Rapid Motor Learning in the Translational Vestibulo-Ocular Reflex

Open loop interval of the TVOR

Figure 1 shows single trial responses to brief rightward translations (Fig. 1A) when the TVOR was in different states. In an unadapted state (Fig. 1B), the compensatory eye movement was not adequate, and a catch up saccade (arrow) was required to bring the eyes onto the world-fixed target after the cessation of translation. In an adapted high gain state (Fig. 1C), however, the compensatory eye movement was nearly perfect, and there was no corrective saccade. When the monkey viewed a head-fixed target, its compensatory eye movements took the eyes away from the target, and a corrective saccade was required to re-fixate the target (Fig. 1D). Further training with repeated TVORC trials nearly suppressed the compensatory eye movements (low gain state) (Fig. 1E).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Single trial responses of the TVOR in different adaptation states. A, The three traces are head position, velocity, and acceleration, respectively. B, Unadapted state. This trial, with a world-fixed target, was obtained before TVOR training. Black and gray traces are eye position and velocity of the left and right eyes, respectively. C, High-gain state. This trial, with a world-fixed target, was obtained after several days of TVOR training. D, TVORC Training. To initiate low gain adaptation, the target was head-fixed. The trial shown was the fifth trial of TVORC training. E, Low-gain state. This trial occurred on the ninth day of low gain training. For eye position traces, the calibration bar (unlabelled) is 30° in B-D and 15° in E. For eye velocity traces, the calibration bar (unlabelled) is 150°/sec in B, D, 90°/sec in C, and 30°/sec in E.

Because a visual target was present during head translations, the observed responses were not entirely driven by vestibular stimuli (Lisberger and Westbrook, 1985; Miles et al., 1986). To determine the effect and latency of any possible visual feedback in this paradigm, monkeys were subjected to TVOR trials with continuous target visibility randomly mixed with TVOR trials during which the target was turned off before head movement and not turned back on until 200 msec after cessation of head movement. The “open loop” interval of the TVOR was estimated by measuring the latency of the divergence between the eye velocity traces obtained in the two conditions, which was 97.9, 125.6, and 113.5 msec for monkeys 3, 4, and 5, respectively. Thus, any effect of visual feedback was excluded in our analysis of TVOR motor learning by its focus on the initial 80 msec of compensatory eye movement (“initial eye speed”).

TVOR learning: time course, retention, latency, and dynamics

Figure 2 illustrates changes in initial compensatory eye speed induced by gain-decrease or gain-increase training over a period of several days. Initially, monkeys were given TVOR trials to maximize their compensatory response (high-gain state). Control data, obtained in this state, are plotted as open circles in the top left panel of Figure 2 (see arrow). After the control trials, monkeys were given repeated TVORC trials to initiate a gain-decrease process (Fig. 2, top panel, day 1, filled circles for TVORC trials). In the first 20 TVORC trials, there was a rapid decrease in initial eye speed evoked by head translation (from ∼60 to 38.2°/sec). This response decrement at the beginning of the adaptation process reflected a parametric adjustment that may have resulted from the animal's recognition of the altered context. An additional adaptive learning process was activated as the monkey completed more TVORC trials. Initial eye speed continued to decrease, following an exponential time course, with a time constant of 152 trials. The learning ratio was 0.46, indicating a 54% reduction in initial eye speed as a result of TVORC training on day 1. The learning process was also very rapid, with a time constant that was equivalent to <1.5 min of accumulated vestibular stimulation. Between test days, monkeys were allowed to move freely in their home cages in normal lighting conditions. When tested on day 2, the decrement in initial eye velocity achieved at the end of day 1 was 90% retained. After day 3 of training, the monkey was left in its home cage for 4 d. When tested on day 7, the reduced gain had been retained without the need of any further reinforcement (day 3: 13.0 ± 3°/sec vs day 7: 13.4 ± 4 °/sec). When two more days (days 8 and 9, data not shown) of TVORC training confirmed that the TVOR reached the low gain state, the monkey was given TVOR trials to initiate a gain-increase process. In day 1 of the TVOR training, an initial set of control TVORC trials (Fig. 2, bottom panel, filled circles, see arrow) were obtained first followed by a block of TVOR trials (open symbols). Within the first few TVOR trials, there was a rapid increase of initial eye velocity (from 5.8 to 16.7°/sec) that we believe was parametric. Further increases in eye velocity followed an exponential time course with a time constant of 345 trials, which was much longer than that observed during TVORC motor learning. On day 2 of the TVOR trials, the enhanced eye speed attained the previous day was 93% retained. Furthermore, the high-gain state achieved after day 2 was retained for 3 d without training (day 2: 41.4 ± 3°/sec vs day 5: 43.1 ± 3°/sec). Table 1 lists learning ratios (VEL_after/VEL_before) for gain decrease and gain increase training for six monkeys, demonstrating the effectiveness of the behavioral paradigm in inducing adaptation of the TVOR. Table 2 lists the retention rates for three monkeys 24 hr after adaptation, demonstrating that the rapidly acquired changes in the TVOR were consolidated over time.

The rate of acquisition of the low gain state during TVORC training and the degree of adaptation finally achieved were relatively independent of any previous experience the monkey may have had with the training paradigm. Figure 3 shows the time course of changes in initial eye speed when monkeys were exposed to a block of TVORC trials on three separate occasions when they were initially in the high gain state. Exponential curves were fitted to the data from each training session, and their time constants and learning ratios are summarized in Table 3. It is evident from these data that there was no progressive increase in the degree of adaptation and no consistent change in the rate of acquisition.

To quantify changes in eye velocity dynamics associated with the adaptive changes during the open-loop period of the TVOR, eye speed was averaged over 100 trials in either the high-gain or low-gain states (Fig. 4, solid gray traces from high-gain state, solid black traces from low-gain state). The latency of the TVOR was 10, 7, and 13 msec for monkeys 3, 4, and 5, respectively. There was a zero-latency anti-compensatory eye movement in the initial eye velocity of monkey 5, which may have resulted from the mechanical compression of the eyeball by linear acceleration (Collewijn and Smeets, 2000). To estimate its latency, we compared the averaged eye velocity traces evoked by the same head translation while the monkey viewed a near (16.5 cm) or a far target (275 cm). The two traces diverged at 13 msec from the onset of head translation, which was taken as an upper estimate of the TVOR latency for monkey 5. Averages were also obtained of the first 20 trials (short dashed traces) and the last 100 trials (long dashed traces) of the gain decrease training on day 1 (Fig. 2, top panel, filled symbols). Comparison of the eye velocity trajectories revealed that there were at least two sequential temporal components during the open-loop interval of the TVOR. The late component (Fig. 4, black arrows) was more labile because it was parametrically modified after only a few trials of a new context (compare solid gray and short dashed traces in Fig. 4). The latency of the late component was 35, 34, and 55 msec for monkeys 3, 4, and 5, respectively (Fig. 4, black arrows). The late component was further adapted by additional training trials presented during the same day. In all three monkeys, the amplitude of the late component was minimized after the low gain states were achieved (solid black traces). In contrast to the lability of the late component, the early component of initial eye velocity appeared to be uninfluenced by changes in task context. When given additional training trials, however, part of the early component (Fig. 4, gray arrows) was systematically modified. This interval was ∼20-35 msec for monkey 3, ∼26-33 msec for monkey 4, and ∼32-55 msec for monkey 5. Because of technical limitations and safety concerns, we limited peak acceleration to <0.3 g. It is possible that an earlier expression of motor learning might be detected if higher accelerations were used (Clendaniel et al., 2001).

The initial eye velocity of the TVOR exhibited different dynamics in high gain (solid gray traces) or low gain state (solid black traces). In the high gain state, linear regression analysis revealed that the initial eye velocity resembled head velocity with a regression coefficient (r²) of 0.99, 0.98, and 0.98 for monkeys 3, 4, and 5, respectively. In the low gain state, however, the regression coefficients were reduced to 0.27, 0.52, and 0.01 for monkeys 3, 4, and 5, respectively. Furthermore, eye velocity in the low gain state quantitatively resembled head acceleration, with a regression coefficient of 0.67, 0.70, and 0.29 for monkeys 3, 4, and 5, respectively.

Specificity of TVOR motor learning

TVOR motor learning is specific to combinations of the head motion and evoked eye response.

In the first experiment, we used TVORC trials to induce reductions in initial eye speed at a heading of 30° to the right, which evoked leftward compensatory eye movements (Fig. 6A, black arrow). Before and after training, we measured initial eye speeds at a heading of 30° to the left (Fig. 6A, white arrow), which evoked rightward compensatory eye movements. In the trained direction, the learning ratio was 0.49 for monkey 4 and 0.48 for monkey 6, indicating a significant reduction in eye velocity resulted from training. However, the learning ratio in the test direction was 0.96 for monkey 4 and 1.01 for monkey 6 (p > 0.05), indicating that motor learning did not transfer from the training direction to the test direction, although the two directions were similar in a vestibular coordinate frame. Thus, adaptation sites are unlikely to be in the sensory limb of the reflex (peripheral end organs and primary afferents), which transforms linear accelerations into neuronal signals and then transmits them into the vestibular nuclei.

In the second experiment, we examined whether adaptation sites are in the motor limb of the reflex (motoneurons and extraocular muscles), which converts the neuronal signals into contractions of extraocular muscles. To achieve this, we took advantage of the fact that the vertical TVOR depends not only on heading direction, but also on the orientation of the head with respect to earth. For example, if a monkey's head is tilted to the left and the animal is translated to the right, the TVOR generates oblique eye movements with upward components (Fig. 6B). Alternatively, if the head is tilted to the right, leftward translation generates oblique eye movements that are also accompanied by upward components (Fig. 6B). If the learning site were located in the motor limb of the reflex, then adaptation should be expressed whenever eye movements are generated in the trained direction, although the heading directions might be different. Two monkeys were trained using TVORC trials with their heads and bodies tilted 20° to the right with leftward translation (Fig. 6B, black arrow). Learning ratios in the vertical eye movements were 0.64 for monkey 3 and 0.65 for monkey 6, indicating a significant adaptation (Fig. 6B, right panel black bars). After training, monkeys were tilted 20° to the left, and initial eye velocity was measured during rightward translations (Fig. 6B, white arrow). In this direction, the evoked compensatory eye movements had the same upward components, as did the trained eye movements. However, learning ratios in vertical eye movements were 0.94 for monkey 3 and 1.02 for monkey 6, indicating that learning did not transfer to the test condition. These data suggest that adaptation does not occur at the motor limb of the reflex. Similar results have been reported in AVOR adaptation in cats (Baker et al., 1987).

In the third experiment, we examined whether motor learning could be induced in an eye that remains motionless during training (Fig. 7). The monkey was required to fixate a near target (10° vergence angle) that was aligned with the left eye while subjected to backward head translations in the nasal-occipital direction. The compensatory eye movements of the right eye (i.e., the converged eye) were rightward (Fig. 7A, middle panel, thick black trace). However, the same head translations evoked much smaller horizontal compensatory eye movements in the left eye (i.e., the aligned eye) (Fig. 7A, bottom panel). The monkey was given a block of TVORC trials that resulted in smaller compensatory eye movements of the right eye (Fig. 7A, middle panel, dotted trace). The effect of training on the left eye was examined by comparing its initial eye speed in a test condition before and after training. In the test condition, the target was placed at the same initial distance from the monkey, but displaced to the left of the training target (Fig. 7B). In this condition, the same head translation used in the trained condition (Fig. 7A) evoked rightward compensatory eye movements in both eyes (Fig. 7B, middle and bottom panels, thick black traces). The dashed traces in Figure 7B shows that learned changes were present in the response of the right eye (middle panel) but not in the response of the left eye (bottom panel), suggesting that the TVOR motor learning is sensitive to the combination of head translation and evoked eye movement used in the training.

TVOR motor learning is not specific for translation acceleration and target distance

The slope of eye velocity versus head acceleration was measured before (Fig. 8A, filled circles in the left panel and black bars in the right panel) and after training using an acceleration of 0.2 g (Fig. 8A, open circles in the left panel and white bars in the right panel). Figure 8A shows that the initial eye speed and translation acceleration were still linearly related but with a reduced slope in the three monkeys tested, suggesting that motor learning transferred to other accelerations.

The initial eye speed of the TVOR is inversely proportional to viewing distance (Schwarz et al., 1989; Snyder and King, 1992). The slope of initial eye velocity versus viewing distance was measured before and after training using a target with a viewing distance of 22 cm. Figure 8B shows that initial eye speeds and inverse viewing distance were still linearly related after training but with a reduced slope in five monkeys tested, suggesting that learning transferred to other viewing distances.

TVOR adaptation involves neural plasticity

Several pieces of evidence support the idea that the observed changes in the TVOR after training are attributable to neural plasticity rather than simply to strategy-specific parametric adjustments. First, the training induced changes in initial compensatory eye velocity that followed an exponential time course (Fig. 2). Second, the effects of training were retained between training sessions without reinforcement (Fig. 2), even during several days of normal visual-vestibular experience in the home cage, suggesting that the learned changes were specific to the stimuli experienced during the training. Third, the rate of acquisition and the degree of motor learning finally achieved were independent of previous exposures to the training condition (Fig. 3). Fourth, motor learning was specific to the direction and modality of head motion and was only expressed in conditions that had the combined head motion and evoked eye movements experienced during training (Figs. 5, 6, 7). These characteristics of TVOR motor learning argue against the hypothesis that the observed motor learning is simply the result of a general cognitive strategy that presets the reflex's amplitude to a higher or lower value.

Motor learning modifies dynamics of the initial compensatory eye speed of the TVOR

In the open-loop interval of the TVOR, we identified at least two temporally distinct components. The later component began ∼34-55 msec after the onset of head motion and was labile. Its amplitude not only could be changed immediately after a new context, but also could be suppressed completely by additional training trials (Figs. 2, 4). It is also evident that there is a substantial difference in the dynamics of the open-loop responses. The profiles of initial eye velocity resembled the profile of head velocity in the high gain state, but resembled the profile of head acceleration in the low gain state. These data suggest that motor learning may target the integration process that converts head acceleration signals into eye velocity signals. In contrast to the late component, the early component apparently is not subject to parametric adjustment and may be only partially modified by additional training trials. Because the measured latency of learning expression is dependent on the intensity of vestibular stimuli in both TVOR (N. Lin, W. Zhou, and W. M. King, unpublished results) and AVOR adaptation studies (Khater et al., 1993; Clendaniel et al., 2001) and our accelerations were limited to 0.3 g, we cannot rule out the possibility that TVOR motor learning may occur within the shortest pathway that generates the TVOR.

Unlike the well established three-neuron arc pathway identified with the AVOR (Lorente De No, 1933; Fernandez and Goldberg, 1971; Scudder and Fuchs, 1992), the neural pathways involved in the TVOR have just begun to be studied (Fernandez and Goldberg, 1976; Uchino et al., 1994, 1996; Angelaki et al., 2001; Chen Huang and McCrea, 1999). Snyder and King (1992) hypothesized that different neural pathways may contribute to the temporal response components of the TVOR. The earliest component of the open loop eye velocity might involve brainstem circuits that include central vestibular and oculomotor neurons via the direct utriculoabducens pathway and the utriculo-vestibulo-oculomotor pathways (Uchino et al., 1994, 1996). In contrast, the latency of the latest component of the open loop eye velocity interval is long enough to involve the vestibulo-cerebellum and even cerebrovestibular pathways. Lisberger et al. (1994a,b,c) and Lisberger and Pavelko (1988) showed that gaze-velocity Purkinje cells project monosynaptically to the vestibular nuclei and inhibit neurons involved in the AVOR. Single-unit recordings by Snyder and King (1996) showed that this cell type also encodes a signal related to translational acceleration, suggesting a similar role for the cerebellum in the modulation of the TVOR. Furthermore, the parametric modulation of the late component after a change of context may involve a cerebrovestibular or cerebrocerebellar vestibular pathway wherein a cortical signal reflecting the animal's intention modifies or interacts with afferent vestibular signals. Several studies demonstrate that neurons in the frontal and parietal eye fields encode attention/intention signals associated with task context (for review, see Colby and Goldberg, 1999; Schall and Thompson, 1999) and project monosynaptically to the brainstem VOR pathways (for review, see Guldin and Grusser 1998), suggesting that they could be involved in parametric adjustment of the TVOR.

Sites of neural plasticity

Figure 9 provides a conceptual framework to understand potential neural mechanisms underlying TVOR learning. Linear acceleration may generate compensatory eye movements in any direction and of any amplitude depending on heading, gaze direction, head orientation, and viewing distance (Paige and Tomko, 1991a,b). Thus, otolith afferent signals must be selectively routed to appropriate premotor circuits to generate the correct eye movement response. Figure 9 schematically illustrates a possible mechanism to transform otolith inflow into upward or downward eye movements depending on head orientation. We assume that eye movement direction is selected or gated by signals that are related to head orientation. This assumption is supported by recent morphological evidence for abundant axo-axonic synapses within the medial vestibular nucleus (Holstein et al., 1999). When the head is tilted to the left, transmission of otolith signals related to rightward translation is blocked by inhibition (−Ø_H) from reaching the downward eye movement pathways. However, otolith signals to the upward eye movement pathways are facilitated by disinhibition. Thus, signals related to head acceleration can be channeled into particular premotor eye movement pathways. A goal of this study was to determine whether learning sites could be localized within the sensory limb, the motor limb, or the central sensory-motor transformation stage (vestibular nuclei, cerebellum, and cortex) of the reflex. Our data suggest that learning sites cannot be located at synapse 1 (Fig. 9); otherwise learned changes should be expressed for translations that are similar to the trained direction regardless of the direction of the elicited eye movements (Figs. 6A, 7). The learning sites cannot be at synapse 3 (Fig. 9); otherwise learned changes would be expressed in the eye movement that was in the trained direction but evoked by untrained stimuli (Fig. 6B,Table 4). By exclusion, we conclude that the learning sites are centrally located, presumably in the brainstem and cerebellum. A novel feature of the model illustrated in Figure 9 is that the learning sites are hypothesized to be synapses located on the dendrites of the postsynaptic neuron (site 2). According to this arrangement, only synapses of the neuron that receive inputs from otolith afferents will be modified during TVOR motor learning. Other synapses of the neuron that receive signals from canal afferents or the smooth pursuit system, remain unchanged. Thus, this model allows for the modification of the response of a neuron to a subset of its inputs, a feature essential to implement modality specificity. Furthermore, Figure 7 shows that otolith activation alone was not adequate to induce motor learning in the TVOR pathway because learning did not take place in the eye without evoked eye movements during training. These data suggest that induction of learning may require the activation of both a vestibular input (presynaptic) and an oculomotor output (postsynaptic), which is characteristic of Hebbian associative learning (for review, see Brown et al., 1990). These characteristics are also shared by rapid saccade adaptation (Fuchs et al., 1996) and smooth pursuit adaptation (Kahlon and Lisberger, 1996, 1999; Chou and Lisberger, 2002). The similarities suggest that motor learning underlying these phenomena may use similar neural mechanisms and that these mechanisms may be widely used by the CNS.

• Download figure
• Open in new tab
• Download powerpoint

Figure 9.

Hebbian model of the TVOR motor learning. Three potential neural learning sites are shown: 1, sensory limb of the reflex; 2, sensory-motor transformation; 3, motor limb of the reflex. At the sensory-motor transformation stage, the TVOR, the AVOR, and smooth pursuit systems may share the same group of neurons located in the vestibular nucleus and/or the nuclei prepositus hypoglossi. However, each system has its own synaptic contacts with target neurons so that motor learning can be induced independently in each subsystem. Four types of signals are proposed to modulate otolith sensory inflow to oculomotor neurons through presynaptic or postsynaptic inhibition. One signal gates otolith sensory signals according to head orientation so that eye movements can be generated in the correct direction depending on head orientation in space (i.e., Ø_H). The second signal is hypothesized to modulate the otolith sensory inflow according to a behavioral context (i.e., VORC condition). The third signal modulates the amplitude of the TVOR according to viewing distance. The fourth signal carries the feedback of its own output.