Mark Shelhamer

Short-term vestibulo-ocular reflex adaptation in humans - I. Effect on the ocular motor velocity-to-position neural integrator

We investigated the effect of short-term vestibulo-ocular reflex (VOR) adaptation in normal human subjects on the dynamic properties of the velocity-to-position ocular motor integrator that holds positions of gaze. Subjects sat in a sinusoidally rotating chair surrounded by an optokinetic nystagmus drum. The movement of the visual surround (drum) was manipulated relative to the chair to produce an increase (× 1.7 viewing), decrease (× 0.5, × 0 viewing), or reversal (× (-2.5) viewing) of VOR gain. Before and after 1 h of training, VOR gain and gaze-holding after eccentric saccades in darkness were measured. Depending on the training paradigm, eccentric saccades could be followed by centrifugal drift (after × 0.5 viewing), implying an unstable integrator, or by centripetal drift [after × 1.7 or × (-2.5) viewing], implying a leaky integrator. The changes in the neural integrator appear to be context specific, so that when the VOR was tested in non-training head orientations, both the adaptive change in VOR gain and the changes in the neural integrator were much smaller. The changes in VOR gain were on the order of 10% and the induced drift velocities were several degrees per secend at 20 deg eccentric positions in the orbit. We propose that (1) the changes in the dynamic properties of the neural integrator reflect an attempt to modify the phase (timing) relationships of the VOR and (2) the relative directions of retinal slip and eye velocity during head rotation determine whether the integrator becomes unstable (and introduces more phase lag) or leaky (and introduces less phase lag).

Short-term vestibulo-ocular reflex adaptation in humans. II. Error signals.

We oscillated humans sinusoidally at 0.2 Hz for 1 h, using various combinations of rotations of the head and visual surround to elicit short-term adaptation of the gain of the vestibulo-ocular reflex (VOR). Before and after each period of training, the gain of the VOR was measured in darkness, in response to a position step of head rotation. A small foveal target served as well as a full-field stimulus at driving VOR adaptation. Oscillation of the visual surround alone produced a substantial increase in the VOR gain. When the visual scene was rotated in phase with the head but with a larger amplitude to produce a reversal of the VOR, the VOR gain increased if the movement of the visual scene was much greater than that of the head, otherwise the gain decreased. We interpreted these results with a model of VOR adaptation that uses as its “error signal” the combination of motion of images on the retina (retinal slip) and any additional slow-phase eye velocity, beyond that generated by the VOR through the vestibular nuclei, necessary to prevent such retinal slip during head rotation. The slow phase velocity generated by the VOR is derived from “inferred head rotation”, a signal based on the discharge of neurons in the vestibular nuclei that receive both labyrinthine and visual (optokinetic) inputs. The amplitude and sign of the ratio of the “error signal” to “inferred head velocity” determine the amplitude and the direction (increase or decrease) of VOR gain adaptation.

Sensorimotor adaptation error signals are derived from realistic predictions of movement outcomes.

Neural systems that control movement maintain accuracy by adaptively altering motor commands in response to errors. It is often assumed that the error signal that drives adaptation is equivalent to the sensory error observed at the conclusion of a movement; for saccades, this is typically the visual (retinal) error. However, we instead propose that the adaptation error signal is derived as the difference between the observed visual error and a realistic prediction of movement outcome. Using a modified saccade-adaptation task in human subjects, we precisely controlled the amount of error experienced at the conclusion of a movement by back-stepping the target so that the saccade is hypometric (positive retinal error), but less hypometric than if the target had not moved (smaller retinal error than expected). This separates prediction error from both visual errors and motor corrections. Despite positive visual errors and forward-directed motor corrections, we found an adaptive decrease in saccade amplitudes, a finding that is well-explained by the employment of a prediction-based error signal. Furthermore, adaptive changes in movement size were linearly correlated to the disparity between the predicted and observed movement outcomes, in agreement with the forward-model hypothesis of motor learning, which states that adaptation error signals incorporate predictions of motor outcomes computed using a copy of the motor command (efference copy).

Context-specific adaptation of saccade gain

Previous studies established that vestibular reflexes can have two adapted states (e.g., gain) simultaneously, and that a context cue (e.g., vertical eye position) can switch between the two states. The present study examined this phenomenon of context-specific adaptation for horizontal saccades, using a variety of contexts. Our overall goal was to assess the efficacy of different context cues in switching between adapted states. A standard double-step paradigm was used to adapt saccade gain. In each experiment, we asked for a simultaneous gain decrease in one context and gain increase in another context, and then determined if a change in the context would invoke switching between the adapted states. Horizontal eye position worked well as a context cue: saccades with the eyes deviated to the right could be made to have higher gains while saccades with the eyes deviated to the left could be made to have lower gains. Vertical eye position was less effective. This suggests that the more closely related a context cue is to the response being adapted, the more effective it is. Roll tilt of the head, and upright versus supine orientations, were somewhat effective in context switching; these paradigms contain orientation of gravity with respect to the head as part of the context.

Recurrence matrices and the preservation of dynamical properties

Abstract We prove that the construction of a recurrence matrix (a matrix of inter-vector distances) preserves the dynamical properties of an observed attractor. Because of this fact, a recurrence matrix is a useful transform for studying high-dimensional systems, “compressing” high-dimensional vector sets into a two-dimensional format without losing relevant information. Practical issues are discussed in an example.

Eye-position dependence of torsional velocity during interaural translation, horizontal pursuit, and yaw-axis rotation in humans

The translational vestibulo-ocular reflex (tVOR) stabilizes an image on the fovea during linear movements of the head. It has been suggested that the tVOR may share pathways with the pursuit system. We asked whether the tVOR and pursuit would be similar in their behavior relative to Listings Law. We compared torsional eye velocity as a function of vertical orbital position during interaural translation, pursuit, and yaw-axis rotation. We found that the eye-position-dependence of torsion was similar during translation and pursuit, which differed from that during yaw-axis rotation. These findings further support a close relationship between the mechanisms that generate pursuit and the tVOR.

Short-term adaptation of the phase of the vestibulo-ocular reflex (VOR) in normal human subjects.

We investigated the effects of short-term vestibulo-ocular reflex (VOR) adaptation on the gain and phase of the VOR, and on eccentric gaze-holding in darkness, in five normal human subjects. For 1 h, subjects sat in a chair that rotated sinusoidally at 0.2 Hz while surrounded by a visual stimulus (optokinetic drum). The drum was rotated relative to the chair, to require a VOR with either a phase lead or lag of 45 deg (with respect to a compensatory phase of zero) with no change in gain, or a gain of 1.7 or 0.5 with no change in phase. Immediately before and after each training session, VOR gain and phase were measured in the dark with 0.2 Hz sinusoidal rotation. Gaze-holding was evaluated following 20 deg eccentric saccades in darkness. Adaptation paradigms that called only for a phase lead produced an adapted VOR with 33% of the required amount of phase change, a 20% decrease in VOR gain, and an increased centripetal drift after eccentric saccades made in darkness. Adaptation paradigms that called for a phase lag produced an adapted VOR with 29% of the required amount of phase change, no significant change in VOR gain, and a centrifugal drift after eccentric saccades. Adaptation paradigms requiring a gain of 1.7 produced a 15% increase in VOR gain with small increases in phase and in centripetal drift. Adaptation paradigms requiring a gain of 0.5 produced a 31% decrease in VOR gain with a 6 deg phase lag and a centrifugal drift. The changes in drift and phase were well correlated across all adaptation paradigms; the changes in phase and gain were not. We attribute the effects on phase and gaze-holding to changes in the time constant of the velocity-to-position ocular motor neural integrator. Phase leads and the corresponding centripetal drift are due to a leaky integrator, and phase lags and the corresponding centrifugal drift are due to an unstable integrator. These results imply that in the short-term adaptation paradigm used here, the control of drift and VOR phase are tightly coupled through the neural integrator, whereas VOR gain is controlled by another mechanism.

Sensory, motor, and combined contexts for context-specific adaptation of saccade gain in humans

Saccadic eye movements can be adapted in a context-specific manner such that their gain can be made to depend on the state of a prevailing context cue. We asked whether context cues are more effective if their nature is primarily sensory, motor, or a combination of sensory and motor. Subjects underwent context-specific adaptation using one of three different context cues: a pure sensory context (head roll-tilt right or left); a pure motor context (changes in saccade direction); or a combined sensory-motor context (head roll-tilt and changes in saccade direction). We observed context-specific adaptation in each condition; the greatest degree of context-specificity occurred in paradigms that used the motor cue, alone or in conjunction with the sensory cue.

An internal clock generates repetitive predictive saccades

Previously we demonstrated the presence of a behavioral phase transition between reactive and predictive eye tracking of alternating targets. Prior studies of repetitive movements have proposed that an “internal clock” is the neural mechanism by which interval timing is achieved. In the present report we tested whether predictive oculomotor (saccade) tracking is based on an internal time reference (clock) by examining the effect of transient perturbations to the periodic pacing stimulus. These perturbations consisted of altering the timing of the stimulus (abruptly increasing or decreasing the inter-stimulus interval) or extinguishing the targets altogether. Although reactive tracking (at low pacing rates) was greatly affected by these timing perturbations, once predictive tracking was established subjects continued to time their eye movement responses at the pre-existing rate despite the perturbation. As expected from certain clock models, inter-stimulus intervals for predictive tracking followed Weber’s law and the scalar property (timing variability increases in proportion to interval duration), but this was not true for reactive tracking. In addition, the perturbation results show that subjects can establish an internal representation of target pacing (the internal clock) in as little as two eye-movement intervals, which suggests that this mechanism is relevant for real-world situations. These findings are consistent with the presence of an internal clock for the generation of these predictive movements, and demonstrate that the neural mechanism responsible for this behavior is temporally accurate and flexible.

Context-specific short-term adaptation of the phase of the vestibulo-ocular reflex

Abstract The phase of the angular vestibulo-ocular reflex (VOR) is subject to adaptive control. We had previously found that adapting the phase of the VOR also produced changes in drift on eccentric gaze-holding, implying a change in the time constant of the velocity-to-position neural integrator. Here we attempted to dissociate changes in gaze-holding drift from changes in the phase of the VOR. In normal human subjects, for 2 h, we alternated 5 min of VOR phase adaptation (sinusoids, 0.2 Hz) with 5 min of making saccades in the light with the head stationary. Afterwards, changes in VOR phase were the same (32% of requested) as those obtained with 1 h of phase adaptation alone, but changes in drift following saccades were much smaller than those found after phase adaptation alone (0.8°/s compared with 5°/s). When measuring drift after VOR steps, however, the changes were closer to those found after phase adaptation alone (3.8°/s). To test the relationship between gaze-holding drift after VOR steps and adaptive changes in VOR phase, we alternated sinusoidal VOR phase adaptation with normal VOR steps in the light. In this paradigm, the adaptive change in VOR phase was about the same as with phase-adaptation alone (35%), but there was now little drift after saccades (1.9°/s) or after VOR steps (0.7°/s). We conclude that the state of the velocity-to-position neural integrator can be altered selectively and rapidly depending upon the task required. Such context-specific adaptation is advantageous, because it allows adjustment of the phase of the VOR without degrading the ability to hold eccentric fixation.