INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Schlag and Schlag-Rey (1985, 1987) initially defined the supplementary eye field (SEF) of the macaque monkey as a discrete region of dorsomedial frontal cortex where saccadic eye movements are electrically elicited with low currents. SEF lies just anterior to the supplementary motor area (SMA) from which skeletalmotor movements can be elicited (e.g., Woolsey et al. 1952) and medial to the better known frontal eye field (FEF) from which both saccades (e.g., Bruce et al. 1985; Robinson and Fuchs 1969) and smooth pursuit (e.g., MacAvoy et al. 1991) can be elicited.

Schlag and Schlag-Rey (1985, 1987) also reported that neurons in SEF responded to the appearance of visual stimuli and in conjunction with saccadic eye movements. This basic result has been confirmed by their subsequent studies (e.g., Schlag-Rey et al. 1997) and in other laboratories, including ours (Russo and Bruce 1996; Schall 1991a,b). Moreover, several interesting and complex aspects of SEF activity indicating several possible specializations have since been examined (see DISCUSSION). However, some basic response properties of SEF neurons, and the functional relationship between the saccades electrically elicited from SEF and the activity of SEF neurons, remain unknown. This void exists, at least in part, because research has not been restricted to the SEF but rather reflected recordings from a larger zone of dorsomedial frontal cortex and has involved either no microstimulation to confirm that recordings were from SEF or stimulation with large currents that can elicit saccades from well outside SEF. In addition, some basic functional issues were unresolved because it was initially reported by Schlag and Schlag-Rey (1987) that the saccades electrically elicited from SEF were not always "constant vector" in nature but rather often seemed "goal-directed." They hypothesized that SEF served to move the eye to particular orbital positions and thus to code saccades in a craniocentric coordinate system. However, after systematically investigating the orbital dependence of saccades electrically elicited from the low-threshold SEF using arrays of fixation positions and testing FEF using the same methods and subjects, we established that saccades elicited from SEF are basically oculocentric with generally modest orbital position effects that are very similar to what is found in FEF (Bruce 1990; Russo and Bruce 1993). Furthermore we later demonstrated that SEF visual receptive fields are oculocentric, not craniocentric, and that SEF movement fields are oculocentric as well (Russo and Bruce 1996).

Although our conclusion that SEF codes saccades oculocentrically has not yet been unanimously accepted, we set out to further study, in an oculocentric framework, the basic neural mechanisms used by SEF to generate saccadic behavior. We mapped the response fields of SEF neurons, both visual receptive fields and saccadic movement fields, and measured the saccades evoked by activating those neurons and their immediate neighbors via electrical microstimulation. Both response-field mapping and electrical stimulation were performed while the monkeys fixated a centrally located fixation point, and both types of data were analyzed in terms of their polar direction and amplitude. A close correspondence between the neural response fields and the dimensions of elicited saccades would indicate a functional linkage between the discharge of a discrete set of SEF neurons and the generation of specific saccade metrics, whereas a lack of correspondence would suggest that SEF is functionally specialized for nonspatial or other more complex aspects of saccade programming. We also investigated the overall physiological organization of SEF by analyzing the saccade parameters encoded by adjacent neurons and stimulation sites (especially those recorded within the same electrode penetration or at the same cortical locus) and the mapping of saccade direction and amplitude across the tangential dimensions of SEF.

We found that neural activity in SEF during visually guided saccades is composed primarily of a visual component and a movement component and that the response fields of these two activities were strongly correlated with each other as well as with the saccades electrically elicited from the recording site. Thus the representation of visual stimulus location and saccade metrics were aligned. We also found similar representations of saccade direction along different cortical depths during the same electrode penetration, indicating a columnar organization. Although the total distribution of response fields and elicited saccades suggested that each SEF contains a complete representation of all possible saccades into the contralateral visual hemifield, saccades were not topographically organized across SEF in a simple way. Instead the representation of saccades was fairly patchy with hints of systematic shifts in direction more indicative of a hypercolumnar-type organization. We conclude that SEF participates in the transformation of visual stimulus location into saccadic commands via the punctuated activity of particular groups of SEF neurons that in turn project to downstream oculomotor structures in a topographic manner. In this regard, the basic sensorimotor mechanisms of SEF are similar to those of FEF and the superior colliculus. Thus any functional specializations of SEF must operate within the context of this core neurophysiological framework.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Surgical and behavioral protocols were approved by the Institutional Animal Care and Use Committee and complied with United States Public Health Service policy on the humane care and use of laboratory animals.

Single-neuron recording

Three female rhesus monkeys (Macaca mulatta) were prepared for chronic single-neuron recording in three separate aseptic surgical procedures. These three monkeys were the same monkeys used in two previous SEF studies (Russo and Bruce 1993, 1996). During experimental sessions, each monkey sat in a primate chair with its head held stationary by a restraining receptacle fixed to the skull. Eye-position coordinates were obtained with a search coil implanted in one eye (Judge et al. 1980). Neurons were recorded with microelectrodes made from either glass-coated Elgiloy wire (tip exposures, 30-50 µm) or glass-coated platinum/iridium wire (tip exposures, 10-30 µm), which were advanced through the intact dura with a hydraulic microdrive (MO-95, Narishige) mounted on recording chambers. The minimum penetration spacing across the cortical surface was 0.5 mm. However, in cases where multiple penetrations were made at the same microdrive coordinate, the slight curvature of the electrode was used to vary the cortical tissue sampled by rotating the electrode in the microdrive ~90° between experimental sessions. Time-amplitude window discriminators (DIS-1, BAK Electronics) sorted action potentials for sampling by the computer.

Behavioral methods

Visual stimuli were small white spots presented on a 27-in color monitor (CS-2669R, Mitsubishi) located 47 cm from the monkey's eyes and subtending 66 by 44° of visual angle. Four tasks were used to analyze presaccadic activity and map response fields. In all four tasks, each trial began when the monkey achieved and maintained fixation of a solitary spot for 0.5 s. At the end of each correctly performed trial, all remaining visual stimuli were extinguished and the monkey was rewarded with ~0.2-ml of dilute fruit drink.

VISUAL-SACCADE TASK. The appearance of a peripheral visual stimulus coincided with the disappearance of the original fixation target, and the monkey was required to saccade directly to the new target. This task was the simplest way to determine if neurons had saccade-related activity, and an interactive version, wherein the experimenter used a joystick to re-position the peripheral target location between trials, was often the first task used to test each neuron.

VISUAL-PROBE TASK. A visual stimulus was presented in the periphery, but in this task, the fixation target remained on and the monkey was rewarded for simply continuing to fixate it. Conversely the trial was terminated if the monkey incorrectly made a saccade away from the original fixation target. This task was used to map neurons with purely visual responses.

DEFERRED-SACCADE TASK. Shortly after the monkey foveated the fixation target, a peripheral target was presented while the fixation target remained on. The monkey was required to saccade to this second target but only after the original fixation target disappeared, usually 0.5-1.0 s after the peripheral target's appearance. By temporally separating the appearance of the peripheral target from the signal to saccade to it, this task dissociated activity related to the initial presentation of the visual stimulus from activity related to the execution of the saccadic eye movement even though the saccade was visually guided.

MEMORY-SACCADE TASK. The monkey was presented with a brief (0.5 s) visual target in the periphery while foveating a continuously illuminated fixation target. After 0.5-1.0 s, the fixation target disappeared, signaling the monkey to saccade to the location where the peripheral target had previously appeared. This task best dissociated visual activity from movement activity because, in addition to temporally separating the presentation of the stimulus and the execution of the saccade, there was no overt target present when the saccade was made.

Microstimulation

After studying a neuron, we tested for electrically elicited saccades by stimulating through the recording microelectrode before advancing it. The stimulation parameters used were the same or similar to the stimulation parameters used in other studies of FEF and SEF. Stimulation consisted of 70-ms trains of 350-Hz biphasic (negative-positive) shocks (thus ~24 shocks per train) with duration of 0.2 ms per phase. Stimulation was applied during fixation of a target at or near the center of the screen, and the threshold of a cortical site was defined as the magnitude of negative-going current necessary to elicit saccadic eye movements on ~50% of trials. Threshold estimation during fixation is quite conservative relative to thresholds measured outside a formal task because attentive fixation raises the threshold for electrically eliciting saccadic eye movements from the FEF (Goldberg et al. 1986) and elsewhere. Although we sought recording sites with low (50 µA) thresholds, sites with slightly higher thresholds (100 µA) were included in our final analysis if robust presaccadic activity was recorded there and low thresholds for eliciting saccades were eventually obtained, either in the same penetration when the electrode tip was advanced into the deeper cortical layers or in other penetrations at the same coordinates or coordinates not more than 1 mm distant. Sites requiring currents 100 µA to obtain elicited saccades were not included in our population summaries even if they were located within the SEF as determined by the cortical boundary defined by the set of low-threshold sites.

Data analysis

Neural discharge rates were computed by estimating the onset and offset of the averaged response using the inflection points of cumulative histograms aligned to either cue onset (visual activity) or saccade beginning (movement activity) and then extracting the trial-by-trial spike rates during this period. The start and end of elicited saccades were found by the computer using an algorithm based on eye velocity.

The optimal polar direction of neural activity was estimated using an array of visual cues all having the same eccentricity but systematically varying in polar direction. The spike rates for each cue direction were fit to the Gaussian function
where f() was discharge frequency,  was stimulus or saccade direction,  estimates the neuron's optimal direction, is an index of the neuron's tuning with respect to direction, B estimates its baseline rate, and R estimates its peak response magnitude. These parameter estimates and their standard error (e.g.,  ± SE) were obtained using the LEASTSQ function in MATLAB (The MathWorks). The optimal direction is designated as _v when visual activity was fit and as _m when movement activity was fit. The visual and movement activities of neurons having both activities were analyzed separately.

The angular-angular correlation (r_aa) between the optimal direction of neural activity (_v or _m) and the median elicited saccade direction obtained at the neuron's site (_e) were computed using the formula of Fisher and Lee (Fisher 1993, p. 151). For all statistics and plots, polar directions obtained from left cerebral hemispheres were transformed into a right-hemisphere convention by the formula ' = 180 -  so that all contralateral directions lie between 90 and 270° regardless of hemisphere.

Optimal eccentricity (i.e., polar amplitude, radius, or distance) of neural activity was estimated using an array of visual cues all having the same polar direction but systematically varying in eccentricity. These data were fit to the Gaussian function
where f(P) is discharge frequency, P is the stimulus or saccade eccentricity,  estimates the neuron's optimal eccentricity, is a index (dimensionless) of the neuron's tuning with respect to eccentricity, B estimates the baseline rate, and R estimates the peak response magnitude. The natural log transform was used because eccentricity tuning appeared to be logarithmically scaled, similar to what had been found in FEF (Bruce and Goldberg 1985). The optimal eccentricity is designated _v when visual activity was fit and as _m when movement activity was fit. In all figures, eccentricity was plotted using a logarithmic scale to maintain sensitivity in the small saccade range.

For some neurons, the formal testing of optimal direction or eccentricity with uniformly spaced arrays of visual cues could not be completed (usually because the neuron was lost or the monkey stopped working before the neuron could be formally tested). In some of these cases, we successfully estimated their optimal direction and eccentricity by fitting the neural activity recorded during our preliminary test that used an interactive joystick to re-position the visual cue between trials. We fit these data to the general Gaussian function of direction and eccentricity
where f(, P) is discharge frequency,  is the stimulus or saccade direction of each trial, P is the stimulus or saccade eccentricity of each trial,  estimates the neuron's optimal direction,  estimates the neuron's optimal eccentricity, B estimates its baseline rate, R estimates its peak response magnitude, is an index of the neuron's tuning with respect to direction, and is a index (dimensionless) of the neuron's tuning with respect to eccentricity.

For the purpose of analyzing the uniformity of a set of directions represented at nearby SEF sites, the mean vector length r (Batschelet 1985, p. 10) was used as an index of "directional-concentration," the formula being
where _i is the ith member of a group of n saccade directions.

Histology

At selected recording sites, electrolytic lesions were made by passing 20 µA of negative current through the electrode for 30 s. Other sites were marked with iron by passing 10 µA of positive current through Elgiloy electrodes for 3 min.

Monkeys AB and SY were deeply anesthetized with pentobarbital sodium and perfused transcardially with saline, followed by 10% formalin in 0.1 M phosphate buffer and a sucrose series. Monkey HK died unexpectedly and could not be perfused. Instead, its brain was fixed by immersion for 7 days in a mixture of 1.25% glutaraldehyde and 1% paraformaldehyde in 0.1 M phosphate buffer followed by 3 days in the same fixative with 30% sucrose. All brains were photographed, blocked in the coronal plane, and sectioned at 50 µm on a freezing microtome. Every other section through the region with electrode penetrations were reacted with ferrocyanide (Perl's reaction) for visibility of the iron deposits and counter stained with neutral red. Individual recording and stimulation sites that had been marked with deposits or lesions were identified.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Location of low-threshold SEF

Both hemispheres of two monkeys (AB and SY) and one hemisphere of one monkey (HK) were studied. For each of these five hemispheres, the low-threshold SEF was defined as the contiguous cortical area whose boundary was not more than 1 mm from an electrode penetration coordinate containing at least one site with a low (50 µA) threshold for eliciting saccadic eye movements. Using this criteria, the low-threshold SEF was typically found to be located in a small region (10-15 mm²) on the dorsomedial convexity of the frontal lobe with its center ~5 mm anterior to the most posterior level of the arcuate sulcus and 2-3 mm lateral to the lip of the longitudinal fissure. Figure 1 shows the location of electrode penetrations in the right hemisphere of monkey AB using different symbols to denote the lowest threshold found at each penetration coordinate. An example of one electrically elicited saccade from one low-threshold SEF site is also shown along with its histological confirmation. We rarely saw evoked movements other than saccades in the low-threshold SEF; the only exceptions were four electrode penetrations where both evoked saccades and pinna movements were observed. Most electrically elicited pinna movements were evoked at sites posterior and lateral to SEF, whereas skeletal movements were evoked several millimeters posterior to SEF. We did not see any elicited smooth eye movements, such as are elicited from the depths of the arcuate sulcus (MacAvoy et al. 1991).

View larger version (36K): [in this window] [in a new window]   Fig. 1. Location of supplementary eye field (SEF). Left: dorsal view of the right frontal lobe of monkey AB showing the thresholds for electrically eliciting saccades at different cortical locations in the dorsomedial region of the frontal lobe. - - -, anterior-posterior location of the histological section shown to the right. Top right: coronal section through the right frontal lobe of monkey AB showing the reaction product from iron deposited at 1 SEF stimulation site (AB267) with a 40-µA threshold. Bottom right: saccades electrically elicited from the stimulation site located at the iron deposit shown above. Saccades were elicited with 50 µA during fixation of a centrally located target. The time course of a single representative trial is shown on the left, and the 2-dimensional trajectories of all 6 trials are shown on the right. The median saccade direction, amplitude, and latency were 257°, 16.4°, and 48 ms, respectively. Starting eye position has been aligned to facilitate comparison of trajectories and end points in this and subsequent 2-dimensional plots of eye position. Stim, time of electrical stimulation; H, horizontal eye position; V, vertical eye position; ||, absolute eye velocity.

The low-threshold saccadic SEF is the subject of this report. Our study is based on elicited saccades from 109 stimulation sites and 160 presaccadic neurons, all located within the low threshold SEF as defined in the preceding text. We first present summaries of the stimulation and neural activity data separately, then present data showing how they are related to each other, and finally show how they are organized across SEF.

Electrically elicited saccades

Figure 2 summarizes our elicited-saccade database pooled across the five hemispheres we studied. Saccades were electrically elicited during attentive fixation of a small spot at the center of the monitor (see METHODS). Thresholds ranged from 10-90 µA (Fig. 2, top left), with a median threshold of 40 µA. Elicited saccade data from stimulation sites with thresholds >50 µA but <100 µA were included in our analyses if they were located within the boundary of low-threshold SEF sites. Neurons at these slightly higher threshold sites usually exhibited robust presaccadic activity similar to that found at low-threshold sites. We would expect that these thresholds would be significantly lower if stimulation had been tested while the monkey alertly looked about without any overt fixation targets; however, we did little testing of this type.

View larger version (30K): [in this window] [in a new window]   Fig. 2. Thresholds, test currents, latencies, and median elicited saccades from the 109 SEF stimulation sites. Top left: thresholds for eliciting saccades at all SEF stimulation sites. Middle left: actual currents used to collect elicited saccade data, typically 10 µA above threshold. Bottom left: median latencies of elicited saccades from each site. Top right: 2-dimensional plots of median elicited saccade vectors from each site. Elicited saccades from left-hemisphere sites are transformed to a right-hemisphere convention (see METHODS). Notice that elicited saccades are distributed throughout the contralateral visual hemifield, but none were ipsilateral. Bottom right: distributions of direction and amplitude components of elicited saccades.

The median cortical depth of the 109 stimulation sites was 1.85 mm below the apparent start of neural activity. Only 7 of the 109 sites was at a depth <0.8 mm as we did not find low thresholds (or robust presaccadic activity) until ~1 mm or more below the apparent entry of the electrode into the cortex.

Because threshold currents elicited saccades on only ~50% of trials, we usually increased the current ~10 µA after determining each site's threshold to quickly obtain representative sets of elicited saccades at fixed currents. These testing currents ranged from 15 to 100 µA, with a median of 50 µA (Fig. 2, middle left).

The latency of electrically elicited saccades, defined as the time from the start of the stimulation train to the start of the saccadic movement, was generally very short (Fig. 2, bottom left). The median latency was 50 ms, with 88% of the stimulation sites having a median latency between 36 and 60 ms. However, the upper tail of the latency distribution is long, and for eight sites, the elicited saccade latency was even greater than the duration of the stimulation train that we used (70 ms). We carefully examined the velocity profiles of elicited saccades and found no indication that elicited saccades with longer latencies were different from elicited saccades with shorter latencies. In particular, they were not prematurely abbreviated by the cessation of 70-ms stimulation train, even when the train ended before the saccade began. Instead, all the saccades electrically elicited from SEF exhibited the classic all-or-none ballistic features originally described for saccades elicited from FEF by Robinson and Fuchs (1969) with consistent amplitudes and directions regardless of latency.

We term the median elicited saccade, taken during central fixation, the characteristic saccade vector for a site. The distribution of characteristic saccade vectors from all 109 sites is shown in Fig. 2, right. For this and subsequent figures, characteristic saccades from left hemispheres are converted to a right-hemisphere convention by mirror reversing them as described in METHODS. For example, elicited saccades with a polar direction of 45° (upward and rightward) obtained from a left-hemisphere site would be converted to 135° (upward and leftward).

All 109 characteristic saccades were contralaterally directed. Their directions (_e) ranged from nearly straight up (minimum 97°) to nearly straight down (maximum 262°), and nearly all directions into the contralateral hemifield seem to be represented. However, almost twice as many saccades were elicited into the upper contralateral quadrant (65%) as into the lower quadrant (35%). Characteristic elicited saccade amplitudes (_e) ranged from small (1.8°) to large (26.9°), with a median of 13.1°.

Presaccadic response fields

We searched for neurons with presaccadic responses while the monkey performed either the visual-saccade or the deferred-saccade task (see METHODS) with the experimenter interactively varying the coordinates of the peripheral visual cue between trials with a joystick. When a SEF neuron with presaccadic activity was isolated, we first performed a set of preliminary tests to estimate the spatial location of the response field and classify its presaccadic activity as visual, movement, or visuomovement. An example of this preliminary testing procedure is illustrated in Fig. 3, A and B. Because a neuron's response field could be located at any point in the visual field, we first attempted to make a rough estimation of its location using the interactive version of the deferred-saccade task while watching the on-line raster display and listening to audio feedback of the neural response. The response field center of this neuron appeared to be directly downward (~270°) and ~10° eccentric. A retrospective quantitative analysis of these data confirmed that our initial estimate was fairly accurate (Fig. 3A, right). The three-dimensional plot shows the target coordinates of the 25 trials from this interactive task plotted along the x and y axes versus the neural response plotted along the z axis. The surface shows the best fit of these data to the Gaussian function for direction and eccentricity described in METHODS. The peak of the function was 263° in polar direction and 8.1° in eccentricity, fairly close to our on-line estimate of 270° in polar direction and 10° in eccentricity.

View larger version (26K): [in this window] [in a new window]   Fig. 3. Standard testing procedure performed on 1 SEF presaccadic neuron (SY024, right hemisphere, visuomovement neuron). A: interactive deferred-saccade task used to perform an on-line estimate of the neuron's response-field location. The 25 target vectors chosen between trials by the experimenter using a joystick are shown to the left. A retrospective analysis of neural activity during these trials is shown to the right. Activity was aligned to the onset of the visual stimulus and the responses (average spike rate for a 182-ms epoch starting 82 ms after visual due onset) were fit to the Gaussian function of direction and eccentricity described in METHODS. The fit was significant (F[24,19] = 2.29, P < 0.05), with a peak at 263° in polar direction and 8.1° in eccentricity. Stems above or below each response marker show the difference between the response and the best fit Gaussian function. B: classification of presaccadic activity. Rasters and peristimulus time histograms (PSTHs) of neural activity during performance of the memory-saccade task in the dark with a target located 270° in direction and 10° in eccentricity. The duration of the visual cue was 0.5 s, and the subsequent delay interval was 1.0-1.8 s. Raster and PSTH on the left are aligned to the appearance of the visual cue and sorted by delay-period length. Raster and PSTH on the right are aligned to the beginning of the memory-guided saccadic eye movement and sorted by latency relative to the offset of the fixation point. Notice that this neuron responded to both the cue (latency 64 ms) and then again in conjunction with the saccadic eye movement directed to the location where the cue was presented (latency 68 ms relative to saccade start). , beginning of saccadic eye movement. *, offset of fixation point. C: quantitative analysis of visual and movement activity optimal direction using the deferred-saccade task. The visual cues were all 10° eccentric with polar directions spaced 60° apart. The peristimulus time histograms represent 10 trials for each of the 5 cue locations. The PSTHs on the left are aligned to the onset of the visual stimulus, and the PSTHs on the right are aligned to the start of the saccadic eye movement. D: response magnitudes, Gaussian fits, and estimated optimal directions of visual and movement activity. Response magnitudes of visual activity are the spike rates during a 200-ms epoch starting 60 ms after the stimulus appeared. Response magnitudes of movement activity are the spike rates during a 200-ms epoch starting 100 ms before saccade initiation. Thick bars show the mean rate ± SE at each direction. , the response rates for each trial. Because single-trial rates in a 200-ms epoch are quantified in steps of 5 Hz, overlapping data points in the visual activity plot were spread out horizontally to show all the points at each direction tested. In the movement activity plot, the monkey's natural variation in saccade direction minimized overlap, and so only a few overlapping data pairs had to be spread. Both Gaussian fits were highly significant (visual activity: F[50,47] = 5.613, P < 0.00001; movement activity: F[49,46] = 4.130, P < 0.00001). Notice that v and m are very similar to each other (258 ± 4 vs. 261 ± 4°), and to the v (263 ± 7°) obtained from the retrospective analysis of our informal test data shown in A.

After approximating a neuron's response-field location, we analyzed the composition of its presaccadic activity using the memory-saccade task with a visual cue located at our initial on-line estimate (Fig. 3B). This neuron was classified as having visuomovement activity because it discharged in conjunction with the appearance of the peripheral visual cue and then again in conjunction with the saccadic eye movement.

Next we attempted to formally analyzed the spatial location of a neuron's response field by determining its optimal direction and optimal eccentricity in separate experiments Response-field direction was analyzed by systematically varying the polar direction of the visual cue around our initial on-line estimate while holding eccentricity constant, and response-field eccentricity was analyzed by systematically varying the eccentricity of the visual cue around our initial on-line estimate while holding direction constant. Figure 3, C and D, shows this procedure for response-field direction. Because this neuron exhibited both visual and movement activities in the memory-saccade task (Fig. 3B), the deferred-saccade task was used so that visual and movement activity could be analyzed separately. These data were fit to the Gaussian function described in METHODS to obtain independent estimates of the neuron's visual activity optimal polar direction (_v) and movement activity optimal polar direction (_m). Notice that the resulting estimate of _v (258°) is very close to the estimate of 263° obtained by fitting the informal, interactive data described in Fig. 3A. However, the formal testing procedure yielded a smaller SE for the _v estimate. Furthermore this neuron's _m could not be accurately estimated with this set of interactive data even though it was well estimated with the formal data. Thus the response-field parameters estimated with our formal tests were generally more accurate; however, some neurons mapped only informally were included in our analysis if the data provided good fits yielding a _v and/or _m that agreed with the on-line estimates.

A total of 160 presaccadic neurons in the low-threshold SEF were successfully classified and mapped. The Venn diagram in Fig. 4 summarizes their response classification: overall, 84% of these neurons had visual activity and 57% had movement activity. There were 68 (43%) with purely visual responses, 26 (16%) with purely movement responses, and 66 (41%) with both visual and movement responses. The median cortical depths for the three neuronal types were 1.75 mm (visual), 2.18 mm (movement), and 1.68 mm (visuomovement). The null hypothesis of equality of depth across neuron types is unlikely (Kruskal-Wallis ²_[2] = 7.39, P < 0.025).

View larger version (40K): [in this window] [in a new window]   Fig. 4. Summary of SEF presaccadic activity. PSTHs showing averaged activity of 104 SEF neurons during the memory-saccade task. Top left: activity is aligned to the onset of the visual cue. Top middle: activity is aligned to the offset of the fixation point. Bottom right: activity is aligned to the beginning of the saccadic eye movement. Notice that the composite activity has a phasic visual response beginning ~102 ms after the response-field cue appears, and a 2nd response beginning ~134 ms after the fixation light goes off. The median memory-saccade latency relative to the fixation point offset was 248 ms. Also notice that the average rate preceding fixation light off (top right) is elevated relative to the fixation rate preceding cue onset (top left). The median delay period between cue-light off and fixation-light off was 1 s. Bin width is 10 ms. Bottom left: Venn diagram showing the classification of presaccadic activity for 160 SEF neurons using the memory-saccade or deferred-saccade tasks. The subset of 104 neurons used for the composite histograms had visual, movement, and visuomovement types in roughly the same proportion.

To illustrate the overall presaccadic responses of SEF, the average response histograms for 104 neurons tested on the memory saccade task are shown in Fig. 4, top (56 of the 160 neurons could not be tested on the memory saccade task because the monkey stopped working or the neuron was lost; they were classified on the basis of activity during the deferred-saccade task). These composite histograms were compiled by first computing histograms of neural activity for each neuron using target locations at or near the optimal location for their presaccadic responses, and then averaging them together. The activity of these 104 neurons aligned to cue onset, fixation offset, and saccade, beginning show that presaccadic activity in SEF is composed of two main components: a burst of activity in response to the appearance of a visual stimulus and a burst of activity preceding and during the saccade. Some tonic mnemonic activity is also evident, indicated by the small but significant elevation in tonic activity (~3 spikes/s) while fixating during the delay period, compared with the period of fixation before the peripheral cue was presented. A very similar pattern of activity was observed when these same neurons were tested with the deferred-saccade task. In general both the deferred-saccade task and the memory-saccade task gave similar results.

When the visual and movement activities of SEF visuomovement neurons were mapped, their response fields were usually closely aligned. Of the 66 visuomovement neurons in our sample, we obtained quantitative estimates of both _v and _m for 44 of them. Figure 5 shows the relationship between _v and _m in these 44 neurons. Both visual and movement activity data were obtained from the same trials during the deferred-saccade or memory-saccade task. However, all estimates of _v and _m were from independent fits of distinct and completely nonoverlapping visual and saccadic bursts. Although some visuomovement neurons exhibited significant differences between _v and _m (the largest was 57°), the median absolute difference was only 8° and with an extremely strong correlation of 0.95. A linear regression of _m on _v yielded a slope not significantly different from 1 (1.1 [0.94, 1.17]), and a y intercept not significantly different from 0 (14 [35, 8]), indicating that the visual and movement fields of SEF visuomovement neurons largely overlapped. This congruence of visual and movement response fields is not only a principal finding but also justifies using the mean of _v and _m as an estimate of a visuomovement neuron's overall optimal direction (_vm) as described in the following text.

View larger version (19K): [in this window] [in a new window]   Fig. 5. Relationship between visual and movement response-field optimal direction of 44 visuomovement neurons. Separate visual (v) and movement (m) response-field optimal directions were computed using the memory-saccade task or deferred-saccade task to dissociate visual activity from saccade-related movement activity. Optimal directions from left-hemisphere neurons were transformed to a right-hemisphere convention (see METHODS). Dashed line with unity slope is drawn to emphasize the alignment of visual and movement response fields. Length and placement of plot frame indicate range of data along each axis.

Figure 6 shows the optimal response-field vectors for all 160 SEF neurons considered in this report. The direction of each neuron's response field was based on its estimated _v, _m, or _vm. The eccentricity of most response fields were based on their optimal stimulus eccentricity (_v) or optimal saccade amplitude (_m) obtained using the log-Gaussian fits described in METHODS (and detailed in the following text) or their mean if both were computed (_vm). However, satisfactory estimates of  could not be computed for 50 of the 160 neurons using either the formal or interactive tests. For these 50 neurons, we used our initial on-line estimate of response-field eccentricity for the plots in Fig. 6; however, these data were not included in further analyses of  (e.g., Fig. 11). As with the electrically elicited saccades, the optimal response-field vectors of left-hemisphere neurons were transformed into a right-hemisphere convention. Compilation of all single-neuron data in this way provides a comprehensive sample of the neural representation of saccades in SEF. Notice that this collection of 160 optimal saccade vectors encompasses virtually all possible contralateral saccades, similar to the analogous plot of elicited saccade vectors (Fig. 2). Unlike the elicited saccades which were all contralateral, however, a few SEF neurons (5%, 8 of the 160) had unequivocally ipsilateral response fields. Another difference between the response fields and the elicited saccades is that the upper and lower contralateral quadrants were equally represented by visual, movement, and visuomovement activity.

View larger version (34K): [in this window] [in a new window]   Fig. 6. Optimal response-field vectors for 160 presaccadic neurons in the low-threshold SEF. Top: 2-dimensional plot of optimal saccade vectors with vectors from left-hemisphere neurons converted to a right-hemisphere convention by mirror reversing them as described in METHODS. Visual receptive field vectors are red and movement field vectors are green. For visuomovement neurons with the optimal vectors for both visual and for movement activity quantified (see Fig. 5), the average of these 2 vectors was used and colored blue. Bottom: distribution of response-field directions (right-hemisphere transformed) and eccentricities. Notice that locations throughout the contralateral visual hemifield are represented but that only 8 of the 160 SEF neurons (5%) were unequivocally ipsilateral (4 visual, 4 movement, 0 visuomovement).

Relationship of presaccadic activity optimal direction to electrically elicited saccade direction

After mapping the presaccadic response field of a neuron, we usually tested for electrically elicited saccades by stimulating through the recording electrode before moving it. Across all SEF sites where both single neurons were mapped and saccades were elicited with currents 100 µA, the neural response fields were highly predictive of the elicited saccade dimensions. Figure 7A shows the analysis of optimal direction for one SEF neuron that exhibited only visual activity and the saccades subsequently elicited with electrical stimulation at the neuron's recording site. The Gaussian fit indicated a quite horizontal _v (183° ± 4°). After recording the activity of this neuron, and without moving the microelectrode, we electrically stimulated through the recording electrode. The 50% threshold for eliciting saccades at this site during central fixation was 20 µA, and the set of 10 elicited saccades shown in Fig. 7A, bottom, were obtained using 25 µA. The median elicited saccade direction (_e) was 185°, very similar to the _v recorded at this stimulation site.

View larger version (22K): [in this window] [in a new window]   Fig. 7. Relationship between the optimal direction of neural activity and the direction of subsequent electrically elicited saccades. A: recording and stimulation at 1 SEF site with a visual neuron. Top: optimal direction was mapped using the visual probe task. Neural responses (latency, 60 ms; duration, 180 ms) and Gaussian fit (F[53,50] = 2.953, P < 0.00001) is plotted in both Cartesian and polar coordinates. Thick bars show the response means ± SE for each visual cue direction. The Cartesian plot shows the individual responses from each trial with overlapping data points slightly separated along the x axis. Solid arrow and dashed lines in the polar plot shows the estimated optimal direction (v) and its 95% confidence interval. Bottom: saccades electrically elicited with 25 µA during fixation of a centrally located target (site threshold, 20 µA). The time course of one representative elicited saccade is shown on the left. The two-dimensional trajectories of 10 elicited saccades are shown on the right. The median elicited saccade direction (e), amplitude, and latency was 185°, 10.1°, and 42 ms, respectively. Notice how v is very close to e. H, horizontal eye position; V, vertical eye position. B: recording and stimulation at 1 SEF site with a movement neuron. m was mapped with the memory-saccade task. Neural response (latency, 100 ms; duration, 200 ms) and Gaussian fit (F[61,58] = 1.964, P < 0.006), elicited saccades using 30 µA (site threshold, 20 µA) are plotted using the same conventions as A. The median elicited saccade direction (e), amplitude, and latency were 126°, 9.4°, and 50 ms, respectively. Notice how m is again very close to e.

Figure 7B illustrates a similar analysis for a neuron with presaccadic movement activity and no visual response. This neuron was tested with the memory-saccade task using an array of eight visual cue directions. The optimal saccade direction for the neuron's movement activity was 119°. The saccades subsequently elicited by stimulation using 30 µA had a median direction of 126°. Notice how _e is very close to _m. Also notice that the elicited saccades are slightly more horizontal than the neuron's optimal direction.

Figure 8 summarizes the relationship between the response-field optimal direction of 77 SEF neurons and electrically elicited saccade direction. As in Fig. 6, a single neuron's optimal direction was derived from estimates of _v, _m, or _vm. The correlation between neural activity optimal direction and elicited saccade direction was highly significant (r_aa = 0.92). A separate analysis of visual and movement activity (using estimates of _v from visual and visuomovement neurons and estimates of _m from visuomovement and movement neurons) yielded similarly strong correlations (visual activity: r_aa = 0.94, n = 59; movement activity: r_aa = 0.88, n = 42).

View larger version (22K): [in this window] [in a new window]   Fig. 8. Relationship between the optimal direction of SEF response fields and the direction of electrically elicited saccades. The response-field optimal direction of 77 SEF neurons computed from visual (v, red), movement (m, green), and visuomovement (vm, blue) activity are plotted vs. the median direction of saccades electrically elicited at the recording site (e). vm is the mean angle of the independently computed estimates of v and m for neurons with both visual and movement activity (see Fig. 6). All directions are right-hemisphere transformed. The dashed line with unity slope is drawn to emphasize the overall similarity between neural response field and electrically elicited saccade direction. The solid line shows the linear regression of elicited saccade direction on response-field direction. Notice the systematic departure from equality for saccade representations near the vertical. Length and placement of plot frame indicates range of data along each axis.

Although SEF neural activity and elicited saccade directions were highly correlated, the precision of their alignment was not uniform across the visual hemifield. In fact, the median absolute difference for the 77 neurons in Fig. 8 was nearly 17°. Although some of this discrepancy could be due to errors of measurement (our estimates of _v and _m typically had an SE <10°), there was a conspicuous trend for sites with neurons that had response fields nearly upward and downward to yield elicited saccades that were slightly more horizontal. This trend was confirmed by computing the linear regression of elicited saccade direction on response-field direction. The slope of the regression line was 0.79 [0.71, 0.87], significantly less than unity inasmuch as its 95% confidence interval does not include 1. Furthermore the complete regression equation predicts that sites where neurons have response fields directly up (90°) will yield elicited saccades that are 107° in polar direction, and sites where neurons have response fields directly down (270°) will yield elicited saccades that are 248° in polar direction. Thus neurons at sites representing vertical directions should yield elicited saccades rotated toward the horizontal an average of ~20°. An example of this phenomenon is shown in Fig. 9. This neuron's _v and _m was nearly straight down (258 and 261°, respectively); however, the median elicited saccade direction was 233°, thus rotated ~27° contralateral from the neuron's response field.

View larger version (20K): [in this window] [in a new window]   Fig. 9. SEF site where 1 presaccadic neuron (SY024, right hemisphere) with nearly vertical (downward) visual and movement response fields yielded oblique electrically elicited saccades. Top: polar plots showing response magnitudes, Gaussian fits, and estimated optimal polar directions for visual (left) and movement activity (right). Response magnitudes of visual activity are the spike rates during a 200-ms epoch starting 60 ms after the stimulus appeared. Response magnitudes of movement activity are the spike rates during a 200-ms epoch starting 100 ms before saccade initiation. All other conventions are the same as Fig. 7. Notice that v and m are both ~10° rotated away from straight down (270°), but e is rotated ~37° from downward. This neuron is the same neuron used to illustrate our basic testing procedure in Fig. 3.

Relationship of presaccadic activity optimal eccentricity to electrically elicited saccade size

Similar to the relationship between the optimal direction of neural activity and the direction of saccades electrically elicited from the recording site, we found a relationship between a neuron's optimal eccentricity and the amplitude of subsequent electrically elicited saccades. However, this relationship was less precise than what we found for polar direction. Figure 10 shows the neural recording and electrical stimulation at three different SEF sites. The visual neuron in Fig. 10, left, preferred moderate visual cue eccentricities (_v = 12.4 ± 1.0°), and medium-sized saccades were subsequently elicited (_e = 8.5 ± 1.0°). The visual neuron in Fig. 10, middle, preferred large eccentricities (_v = 29.6 ± 1.1°), and fairly large elicited saccades were subsequently elicited (_e = 17.1 ± 0.7°). The movement neuron in Fig. 10, right, clearly preferred very large saccades but had very poor tuning for saccade size (_m = 47.5 ± 11.3°), and the subsequent elicited saccades were very large (_e = 25.9 ± 2.5°). Figure 11 shows the optimal eccentricity of 60 neurons plotted against the median saccade amplitudes electrically elicited from their recording sites. The correlation coefficient (r = 0.49) is highly significant, but only about half the correlation that was observed for the analysis of response field versus elicited saccade direction. As in the analysis of saccade direction, similar results were obtained when visual and movement activity were considered separately. _v obtained from visual and visuomovement neurons were significantly correlated with _e (r = 0.48, P < 0.005, n = 44), as were _m and _e obtained from visuomovement and movement neurons (r = 0.39, P < 0.05, n = 26).

View larger version (24K): [in this window] [in a new window]   Fig. 10. Relationship between the optimal eccentricity of neural activity and the amplitude of subsequent electrically elicited saccades at 3 different SEF sites. Left: SEF site where the visual activity of 1 visuomovement neuron (AB231, left hemisphere) was tested with 5 eccentricities along a direction ( = 345°) close to the neuron's optimal direction (v = 346 ± 3°). PSTHs are aligned to the stimulus appearance during the visual-probe task (the movement activity of this neuron was deemed too weak for analysis). Each visual cue location was tested with ~6 trials, and the fixation point was at the center of the screen. Below the PSTHs the responses (110 ms epoch starting 40 ms after stimulus onset) and are plotted as the means ± SE, along with the best fit log-Gaussian function (see METHODS). Solid line shows the estimated optimal eccentricity (v = 12.4°); dashed lines show its SE (1.0°). Saccades were electrically elicited from the site of this using 75 µA (threshold, 60 µA) while the monkey fixated the same central fixation point used to test the neuron's response-field eccentricity. The median amplitude (e), direction, and latency of the elicited saccades were 8.5°, 344°, and 46 ms, respectively. Middle: SEF site where a visual neuron (AB135, right hemisphere) exhibited maximal responses to stimuli far from the fixation point. PSTHs are aligned to the stimulus appearance during the visual-probe task. Each visual cue location was tested with ~10 trials with the fixation point at the center of the screen. Spike rates were taken from a 150-ms epoch starting 60 ms after the stimulus onset. Saccades were electrically elicited from this site using 150 µA (threshold, 130 µA) while the monkey fixated at the center of the monitor. Because such a high current was needed to elicit saccades at this site, these elicited saccade data are not included in any of the summary figures and statistics even though the neuron was located in SEF. Right: SEF site where the response field of 1 visuomovement neuron (AB197, right hemisphere) exhibited an extremely eccentric response field. PSTHs are aligned to the beginning of the saccadic eye movement in the deferred-saccade task (the visual activity of this neuron was too weak to analyze). Because this neuron responded best in conjunction with very large saccades, it was tested with the fixation point located 25° down (21.7°) and right (12.5°) to allow larger amplitude saccades than would otherwise be possible. Each visual cue location was tested with ~10 trials. Spike rates were taken from a 180-ms epoch starting 90 ms before the beginning of the saccadic eye movement. Saccades were electrically elicited from this site using 70 µA (site threshold 60 µA) while the monkey fixated the same fixation point used to test the neuron's m. The median amplitude (e), polar direction, and latency of elicited saccades were 25.8°, 115° and 44 ms, respectively. This was the only case where neural activity and elicited saccades were not tested with a centrally located fixation point.

View larger version (19K): [in this window] [in a new window]   Fig. 11. Relationship between the optimal eccentricity for SEF presaccadic activity and the amplitude of electrically elicited saccades. The response-field optimal eccentricity of 60 SEF neurons computed from visual (v, red), movement (m, green), and visuomovement (vm, blue) activity are plotted vs. the median amplitude of saccades electrically elicited at the recording site (e). Note that both axes are logarithmic and the correlations were based on log-transformed data. The dashed line with unity slope is drawn to emphasize how elicited saccade amplitude was generally smaller than the corresponding optimal eccentricity of neurons recorded at the stimulation site. Length and placement of plot frames indicates range of data along each axis.

Notice that most of the data points in Fig. 11 were below the x = y diagonal, indicating that estimates of optimal eccentricity were generally larger than elicited saccade amplitude. One reason is that many SEF response fields were open ended with responses that fell off very gradually, if at all, with increasing eccentricity. As a result, the estimates _v and _m for these eccentric response fields generally had a correspondingly larger SE (see Fig. 10), making it difficult to obtain a single accurate estimate of response-field optimal eccentricity that could be considered truly "optimal." Similar observations of open response fields have been made in the superior colliculus (Munoz and Wurtz 1995) and the FEF (Bruce and Goldberg 1985). Like _v and _m of visuomovement neurons, _v and _m were also highly correlated (r = 0.84, P < 0.005, n = 15). Such a small sample of visuomovement neurons with both _v and _m estimated again reflects the difficulty in estimating the optimal response-field eccentricity of SEF presaccadic activity.

Relationship of presaccadic activity type to electrically elicited saccade threshold

We compared the likelihood and ease of obtaining elicited saccades at the sites of purely visual neurons, visuomovement neurons, and movement neurons. Low-thresholds are more likely where FEF neurons have movement activity (Bruce et al. 1985); however, we were surprised to find that microstimulation was uniformly effective in eliciting saccades at SEF sites, regardless of the type of presaccadic activity there. Of the 68 SEF sites where visual neurons were recorded, 51 were tested with electrical stimulation. Of these tests, 82% (42/51) elicited saccades, and the median threshold (regarding threshold at the 9 unexcitable sites as large) was 52.5 µA. Of the 66 SEF sites where visuomovement neurons were recorded, 53 were tested with electrical stimulation and 85% (45/53) elicited saccades with a median threshold of 55 µA. Of the 26 SEF sites where movement neurons were recorded, 21 were tested with electrical stimulation and 86% (18/21) elicited saccades with a median threshold of 52.5 µA. A ² test failed to indicate that these percentages of elicited saccades differ significantly across neuron types (²_[2] = 0.0239, P > 0.5), and a Kruskal-Wallis test failed to indicate that thresholds differ across neuron types (²_[2] = 0.7575, P > 0.5).

Topographic organization

The representation of saccades in SC has long been known to have a straightforward topography, with small saccades represented anterior, large saccades posterior, upward saccades medial, and downward saccades lateral (Robinson 1972). In FEF saccade amplitude is also topographically organized with large saccades represented dorsomedially and small saccades ventrolaterally, but the representation of saccade direction is more complex with gradual changes in saccade direction as an electrode is advanced parallel to the cortical surface resembling a hypercolumnar organization (Bruce et al. 1985). To determine what type of saccade topography, if any, the low-threshold SEF has, we analyzed neural response-field vectors and electrically elicited saccade dimensions with respect to the relative location of the electrode tip within the cortex. Although we did not find a systematic global topographic organization of saccades across SEF, there was continuity of saccade direction across short distances and evidence of columnar and hypercolumnar organization with respect to polar direction.

As the electrode was advanced perpendicular to the cortical surface, neural response fields and elicited saccades represented similar saccade directions, indicating a columnar organization with respect to saccade direction. One example of this finding for neural activity is illustrated in Fig. 12, top left. In this electrode penetration, two superficially located neurons that were mapped simultaneously had response-field directions of 140 and 143°. Furthermore a neuron 0.45 mm deeper that was also mapped during the same experimental session had a response-field direction of 147°, very close to response-field direction of the two neurons recorded above them. Similar results were found when analyzing electrically elicited saccades. Figure 12, bottom left, shows the characteristic direction of saccades elicited from two different sites in the same electrode penetration. The characteristic elicited saccade direction from the more superficial stimulation site was 134° (threshold, 45 µA), and the characteristic elicited saccade direction from the stimulation site 1.3 mm deeper (threshold, 20 µA) was an almost identical 135°.

View larger version (18K): [in this window] [in a new window]   Fig. 12. Columnar organization of SEF. Top left: response-field optimal direction of 3 neurons recorded during a single-electrode penetration (penetration ABSR044, monkey AB, right hemisphere) are indicated as arrows originating from the location along the electrode path where the neurons were recorded. All 3 neurons exhibited only visual activity. Top right: distribution of directional-concentration indexes (DCI, see METHODS) from all electrode penetrations with multiple neurons mapped (36 penetrations, 84 neurons). The median DCI (- - -) was 0.99. Median control DCI () computed from randomly shuffled data was 0.686. Notice that most DCIs were substantially higher than the control, indicating that neurons recorded on the same electrode penetration had very similar optimal directions. Bottom left: characteristic directions of elicited saccades from 2 stimulation sites studied during a single-electrode penetration (penetration ABSR128, monkey AB, right hemisphere). Bottom right: distribution of DCI from all electrode penetrations with multiple stimulation sites (11 penetrations, 22 sites). The median DCI (- - -) was 0.999. Median control DCI () was 0.804. Again, notice that nearly all DCIs are larger than the control median.

To examine columnar organization for all electrode penetrations with more than one neuron or stimulation site, the mean vector length of their characteristic directions was computed (see METHODS). Because the mean vector length indexes the concentration of directions around the circular mean, we call these measures directional-concentration indexes (DCI). Indexes near 1 indicates a very small deviation around the mean (and hence similar directions), whereas smaller indexes indicate a larger deviation around the mean (and hence diverse directions). The DCI of the penetration with three neurons illustrated in Fig. 12, top left, was 0.998, consistent with their similar response-field directions. The DCI of the penetration with two stimulation sites illustrated in Fig. 12, bottom left, was 0.99996, consistent with their almost identical characteristic directions. The histograms in Fig. 12, right, show the distribution of DCIs for all penetrations with multiple neurons and stimulation sites. The vast majority of DCIs were between 0.95 and 1, with a median of 0.990 for response fields (37 electrode penetrations) and a median of 0.999 for elicited saccades (11 electrode penetrations). To check whether such large index values simply reflected the strong contralateral bias of SEF, we used a bootstrap technique to determine the distribution of DCIs expected by chance. The median control DCIs computed from randomly shuffled data (Fig. 12, -) were substantially smaller than most experimental DCIs, and no control median (in 100 different