 in Figures 3 4 5 . Figure 3 compares tracking during the initial several hundred milliseconds of pursuit, between the beginning and the end of the training period. First, note the pattern of tracking with the decreasing paradigm when the eyes were up 5° in the orbit. When training began (Fig. 3 , lefthand panels, Early in Training), the initial eye velocity at the end of the open-loop period was commensurate with the initial target velocity so that the eye then had to slow down to match the movement of the target at its final (halved) velocity. At the end of the training period, however, the acceleration and velocity during the initial few hundred milliseconds of pursuit were less (Fig. 3 , lefthand panels, Late in Training), indicating that the initial command for pursuit had been modified toward the lower, second target speed. An analogous pattern of change was observed with the increasing paradigm when the eyes were down 5° in the orbit, only in this case (Fig. 3 , righthand panels) the acceleration and velocity during the initial few hundred milliseconds of pursuit were greater at the end of the training period. In sum, the results shown in Figure 3 indicate that different patterns of adaptive changes in horizontal pursuit initiation can be induced simultaneously and that the adaptive changes can even be in the opposite sense, one for an increase and the other for a decrease. Which adaptive change was invoked depended on the vertical position of the eye in the orbit. 

Figure 4 shows the changes in average acceleration in the first 100 msec of tracking during the same adaptation session for which data are shown in Figure 3 . As expected, for tracking in up gaze the average acceleration gradually decreased, whereas for tracking in down gaze it gradually increased. To quantify this change, we fit a regression line to the data and used the fitted values at the beginning and the end of training to compute a percentage change, which is also shown in Figure 4 . For this particular record, the change was −11.4% (correlation coefficient = 0.24, n = 32, P = 0.19) for up gaze and +47.1% (0.57, n = 36, P < 0.0001) for down gaze. 

The effects of training could also be seen when comparing pretraining and posttraining values of average acceleration in the open-loop period. Figure 5 shows examples from subject 3 while looking up (right panel) and down (left panel), for two different initial target speeds. Again, there was a change in average acceleration in the open-loop period that was specific to the vertical position of the eye in which the training took place. 

Figure 6 summarizes the adaptive change in average acceleration in the open-loop period for all subjects and experiments. These numbers were obtained from the slope of the regression line during the entire adaptation session as was calculated for the data presented in Figure 4 . Although the general pattern of response was qualitatively similar among all subjects—the adaptive response in the initial acceleration of horizontal tracking depended on vertical eye position, there were quantitative differences. In both experiments 1 and 2, different amounts of adaptation occurred, depending on whether the subject was looking up or down. The mean difference between the adaptation values for looking up and down, for the corresponding trial type for each subject is given in the right upper corner of each panel. For experiment 1, a large increase in acceleration was seen during down gaze in subjects 1 and 3, and a large decrease in acceleration was seen during up gaze in subject 2. For subject 4, the overall adaptive changes were smaller, and there was little difference in the amount between up and down gaze. There was a statistically significant difference using the paired t-test between corresponding trial types for all cases (P = 0.0004, n = 16). The percentage differences for experiment 1 ranged between 15.9% and 37.7%. On average, the difference was 26.6% for all cases. In experiment 2, acceleration was increased for looking upward and decreased for looking downward in subjects 2 and 3. In subjects 1 and 4, the decrease in acceleration was greater for looking down. There was a statistically significant difference using the paired t-test between corresponding trial types for all cases (P < 0.001, n = 16). The percentage differences for experiment 2 ranged between 11.3% and 55.1%. On average, the difference was 33.4% for all cases. 

In experiment 3, a similar analysis is presented; in this case adaptation is compared with the contextual cue being the color of the target. There was little difference in adaptation between the red target and green targets in this experiment (on average 3.0%), and the differences were not significant (P = 0.47, n = 16). This result indicates that unlike vertical orbital position, target color alone is not an effective cue for context-specific pursuit adaptation. 

Discussion

The main finding of this study is that for smooth pursuit, multiple relationships between the value of eye acceleration in the initial open-loop period of tracking and the velocity of the target can be learned simultaneously and gated in and out according to a contextual cue. Our specific paradigm linked horizontal pursuit to vertical eye position and called for an increase in eye acceleration in one vertical eye position and a decrease in the other. Although the responses were not always tailored exactly to the combinations of stimuli presented, in all cases the general pattern of the response, that is, a relatively greater increase in pursuit acceleration when an increase in acceleration was called for in a particular vertical eye position or a relatively greater decrease in pursuit acceleration when a decrease in acceleration was called for in a particular vertical eye position, conformed to the requirements of the stimulus. Considering how brief the training period was and how close the contextual cues for vertical eye position were (they were separated by only 10°, so that learning in one position could have interfered with learning in the other), the context-specific effect was robust and unequivocal. The finding of context-driven learning for smooth pursuit is not surprising because such learning has already been demonstrated in other open-loop eye movement subsystems that have been shown to be under adaptive control, such as the vestibulo-ocular reflex ^{15 19 23} and saccades. ^{16 17 18 20 21 22 24} Likewise vertical eye position has also been shown to be an effective cue for context-specific adaptation of the horizontal vestibulo-ocular reflex ¹⁴ and for horizontal saccades. ²¹ We also found, however, that unlike eye position, target color was not an effective cue for context specific adaptation. A similar dichotomy has been reported previously for context-specific learning in saccades. ¹⁷  

At first glance, one would think that the eye position cue is“ closer” than the color cue and perhaps more necessary to the central mechanisms that generate pursuit eye movements. Premotor commands for all types of eye movements must take into account eye position in the orbit because of the nonlinear relationship between eye muscle strength and orbital position. ²⁵ Adaptation in the presence of a weak muscle, for example, must be orbital-position specific. ⁶ One can ask, however, why would vertical eye position be important for horizontal pursuit adaptation because the change in the cue itself—the position of the eye in the orbit—is orthogonal to the direction of eye motion that must be recalibrated. One possible explanation is that in natural circumstances targets rarely move across the visual field in a purely horizontal or vertical direction (i.e., relative to the orbit) but rather have an oblique component. With a change in vertical eye position, the innervation to produce the horizontal component of pursuit might have to be modified, because the pulling directions of the horizontal muscles change with the vertical position of the eye in the orbit. Thus, vertical eye position could be an important cue for generating the correct horizontal component of pursuit, as the eye traverses an oblique trajectory. Another possible factor would be if smooth pursuit learning were applied in polar rather than Cartesian coordinates and so generalize to target (and eye) trajectories with vectors that were slightly oblique relative to the training direction. ^{7 26} This spread of adaptation to target trajectories that differ in their direction from the training axis has also been shown to be the case for saccade adaptation. ^{16 22} In this case, too, vertical eye position would become an important cue for the horizontal component of eye motion and would be expected to influence adaptive responses even to purely horizontal target motion. With respect to both these ideas, it would have been of interest to test oblique tracking before and after training even though the adaptive stimuli were for pure horizontally directed target and eye motion. 

Finally, it has been reported that the initial “open-loop component” of pursuit is kept accurate regardless of the starting position of the eye in the orbit, ²⁷ although the same is not true for the closed-loop sustained portion of tracking in which there are changes in accuracy that depend on orbital position. ²⁸ Presumably, accuracy in the initial phase of pursuit tracking is the most critical for object identification, so that the open-loop period is always being calibrated as precisely as possible. 

Why was target color not an effective cue? Other studies of adaptation have shown that cues apart from eye position can be used to guide ocular motor learning, such as frames on/off for eye glasses ¹³ and conditional adaptation (with the experimental setting itself becoming the contextual cue). ^{14 17} Perhaps color is too remote from the functional motor mechanisms underlying accurate pursuit and might not work without a much longer training period. In a sense, though, the inability of target color to serve as a contextual cue for pursuit learning in our short-term learning paradigm indicates that mental effort alone, to increase or decrease gain, is not the cause of separate adaptive responses in experiments 1 and 2. Taken together with the nonpredictability of the direction and the velocity of the target, the changes in the very early part of the pursuit response cannot be explained by a simple higher-level cognitive strategy such as“ anticipation,” “intention,” or “prediction.” 

What might be the neural mechanisms underlying context-specific adaptation for smooth pursuit? For the type of cue used here, vertical eye position, a likely locus is the cerebellum. The cerebellar cortex, including the dorsal vermis ¹² and the flocculus or ventral paraflocculus, ²⁹ has been implicated in pursuit adaptation, and with its rich supply of visual and eye position information (both proprioceptive afference and efference copy), ^{30 31} the cerebellar cortex is ideally poised to recognize contexts that could be used to gate the corrective adaptive response. Such a role for the cerebellum in pursuit adaptation does not preclude a role for other structures, either in the type of learning demonstrated in our study or in higher-level more cognitively driven adaptation. Deubel ¹⁷ has shown, for example, that learning in the saccadic system can be specific to the type of saccade, be it to a novel stimulus, to a remembered target, or self-generated. It remains to be shown how the cerebellum and other cerebral structures, including the thalamus, basal ganglia, and cerebral cortex, contribute to ocular motor learning in general and context-specific learning in particular. 

 Supported by Grant-in-Aid 09771413 for Encouragement of Young Scientists (MT), Scientific Research Grant 11671727 (the Japan Ministry of Education, Science, Sports and Culture), National Institutes of Health Grant R01-EY01849, and NASA through Cooperative Agreement NCC 9-58 with the National Biomedical Research Institute (DSZ).

 Submitted for publication January 26, 2000; revised June 5, 2000; accepted July 5, 2000.

 Commercial relationships policy: N.

 

Figure 1.

 

Experimental paradigms used to induce horizontal pursuit adaptation. First, the target jumped away (step) from the center fixation point and then began moving at a constant velocity (ramp, 23.3 or 15.6°/sec) back toward the center, but just after it crossed the center (230 msec after the onset), target velocity doubled (A, increasing paradigm) or halved (B, decreasing paradigm).

Figure 2.

 

Configuration of the three experiments. In experiment 1 (top), when the target appeared up 5°, training was for decreasing adaptation; when it appeared down 5°, training was for increasing adaptation. In experiment 2 (middle), the training was opposite relative to vertical eye position, with the target up 5°, increasing, and down 5°, decreasing. In experiment 3 (bottom), the target appeared in the straight-ahead position, but its color was red or green. When red, training was for decreasing adaptation, and when green, for increasing adaptation.

Figure 3.

 

Representative pursuit traces from subject 3 during the training session of experiment 1 (the initial velocity of the target was leftward 23.3°/sec). Downward denotes leftward movement. The first five and the last five trials during the training session are shown. Left: decreasing paradigm with the eyes up 5°; right: increasing paradigm with the eyes down 5°. Top: position traces. Thick solid lines, target movement; dashed lines, what target position would have been if target velocity had not changed. In the decreasing paradigm, initially during training the eye had to slow down considerably. At the end of training period, however, the acceleration in the open-loop period decreased, and eye velocity overshot target velocity by much less. In the increasing paradigm, initially during training the eye considerably lagged the target. At the end of the training, however, the acceleration in the open-loop period increased, and the eye did not lag the target by as much. Bottom: velocity traces. The change in the acceleration in the open-loop period is seen by comparing the slopes of the velocity traces.

Figure 4.

 

The time course of change in the average acceleration in the open-loop period from the same data shown in Figure 3 (subject 3, experiment 1, left 23.3°/sec). Left: decreasing paradigm with the eyes up 5°; right: increasing paradigm with the eyes down 5°. Abscissa: trial number; ordinate: average acceleration during the initial 100 msec of pursuit tracking. Solid line, the regression line. Left: average acceleration gradually decreased; the change during 100 trials based on the first and last values from the regression line was −14.4%; right: average acceleration gradually increased, with the change being 47.1%.

Figure 5.

 

Comparison of acceleration of initial 100 msec between preadaptation session (▴) and postadaptation (○) session from subject 3, experiment 2. Left: up gaze, increasing paradigm; right: down gaze, decreasing paradigm. On the y-axis is the average acceleration during the initial 100 msec of tracking. R1, right tracking, lower target speed; R2, right tracking, higher target speed; L1, left tracking, lower target speed; L2, left tracking, higher target speed.

Figure 6.

 

Change in acceleration of initial 100 msec during adaptation session, which is percentage change in accelerations as calculated from the slope of linear regressions performed on acceleration as a function of trial number. Left, middle, and right panels: experiments 1, 2, and 3, respectively; Nos. 1, 2, 3, and 4 denote subjects. Left and middle panels: left bars are for up 5° and right bars are for down 5° in order of left 23.3°/sec (L2), left 15.6°/sec (L1), right 15.6°/sec (R1), and right 23.3°/sec (R2) as initial target velocity. Right panels: left bars are for red target, and right bars are for green target in the same order. Upward arrow, increasing paradigm; downward arrow, decreasing paradigm. The number in the right upper corner of each panel indicates mean difference of percentage change in accelerations for the corresponding four trial types.