To keep a stable internal representation of the environment as we move, extraretinal sensory or motor cues are critical for updating neural maps of visual space. Using a memory-saccade task, we studied whether visuospatial updating uses vestibular information. Specifically, we tested whether trained rhesus monkeys maintain the ability to update the conjugate and vergence components of memory-guided eye movements in response to passive translational or rotational head and body movements after bilateral labyrinthine lesion. We found that lesioned animals were acutely compromised in generating the appropriate horizontal versional responses necessary to update the directional goal of memory-guided eye movements after leftward or rightward rotation/translation. This compromised function recovered in the long term, likely using extravestibular (e.g., somatosensory) signals, such that nearly normal performance was observed 4 mo after the lesion. Animals also lost their ability to adjust memory vergence to account for relative distance changes after motion in depth. Not only were these depth deficits larger than the respective effects on version, but they also showed little recovery. We conclude that intact labyrinthine signals are functionally useful for proper visuospatial memory updating during passive head and body movements.
To compute the spatial locations of objects, the brain relies primarily on visual information from the retina. However, because these spatial locations are encoded relative to the current direction of the eyes in space, these retinal coordinates become obsolete as soon as the eyes or head move. Using remembered visual information and by taking into account the amplitude/direction of the intervening eye and head motion, both humans and monkeys can look accurately at the remembered location of a flash, even when an eye or head movement had intervened between the target flash and the subsequent saccade (Hallett and Lightstone 1976; Herter and Guitton 1998; Klier et al. 2005; Li and Andersen 2001; McKenzie and Lisberger 1986; Medendorp et al. 2002, 2003; Ohtsuka 1994; Pelisson et al. 1989; Schlag et al. 1990; Sparks and Mays 1983; Zivotofsky et al. 1996). This process, which has been extensively studied during double-step saccades, is referred to as visuospatial “updating,” because the saccadic goal has to be modified from the original retinal error corresponding to the flash. The presumed extraretinal signal necessary for visuospatial updating during double-step saccades is typically believed to arise as a corollary discharge of the evoked eye movement that is subtracted from the retinotopic retinal error (Duhamel et al. 1992; Goldberg and Bruce 1990; Sommer and Wurtz 2004) or used to convert spatial locations in extraretinal (i.e., head, body or world-fixed) reference frames (Brotchie et al. 1995; Li and Andersen 2001; Snyder et al. 1998; Xing and Andersen 2000; Zipser and Andersen 1988).
Visuospatial updating also occurs after intervened head and body movements. For example, human subjects can update the sensorimotor goal after a head/body reorientation (Blouin et al. 1998; Israël et al. 1999; Klier et al. 2005; Medendorp et al. 2002; Van Pelt et al. 2005) or translation (Medendorp et al. 2003). Similarly, trained monkeys can execute proper memory eye movements after intervened rotational or translational movements, although with greater variability than double-step or retinotopic memory saccades (Baker et al. 2003; Li and Angelaki 2005; Li et al. 2005).
Whenever gaze in space changes through a displacement of the head in space, vestibular signals might contribute by providing the brain with information about the amplitude and direction of the intervened movement. This hypothesis has been supported by patient studies, which have reported deficits in vestibular perception and path integration functions in labyrinthine-defective subjects (Grasso et al. 1999; Israël and Berthoz 1989; Kanayama et al. 1995). More recently, intact vestibular labyrinths were shown to be critical for visuospatial updating during motion in depth (Li and Angelaki 2005). When tested within a week after bilateral labyrinthectomy, trained animals had completely lost the ability to adjust memory vergence as required by the intervening forward or backward displacements. Here we study whether vestibular signals are also critical for visuospatial constancy during lateral translation and yaw rotation, movements that require mostly updating of target direction (rather than depth). We report differences in the role of vestibular information in visuospatial constancy for left/right versus depth movements, not only in the acute deficits, but also in the extent of functional recovery.
Data were collected from two rhesus monkeys with chronically implanted head ring and scleral eye coils. Details for binocular eye movement recordings, experimental setup, and protocols are identical to those in Li et al. (2005) and Li and Angelaki (2005). Briefly, animals were trained to perform memory-guided eye movements from a central, far target (a head-fixed LED located 1 m away—fixation target) to one of several briefly flashed peripheral targets that remained fixed in space (Fig. 1A). The peripheral targets were either one of two LEDs lined up directly in front of the monkey at distances of 17 or 27 cm (lateral translation/rotation task) or one of four eccentric LEDs symmetrically placed 3 cm up, down, left, and right at the same distances (forward/backward motion task).
Monkeys performed randomly interleaved motion and stationary memory eye movement tasks within the same block of trials (see Li and Angelaki 2005; Li et al. 2005 for details). For motion tasks, animals were passively moved during the delay period, such that the spatial goal of the memory eye movement differed from the retinal location of the flash. The motion (5 cm for translation; 10° for rotation) lasted 0.5 s (with a peak acceleration of 2.5 m/s2 and velocity of 19°/s) and was completed before the central fixation target was turned off. Because animals were required to continue fixating the central, head-fixed LED (in an otherwise totally dark room) during motion, they suppressed the vestibulo-ocular reflex (VOR). Prelesion mean eye position changes (computed across 20 ms at the beginning and end of motion) averaged –0.08 ± 0.29° (animal 1) and –0.01 ± 0.22° (animal 2) for version and –0.09 ± 0.21° (animal 1) and –0.26 ± 0.31° (animal 2) for vergence.
For the stationary task, which used the same targets and were executed from the same initial (initial-S) or final (final-S) position as the motion trials, monkeys made memory-guided eye movements while stationary in space. Because the required memory eye movement was toward the location of the flash and no updating was required, these stationary trials served as controls to evaluate the animal’s updating capacity for the respective motion trials. These stationary task trials were in any other way identical to the corresponding motion trials.
Both animals were extensively trained with these tasks for ∼1 yr. Eye movement performance was monitored on-line using behavioral windows for both version and vergence (with fixation windows for visible targets of ±2° for version and ±0.75° for vergence). For all runs in which the animals fixated the relit target, a drop of juice was given at the end of the trial. Furthermore, to motivate accuracy in the memory-guided eye movements, whenever memory version or vergence fell within ±8 and ±4°, respectively, of the memory target for 1,000 ms, an additional drop of juice was given. Nevertheless, all data (both control and lesion), regardless of behavioral performance during the memory period (i.e., independently of whether animals received single or double juice), were saved for off-line analyses. This was important since our goal was to quantify differences in the accuracy of memory-guided eye movements in trained animals before and after labyrinthine lesions, without imposing any a priori behavioral requirement during the memory period itself.
Once normal responses were obtained, the vestibular labyrinths were lesioned bilaterally (Angelaki et al. 2000; Newlands et al. 2002). The animals were allowed 2 days to recover from surgery before returning to controlled water intake and the memory eye movement tasks. The first data set was collected 3–7 days after the operation (1 wk of testing). Both animals were retested with an identical task on regular intervals (2 or 4 wk) ≤4 mo after the surgery.
Off-line, a semi-automatic procedure was used to identify saccades when eye velocity exceeded (or fell below) 25°/s. Because there were no behavioral criteria to accept or reject trials based on the memory period performance on-line, an off-line criterion was used to ensure that animals actually made a memory eye movement. Thus only runs with memory saccade latencies between 150 and 350 ms (94% of all saved runs) were included in further analyses. Data were aligned at the onset of the first memory saccade, and versional/vergence eye positions were computed from left (L) and right (R) eye positions, as (R + L)/2 and R − L (convergence is positive), respectively. To visualize the effects of the lesion, mean version and vergence responses were calculated from superimposed single trial responses. For quantitative analysis, for each experimental run, a memory eye position (EP) was computed by averaging the responses 50–70 ms before the reillumination of the memory target (for single juice trials) or 50–70 ms before the first juice was delivered (for double juice trials). To quantify and compare the animals’ performance over time, an updating index was computed as follows: updating index = (EPmotion − MEPstat.initial)/(MEPstat.final − MEPstat.initial). EPmotion corresponds to individual motion trial values, whereas MEPstat.initial and MEPstat.final are the mean EP values computed from the corresponding interleaved stationary runs in initial and final positions, respectively. The advantage of this analysis and the direct comparison between motion and stationary trials is that we did not need to estimate geometrically what the effect of the movement is, because we had a direct measure of the effect of the displacement during the memory period from the final-S trials.
The efficacy of labyrinthectomy was verified physiologically by monitoring the rotational and translational vestibulo-ocular reflex (RVOR and TVOR, respectively) before, after 1 wk and at regular intervals after labyrinthectomy. The RVOR was tested during yaw rotation (4 Hz: 0.5°; 2 Hz: 1.8°) about an earth-vertical axis in the midsagittal plane, intersecting the line connecting the two auditory meatii (6–7 cm behind the eyes). The TVOR was tested during lateral motion (4 Hz: 10 cm/s; 2 Hz: 24 cm/s). For both stimuli, the animal had to maintain fixation (2° window) for ∼1 s on a near target at a distance of 20 cm; the fixation point was then extinguished, and the monkey was required to maintain fixation of the memorized target (6° window) for another 1 s in darkness. Animals were rewarded after both fixation and motion in darkness intervals, as long as eye position stayed within the specified windows. The magnitude of the eye velocity modulation was quantified by fitting a sum-of-sinusoids (1st and 2nd harmonics) to the data using a nonlinear, least-squares algorithm based on the Levenberg-Marquardt method.
Two monkeys were trained to perform memory-guided eye movements to targets flashed at a distance of 17 or 27 cm, with the animals either stationary or after being passively moved during the delay period (Fig. 1A). Because at the end of the rotational/translational motion trials, animals ended up at a different angle and/or distance relative to the flashed target, the retinal location of the flash was no longer appropriate as a motor error for the memory eye movement. For the monkeys to perform this task accurately and land on the world-fixed location of the memorized flash, extraretinal sensory cues related to the intervening movement were required to compute the updated motor error for the memory eye movement. In contrast, motor error is approximately equal to retinal error for stationary trials and, because no updating is required, they can serve as controls.
Indeed, trained animals appropriately updated the goal of the memory-guided eye movements to land closer to the landing points of the memory saccades of the corresponding stationary runs from the final (final-S), but not initial (initial-S), position (Li and Angelaki 2005; Li et al. 2005), as shown with examples during leftward and backward displacements in Fig. 1, B and C, respectively. Notice that the memory-guided eye movement landed closer to the final-S rather than initial-S mean responses (Fig. 1, B and C, horizontal solid vs. dotted lines). Because the fixation and flashed targets differed in both direction and depth, eye movements consisted of a combination of version and vergence responses. However, differences between initial-S and final-S values were small for vergence during lateral motion/yaw rotation and for version during forward/backward movements. Thus for the following quantitative analyses, we focus on version responses during lateral translation/yaw rotation and vergence responses during forward/backward motion. We compared the versional or vergence eye movements during the motion trials with those of the corresponding interleaved initial and final position stationary runs, before, acutely after (i.e., within the first week), and at different times (≤4 mo) after bilateral labyrinthectomy.
Deficits and recovery during lateral translation/yaw rotation
Mean memory eye movement (version) responses from M1 after being passively rotated or translated to the left or right are summarized in Fig. 2A (Rotation Prelesion and Translation Prelesion traces). In fact, these motion task responses were more similar to those evoked during stationary trials from the same final than initial position (Fig. 2A, cf. Rotation and Translation with Stationary responses), showing that animals updated the goal of the memory eye movement toward the now eccentric spatial location of the flash. In contrast, after bilateral labyrinthectomy, the horizontal versional component in the motion trials was small, more closely resembling those for the initial than final position stationary task responses and suggesting little updating capacity (Fig. 2B). The deficits appeared more severe for the rotation than lateral translation tasks. On the other hand, both initial and final position stationary task responses remained qualitatively similar after the lesion (Fig. 2, Stationary, Postlesion). There was no significant difference in the memory saccade latency distributions of the prelesion and postlesion data (t-test, P > 0.05).
The ability of the monkeys to update the goal of the memory-guided eye movement recovered over time, as shown in Fig. 2C, where Rotation and Translation responses 4 mo later were similar to those before the lesion. These findings have been quantified in Fig. 3, A and B, using an updating index, defined as the difference in memory version between motion and initial position stationary trials relative to the corresponding difference between final and initial position stationary trials (see methods). An updating index of 1 indicates perfect updating (i.e., motion trials = final position stationary trials), whereas an updating index of 0 shows no updating (i.e., motion trials = initial position stationary trials).
Updating indexes for lateral translation and yaw rotation before the lesion were generally <1. This is typical of lateral translation/yaw rotation updating (Klier et al. 2006; Li et al. 2005) and perhaps exaggerated by the presence of an occasional bias for rightward or leftward directions that could vary from day to day. When tested during the first week after labyrinthectomy, the updating ratio was significantly decreased compared with control values [ANOVA, F(1,1848) = 270, P ≪ 0.001], with deficits being larger for rotational than translational movements [ANOVA, F(1,1848) = 10.7, P < 0.001]. The animal’s updating capacity recovered significantly over time [ANCOVA, F(1,3423) = 233, P ≪ 0.001], such that responses 3 and 4 mo after the lesion were similar to prelesion values. Importantly, both the yaw rotational (RVOR) and translational (TVOR) vestibulo-ocular reflexes remained compromised throughout the 4-mo period after labyrinthectomy (Fig. 3C), suggesting that the observed functional recovery in the memory-saccade tasks probably involves extravestibular signals.
Deficits and recovery during motion in depth
Trained animals also appropriately adjusted memory vergence to account for the change in viewing distance during motion in depth (Fig. 4A; see also Li and Angelaki 2005). Specifically, for motion trials, when the monkey was passively moved 5 cm forward or backward (17→12 or 17→22 cm) during the delay period, the vergence angle change for motion trials (green traces) was more similar to that for final (red traces) than initial (blue traces) position stationary trials (Fig. 4A, Prelesion). The ability to adjust memory vergence according to the distance traveled was completely compromised in labyrinthine-lesioned animals (Fig. 4B, 1 wk; see also Li and Angelaki 2005). A week after bilateral labyrinthectomy, motion task data (green traces) were closer to those from the initial (blue traces) than final (red traces) position stationary task responses. This compromise in function persisted throughout the monitored period, as shown in Fig. 4C, which shows corresponding mean responses 4 mo after the operation.
An updating index for all distance/motion combinations (17→12, 17→22, 27→22, and 27→32 cm) before and after the lesion have been plotted as a function of time in Fig. 4D. Although in labyrinthine-intact animals, the updating index was indistinguishable from a value of 1 (t-test, P ≫ 0.05), it dropped to zero values during the first week after labyrinthectomy (t-test, P ≫ 0.05). The animals’ updating capacity during motion in depth showed some recovery over time [ANCOVA, F(1,3594) = 7.3, P = 0.007, Fig. 4D]. Four months after the lesion, the updating index remained indistinguishable from 0 in animal M1 (t-test, P > 0.05), but not in animal M2 (t-test, P < 0.001). Importantly, the updating capacity continued to be compromised, with the index being significantly different from 1 for both animals (t-test, P ≪ 0.001). Thus despite some small recovery in animal M2, labyrinthine-lesioned animals continued to exhibit compromised updating capacity during changes in relative target distance.
These studies have directly tested for a causal role of intact vestibular information on visuospatial constancy. This was done by first training rhesus monkeys to perform memory-guided eye movements with or without intervening motion and testing whether and how this ability was compromised after bilateral destruction of the vestibular labyrinths. Deficits were more severe and showed little recovery of function for head and body movements that changed the relative depth, as opposed to the relative direction, of the memorized visual target. The results strongly support that vestibular information is functionally used for visuospatial updating during passive head/body movements. Possibly, extralabyrinthine information involving tactile or body proprioceptive cues (e.g., trunk inertial sensors, Mittelstaedt 1997) might substitute the missing labyrinthine function in directional visuospatial updating. Notably, body tactile/proprioceptive signals did not similarly substitute vestibular function for memory vergence updating during motion in depth. Thus whatever mechanism takes over for the vestibular signals to provide versional updating during passive lateral displacement does not extend to depth updating.
How could vestibular signals be used for space constancy? Two mechanisms have been proposed for how spatial constancy can be achieved during intervening movement. On one hand, the brain may convert visual information from its original retinal frame to a more stable frame, computing and storing the locations of objects relative to the earth (or torso or inertial frame), so that the stored coordinates remain correct when the eyes and head move (Andersen et al. 1985; Karn et al. 1997; Snyder et al. 1998; Tillery et al. 1991; Van Pelt et al. 2005). Alternatively, the brain may store locations of objects in a retinal frame (Colby et al. 1995; Duhamel et al. 1992; Goldberg and Bruce 1990; Walker et al. 1995). In the latter case, stored eye-centered locations must be continuously updated, whereas in the former case, retinal information must first be converted into a more stable frame (which, again, requires extraretinal signals) before being stored for spatial perception or motor planning.
Which, if any, alternative is used by the brain remains controversial. Accumulating evidence indicates that, at least under certain conditions, spatial locations are stored in an eye-centered reference frame (Baker et al. 2003; Batista et al. 1999; Colby and Goldberg 1999; Henriques et al. 1998; Snyder 2000). If true, when a salient visual stimulus is presented shortly before a saccade, the representation of that stimulus must be updated, or remapped, at the time of the eye, head or body movement. This property has now been shown for saccadic changes in eye position at the level of single cells in posterior parietal cortex (Colby and Goldberg 1999; Duhamel et al. 1992), extrastriate visual areas (Nakamura and Colby 2002), frontal cortex (Goldberg and Bruce 1990; Umeno and Goldberg 1997), and superior colliculus (Walker et al. 1995). Specifically, when the eyes move so that the receptive field of a neuron lands on a recently stimulated screen location, the cell fires as though a stimulus were still present, even though the screen is blank. This property, which has been called remapping to emphasize that visual information is being shifted from the coordinates of the initial eye position to the coordinates of the next eye position, may contribute to maintaining the spatial alignment between the external world and its internal representation.
Can a similar mechanism to that proposed for visuospatial updating during saccadic eye movements also work for vestibular signals? For this to be true, visual receptive fields in visuomotor cortical or superior colliculus neurons should shift during head rotation/translation according to an eye-in-space signal that reflects the suppressed rotational or translational vestibulo-ocular response (see Medendorp et al. 2003 for such a model of updating in eye coordinates). Such signals (often referred to as gaze velocity) have been described not only in the cerebellar flocculus (Lisberger and Fuchs 1978) and oculomotor vermis/fastigial nucleus (Buttner et al. 1991; Sato and Noda 1992; Suzuki and Keller 1988), but also in cortical visuomotor areas, like medial superior temporal area (Thier and Erickson 1992) and the posterior bank and fundus of the arcuate sulcus (frontal pursuit area; Fukushima et al. 2000). Whether gaze-related information is used to implement remapping during head and body movements has yet to be studied (but see Powell and Goldberg 1997).
If visual receptive fields can be remapped using vestibular information signaling, a relative change in target direction, what about relative changes in target depth? There is growing evidence that neurons in both parietal and frontal visuomotor areas have three-dimensional receptive fields (Ferraina et al. 2000; Fukushima et al. 2002; Gnadt and Beyer 1998; Gnadt and Mays 1995). Thus it is conceivable that the process of vestibularly-driven visuospatial remapping may operate not just in the frontoparallel plane but also in depth. Importantly, target distance information is also necessary to implement the distance-dependent component of version (saccadic) updating (Medendorp et al. 2003). At present, despite the presence of vestibular signals in sensorimotor cortex (de Waele et al. 2001; Ebata et al. 2004; Fukushima et al. 2000, 2004), their functional contributions remain obscure.
Finally, it is important to emphasize that, while the above discussion was centered on the hypothesis that depth updating is cortically mediated as part of a multicue saccadic-like remapping in three dimensions, there is an alternative possibility. Depth updating could be related instead to a low level management of the memorized vergence motor error, by feeding the vergence contributions from the vestibular system directly into the hypothesized local vergence feedback loop (Busettini and Mays 2005; Zee et al. 1992). This could be even done without cortical intervention and no need of a three-dimensional update mechanism of extrapersonal space. Further experiments at the subcortical and cortical level are needed to understand the contributions of the vestibular-driven vergence and version responses to visuospatial updating.
This work was supported by National Institutes of Health Grants EY-12814, DC-04260, and DC-07620.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2006 by the American Physiological Society