INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Primates use rapid saccadic eye movements to move their line of sight to a newly appearing object in the peripheral visual field. In this "visual grasp reflex" (Hess et al. 1946), the stimulus is also the target for the movement. Both humans (Hallett and Adams 1980) and monkeys (Amador et al. 1998; Bell et al. 2000), however, can be instructed not to use the stimulus as the target for a saccade but as a landmark for a saccade to its mirror location (Schlag-Rey et al. 1997). This task, known as the anti-saccade task, requires the active inhibition of the prepotent response to look to the stimulus in favor of the generation of a voluntary saccade to an empty location in the visual field. The anti-saccade task has become a popular test of frontal function because patients with frontal lobe damage (Fukushima et al. 1994; Guitton et al. 1985; Pierrot-Deseilligny et al. 1991; Walker et al. 1998) and schizophrenic patients (Clementz et al. 1994; Fukushima et al. 1990) often fail to suppress a reflexive saccade toward the stimulus (for review see Everling and Fischer 1998).

PET studies, however, have demonstrated that cerebral blood flow is not only increased in the frontal lobe, but in a large network of cortical areas when subjects perform blocks of anti-saccade trials compared with blocks of pro-saccade trials (Doricchi et al. 1997; O'Driscoll et al. 1995; Paus et al. 1993; Sweeney et al. 1996). These brain areas include dorsolateral prefrontal cortex (DLPFC), frontal eye fields (FEF), supplementary eye fields (SEF), superior and inferior parietal cortex, and primary visual cortex. Recent functional MRI (fMRI) studies have yielded similar results (Connolly et al. 2000; Kimmig et al. 2001).

Although these studies have shown that a number of cortical areas were more active during anti-saccade trials compared with pro-saccade trials, they have not provided any direct information about when these areas were activated during the task. Thus no imaging study has separated preparatory signals from those involved in executing the response (for review see Corbetta and Shulman 2002). The increased FEF and SEF activation, for example, is often interpreted as evidence for a more prominent role of these cortical areas in anti- compared with pro-saccade generation (Gaymard et al. 1998). The differences, however, may also originate from different activity levels between pro- and anti-saccades before stimulus presentation and saccade generation. In fact, single neuron recordings in monkeys have demonstrated that FEF (Everling and Munoz 2000) and SEF neurons (Schlag-Rey et al. 1997) already show differences in their baseline discharge rate during visual fixation when the monkey is instructed about the saccade response. These differences have been interpreted as neural correlates for different preparatory sets between pro- and anti-saccades. To address the question whether variations in cortical fMRI activation between pro- and anti-saccades reflect differences in the motor signal for the saccade or in preparatory set, we measured event-related fMRI in a paradigm that allowed us to separate instruction-related brain activity from saccade-related brain activity.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects and visual displays

Ten subjects were paid volunteers in this study (6 women and 4 men; mean age, 26.6 ± 1.0 yr). Subjects gave informed written consent, and the University of Western Ontario Ethics Review Board approved all procedures. Each subject performed three training sessions to ensure they could execute each experiment (1 day before testing, immediately before entering the magnet bore, and just after entering the magnet before imaging). Thus each subject performed approximately 20-30 pro- and anti-saccades in total during the training sessions. Subjects lay supine and viewed computer generated images. Six subjects viewed images on a back-projection screen (Da-Lite, Warsaw, IN) with the aide of a mirror. Four subjects viewed the visual stimuli using Visible Eye Fully Integrated Eye Tracking (SensoMotoric Instruments, Needham/Boston, MA) and Visual Stimulation (Avotec, Stuart, FL).

Eye tracking/calibration

In four subjects, eye movements were recorded monocularly at 60 Hz while subjects performed the experiments in the magnet using Visible Eye Fully Integrated Eye Tracking and Visual Stimulation. After the subjects were placed on the MR bed, the video display and tracking apparatus were adjusted manually for stimulus viewing and eye tracking. Stimulus projection was binocular, and each LCD display could be moved independently about three axes for each eye. Once each subject was comfortable viewing the display, the infrared video eye tracker was calibrated at center position and five eccentric points for the subject's right eye. Additionally, each subject's eye was viewed by the experimenters on a computer monitor during data collection as each event-related trial was completed. Analysis of the eye movement traces was performed off-line.

Blocked design experiment

To identify eye movement regions activated by pro- and anti-saccades, each subject made pro-saccades and anti-saccades in a blocked design similar to previous studies (Connolly et al. 2000; Doricchi et al. 1997; Kimmig et al. 2001; O'Driscoll et al. 1995; Paus et al. 1993; Sweeney et al. 1996). Pro- and anti-saccades blocks were alternated with fixation control blocks (Fig. 1A) with a fixation control block at the beginning and the end of a scan. A green fixation cross signaled a pro-saccade, a red fixation cross signaled an anti-saccade, and a white fixation cross signaled the subject to maintain visual fixation. At the beginning of each pro/anti-saccade block (Fig. 1A), the fixation cross changed color from a white cross to either green or red cross. After 1,500 ms, a peripheral stimulus (white square, 3° × 3°) was flashed for 500 ms either 10° to the right or 10° to the left of the fixation. Subjects were instructed to look toward the stimulus on pro-saccade trials and to look from away from the stimulus to its mirror position in the opposite hemi field on anti-saccade trials. They then looked back at the central fixation cross. Each trial lasted 3 s; thus five pro-saccades (or 5 anti-saccades) were made within each 16-s block. Each functional scan took 4.6 min. This experiment was repeated two to four times and then averaged for data analysis. The order of pro- and anti-saccade blocks was pseudo-randomly alternated across subjects.

View larger version (23K): [in this window] [in a new window]   Fig. 1. The event-related visual stimulus sequence protocol for the blocked experiment and the event-related experiment. A: blocks were 16 s long. For the fixation blocks, a white fixation cross was presented. A green fixation cross signaled a pro-saccade trial and a red cross signaled an anti-saccade trial. The time course of 1 pro-saccade trial and anti-saccade trial was depicted above the respective blocks. There were 5 pro-saccade or 5 anti-saccade trials in each block. The dotted circle represents where the subject was fixating after saccade execution. B: visual stimulus sequence for the event-related experiment. In panel 1, a white circle stimulus was displayed for 2 s. In panel 2, this changed to either a green (PRO) or red cross (ANTI), which indicated whether to plan a pro- or anti-saccade, respectively. This instruction stimulus was presented for 1 of 3 possible times (6, 10, or 14 s). A peripheral stimulus (3° white square) appeared either 10° to the right (or 10° to the left) of center (panel 3). This stimulus was flashed for 500 ms. Subjects were instructed to immediately look toward the stimulus on pro-saccade trials and away from the stimulus to its mirror location on anti-saccade trials (in panel 3). The peripheral cue reappeared at the location of the correct saccade. They maintained fixation at this peripheral stimulus location for 12 s (panel 4) and made a saccade-to-center when appropriately cued (panel 5).

Event-related experiment

The visual display and timing sequences are depicted in Fig. 1B. Subjects fixated the white cross for 2,000 ms (Fig. 1B1) until the instruction stimulus changed its color (Fig. 1B2). A green cross signaled a pro-saccade trial and a red cross signaled an anti-saccade trial. The order of trials was pseudo-randomly interleaved. After a 6-, 10-, or 14-s instruction period, a peripheral stimulus (white square, 3° × 3°) was flashed for 500 ms either 10° to the left or 10° to the right of the fixation cross. Subjects were previously trained to look toward the peripheral stimulus on pro-saccade trials and to its mirror location in the opposite hemifield on anti-saccade trials (Fig. 1B3). When subjects noticed they had made an error (pro-saccade on anti-saccade trials or anti-saccade on pro-saccade trials) they pressed a button. The number of reported errors was very low (8/492 or 1.6%) and all but one reported error was a pro-saccade on an anti-saccade trial. We designed our stimulus to reduce the potential of error by employing an overlap saccade task compared with a gap saccade task (Bell et al. 2000; Fischer and Weber 1997). Two seconds later, the appropriate peripheral stimulus was turned on (Fig. 1B4), and subjects maintained fixation on this peripheral stimulus. The simultaneous offset of the peripheral stimulus with onset of the central cross (Fig. 1B5) instructed subjects to make a visually guided saccade back to the central position, ready to begin the next trial. A transistor-transistor logic (TTL) pulse from the imaging control computer synchronized visual stimulation to the imaging volume times.

For the event-related scans, each subject performed six pro- and six anti-saccade trials that were pseudo-randomly interleaved. Each scan was 6.8 min long. Four or five event-related scans were completed for each subject, except for one subject who completed three scans. The scans were averaged within subjects before data analysis. In the off-line image analysis, 2 of 43 scans were excluded from the image analysis because of movement artifacts and/or instrumental instability. Equal numbers of pro- and anti-saccades to left and right were made at each of the three (6, 10, and 14 s) instruction periods.

Data acquisition

All imaging was conducted using a 4 Tesla Varian (Palo Alto, CA) Unity Inova whole-body MRI system equipped with a Siemens Sonata Gradient chain (Siemens, Erlangen, Germany) with a quadrature head coil. Functional data were collected using BOLD (blood-oxygenation level-dependent) navigator echo corrected T^2*-weighted segmented gradient echoplanar imaging [16 slices, 64 × 64 resolution, 19.2 × 19.2 cm in-plane field of view (FOV), echo time (TE) of 15 ms, flip angle of 30°, and 4 segments per slice, thus achieving a total volume acquisition time of 2.0 s and voxel sizes of 3 × 3 × 5 mm]. A full k-space phase reference was acquired for each slice, which suppressed high field geometric distortions.

Functional data were superimposed on high-resolution inversion prepared three-dimensional T¹-weighted anatomical images of the brain collected immediately after functional images using the same in-plane FOV (128 slices, 256 × 256, TE = 5.4 ms, TR = 9.8 ms, flip angle = 15°). Anatomical images for each subject were segmented at the gray/white matter boundary, rendered, and inflated for visualization purposes (Goebel et al. 1998).

Image analysis

Analysis was carried out using BrainVoyager 2000 version 4.4 (Brain Innovation, Maastricht, The Netherlands). All functional images underwent motion correction (within-slice), linear trend removal and a spatial band-pass filtering (full-width half-maximum of 2) before being transformed into the stereotaxic frame of (Talairach and Tournoux 1988). The brain regions were defined using the general linear model with separate predictors for pro-saccades and anti-saccades from the blocked design experiments. We then convolved these two predictors with the hemodynamic response function. The resulting averaged functional map across subjects had a threshold of corrected P value (P < 0.00001). These functionally mapped brain regions were then used to examine the BOLD signal changes from each subject's event-related experiments.

To analyze the signal intensity changes from the single event-related experiments, we aligned each trial to onset of 1) the instruction stimulus or to the onset of 2) the peripheral stimulus. The data point at the onset of the instruction stimulus was defined as the baseline for each event-related trial (time point 0 for each instruction stimulus in Fig. 2).

View larger version (26K): [in this window] [in a new window]   Fig. 2. Averaged BOLD signals (n = 10) during the 3 instruction period times for the event-related experiment (6, 10, and 14 s) from left hemisphere frontal eye fields (FEF). The instruction onset stimulus occurred at the 0 time point. This time point at 0 was defined as the baseline activation. This was standardized for each individual trial, thus there was no SE at the 0 time point. The SE bars signify SE across subjects at each time point. The gray periods signify the time points used to average each instruction period within each subject.

Statistical analysis of the event-related trials

For each of the instruction periods (6, 10, and 14 s), we shifted our statistical region 2 s forward in time and excluded the first data point (at 0 in Fig. 2) to accommodate the hemodynamic response. Thus for the 6-s instruction period, our statistical region included the data points from 4 to 8 s; for the 10-s instruction period, our statistical region included the data points from 4 to 12 s; and for the 14-s instruction period, our statistical region included the data points from 4 to 16 s. The BOLD percent signal change during these three instruction times (6, 10, and 14 s) was computed by averaging the data points for each instruction period and then averaging these three values within each subject (schematically shown as the grayed in regions for the average data in Fig. 2).

For the movement periods, we excluded the first 4 s after peripheral stimulus onset (thus including an epoch of 4-12 s). Trials with leftward and rightward saccades were combined for pro- and anti-saccade trials. The right panels in Figs. 6-8 were computed by averaging the pro- and anti-saccade instruction or movement epochs within each subject and then across the subjects. The statistical test used for comparison between pro-saccade and anti-saccade signal intensities across subjects was the Student's paired t-test.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Functional mapping of eye movement regions using blocked experiment

We first identified regions of the cortex that were selectively activated for saccades (both pro and anti) compared with the fixation control across our subjects (corrected P < 0.00001). This subtraction revealed functional activation within the right hemisphere DLPFC, bilateral activation of the FEF and SEF, parietal activation along the cortex lining the intraparietal sulcus (IPS), and early visual areas along the calcarine sulcus (V1/V2; Table 1). Figure 3 shows three anatomical views of the averaged functional map of saccades compared with the fixation control at a viewpoint centered on the DLPFC activation (Talairach coordinates x = 33, y = 40, z = 37; Fig. 3, A and B). This functional activation map was rendered onto the inflated right hemisphere demonstrating the location of the DLPFC in relation to the precentral sulcus activation of FEF and SEF (Fig. 3, C and D). When the signal intensity during the pro- and anti-saccade blocks were compared for each brain area, only FEF, SEF, and IPS showed significantly increased activation for anti-saccades compared with pro-saccades across our subjects (P < 0.05; Fig. 4).

View larger version (82K): [in this window] [in a new window]   Fig. 3. Functional magnetic resonance imaging (fMRI) BOLD activation from saccade blocks (pro and anti) compared with fixation control blocks (n = 10, P < 0.00001). A: parasagittal view centered through the dorsolateral prefrontal cortex (DLPFC) activation (at Talairach coordinates x = 33, y = 40, z = 37). B: coronal view through the right hemisphere DLPFC activation. Right of the image is the right hemisphere. C: functional map was rendered onto the Talairach transformed cortical surface of a subject (right hemispherelateral and medial views shown). Gyri are represented by lighter areas, and sulci, by darker areas.

View larger version (15K): [in this window] [in a new window]   Fig. 4. Comparison of the pro- and anti-saccade blocked design experiment. Data were averaged across each pro- or anti-saccade epoch within each subject before averaging across the 10 subjects. The fixation control blocks were defined as the baseline. Error bars signify the SE across subjects. **P < 0.01 for the paired t-test comparison across subjects; *P < 0.05.

Eye movement recordings

In those subjects in whom eye movements were recorded, no extraneous eye movements were found during the instruction period (blinks excluded). Figure 5 shows the three instruction time periods and the subsequent pro-saccade or anti-saccade eye movements after the peripheral stimulus appearance (T). The demonstration of no extraneous eye movements during the instruction periods suggests that differences found in the BOLD signals for pro-saccade or anti-saccades could not be due to different numbers of eye movements made during those periods. Figure 5B shows the eye traces for 1,000 ms after the peripheral stimulus onset (the gray region in Fig. 5A).

View larger version (20K): [in this window] [in a new window]   Fig. 5. Eye movement traces from a subject's scanning session during 1 event-related experiment scan. A: data were aligned to the onset of the peripheral stimulus after the 3 instruction periods (6, 10, and 14 s). This was representative of the subjects in which we recorded eye movements. Blinks were removed from the eye traces and were not replaced with smoothed data, thus there were some short spaces in the eye position traces. B: eye traces for 1,000 ms after the peripheral stimulus onset (from the gray region in Fig. 5A). Eh, horizontal eye position; FP, fixation point; T, peripheral stimulus.

Event-related signals during pro- and anti-saccades

We used these functionally localized areas from the blocked experiment to analyze the BOLD signal intensities from the event-related experiments. We examined the same brain areas as were previously shown in Fig. 4 for the blocked experiment except now we examined the signal intensity from the event-related experiments. The signal intensity for the three instruction times periods (shown in Fig. 2) was then averaged within each subject. The averaged time course in Fig. 6A was used for demonstration purposes; the statistics for the instruction period was computed from the instruction periods within the gray regions in Fig. 2. The left panel shows the two waveforms for pro- and anti-saccades diverging after the instruction stimulus in left hemisphere FEF. This instruction stimulus change from a white fixation cross to a green (or red) stimulus did not produce a significant difference in V1/V2 (data not shown), which suggests that these differences in FEF were not due to lower level stimulus properties. When the instruction periods were averaged across subjects (as described in METHODS), a significantly higher signal intensity was found for the anti-saccade instruction period compared with the pro-saccade instruction period (t₍₉₎ = 3.00; P < 0.05; Fig. 6A, right panel). This increase in signal intensity for the anti-saccade instruction was carried forward in time as the peripheral stimulus was flashed and the appropriate motor response was made in Fig. 6B (left panel). This instruction period difference in signals was evident between the anti- and pro-saccade waveforms at time point zero in Fig. 6B. The right panel shows the average across subjects for the movement period. There was a significantly higher signal for the anti-saccade movement period when compared with the pro-saccade movement period (t₍₉₎ = 2.65; P < 0.05).

View larger version (33K): [in this window] [in a new window]   Fig. 6. Averaged BOLD signals (n = 10) during the event-related experiment from left hemisphere FEF. A: instruction onset stimulus occurred at the 0 time point. This time point was also defined as the baseline activation for Fig. B and C. The right panel shows the instruction periods averaged across subjects as described in Fig. 2. P values represent significance level for the 2-tailed paired t-test comparison. B: data aligned on the peripheral stimulus onset. SE bars signify the SE across subjects. C: data in B that were baseline corrected at the 2-s time point.

To examine whether this difference in signal (anti > pro) was indeed due to the appearance of the peripheral stimulus and saccade execution or due to the residual difference remaining from the previous instruction period, we baseline corrected the signal at the two second time point within each subject in Fig. 6B. This time point was chosen to accommodate the hemodynamic response from the previous preparatory activity. Similar results were obtained (data not shown) when we baseline corrected the signal at the 4-s time point within each subject. The results of this analysis show that there was no signal difference across subjects (t₍₉₎ = 0.35; P = 0.74; Fig. 6C, right panel). This suggests that any differences during the movement period (Fig. 6B) were due to a residual difference from the previous instruction period.

The instruction period for the right hemisphere FEF resembled the left hemisphere's activation pattern, with the anti-saccade instruction signal being greater than the pro-saccade instruction signal (Fig. 7A, left panel: t₍₉₎ = 2.72; P < 0.025; Fig. 7A, right panel). Similarly, the movement period showed a significantly higher signal intensity for anti-saccades compared with pro-saccades before baseline correction (t₍₉₎ = 3.24; P < 0.01; Fig. 7B, right panel) but not after the movement baseline correction (t₍₉₎ = 0.88, P = 0.40, Fig. 7C, right panel).

View larger version (35K): [in this window] [in a new window]   Fig. 7. Same as in Fig. 6 except that data are from right hemisphere FEF.

Besides the FEF, the only other brain area that showed a significantly increased signal for the anti-saccade instruction compared with the pro-saccade instruction was the right hemisphere DLPFC (Fig. 8A). The difference between the anti-saccade instruction signals and the pro-saccade instruction signals was not as striking as in FEF but nonetheless was significant (t₍₉₎ = 2.55; P < 0.05; Fig. 8A, right panel). The movement period analysis also showed the identical pattern to that of FEF, with a significantly increased signal for anti-saccades compared with pro-saccades during the movement period (t₍₉₎ = 3.12; P < 0.01; Fig. 8B, right panel), but in the movement baseline correction analysis, this difference disappeared (t₍₉₎ = 0.36; P = 0.72; Fig. 8C, right panel).

View larger version (36K): [in this window] [in a new window]   Fig. 8. Same as in Fig. 6 except that data are from right hemisphere DLPFC.

The parietal regions showed a trend toward a higher instruction period signal for anti-saccade compared with pro-saccade trials (left IPS: t₍₉₎ = 2.26, P = 0.05; right IPS: t₍₉₎ = 2.11, P = 0.06; Fig. 9A, center panel). The left hemisphere IPS was the only other area (from Fig. 9) that showed a significantly higher movement period for anti-saccades compared with pro-saccades (t₍₉₎ = 2.76; P < 0.05; Fig. 9B, center panel), but this movement period difference disappeared after movement baseline correction (t₍₉₎ = 0.67; P = 0.52; Fig. 9C, center panel). There were no statistically significant differences in SEF and V1/V2, with the later demonstrating that low-level physical properties of the fixation stimulus did not confound the instruction period differences.

View larger version (17K): [in this window] [in a new window]   Fig. 9. Averaged BOLD signals (n = 10) during the event-related experiment from supplementary eye fields (SEF), intraparietal sulcus (IPS), and calcarine sulcus (V1/V2). Same as in Fig. 6 except only the right panels were shown from Fig. 6, A-C, for the respective brain areas. P values represent significance level for the 2-tailed paired t-test comparison.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In this study we provided evidence that areas in the human frontal lobe carry preparatory set-related activity for anti-saccades. The left and right FEF and the right DLPFC showed a greater activation during the instruction period on anti-saccade trials than on pro-saccade trials. The differences in the right and left IPS approached significance. No differences were found during this period in SEF or in visual areas lining the calcarine sulcus (V1/V2). The comparison of activation between pro- and anti-saccades evoked by the presentation of the peripheral stimulus and the saccade revealed no differences in any of the investigated areas (DLPFC, FEF, SEF, IPS, V1/V2) after the movement period was baseline corrected. The increased activation in frontal areas during anti-saccade trials compared with pro-saccade trials is consistent with previous PET and fMRI data (Connolly et al. 2000; Doricchi et al. 1997; Kimmig et al. 2001; O'Driscoll et al. 1995; Paus et al. 1993; Sweeney et al. 1996). The finding that FEF and DLPFC only show differences during the instruction period, but not in response to the peripheral stimulus and the saccade, suggests that the differences between anti- and pro-saccades that are typically found in blocked designs originate primarily from modulations of preparatory-activity and not movement-related activity.

We hypothesize that the differences in FEF and DLPFC during the instruction period reflect differences in subjects' preparatory set (Evarts et al. 1984; Hebb 1972) between anti- and pro-saccade trials. Single neuron recordings in the SEF and FEF in nonhuman primates have previously demonstrated preparatory set-related activity for saccades. Schlag-Rey et al. (1997) reported that SEF neurons that had stronger visual responses for anti-saccades already showed a higher activation level on anti-saccade trials before the stimulus appears. Everling and Munoz (2000) recorded from identified corticotectal FEF neurons and found that these saccade-related neurons had lower discharge rates during the instruction period on anti-saccade trials than on pro-saccade trials. The same was found for saccade-related neurons in the superior colliculus (SC) (Everling et al. 1999). The differences in FEF and SEF discharge were interpreted as evidence that the correct performance of the anti-saccade task depends on a top-down control of the SC (Everling and Munoz 2000; Schlag-Rey et al. 1997). Saccade-related neurons in the SC had a higher level of preparatory activity and a strong visual burst on trials when the monkey failed to suppress a reflexive saccade toward the stimulus (Everling et al. 1998).

Neurons with preparatory set activity are also common in the lateral prefrontal cortex (Fuster et al. 1982; Quintana and Fuster 1992), and rule learning and representation are considered one of its cardinal functions (Passingham 1993; Wise et al. 1996). In a particular relevant example, Asaad et al. (2000) recently reported that prefrontal neurons show task-dependent differences in their baseline activity. Interestingly, the authors hypothesized that this activity could provide a signal that allows conflicting sensory input to be mapped to the appropriate motor output (see Miller and Cohen 2001 for a model of prefrontal function). In accordance with this model, we hypothesize that the increased activation in the FEF and DLPFC on anti-saccade trials represents top-down control signals that are involved in suppressing preparatory saccade-related activity in the SC to avoid a reflexive pro-saccade toward the stimulus.

How can the increased activation of the FEF in the current fMRI study be reconciled with the reduced discharge rate in the previous single neuron recording study (Everling and Munoz 2000)? A recent comparison between neural discharge rate, local field potentials, and the BOLD effect has shown that the BOLD signal is better correlated with local field potentials than with neural discharge rate (Logothetis et al. 2001). The authors concluded that activation in fMRI is more likely to reflect the input to an area and processing within an area than the output signal of an area. Therefore the activation that we found in the FEF may rather indicate inhibitory input into the FEF than an increased output coming from the FEF on anti-saccade trials. Further direct comparisons between fMRI and single neuron recordings are needed to resolve the discrepancies between the two techniques.

We were surprised to see no differences in the activation evoked by stimulus presentation and saccadic eye movement between pro- and anti-saccades in the FEF and SEF, once we subtracted the existing differences in preparatory set-related activity. Single neuron recordings have found that almost all saccade-related neurons in the FEF have higher motor bursts for pro-saccades than for anti-saccades (Everling and Munoz 2000). At the moment, we can only speculate about why we did not detect any differences in our study. The simplest explanation would be that FEF neurons in humans do not have different motor bursts for pro- and anti-saccades. We believe that this is unlikely, given the close resemblance of pro- and anti-saccades in humans and monkeys (Amador et al. 1998; Bell et al. 2000; Everling and Fischer 1998). A possible scenario for similar activation between pro- and anti-saccades could be that the lower motor burst of FEF neurons for anti-saccades is associated with the activation of a larger number of FEF neurons. However, there is currently no evidence from single neuron electrophysiology to support this hypothesis. Another explanation may be that fMRI is not sensitive enough to detect the relatively small difference in discharge in FEF neurons between pro- and anti-saccades (approximately 80 spikes/s for pro-saccades vs. approximately 50 spikes/s for anti-saccades) that lasts for <100 ms (Everling and Munoz 2000). In addition to the small differences, it has been found that the activity of other cells in the FEF are suppressed prior to saccade onset if the saccade target is located outside of their response field (Everling and Munoz 2000; Schall et al. 1995). Indeed, an optical imaging study employing electrical stimulation of FEF and neighboring area 8Ar showed a rapid and widespread depolarization followed by a large and prolonged hyperpolarization (Seidemann et al. 2002). Thus the metabolic activity associated with the suppression of a large population of cells may be stronger than the small differences in motor activity between pro- and anti-saccades and may prevent the detection of such differences with fMRI.

Our study has demonstrated significant differences in activation levels in frontal brain areas during the instruction period between pro- and anti-saccades. We suggest that the increased activation on anti-saccade trials reflect the preparatory set that is necessary to suppress incorrect reflexive saccades. Dysfunction of these areas may disturb this top-down control and may underlie the poor performance of patients with schizophrenia (Clementz et al. 1994; Fukushima et al. 1990) and frontal lobe lesions (Fukushima et al. 1994; Guitton et al. 1985; Pierrot-Deseilligny et al. 1991; Walker et al. 1998) in the anti-saccade task.