| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1Department of Psychology, 2Graduate Program in Neuroscience, 3Department of Physiology and Pharmacology, University of Western Ontario, London, Ontario N6A 5C1; and 4Robarts Research Institute, London, Ontario N6A 5K8, Canada
Submitted 16 April 2003; accepted in final form 23 September 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
, 1991; Fuster 1991; Hikosaka and Wurtz 1983; Pierrot-Deseilligny et al. 2002). This task requires a subject to fixate a central fixation point while a stimulus is briefly flashed in the subject's peripheral vision. The subject must suppress the reflex to look at the stimulus and must also remember the stimulus's location over an ensuing delay period, which can last from a few hundred milliseconds to tens of seconds. At the end of the delay period, a cue, usually the offset of the central fixation point, instructs the subject to make a memory-guided saccade to the blank space where the stimulus was previously flashed. This task allows for the separation of sensory processing of the flashed stimulus from motor processes associated with saccade generation, and it has been used for this purpose extensively in neuronal recording studies in nonhuman primates (for examples see Barash et al. 1991; Bruce and Goldberg 1985; Hikosaka and Wurtz 1983). In addition, the memory-guided saccade task introduces cognitive demands in the form of working memory during the delay period (Baddeley 1986). Working memory is a form of shortterm memory involving on-line manipulation and maintenance of information no longer available from sensory input, and it is important for diverse activities from playing chess to washing dishes (Baddeley 1983, 1986; Levy and Goldman-Rakic 2000).
To investigate working memory processes, previous functional imaging studies used positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) to compare human brain activation during the performance of memory-guided saccades and visually guided saccades, which require a subject to look at a visible stimulus and presumably do not recruit working memory for stimulus location. Three cortical areas, the parietal eye field (PEF), frontal eye field (FEF), and supplementary eye field (SEF), exhibit greater activity on memory-guided saccades than on visually guided saccades (Anderson et al. 1994; Greenlee et al. 2001; Sweeney et al. 1996a,b). Activity in the prefrontal cortex is more ambiguous. Anderson and colleagues (1994) found that visually and memory-guided saccades evoked equivalent regional cerebral blood flow in the middle frontal gyrus, whereas Sweeney and colleagues (1996b) found greater regional cerebral flow for memory-guided saccades than for visually guided saccades in this region.
Previous functional imaging studies used blocked trial designs, in which subjects performed blocks of saccade trials, all memory-guided or all visually guided, and the experimenters then compared activation summated across entire blocks. Because blocked designs do not allow us to localize differences in brain activity to specific epochs within the task sequence, previous imaging studies were unable to determine whether greater activity for memory-guided versus visually guided saccade blocks was related to stimulus presentation, memory-function during the delay period, saccade execution, or a combination of these. Here, we used event-related fMRI to compare the activation time courses from individual visually and memory-guided saccade trials, allowing us to determine during which epochs of the trial sequence activation differences arise between memory- and visually guided saccades for different brain areas. Our hypothesis was that greater activation in cortical saccade regions during blocks of memory-guided saccades is at least partly attributable to memory processes occurring in the delay period and that those memory processes would be reflected by greater activation in the event-related memory-guided saccade task than in the visually guided saccade task during the delay period.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Subjects
Nine subjects, 4 female, ages 21 to 31 yr with a mean age of 25, participated in our study. All subjects gave informed written consent before testing. Subjects did not report any history of neurological disorder, all had normal vision or corrected-to-normal vision, and all described themselves as right-handed.
Scanning procedure
Scans were performed at the Robarts Research Institute in London, Ontario using a 4-tesla Varian (Palo Alto, CA) Unity Inova whole body MRI scanner equipped with a Siemens Sonata Gradient system (Siemens, Erlangen, Germany) and a quadrature head coil. We collected functional data using blood oxygenation level-dependent (BOLD) navigator echo-corrected, T2*-weighted, segmented gradient echoplanar imaging [EPI; 11 slices, 64 x 64 matrix, 19.2 x 19.2 cm in-plane field of view (FOV), 5 mm slice thickness, 0 mm gap, echo time (TE) of 15 ms, flip angle of 30°; 2 segments per slice, resulting in a total volume acquisition time of 1.0 s and an EPI voxel size of 3 x 3 x 5 mm]. Slices were oriented axially. 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 3D T1-weighted anatomical images of the brain collected immediately after the functional scans using the same in-plane FOV (128 slices, 256 x 256, voxel size = 0.75 x 0.75 x 1.25 mm, TE = 5.4 ms, TR = 9.8 ms, flip angle = 15°).
Stimulus presentation and eye tracking procedure
Stimulus presentation was performed using a Silent Vision SV-4021 projection system (Avotec, Stuart, FL). This system includes an MEyeTrack-SV (Silent Vision) eye tracker (SensoMotoric Instruments GmbH, Teltow, Germany). This device uses fiber optics housed in dual stalks that sit in front of a subject's eyes, allowing presentation of visual stimuli and simultaneous CCD video-based infrared eye tracking. The visual display subtends 30° horizontally by 23° vertically with a resolution of 800 x 600 pixels and a refresh rate of 60 Hz. Eye position recordings were made at a 60-Hz sampling rate. Before scanning, we calibrated the system with a 5-point system native calibration.
Training
On the day before scanning, each subject was instructed on the tasks and practiced the tasks for 15 min. On the day of scanning, we explained the tasks to subjects again and gave them 5 min of practice during the initial scanner setup. To verify correct task performance, we observed subjects' eye movements during scanning on a computer monitor and recorded them for later off-line analysis.
Blocked memory- and visually guided saccade paradigm
In our first experiment, subjects performed memory-guided saccades and visually guided saccades arranged in a blocked design. We ran this experiment to confirm results from previous studies employing similar paradigms (Anderson et al. 1994; Greenlee et al. 2001; Sweeney et al. 1996a,b) and to compare the results from the event-related experiment described below with those obtained using a blocked design in the same subjects.
Functional scans for the blocked experiment consisted of four 32-s blocks of trials interleaved with five 32-s fixation intervals for a total of 296 s. Each block was composed of 8 memory-guided saccade trials or 8 visually guided saccade trials, and each run included 2 visually guided saccade blocks and 2 memory-guided saccade blocks with order randomized across runs and subjects. During fixation intervals, the subject fixated a central fixation circle. The individual trials consisting of the visually and memory-guided saccade blocks lasted 4 s. At the start of visually guided trials, the subject fixated a white central fixation cross for between 2,250 and 2,750 ms (uniform distribution), after which the fixation cross disappeared and a peripheral stimulus appeared simultaneously (Fig. 1A, left panel). The peripheral stimulus remained visible for 200 ms, during which the subject made a visually guided saccade to the stimulus followed by an immediate saccade back to the visual display's center. By 1,050 to 1,550 ms after stimulus disappearance, the fixation cross reappeared, starting the next trial. In memory-guided saccade trials, the subject initially fixated the fixation cross for 1,050 to 1,550 ms (Fig. 1A, right panel). A peripheral stimulus was then flashed for 200 ms, and the subject had to maintain central fixation and remember the location of the flashed stimulus. After a 1,000-ms delay period, the fixation cross disappeared, cuing the subject to execute a memory-guided saccade to the remembered stimulus location and then immediately to return gaze to center. The reappearance of the fixation cross 1,250 to 1,750 ms after fixation cross offset indicated the start of the next trial. Each subject completed 4 functional scans in the blocked experiment for a total of 8 memory-guided and 8 visually guided saccade blocks.
|
Event-related memory- and visually guided saccade paradigm
In the second experiment, we measured the BOLD response to single memory- and visually guided saccade trials arranged in an event-related design. Each trial lasted 13 s and was followed by a 12-s fixation interval. Individual scanning runs lasted 312 s and included 6 visually guided and 6 memory-guided trials in random order interleaved with thirteen 12-s fixation intervals, during which the subject fixated a central fixation circle. Each trial started when the circle changed to a green or red cross, where color identified the trial as either a visually or memory-guided one (Fig. 2). For 5 subjects, a red cross represented a memory-guided trial, and for 4 subjects, a red cross represented a visually guided trial. Subjects fixated the central cross for 2 s, after which a peripheral stimulus was flashed for 200 ms at one of the 4 locations described above for the blocked design experiment (Fig. 1B). Subjects had to maintain central fixation during the stimulus flash. On visually guided saccade trials, the stimulus was a distractor, and subjects were instructed to ignore it, whereas on memory-guided saccade trials, subjects had to remember the stimulus's location for the subsequent saccade. After a 9.8-s delay period, the fixation cross disappeared. On visually guided saccade trials, a second stimulus was flashed for 200 ms starting on fixation cross offset, and the subject made a visually guided saccade to this stimulus. The second stimulus appeared at one of the 4 locations described above, and this location was unrelated to the distractor stimulus's location so that subjects were not able to use the distractor to predict where the second target would appear. On memory-guided trials, no second stimulus was presented, and fixation cross offset was the cue to make a memory-guided saccade to the remembered target location. The central fixation circle, the central fixation cross, and peripheral stimuli had the same dimensions as in the blocked experiment. Stimulus locations were randomized across subjects. For 3 of the 9 subjects, we tested the effects of different delay period lengths. Those subjects performed 8 visually guided and 8 memory-guided trials for each of 3 different delay lengths including 5.8, 7.8, and 9.8 s. We found that using the 9.8-s delays allowed for better separation of BOLD activation for delay-related processes and saccade execution than using the 5.8- and 7.8-s delays, and the remaining 6 subjects each made a total of 24 visually guided and 24 memory-guided saccades all with 9.8-s delay periods.
|
We used the BrainVoyager v4.6.3 software package (Brain Innovations, Maastricht, The Netherlands) for data analysis. Functional data were superimposed on anatomical brain images, aligned on the anterior commissure-posterior commissure line, and transformed into Talairach space (Talairach and Tournoux 1988). Functional data were spatially smoothed in the Fourier domain using a low-pass filter with a cutoff of 0.2414 cycles/mm, which produces smoothing similar to that of a nonlinear filter with a Gaussian kernel of 4-mm function width at half-maximum in the spatial domain. Functional data were also corrected for temporal linear trends and temporally smoothed using a band-pass filter in the Fourier domain with respective windows for blocked and event-related runs of 2-75 cycles/run (0.0068-0.2534 Hz) and 6-75 cycles/run (0.0192-0.2404 Hz). To preserve the low-frequency components of the data, we used different high-pass filter values (2 cycles/run for the blocked experiment and 6 cycles/run for the event-related experiment) because each scanning run in the blocked experiment contained 4 blocks, whereas each scanning run in the event-related experiment contained 12 trials.
Blocked experiment analysis
We first used Matlab v6.1 (The MathWorks, Natick, MA) to extract all error blocks and error trials from the functional data files. To compare BOLD activation from memory- and visually guided blocks or trials, we used the general linear model (GLM) as implemented in BrainVoyager, incorporating all nonrejected data from all subjects normalized to Talairach space.
For the blocked experiment, we used 2 predictor (or regressor) curves per subject, one for memory-guided saccade blocks and one for visually guided saccade blocks, resulting in a total of 18 predictor curves. We used separate predictors for each subject to balance subjects' influence on the results, given that they contributed different amounts of data after exclusion of error blocks. Predictor curves were the convolution of boxcar functions, consisting of 32-s-duration square waves centered on each block of the appropriate type for a given subject, with BrainVoyager's model of the hemodynamic response, based on Boynton and colleagues (1996). BrainVoyager also automatically added a constant predictor curve for each functional data file to account for between-run signal baseline differences. The set of predictors was fitted to the blocked experiment functional data using least-squares regression to produce a beta weight for each predictor curve for each voxel. To compare memory- and visually guided saccade block activation, we used a contrast function that was the sum of the beta weights for all memory-guided saccade block predictors minus the sum of the beta weights for the visually guided saccade block predictors from all 9 subjects to compute a t-statistic for each voxel. The resulting map was Bonferroni-corrected for multiple comparisons over the voxel population by multiplying the P-values by 44,303 and thresholded at t = 4.871, resulting in a corrected P < 0.05 [degrees of freedom (df) = 6,863]. Based on the statistical map, we defined significantly activated brain regions and used anatomical landmarks to identify them. We included in our analysis only those regions with volumes 
100 mm3.
We also computed mean activation differences between memory- and visually guided saccade blocks for each region. For each subject, we derived percentage signal change activation time courses from the mean activation values taken across all voxels in a region using the 16-s interval occupying the last half of the preceding interblock interval as the baseline for each block. We averaged the time courses across all runs in a given subject and then calculated the mean activation across the 32-s block. From these values we computed the difference between memory- and visually guided saccade block activation for each subject, and we took the between-subject mean difference as a measure of the magnitude of the BOLD signal difference between the memory- and visually guided saccade blocks.
Event-related experiment analysis
For the event-related experiment, we modeled BOLD activation using the GLM with 6 predictor curves for each subject. Three curves modeled stimulus-, delay-, and saccade-related BOLD activation for memory-guided saccade trials, and 3 curves modeled stimulus-, delay-, and saccade-related activation for visually guided saccade trials. We used separate subject predictors to balance subjects' influence on the results given exclusion of error trials. Each predictor curve was a series of Gaussian curves aligned on specific trial epochs associated with stimulus-, delay-, or saccade-related BOLD activation (Fig. 3). Predictors modeling stimulus-related activation had normal curves centered 4.5 s after stimulus presentation. Predictors modeling delay-related activation had normal curves centered 7.5, 8.5, and 9.5 s after stimulus presentation for trials with 5.8-, 7.8-, and 9.8-s delay lengths, respectively. Predictors modeling saccade-related activation had normal curves centered 10.5, 12.5, and 14.5 s after stimulus presentation for 5.8-, 7.8-, and 9.8-s delays, in that order. The values for the means of the Gaussian curves were based on visual inspection of mean activation traces computed from the event-related data using the statistical activation map from the blocked experiment. We found that most regions exhibited bimodal BOLD signal traces with the first mode occurring 4 to 5 s after stimulus presentation and the second mode occurring 4 to 5 s after saccade execution at the end of the delay period. On this basis, we chose to center the predictors for stimulus- and saccade-related activity 4.5 s after stimulus presentation and saccade execution, respectively. We centered the predictor for delay-related activity halfway between those for the stimulus and saccade predictors. All Gaussian curves had a 2-s SD. BrainVoyager also added constant predictor curves to model mean activation for each functional data file. The full set of predictors was fit to the functional data by least-squares regression.
|
100 mm3. We also did control comparisons of the delay period predictor weights from the memory-guided saccade task to baseline and the delay predictor weights from the visually guided saccade task to baseline. We constructed 2 maps using contrast functions, which were the sum of all 9 subjects' delay-period predictor weights for the memory-guided saccade task or for the visually guided saccade task, as appropriate. We used a threshold of t = 4.803 (df = 8,812), giving a Bonferroni-corrected P < 0.05. For each region activated in the delay comparison (subtraction) map, we then computed the degree of overlap, as a percentage, of the given region with voxels found to be active in the 2 control maps.
Our decision to model the event-related data using 6 Gaussian curves per subject, rather than representing neuronal activity with square waves and convolving them with an estimate of the hemodynamic response to model the BOLD signal, was based on the difficulty of defining appropriate square waves given current limitations in our understanding of the neuronal processes involved in performing the memory-guided saccade task. Neuronal recording studies using this task in nonhuman primates have found neurons that discharge at high rates during stimulus presentation or saccade execution as well as neurons that maintain tonic, elevated discharge rates during the delay period (Barash et al. 1991; Bruce and Goldberg 1985; Funahashi et al. 1989, 1991; Russo and Bruce 1996). Many neurons exhibit a combination of 2 or all 3 of these activity patterns. However, the exact numbers of neurons exhibiting specific discharge profiles in various regions, their distributions, and their separate contributions to the BOLD signal are unknown. It is also unknown whether sustained delay-period activity represents a memory for stimulus location, a motor plan, or some other computational structure or combination of structures. Given these issues, we decided to use evenly spaced Gaussian curves with identical shapes centered on different parts of the trial sequence to model stimulus-, delay-, and saccade-related activity for the memory- and visually guided saccade tasks.
To illustrate the magnitudes of the effects tested for by the above stimulus, delay, and saccade comparisons, we computed mean differences between BOLD activation levels for memory- and visually guided saccade trials related to stimulus, delay, and saccade for each region. For each subject, we derived percentage signal change activation time courses from the mean activation values taken across all voxels in a region using the interval from 0 to 2 s, where 0 s marks trial start, as the baseline for each trial. We computed the average activation over the 2 time points occupying the peak of each predictor curve for each subject. From these values, we computed the difference between memory- and visually guided saccade trial activation related to stimulus, delay, and saccade for each subject, and we took the between-subject mean differences as a measure of the magnitude of the BOLD signal difference between the memory- and visually guided saccade trials associated with stimulus, delay, and saccade.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Based on recorded eye movement information (Fig. 4), we excluded saccade blocks or saccade trials with errors. Eye position traces indicated that subjects were able to maintain fixation during target presentation, delay period, and intertrial interval, as well as interblock interval for the blocked experiment. They could also make correct visually and memory-guided saccades. For the blocked experiment, we excluded blocks containing one or more trials for which these criteria were not met, and we excluded single trials with errors for the event-related experiment.
|
|
We excluded 5 functional runs out of a total of 36 from the blocked experiment because they were damaged by equipment trouble. For the event-related experiment, we excluded 9 out of a total of 432 trials containing artifacts in the fMRI signal caused by equipment trouble. We also excluded 3 trials containing head movements by the subject, where a head movement was defined as a cranial displacement of 0.5 mm between successive functional images.
Of all trial blocks collected from all 9 subjects in the blocked experiment, 25.0% were excluded for one or more of the above reasons. Of all trials collected from all 9 subjects in the event-related experiment, 15.1% were excluded for one or more of the above reasons.
It would have been interesting to compare BOLD activation from correct event-related memory-guided saccade trials and memory-guided trials in which subjects executed saccades to the wrong location. Unfortunately, this type of error was rare, accounting for only 4.5% of the total errors on memory-guided saccades in the event-related experiment (Table 1, line 5), and we did not have enough data to make a meaningful comparison.
Blocked experiment
We compared BOLD activation for memory- and visually guided saccade blocks using a GLM across all 9 subjects' data aligned in Talairach space. The following regions exhibited significantly greater activation for memory-guided than for visually guided saccade blocks (P < 0.05, Bonferroni-corrected, t 4.871, df = 6,863) (Fig. 5, A-F; Table 2): middle frontal gyrus (MFG), posterior inferior frontal gyrus (pIFG), frontal eye field (FEF), supplementary motor area including supplementary eye field (SEF), posterior medial prefrontal cortex contiguous with the supplementary motor area, rostral intraparietal sulcus (rIPS), caudal intraparietal sulcus, right ventral intraparietal sulcus (vIPS) where it meets the transverse occipital sulcus, right temporo-parieto-occipital junction (TPOJ), precuneus, dorsal part of the caudate body, and thalamus. All regions were active bilaterally unless otherwise noted. In accordance with Paus (1996), FEF extended from the surface of the precentral sulcus to its depth where it joins with the posterior fundus of the superior frontal sulcus (Fig. 5, A and C-E; Table 2, lines 7-8). The medial aspect of the brain featured a large activated area encompassing the supplementary motor area, the anterior portion of which contains SEF (Carter and Zee 1997; Grosbras et al. 1999; Schlag and Schlag-Rey 1987), as well as the posterior part of the medial frontal gyrus (Fig. 5, B-E). The ventral border of this medial activation focus extended into the cingulate sulcus, but it did not extend onto the cingulate gyrus. The rostral part of the intraparietal sulcus contained an activation focus, which we called rIPS (Fig. 5, A and E; Table 2, lines 12-13) and which was consistent with foci of activation identified as the parietal eye field (PEF) by Muri and colleagues (1996a), Kawashima and colleagues (1996), and Rushworth and colleagues (2001). We were unable to analyze primary visual cortex because the functional imaging slice plan did not extend low enough to include it.
|
|
|
-4.871, df = 6,863). These areas were located in the left frontal pole (Table 2, line 22), in the anterior cingulate gyrus on both sides of the midline (Table 2, line 23), in right anterior medial frontal gyrus (Table 2, line 24), and in the ventral bank of the left inferior frontal sulcus midway along its anteroposterior extent (Table 2, line 25). Event-related experiment
To analyze the event-related experiment, we compared the BOLD signal from memory- and visually guided saccade trials at 3 different epochs within the task sequence including the stimulus epoch at the beginning of the delay period, the delay epoch at the end of the delay period, and the saccade epoch at the beginning of the intertrial interval after saccade execution. We used a GLM with 6 predictor curves for each subject. Two predictor curves modeled stimulus-related BOLD activation for memory- and visually guided saccade trials, 2 predictors modeled delay-related BOLD activation for memory- and visually guided saccade trials, and 2 predictors modeled saccade-related activation for the 2 trial types (see METHODS for details).
Stimulus comparison for event-related experiment
We compared predictors for stimulus-related activation from the memory- and visually guided conditions and found no regions that were significantly more active for memory-guided than for visually guided saccade trials (P > 0.05, Bonferroni-corrected, t < 5.021, df = 8,812). We did find several regions that were somewhat more active for visually guided than for memory-guided saccade trials (P < 0.05, Bonferroni-corrected, t -5.021, df = 8,812) (Fig. 7, A-F; Table 3, lines 1-5). Prefrontal cortex contained 3 small regions of activation located in right superior frontal gyrus (SFG), in right MFG, and in the medial part of pIFG (Fig. 7, A-D; Table 3, lines 1-3). Activation in right MFG was located slightly posterior to and did not overlap with the right MFG region defined from the blocked experiment. The pIFG region was located in the ventral bank of the fundus of the right posterior inferior frontal sulcus. The fundus of right rIPS contained a large activated region (Fig. 7, E and F; Table 3, line 4). Although this region was located at the same anteroposterior position within the intraparietal sulcus as the rIPS region defined from the blocked experiment, the two regions did not overlap, given that the area from the event-related experiment was situated deeper within the intraparietal sulcus than the area from the blocked experiment. The map also included an activated focus in right precuneus (Table 3, line 5).
|
|
In a comparison of the predictors modeling activation during the delay period, the following regions displayed significantly stronger activation for the memory-guided than for the visually guided saccade task (P < 0.05, Bonferroni-corrected, t 5.021, df = 8,812) (Fig. 8, A-F; Table 3, lines 6-13). We found a large activation focus in pIFG on the right side in the same location as that found to be active in the blocked experiment (Fig. 8, A and C; Table 3, line 6). The small medial pIFG region activated in the stimulus predictor comparison was located on the medial edge of this large pIFG region. SEF was active bilaterally around the paracentral sulcus on the medial aspect of the frontal lobe (Fig. 8, B and D; Table 3, line 7) in accordance with Grosbras and colleagues (1999). An activation focus in right medial FEF (mFEF) occupied the fundus of the right medial precentral sulcus where it joins the superior frontal sulcus (Fig. 8, B and D; Table 3, line 8). We found 3 activated foci in the right postcentral sulcus and adjoining rostral intraparietal sulcus (Fig. 8, E and F; Table 3, lines 9-11). Two of the foci fell within the postcentral sulcus, one near the cortical surface and one in the fundus of the sulcus. The other focus occupied rIPS at a depth intermediate between the cortical surface and the fundus (Fig. 8F). This region was located in the same position as and overlapped almost entirely with the rIPS region found to be active in the blocked experiment. Right vIPS where it crosses the transverse occipital sulcus contained an activated region in its fundus (Table 3, line 12). We found a similar region of activation in the blocked experiment, but that region was located more superficially in the intraparietal sulcus near the cortical surface. There was also a small, activated region in the right supramarginal gyrus. The delay comparison revealed no activation in left prefrontal cortex, left FEF, or left intraparietal sulcus, nor were any regions significantly more active for visually guided versus memory-guided saccade trials in this comparison (P > 0.05, Bonferroni-corrected, t < 5.021, df = 8,812).
|
|
We did not find an activation difference between memory- and visually guided saccades during the delay period in MFG or left pIFG. We examined activation patterns from the event-related experiment in the right MFG and left pIFG regions identified in the blocked experiment. Both left pIFG and right MFG exhibited bimodal activation profiles that were qualitatively similar to that of right pIFG. Both regions had slightly greater activation for memory-guided than for visually guided saccades during the delay period, but these differences were not significant because no voxels within left pIFG or right MFG were found to be active in the delay comparison.
Saccade comparison for event-related experiment
We compared saccade-related activation from the memory- and visually guided conditions and found that the following regions displayed greater saccade-related BOLD signal on memory-guided versus visually guided saccade trials (P < 0.05, Bonferroni-corrected, t 5.021, df = 8,812). A large region of activation was located in the right precentral gyrus (Fig. 10, A and B; Table 3, line 14). A large area was active in the anterior end of right rIPS where it joins the postcentral sulcus and extended back into right rIPS (Fig. 10, A-C; Table 3, line 15). This region was located superficially within the sulcus and did not overlap with the rIPS region activated in the blocked experiment. The left postcentral sulcus contained 2 activated foci, one superficial and one deep (Table 3, lines 16-17). No regions were more active on visually guided than on memory-guided saccade trials in the saccade comparison (P > 0.05, Bonferroni-corrected, t < 5.021, df = 8,812).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
; Luna and Sweeney 1999; Sweeney et al. 1996b). Because blocked designs compare activation from entire blocks, not individual trials or components of trials, these studies could not discern whether greater activation on memory-guided saccade blocks was attributable to processes associated with stimulus presentation, the delay period, saccade execution, or a combination of these. To examine this issue, we used an event-related trial design to compare fMRI BOLD signal from visually and memory-guided saccades. We identified various cortical and subcortical regions expressing greater BOLD activation in either the memory-guided or the visually guided saccade trials during the stimulus epoch, the delay period epoch, or the saccade epoch. Unlike most previous imaging studies on memory-guided saccades, we recorded subjects' eye movements simultaneously with the collection of fMRI measurements and excluded error trials on the basis of the eye movement recordings. Blocked experiment
Our blocked experiment revealed a greater BOLD signal for memory-guided saccade blocks than for visually guided saccade blocks in a large number of cortical and subcortical regions. Previous functional imaging studies using PET found similar results (Anderson et al. 1994; Sweeney et al. 1996b), although they did not find activity in posterior inferior frontal gyrus (pIFG), as we did. Blocks of memory- and visually guided saccades were also compared using fMRI, and these results were published as abstracts (Greenlee et al. 2001; Sweeney et al. 1996a). The current project provides further replication of the PET results using fMRI.
Stimulus presentation and encoding processes for working memory
In the event-related experiment, comparison of stimulus-related activity in the beginning of the delay period revealed 3 small prefrontal regions, in right pIFG, right MFG, and right superior frontal gyrus (SFG), as well as an area in right rostral intraparietal sulcus (rIPS) with greater activation for visually guided than for memory-guided saccade trials. Although these results are not particularly strong, given the small size of the activation foci (Table 3), we propose the following. It is possible that subjects, having been instructed to ignore the distractor in the visually guided saccade task, were actively inhibiting the formation of a working memory trace for distractor location. By this view, the brain's default procedure is to store the locations of relevant objects in working memory, and to avoid formation of those memory traces requires active inhibition, which could be the function of the 4 brain regions listed above. Such a view is consistent with the hypothesis that the prefrontal cortex provides inhibitory control over computational processes occurring in other brain regions (for review see Miller and Cohen 2001). It is also revealing that the regions in right SFG, right MFG, and right rIPS that were activated in the event-related experiment were not activated in the blocked experiment, which did not present subjects with distractors that had to be ignored.
Delay period activity and maintenance functions in prefrontal cortex
In right pIFG, we found greater activation for memory-guided than for visually guided saccade trials during the end of the delay period. We interpret this as the signature of working memory processes during the delay. Previous functional imaging studies also found evidence for working memory processes in inferior frontal gyrus (for review see Duncan and Owen 2000; Goldman-Rakic 2000).
For 45.1% of the voxels in right pIFG, activation during the delay period in the visually guided saccade task undershot the baseline. Perhaps this reflects the fact that, during the delay period, subjects knew they did not have to remember any target locations for the duration of the trial, whereas during the intertrial interval they might have maintained a degree of preparedness to encode a new target location in working memory in case the subsequent trial proved to be a memory-guided saccade trial.
Much research in nonhuman primates has investigated the neural processes associated with working memory in the dorsolateral prefrontal cortex (dlPFC) (for reviews see Fuster 1997; Goldman-Rakic 1987). Neuronal recording studies in dlPFC in monkeys identified neurons that maintain tonic, elevated discharge rates during the delay period of the memory-guided saccade task when the stimulus is flashed in their response fields (Funahashi et al. 1989, 1991). Lesions and chemical inactivation of dlPFC impair memory-guided but not visually guided saccades (Funahashi et al. 1993; Sawaguchi and Iba 2001).
Our blocked experiment, as well as previous experiments (Anderson et al. 1994; Luna and Sweeney 1999; Sweeney et al. 1996b), found greater activation in MFG for memory-guided than for visually guided saccade blocks. The region identified as right MFG from the blocked experiment data exhibited activation patterns in the event-related experiment that were qualitatively similar to those from right pIFG with BOLD signal levels during the delay period being greater for memory-guided than for visually guided saccades, although this trend was not significant. One possibility is that the working memory load of a task determines the degree of activation in MFG, in which case our event-related task design might not have imposed sufficient working memory demand to elicit detectably greater BOLD activity in the memory-guided than in the visually guided saccade task. This interpretation is consistent with the finding by Leung and colleagues (2002) that middle frontal gyrus exhibited elevated, sustained BOLD activation during the delay period of a working memory task when subjects had to remember 5 target locations but not when they had to remember only 3 locations. It is also consistent with experiments that varied working memory demand using the n-back task with different n-values and demonstrated that BOLD activity in MFG depends on memory load (Braver et al. 1997; Carlson et al. 1998; Cohen et al. 1997).
An alternative explanation for differences in activation patterns in MFG between the blocked and the event-related experiments is that the type of working memory requirement as opposed to the overall working memory demand of a task might determine MFG involvement in the task. Petrides has proposed that dlPFC, which includes MFG, is specialized for monitoring and manipulating the contents of working memory as opposed to maintaining working memory traces and that maintenance functions are performed by ventrolateral prefrontal cortex, including pIFG (Petrides 1994, 1995; see also D'Esposito et al. 1998, 1999; Owen et al. 1996, 1999; Rowe and Passingham 2001; Rowe et al. 2000; Stern et al. 2000). This hypothesis predicts that MFG should not be recruited substantially by the memory-guided saccade task, which required subjects to maintain a memory of target location but did not impose significant monitoring or manipulative requirements. Petrides's hypothesis (Petrides 1994, 1995) could also explain our finding from the blocked experiment that MFG exhibited stronger activation for memory-guided than for visually guided saccade blocks. The blocked experiment involved a higher frequency of trial incidence than the event-related experiment, and performing trials in rapid succession could impose a heavy demand on processes that update working memory contents by discarding old memories and encoding new ones, resulting in greater recruitment of MFG in blocks of memory-guided saccades. It has also been shown that MFG is activated by tasks requiring response selection (Rowe and Passingham 2001; Rowe et al. 2000), and this could explain the presence of greater activation on memory-guided than on visually guided saccades in our blocked experiment and the lack of such a difference in our event-related experiment. In the blocked experiment, subjects had to choose to look at target stimuli for the visually guided saccade task or to remember target location without breaking central fixation on the memory-guided saccade task. The event-related experiment required subjects to maintain central fixation during stimulus presentation at the beginning of the delay period regardless of the condition. The response selection component of the blocked experiment could account for greater activation in MFG for memory-guided saccade blocks, whereas the lack of response selection in the event-related experiment could account for the absence of task-related activation differences in MFG for that experiment.
One other explanation for task-related differences in MFG activation between the blocked and event-related experiments involves MFG's putative role in providing inhibitory modulation of processing in other brain regions (for review see Miller and Cohen 2001). For example, MFG exhibits greater BOLD signal when a subject knows she/he will make an antisaccade, which involves inhibition of the reflexive saccade on stimulus presentation followed by execution of a voluntary saccade to the stimulus's mirror location, compared with knowing she/he will make a prosaccade (Desouza et al. 2003). Our blocked experiment required subjects to inhibit reflexive saccades to presented target stimuli in the memory-guided task but not in the visually guided task, whereas our event-related experiment required subjects to inhibit the reflex to look at the target appearing just before the delay period in both the memory-guided and the visually guided saccade tasks. Greater activation in MFG for memory-guided than for visually guided saccades in the blocked experiment and the lack of such a difference in the event-related experiment could be explained by the presence of different inhibitory components in the blocked memory- and visually guided saccade tasks combined with the equivalent inhibitory components in the event-related memory- and visually guided saccades tasks.
Transcranial magnetic stimulation (TMS) over MFG during the delay period of the memory-guided saccade task impairs saccadic accuracy, whereas TMS over MFG around the time of target presentation or saccade execution does not (Brandt et al. 1998; Muri et al. 1996b, 2000). These results were taken as evidence that dlPFC is involved in maintaining a working memory trace for target location in the memory-guided saccade task. There are at least 2 ways to reconcile these results with our finding that MFG exhibited only slightly and nonsignificantly greater activation on memory-guided than on visually guided saccades during the delay period. MFG might have a real but small role in maintaining a working memory trace for target location, in which case application of TMS over MFG during the delay period of the memory-guided saccade task would directly disrupt the internal memory trace of target location. The robust activation we found in pIFG for the delay comparison suggests that pIFG should play a much more important role in remembering target location and that applying TMS over pIFG should cause a greater impairment of memory-guided saccade accuracy than TMS over MFG. Of course, such a finding would not, on its own, establish maintenance of working memory traces as the exclusive purview of pIFG. Alternately, MFG might not be involved in maintaining a memory trace of target location; it could subserve other functions such as monitoring processes for working memory as proposed by Petrides (1994, 1995). In this case, TMS over MFG might have produced memory-guided saccade deficits through peripheral disruption of pIFG, which is adjacent to MFG, especially for the studies by Muri and colleagues (1996b, 2000), both of which used circular stimulation coils as opposed to the more focal figure 8-coil design. Even if direct disruption by TMS was limited to MFG, aberrant neuronal signals might have spread from MFG to pIFG by known connections between the two (Pandya and Yeterian 1996), causing a secondary perturbation of working memory processes in pIFG.
Delay period activity and maintenance processes in FEF, SEF, and IPS
The frontoparietal network composed of the frontal eye field, supplementary eye field, and intraparietal sulcus (IPS) is known to be involved in saccadic behavior and is thought to play a role in working memory based on comparisons of blocked memory- and visually guided saccades (Anderson et al. 1994; Luna and Sweeney 1999; Sweeney et al. 1996b). In our event-related experiment, we found greater delay-period BOLD activity on the memory-guided saccade task than on the visually guided saccade task in right rIPS, right ventral intraparietal sulcus (vIPS), right medial frontal eye field (mFEF), and SEF, suggesting that greater activation observed for blocks of memory-guided saccades than for visually guided saccade blocks is attributable at least in part to working memory processes during the delay period of the memory-guided saccade task.
The activation foci we identified in rIPS from the blocked experiment and the delay comparison of the event-related experiment are consistent with foci of activation previously identified as PEF (Kawashima et al. 1996; Muri et al. 1996a; Rushworth et al. 2001), which is the proposed homolog of monkey lateral intraparietal area (LIP) (Muri et al. 1996a). The rIPS and vIPS regions also coincide, respectively, with two regions that have been called anterior IPS and ventral IPS and that have been proposed to be part of a frontoparietal network involved both in generating saccadic eye movements and in mediating shifts of visuospatial attention (for review see Corbetta and Shulman 2002).
A role in working memory for mFEF, SEF, and rIPS is also supported by electrophysiological and inactivation studies. Neuronal recording studies in FEF, SEF, and LIP in monkeys identified neurons that maintain tonic, elevated discharge rates during the delay period of the memory-guided saccade task when the stimulus is flashed in their response fields (Barash et al. 1991; Bruce and Goldberg 1985; Russo and Bruce 1996). The functional importance of FEF and LIP in generating memory-guided saccades was demonstrated by studies that found that inactivation of either cortical area in monkeys causes targeting errors and increases latencies on the memory-guided saccade task (Dias and Segraves 1999; Li et al. 1999; Sommer and Tehovnik 1997). Inactivation of SEF does not affect single-step memory-guided saccades but it does cause increased latencies and targeting errors in the double-step memory-guided saccade task (Sommer and Tehovnik 1999). Lesion studies in humans found similar results (Gaymard et al. 1993; Pierrot-Deseilligny et al. 1991a,b; Ploner et al. 1999; Rivaud et al. 1994).
Application of TMS over right posterior parietal cortex (PPC) during stimulus presentation has been found to impair memory-guided saccade accuracy, and TMS over left or right PPC around the time of saccade execution increased memory-guided saccade latency (Brandt et al. 1998; Muri et al. 1996b, 2000). However, TMS over left or right PPC during the delay period did not affect memory-guided saccade performance. Pierrot-Deseilligny and colleagues (2002) interpreted these results as evidence that PPC has a role in target perception and saccade execution but not in remembering target location during the delay of the memory-guided saccade task. We propose that storage of target location in multiple brain areas, including right pIFG, right rIPS, right mFEF, and possibly SEF, during the delay period could explain the lack of targeting error with TMS over PPC during the delay. This could be tested by simultaneous application of TMS over multiple regions, FEF and PPC for example, during the delay period of the memory-guided saccade task.
Visuospatial sensory memory versus saccade motor plan
Two possible components of working memory function during the delay period are the storage of visuospatial sensory information related to simple stimulus location or possibly to visual attention and the maintenance of a motor plan to generate a saccade to the stimulus location. One fMRI study (D'Esposito et al. 2000) that attempted to distinguish between these two possibilities in prefrontal cortex did not find evidence for a prefrontal specialization toward either one or the other. The relative merits of the sensory memory and motor plan ideas have been debated extensively with regard to monkey LIP. Some reports (Goldberg et al. 2002) suggest that tonic firing activity in LIP neurons during the delay period codes for the locus of visual attention, citing evidence that this activity is modulated by the dynamics of salient visual objects and not by impending saccadic behavior. Others (Snyder et al. 2000) point out instances in which LIP neuronal activity does predict saccade performance, and they emphasize LIP's role in storing motor plans during the delay period. Many of the neurons in LIP, FEF, and SEF that maintain elevated discharge rates during the delay period of the memory-guided saccade task also exhibit either visual responsiveness, reacting to the presentation of visual stimuli in their receptive fields with transient elevations in discharge rates, or saccade responsiveness, discharging at elevated firing rates during the execution of saccades into their response fields (Barash et al. 1991; Bruce and Goldberg 1985; Russo and Bruce 1996). Based on the above discussion, it seems that working memory processes during the delay period in pIFG, FEF, SEF, rIPS, and vIPS cannot be differentiated on the basis of sensory memory as opposed to motor plan. Instead, it seems more plausible that these areas encode both processes.
Saccade execution and working memory retrieval
Comparison of the saccade-related activity from the memory- and visually guided saccade trials revealed greater BOLD activation for the memory-guided condition in right precentral gyrus and near the cortical surface in right rIPS, where it joins the postcentral sulcus. The precentral gyrus activation was located in that part of the motor strip representing the neck and head (Rotte et al. 2002) and might reflect suppression of head movements during saccade execution, given that eye and head movements often occur in concert, whereas our subjects were instructed to move their eyes and not their heads in the saccade tasks. We currently do not have a good explanation for why memory-guided saccades evoked more activation in precentral gyrus. The activation focus in right rIPS was located within the anterior end of the intraparietal sulcus where it joins the postcentral sulcus, which placed it anterior and superficial to PEF as identified by Muri and colleagues (1996a). It is possible that this region is part of Brodmann's Area 5 and that it serves a somatosensory function, possibly related to the requirement to keep the head still during saccade execution.
In our event-related visually guided saccade task, subjects generated saccades to visible targets, which appeared simultaneously with fixation offset whereas, in our memory-guided saccade task, subjects made saccades to remembered target location in the absence of a visible target. Some brain regions might have expressed greater BOLD signal related to retrieval and motor processes for memory-guided saccades than for visually guided saccades, but such a difference might have been masked by the additional sensory-
