| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Laboratory of Sensorimotor Research, National Eye Institute, National Institutes of Health, Bethesda, Maryland 20892-4435
Submitted 31 July 2003; accepted in final form 24 September 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
2 ms from SC to FEF) and topographically organized (SC neurons drove MD and FEF neurons having similarly eccentric visual and movement fields). Our analysis of identified neurons in one pathway from brain stem to frontal cortex thus demonstrates that multiple signals are sent from SC to FEF with presaccadic activity being prominent. We hypothesize that a major signal conveyed by the pathway is corollary discharge information about the vector of impending saccades. |
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Anatomical work has indicated that one such pathway in monkeys ascends from midbrain to frontal cortex via a thalamic relay (Fig. 1A). Studies showed, first, that some projections from the intermediate layers of the superior colliculus (SC) terminate at the lateral edge of the mediodorsal nucleus (MD) of the thalamus (Benevento and Fallon 1975
; Harting et al. 1980) and, second, that some neurons in this part of MD project to a prefrontal area important for oculomotor control, the frontal eye field (FEF) (Barbas and Mesulam 1981; Goldman-Rakic and Porrino 1985; Kievit and Kuypers 1975; Le Gros Clark and Boggon 1935; Walker 1940). Subsequent transynaptic retrograde tracing experiments using herpes virus injections in FEF found first-order labeling in thalamus, including in lateral MD, and second-order labeling in the SC intermediate layers (Lynch et al. 1994). Taken together, these results provided strong evidence for an SC-MD-FEF pathway. In the present study, we examined the signals sent through this pathway.
|
; Evarts 1968). To identify neurons that receive input from a specific brain area, investigators have used orthodromic activation (synaptically driving neurons with stimulation) (e.g., Raybourn and Keller 1977). In the present study, we used these methods as follows. To identify SC neurons projecting into the ascending pathway, we antidromically activated them from MD. To identify FEF neurons receiving input from the pathway, we orthodromically activated them from the SC. To identify the crucial MD relay neurons linking the SC to the FEF, we orthodromically activated them from the SC and antidromically activated them from the FEF. The use of antidromic and orthodromic activation techniques yielded two further benefits as well: by analyzing activation latencies, we could determine the speed of signal transmission in the pathway, and by comparing the ability to activate MD or FEF neurons from different parts of the SC, we could establish whether neurons along the pathway were connected logically as a function of the well-known topography in the SC (rostral SC represents small and caudal SC large saccades) (Robinson 1972).
Our second main goal was to characterize the signals conveyed through the pathway. Previous researchers hypothesized that the pathway might carry visual signals (Suzuki and Azuma 1983; Wurtz and Mohler 1976), activity related to cognitive processes (Leichnetz et al. 1981), and/or presaccadic activity (Goldberg and Bushnell 1981; Lynch et al. 1994). All of these hypotheses are plausible because recordings in the general populations of SC, MD, and FEF neurons showed that neurons in all three structures can have visual-, cognitive-, or saccade-related activity (FEF: reviewed by Schall 1997; MD: Schlag and Schlag-Rey 1984; Schlag-Rey and Schlag 1984; Tanibuchi and Goldman-Rakic 2003; Wyder et al. 2003; SC: reviewed by Wurtz et al. 2000). Our identification of the specific SC neurons that feed into the pathway, the specific MD neurons that serve as relays, and the specific FEF neurons that receive input from the pathway allows us to test these hypotheses by explicitly determining what signals are sent from SC to FEF and how they change along the way. To evaluate the signals, we recorded from identified neurons while monkeys performed tasks that included making delayed saccades to visual or remembered targets. We focused our analyses on visual responses, delay activity, and saccade-related activity.
We found that the pathway from SC to MD to FEF is fast conducting, topographically connected, and rich in many types of signals from purely visual to purely saccadic. The saccade-related activity seemed especially important because unlike the other signals it coursed through the pathway with no detectable change in its strength or timing. We conclude that a major role of the pathway from SC to FEF may be to provide feedback of saccadic commands, or corollary discharge, to cerebral cortex. In the accompanying paper (Sommer and Wurtz 2004b), we describe experiments in which we tested this hypothesized function by transiently inactivating the pathway.
Some of these results were previously reported in brief articles and abstracts (Sommer and Wurtz 1998, 2000a, 2002; Wurtz and Sommer 2000).
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
The first main goal of this study was to physiologically identify neurons throughout the pathway from SC to MD to FEF. In three 5-10 kg monkeys (Macaca mulatta), we implanted scleral search coils for measuring eye position, recording cylinders for accessing the brain, and a post for immobilizing the head during experiments (see Sommer and Wurtz 2000b for details). All procedures were approved by the Institute for Animal Care and Use Committee and complied with Public Health Service Policy on the humane care and use of laboratory animals. In one monkey (C), we studied identified neurons at all three levels of the pathway, in another (B), we studied identified neurons in SC and MD only, and in the third (H), we studied identified neurons in FEF only. Data were collected from within the left hemisphere of monkeys C and H and from within the right hemisphere of monkey B. We found the FEF and SC using recording and stimulation criteria (presaccadic activity and <50 µA current thresholds for evoking saccades) and confirmed the localizations with magnetic resonance imaging (MRI). To find MD, we used anatomical studies (Barbas and Mesulam 1981; Goldman-Rakic and Porrino 1985; Lynch et al. 1994) and stereotaxic coordinates (Martin and Bowden 1997; Olszewski 1952) as guides. We positioned MD cylinders at A8, L3 and searched in them until we found identified relay neurons (the process of identification is described next). We always inserted recording and stimulating electrodes through guide tubes (23 gauge) aimed using a grid (Crist et al. 1988) having holes 1 mm apart (even finer resolution was achieved in some recordings using grids with holes offset 0.5 mm from the normal pattern).
ANTIDROMIC AND ORTHODROMIC STIMULATION. We used standard techniques to record extracellularly from single neurons (Sommer and Wurtz 2000b). After isolating a neuron, we identified its connections using antidromic and/or orthodromic activation (for reviews, see Lemon 1984; Lipski 1981). Activation means that a neuron produced an action potential shortly after we electrically stimulated a distant structure (for examples see Fig. 1B, top row). Antidromic activation, or backfiring a neuron through its own axon, indicated that the neuron projected to the distant structure. Orthodromic activation, or synaptically driving the neuron, indicated that the neuron received input from the distant structure. We distinguished anti-from orthodromic activation primarily with the collision test in which electrical stimulation is timed to occur just after a spontaneous action potential of the neuron (Fig. 1B, bottom). If the neuron projects to the area containing the stimulating electrode, then the spontaneous action potential moving forward in the axon will meet the stimulation-evoked action potential moving backward in the same axon, and neither will travel further because the axon will be in its absolute refractory period on both sides of the meeting point. Hence the stimulation-evoked action potential will not appear at the recording electrode (Fig. 1B, bottom right, *), whereas normally it would (Fig. 1B, top right). If the delay between onset of the spontaneous action potential and onset of stimulation is increased past a certain interval, 0.2-0.6 ms longer than the activation latency, then the collision effect ceases (not shown). If all this occurs, then the activation passes the collision test, and the neuron is considered to be antidromically activated; it projects to the distant structure holding the stimulating electrode. If the activation fails the collision test (Fig. 1B, compare top left and bottom left), then the neuron is considered to be orthodromically activated; it receives input from the distant structure. To complement the collision test, we also examined whether the activation latency was stable or not; stable activation latencies, exhibiting a range 
0.1 ms (Fig. 1B, top right), are typical of antidromic activation, but jittery latencies, range >0.1 ms (Fig. 1B, top left), are typical of orthodromic activation. In rare cases when we were unsure whether the activation was antidromic versus orthodromic, we abandoned the neuron and searched for a new one.
Neurons were identified as belonging to the ascending pathway according to the following criteria: SC neurons had to be antidromically activated from MD, MD neurons had to be orthodromically activated from the SC and antidromically activated from the FEF, and FEF neurons had to be orthodromically activated from the SC. To activate neurons, we stimulated through tungsten microelectrodes (100 k
at 1,000 Hz; Frederick-Haer) using single, biphasic, negative-positive current pulses of 0.15 ms/phase. We implanted these monopolar stimulating electrodes semi-chronically for several weeks except during some experiments (as mentioned in RESULTS) in which we used moveable electrodes. Accurate placement of the stimulating electrodes was critical to this study, and therefore details about this are provided next.
SC-STIMULATING ELECTRODES. We implanted stimulating electrodes in the SC (for orthodromically activating MD or FEF neurons) with their tips in the intermediate layers in all three monkeys. We always implanted a pair of SC-stimulating electrodes with one relatively rostral (in or near the "fixation" zone; Munoz and Wurtz 1993) and one relatively caudal (typically 10-20° eccentricity on the SC topographic map) (Robinson 1972), except in a few experiments when we used a single, moveable electrode to find the current threshold profile as a function of depth into the SC (e.g., Fig. 4, E and F). To confine stimulating currents as much as possible to the SC, we positioned stimulating electrodes on or near the representation of the horizontal meridian that runs through the center of the SC map (see Fig. 4C, 0° direction contour). Before epoxying an electrode in place, we always recorded through it to ensure that its tip was amid neurons having fixation-related, foveal visual, and/or small-saccade-related activity (for the rostral electrode) or peripheral visual responses and large-saccade-related activity (for the caudal electrode), and we stimulated through the electrode using trains of pulses to ensure that low currents (<20 µA at 350 Hz for 70 ms using biphasic pulses having durations of 0.25 ms/phase) fixed the eyes in place or evoked small saccades (for the rostral electrode) or evoked larger saccades (for the caudal electrode). After several weeks, the implanted electrodes became unusable (they failed to pass current reliably), at which time we replaced them with new electrodes at slightly different SC locations. Over the course of the study, we used a total of 10 SC electrode pairs in the three monkeys. On average, we placed rostral electrode tips at 1.7° eccentricity on the SC map and 1.8 mm depth below the SC surface and caudal electrode tips at 13.0° eccentricity and 1.7 mm depth. The current thresholds for orthodromically activating neurons using these SC electrodes were on average 264 µA (range: 7-718 µA, n = 45) for MD neurons and 110 µA (range: 16-540 µA, n = 47) for FEF neurons. The difference in these averages probably was not important; it may have been an artifact of slightly different SC stimulating electrode placements in each of the three monkeys.
|
), the average threshold of 1,298 µA at a pulse duration of 0.15 ms should activate low-threshold axons (K 
381 µA/mm2) only within 1.8 mm and high-threshold axons (K 4,844 µA /mm2) only within 0.5 mm of the electrode tip. MD-STIMULATING ELECTRODES. We implanted stimulating electrodes in MD (for antidromically activating SC neurons) in two monkeys at sites of previously recorded MD relay neurons. To minimize damage to the relay neurons and permit further experiments in the region (i.e., reversible inactivations; Sommer and Wurtz 2004b), we used only a single stimulating electrode in the MD of each monkey. We placed it at the most posterior site found to contain relay neurons, reasoning that SC afferents follow a posterior-anterior trajectory so that most should pass near this electrode before terminating in MD. Current thresholds for antidromically activating SC neurons from MD were on average 270 µA (range: 35-1,284 µA, n = 48).
TOPOGRAPHY OF PROJECTIONS. We also used orthodromic activation to search for projection topographies in the pathway. The concept was to see if stimulation in rostral (or caudal) SC preferentially drove MD or FEF neurons that represented small (or large) saccades as would be logical according to the SC map (Robinson 1972). This required measuring the visual and/or movement field of each MD or FEF neuron, and our procedure for doing this is described in the VISUAL AND MOVEMENT FIELDS section. We also had to estimate what part of the SC provided input to the neuron, and to do this, we calculated for each MD or FEF neuron a contrast ratio called the electrode preference index (EPI). The principle behind the EPI was as follows. Because we typically used two stimulating electrodes in the SC, one rostral and one caudal, we could compare the ability to drive an MD or FEF neuron from rostral versus caudal SC. If stimulation through one electrode activated the MD or FEF neuron at a lower current threshold than did stimulation through the other, then the former, "better" electrode was probably located nearer to those SC neurons that drove the MD or FEF neuron. EPI = (Ir - Ic)/(Ir + Ic), where Ir and Ic were the current thresholds for activating the recorded neuron from rostral and caudal SC (threshold was defined as the current causing activation in 50% of trials). An EPI closer to -1 or +1 suggested that the neuron was preferentially driven from projections originating closer to rostral or caudal SC, respectively.
Characterizing signals in the pathway
Our second main goal was to evaluate the signals sent from SC to MD to FEF. After physiologically identifying a neuron as belonging to the pathway as described in the preceding text, we analyzed its signals by having the monkey perform a series of tasks. Details of the testing apparatus were described previously (Sommer and Wurtz 2000b). Briefly, the monkey faced a tangent screen on which visual stimuli were projected by an LCD monitor. Visual stimuli were 0.3 x 0.3° blue or red spots (0.6 cd/m2) presented on a dark background (0.1 cd/m2) with dim ambient room light. Personal computers controlled the presentation of visual stimuli and recorded at 1 kHz the eye position, the occurrence of action potentials, and the timing of task events.
DELAYED-SACCADE TASKS. The purpose of these tasks was to search for visual responses, presaccadic activity, and tonic activity known as delay activity that seems related to cognitive processes such as memory or planning (e.g., see Sommer and Wurtz 2001). The delayed-saccade tasks are diagrammed in Fig. 10A. All trials began with a fixation point appearing in the center of the screen. After the monkey foveated it for 500-800 ms (pseudorandomly varied, like all task timings in this study), a target appeared at the estimated center of the visual and/or movement field. In visual trials of the task (Fig. 10A, Vis.), the target remained lit for the rest of the trial; in memory trials (Fig. 10A, Mem.), the target disappeared after 100 ms. In all trials, after a delay period of 500-1,000 ms the fixation spot disappeared, cueing the monkey to make a saccade to the location of the target after which a water reward was delivered.
|
GAP TASK. The purpose of this task was to look for "gap activity" during a brief period (the gap) after a fixation spot disappears but before a saccadic target appears. Such activity may be related to cognitive processes such as disengaging fixation or preparing to make a saccade (e.g., Dias and Bruce 1994; Munoz et al. 2000). In the gap task (Fig. 12A), the monkey fixated a spot for 500-800 ms and then the spot disappeared; the monkey had to maintain fixation on the blank screen, and then, after a 200-ms gap period, a target was presented at the estimated center of the visual and/or movement field; the monkey then looked at the target to receive its reward. We compared the mean firing rate during the gap period epoch, from 50 ms before target onset to 50 ms after (Fig. 12A), with the firing rate during a baseline epoch 500-200 ms before start of the gap. Gap activity occurred if the gap period activity exceeded the baseline activity.
|
VISUAL AND MOVEMENT FIELDS. For every identified neuron, we measured the range of locations where visual stimuli caused it to fire (its visual field) and the range of saccadic vectors for which it fired presaccadically (its movement field). As with all testing in this study, we performed these measurements while the monkey's head was held stationary. We first estimated these fields on-line (see Fig. 7A, pink) by having the monkey make saccades to a variety of target locations while we inspected rasters of neuronal activity. For neurons having both visual and movement fields, it was clear by inspection that the fields were always highly coincident, so a single estimated field accurately represented both component fields. From this initial testing, we found the location of the target that evoked maximal activity, and we presented targets at this estimated best target location during the delayed-saccade and gap tasks described in the preceding text.
|
; Munoz and Wurtz 1995), e.g., 20° leftward saccades were elicited by placing the fixation spot 10° right of center and the saccadic target 10° left of center. We did this for all target eccentricities except for the two smallest for which we used a central fixation spot.
For every neuron having a significant visual response (as determined quantitatively using the delayed-saccade tasks), we reconstructed its visual field off-line by measuring average visual activity 50-150 ms after target onset in the direction and eccentricity series data. Similarly, for every neuron having significant presaccadic activity, we reconstructed its movement field by measuring average presaccadic activity 50-0 ms before the saccade in the direction and eccentricity series data. We fitted Gaussians to the direction series data (least-squares approximation) separately for the visual activity and the presaccadic activity, yielding direction cross-sections through the visual and movement fields (Fig. 7, B and C, left). We measured the direction range as the range of directions for which the Gaussian was >2 SDs above the mean baseline firing rate (we measured baselines 300-0 ms before target onset). The direction series data were shifted before fitting the Gaussian so that the fitted curve would peak near the center of the range; periodic curve fits, e.g., cosines, were attempted but the fields were too narrow to be well fit by them. We fitted splines to the eccentricity series data (Munoz and Wurtz 1995), yielding eccentricity cross-sections through the visual and movement fields (Fig. 7, B and C, right). We measured the eccentricity range as the range of eccentricities for which the spline was >2 SDs above the mean baseline firing rate. We used cubic spline fits except when they were clearly inadequate due to sharply varying data points (e.g., Fig. 7D, left and middle), in which case we used a ninth-order spline. We performed all curve fitting with Matlab (The MathWorks).
We note that for quantifying movement fields, activity was plotted against the actual direction or amplitude of each saccade, which is why data clusters in Fig. 7, C and D, exhibit scatter in both the x- and y-axis directions. For movement field eccentricity, therefore, spline fits were least-squares approximations. For visual fields, in contrast, we calculated the mean firing rate for each target location, which is why there is a single data point at each direction or eccentricity in Fig. 7B. Also, for simplicity we use the term "eccentricity" to describe radial extent for visual and movement fields, although "amplitude" would be more precise for movement fields.
DETECTION OF SACCADES AND FIXATIONS. On-line, we identified saccades and fixations using real-time detection software written in-house. A saccade was accurate if it landed in a rectangular virtual window surrounding the target location that ranged in size from 1° horizontally x 2° vertically for targets at small eccentricity (e.g., 4°) to 10 x 20° for targets located at large eccentricity (e.g., 40°). Windows for enforcing fixation were 2 x 2° squares around the fixation point. Off-line, we used software running a template-matching algorithm to automatically detect saccades in eye-position records. We verified the accuracy of saccade detection by visually inspecting the data from every trial.
STATISTICS. Unless explicitly noted otherwise, throughout this paper we compare data sets using Student's t-test, if judged normal by the Kolmogorov-Smirnov test and of equal variance by the Levene Median test, or else the Mann-Whitney rank sum test, and we analyze correlations using Pearson's test.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
We studied a total of 151 identified neurons: 48 SC source neurons projecting into the ascending pathway, 47 MD relay neurons linking SC to FEF, and 56 FEF recipient neurons targeted by the pathway.
MD RELAY NEURONS. Identification of MD relay neurons was the crucial first step in our study because it explicitly confirmed the presence of a pathway from SC to MD to FEF. Every time we isolated a neuron in MD, we tried to activate it using stimulating electrodes in the SC and the FEF (Fig. 1A). All 47 MD relay neurons in our sample (17 from monkey B, 30 from monkey C) were orthodromically activated from the SC as evidenced by activations that always failed the collision test (Fig. 1B, bottom left) and stimulation-evoked action potentials that nearly always had jittery latencies (Fig. 1B, top and bottom left). All the MD neurons also were antidromically activated from the FEF, as evidenced by activations that always passed the collision test (Fig. 1B, bottom right) and stimulation-evoked action potentials that always had stable latencies (Fig. 1B, top right). These individual MD neurons therefore received SC input and projected to the FEF, verifying the existence of a pathway from SC to MD to FEF.
We recorded only from well-isolated MD neurons and were always confident that both the orthodromically and antidromically activated action potentials were produced by the same neuron. To confirm this, when time permitted we also used the relay collision test (Fig. 1C), which has been used previously for identifying thalamic relay neurons of other pathways (e.g., Deschênes et al. 1982; Zhu and Lo 1998). Instead of waiting for the MD neuron to fire spontaneously to perform the collision test, we made it fire by orthodromically activating it from the SC. The orthodromically activated action potential could annihilate an action potential evoked antidromically from the FEF only if both sets of action potentials were produced by the same MD neuron. This always occurred; every neuron tested (11/11) passed the relay collision test.
In both monkeys, we found MD relay neurons 3 mm lateral of the midline and 7-9 mm anterior to the interaural line (Fig. 2A). In monkey B, we took particular care to search outside this narrow zone but found no other relay neurons (Fig. 2A, right). Many of the surrounding neurons could be orthodromically activated from the SC or antidromically activated from the FEF but not both (Fig. 2B); we do not know to where these SC-recipient neurons projected or from where these FEF-projecting neurons received their input, and we did not study their signals. We found MD relay neurons 18-23 mm below the top of the brain, slightly deeper (by 3 mm) in one monkey than in the other (Fig. 2C). The coordinates of the MD relay neurons corresponded well with the lateral edge of MD as described in stereotaxic atlases (Martin and Bowden 1997; Olszewski 1952) and as predicted by anatomical studies (Barbas and Mesulam 1981; Goldman-Rakic and Porrino 1985; Lynch et al. 1994). This lateral MD location was histologically verified in both monkeys (Fig. 3 shows the results from monkey C).
|
|
4-5 mm below. We found source neurons primarily 1-3 mm below the surface (Fig. 4D), corresponding approximately to the intermediate layers. To test this assessment that SC intermediate layer neurons drove the MD relay neurons, we also measured the optimal depths of stimulation within the SC for orthodromically activating MD relay neurons. We did this for eight MD relay neurons by moving a stimulating electrode through the SC and finding the depth of lowest current threshold for activation. Figure 4E shows a current threshold profile for one experiment: the minimum was in or near the intermediate layers where we found strong visual- and saccade-related activity during recordings in the same penetration. In most cases (7/8), threshold minima were in the intermediate layers (Fig. 4F), whereas in one case, the threshold minimum was deeper. We never found a threshold minimum in the purely visual superficial layers. These current threshold minima correspond well with the depths of the SC source neurons (cf. Fig. 4D), and therefore both lines of evidence agree that SC input to the MD relay neurons arises primarily from the intermediate layers.
FEF RECIPIENT NEURONS. All 56 FEF recipient neurons (13 from monkey C, 43 from monkey H) were orthodromically activated from the SC (Fig. 5, A and B), and, as we previously reported, current thresholds for driving them were lowest in the SC intermediate layers (Sommer and Wurtz 1998). Note that it was crucial to activate FEF recipient neurons from the SC, not from MD; as was shown in Fig. 2B (thin, solid-line boundaries), many MD neurons that project to the FEF do not seem to get input from the SC, and therefore stimulating MD instead of SC could mistakenly identify FEF neurons that are targeted by pathways other than that originating in the SC. Recipient neurons were located throughout most of the mediolateral range of the FEF (Fig. 5C) and from near the top of cortex to 9 mm deep (as the electrode traversed the anterior bank of the arcuate sulcus; Fig. 5D). We often found FEF recipient neurons in the same penetrations that yielded FEF SC-projecting neurons (Sommer and Wurtz 2000b), but these two classes of neurons were not intermingled; they were always separated from each other by at least a few hundred microns in depth. The FEF recipient neurons were likely in layer IV, where most thalamic afferents terminate (Giguere and Goldman-Rakic 1988). FEF SC-projecting neurons are found only in layer V (Fries 1984; Leichnetz et al. 1981).
|
|
PROJECTION TOPOGRAPHY IN THE PATHWAY. We also used our activation methods to examine whether neurons in the pathway were linked together in a logical manner according to the topography of the SC. We expected MD and FEF neurons having small eccentricity fields to receive input primarily from rostral SC and those having large eccentricity fields to receive input primarily from caudal SC (see Fig. 4C for SC topography) (Robinson 1972). First we document the general characteristics of visual and movement fields throughout the pathway, and then we analyze the projection topography.
Figure 7, A-C, summarizes how we measured the visual and movement fields for one example MD relay neuron (see METHODS for details). It was clear from initial testing that the neuron had both visual- and saccade-related activity and that its visual and movement fields were almost perfectly coincident. After estimating the average location and size of the fields (Fig. 7A, pink), we had the monkey make saccades to targets in a direction series (Fig. 7A, orange) and eccentricity series (Fig. 7A, blue) that cut through the estimated fields. We plotted visual responses as a function of direction (Fig. 7B, left) and eccentricity (Fig. 7B, right) to reconstruct the visual field and similarly plotted presaccadic activity to reconstruct the movement field (Fig. 7C). This example neuron had a visual field (Fig. 7B) with best direction 35°, direction range 90°, best eccentricity 16°, and eccentricity range 37°. Its movement field was nearly identical (Fig. 7C), having a best direction of 27°, direction range of 128°, best eccentricity of 16°, and eccentricity range of 31°. The correspondence of this neuron's visual and movement fields was typical; similar overlap was seen in all of our MD, SC, and FEF neurons that had both visual and presaccadic activity.
The visual and movement fields for this example neuron were "closed", i.e., bounded in eccentricity (Fig. 7, B and C, right), but some other neurons had "open" visual or movement fields in that visual or presaccadic activity was significantly elevated even at the furthest eccentricity tested. Open fields have been described previously for SC and FEF neurons (Bruce and Goldberg 1985; Munoz and Wurtz 1995). We found open visual fields and movement fields, respectively, in 18% (6/33) and 53% (18/34) of SC neurons, 57% (16/28) and 76% (22/29) of MD neurons, and 18% (6/33) and 43% (10/23) of FEF neurons. There were three basic kinds of open visual or movement fields as illustrated by representative examples in Fig. 7D. In some neurons, activity rose and then stayed at a plateau level for as large of eccentricities that we could test (Fig. 7D, left); these fields had no distinct peak. In other neurons, there was a distinct peak, but the firing rate then never dropped below the significance criterion level even at very large eccentricities (Fig. 7D, middle). Finally, in some neurons, the firing rate increased monotonically with eccentricity (Fig. 7D, right).
Visual or movement fields having distinct peaks of activity (e.g., Fig. 7, B and C, and D, middle) appeared to encode a specific target location or saccadic vector, and those having firing rates that increased steadily with eccentricity (Fig. 7D, right) seemed to encode the distance of the target from the fovea (or the length of the saccade). Such visual and movement fields, carrying information about the target location or saccade vector, were found, respectively, in 100% (33/33) and 88% (30/34) of the SC source neurons, 61% (17/28) and 79% (23/29) of the MD relay neurons, and 100% (33/33) and 96% (22/23) of the FEF recipient neurons. Signals encoding the target location or saccadic vector therefore were abundant throughout the pathway.
Quantitative comparison of the sizes and spatial locations of the visual fields (Fig. 8A) and movement fields (Fig. 8B) showed that they remained similar throughout the pathway. Looking for differences from SC to MD and from MD to FEF, we found only that the best eccentricities of visual fields in MD were slightly larger than in the SC or the FEF (Fig. 8A, *; P < 0.025 criterion, corrected from P < 0.05 because the MD data were used in 2 comparisons). Throughout the pathway nearly all visual and movement fields had contralateral best directions (Fig. 8, A and B, far left).
|
; 
, irrelevant here, is described in DISCUSSION).
|
To summarize the results, we pooled all the data (Fig. 9C) and used the resulting regression equation and the average locations of the rostral and caudal stimulating electrodes (see METHODS) to illustrate the projection topography schematically (Fig. 9D). The topography was reasonably precise: projections from 2° eccentricity on the SC map preferentially drove MD or FEF neurons having visual or movement fields of 1° best eccentricity, projections from 5° eccentricity preferentially drove neurons having fields of 6° eccentricity, and projections from 13° eccentricity preferentially drove neurons having fields of 23° eccentricity.
Signals conveyed through the pathway
VARIETY OF SIGNAL TYPES CARRIED BY THE NEURONS. We evaluated the neuronal signals primarily with delayed-saccade tasks (Fig. 10A). After Bruce and Goldberg (1985), we classified neurons into three major categories: visual neurons (Fig. 10B) having phasic or tonic visual activity but no presaccadic activity, visuomovement neurons (Fig. 10C) having phasic or tonic visual activity plus presaccadic activity, and movement neurons (Fig. 10D) having presaccadic activity but no phasic or tonic visual activity. We found all three categories of neurons throughout the pathway, except that there were no movement neurons in the FEF recipient neuron sample (Fig. 10D).
Figure 11A shows the distributions of visual, visuomovement, and movement neurons at each stage of the pathway. Nearly every neuron in the pathway was active in the delayed-saccade tasks with only a small proportion at each stage unmodulated ("other" neurons). Visuomovement neurons were the most common type at every stage. We compared the distributions of neuron types from SC to MD and from MD to FEF (
2 tests, P < 0.025 criterion because MD data were tested twice). From SC to MD, there was no significant change in the distribution of neuron types, but from MD to FEF, the distribution changed significantly: the FEF distribution was much more visual in nature than would be expected from its MD input.
|
; Goldman-Rakic 1995). As examples, delay activity was present in all of the visuomovement neurons in Fig. 10C (see orange data representing memory trials). From SC to MD, the proportion of neurons with delay activity dropped significantly (Fig. 11B), but from MD to FEF, it did not change. Because the proportions of visual, visuomovement, and movement neurons stayed the same from SC to MD (Fig. 11A), but the proportion of neurons having delay activity dropped (Fig. 11B), the connection from SC to MD seemed to act like a high-pass filter. That is, the connection seems to let bursts of action potentials through more readily than sustained activity (visual, visuomovement, and movement neurons nearly always had sharp visual- and saccade-related bursts of activity; see Fig. 10, B-D). To test this idea, we also examined whether the proportion of neurons carrying another sustained signal, tonic visual activity, similarly decreased from SC to MD. Figure 11C shows that it did. Moreover, we found that this was not a trivial result of delay activity and tonic visual activity tending to co-occur in the same neurons (analysis not shown).
Notably, however, the proportion of tonic visual signals increased from MD to FEF (Fig. 11C). Taken together with the result of Fig. 11A, that FEF recipient neurons are generally more visual than expected from their MD input, this suggests that FEF recipient neurons receive additional visual signals from elsewhere (probably from extrastriate cortex).
Many neurons in the SC and the FEF have gap activity presumably related to cognitive aspects of preparing to move or disengaging from fixation (Dias and Bruce 1994; Munoz et al. 2000), so we also used a gap task (Fig. 12A) to see if the pathway carried this signal. Gap activity is a slowly rising firing rate at the end of a gap period, occurring after a foveated spot disappears and before the neuron can possibly respond to the onset of the visual target for a saccade. Some neurons at every stage of the ascending pathway had gap activity (Fig. 12B). The occurrence of gap activity, however, diminished significantly from SC to MD (Fig. 12C). This provides even further evidence for a high-pass filter at the SC-MD synapse, given the slowly varying nature of gap activity (albeit not as sustained as delay or tonic visual activity, it nevertheless is a much more slowly varying signal than the visual or saccadic bursts). From MD to FEF, there was no significant change in the proportions of neurons having gap activity (Fig. 12C).
Finally, we also looked for two other kinds of signals that were relatively rare: fixation-related and postsaccadic activity. A few neurons clearly changed their firing rate at the start of fixation, and to further study them, we ran monkeys on a standard fixation blink task in which the fixation spot disappeared for 400 ms, allowing us to distinguish activity related to the motor act of fixating from activity related to foveal visual responsiveness (for details, see Sommer and Wurtz 2000b). We found 5 SC neurons with foveal visual responses, 3 of which also had a fixation signal, 3 MD neurons with fixation signals, 2 of which had a foveal visual response, and 10 FEF neurons having the following distribution of signals: 2 had a foveal visual response but not a fixation signal, 4 had a fixation signal but not a foveal visual response, and 4 had both a foveal visual response and a fixation signal. We also searched for postsaccadic activity using the delayed-saccade tasks (see METHODS), and found three neurons in each of the SC, MD, and FEF samples that exhibited this activity.
STRENGTH AND TIMING OF SIGNALS. The preceding analyses were concerned with the percentages of various signals at different levels of the pathway; now we consider the activity profiles of the signals. We focused on visual- and saccade-related bursts of activity because these signals were plentiful throughout the pathway. For each neuron, we constructed spike density functions to summarize its firing rate in the visual version of the delayed-saccade task. To analyze visual bursts (Fig. 13A, left), we used Gaussians of width 
= 2 ms for constructing spike density functions (narrower Gaussians resulted in data too noisy for reliably analyzing every neuron, and wider Gaussians resulted in unacceptable underestimates of the visual latency and visual burst magnitude). The latency of a visual burst was the duration from target onset until the spike density function crossed a significance threshold set to 2 SDs above mean baseline activity (measured from 40 ms before to 40 ms after target onset). The burst magnitude was the peak magnitude of the burst minus the mean baseline activity. The peak time was the duration from target onset until the peak magnitude of the burst. In this example (Fig. 13A, left), the visual burst latency was 75 ms, the burst magnitude was 475 spikes/s, and the peak time was 90 ms after target onset (also, the baseline activity had a mean of 8 spikes/s and an SD of 11 spikes/s, yielding a significance threshold of 30 spikes/s). To analyze saccadic bursts (Fig. 13A, right), we used similar methods and the same descriptors (latency, burst magnitude, and peak time) except that the spike density function Gaussian was set to 10 ms and the baseline activity was measured during the immediately preceding Delay epoch (see Fig. 10A). We tried narrower Gaussians but they often caused spurious, temporary rises above the baseline activity prior to the saccade. In this example (Fig. 13A, right), the saccadic burst latency was -96 ms, the burst magnitude was 194 spikes/s, and the peak time was -8 ms, where negative timing values represent periods before saccade initiation.
|
65 ms. Thus although Fig. 13B accurately depicts the relative differences between the SC, MD, and FEF data, one should consult Table 1 for exact timing values. The most striking result in the visual burst data were that the FEF latencies occurred earlier than the MD latencies (Fig. 13B, left; Table 1, top row). In other words, the visual response of FEF recipient neurons preceded that of the MD neurons projecting to them, and therefore the primary visual drive of the FEF recipient neurons cannot be coming from the ascending pathway. In contrast, from SC to MD, the visual burst latencies were not different (Fig. 13B, left; Table 1, top row), consistent with the SC provid
) on the standard SC map of Robinson (1972
) was 265 µA, at 2.8 mm deep. F: results of all 8 tests in which the current threshold for driving MD relay neurons was evaluated as a function of electrode tip depth in the SC.
