Cross-Axis Adaptation of Pursuit Initiation in Humans

Abstract

purpose. The initial acceleration of pursuit in the open-loop period is under adaptive control and undergoes motor learning. The current study was undertaken to examine the hypothesis that the direction of pursuit initiation can also be adaptively modified.

methods. Four neurologically and ophthalmologically normal subjects participated in the experiment. A modified step-ramp paradigm was used to induce cross-axis adaptation, in which a ramp target changed its direction orthogonally just after the target crossed the center. Four direction changes were tested in separate experiments: left to up, left to down, down to left, and up to left. During a 30-minute adaptation session, the target moved in one of two randomly chosen directions (right to left or up to down) at one of two randomly chosen speeds (15.6 or 22.3 deg/sec), but the target changed orthogonally in only one direction. A linear regression fit to the initial 100-msec segment of the pursuit trace was used to determine the direction of pursuit initiation.

results. In all cases, an adaptive change in pursuit initiation was gradually induced in the direction called for by the training paradigm. Adaptation was usually completed (90° shift) within the 30-minute training session but declined quickly to an approximate 30°-shift after training. The latency and vectorial amplitude of the initial acceleration remained unchanged. The adaptation was specific for the direction but not the velocity of the target.

conclusions. This study showed that the direction of pursuit initiation is under adaptive control, as has been shown for saccadic eye movements and the vestibulo-ocular reflex.

Smooth pursuit is a tracking eye movement performed to retain the image of a moving target on the fovea. The image of a target that passes across the fovea (retinal slip), triggers the pursuit system. The pursuit system has been classically argued to be under closed-loop control—that is, the accuracy of pursuit eye movements is maintained using a feedback error signal based on the difference between target and eye motion. The initiation of pursuit eye movement, however, operates in an open-loop fashion for roughly the first 130 msec after onset of movement, before visual feedback can influence the motor response. ¹ It is known that the initial acceleration of pursuit in the open-loop period is adaptively controlled with motor error learning. ^{2 3 4 5 6} In the natural environment, a target often suddenly changes its direction of movement, as is often observed when a prey is being chased by a predator. This suggests the need for a mechanism to keep the direction of pursuit initiation accurate. Both the magnitude of acceleration and the direction of pursuit initiation may be preprogrammed in the central nervous system and adaptively controlled to be appropriate for the target speed and direction. Directional adaptation has already been demonstrated for saccadic eye movements ⁷ and the vestibulo-ocular reflex. ^{8 9}  

In this experiment, we investigated cross-axis adaptation in the initiation of pursuit eye movements in which an error signal in direction was introduced artificially. 

Methods

Subjects

Four human subjects (age range, 29–37 years) participated in the experiments after providing informed consent and were neurologically and ophthalmologically normal except for refractive errors. Their heads were immobilized with a dental bite bar, and they were asked to follow a target on the video monitor. The procedure of this study conformed with the Declaration of Helsinki (1964) and was approved by the Ethics Committee of Niigata University School of Medicine. 

Experimental Paradigm

The experiments were performed in a dark room. The visual target was a white square measuring 2 × 2 mm (subtending an angle of 0.20 × 0.20 minutes) displayed on a video monitor placed 60 cm away from the subject. A modified step-ramp paradigm was used to induce cross-axis adaptation: A target appeared for 1.5 seconds at the center of the monitor, jumped 3.6° or 5.1° away from center, and then moved toward the center at a constant velocity of 15.6 (slow target movement) or 22.3 deg/sec (fast target movement). When the target crossed the center 230 msec after the onset of the ramp movement, it changed direction, horizontal to vertical or vertical to horizontal, with the same speed. Thus, just after pursuit eye movements started from the center in response to the first target ramp movement, the target was moving orthogonal to the pursuit response. 

We tested four kinds of cross-axis adaptation (Fig. 1) . In each experiment, the target first moved in one direction, right to left or up to down, randomly, but the target changed direction orthogonally in only one direction: from down to left, from left to up, from up to left, and from left to down (adaptation side). In the opposite direction (reference side), the direction of the target did not change throughout the trial. The ramp velocity was also randomized so that the subject could not predict the direction or velocity of the target. The four experiments were conducted on different days. Each experimental run consisted of preadaptation, adaptation, and postadaptation sessions. In the preadaptation session, control data were collected before cross-axis adaptation. Thirty-two trials of the step-ramp paradigm without directional change were tested. In the adaptation session, the modified step-ramp paradigm was tested for approximately 30 minutes. One training subsession consisted of 48 trials followed by a brief pause of approximately 1 minute, and six subsessions were conducted with a total of 144 trials (72 slow and 72 fast targets) for each adaptation and reference side. The postadaptation session consisted of the same 32 trials used in the preadaptation session—that is, the direction of the target did not change throughout the trial. 

Eye Movement Recording and Data Analysis

Horizontal and vertical movements of both eyes were recorded simultaneously with an infrared system (Ober 2; Iota, Sundsvall, Sweden) and stored on a hard disk at 600 Hz for off-line analysis. For calibration, subjects were required to fixate sequentially 17 points around the center arranged in a grid pattern separated by 2.5° horizontally and vertically so that a calibration matrix could be obtained. This brief calibration session was performed every 5 minutes during the experiment. After two-dimensional interpolation of the matrix by the inverse distance method, ¹⁰ calibration was applied to raw eye movement data to correct for the two-dimensional distortion of the recording system. Conjugate eye position traces (average of right and left horizontal traces) were filtered and differentiated with a single-pole analog filter with a cutoff frequency of 15 Hz. We defined pursuit onset as the time when the vectorial eye velocity exceeded 3 deg/sec. Each conjugate trace of pursuit eye movement was plotted in two dimensions, and linear regression was used to fit the eye movement traces for the first 100-msec segment after the pursuit onset using the method of least squares. The angle of this regression line was used to determine the initiation angle of pursuit eye movement. The latency of pursuit and average initial acceleration during the first 100-msec segment after the onset were also calculated. Trials were rejected if eye traces were disturbed by blinks or if saccades occurred within 100 msec after pursuit onset (less than 5%). All the data were processed using data analysis software (Matlab; The MathWorks, Natick, MA). 

Results

Figure 2 shows representative pursuit traces for the first 500-msec segment after the onset of pursuit before, early in, late in, and after the training session. In this experiment, the target moved first left and then down on the adaptation side, but moved right throughout on the reference side. Before training of cross-axis adaptation, eye movements were induced exactly to the left (184.2° ± 6.5°, eight traces) on the adaptation side. At the beginning of training, the eye first moved left (193.6° ± 7.2°, first 12 traces) to follow the initial leftward target movement but had to catch up with the target moving down with oblique corrective saccades approximately 200 to 300 msec after the onset of pursuit. At the end of training, the eye started moving down from the beginning (286.6° ± 6.7°, last 12 traces) although the target first moved left. Approximately 30 minutes after training, the direction of initial pursuit had shifted down (237.4°± 30.7°, eight traces) in response to the leftward step-ramp target movement without direction change. 

On the reference side, there was no effect of the training. The Mann–Whitney rank sum test revealed a significant difference in the angle of pursuit initiation before and after training on the adaptation side (P = 0.008), but not on the reference side (P = 0.64). Figure 3 shows horizontal and vertical components of pursuit traces for the adaptation side from the data shown in Figure 2 . It is evident that even in the late phase of the adaptation session and postadaptation session, directionally adapted pursuit started smoothly with a latency of approximately 150 msec. In this case, pursuit latencies were 152.6 ± 5.7, 155.4 ± 7.8, 154.6 ± 12.6, 154.1 ± 7.5 msec before adaptation, in the early and late phases of adaptation, and after adaptation, respectively. Using Kruskal–Wallis one-way analysis of variance on ranks, no significant difference was found among them. Similarly, in all cases we tested the differences in latency among the four conditions (before, in the early and late phases, and after adaptation) but no significant difference was found, the latencies always being approximately 150 msec. 

Figure 4 shows the time course of changes in angle during the training session in the same subject as is shown in Figure 2 . The angles of pursuit initiation were plotted for each of 72 trials for slow and fast targets during the adaptation session. On the adaptation side, the initial angle direction changed gradually from left (180°) to down (270°); however, on the reference side the initial angle direction did not change for either target speed. An exponential curve was fitted to all data points as a function of trial number in each panel, using the least-squares method, and then to evaluate changes in the angle of initial pursuit direction during the training session, the difference in angle between trials 1 and 72 was calculated using the exponential curve. In this experiment, changes in the angle on the adaptation side were 87.9° and 95.4° for slow and fast target movements, respectively. In contrast, on the reference side, the respective changes in angle were only 2.6° and 0.7°. On the adaptation side, the extent and time course of directional change were almost the same for both target speeds, as seen in panels L1 and L2. The time constants of the exponential curve were 30.2 trials for L1 and 32.1 trials for L2 in these cases. In general, the time constants of directional adaptation were very similar among experiments and subjects. They were 32.1 ± 3.0 (average of slow and fast target for four directions, i.e., eight cases), 32.4 ± 2.9, 30.9 ± 1.9 and 31.4 ± 2.2 trials for subjects 1 through 4, respectively. 

Similar results were obtained in all cases (Fig. 5) . Changes in angle of almost 90° were obtained (85.0 ± 14.9° for 32 cases) during training sessions on the adaptation side, but changes were very small on the reference side (7.5 ± 5.2°). The paired t-test was performed to compare the change between the slow and fast target movements. No significant difference was found on either the adaptation side (P = 0.97, n = 16) or the reference side (P = 0.54, n = 16). Figure 6 shows the changes in angle of pursuit initiation in the posttraining sessions for all cases. The changes in angle on the adaptation side were, on average, 34.4° ± 19.0° for the slow target movement and 27.3° ± 20.0° for the fast target movement. Large changes in angle of approximately 90° disappeared quickly. However, changes in angle of approximately 30° persisted or declined slowly during the postadaptation session. On the reference side, the changes after training were only 1.86° ± 5.87° for the slow and 0.60° ± 4.87° for the fast target movement. The Mann–Whitney rank sum test revealed a significant difference in initiation angle between preadaptation and postadaptation sessions in 14 of 16 cases; there was no significant difference on the reference side. 

In summary, it was possible to induce directional adaptation of pursuit initiation without changes in the latency or acceleration by providing artificial direction error in the open-loop period of the pursuit eye movement response. Adaptation was specific for the target’s direction but not for its velocity. 

Discussion

Recently, considerable evidence has been accumulated for an adaptive capability for pursuit acceleration in its initial open-loop segment. Optican et al. ² showed that adaptation of acceleration in the open-loop period of pursuit eye movement occurred in response to anisotropic eye movement deficits caused by ocular muscle weakness in patients with ocular motor nerve palsies. Kahlon and Lisberger ³ observed that the acceleration of initial pursuit responses in monkeys changed adaptively on repeated presentation of targets that moved at one speed for 100 msec and then changed to a second higher or lower speed. Similar adaptive changes were also reported in humans. ^{4 5 6} Reports of directional adaptation in saccadic eye movement and vestibulo-ocular reflex (VOR) demonstrate that adaptation applies to vector as well as scalar (gain) values. Deubel ⁷ demonstrated that saccadic direction in primates was equally as adaptable as saccadic gain by using a target that was displayed orthogonally to the direction of the initial step. Adaptive changes of direction in VOR after training were found in animals by using an optokinetic drum ⁸ or a target spot ⁹ moved horizontally in synchrony with vertical head oscillations. 

The paradigm used in this experiment seems to be appropriate for providing artificial directional motor error during the open-loop period, because the latency of pursuit eye movement remained unchanged at approximately 150 msec after adaptation for all subjects and the target’s direction changed within the open-loop period, at approximately 200 to 300 msec after onset of pursuit. We consider our results to represent adaptation by motor learning for the following reasons: First, the change in the direction of eye movement was not an artifact of the recording system. Although the raw eye movement data recorded by our monitor system were usually affected by two-dimensional distortion, a two-dimensional calibration was performed every 5 minutes during the experiment to monitor the exact direction. Second, it was not a predictive response, because the direction and speed of the target were randomized to prevent prediction by the subjects. Furthermore, whenever the subjects made a predictive pursuit response, an increase of eye acceleration and decrease of the latency in pursuit initiation could be expected. ¹¹ No such changes in acceleration and latency were found. Rather, constant latencies throughout the experiment strongly suggested that the response was driven reflexively by the first retinal slip signal. Third, it is not a voluntary directional change arising from a cognitive strategy. The time course of the directional change during training was not abrupt, as would be expected from a voluntary change, but was exponential, as is common for motor error learning. 

In the present experiment, directional changes did not transfer to the reference side. No differences in adaptation between the two levels of target velocity were observed. Similar results for adaptation of initial acceleration were described by Kahlon and Lisberger ³ and Scheuerer et al. ⁶ What differs from gain adaptations such as saccade size and pursuit initial acceleration is the fast induction, the time constant being approximately 30 trials and orthogonal adaptation being completed within 72 trials (Fig. 3) . Similar rapid cross-axis adaptation is known for VOR. An orthogonal eye movement response to body rotation in monkeys appeared after approximately 30 minutes of training. ⁹ In contrast, gain adaptation requires more training. At least 60 trials were necessary for acquisition of visible saccadic and pursuit adaptation in humans ^{6 7} and approximately 200 trials in primates. ^{3 7} In the postadaptation session, which began 1 to 2 minutes after the end of training, the large directional shift of 90° disappeared rapidly, but a 30° shift persisted during the session. There are two possible explanations for this result. Directional adaptation can consist of an immediate component and a slower, long-lasting component. Alternatively, some deadaptation could be caused by normal VOR during the interval between sessions, because the subject’s head was not restricted, and it was possible to glance at the dimly illuminated room during the interval. Also, because the target presented in the postadaptation did not change its direction, the trials would have served as deadapting stimuli. 

The central nervous system structures related to pursuit eye movements have been investigated in monkeys. At present, the middle temporal (MT) area and the medial superior temporal (MST) area in the cerebral cortex, the ventral paraflocculus, the posterior vermis and underlying fastigial nucleus in the cerebellum, the dorsolateral pontine nucleus, and the nucleus reticularis tegmenti pontis in the brain stem are believed to be the main structures controlling pursuit. ^{12 13} However, the mechanism for pursuit adaptation is still unclear. The neuronal activity related to adaptation of acceleration in pursuit initiation has been found in the dorsoventral paraflocculus. ¹⁴ In addition, lesions of the VII lobule of the cerebellar vermis cause partially impaired adaptation of initial pursuit acceleration. ¹⁵ On the basis of these reports, the current hypothesis is that gain adaptation of pursuit initiation is controlled by both the posterior vermis and the dorsoventral paraflocculus. There is also much debate about the mechanism involved in directional adaptation of eye movements. In cats, adaptive changes in VOR direction were disturbed after removal of the paraflocculus and lobule VII and part of VI in the vermis. ⁸ However, the central nervous structures involved in directional adaptation in saccade and pursuit eye movements are not known. Directional adaptation in pursuit may involve a different mechanism from that in saccades. In oblique saccades, various types of saccadic pulse generators are associated with different directions. ¹⁶ The outputs of vectorial pulse generators are transmitted to horizontal and vertical output channels by different synaptic weightings onto motor neurons. ¹⁷ In contrast, in directional tuning of pursuit eye movements, the best directions elicited by microstimulation of PCs and mossy fibers are split into pure horizontal and vertical components. Directional adaptation in initial pursuit may not be a simple adaptive change in angle information but may be processed analogously to adaptation in acceleration after being segregated into horizontal and vertical channels in the same way as ordinary pursuit eye movement. ¹⁸  

Clinical evidence shows that changes in the ocular motor plant, such as orbital mass lesions or extraocular muscle weakness, can lead to adaptive increases in motor signals. ² In other words, when the direction of eye movement is distorted by eye disease or some other disorder, the central nervous system must correct and maintain eye movement. Even in the normal eye, differences in the viscous forces on the eye arising from differences in the stiffness of nasal and temporal tissue are known to occur. ¹⁹ In the eccentric gaze, elastic forces move the eye toward the center of the orbit. ²⁰ Thus, the direction of pursuit initiation could be distorted by orbital elastic forces, and directional adaptation mechanisms would be required to maintain accurate pursuit initiation, regardless of the starting eye position in the orbit. Consequently, directional adaptation in the central nervous system maintains the accuracy of the initial pursuit direction during eye movement disorders as well as in the normal condition. 

 Supported by Grants-in-Aid for Encouragement of Young Scientists 09771413 (MT) and for Scientific Research 11671727 from The Japan Ministry of Education, Science, Sports and Culture.

 Submitted for publication May 18, 2000; revised September 28, 2000; accepted October 16, 2000.

 Commercial relationships policy: N.

 

Figure 1.

 

Two-dimensional schematic illustration of target movement during an adaptation session. In each experiment, one of four kinds of training paradigms was used. The target first appeared in the center and then jumped away from the center (step) and began moving at constant velocity (15.6 or 22.3 deg/sec, ramp). On the adaptation side, the target always changed direction orthogonally when it crossed the center 230 msec after onset of the ramp (solid lines): (A) down to left, (B) left to up, (C) up to left, and (D) left to down. On the reference side, the target did not change direction (dotted lines). These experiments were performed on different days for each subject.

Figure 2.

 

Two-dimensional pursuit traces before, early in, late in, and after the training sessions (subject 3; target direction changed from left to down). Top: Adaptation side; bottom: reference side. The traces representing the early and late phases of training, the first and last 12 traces during training sessions are shown. The ordinates and the abscissas of the graphs represent vertical and horizontal angles (in degrees) for pursuit traces for the 500-msec segment after the target onset, respectively. On the adaptation side, left horizontal eye movements were induced before training. Early in training, the eye began moving left in response to the initial leftward target movement. However, because the target changed its direction orthogonally down, the eye had to catch up with the target using saccades and thereafter kept moving down. Later in training, pursuit began moving not left but down from the beginning, even though the target first moved left. After subjects trained for 30 minutes, pursuit initiation shifted down in response to the leftward target movement. Eye movements were not affected on the reference side.

Figure 3.

 

Horizontal (top) and vertical (bottom) components of pursuit traces on the adaptation side presented in Figure 2 . Similarly, traces of the 500-msec segment after the target onset before, early in, late in, and after the training sessions from the left. For the early and late phases of training, the first and last 12 traces during the training sessions are shown. The ordinates and abscissas represent horizontal or vertical angles (in degrees) and time after the target onset, respectively. Note that the latencies of pursuit initiation were almost constant throughout the experiment.

Figure 4.

 

The time course of directional change of pursuit initiation during a training session (subject 3; change in target direction from left to down) for the (top) adaptation side and the (bottom) reference side (change to the right). The ordinates and the abscissas of the graphs represent pursuit initiation angle (in degrees) and number of trials in the experiment, respectively. The target first moved left at 15.6 deg/sec or 22.3 deg/sec (L1, L2, respectively) or right at 15.6 deg/sec or 22.3 deg/sec (R1, R2, respectively). An exponential curve was fitted as a function of trial number for each panel. Changes in pursuit initiation angle were calculated as the difference between the beginning and end of training trials using the exponential curve.

Figure 5.

 

Changes in pursuit initiation angle (in degrees) calculated from the exponential curve are shown for all cases. Solid bars: adaptation side; clear bars: reference side. L, R, D, U: Direction in which target first moved (left, right, down, up, respectively); 1, 2: mean target velocities of 15.6 and 22.3 deg/sec, respectively. In all subjects, changes in angle of approximately 90° were found during the training session on the adaptation side. In contrast, no changes in angle were observed on the reference side.

Figure 6.

 

Angle shift of pursuit initiation from the horizontal meridian in the postadaptation session for all cases. Results of eight trials for each trial type on the adaptation side are shown in the order of trial progression from the left. S, Trials for the slow target; F, trials for the fast target. Significant differences between initial angles between the pre- and postadaptation sessions using the Mann–Whitney rank sum test (*P < 0.05; **P < 0.01; N.S., not significant).

The authors thank David S. Zee and Takehiko Bando for critical review of the manuscript.