| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1Neuroscience Program and 2Department of Psychology, University of Illinois, Champaign, Illinois 61820
Submitted 26 November 2002; accepted in final form 18 February 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXACKNOWLEDGMENTSREFERENCES |
|---|
). The main interaction is through inputs from superficial layers of the ipsilateral SC and projections back to superficial layers of both ipsilateral and contralateral SC (Baizer et al. 1991; Graybiel 1978; Sherk 1979b). However, cells of deeper SC layers may also be involved, providing some of the input to the PBN (Baleydier and Magnin 1979) and receiving PBN terminals on their dendrites (Roldán et al. 1983). The return projection to the SC is cholinergic (Mufson et al. 1986), and in the rat, it activates inhibitory interneurons (Binns and Salt 2000; Lee et al. 2001) that inhibit tectal-thalamic cells (Lee et al. 2001). In anesthetized cats, PBN cells respond briskly to visual stimuli and are organized into a retinotopic map mirroring the SC: most of the PBN maps the contralateral hemifield; a smaller anterior segment maps the ipsilateral hemifield (Sherk 1979a). Monkey SC cells projecting to the PBN have been examined in the anesthetized animal and some show strong direction selectivity (Marrocco et al. 1981). The intimate association of the PBN with the SC suggests it plays a role in detecting saccade targets and/or facilitating eye movements to them, but we are unaware of published descriptions of PBN activity in any awake, behaving species. The purpose of this study was to investigate PBN contributions to visuomotor behavior by recording responses of PBN cells in awake cats trained in oculomotor tasks.
Given that the SC is primarily associated with saccades, we initially concentrated on paradigms requiring saccades to stationary targets. However, a tracking task produced the greatest insights by far, and we will devote most of this report to PBN activity during pursuit. Cats tend to track moving objects primarily with small catch-up saccades supplemented by low-gain smooth pursuit. As described in the following text, these eye movements were sufficiently accurate to keep the target within the area centralis most of the time.1 Because "pursuit" has often been used interchangeably with "smooth pursuit," we refer to such close following of the target, however accomplished, as "close-order tracking." We found that PBN cells continuously fire at rates that are very sensitive to small variations in the angular distance between an attended target and the center of the area centralis, particularly for the small distances that characterize close-order tracking. The terms "gaze position error" (Bergeron and Guitton 2002) and "position error" (Krauzlis et al. 2000) have been used in conjunction with similar phenomena in the SC. We will use "retinal position error" (RPE) to denote the distance between eye and target. This term defines a purely geometric distance on the retina and should not be taken to indicate that PBN activity is used directly as a motor signal to correct this "error." Our results indicate that PBN cells encode the distance to a potential fixation target at any point in an intersaccade interval but do not trigger saccades or determine their metrics.2
|
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXACKNOWLEDGMENTSREFERENCES |
|---|
Under general anesthesia, the skull was exposed and several titanium studs threaded into the bone. An aluminum fixture was implanted around the calvarium by bonding it to the skull and the titanium studs with an orthopedic acrylic. Miniature microdrive bases with integral guide tubes were implanted through the acrylic and skull over the PBN. The guide tube was held rigidly in the base by a low-melting point alloy. The alloy could be melted without discomfort to the animal, and the guide tube swiveled up to 5° in any direction from the initial aiming point—a sufficient range to locate both the PBN and most of the SC. To redirect the guide tube, a narrow 10-cm-long pointed rod was threaded onto the top of the guide tube, coaxial with the guide tube. The point of the rod was lightly pressed into a small hole in a plunger attached to a stereotaxic carrier. The plunger was spring-loaded against the rod, which kept the two in reliable contact as the former was translated in a horizontal plane, swiveling the guide tube to a new location. Because the distance from the microdrive base to the point of the rod was approximately four times the distance from the base to the PBN, a 100-µm translation of the top of the rod resulted in approximately a 25-µm shift of the aiming point in the PBN.
During recording sessions, miniature microelectrode drives with preloaded microelectrodes were heat-sterilized and mounted on the guide tube. When micropipettes were used, they were sterilized in the same manner as the microelectrodes and then filled with sterile normal saline by pressure from the rear. The microelectrode or micropipette was advanced into the brain remotely via a stepping-motor driven shaft. Four threaded studs protruding from the aluminum skull fixture were used to hold a cap that protected the microdrive bases between training and recording sessions, and to immobilize the head during sessions. This recording system is identical to one previously described (Malpeli et al. 1992) except that the microdrive was modified for longer travel.
During training and data-collection sessions, the cats lay in a loosely-fitted cloth bag with their heads fixed to a thick, rigid, acrylic plate, tilted forward 5° relative to the Horsley-Clark horizontal plane. They faced a rear-projection screen subtending 60° horizontally and 50° vertically, positioned 70 cm from their eyes. Laser spots approximately 0.1° in diameter, positioned on the screen by computer-controlled mirror galvanometers, served as fixation targets. Rectangles produced on a computer-driven LCD projector were superimposed on the same screen and used for mapping receptive fields. These could be manually directed or presented in an automated fashion under computer control. They were never behaviorally relevant. The cats were trained extensively in their presence, and they soon appeared to ignore them.
The screen was illuminated with a uniform background of 0.021 cd/m2. When the LCD projector and lasers were turned off the room was completely dark. The animals were monitored at all times via an infrared-sensitive camera and infrared light sources.
Eye movements were monitored with the double magnetic induction method (Allik et al. 1981; Bour et al. 1984; Reulen and Bakker 1982) as implemented by Malpeli (1998), sampled every 2 ms, and smoothed with a Gaussian filter of fixed SD (SD = 4 ms). The circuit described by Remmel (1984) was used to drive the field coil and measure the output of the detection coil. The output of the system was linearized over the range of target positions by the method of Kang and Malpeli (2003). Final calibration of eye position signals was done post hoc, making the assumptions that, on average, the end points of saccades for tracking in both directions along a given target trajectory were symmetrically placed to either side of the target trajectory and that the mean position of the eye during a number of fixations of a central, stationary target denoted both the target position and the center of the area centralis. Saccade start was taken as the point where eye velocity exceeded 25°/s; saccade end as the point where velocity dropped less than 10°/s. Eye movements whose speed never reached 25°/s were not counted as saccades. The cats were trained to fixate laser targets by rewarding them with soft cat food delivered through a metal tube when they fixated the target within a certain tolerance window whose dimensions were reduced as training progressed. The food was delivered through a tube that moved away from the animal during trials to avoid licking artifacts. Nevertheless they often licked at inopportune moments, and these segments of data, which were easily recognized by a characteristic oscillation of the eye movement record, were discarded.
Three behavioral paradigms were employed for these experiments. For saccades to stationary targets, a brief warning tone sounded, followed in 500 ms by the appearance of a central fixation point (a laser spot). The cat was given 2,000 ms to fixate the target within a tolerance window of ±2 to ±3°. After a delay of 1,000–2,000 ms after acquisition of the central fixation point, it was turned off and the saccade target came on. The cat was given 1,500 ms to fixate the peripheral target with a single saccade terminating in a tolerance window whose dimensions varied with target eccentricity (typically, ±3 to ±8°). Cats can readily perform with much shorter intervals allowed for initiating saccades, but we gave them a long time to do so to encourage variations in saccade latency. Such variations in saccade latency are useful for distinguishing stimulus-evoked responses from motor-related responses. Nevertheless, they usually made saccades promptly, with a mean latency of 199.6 ± 85.4 (SD) ms. If the eye stayed in the window around the peripheral target for 1,000–2,000 ms, the trial terminated and a reward was delivered. If the animal failed to meet these criteria at any point, the trial was terminated without reward. Intertrial intervals varied from 5,000 to 8,000 ms. The number of potential target locations in a series of trials varied from 2 to 6 with locations determined pseudorandomly. At least one target location was always devoted to the anti-preferred hemifield, although we often biased the number of targets toward the preferred hemifield to maximize the number of trials associated with evoked responses.
In the close-order tracking paradigm, trials began with a warning tone, followed after 500 ms by a target that appeared peripherally and moved along meridians (chosen as described in RESULTS) at constant velocity through the center of the display to an eccentric location. Sometimes, to increase the data obtained for large RPE, the target was kept stationary for 500–1,500 ms before beginning to move. Large ranges of RPE were also encouraged by using a large tolerance window for fixation (typically ±7.5°) and by giving the animal a long time to acquire the target (2,000 ms). If the target was acquired and successfully tracked along its full trajectory of 20–30°, a reward was delivered at trial end. Target speeds of 2–10°/s were used, although data were collected most often at 3 and 6°/s.
For saccades in the dark, cats were randomly rewarded without behavioral contingencies. Compact-disk recordings of a variety of animal sounds were sometimes played from one of several speakers arrayed around the cat to encourage saccades. Thus it could be said that many of the saccades were not truly spontaneous because they may have been directed to auditory targets. However, we detected no saccade-related responses in the dark, so this is a moot point.
Once cats were adequately trained, we began searching for the PBN. Recordings were made via glass micropipettes or tungsten-platinum quartz-insulated microelectrodes (Reitboeck 1983) mounted in large-bore glass micropipettes that were ground to a smooth conical taper at the interface between pipette and electrode. The metal microelectrodes could be reused across several sessions without removing them from the microdrive, resulting in more repeatable penetration trajectories across sessions.
The PBN is too small and awkwardly placed (on the lateral edge of the midbrain) to be found by relying on stereotaxic coordinates alone. Following a search procedure developed by Sherk (1979a), we first recorded from the center of the SC, and guided by SC retinotopy, repositioned the electrode until we found the spot on the lateral edge of the SC representing a point in visual space 5° into the contralateral hemifield and 50° below the horizontal meridian. The PBN is approximately 1 mm lateral to this point, 5–7 mm further ventral, tucked under the brachium of the inferior colliculus (Fig. 1). After locating the desired point in the SC, the PBN was found with series of passes made along a medial-lateral plane (approximately 100 µm apart at the depth of the PBN). To avoid excessive damage to the midbrain by such closely spaced passes, these series were begun lateral to the PBN and continued lateral to medial.
PBN responses were quite distinctive and could easily be distinguished from those of nearby structures. Responses in the superficial layers of the SC and the PBN differed dramatically in that visually evoked activity in the PBN was selective for attended saccade targets, whereas responses in the SC were not. We readily mapped receptive fields in the SC either with manually directed stimuli or with retinally stabilized, randomly interleaved, horizontal and vertical, light or dark bars, flashed on and off. These produced vigorous responses in the SC just as they do in the anesthetized animal. In contrast, the same stimuli usually evoked vague and irregular responses from PBN cells. When used with hand mapping, the same stimuli sometimes influenced activity sufficiently to roughly localize responses to a particular quadrant of visual space. Automated mapping with behaviorally irrelevant stimuli rarely identified discrete response fields, but it is unclear if this is because the mapping stimuli were unattended or if it was because they were inappropriate visual stimuli for PBN cells. In sharp contrast, laser spots used for saccade targets always evoked good PBN responses while the cats tracked moving targets or when stationary targets were presented in the appropriate hemifield, whether or not they elicited saccades. On occasion, a fiber with PBN characteristics was recorded dorsal to the PBN, but the point of entry into the nucleus itself was always obvious because of the sudden appearance of multiple-unit and background activity briskly driven by the laser fixation target. We histologically confirmed electrode locations in both PBN of one cat (Fig. 1); the other two animals are still alive.
We made no attempt to sample systematically the entire PBN. Although the guide tube could be redirected with great precision, variations in the tightness of fit of the electrode shaft in the guide tube and slight curvatures of the tapered part of the electrode resulted in random deviations in trajectory that were a sizable fraction of the dimensions of the PBN. After having located the PBN, replacing a microelectrode with a new one often resulted in the PBN being missed entirely, necessitating another series of search passes. Consequently we tried as much as possible to aim for thickest part of the PBN and probably did not record from the most posterior portion, which contains eccentric receptive fields in the anesthetized animal (Sherk 1979a). The coverage of electrode passes could not be determined for the two PBN for which histology is currently available because these experiments extended over many months and only recent electrode tracks were visible.
Data were obtained from 36 isolated cells and 71 multiple-unit recordings, although most cells were tested thoroughly with only one paradigm. Multiple-unit responses were virtually identical to single-cell responses recorded at the same PBN location. Only data from isolated cells are presented in this report. Spike data were smoothed with Gaussian filters, with the exception of those data in Fig. 9 and 11. For those analyses for which activity was examined as a function of time, Gaussian filters of fixed SD were used, and in each case, the SD is noted in the legend. For analyses in which activity was examined as a function of RPE, the density of data varied greatly with RPE because tracking was good enough to keep the target within the area centralis most of the time. Consequently, the SD of the Gaussian was varied inversely with data density (i.e., adaptive Gaussian filtering). How activity as a function of RPE was extracted from the raw data, and how adaptive Gaussian filtering was used to smooth these data are explained in the APPENDIX.
|
|
|
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXACKNOWLEDGMENTSREFERENCES |
|---|
Stationary saccade targets were always presented outside the area centralis, typically at eccentricities between 4 and 16°, because the cats did not reliably shift their gaze to saccade targets appearing within a couple of degrees of their current fixation point. Brisk responses were evoked by the appearance of stationary saccade targets in a large region of the appropriate hemifield, with latencies averaging 69.2 ± 8.7 ms (range = 59.0–87.5 ms, n = 20). A cell with response latency near the average latency is illustrated in Fig. 2. Usually, there was an initial high burst of activity (Fig. 2A) that declined within 100 ms of target onset to a plateau that was maintained until the saccade began (Fig. 2B). This response began to drop at about peak saccade velocity (Fig. 2C) and terminated around saccade end (Fig. 2D).
|
|
Given that stationary saccade targets evoked good responses, these are probably the best stimuli for defining the size and location of excitatory regions of PBN response fields. We did not routinely attempt this because most cells would have been lost before a systematic mapping of the response field could have been completed. However, most cells responded briskly to targets placed anywhere in the preferred hemifield, out to the limits of these cat's effective oculomotor range—i.e., the most eccentric saccade target for which they performed the task reliably. This was about 15°, depending on direction and the individual cat, so it is clear that response fields usually extended further than this from the area centralis (the 3 cells of Fig. 10B were exceptions).
Activity during close-order tracking of moving targets
The cats' oculomotor control proved adequate to keep moving targets well centered on the retina. For all tracking data accumulated while recording PBN cells in this paradigm, the target was kept within 2.5° of the center of the area centralis 83.7 and 68.4% of the time spent tracking targets moving at 3 and 6°/s, respectively.3 Often, saccades overshot the target. Then the eye remained stationary as the target moved across the area centralis, and another corrective saccade was launched to repeat the process (Fig. 3, top). Such behavior keeps the target within the area centralis with a minimum number of catch-up saccades. Although most catch-up saccades overshot the target, a substantial fraction undershot (Fig. 3A, latter half of top). These cycles of saccade and intersaccade intervals during close-order tracking produced saw-tooth patterns of RPE (Fig. 3, A and B, middle).
|
|
). Ipsilateral and contralateral components were preferred in the anterior and posterior parts of the PBN, respectively (e.g., the left posterior PBN was active when the eye lagged a target moving left to right). Cells at the border between these two regions preferred vertical components. Cells encountered first in a microelectrode penetration typically preferred upward RPE, and the best direction of RPE usually rotated through horizontal to downward as the penetration progressed. We did not quantitatively determine tightness of tuning for RPE direction, but tuning had to be fairly tight because PBN cells responded poorly, if at all, during tracking of targets moving at 90° to the best estimated RPE. For all data presented here, RPE refers to the component of target position on the retina parallel to the direction of best target motion, so for example, if fixation was directly to the right or left of a cell that preferred upward RPE, the component of RPE used in our analyses would have zero magnitude. RPE in the preferred direction is considered positive and that in the anti-preferred direction negative.
Several common characteristics of our sample of PBN cells are illustrated by the cell of Fig. 3, which was unusual only in the large amount of pursuit data obtained for it. It will be referred to as the "exemplar cell" and used to illustrate many of the points made for PBN activity during close-order tracking of moving targets.4 Its activity was best modulated during tracking of a target moving vertically. For upward target movement, activity ramped higher when the eye lagged the target during intersaccade intervals, and rapidly plummeted during saccades toward the target (Fig. 3A). The first three saccades overshot the target, reversing the direction of RPE and terminating activity until the target moved across the center of the area centralis and the eye once again lagged the target. Thus both direction and magnitude of RPE determine firing rate. Subsequent saccades fell short of the target, and although activity dropped with each saccade, it did not generally reach zero. Occasionally (but not for this cell), a saccade was made in the wrong direction, which increased RPE. This resulted in a sudden increase rather than a decrease in activity (Fig. 4). Thus saccades do not simply inhibit PBN activity, and sudden drops in activity were not related to saccade start per se but to the reduction of RPE. These points are further illustrated by tracking in the anti-preferred direction for the exemplar cell (i.e., downward; Fig. 3B). Here the temporal relation between activity and cycles of catch-up saccades is reversed: when the eye lagged the target, the cell was silent, and it started firing abruptly shortly after each saccade, all of which overshot the target.
|
The two trials chosen for Fig. 3 are entirely representative. To allow a more direct comparison of unprocessed spike trains with RPE, we present every other trial for 3°/s tracking for this cell at a higher temporal resolution in Fig. 5. Inspection of these data illustrate that when RPE was positive, regardless of the direction of target motion, the cell was active, although there was some variability in the relationship between the level of activity and the precise moment when zero RPE was crossed. Intervals of large positive or negative RPE at the start of each trial occurred because the eye was usually centered on the display screen when the target appeared peripherally. For upward tracking, at target onset, spontaneous activity was quickly driven to zero, and the cell subsequently fired whenever RPE became positive. For downward tracking, activity gradually decreased after target onset, going to minimal levels when the target was first acquired and then reappearing whenever RPE was positive.
|
The largest RPE did not necessarily evoke the highest firing rate for either direction of motion. Although this certainly reflects some random variation in response, to a large extent, it resulted from response saturation at relatively low RPE. The continuous curve in Fig. 6 shows how the activity of this cell varied with RPE during close-order tracking, pooled for both directions of target movement (the methods used to extract and smooth these data are explained in the APPENDIX). For RPE from 0 to approximately 1.5°, the response increased rapidly with RPE at approximately 60 spikes/s/deg, then slowly declined with further increases in RPE (see bold curve in Fig. 10A for the full activity-RPE profile for this cell). It is for this reason that the large positive RPE achieved at the end of the third intersaccade interval of Fig. 3A and at the beginning of the first and second intersaccade intervals of Fig. 3B did not evoke proportionally large responses.
As noted previously with respect to specific trials in Figs. 3 and 4, abrupt changes in activity associated with saccades often quickly settled at a level appropriate to the RPE achieved by the saccade regardless of the direction of target motion or the direction of the saccade. To examine this more thoroughly, we plotted activity in the intervals 0–60 ms after saccade end for all saccades made during close-order tracking (Fig. 6, 
), and compared it to the continuous activity-RPE function for this same cell (Fig. 6, —). Considerable variability in these data are guaranteed, because at typical firing rates, only a few spikes are expected to occur in a 60-ms interval (e.g., 1, 2, or 3 spikes at firing rates of 16.7, 33.3, or 50 spikes/s, respectively). Nevertheless, the average activity in these intervals (Fig. 6, - - -) is fairly close to the continuous activity-RPE function for all epochs of close-order tracking, supporting the conclusion that activity is quickly reset to a level appropriate to RPE at saccade end.
To examine the time course of activity changes occurring during catch-up saccades, for our exemplar cell, we identified those saccades that were made in a direction that reduced RPE, had similar amplitudes, and terminated close to the target. As with saccades to stationary targets, activity began to change at about peak saccade velocity, and rapidly dropped as the saccade terminated (Fig. 7). Figure 7 does not directly illustrate the relationship between activity and RPE during intersaccade intervals because RPE changed sign at a different point in time prior to each saccade: one must synchronize the data on spatial, rather than temporal events (such as Fig. 6, —).
|
Responses averaged across time encode RPE with considerable precision. To obtain a more general picture of the precision of encoding, we examined variations in RPE at points in time during intersaccade intervals when activity went to zero ("activity zero crossings"). Regardless of tracking direction, activity zero crossings are tightly clustered around zero RPE: 81% of zero crossings are within ±0.5° of zero RPE (Fig. 8A). The distribution of these zero crossings is tighter by a factor of 2.5 than the distribution of saccade overshoots and undershoots (Fig. 8B; SD 0.45 vs. 1.11° for neural and behavioral data, respectively), and so the neural activity appears sufficient to account for behavioral saccade accuracy (although we do not mean to imply a causal relationship). Given that eye position cannot be specified perfectly, the actual clustering of RPE at activity zero crossings is probably even tighter.
Although PBN activity is strongly related to RPE, it might also be related to other parameters. However, during close-order tracking, obvious candidates, such as absolute position of eye in orbit, hemifield of target presentation, target speed, relative velocity between eye and target, and direction of target motion (across trials) changed continually in ways that were not fixed in relation to RPE. Thus it is unlikely that any of these other parameters could strongly impact activity without seriously perturbing the relationship between activity and RPE. For example, activity changed radically during each trial, whereas target speed was constant throughout, and eye speed was very low during many intersaccade intervals (Figs. 3 and 5). Also, the same relationship between activity and RPE appears to hold despite the large overall movements of the eye as it tracked the target across the display. We examined these apparent invariances quantitatively for the exemplar cell for the range of RPE over which the activity-RPE function is fairly linear (Fig. 9). The slopes of the regression lines relating activity to RPE in this linear range for two target speeds (3 and 6°/s) are virtually identical, indicating that the activity-RPE relationship is not strongly affected by target speed, at least for relatively low speeds (Fig. 9A; see also Fig. 11B, for which target speeds of 0 and 3°/s are compared). We also examined the effect of absolute target position on the activity-RPE function, and found none (Fig. 9B). Because the eye tracked the target closely over most of the total target trajectory (30°), this means that absolute position of eye in orbit is also inconsequential for activity. Thus neither eye position nor target position, per se, determine PBN activity, except as they affect RPE.
Every PBN cell's activity varied with RPE, and the relationship between activity and RPE was usually steepest within the area centralis. We obtained enough data to define this relationship over various ranges of RPE for 20 cells (Fig. 10). The data provided by the close-order tracking paradigm were dominated by small RPE because the cats' excellent saccadic tracking behavior kept fixation close to the target most of the time. Fortunately, a common tactic of the cats yielded some data for large RPE. On target onset, the cats would often make a small saccade toward, but not to the target, and then wait until it moved closer before intercepting it. This provided occasional sweeps of large-amplitude positive or negative RPE at trial start, although at low data density because there was at most only one such sweep per trial.5
Most commonly, PBN cells (10 of 20) increased their firing rates sharply with increasing RPE for central visual fields, particularly within the area centralis (Fig. 10A). The response usually peaked at low RPE (1–5°) and then gradually declined with increasing RPE. The second group consists of three cells whose activity peaked sharply at a particular RPE in the area centralis and declined sharply and fairly symmetrically for RPE above and below this peak (Fig. 10B). The final group consists of seven cells whose activity changed relatively slowly with RPE (Fig. 10C). An additional single cell for which we had insufficient data to produce an activity-RPE function, as well as multiple-unit records obtained at 28 sites in the PBN, also displayed similar modulation during close-order tracking.
Activity during gaze adjustments around stationary targets
The apparent invariance of activity as a function of RPE over a range of target speeds suggests that as target speed is reduced to zero, the same relationship between activity and RPE should be maintained. As it turns out, the initial phase of the paradigm using saccades to stationary targets provides some data relevant to this point. In this paradigm, cats had to first fixate a central, stationary laser target for 1,000–2,000 ms before the peripheral saccade target came on. While viewing the central target, they often adjusted their gaze with small saccades, and during this time, activity mirrored RPE in a way similar to that described for close-order tracking. An illustrative trial for the exemplar cell is presented in Fig. 11A. Before the central fixation target appeared, the cell fired at a low rate. A nonzero spontaneous rate of firing in the absence of any target was typical for this paradigm, as well as in the dark (see Fig. 13). Soon after the target appeared, activity dropped to zero because RPE was negative (i.e., in the anti-preferred direction). We interpret this as active inhibition of the spontaneous rate of firing, similar to the inhibition observed for the appearance of eccentric targets in the anti-preferred hemifield (Fig. 2, E–H) or for negative RPE during close-order tracking (Figs. 3 and 5). About 400 ms later, a small saccade was made to the other side of the target, and the cell began to fire. For this cell, we had enough data from saccade and close-order tracking paradigms to compare directly activity during viewing of a stationary target with activity during close-order tracking: The two were strikingly similar (Fig. 11B). It is premature to conclude that activity for stationary and moving targets is the same for all PBN cells. However, this comparison does add some additional support to the conclusion, derived from moving targets, that neither eye nor target speed are major factors for determining PBN activity. It appears likely that the PBN is involved in all close viewing of targets, whether moving or stationary, although the quantitative relationship between activity and RPE was more apparent with moving targets because of the striking ramps of activity characterizing intersaccade intervals.
|
PBN premotor activity
Here we consider the possibility that PBN activity directly triggers saccades or is an obligatory prerequisite to saccades. Because PBN activity for tracking in the preferred direction increases with RPE, one might posit that corrective saccades are triggered at some particular threshold level of firing. However, the relationship between firing rate and saccade initiation is quite loose. This can be appreciated by inspection of Fig. 3A, for which firing rates at saccade initiation vary by more than a factor of 2. It is demonstrated more directly from the frequency histogram of firing rate at saccade start for all catch-up saccades for this cell in the preferred direction (Fig. 12). Also, for saccades to stationary targets, the initial burst of activity is more sharply synchronized on target onset than on saccade start (compare Fig. 2, A and B).
|
One might reasonably wonder if the magnitude of PBN activity at saccade start is predictive of the size of the ensuing saccade. However, most saccades either undershot or overshot the target (Fig. 8B), so to the extent that PBN activity is related to instantaneous distance to the target, this is a logical impossibility. We checked this by looking at the relationship between average activity in the 60 ms prior to saccade onset and saccade amplitude: there was little, if any (slope = 5.6 spikes/s/deg, r = 0.13 for the exemplar cell—recall that the equivalent slope for activity vs. RPE was 60 spikes/s/deg for this cell).
Finally, we could find no consistent responses that were temporally related to saccades in the dark (Fig. 13). We conclude that PBN activity in intersaccade intervals is derived from sensory input and may be used to locate the position of a potential target but does not directly drive saccades or determine their metrics.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXACKNOWLEDGMENTSREFERENCES |
|---|
The most salient aspects of PBN activity identified in this study are specificity for potential saccade targets, large response fields, and precise encoding by firing rate of small RPE during intersaccade intervals of close-order tracking of visual targets. The PBN is not a motor structure: no bursts of activity precede saccades in the dark; there is no fixed threshold of activity at which saccades are triggered; the level of activity at the start of the saccade does not predict saccade metrics; abrupt changes in activity associated with saccades lag rather than lead saccade start. The PBN responds to target onset with a typical sensory delay of 60–80 ms and obviously extracts RPE-related activity from sensory signals. Its responses are independent of the absolute position of the eye in the orbit. However, PBN activity is probably not simply sensory, for responses evoked by attended fixation targets were always brisk and reliable, whereas the behaviorally irrelevant stimuli intended for mapping receptive fields were relatively ineffective. In contrast, under identical conditions, and for the same individual animals, the identical behaviorally irrelevant stimuli evoked brisk responses in the SC, which is the only known source of visual input to the PBN. Presently, it is unclear if the irrelevant stimuli were ineffective because they were unattended or because PBN cells are specifically selective for small stimuli such as the laser targets.
An important unanswered question is the range of RPE encoded by the PBN. The PBN receives input from the entire SC, and so it should respond to targets in any region of the visual field mapped by the SC. This appears to be the case because most PBN cells responded well to targets appearing over large regions of the visual field. Although we did not examine response fields sufficiently to map their extent, we found at least a crude PBN retinotopy consistent with that described in the anesthetized cat (Sherk 1979a). For example, in the posterior PBN, which responds to stimuli presented in the contralateral hemifield in the anesthetized cat, saccade targets excited PBN cells in the contralateral hemifield and inhibited them in the ipsilateral hemifield. The reverse was found in the anterior PBN, which maps the ipsilateral hemifield. On the other hand, most cells varied their firing rate in a monotonic fashion only for small RPE, and we encountered none that responded only to eccentric targets. However, the activity-RPE functions of some cells had low slope (Fig. 10C), and this suggests that the PBN may encode larger ranges of RPE. We could not systematically sample the PBN in these first experiments, and the parts encoding the most peripheral visual field were the least likely to have been explored, so it is quite possible that we entirely missed regions that encode larger RPE.
Whereas PBN activity during intersaccade intervals is undoubtedly derived from sensory signals, the factors determining activity during saccades are unknown. Presently, we cannot determine if the resetting of activity associated with saccades to stationary and moving targets is due to rapid visual feedback, feed-forward circuits, or feedback from a structure providing an internally derived estimate of how far the eye has moved (e.g., an integrator of eye velocity).
Interactions with the superior colliculus
The PBN probably receives all of its input from the SC, and its main output is fed back to the SC (Graybiel 1978; Sherk 1979b), so it is doubtful that the contributions of the PBN to visuomotor behavior can be understood outside of the context of their interactions. PBN activity is presumably derived from sensory information received by the SC. The direction selectivity of some SC cells projecting to the monkey PBN (Marrocco et al. 1981) may be related to the RPE direction preference of PBN cells. It is not clear if PBN properties are supplied in final form by the SC or developed in the course of SC-PBN interactions. We favor the latter possibility because it is difficult to see what purpose this feedback from the PBN to the SC would serve were it an unmodified replica of SC activity.
Given the close two-way communication between the two structures, a subset of SC cells is likely to share some characteristics with the PBN. Indeed, cells in the rostral SC of both cat (Bergeron and Guitton 2002) and monkey (Krauzlis et al. 2000) have been reported to encode RPE by firing rate. Although originally suggested to gate saccades, these cells are active in circumstances other than steady fixation, and it has been suggested that they form a continuum with premotor cells in the rest of the SC (Bergeron and Guitton 2002; Krauzlis et al. 2000). Ironically, the better match to cat PBN cells is found in the monkey data. The rostral monkey SC cells encode RPE during smooth pursuit of moving targets when the targets are close to the fovea. The functions relating their activity to target distance are quite similar to those of PBN cells, rising steeply from very low activity through the central fovea to peak at small RPE, and then falling off at a slower rate with further increases in RPE (Fig. 1 of Krauzlis et al. 1997). In contrast, in the head-free cat, activity appears to peak for zero RPE, and to monotonically decline with distance between gaze and target (see Fig. 7, A–C of Bergeron and Guitton 2002). However, this difference may not be real. Bergeron and Guitton (2002) caution that because of a high degree of variability in neural response, the peak activity of their cells could only be identified as occurring when gaze was within 3° of the target, so the actual peaks are not necessarily located at zero RPE. This problem may also have been exacerbated by the difficulties of determining gaze position precisely in head-free cats. If the same limitations had applied to our head-fixed cats, the activity-RPE functions would have appeared to peak at zero RPE at a value equal to the average activity for the central 3°. Therefore it is possible that activity is actually at a nadir for most of these SC cells for targets close to the center of gaze. Even so, it should be kept in mind that PBN cells differ from these SC cells in that they have no build-up or motor-associated activity, whereas the latter do (Bergeron and Guitton 2002; Krauzlis et al. 2000).
Possible functions of the PBN
Studies of gaze control in freely moving cats suggest that virtually all gaze saccades are accomplished with combined head/eye movements, that the eye seldom deviates more than 10° from the center of the orbit and that at completion of a gaze saccade the eye is seldom more than 2° from the center of the orbit (Guitton et al. 1984). In comparison to the primate fovea, the cat area centralis is relatively uniform (Hughes 1975). The cat's acuity demands do not evoke frequent adjustments of fixation if the target is within 2–3° of the center of the area centralis, and its oculomotor behavior reflects this both for tracking moving targets and viewing fixed targets. The sensitivity with which RPE is related to activity for targets within the area centralis is clearly sufficient to support the animal's fixation requirements. The instantaneous activity of an individual cell is certainly too variable to do so, but averaged across trials for a single cell, RPE is encoded with a precision considerably better than the behavioral precision (Fig. 8). Population averages across a small number of PBN cells would afford similar or better precision on an instantaneous basis. In this respect, it should be noted that responses of individual SC cells are also too variable to support behavior, and the SC place code for saccade metrics depends on population averaging over a fairly large region of the SC (Lee et al. 1988).
Although the firing rate of PBN cells could provide an internal code for the amplitude and direction of catch-up saccades when averaged across several PBN cells with similar characteristics, it is not clear how the SC makes use of such information. The rate at which SC cells fire during bursts that precede saccade start is believed to determine the speed of saccades not their amplitude (Scudder et al. 2002). Instead, saccade metrics are determined by a place code: the location of the active population of cells in the motor map of deeper SC layers (Scudder et al. 2002). Waitzman et al. (1991) have argued that the SC uses both a rate and a place code simultaneously, but it is difficult to reconcile a rate code with the observation that electrical stimulation of the SC at different rates evokes saccades of the same size and direction (Scudder et al. 2002; see, however, Pélisson et al. 2001). Krauzlis et al. (1997) proposed that the RPE rate code of monkey rostral SC cells coordinates the saccade and smooth pursuit systems. The idea that these signals are somehow passed on to the smooth pursuit system avoids the conundrum of a rate code for RPE in a structure that presumably carries out its main function via a place code. The same logic could be applied to PBN cells. However, the cat smooth pursuit system is rudimentary in comparison to the monkey's, and cats made negligible smooth pursuit movements for the paradigms we employed. This suggests that PBN activity in the cat is related to the distance to potential saccade targets, at least during close-order tracking. On the other hand, the head-free cat can smoothly pursue targets moving at speeds of 25°/s or higher (Missal et al. 1995), so smooth pursuit should play a more prominent role in close-order tracking in natural conditions. Although the technical demands of these experiments did not allow us to examine PBN activity in the head-free condition, one should not neglect a possible role of the PBN in gaze control. Bergeron and Guitton (2002) proposed that the rate code for RPE displayed by cat SC cells reflects feedback to the SC related to the coordination of head and eye movements during gaze saccades. To the extent that the PBN contributes to this activity, it is subject to the same explanation.
Our current uncertainty about the range of RPE encoded in the PBN is a critical impediment to assessing the role of the PBN in oculomotor behavior. If it turns out that the patterns of activity displayed by most of our limited sample are characteristic of the entire PBN, this would imply that the PBN is specialized for close-order tracking or scanning of targets. Response to saccade targets outside of this range might signal target direction but target distance only crudely. Alternatively, if the PBN encodes RPE at least for the full oculomotor range, one should consider a more global role for it in oculomotor behavior.
Because the activity-RPE functions of most PBN cells rose to a peak and then fell, a given activity level often corresponded to two different RPE. This means that an individual PBN cell does not generally provide unambiguous information about RPE (especially for the 3 cells in Fig. 10B). Although this is a hardly a unique situation (neither orientation of edges nor direction of motion is uniquely specified by single cells in the visual system), if PBN activity is used to signal RPE, there would have to be a means to extract this information from the population of cells. The summed activity of many PBN cells encoding a particular direction of best RPE could produce a population activity-RPE function that monotonically rises with RPE over the cat's oculomotor range. Another possibility is that such ambiguities are worked out in the context of the topographic arrangement of interconnections with the SC. Finally, the peak of the activity-RPE function may be the key— i.e., cells may be tuned for a given RPE, as suggested by Krauzlis et al. (1997) for monkey SC cells. It is premature to speculate further until the role of the PBN in oculomotor behavior is more directly demonstrated with lesions, stimulation, or other experimental interventions.
The PBN is a highly conserved structure, and a full understanding of its role should benefit from comparative studies. The mammalian PBN and SC are homologous to the nucleus isthmi and optic tectum, respectively, of other vertebrates. In addition to parallels in anatomical structure and connectivity, the PBN and nucleus isthmi provide the only major cholinergic projection to the SC and optic tectum (Desan et al. 1987; Mufson et al. 1986). The bird nucleus isthmi is the most elaborately evolved. It appears to contribute profoundly to the receptive field organization of tectal neurons, and thoughts about its function have centered on possible contributions to vision (e.g., Wang et al. 2000). For amphibians and reptiles, the most common speculation is that the nucleus isthmi is important for tracking prey in three dimensions (e.g., Weber et al. 1996). Damage to the nucleus isthmi interferes with frogs' ability to orient toward and capture prey, particularly when prey are presented laterally (Gruberg et al. 1991). The PBN and its homologues may be generally important for providing a measure of target location during visual tracking, whether the goal is to put a target on the fovea or the tongue on a fly.
Conclusions
This first examination of PBN activity in the awake animal must be considered a preliminary survey. The properties most firmly established are activity strongly modulated by potential saccade targets; rates of firing varying dramatically over the small ranges of RPE typical of close-order tracking; brisk activity evoked by saccade targets over large response fields; and selectivity for direction of RPE. These data also raise issues that will require more experiments to settle, including: the extent to which PBN activity is selective for attended saccade targets; the timing and strength of inhibition evoked by stimuli presented in the anti-preferred hemifield; the nature of the signals that rapidly reset activity at saccade end; the range of RPE amplitudes represented by rate codes; the topographic representation of RPE amplitude and direction; whether the PBN is required for normal oculomotor function. Many of the experimental paradigms that have been applied to the SC over the last 40 yr might be applicable to the PBN, so it may take some time to firmly establish its response properties and the relationship between these two structures. The strongest hint about this relationship is the similarity of PBN responses to cells in the rostral SC that also appear to encode target distance by firing rate (Bergeron and Guitton 2002; Krauzlis et al. 2000). This raises the possibility that the SC-PBN loop has an important role in establishing the properties of these SC cells. For now, one can say that the PBN is an integral part of a midbrain circuit that generates a sensitive rate code for targets in or near the area centralis. We suggest that this code is used in some way to guide saccades, at least for the catch-up saccades characterizing close-order tracking, but this remains to be proven.
|
APPENDIX |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXACKNOWLEDGMENTSREFERENCES |
|---|
|
The time spent in each 0.1° bin of RPE for this same cell is shown in Fig. A1B. Obviously, the activity-RPE function cannot be meaningfully defined for ranges of RPE for which
), passing though the parabigeminal nucleus (PBN; - - -). Inset: the position of the PBN in the midbrain. Frozen sections were cut at 25 µm and stained with cresyl violet. For subsequent PBN, the microdrive base was placed more laterally, so that the usual angle of approach into the PBN was more nearly vertical, to minimize damage to the midbrain by the microelectrode penetration. Scale bar = 1 mm.
40 ms. C: width of Gaussian filter (
, in °) used for smoothing spike data. This varied inversely with the square root of the time spent at each RPE. D: final result of smoothing with adaptive Gaussian. Firing rates were low at both small and large RPE, but because of the high data density at low RPE, a narrow filter preserves good resolution in this critical area. In contrast, a broad filter is applied at large RPE, but the resulting loss of resolution is not critical because activity usually changes slowly with RPE in this range.