| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1York Centre for Vision Research, 2Canadian Institutes of Health Research Group for Action and Perception, Departments of 3Psychology, 4Biology, and 5Kinesiology and Health Sciences, York University, Toronto, Ontario M3J 1P3, Canada
Submitted 3 February 2004; accepted in final form 2 June 2004
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
), and sometimes using the head more, as in driving (Land 1992). However, the neural mechanisms of these adaptive processes remain unknown. For example, the deep motor layers of the superior colliculus (SC) are often said to code gaze, with coordination of the eye and head taking place downstream, perhaps in the brain stem and cerebellum (Freedman 2001; Fuchs et al. 1985; Guitton 1992; Optican and Robinson 1980; Pelisson et al. 2003; Robinson and Fuchs 2001; Sparks et al. 2001; Waitzman et al. 2002), but it is much less clear whether these lower-level mechanisms are themselves responsible for the adaptive and context-dependent aspects of eye–head coordination, or whether higher-level inputs from the cerebral cortex are required.
A number of studies have looked for the anatomic sites and brain stem mechanisms of adaptation for saccades (Edelman and Goldberg 2002; Optican 1985; Optican and Robinson 1980; Zee and Optican 1985) as well as for the vestibuloocular reflex (VOR) (Galiana 1986; Miles et al. 1985; Viirre et al. 1988). Several studies have looked at the behavioral aspects of adaptation in eye–head coordination (Ceylan et al. 2000; Crawford and Guitton 1997; Mellvill Jones et al. 1988; Misslisch et al. 1998; Pathmanathan et al. 2001; Phillips et al. 1997; Stahl 2001). However, very few studies have looked directly at the adaptive neural mechanisms for eye–head coordination during gaze shifts. In contrast to behavioral controls (Coimbra et al. 2000), gaze shifts evoked by stimulating the SC of the cat failed to compensate for acutely increased loads on the head. However, this task involved the preservation of normal gaze kinematics rather than the long-term acquisition of new, context-dependent patterns of eye–head coordination.
One context-dependent way to change the relative contributions of the eye and head to gaze shifts in the primate is to train subjects to make head-free gaze shifts while looking through a small head-fixed aperture—the "goggles task" (Ceylan et al. 2000; Crawford and Guitton 1997; Mislisch et al. 1998; Stahl 2001). Learning this task forces 2 main changes in the neural control system. First, the amount of head contribution to the gaze shift must increase so that the head moves directly toward the target by about the same amount as the desired gaze shift. Second, in the goggles task, the eye must look through the aperture to foveate the target.
The latter requirement is not as trivial as it might sound. Because the eye and head do not generally rotate in exactly the same direction during oblique gaze shifts (Tweed et al. 1995), an increased head rotation will not automatically recenter the eye. After goggle training, the eye does not simply stay stuck at one position (i.e., by reducing saccade and VOR gain to zero), nor does it simply roll back further in the head because of the increased head movement. Rather, the eye engages in a coordinated pattern of saccade and VOR movements such that it ends up at the trained location. If the eye position is initialized at a new aperture location, monkeys trained on a different aperture location continued to drive the eye toward the old (now occluded) aperture location until retrained (Crawford and Guitton 1997). However, despite the complexity of these learning processes, once trained, monkeys are able to switch between the goggles strategy and the normal default eye–head coordination strategy as a function of context (Crawford et al. 1999).
The purpose of this study was to determine whether these context-dependent learning strategies were implemented upstream or downstream from the SC motor command, by training monkeys to switch rapidly between the normal and "goggle" strategies, and then evoking gaze shifts by electrically stimulating the SC in each condition. As shown schematically in Fig. 1A, any context-dependent adaptation in eye–head coordination that is implemented downstream from the SC (i.e., in structures accessed by a fixed output from the SC) should be preserved during SC stimulation (Freedman et al. 1996; Guillaume and Pelisson 2001; Paré et al. 1994). Conversely, context-dependent mechanisms that are implemented only upstream (Fig. 1B) or parallel (Fig. 1C) to the site of stimulation should not be preserved during this paradigm.
|
, 2003). |
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
All surgical and experimental procedures were approved by the York University Animal Care Committee and were in compliance with the Canadian Council on Animal Care policy on the use of laboratory animals. Two juvenile female monkeys, Macaca fascicularis (henceforth designated M1 and M2), underwent aseptic surgery under general anesthesia (isoflurane 1.5% and ketamine 10 mg/kg). The procedure involved the implantation on the skull of 3 types of devices, fixated with the aid of a dental acrylic cap: 1) one stainless steel recording chamber was placed on the skull at 5 mm anterior and 0 lateral in stereotaxic coordinates, aligned such that both the 3rd cranial (oculomotor) nucleus and both colliculi were accessible; 2) a head restraint consisting of a stainless steel cylinder with a rapid-release mechanism was placed frontally (for details see Crawford et al. 1999); and 3) sockets for cable connection were placed dorsally. The animals were also fitted with a pair of scleral search coils, 5 mm in diameter, implanted subconjunctivally in the nasal quadrants of one eye for 3-dimensional (3-D) measurement of eye positions (for details see Tweed et al. 1990). After the surgery, the monkeys were allowed 2 wk for recovery. During the experiments, we temporarily attached 2 orthogonal coils, of 1 cm diameter, to the acrylic skull-cup using a custom-fitted socket, for the measurement of the head position.
Before experimentation, the animals were trained to sit upright in a modified Crist primate chair (Crist Instruments, Hagerstown, MD), which allowed for a large variety of head movements. The device obstructed gaze directions >50° downward, but left the remaining visual range unobstructed, by restraining the animal with only a canvas vest that extended from the neck to the inside of the chair. This primate chair was placed near the center (±15 cm) of 3 orthogonal magnetic fields (Tweed et al. 1990). Coil signals were recorded at a sampling frequency of 100 Hz.
Pinhole goggles task and training
The primates were trained only with the head unrestrained. They first learned to fixate and follow "treats" held at a distance of 80 to 100 cm throughout a large range of eye and head movements (Crawford and Guitton 1997). After this initial training, we fit the animals with a pair of opaque plastic goggles, shaped to the contour of each monkey's face. Initially, a single round aperture was positioned at our best estimate of the center of the mechanical range of the eye, as given by the intersection of a visually measured line between the center of the eye and the auditory canal and a line projected from the pupil, when the monkey was looking straight ahead toward a distant target, with its head restrained. This standard aperture gave the eye a useful visual range of only 10°. We trained the animals to fixate treats through the aperture and then to make horizontal and vertical gaze shifts following the treats, and finally to make oblique gaze shifts. The animals were rewarded with the treats after 4 or 5 gaze shifts. A large and varied range of gaze movements was obtained in an effort to match the large variety of naturally occurring gaze shifts.
The animals were trained in 1-h sessions, until they were able to crisply move the head with the goggles on and look through the aperture for the treats. It took approximately 3 wk of training to establish the basic elements of this task (Crawford and Guitton 1997). The training was continued throughout the subsequent weeks of experiments, and thus the learned behavior was maintained.
Experiments
We began experiments when the animals were fully trained on the goggles task, and could switch between the normal and "goggle" strategy of eye–head coordination, and when we identified the location of the SC sites. Before each stimulation experiment, we reinforced the learned goggles behavior and recorded behavioral data (i.e., the eye and head movements evoked by a visual stimulus) in the 2 conditions: without goggles (no goggles: NG) and with goggles (WG).
Single-unit recordings and microstimulations were performed with the use of tungsten microelectrodes (0.5–2 m
impedance at 1 kHz), which were advanced using a Narishige model MO-95S hydraulic microdrive modified to fit onto the stage of an MO-99 X–Y placement system. The weight of this entire system was 68 g. Once experiments began, animals bore this same weight on their head during all head-free recording conditions.
We identified SC sites by following a 3-step method. First, we identified the interstitial nucleus of Cajal and oculomotor nucleus (Klier et al. 2002), using the clear saccade-related burst-tonic activity of these nuclei to confirm our stereotaxic coordinate system. Second, we moved our electrode "tracks" posterior toward the expected center of the SC topographic saccade map. At the beginning of each SC experiment we recorded unit activity with the animal's head restrained to identify any saccade-related burst activity, Finally, electrical stimulations, of various durations, were delivered through the same electrodes at 0.5-mm intervals. Sites that showed bursting activity during contralateral saccades and that evoked contralateral saccades when stimulated at 50 µA, were classified as SC motor sites and investigated further. Exploration of further sites on subsequent days was aided by following the expected topography of the SC (Freedman et al. 1996; Stanford et al. 1996).
Once the SC site was identified, the head was then freed and each identified site was further explored by use of electrical microstimulation trains (50 µA, 200 Hz). In conditions established for our investigations (Klier et al. 2001, 2003) these are the minimal required parameters to consistently evoke maximal eye–head gaze shifts from most SC sites. We held this current strength constant to provide minimal current spread and consistent comparisons between sites and conditions. Stimulation train duration was initially varied in 100-ms steps to find the minimum duration that most consistently produced maximum amplitude (single-step) gaze shifts that resembled natural gaze shifts. In animal M1 this "optimal duration" was 300 ms and in animal M2, 200 ms. These durations were then held constant (with and without goggles). These stimulation parameters are unlikely to evoke gaze shifts from the superficial sensory layers of the SC (Van Opstal et al. 1990) but have been shown to evoke natural-looking eye–head gaze shifts from the deeper motor layers in the monkey (Freedman et al. 1996; Klier et al. 2001, 2003; Van Opstal et al. 1990). For each site, evoked eye and head movements were recorded in the 2 conditions: "with goggles stimulation" (WGS) and "no goggles stimulation" (NGS). In the NGS condition, we stimulated each SC site (30 to 60 stimulations per site) and recorded the data while the monkey was using an eye–head coordination strategy of its own choosing. In the WGS, we donned the goggles and stimulated the SC again. We moved the electrode by 0.5 mm vertically and repeated the process, in both conditions. After these procedures, we investigated 19 sites from M1 (in 4 tracks) and 38 sites from M2 (in 10 tracks). However, for data analysis we used only those sites that consistently evoked eye + head gaze shifts >10° (i.e., the aperture diameter), which resulted in 12 sites from M1 and 31 sites from M2.
With M2 we also performed the "switching aperture experiment," which took place at the end of a regular experiment. Four additional 10° goggles apertures were positioned 20° up, down, left, and right from the center of the middle aperture. Up until the time of this "switching holes" experiment, these 4 holes were occluded so that the animal was trained only on the center aperture. However, during this experiment we covered the standard central aperture and uncovered the rest of the apertures, one at a time, while monitoring the eye and head positions on the computer. We then delivered the electrical microstimulation when the monkey was looking through each new aperture. This was repeated for each of the 5 apertures, at 7 SC sites.
When the experiments were finished the SC sites were marked with a localized lesion (using a current pulse of 1.5 mA for 15 s) and the brains were removed for histological verification. This confirmed that the stimulation sites reported in this study were situated in the deep motor layers of the SC, extending rostrocaudally from its middle to its caudal border.
Data analysis
A computer program was used to convert the coil signals into eye and head position quaternions for head relative to space (Head) and eyes relative to space (Gaze), and then these values were used to calculate the position for the eye relative to head (Eye) (Crawford et al. 1999; Tweed et al. 1990). Quaternions represent positions as a fixed-axis rotation from a reference position (Westheimer 1957). This reference position was chosen when the monkey was looking straight ahead (in the same directions as the forward-pointing magnetic field), in the NG condition. However, for some analyses we required a "WG reference position," taken when the monkey was wearing the goggles and was looking straight ahead and through the aperture.
Quaternions were used because they provide an accurate and convenient representation of eye and head orientations throughout the entire 360° range in all 3 dimensions. Gaze trajectories, angles of rotation, and angular velocities were computed from these quaternions (Crawford et al. 1999; Tweed et al. 1990). The beginning and the end of the stimulation-evoked eye, head, and gaze movements were selected by an experimenter during visual inspection of the movement traces on a computer screen. The saccade and the VOR components of gaze shifts were differentiated visually by marking (on a computer screen) the inflection point at which eye-in-head velocity reversed, from the eye proceeding in the same direction as head movement to receding in the direction opposite to head movement. We never observed "plateau phases" where eye-in-head position held steady between the saccade and VOR, so this inflection point was always clear. To characterize the typical results of stimulating each SC site, we also calculated the "characteristic vector" (CV) for eye, head, and gaze, using a method described previously (Klier et al. 2001; Martinez-Trujillo et al. 2003). This "characteristic vector" expresses the gaze, head, or eye trajectory that would be expected if the eyes and head were pointing straight ahead at the beginning of the stimulation.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
The purpose of this section is to provide an overview of the behavioral differences between gaze shifts in NG and WG conditions after training, as recorded at the beginning of each experiment. In subsequent sections we will see which aspects of these context-dependent changes are preserved during SC stimulation. The 2 top panels in Fig. 2 show the horizontal components of 2 similar-sized gaze shifts, without (A) and with (B) the goggles, plotted as a function of time. Gaze is plotted along with its constituent eye and head components. As in most head-free gaze shifts, the eye and head move in the same direction during the rapid-saccade portion of the gaze shift (between the dashed vertical lines), and then the eye rolls backward during the subsequent VOR phase. However, during the WG condition (B), notice that the head movement is larger and faster, and that the eye rolls back toward its initial position (at the aperture), such that the overall saccade + VOR eye movement is negligible.
|
) correspond to the end of the movement trajectories. These trajectories are represented by unit-length vectors parallel to the pointing direction of the eye-in-space, head-in-space, and eye-in-head (i.e., showing gaze direction, facing direction, and eye direction, respectively). The horizontal and vertical components of these vectors are plotted, as one would view them from behind the monkey. These vectors thus define a ±90° sphere indicated by the large circles in A and B. The gaze shifts illustrated here were selected so that their gaze amplitudes were similar in the 2 behavioral conditions (mean 33.5° in WG and 32.4° in NG), to allow for comparison between the corresponding eye and head components.
|
) was still relatively large (left column). In contrast, in the WG condition, the eyes rolled back toward the center, landing within the location of the visual aperture (indicated by the central gray circle).
These basic results (quantified in more detail below) are consistent with previous behavioral observations using the goggles task (Crawford and Guitton 1997) as follows: 1) an increase in the size of head movement for a given gaze shift and 2) a restriction in the final eye-in-head position range. This suggests that the animals had learned 2 qualitatively and quantitatively different patterns of eye–head coordination. Once this was learned, the animals could rapidly switch between these 2 patterns simply by donning or removing the goggles. This allowed us to perform the planned electrophysiological experiments.
Overview of stimulation data
One simple hypothesis could be that all of these state-dependent behavioral changes were implemented downstream from the SC (in a functional sense) and that these changes were "switched ON-OFF" by contextual signals, depending on whether the monkey wore the goggles. Previous studies suggest that electrical stimulation of the SC activates a local population of neurons whose response profile is graded with distance from the electrode tip, presumably simulating a fixed motor output from the SC (Bergeron et al. 2003; Freedman et al. 1996; Guitton et al. 1993; Port et al. 2000; Ranck 1981; Sparks 1993). Therefore if the behavioral adaptations described above are implemented downstream from the SC, they should be preserved during stimulation of its intermediate and deep layers.
Without goggles, the average amplitudes of the stimulation-evoked gaze shifts (averaged for each particular stimulation sites and across individual trials for each site) that met our analysis criteria (eye and head gaze shifts >10°) ranged from 10.3 to 90.4°, with corresponding average head movement amplitude ranging from 4.5 to 80.3° and corresponding average evoked-saccade amplitudes ranged between 4.5 and 22.9°. Here we provide a qualitative overview of the main results before embarking on a quantitative comparison of the data.
The bottom panel of Fig. 2 shows typical horizontal movement trajectories during movements evoked from one SC stimulation site, in the NGS condition (C) and WGS condition (D), matched in size to the behavioral gaze shifts shown in the top panel. As in the NG behavioral data (A), in the NGS data (C) one can see the natural sequence of a saccadic gaze shift, accompanied by a small head movement, and a slight VOR-related rollback of the eye at the end of the movement. Moreover, this pattern is changed in the WGS condition (D), where the eye rolls back further toward its initial position (at the aperture), much like the behavioral WG data (B). The effect on head movement is less clear; in this example (D) the head also seems to move a bit more, but as we shall see, this result was not consistent.
Figure 4 extends these observations by plotting the 2-D trajectories of stimulation-evoked gaze shifts from one SC site in the NGS (left column) and WGS (right column) conditions, using the same conditions shown previously in Fig. 3. The empty circles () correspond to the end of the stimulus-evoked movement. On casual inspection, there does not appear to be much difference between the gaze trajectories evoked in the NGS and WGS (A) conditions. More important, unlike the behavioral data, there does not appear to be much difference between the corresponding head movement trajectories in these 2 stimulation conditions (B).
|
Amplitude and direction of stimulus-evoked gaze, head, and eye movements
Previous studies have shown that the amplitude of gaze, eye, and head movements evoked from SC stimulation depend on initial eye and head position (Freedman et al. 1996; Klier et al. 2001; Roucoux et al. 1980). Typical position dependencies can be observed in Fig. 4, but these tend to be much greater for large-amplitude stimulus-evoked movements. We could not independently manipulate eye and head position in this experiment without disrupting our 2 tasks (natural gaze shifts vs. goggles paradigm). Moreover, these tasks could not guarantee a homogeneous and comparable distribution of initial eye and head positions (training monkeys to look at LEDs with goggles becomes unwieldy for more than a few targets). These factors rule out comparisons of movement amplitude based on raw or averaged data. Therefore we calculated the "characteristic gaze vector" (CV) for each site. This CV represents the theoretical movement that would be evoked by stimulation of that site while the eyes and head are looking straight ahead, as calculated by a multiple linear regression on all of the gaze trajectories and initial gaze positions from each particular site (Klier et al. 2001; Martinez-Trujillo et al. 2003). The CV thus accounts for the near-linear dependency of evoked gaze, eye, and head movement components on their initial-position components (Freedman et al. 1996; Klier et al. 2001; Paré et al. 1994; Roucoux et al. 1980), correcting for nonuniform distributions of initial position.
GAZE PREDICTION.
Our paradigm does not provide any specific predictions about the effect of the WG condition on the overall gaze shift. However, most "gaze models" of the SC would predict no change (i.e., the goggles would not affect the gaze shifts; Galiana and Guitton 1992; Guitton 1992; Sparks 1999). Characteristic gaze vectors (CVgaze) are plotted in Fig. 5 for NGS (A) and WGS (B), for both M1 () and M2 (
). Typical of SC stimulation (Moschovakis 1996; Stanford et al. 1996), the control NGS gaze shifts (A) were mainly horizontal (contralateral to the stimulation site) with various smaller vertical components, and the same was true in the WGS condition (B).
|
HEAD PREDICTION. In the behavioral data, the new eye–head coordination pattern in the WG condition included a clear and significant increase in the size of the head movement for a given gaze shift (Figs. 2 and 3). For example, when we pooled the head movements for all of the behavioral data collected from both monkeys, average head amplitude increased from 19.0° (M1) and 14.6° (M2) in the NG condition to 32.4° (M1) and 19.3° (M2) in the WG condition. This increase was statistically significant in both monkeys (Student's t-test, P < 0.001).
If this increase was implemented downstream from the SC, one would expect it to occur in our stimulation data as well. However, this was not the case (see also Fig. 4). Figure 5D shows the characteristic head movement vectors (CVhead) computed for each SC stimulation site for the NGS condition (as described above for gaze). Notice the wide distribution of directions and amplitudes. When these were computed for the WGS condition, we observed no increase between the WGS (E) and the NGS (D) conditions. Indeed, our quantitative analysis revealed a very small but significant decrease in the size of the head movements in the WGS condition. The vectors have a significantly smaller magnitude in WGS (CVhead = 15.9 ± 2.5° for M1, CVhead = 14.1 ± 1.9° for M2) than in NGS (CVhead = 19.9 ± 3.4° for M1, CVhead = 16.5 ± 2.5° for M2) (Student's t-test, P = 0.008 for M1 and P = 0.009 for M2). Again, these results are summarized in Fig. 5F, which shows a reduced slope in the CVhead amplitudes for WGS versus NGS in both animals (0.683 for M1 and 0.727 for M2 with correlation factors of 0.886 for M1 and 0.892 for M2). Thus the stimulation-evoked head movements failed to show the increased contribution to the gaze shift observed in the behavioral WG condition.
SACCADE PREDICTION. Based on the behavioral data, it was not clear what to predict for the effects of the WGS condition on saccade amplitude. The goggles paradigm requires that a coordinated saccade–VOR movement place the eye on the aperture location. There are several ways this could happen. Although Figs. 2 and 3 would suggest otherwise, one way this could occur is that there could be a dramatic reduction in saccade gain (i.e., no saccade and no VOR). Therefore for the sake of completeness we also analyzed the CV for the eye-in-head movement up to the end of the saccade part of the gaze shift (or to the end of the first saccade when stimulation produced multiple steps movements), in the same way as above.
Figure 5, G and H show that the CVsaccade vectors were smaller than the corresponding overall head movements (D, E) and less compressed in direction along the horizontal axis. Again, there was no obvious difference between the CVs for the NGS and WGS conditions on casual inspection. Statistical analysis showed a slight decrease in the CVsaccade amplitude, from 9.3 ± 1.0° for M1 and 6.0 ± 0.7° for M2 in NGS to 4.9 ± 1.0° for M1 and 6.5 ± 0.5° for M2 in WGS. This is summarized in Fig. 5I, which shows that the relationship between saccade amplitudes in the WGS to NGS conditions remained quite linear but with a small offset or reduction between the 2 tasks (with slopes of 0.86 for M1 and 0.75 for M2 and correlations of 0.75 and 0.75, respectively). Clearly, this reduction in saccade size would contribute little to directing the eye toward the aperture. Again, however, placing the eye on the aperture location depends on the coordination of both the saccade and VOR components, which will be addressed in a subsequent section.
Latencies of the stimulation-evoked movements
One explanation for the slight reduction of the WGS head and gaze movement amplitudes could be that the goggle paradigm produces greater activation of a head (and perhaps eye) fixation system, making it harder to evoke a movement (Tehovnik et al. 1999). If so, then one would expect the latency of stimulus-evoked head and gaze movements to increase with the goggles on. However, this did not occur. The histograms in Fig. 6 illustrate that the population of stimulation-evoked gaze movements (here pooled from both monkeys) had similar latency profiles in the 2 conditions (A and C). The graphs are supported by statistical analysis showing that gaze movement latencies (means ± SE) without goggles (42.9 ± 4.7 ms for M1 and 38.0 ± 1.6 ms for M2) were comparable with the gaze movement latencies with goggles (42.7 ± 4.6 ms for M1 and 37.9 ± 1.8 ms for M2) with P values of 0.94 for M1 and 0.92 for M2. Furthermore, the corresponding stimulation-evoked head movement latencies were comparable in the 2 conditions (Fig. 6, B and D). Analysis of the corresponding head movement latencies for each animal confirmed that the latencies without goggles (84.7 ± 7.6 ms for M1 and 55.8 ± 2.3 ms for M2) were not significantly different (P = 0.36 for M1 and P = 0.67 for M2) than the head movement latencies in the goggle condition (with an average of 92.2 ± 7.5 ms for M1 and average of 54.7 ± 2.0 ms for M2). Thus the reduction in gaze and head amplitudes in the WGS condition was not the result of increased latencies of response to the stimulus.
|
The preceding analysis implicitly assumes that the same SC sites are activated physiologically during the NG and WG tasks. However, it is conceivable that the topographic profile of SC activity changed between these conditions so that one cannot make within-site comparisons. This conjecture, although seemingly unlikely, is consistent with the observed change in gaze-shift amplitude between the NGS and WGS conditions (Fig. 5). To control for this, one has to compare eye–head contributions for NG and WG as a function of the amplitude of the gaze shift, for both the behavioral and stimulation data. To do this, we sorted gaze-displacement amplitudes into bins of 10°. Then, within each of these bins we calculated the average gaze, eye, and head amplitudes. Finally, we plotted the average eye and head amplitudes as a function of the average gaze amplitudes (Martinez-Trujillo et al. 2003). Included in this analysis was the entire set of stimulation trajectories for each animal and behavioral data, matched for the size and distribution of gaze shifts. The left column of Fig. 7 shows these plots for the behavioral data, comparing the NG data set (, 
) to the WG data set (, 
) in M1 (A) and M2 (C). Corresponding data for the NGS and WGS conditions are shown in B and D.
|
) makes about the same contribution as the head () up until gaze shift exceeded 25° in M1 and 15° in M2. After that, the relative eye contribution begins to plateau. The main effect of the trained WG condition is to increase the gain of the head contribution () and decrease the saccadic eye contribution () so that it plateaus much earlier. Statistical analysis (ANCOVA) showed significant differences between the behavioral NG and WG with P < 0.001, in both monkeys. If these context-dependent patterns were implemented downstream from the SC, one would expect them to persist during SC stimulation. However, there was no statistical difference between the NGS and WGS curves in panels B and D. During stimulation, the 2 curves appear to be essentially identical (ANCOVA, P > 0.5). (Interestingly, they appear to fall between the NG and WG curves for the behavioral data.) Thus according to both analyses—site-by-site (Fig. 5) and gaze-by-gaze (Fig. 7)—there was no preservation of the behavioral context dependency in amplitude contribution of saccades and head movements during stimulation-evoked gaze shifts.
Position-dependent patterns of gaze and head convergence
As noted earlier, a number of studies have noted that gaze shifts evoked from the superior colliculus converge as a function of initial position (Freedman et al. 1996; Gandhi and Keller 1999; Moschovakis et al. 1998; Pelisson et al. 1989). Our preceding analysis accounted for this, but did not look directly at the position dependencies. Such dependencies have recently been quantified to probe reference frame coding in the SC (Klier et al. 2001), and to show the differences in movements evoked with the head fixed and head free (Martinez-Trujillo et al. 2003). Here we used similar methods to ask whether the pattern was influenced by the goggles paradigm. In particular, we were interested to see whether there was any systematic change in reference frame coding by looking at position dependencies in the evoked movements.
As described elsewhere (Martinez-Trujillo et al. 2003) for each stimulation site we calculated a convergence index (CI), which was the slope of the stimulus-evoked displacement as a function of initial position. This was done separately for components orthogonal to the CV (off-axis) as a measure of direction dependency and for components parallel to the CV (on-axis) as a measure of amplitude dependency on position. For reference, a CIgaze of 0 denotes fixed-vector coding and –1.0 denotes a goal-in-space code, whereas a retinal code results in off-axis values ranging from 0 to –1.0 as a function of gaze amplitude. This analysis was done for both the gaze shifts and the head movements for comparison between the NGS and WGS conditions. It was not done for the eye-in-head saccades because, in the goggles paradigm, the range of initial eye-in-head positions was not wide enough to calculate a reliable slope.
Having made these calculations, we then compared the CIgaze for NGS and WGS using the various plotting schemes described in Klier et al. (2001) and Martinez-Trujillo et al. (2003). As these authors reported, the off-axis CIgaze increased as a function of CVgaze magnitude in our NGS data, in a manner consistent with an eye-centered coding scheme. We found no systematic change in this pattern (or any other measures of position-dependent convergence in gaze and head movements) in our WGS data. In both NGS and WGS, SC sites coding small to medium gaze shifts appeared more "fixed vector" (Fig. 4A), whereas sites encoding larger gaze shifts were much more convergent, as reported previously (e.g., Klier et al. 2001).
Because these results were negative and not directly related to our current hypotheses, they are only briefly summarized in Fig. 8, which plots CIgaze(WGS) as a function of CIgaze(NGS) (top row) and CIhead(WGS) as a function of CIhead(NGS) (bottom row), along both the "on axis" (left column) and "off axis" (right column). As shown in the figure and verified statistically (see figure legend) a few individual stimulation sites showed a substantial change across conditions, but there were no consistently significant trends across sites or between animals (as we saw in our various other "reference frame plots"). Thus we found no evidence that the default position dependency or eye-centered gaze frame of the SC motor code was altered in the WGS condition.
|
Our behavioral tasks were not designed to directly manipulate the relationships between movement amplitude and velocity, but we reasoned that monkeys might make faster head movements with the goggles to get the visual aperture onto the target more quickly. In the behavioral data set, we observed an increase in head velocity during head movements from an average of 117.1°/s in NG to 177.2°/s in WG (Student's t-test, P < 0.001) for M1 and an average of 127.5°/s in NG to 195.4°/s in WG (ANOVA, P < 0.001) for M2. This alone was not surprising, given that head velocity varies with head-movement amplitude (Freedman and Sparks 1997; Zangemeister et al. 1981), which was larger in the WG condition. However, even though average head displacements were slightly smaller in the WGS condition, average head velocity (v) was still significantly higher (ANOVA, P = 0.002 for both monkeys) (v = 132.0°/s for M1 and v = 118.8°/s for M2) compared with the control NGS condition (v = 111.5°/s for M1 and v = 110.9°/s for M2).
Having made this initial observation, we set out to compare eye, head, and gaze velocities between the 2 paradigms on a more even footing (i.e., by plotting them as a function of amplitude). Such plots normally show a characteristic relationship between velocity and amplitude known as the main sequence (Bahill et al. 1975; Freedman and Sparks 1997; Gandhi and Sparks 2001; Zangemeister et al. 1981). Main sequences for the gaze, eye-in-head saccade, and head behavioral data are shown in Fig. 9 (rows 1 and 3). These data have been grouped into bins to simplify the graphic plot, but statistics were done on the raw data. Computing an ANCOVA test for the amplitude and velocity in each condition, we found for the head (right column) a slight but significant increase of velocity as a function of amplitude in the WG head data (open circles) than the NG data (filled circles) with P < 0.001. In contrast, the eye data (middle column) showed a significant decrease in velocity as a function of amplitude (ANCOVA, P < 0.001) in the WG condition, whereas the resulting gaze data (left column) showed a slight and significant decrease (ANCOVA, P < 0.01) in WG data. These observations were true for the behavioral data in both M1 (1st row) and M2 (3rd row).
|
). Next we focus on preserved effects that are more central to the adaptations required in the goggles paradigm. Preservation of the adapted eye-in-head range under stimulation
Thus far, our quantitative data (Figs. 5–9) have largely shown the negative result inconsistent with a downstream implementation of adaptive eye–head coordination. However, we now turn our attention to the most basic aspect of the goggles task: placement of eye position at the aperture at the end of the gaze shift. Figures 2 and 3 suggest that the WG condition caused the saccades and VOR to coordinate in such a way that eye positions end up near the goggles aperture, and that this part of the behavior may be preserved during SC stimulation (Fig. 4).
To illustrate this more directly, Fig. 10 plots post-VOR eye position ranges without (A–C) and with (D–F) the goggles. These plots show data recorded between head-free gaze shifts (i.e., during fixations) when both gaze and head velocity dropped below 10°/s. Each square represents the tip of a 2-D eye "pointing vector," plotted on a vertical versus horizontal coordinate system. The left column shows a typical example of the behavioral result. Whereas the natural NG range was relatively distributed, particularly in the vertical dimension (A), the post-VOR range for the WG condition was reduced to the size of the central 10° aperture (D). The corresponding frequency-bin histogram for vertical eye position (Fig. 10G) shows that the eye landed more often on the location of the aperture (horizontal black bar) when the animal was wearing goggles.
|
To extend these observations to 2-D and document them across all of our stimulation sites, we used a standard Matlab (The MathWorks, Natick, MA) function to fit ellipses to the horizontal and vertical components of the final eye-in-head positions of the NGS data (Fig. 11A) and WGS data (Fig. 11B). For this analysis, all eye positions were normalized with respect to the goggles aperture by using a reference eye position collected while the eye was looking straight ahead through the aperture (see METHODS). Thus a position at 0, 0 in these coordinates would place the eye right through the center of the aperture. Rows 2 and 3 of Fig. 11 show such elliptical fits for all stimulation sites in animals M1 (2nd row) and M2 (3rd row). Comparing the NGS data (left column) to the WGS data (right column) 2 trends should be evident: the WGS ellipses tend to be smaller and better aligned with the origin (the aperture center).
|
; M2, ). These averaged eye endpoints covered a wide range in the NGS condition (A). In contrast, the centers of the WGS ranges (B) were more narrowly distributed and largely confined within the range of the aperture. This effect was statistically significant (Student's t-test, P < 0.05) for both the vertical and horizontal components.
|
Position-dependent aspects of eye–head coordination
With the head free, final eye-in-head position at the end of a gaze shift is determined by a combination of saccade and VOR trajectories (Fuller 1996). In the NGS stimulation condition, this did not result in a recentering of the eye, apparently because the eye and the head had variable starting positions and the saccades and head movements had different amplitudes and directions (Fig. 4). Could it be that the recentering of the eye in the WGS condition was the result of the eye always starting at the aperture location, such that the subsequent recentering was simply a by-product of the saccade and head movement (perhaps) rotating about the same axis, so that the VOR negated the saccade? Or was the eye aimed toward the aperture by a more sophisticated mechanism? As noted above, the initial set of eye-in-head positions in the goggles paradigm was too restrictive to test the position-dependent aspects of this behavior. Therefore to test between these 2 possibilities, we used the "switching holes" paradigm in animal M2.
It was previously shown (Crawford and Guitton 1997) that, once a monkey is trained on one aperture location, it will initially continue to drive the eye toward the old aperture site when confronted with a new aperture location. To find out what happens when the eye started from a different location here, we covered the central aperture and uncovered top, bottom, right, and left apertures, one by one. When the animal was fixating through the new apertures we applied the electrical stimulus to see whether the evoked moments would show the same effect; that is, would the saccade + VOR bring the eye back toward the location of the initial, now covered, aperture? Or, would the saccade + VOR bring the eye back to the new, uncovered, aperture? We were able to do this with 7 stimulation sites before the animal showed signs of "unlearning" the original task, that is, reducing its accuracy for aiming the eye with the original aperture.
Figure 13A illustrates examples results of this experiment for eye movements evoked by stimulating one of the SC sites with the eye initialized at 3 different aperture locations: right (1), left (2), and down (3). Again, the reference position has been adjusted here so that the location of the original aperture is centered at zero in this coordinate system. Independent of initial eye position, the final eye positions were driven toward the central (now occluded) aperture for which the monkey had been trained. In these examples, the eye obtained the original aperture through a series of small multistep saccades (and intervening VOR segments) that drove the eye well away from the direction of head movement. Once again, because the central aperture was now occluded from vision, this coordinated pattern of saccades and VOR slow phases can be attributed to the perseverance of some context-dependent, learned response rather than just a visual stimulus. Further, this context-dependent adaptation would be situated downstream from the stimulation location (i.e., downstream from the SC).
|
The top panels of Fig. 13B show the results from one site, whereas the bottom panels show the regression fits to all 7 sites. On average, slopes were –0.57 ± 0.09 for the vertical and –0.48 ± 0.1 for the horizontal component, with average correlations of 0.45 and 0.25, respectively. This population of 7 sites is perhaps not large enough to prove that initial eye-position dependencies are significantly changed by the goggles paradigm. However, this does cast doubt on the idea that the clustering of final eye positions near the aperture in our larger WGS population (Figs. 4, 8–11) was attributed only to an initial eye position effect—clearly the system was not limited to bringing eye position back to where it started.
Head movements were also influenced by the initial eye position in the switching-aperture paradigm. Moreover, this influence was correlated to the eye movement in a way that is consistent with gaze models of the superior colliculus (Freedman and Sparks 1997, 2000; Guitton et al. 1990). For example, although the gaze shifts evoked from the SC shown in Fig. 13A were all leftward, the overall eye movement and the head movement could be in the same direction (trajectory 1) or—most dramatically—in the opposite direction (trajectory 2). In the latter movement, the overall eye movement was rightward and the overall head movement was leftward. Why would this happen? Viewing Fig. 13A, one can see that when the overall eye movement reversed from leftward (eye trajectory 1) to rightward (eye trajectory 2), the leftward head movement (head trajectories 1 and 2) increased to compensate. Likewise, when the overall eye movement took on an upward component (eye trajectory 3), the accompanying head movement took on a downward head component (head trajectory 3), with the result that the gaze kinematics (gaze trajectories 1, 2, and 3) remained consistent in size and direction.
Figure 13C quantifies this inverse eye–head relationship by plotting overall eye displacement as a function of head displacement for individual stimulation trials. Vertical displacements are plotted in the first column and horizontal displacements in the second column, pooled across all of the "switching holes" experiments for that site. Clearly, these eye and head data show a strong negative correlation. Regression fits for all 7 stimulation sites tested this way are shown in the bottom panels of Fig. 13C, confirming this trend. Average slopes1were –0.76 ± 0.12 for vertical and –1.04 ± 0.13 for horizontal displacements, with average correlations of 0.26 and 0.41, respectively. Thus during the goggles task in this animal, SC stimulation elicited eye (saccade + VOR)–head movements that compensated for initial position and were specifically coordinated to land the eye on the learned aperture while maintaining gaze kinematics.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTION |
|---|
