INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The importance of the frontal eye field (FEF) and the superior colliculus (SC) to oculomotor behavior is well established. Electrical stimulation using low currents in either structure elicits saccadic eye movements (Bruce et al. 1985; Robinson 1972; Robinson and Fuchs 1969; Schiller and Stryker 1972) or inhibits them such that the eyes stay fixed (Burman and Bruce 1997; Munoz and Wurtz 1993b). Neurons in both FEF and SC discharge immediately before saccade initiation (Bruce and Goldberg 1985; Schiller and Koerner 1971; Wurtz and Goldberg 1971) or during fixation (Bizzi 1968; Munoz and Wurtz 1993a; Suzuki et al. 1979). Reversible inactivation of either structure severely disrupts saccades and fixations (Dias and Segraves 1999; Hikosaka and Wurtz 1985, 1986; Schiller et al. 1987; Sommer and Tehovnik 1997). Ablation of either structure impairs saccades and fixations for a few days or weeks, but animals then recover and exhibit only a few types of long-term deficits (e.g., Deng et al. 1986; Schiller et al. 1987). If both the FEF and the SC are bilaterally ablated, however, the ability to make saccades and fixations is permanently devastated (Schiller et al. 1980).

The FEF projects to the SC (reviewed by Leichnetz and Goldberg 1988), and there is evidence that much of the FEF's influence over oculomotor behavior is mediated by this projection. The FEF relays oculomotor signals to the SC (monkey: Segraves and Goldberg 1987; cat: Weyand and Gafka 1998b), electrical stimulation of FEF modulates neuronal activity in the SC (monkey: Schlag-Rey et al. 1992; cat: Guitton and Mandl 1974), and if the SC is reversibly inactivated, the ability to evoke saccades electrically from the FEF is impaired (Hanes and Wurtz 1999).

To understand the role of the FEF in the oculomotor system, therefore, it is important to elucidate the nature of the signals relayed from FEF to SC. One way to accomplish this is to record from neurons in the FEF that are identified as projecting to the SC by virtue of their antidromic activation after stimulation of the SC. In a landmark application of this method, Segraves and Goldberg (1987) concluded that the output of FEF to the SC primarily consists of signals related to saccade generation or suppression. Because the FEF is known to contain a wide variety of signals, from purely visual to purely motor in nature (Bruce and Goldberg 1985; Schall 1991), the FEF's output to SC appeared to be "selectively enriched" in motor-related signals (Goldberg and Segraves 1990; Segraves and Goldberg 1987).

When we began the present study, no one had examined the projection from FEF to SC in the monkey since Segraves and Goldberg (1987). Of necessity, therefore our first goal was to determine the composition of signals in this projection so as to replicate the previous findings. To characterize the signals flowing from FEF to SC as carefully as possible, we took advantage of techniques that have become common in the past decade: 1) use of a grid system for aiming electrodes that facilitates the systematic exploration of a cortical region and 2) statistical analysis of spike trains that facilitates the objective classification of signals carried by neurons. Also, we tried to correct for sampling bias caused by cell size variation using a method adapted from primary motor cortex research.

The present study was primarily motivated by three recent findings regarding SC neurons. First, there are tonic neuronal discharges in the SC that intervene after the phasic response to a visual target for a saccade and before the phasic discharge that is synchronized to saccade initiation. These tonic delay signals can predict hundreds of milliseconds in advance where, or whether, a saccade will be made, suggesting that they help to mediate attention, memory, or planning (Basso and Wurtz 1998; Glimcher and Sparks 1992; Kojima et al. 1996; Munoz and Wurtz 1995; Sommer et al. 1997). Second, neuronal discharges in the SC can increase during a temporal gap that elapses after a foveated spot disappears and before a visual target for a saccade appears in the periphery (Dorris and Munoz 1998; Munoz and Wurtz 1995); such gap signals are correlated with relatively fast reaction times (Saslow 1967) and therefore may help mediate fixation disengagement or movement planning (e.g., Paré and Munoz 1996; Reuter-Lorenz et al. 1995). Third, neurons concentrated in the rostral pole of the monkey SC exhibit tonic discharges during steady fixation (Munoz and Wurtz 1993a). Often, neurons with these fixation signals also pause during saccades and show an increase in activity after saccades, all of which suggests that the neurons are important for keeping the eyes still. All three of these recently studied types of signals in the SC also have been found in the FEF (Bizzi 1968; Bruce and Goldberg 1985; Dias and Bruce 1994; Funahashi et al. 1989; Hanes et al. 1998; Joseph and Barone 1987).

The second goal of this study, therefore, was to examine the properties of delay, gap, and fixation signals that may be relayed from FEF to SC. The report of Segraves and Goldberg (1987) did not discuss delay and gap signals. Both types of signals are thought to help mediate cognitive operations, as noted above, and if neurons projecting from FEF to SC were found to carry these signals, this would reveal a distinct subcortical route through which frontal lobe cognitive operations might influence oculomotor behavior. Segraves and Goldberg (1987) previously demonstrated that fixation-related signals are relayed from FEF to SC, and they briefly commented that FEF neurons with such signals could be activated from "a wide range of points" on the SC map (their p. 1399). We followed up on this note by systematically testing whether fixation-related signals might project preferentially to the rostral as opposed to the caudal SC. If a rostral bias were present, this would provide strong evidence that the FEF is a source of the fixation-related signals carried by rostral SC neurons.

Our third goal was to analyze the topography of visual and movement fields in the FEF and determine how this topography projects onto the map of space within the SC. It has long been known that visual and movement space is represented topographically across the SC; in particular, the eccentricities of visual receptive fields and movement fields decrease gradually from caudal to rostral (Apter 1945; Cynader and Berman 1972; Sparks et al. 1976). In the FEF, the eccentricities of visual and movement fields appear to decrease from mediodorsal to ventrolateral (Bruce et al. 1985). Three important aspects of the FEF eccentricity gradient, however, are still unknown. First, it is not known which laminae in the FEF contain this eccentricity map. We examined whether an eccentricity map exists specifically in the corticotectal population of FEF neurons concentrated in layer V (Fries 1984; Leichnetz et al. 1981). Second, the exact angle of the eccentricity gradient across the two-dimensional area of the FEF is unknown. To explicitly determine this angle, we plotted eccentricity as a function of two-dimensional location in the FEF. Third, it is unknown whether the FEF's eccentricity gradient projects directly onto the known gradient of eccentricity in the SC. We analyzed this by comparing the eccentricities represented by FEF neurons to the collicular termination zones of these neurons. Prior, less direct evidence for a superposition of FEF and SC eccentricity maps has come from a variety of anatomic (Komatsu and Suzuki 1985; Stanton et al. 1988) and electrophysiological studies (Schlag-Rey et al. 1992; Segraves and Goldberg 1987).

In the present study, we first physiologically identified the FEF and SC (Fig. 1A) and implanted stimulating electrodes in the rostral and caudal SC (Fig. 1B). We then characterized the task-related signals of FEF neurons that were antidromically activated from the SC (Fig. 1C), estimated the locations of these neurons within the FEF, and estimated their termination locations along the rostrocaudal axis of the SC. We found that essentially all the neuronal types previously identified in the FEF can be antidromically activated from the SC, suggesting that the output of the FEF to the SC is not selective but reflects the general population of FEF signals. A large proportion of FEF corticotectal neurons exhibited delay signals, gap signals, or both. Surprisingly, FEF neurons with activity strongly related to the act of fixating did not project preferentially to rostral SC, although other types of corticotectal neurons did. We found a two-dimensional map of eccentricity in the FEF corticotectal neuron population and showed that it projects in an orderly manner onto the map within the SC.

View larger version (15K): [in this window] [in a new window]   Fig. 1. Techniques used for studying signals sent from frontal eye field (FEF) to superior colliculus (SC). A: lateral view of the macaque brain is shown, with brain stem revealed in a cutaway view. FEF and SC are labeled. For reference in subsequent figures, the principal sulcus (Pr) and the arcuate sulcus's superior (As) and inferior (Ai) branches are indicated. B: the FEF, in the rostral bank of the arcuate sulcus, is depicted as if removed from the rest of the brain. We isolated and studied single neurons in the FEF that were activated antidromically from electrodes implanted in the rostral and the caudal SC. C: action potential waveforms are shown to illustrate application of the collision test. Top: a single biphasic pulse of electrical stimulation in the SC (beginning at time 0 and lasting for 0.3 ms as indicated by solid bar) reliably caused the neuron, recorded in FEF, to fire at 1.0 ms latency (black arrow). Stimulus artifact is erased for clarity. Bottom: if stimulation in the SC was synchronized to occur 1.4 ms after the FEF neuron fired spontaneously, then stimulation failed to evoke an action potential in the FEF neuron (white arrow). The spontaneous spike traveling away from the cell body collided with and annihilated the stimulation-evoked spike traveling toward the cell body. This neuron therefore passed the collision test, demonstrating that it projected to the SC.

Brief reports pertaining to some of these data have appeared previously (Sommer and Wurtz 1998a,b, 1999; Wurtz and Sommer 1998).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Surgery

Two monkeys (Macaca mulatta) were surgically prepared in aseptic conditions using isofluorothane anesthesia. We inserted eye coils subconjunctivally (Judge et al. 1980), drilled and tapped holes in the skull for the placement of screws, and trephined holes for accessing the FEF (this hole was centered at A25, L20 for monkey H and at A23, L18 for monkey C) and the SC (this hole was centered on the midline and angled 42° back from vertical so that electrode penetrations would approach the SC approximately orthogonally to its surface, for both monkeys). Recording chambers were placed over the trephinations, plugs were attached for accessing eye coil leads, and dental acrylic was applied so that the cylinders, eye coil plugs, and a post for head restraint all were held securely, and so that the entire implant was connected firmly to the skull via the screws. To permit magnetic resonance images (MRIs) of the monkeys, screws were titanium and the chambers and head holder were plastic. Monkeys received analgesics and antibiotics postoperatively. All procedures were approved by the Institute Animal Care and Use Committee and complied with Public Health Service Policy on the humane care and use of laboratory animals.

Antidromic stimulation

The FEF and the SC (Fig. 1A) were located physiologically. To find the FEF, we explored the cortex rostral to the arcuate sulcus (the sulcus was visible through the dura mater during surgery). We defined the mediolateral range of FEF in our monkeys as those sites just rostral to the arcuate sulcus where penetrations yielded saccade-related corticotectal neurons (Visuomovement or Movement Neurons as defined in RESULTS; see Fig. 7, A and D). We verified that these sites were within the FEF of Bruce et al. (1985) by electrically evoking saccades from these sites, or immediately adjacent sites, at low current threshold (<50 µA using 70-ms trains of biphasic pulses, 0.25 ms/phase, at 350 Hz). Threshold was defined as current that evoked saccades on 50% of trials. Stimulation began 200 ms after disappearance of a foveated light while the monkey's task was to maintain fixation on the blank screen (during the fixation task described in Behavioral procedures). The SC was identified physiologically by its characteristic lamination, with visually responsive neurons located dorsal to neurons discharging before and during saccade generation (e.g., Sparks and Hartwich-Young 1989), and by its topographic map of stimulation-evoked saccades (Robinson 1972).

Stimulating electrodes were implanted in rostral and caudal SC (Fig. 1B). They were used for 1-3 mo and then replaced at slightly different locations once they began to fail (i.e., when their ability to conduct current degraded). Electrodes in rostral SC all were within the 3° amplitude representation on the SC map (Robinson 1972), and those in caudal SC all were between the 7 and 20° amplitude representation. Electrodes were placed near the representation of the horizontal meridian. We chose the depth of stimulation according to the following criteria. Rostral SC electrode tips were placed at depths in the SC where visual receptive fields were foveal, where activity continued while the monkey fixated a spot that blinked off for several hundred milliseconds, and where we could delay ipsiversive visually guided saccades, evoke small contraversive saccades, or both, using <20 µA. In practice, these criteria led to placement of rostral electrode tips 2.4 ± 0.6 (SD) mm below the SC surface (range 1.6-3.3 mm), where saccades were inhibited or <3° amplitude saccades were evoked at thresholds of 9.5 ± 5.1 µA (range 3-14 µA). When placing caudal SC electrode tips, we chose sites where large contraversive saccades were evoked using <10 µA and where the dominant neuronal discharge was presaccadic burst activity. These criteria led to placement of caudal electrode tips 1.8 ± 0.5 mm below the SC surface (range 1.5-2.7 mm) where saccades of amplitude 12 ± 6° (range 720°) were evoked using current thresholds of 4.3 ± 1.9 µA (range 2-7 µA). Distances between the rostral and caudal electrode tips in the SC, as estimated from inter-electrode distances in the grid, were 1.8 ± 0.4 mm (range 1.4-2.2 mm). The characteristics of the rostral SC stimulation sites indicated that they were in the intermediate gray layer zone where "fixation" neurons are found (Munoz and Wurtz 1993a), and the characteristics of caudal sites indicated that they were in the intermediate gray layer region where saccade-related "burst" and "buildup" neurons are found (Munoz and Wurtz 1995).

Once the SC stimulating electrodes were cemented into place (by applying epoxy to bind together the electrode shafts, guide tubes, grid, and implanted cylinder), near daily recording sessions commenced. During a penetration through the FEF, we first isolated a neuron and then attempted to activate it from the rostral or the caudal SC using a single biphasic pulse of current (cathodal-anodal, 0.15 ms and 600 µA per phase). Once an activated neuron was found we lowered the current to find the threshold for activating it from each electrode (threshold defined as current for which activation occurred 50% of the time). Antidromic activation was confirmed for all neurons in this study using the collision test (Fig. 1C) (see Lemon 1984 for review of the collision test). Activation latency was measured from the start of the stimulation artifact until the start of the evoked action potential. Conduction velocity was calculated using the formula D/(L  u), where D is the axon distance from FEF to SC (estimated to be 40.5 mm) (Segraves and Goldberg 1987), L is the antidromic activation latency, and u is the utilization time, i.e., the time it takes for electrical stimulation to elicit an action potential, estimated to be 0.2 ms (reviewed by Lemon 1984).

We used tungsten microelectrodes (Frederick Haer) for recording and stimulating (impedances were 300-1,200 k and 90-110 k at 1,000 Hz, respectively). Electrodes were inserted through a guide tube held by a grid (Crist et al. 1988) that was attached within the implanted chamber. Use of the grid was especially important for recordings, because it permitted systematic exploration of the cortex within and surrounding the FEF. For the FEF, guide tubes were made as short as possible, so that they barely passed through the dura (to avoid damaging the cortex). For the SC, guide tubes were inserted so their ends were ~4 mm above the SC surface.

Behavioral procedures

During experimental sessions, a monkey sat in a primate chair centered within magnetic fields used for detecting eye position (Robinson 1963). Visual stimuli (0.3 × 0.3° blue or red spots on a dark background) were back-projected onto a tangent screen 57 cm in front of the monkey using an LCD projector (Sharp model 850). Ambient room light was dim. Coverage of the visual field was 80° horizontally and 60° vertically, centered on a point straight ahead from the midpoint of the monkey's eyes. A personal computer controlled the presentation of visual stimuli, and this computer in turn was controlled by a personal computer running a QNX-based real time experimentation data acquisition system (REX) (Hays et al. 1982). A third personal computer ran in-house software that served as a digital oscilloscope (50 kHz), allowing us to separate action potential waveforms using time and amplitude windows. The REX system recorded at 1 kHz the eye position, the occurrence of action potentials, and the timing of task events. Visual stimuli actually appeared on the screen an average of 12 ms after the times noted in data files, as reported previously (Basso and Wurtz 1998; Eifuku and Wurtz 1998), and we accounted for this by shifting these times in the data files later by 12 ms.

Monkeys performed three tasks that allowed us to characterize the signals carried by neurons: the delayed-saccade task, the gap task, and the fixation task.

DELAYED-SACCADE TASK. At the beginning of the delayed-saccade task (Fig. 2A), a monkey was required to fixate a central spot of light for a random duration (500-800 ms), and then a target appeared in the periphery. In visual trials (Vis. in Fig. 2A), the target remained lit for the rest of the trial; in memory trials (Mem. in Fig. 2A), the target disappeared after 100 ms. We randomly interleaved these two versions of the delayed-saccade task to help us identify tonic visual responses, as described in RESULTS. After a random delay period of 500-1,000 ms, the fixation light disappeared, providing the cue to make a saccade to the target's location. When analyzing results, the mean firing rates during five periods (Analysis Epochs in Fig. 2A) were quantified and compared (statistical analysis is described in STATISTICAL ANALYSIS OF NEURONAL SIGNALS). The Baseline epoch was 500-200 ms before target onset, the Visual epoch was 50-150 ms after target onset, the Delay Period epoch was 300-0 ms before the cue to move, the Presaccadic epoch was 50-0 ms before saccade initiation, and the Postsaccadic epoch was 50-150 ms after saccade termination.

View larger version (51K): [in this window] [in a new window]   Fig. 2. Delayed-saccade task and some of the signals and neuron types identified using it. A: schematic of the task is shown. After the monkey fixated a light (Fix Spot) for a random duration, a target (Targ) appeared in the periphery. In Visual Trials (Vis.), the target remained lit for the rest of the trial; in Memory Trials (Mem.), the target disappeared after 100 ms. The monkey was allowed to make a saccade to the target location (Eye Pos.) only after a randomized delay period elapsed. The cue to move was disappearance of the Fix Spot. Firing rates during 5 periods (Analysis Epochs) were statistically compared: Base, Baseline epoch; Vis, Visual epoch; Del, Delay Period epoch; Pr, Presaccadic epoch; Po, Postsaccadic epoch. Time scale in all panels is 100 ms per tic. B: activity of a neuron carrying a delay signal. Only results from memory trials are shown, because only these trials were used to identify delay signals. Neuronal discharges in successive trials are represented by rasters, and the average firing rate is shown using a spike density function formed by convolving the individual rasters with a Gaussian of  = 10 ms and summing the resultant curves. Eye traces (Eh, horizontal component; Ev, vertical component) are shown below the rasters. Eye position and firing rate scales are shown at right. Rasters and spike density functions are aligned with respect to target onset (Target On), fixation spot offset (Cue to Move), and saccade initiation (Saccade). C: activity of 2 Visual Neurons. Top: a neuron carrying only a phasic peripheral visual signal. Bottom: a neuron carrying a strong tonic peripheral visual signal. Spike density functions from visual trials (thick lines) and memory trials (thin lines) are superimposed and rasters are omitted for clarity. Eye movements shown are from visual trials. D and E: activity of a Movement Neuron and a Visuomovement Neuron, respectively. In D and E, data from visual trials are shown above and data from memory trials are shown below.

GAP TASK. In the gap task (Fig. 3A), the monkey first fixated a spot for a random duration of 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; the monkey could then look at the target with no further imposed delay. Firing rate during a Gap Period epoch, from 50 ms before target onset to 50 ms after, was compared with firing rate during a Baseline epoch 500-200 ms before start of the gap. We inspected all the data to ensure that the Gap Period epoch did not overlap with periods of phasic peripheral visual activity occurring after target onset.

View larger version (27K): [in this window] [in a new window]   Fig. 3. Gap task and the signals identified using it. A: schematic of the task. After initial foveation of a fixation light (Fix Spot) for a random duration, the light was extinguished and fixation had to be maintained on the blank screen. After a gap period of 200 ms, a Target appeared in the periphery. The monkey was allowed to make a saccade (Eye Pos.) as soon as the Target appeared. Analysis Epochs shown at bottom: Base, Baseline epoch; Gap, Gap Period epoch. Time scale in all panels is 100 ms per tic. B: activity of a neuron carrying a gap increase signal. Shown are the rasters, eye movement data, and spike density function (drawn with bold line) for trials in which the target was presented in the movement field. Superimposed is the spike density function (drawn with thin line) summarizing neuronal activity when the target was presented, in randomized trials, out of the movement field; note that gap increase signals are present in both trial types, but the subsequent presaccadic burst occurs only for saccades made into the movement field. C: activity of a neuron carrying a gap decrease signal.  = 10 ms for all the spike density functions.

FIXATION TASK. In the fixation task (Fig. 4A), the monkey foveated a spot for a random duration of 500-1,000 ms, and then the spot disappeared for a random duration of 400-600 ms while the monkey had to maintain fixation. Then the spot reappeared at the same place for an additional random duration of 500-1,000 ms. Five analysis epochs were defined. The Baseline epoch was during the intertrial period, 300-0 ms before fixation spot onset, the First Fixation epoch was 100-300 ms after start of fixation, the Foveal Visual Offset epoch was 100-300 ms after disappearance of the fixation spot, the Second Fixation Epoch was 300-0 ms before fixation spot reappearance, and the Foveal Visual Onset epoch was 100-300 ms after fixation spot reappearance. Firing rates during all five epochs were quantified and compared with each other statistically, as described below, but only a few of the resulting comparisons were found to be useful for characterizing the neurons, as described in RESULTS.

View larger version (35K): [in this window] [in a new window]   Fig. 4. Fixation task and the signals and neuron types identified using it. A: schematic of the task. A fixation spot (Fix Spot) appeared and the monkey foveated it (Eye Pos.) for a random duration. The fixation spot disappeared for 400-600 ms and then reappeared. Throughout the trial the monkey had to maintain steady fixation. Analysis Epochs shown at bottom: Base, Baseline epoch; Fix1, First Fixation epoch; VisOff, Foveal Visual Offset epoch; Fix2, Second Fixation epoch; VisOn, Foveal Visual Onset epoch. Time scale in all panels is 100 ms per tic. B: activity of a neuron carrying a fixation-related signal. C: activity of a Pure Fixation Neuron, a neuron that carried a fixation-related signal and also fired fairly steadily throughout the entire fixation trial. D: activity of a Pure Foveal Visual Neuron. Examples of eye positions recorded during the fixation task are shown for the data in D.  = 10 ms for all the spike density functions.

GENERAL TESTING PROCEDURE. The basic protocol for characterizing the signals carried by each antidromically activated neuron was as follows. First, we determined the extent of the neuron's visual receptive field or movement field by having the monkey perform the delayed-saccade or gap task while the position of the target was varied throughout the testing space until a location was found that evoked maximal visual- or movement-related firing (as judged by on-line inspection of action potential rasters and histograms). This location was termed the best location for the neuron. Then, during formal testing using the delayed-saccade and gap task, the visual target was presented at this best location (for a minority of neurons, visual- and movement-related activity did not vary throughout the testing space, i.e., there was no best location, so during formal testing of these neurons we arbitrarily chose to present the target contralaterally, 10° eccentric on the horizontal meridian). Often a separate block of gap task trials also was run in which the target location was randomized twofold (described in RESULTS), to see whether gap-related signals depended on knowledge of eventual target location and to help facilitate comparison of our results with those of Dias and Bruce (1994). If a neuron appeared to change its firing rate at the start of fixation during the delayed-saccade or gap tasks, the monkey was then run on the fixation task to better characterize the foveal-related signals. Eye position tolerance windows around fixation spots were 2 × 2°, and those around target stimuli typically were 5 × 5° (sometimes larger for targets in the far periphery). Correct responses were rewarded with drops of water during experiments (water intake was controlled in the monkey's home cage).

STATISTICAL ANALYSIS OF NEURONAL SIGNALS. For each neuron and each type of task, the data set consisted of mean firing rates during the series of epochs associated with the task events. First, to see whether the neuron's activity varied at all during the task, we ran an ANOVA on the data. If this was significant (P < 0.01), we then performed an all-pairwise multiple comparison test (Student-Newman Keuls or Dunn's) so that we could determine whether firing rates in any two epochs differed from each other (P < 0.05). Specific types of signals (e.g., delay signals) were defined according to comparisons between epochs, as presented in RESULTS.

Estimating cell body and axon termination locations

To estimate the location of a neuron's soma in the FEF, we moved the recording electrode carefully up and down until the action potential voltage was peak-to-peak maximal and then recorded the depth of the electrode tip with respect to the end of the guide tube. Over the course of the study, after numerous penetrations, it became evident that the ends of the guide tubes for monkey C rested on top of the cortex (because 1st neurons typically were encountered 0-500 µm below the end of the guide tube even though the guide tube was barely through the dura), whereas the ends of the guide tubes for monkey H were 1 mm above the cortex. During data analysis, the noted depths of all the FEF neurons were adjusted using this corrective information to estimate how deep the neuron was located with respect to the top of the cortex. To analyze topographies within the FEF, cell body locations in FEF were plotted on a standard two-dimensional map [similar to the practice of plotting SC neurons on a standard map (e.g., Anderson et al. 1998)]. Derivation of the FEF standard map is described in RESULTS.

The rostrocaudal location of axon termination in the SC was estimated by comparing the current threshold, I, for activating each FEF neuron from the rostral versus the caudal SC electrodes. We preferred to use a quantity that increased with increasing ease of activating a neuron; therefore we defined an ability to activate, A, as I¹. As examples, A = 0 meant the neuron could not be activated antidromically from an electrode, A = 2,500 meant that a neuron could be activated, but at relatively high current threshold (400 µA), and A = 100,000 meant that it could be activated very easily (using only 10 µA). To compare the ability to activate a neuron at the rostral and caudal electrodes, we defined an Electrode Preference Index, EPI, using a standard contrast ratio: EPI = (A_caudal  A_rostral)/(A_caudal + A_rostral). Therefore EPI = 1 meant that the neuron was activated only from the caudal electrode, EPI = 1 meant that it was activated only from the rostral electrode, and EPI = 0 meant that it was activated with equal ease from both electrodes. These techniques appeared to provide a good estimate of where a neuron's axon terminated along the rostrocaudal axis of the SC, as reviewed in the DISCUSSION. [Note, if one prefers to think in terms of current threshold, I, rather than its reciprocal, A, then the above equation can be rearranged to yield EPI = (I_rostral  I_caudal)/(I_rostral + I_caudal).]

Correction for sampling bias

Estimates of signal composition in a population of neurons can suffer from sampling bias due to the preferential recording of larger neurons (Towe and Harding 1970). In terms of quantities that are measurable during in vivo extracellular recordings, neurons with higher conduction velocities are oversampled (conduction velocity is directly related to cell size) (Cullheim 1978; Gasser 1941; Kernell and Zwaagstra 1981). To correct for this sampling bias we used the method described in detail by Humphrey et al. (1978) and applied to primary motor cortex data by Humphrey and Corrie (1978). Their premise is that, "with equivalent transmembrane action potentials, the discharge of a large neuron will generate a greater flow of membrane current, a larger extracellular spike, and a potential field that is above recording noise levels over a greater distance than will a small cell. Thus the larger a neuron, the greater is the distance that it may lie from an exploring electrode before its spike becomes undetectable or too small to observe reliably. Because of this relationship, the effective volume of neural tissue that is `sampled' or `observed' during a given microelectrode penetration is not a constant, but is instead larger when recording extracellularly from large cells than when recording from small cells. In order to estimate the true relative densities of cells of different sizes, it is necessary, therefore to divide observed measures of their relative densities or frequencies within a given sample of units by estimates of their relative, effective recording volumes (V_eff). For example, if N_o() is an experimentally observed distribution of cellular conduction velocities (), then the true or unbiased distribution, N_t(), would be given by N_t() = N_o()/V_eff() where V_eff() is now considered to be an explicit function of axonal conduction velocity, rather than that of the closely related quantity, cell size." (Humphrey and Corrie 1978, p. 234). Therefore the key to correcting for sampling bias due to cell size variation is to find V_eff(). Humphrey et al. (1978) derived an equation for pyramidal neurons that described the extracellular spike amplitude as a function of various neuronal characteristics (e.g., dendritic geometry), extracellular conductivity, and distance from the cell body's center to the recording site. Empirical measurements verified that the equation was accurate (Humphrey et al. 1978), and therefore it was used to evaluate V_eff. Further models and physiological experiments revealed a relation between extracellular spike amplitude and conduction velocity, and by combining this result with the equation noted above, Humphrey et al. (1978) concluded that V_eff() = k^3/2, where k is a constant.

We used the above expression for V_eff() to correct for sampling bias. First, the experimentally observed distribution of conduction velocities for our sample of antidromically activated neurons was expressed as a histogram N_o(_i), representing the number of neurons, N_o, that had conduction velocities within each bin _i (18 bins of 5 m/s width were used, spanning 0 to 90 m/s; see Fig. 5). The estimated true distribution, therefore was N_t(_i) = N_o(_i)/k_i^3/2. The value of k is unknown, but it can be canceled out by converting the numerical histogram, N_t, into a histogram of proportions, P, of neurons that are in each conduction velocity bin
The value of each _i was set to the midpoint of each velocity bin.

View larger version (27K): [in this window] [in a new window]   Fig. 5. Conduction velocity histograms for FEF corticotectal neurons. Graphed are the percentages of neurons in each population that had conduction velocities represented along the abscissa. To facilitate comparison between histogram pairs, in each panel one histogram is inverted and placed beneath the other. A: conduction velocity distributions for the 88 neurons tested on behavioral tasks and for the entire population of 138 neurons [note that a histogram and statistical summary of the antidromic latencies for most of these neurons (n = 133) was previously published (Sommer and Wurtz 1998a)]. B: conduction velocity distributions for Movement Neurons and for other neurons. C: conduction velocity distributions for all neurons carrying a presaccadic burst signal and for other neurons.

In summary, this corrected distribution estimates the actual proportion of neurons in the underlying population that have various conduction velocities. The major assumption of the method is that the neurons are pyramidal such that their effective recording volumes (V_eff) are cylindrical, aligned with the apical dendrite. Although FEF corticotectal neuron morphology is not known in detail, most neurons in FEF layer V appear to be pyramidal (e.g., Stanton et al. 1989; Walker 1940), and studies of partially filled FEF corticotectal neurons suggest that most of them are pyramidal (Fries 1984; Leichnetz et al. 1981).

Using the corrected distribution of conduction velocities, the corrected proportion of each functionally defined class of neuron can then be calculated. To illustrate this procedure, consider the simple case in which there are two conduction velocity bins, ₁ and ₂, and three neurons in the recorded sample. Assume that one neuron fell in bin ₁ and two neurons fell in bin ₂. After correction, assume that the corrected distribution indicates that 80% of neurons in the underlying population actually fall into bin ₁ and 20% fall into bin ₂. Now, assume that only one of the recorded neurons is of cell type X (e.g., the class of neurons carrying delay signals) and that this neuron fell into bin ₂. In the experimentally observed data, therefore cell type X made up 50% of the cells in bin ₂, or 33.3% of the entire sample. After correction, cell type X still accounts for 50% of the data that fall into bin ₂, but now it has been calculated that cells with conduction velocity in bin ₂ actually make up only 20% of the underlying population. Therefore the corrected proportion of cell type X in the population is 50% of 20%, or 10%. This procedure is easily extended to cases where there are arbitrary numbers of cell classes or conduction velocity bins. The only assumption is that neurons in the same velocity bin (i.e., neurons of similar sizes) are sampled with equal likelihood; this assumption seems valid because sampling bias appears to be caused primarily by cell size variation (Towe and Harding 1970).

Anatomic verification of FEF and SC sites

MRIs (1.5 Tesla) were taken of the brain in both monkeys, and frontal and parasagittal planes were inspected at 1-mm intervals. A few days before taking the MRI we implanted electrodes with their tips at locations in the FEF that had yielded many corticotectal neurons so that we could visualize these locations. For monkey H, over a series of days near the end of the experiment, we made marking lesions at the sites of antidromically activated neurons in the FEF and through the tips of the stimulating electrodes in the SC, using DC of 10 µA for 20 s (for FEF) and 20 µA for 60 s (for SC). About 1 wk later we overdosed the monkey with pentobarbital sodium, inserted several guide pins into the brain through reference holes in the FEF chamber grid, and perfused the animal transcardially with 10% neutral buffered Formalin. The guide pins, fixed in place, were then used to direct blocking cuts of the FEF. We sectioned the FEF block in a plane normal to the cortical surface and parallel to the principal sulcus in 30-µm sections. The SC was sectioned coronally every 30 µm. For both FEF and SC, in every three consecutive sections, two were stained for cell bodies (thionin) and one for myelin (modified protocol of Gallyas 1979) to aid in recovery of marking lesions. The other monkey is being used for further experiments.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We isolated 138 neurons in the FEF that were activated antidromically from the SC (monkey H, n = 82; monkey C, n = 56). Of these, 88 were analyzed using our behavioral tasks. Of the remaining 50 neurons, 4 were not modulated by any of our tasks, and 46 were lost before they could be fully tested on the tasks.

Composition of signals sent from FEF to SC

We analyzed neuronal discharges that occurred in relation to visual stimulation, saccade generation, fixation, delay periods, and gap periods. These discharges were the "signals" carried by neurons. Neurons that carried specific combinations of signals were grouped into "neuron types." For example, neurons discharging just after visual stimulation and also just before saccade generation were termed Visuomovement Neurons (after the nomenclature of Bruce and Goldberg 1985). Summary data for each signal type and each neuron type are listed in Tables 1 and 2, respectively.

SIGNALS RELATED TO DELAY PERIODS, PERIPHERAL VISUAL STIMULATION, OR MOVEMENT. We used the delayed-saccade task to detect signals related to delay periods, peripheral visual stimulation, and saccade generation. A neuron had a delay signal if, in memory trials, its firing rate during the delay period (Del epoch, Fig. 2A) differed from its baseline firing rate (Base epoch). Only memory trials were considered because, in the delay period and baseline epochs of these trials, visual stimulation (fixation spot on, target off) and motor behavior (steady fixation) were identical. Firing rate differences between the epochs therefore may primarily reflect cognitive processes (e.g., see Basso 1998; Fuster 1973). Of the 88 neurons tested, 33 had a delay signal (Fig. 2B). For 21 of the neurons the delay period firing rate was higher than baseline, and for the remaining 12 neurons it was lower than baseline [both elevated and suppressed delay signals are thought to play a role in cognitive operations (e.g., Funahashi et al. 1989; Fuster et al. 1982; Niki 1974)]. It has been proposed that delay signals related to a restricted range of target locations or saccade vectors may reflect spatially restricted attention, memory, or planning (e.g., Niki 1974; Niki and Watanabe 1976). Thus we tested most of the neurons that had delay signals with multiple target locations and found that 90% (26/29) did exhibit spatially restricted delay signals (spatial regions associated with the delay signals were contralateral for 80% of neurons, ipsilateral for 8%, and on the vertical meridian for 12%).

GAP-RELATED SIGNALS. We used the gap task to identify signals during gap periods that may be associated with cognitive processes such as fixation disengagement (e.g., Dias and Bruce 1994). A neuron had a gap increase signal if it increased its firing rate during the gap period (Gap epoch, Fig. 3A) compared with baseline (Base epoch). Of the 88 neurons tested, 34 exhibited this type of signal (Fig. 3B). We note three important characteristics of these neurons. First, many neurons with gap increase signals also exhibited peripheral visual signals (62%, 21/34), presaccadic burst signals (65%, 22/34), or delay signals (44%, 15/34) when tested with the delayed-saccade task. In terms of neuron types, gap increase signals were carried by 41% (7/17) of Visual Neurons, by 61% (14/23) of Visuomovement Neurons, and by 44% (8/18) of Movement Neurons. Second, gap increase signals usually occurred even if target location was randomized. We tested 25 of the neurons with gap increase signals in a separate block of gap task trials in which the target appeared either within the visual or movement field of the neuron or, randomly on 50% of the trials, at the same eccentricity but 180° opposite in direction. Most neurons (88%, 22/25) still exhibited a gap increase signal in this block of trials. Third, when the target appeared outside the visual or movement field, elevated gap discharges quickly ceased after the target appeared (Fig. 3B). With respect to all of these discharge characteristics, our neurons with gap increase signals appear to be very similar to the FEF neurons described by Dias and Bruce (1994).

FOVEAL-RELATED SIGNALS. While testing neurons using the delayed-saccade and gap tasks, we noticed that 39% (34/88) changed their firing rate at the start of fixation. This foveal-related activity was quantified using the fixation task (Fig. 4A), and two kinds of signals were defined. A neuron had a fixation-related signal (Fig. 4B) if its activity during the late blink period (Fix2 epoch, Fig. 4A) was different (greater or less than) baseline (Base epoch). This activity was not a foveal visual response for two reasons: first, it was not a foveal off-response because the change in activity persisted for hundreds of milliseconds after fixation spot disappearance; second, it was not a response to the diffuse light on the screen because activity during the Fix2 epoch was different from baseline activity even though the fovea was illuminated with the same diffuse light during both periods. The only difference between the Base epoch and the Fix2 epoch was the requirement to maintain fixation at the center of the screen during the latter, and therefore we interpreted the type of activity shown in Fig. 4B as related to the motor act of fixating. On the other hand, some neurons did carry signals clearly related to foveal visual stimulation. We considered a neuron to have a foveal visual signal (Fig. 4D) if it increased its activity just after fixation spot reappearance (VisOn epoch) compared both to baseline and to the late blink period. Twenty-four neurons had fixation-related or foveal visual signals (20 had fixation-related signals, 7 had foveal visual signals, and 3 had both). The remaining 10 neurons exhibited discharges during the fixation task that resisted simple classification.

CONDUCTION VELOCITY, CELL SIZE, AND SAMPLING BIAS. Conduction velocity distributions for the 88 neurons tested on behavioral tasks and for all 138 neurons are shown in Fig. 5A (medians 30 and 26 m/s, respectively; not significantly different). We compared the conduction velocities of neurons carrying each type of signal or belonging to each defined neuron type to the conduction velocities of all the other neurons analyzed with behavioral tasks. The only significant results were that Movement Neurons had higher conduction velocities than other neurons (Fig. 5B; medians 43 vs. 25 m/s, P = 0.007) and that the general class of all neurons carrying a presaccadic burst signal had higher conduction velocities than other neurons (Fig. 5C; medians 37 vs. 21 m/s, P = 0.002). These two categories of neurons, therefore probably had larger axons and cell bodies than the other neurons (Cullheim 1978; Gasser 1941; Kernell and Zwaagstra 1981).

View larger version (38K): [in this window] [in a new window]   Fig. 6. Correcting for sampling bias using the method of Humphrey et al. (1978). A: distribution of conduction velocities (mean ± SE) for neurons carrying each type of signal, arrayed in decreasing order from left to right. B: observed and corrected percentages of neurons carrying each type of signal (legend at bottom right). C and D: distribution of conduction velocities (C) and observed and corrected percentages (D) of each defined neuron type. For key to abbreviations see legends of Tables 1 and 2.

Topographic organization of signals sent from FEF to SC

DISTRIBUTIONS OF CELL BODIES IN FEF AND AXON TERMINALS IN SC. To analyze the distribution of corticotectal cell bodies, we first constructed a standard map of the FEF. Successful penetration entrance sites in the FEF (i.e., those yielding saccade-related corticotectal neurons) are shown for monkey C in Fig. 7A and for monkey H in Fig. 7D. MRIs verified that penetration trajectories went through the rostral bank of the arcuate sulcus (MRI of monkey C is shown in Fig. 7B). For monkey H this was further confirmed by inspection of marking lesions and electrode tracks in histological sections (not shown). Low-threshold electrical stimulation (<50 µA) within these recording sites or adjacent sites (×, Fig. 7, A and D) evoked saccades at short latency (Fig. 7C). The amplitude of evoked saccades decreased from medial to lateral (Fig. 7C) and also from dorsal to ventral within a penetration (not shown). Our recording sites therefore were in the FEF as classically defined (e.g., Bruce et al. 1985; Robinson and Fuchs 1969). The recording sites tended to form a curve that paralleled the arcuate sulcus (Fig. 7, A and D), undoubtedly because to yield corticotectal neurons the penetration trajectories had to intersect with or follow the contour of layer V (Fries 1984; Leichnetz et al. 1981), which runs parallel to the sulcus. For each monkey, we drew a curve representing the top edge of layer V onto the map of penetrations (Fig. 7, A and D). This curve defined a mediolateral axis (Fig. 7E, top), with zero at the medial edge of the FEF (*, Figs. 7, A and D) and with values increasing toward the more lateral FEF. The location of each recording site was described in relation to the mediolateral axis using orthogonal projection as shown schematically in Fig. 7D. Recording depth below the cortical surface defined a second, dorsoventral axis (Fig. 7E, top), with zero at the cortical surface and with values increasing down through the bank, parallel to the sulcus. Note that the mediolateral and dorsoventral axes used in this report are rotated from the conventional stereotaxic axes (Fig. 7E, bottom).

View larger version (46K): [in this window] [in a new window]   Fig. 7. Recording sites in the FEF. A: recording chamber for monkey C. Entrance sites of penetrations that yielded saccade-related corticotectal neurons are shown with dots. Curved line between the penetration sites is the estimated top edge of layer V. Asterisk shows the medial edge of the functionally defined FEF. Sites where reference electrodes were inserted for MRI are labeled 1 and 2, respectively. ×, sites where saccades were electrically evoked at low threshold (<50 µA; these include sites labeled 1 and 3, discussed further in C). B: examples of coronal plane MRIs for monkey C, showing the recording chamber and grid (top left in each MRI) and reference electrodes (dark lines extending from chamber into the brain). Grid holes were visualized by filling the grid and chamber with betadine ointment ~1 h before taking the MRI. Top: reference electrode at site 1 as labeled in A. Bottom: reference electrode at site 2. C: examples of electrically evoked saccades. In the center, trajectories of saccades evoked from 2 recording sites (labeled 1 and 3 in A) are illustrated. Dotted box shows the fixation window (monkey was trained to keep its eyes still within this window, but stimulation moved the eyes out of it). In the large boxes at left and right, saccades from each site are decomposed into horizontal (Eh) and vertical (Ev) components and plotted against time; bars below the eye traces depict stimulation onset and duration. The mean amplitudes of saccades evoked from each site, and the suprathreshold currents used for evoking these saccades, are noted. Current threshold was 3 µA at site 1 and 20 µA at site 3. Amplitude and time scales at bottom. D: recording chamber for monkey H. The method of projecting penetration entrance sites orthogonally onto the curve of layer V is diagrammed in the dotted-outline box (projected locations are shown as white dots). E: standard map of the FEF. At top, layer V is depicted as a sheet, electrode trajectories are depicted as vertical lines running tangential to this sheet, and recorded cells are depicted as small circles (note, these are not actual data, but are shown only to help illustrate the method). Each cell was mapped in 2 dimensions by describing its position with respect to the mediodorsal axis, which is the top edge of layer V as drawn in A and D, and the dorsoventral axis, which is parallel to the penetration trajectory. Zero on the mediolateral axis is the medial edge of the FEF (* in A and D), and zero on the dorsoventral axis is the top of the cortex. At bottom, the mediolateral and dorsoventral axes are shown in reference to a coronal section of the FEF, which was cropped from the top MRI in B and enlarged. Ai, As, inferior and superior limbs of the arcuate sulcus, respectively; Pr, principal sulcus; A, anterior; P, posterior; M, medial; L, lateral; D, dorsal; V, ventral.

View larger version (32K): [in this window] [in a new window]   Fig. 8. Distribution of cell body locations (circles) in the FEF for each monkey. Note that 131 neurons were plotted, but due to clustering of the recorded neurons many of the symbols are occluded by others. Polygons depict the general area of sampling for each monkey. Explicit diagram of how this graph corresponds to gross anatomy is shown at bottom.

View larger version (27K): [in this window] [in a new window]   Fig. 9. Stimulation sites in the SC. A: examples of rostral and caudal stimulating electrode locations (dots) in the SC with respect to its topographic map (Robinson 1972). Rostral and caudal directions, as well as isoeccentricity and isodirection contours, are labeled. B: depths of the electrode tips for the SC penetration sites of A, shown in coronal sections. Marking lesions (black marks at the middle of circles) were made at the end of the experiment through the rostral (top) and caudal (bottom) electrodes. The approximate dorsal and ventral boundaries of the stratum griseum intermediale (SGI), as determined by examination of successive cell- and myelin-stained sections through the entire rostrocaudal extent of the SC, are shown with curved dotted lines. In the caudal section the marking lesion appears relatively deep and the SGI relatively wide; these are artifacts due to the plane of section being somewhat oblique to the SC surface. In A and B, estimates of the mean effective current spread are shown with circles, as calculated using the equation Distance = (Current/K)0.5 (Tehovnik 1996), where K  381 µA/mm2 (Sommer and Wurtz 1998a). Mean current thresholds for antidromic activation were 208 ± 144 µA from the rostral electrode and 198 ± 122 µA from the caudal electrode, suggesting that we activated neural elements on average within 0.74 ± 0.62 mm from the rostral electrode tip and within 0.72 ± 0.57 mm from the caudal electrode tip. C: distribution of the rostrocaudal extent of axon termination in the SC (quantified with Electrode Preference Index, EPI) for each monkey. EPI histogram for monkey C () is inverted and placed directly below that from monkey H () to permit direct comparison.

DISTRIBUTIONS OF TASK-RELATED SIGNALS IN CORTEX. Using the FEF standard map, we compared the FEF locations of neurons carrying each type of signal or belonging to each defined neuron type to the FEF locations of all the other neurons analyzed with behavioral tasks. Comparisons were made in the mediolateral and the dorsoventral directions (Student's t-test or Mann-Whitney rank sum tests were used as appropriate, and because we tested the data twice, along orthogonal axes, the significance criterion was adjusted to P < 0.05/2 = 0.025). Signals putatively related to cognitive operations (delay and gap increase signals) were carried by neurons located more dorsally in the FEF than other neurons (Fig. 10A; medians 2.7 vs. 3.6 mm along the dorsoventral axis, P < 0.001). This dorsal bias was significant for each of the component signal types, too (i.e., for the neurons carrying a delay signal as well as for those carrying a gap increase signal). Neurons carrying a peripheral visual signal but not a presaccadic burst signal (Visual Neurons) were biased laterally in the FEF compared with other neurons (Fig. 10B; medians 3.5 vs. 2.5 mm along the mediolateral axis,