van Mier, H., L. W. Tempel, J. S. Perlmutter, M. E. Raichle, and S. E. Petersen. Changes in brain activity during motor learning measured with PET: effects of hand of performance and practice. J. Neurophysiol. 80: 2177–2199, 1998. The aim of this study is to assess brain activity measured during continuous performance of design tracing tasks. Three issues were addressed: identification of brain areas involved in performing maze and square tracing tasks, investigation of differences and similarities in these areas related to dominant and nondominant hand performance, and most importantly, examination of the effects of practice in these areas. A total of 32 normal, right-handed subjects were instructed to move a pen with the dominant right hand (16 subjects) or nondominant left hand (16 subjects) continuously through cut-out maze and square patterns with their eyes closed during a 40-s positron emission tomography (PET) scan to measure regional blood flow. There were six conditions: 1) holding the pen on a writing tablet without moving it (rest condition); 2) tracing a maze without practice; 3) tracing the same maze after 10 min of practice; 4) tracing a novel maze; and tracing an easily learned square design at 5) high or 6) low speed. To identify brain areas generally related to continuous tracing, data analyses were performed on the combined data acquired during the five tracing scans minus rest conditions. Areas activated included: primary and secondary motor areas, somatosensory, parietal, and inferior frontal cortex, thalamus, and several cerebellar regions. Then comparisons were made between right- and left-hand performance. There were no significant differences in performance. As for brain activations, only primary motor cortex and anterior cerebellum showed activations that switched with hand of performance. All other areas, with the exception of the midbrain, showed activations that were common for both right- and left-hand performance. These areas were further analyzed for significant conditional effects. We found patterns of activation related to velocity in the contralateral primary motor cortex, related to unskilled performance in right premotor and parietal areas and left cerebellum, related to skilled performance in supplementary motor area (SMA), and related to the level of capacity at which subjects were performing in left premotor cortex, ipsilateral anterior cerebellum, right posterior cerebellum and right dentate nucleus. These findings demonstrate two important principles: 1) practice produces a shift in activity from one set of areas to a different area and 2) practice-related activations appeared in the same hemisphere regardless of the hand used, suggesting that some of the areas related to maze learning must code information at an abstract level that is distinct from the motor performance of the task itself.
Motor performance changes with practice from initially slow, inaccurate, and uncoordinated movements to fast, accurate, and coordinated actions (Schmidt 1988). In studies of motor performance and cognitive psychological studies of skill acquisition, these changes often have been attributed to shifts in underlying processes, such as motor planning and programming (e.g., Fitts 1964; Posner and Keele 1968; van Mier and Hulstijn 1993; van Mier et al. 1993). Neural mechanisms related to these processing shifts might be identified with functional neuroimaging, by measuring modifications in brain-activation patterns.
For example, Raichle and colleagues (1994) demonstrated that practiced performance of a verb generation task activated areas that were different from the areas activated during naive performance. Areas activated after practice were the same as areas activated during more or less overlearned noun reading. This suggested that two distinct pathways can be used for verbal-response selection, depending on the level of skill associated with the response. Similarly, Sakai et al. (1998) using functional magnetic resonance imaging (fMRI), showed a shift in brain activations in a visuomotor learning task. Whereas frontal areas were activated mostly during early learning, involvement of the parietal cortex increased in the more advanced learning stage.
Many other neuroimaging studies have shown changes in activation related to the level of motor skill as well. Several studies showed activation of prefrontal areas, supplementary motor area (SMA), premotor cortex, and cerebellum when novel motor tasks had to be performed (Decety et al. 1990; Friston et al. 1992; Jueptner et al. 1997b; Roland and Seitz 1989) or when a movement had to be selected based on internal or external cues (Deiber et al. 1991). Activity decreased in these areas after practice (Grafton et al. 1992a; Haier et al. 1992; Jenkins et al. 1994; Jueptner et al. 1997b; Mazziotta et al. 1991). Cerebellar activity was found during the learning of a motor task (Grafton et al. 1992a; Roland et al. 1989), whereas activity in the basal ganglia was found only when the movement was highly overlearned and automatic (Mazziotta et al. 1985, 1991). When subjects had to select a motor response randomly, i.e., when they had to choose freely which movement to make, the anterior part of the cingulate was activated (Deiber et al. 1991; Frith et al. 1991; Jueptner et al. 1997a). Aizawa et al. (1991) and Roland et al. (1980a) found little or no activity in SMA when an extensively overlearned motor task was performed.
Our goal in the present study was to use a set of tasks that allowed some control for the gross changes in movement parameters that accompany learning and that allowed for a full cycle of learning to be observed in the scanning session (from novel to practiced to novel). This task also emphasized more strongly the use of somatosensory and proprioceptive feedback during learning and as such could extend the domain of learning effects described in the above mentioned studies and literature. This was accomplished by using continuous maze and square tracing tasks (van Mier et al. 1993) without visual feedback.
A total of 32 normal adult volunteers, divided in two groups of 16 subjects each, participated in the study. Each group consisted of nine females and seven males. For both males and females, age varied from 19 to 35 yr with an average age of 24 yr for both. All subjects were strongly right handed based on the Edinburgh handedness inventory (Raczkowski et al. 1974). None of the subjects had a history of neurological disorders. The Human Studies Committee as well as the Radioactive Drug Research Committee of Washington University approved the study. Subjects gave informed consent conforming to the guidelines and procedures set forth by these committees. Subjects were paid for participation.
The two independent groups differed only with respect to which hand was used to perform the tasks. In the first group, which will be referred to as right-hand (RH) performance group, subjects performed the tasks with their dominant right hand, whereas performance in the second group, referred to as left-hand (LH) performance group, was done with the nondominant left hand.
Positron emission tomography (PET) imaging
Subjects were positioned in a supine position on an adjustable table. A venous catheter was inserted in the left arm of those subjects who performed the tasks with their right hand and in the right arm of the subjects performing the tasks with the left hand. To restrict head movement during the experiment, an individually molded plastic mask was securely and closely fit over the subject's face. A lateral skull radiograph permitted assessment of head alignment in the scanner.
PET scans were performed using a Siemens-CTI 953B PET scanner (Spinks et al. 1992), that collects data for 31 axial planes with center-to-center slice separation of 3.38 mm. Data were collected in the three-dimensional mode (Kinahan and Rogers 1989), which provided an axial field of view of 10.5 cm. Because of this somewhat limited axial field of view, we scanned one group of 16 subjects (8 performing the tasks with the right hand and 8 with the left hand) while they were positioned low in the scanner, thereby ensuring coverage of the upper two-thirds of the brain. In the same way, a second group of 16 subjects was positioned high in the scanner, thereby ensuring coverage of the lower two-thirds of the brain. Combining the data acquired from these two groups optimized data collection for the supplementary and lateral dorsal premotor areas as well as most of the cerebellum (see Fig. 1).
A 10-min transmission scan, performed on each subject after adequate stabilization and alignment of the head, was used to correct for radiation attenuation by the tissues of the head. This scan was collected during exposure of three rotating rod sources containing Germanium-68. Water labeled with 15O acting as a blood-flow tracer was administered intravenously as a bolus of 8–10 ml of saline containing 15 mCi (Herscovitch et al. 1983; Raichle et al. 1983). As soon as the bolus entered the brain, as noted by a sudden rise in the system count rate, a 40-s scan was initiated. The short half-life of H2 15O (∼2 min), the short scanning time of 40 s, and the relatively low dose of radiation made it possible to perform 10 scans within a subject in a single session.
One square design and three maze designs were used in the experiment (see Fig. 2). The square and mazes were cut-out designs, creating paths for the pen to traverse. The square consisted of 4 segments of 6 cm each, the maze designs consisted of 8 or 10 line segments. The complete path length of each maze, if traced correctly, was 24 cm. The width of the paths was 0.5 cm, the depth was 0.15 cm. To simplify decision making, the maze designs consisted of straight lines with 90° angles. Only two opposite-direction choices could be made at each intersection, one of which came to a dead end. The length of each dead end path was 0.5 cm. All designs formed closed loops, i.e., the endpoint of one loop was the starting-point of the next loop and the designs could be traced continuously. The mazes that were used in the LH performance group were mirror images of the mazes used in the RH performance group. For each subject, two versions were used for the eight-segment maze designs, versions A and B. The B versions were 90° rotations of the A versions. One version of the eight-segment maze was traced during naive and practiced performance, the other version during novel performance, with the version of maze presentation counterbalanced across subjects.
A Calcomp 2500 digitizer and a specially designed pen (Maarse et al. 1988), both connected to an IBM PS2/30 microcomputer, were used to record the x and y coordinates of the pen movements. During each experimental scan, movements were registered for a period of 1 min, starting 10 s before and finishing 10 s after each scan. Recording was done at a precision of 1 mm and a frequency of 100 Hz. In other words, the position of the pen was recorded every 10 ms.
Task description and procedure
Subjects performed the tasks with eyes closed while lying in the PET-scanner with their arms lying in arm rests. The arm that was used to perform the tasks was fixed with straps at the upper arm to minimize movement of the upper arm, shoulder, and head. The cut-out designs were taped on a writing tablet positioned on a table to the right or left of the subject (see Fig. 3). The height of the table and the position of the writing tablet on the table were adjusted for each subject to achieve the most comfortable drawing position. Hand, wrist, and forearm were positioned freely above the writing tablet to minimize tactile stimulation in these areas. The pen was kept on the writing tablet during all of the scans.
Deltoid muscle activity from the performing arm was recorded by electromyographs (EMGs) to detect possible changes in proximal or distal movements as an effect of practice. To minimize possible confounding of neuronal activity related to eye movements, subjects were asked to keep their eyes still for the duration of the scan and compliance was verified with electrooculograms (EOG) recorded from electrodes placed on the left and right temples close to the outer canthus of the eyes. EMG and EOGs were tested before the experiment started, and subjects were asked to lift their upper right or left arm and to move their eyes to the left, right, up, and down. During each scan, activity from the eye and deltoid muscles was recorded. Deltoid activity was recorded from the arm that was used during the performance, except during the first scan in the LH performance group. During that scan, deltoid activity from the left arm was recorded although maze tracing was performed with the right hand. EMG and EOG data were analyzed by visual inspection of the tracings.
The position of the subjects in the scanner prevented them from seeing the maze designs, but to ensure subjects had no visual feedback during the task, tracing of the designs was done with eyes closed. Subjects were also not allowed to inspect the maze designs by touch. The experimenter guided the subjects' hand, holding the pen, into the proper starting position of the maze. The instruction to the subjects was to move continuously through the square and maze design. Furthermore, for most conditions, subjects were instructed to trace the designs as quickly as possible. They were told that the first movement was always upward, that each maze was a loop, and that they should try to avoid retracing a path once they experienced a forced stop. Before subjects were put in the scanner, a maze, different from the ones used in the experiment, was shown as an example and the two-choice principle at intersections was explained.
Ten PET scans were performed in each subject: seven experimental scans and three rest scans (see Table 1), each separated by a 10-min interscan interval. The scans were presented in the same fixed order in all subjects. Duration of the performance was 1 min. Subjects received the instruction to start tracing at the time of the injection, which was ∼10 s before the onset of the scan. The instruction to stop was given 10 s after the scan was finished. In the RH performance group, performance during all scans was done with the right hand and in a generally clockwise direction. In the LH performance group, subjects performed with the left hand and in a counterclockwise direction (except during the 1st scan, which was performed with the right hand and in a clockwise direction). During all the scans, subjects were instructed to keep their eyes closed and still for the duration of the scan.
During the first scan (maze 10), all subjects (both RH and LH performance groups) were instructed to trace a 10 segment maze as quickly and accurately as possible with the right hand in a clockwise direction. Although subjects were instructed clearly how to perform the task by showing them a completely different maze than the ones used in the experiment, by including this maze, subjects could get used to the tracing and the two-choice principle and get a feeling for the overall size of the maze. In this way, general task-related aspects, which might have affected performance and brain activity, would be reduced during the performance of the naive eight-segment experimental maze. During the interscan interval between scans 1 and 2, subjects in the LH performance group traced the mirror image of the 10-segment maze with their left hand in counterclockwise direction for a period of 1 min. The latter was included so that subjects in the LH group would be exposed to the same amount of left hand and counterclockwise maze tracing as clockwise right hand tracing in the RH performance group.
During the second scan (rest), subjects were instructed to keep the pen stationary on the writing tablet without moving it. Subjects in the RH performance group held the pen in the right hand, subjects in the LH performance group in the left hand.
During the third scan (sq. fast), subjects were instructed to trace a square design as quickly as possible. The square was included because it is familiar and quickly “learned” (van Mier et al. 1993). The square design was taped on the writing tablet and was shown during the instruction to the subject before the latter was put in the scanner. Subjects also traced the square during adjustment of the height of the table and the positioning of the writing tablet on the table to find the most comfortable drawing position for each subject. Once this position was established, subjects were asked to trace the square as fast as possible for a short period. So subjects had seen and traced the square before the experiment started and were therefore very familiar with it. All subjects were instructed to trace the square in a clockwise direction, subjects in the RH group with their right hand, subjects in the LH group with their left hand. Between scans 3 and 4 subjects in the LH group traced the square counterclockwise for 1 min. As stated before, this was included so that subjects in the LH group would be exposed to the same amount of counterclockwise left hand tracing as clockwise right hand tracing in the RH group.
During the fourth scan (naive) subjects had to trace an eight-segment maze as quickly and accurately as possible. After this scan, subjects practiced the same maze for 10 min during the interscan interval. To prevent subjects from slowing down as an effect of fatigue a resting period of 30 s was included after every 2 min of practice.
During the fifth scan (prac), subjects were once more instructed to trace the same maze as quickly and accurately as possible.
The sixth scan (rest) was again a rest scan, during which subjects held the pen on the writing tablet without moving it.
The seventh scan (sq. fast) was a repetition of the third scan, and subjects were again instructed to trace the square as quickly as possible. Right-hand square tracing had to be done in a clockwise direction, left-hand tracing in a counterclockwise direction.
The instructions given during the fourth scan were repeated during the eighth scan (novel) when subjects had to trace a novel maze design, that is, the other version of the eight-segment maze.
A third rest scan was included during the ninth scan (rest) during which subjects held the pen on the writing tablet without moving it.
Finally, during the 10th scan (sq. slow) subjects were instructed to trace the square slowly and only change direction after bumping into the side. Large differences were expected between the naive and prac conditions both in the speed at which the tasks were performed and the number of stops observed. The slow square tracing was included as a control condition to account for velocity and somatosensory effects of bumping into dead ends and stopping. The effect of speed was controlled by asking subjects to perform the square tracing slowly. By instructing subjects to change direction only after bumping into the sides at the end of each segment, they were forced to stop and use somatosensory feedback before changing direction. So the major differences between the fast and slow condition were the velocity and somatosensory feedback. These effects on brain activity could be studied directly by comparing cerebral blood flow during sq. fast and sq. slow.
As mentioned above, the LH performance group was instructed to trace maze 10 in the first scan with the right hand and the square in scan 3 in clockwise direction. This would make it possible to check for effects of subject group and/or drawing direction if differences between left- and right-hand performance were found.
Data records for all conditions were low-pass filtered at 10 Hz, and the drawing trajectories were displayed on the computer screen. The beginning and end of the 1-min tracing movement was determined by means of an interactive computer program. The mean tracing velocity was calculated for the total period of 1 min. For each condition, one maze or square loop, traced at the middle of the 1-min period, was segmented in line and angle segments based on velocity minima and x and y displacements. Duration, mean, and minimum velocity were determined for each angle segment. The tracing also was checked for stops and incorrect turns. A stop was assigned when the velocity at an angle was lower than 1 cm/s for a period >100 ms. When a dead-end path was traced, indicating that an incorrect turn was taken or when a segment was retraced, an error was assigned (see also van Mier et al. 1993). For all angle segments, the number of errors, the percentage of stops and the mean stop duration was calculated. For each of the four dependent variables, analyses of variance (ANOVAs) were performed to study effects of hand of performance, scan position, and condition (superANOVA, version 1.1, Abacus Concepts). When a significant effect of condition was observed, specific comparisons were examined post hoc to assess the pattern of significant differences between particular conditions. Probabilities were adjusted using the Greenhouse-Geisser epsilon.
General. The PET images were reconstructed with filtered backprojection using a Butterworth 0.5 filter with an order of 5, producing a transaxial resolution of 14-mm full width at half-maximum (FWHM). A linear normalization was applied to the images to negate the effects of global fluctuations in blood flow due to changes in arterial pCO2 and variations in the amount of tracer administered. Because we did not measure arterial blood radioactivity, our data represent changes in the distribution of radioactivity. In the text we refer to these changes as changes in blood flow because there is a tight linear relationship between blood flow and radioactivity in brain tissue under normal physiological conditions (Fox and Mintun 1989; Herscovitch et al. 1983).
The intercommissural AC-PC line was identified in each subject's skull radiograph and was used for anatomic standardization of the images (Fox et al. 1985) into stereotaxic space defined in Talairach and Tournoux's atlas (Talairach and Tournoux 1988). At the time of testing, no MRI data were collected; however, we could collect high-resolution anatomic MRI data for six subjects at a later time. Comparison of Talairach coordinates for peak activations in these six subjects based on the skull X-ray method were within 5 mm distance from coordinates obtained by the MRI method and are most likely due to methodological variability (Hunton et al. 1996), indicating that the use of a skull X-ray for anatomic standardization was also valid for the high-resolution scanner that was used for the experiment. For each subject, images were grouped in experimental-control pairs and subtracted from each other. All subtracted image pairs were checked for movement artifacts and eliminated if excessive movement was found. To improve signal to noise, single subtraction images of identical conditions were averaged across subjects (Fox et al. 1988; Mintun et al. 1989). Averaged subtraction images then were searched by an automatic maximum-detection algorithm to identify and record all positive and negative maxima by location in stereotaxic coordinates and by magnitude in PET counts (Mintun et al. 1989).
In both groups, specific subtractions were averaged over all 16 subjects in such a way that the top and the bottom of the brain included data from 8 subjects, whereas a small area in the middle of the brain included data from all subjects. In the RH performance group, two subjects could not be positioned far enough in the scanner to include the whole cerebellum and yet maintain a comfortable drawing position, so the lowest parts of the brain contain data from only six subjects. Of the 31 PET slices that were obtained, we excluded the very noisy outer slices so that only data from slices 6–26 were included in the analysis (see also Fig. 1).
An objective way to identify activations and assess their reliability and reproducibility is to determine if response locations defined in one group of subjects replicate in a second independent group performing under identical conditions (Buckner et al. 1995; Hunton et al. 1996) or replicate in the same subject group but performing under identical or comparable conditions (Corbetta et al. 1993). Activations that reproduce across groups are the best indication that the activations are reliable and can be attributed to the particular task demands and not to noise or other irrelevant factors. In this study, a combination of above-mentioned replication methods was used.
The first step was to determine the magnitude cutoff that would most reliably distinguish between signal and noise peaks. We used the method described by Hunton et al. (1996) for PETT VI data, to establish the magnitude cutoff for our 953B data. Based on a small sample set (6 < n < 12, depending on the peak location) of our data acquired in the RH performance group, noise peaks were acquired in a control data set in which the same tasks were performed during the active and reference scan. Magnitudes of noise peaks were all under 80 PET counts, and “noise” peaks did not significantly replicate at a magnitude of ≥50 (P > 0.05). Signal peaks also were acquired in a small experimental data set (6 < n < 10) acquired in the RH performance group, in which different tasks were performed during the active and reference scan. Of the peaks with a PET count of ≥50, 61% replicated, areas with peaks of ≥100 PET counts replicated 100%. On the basis of these data and on the visual inspection of several image sets, a magnitude cutoff of 50 PET counts was used in the replication analysis.
Three different data analyses were performed on the activation data: 1) to identify brain activations generally related to the tracing tasks, data acquired during the five tracing scans minus rest were combined in each group and were checked for brain areas showing peak activations of ≥50 PET counts; 2) to assess reliability of the activations and examine the effect of hand of performance, in each group brain areas showing magnitudes of ≥50 PET counts were tested for replication in the other group; and 3) to examine the effect of practice, differences in magnitudes between conditions were examined in areas showing reliability and/or replication.
Areas commonly activated across hand of performance. To identify areas that generally were activated during maze and square tracing, in each group activations acquired during rest conditions were subtracted from activations measured during the five tracing scans, which then were combined into a composite summed difference image, and areas with activations of ≥50 PET counts were specified. To assess within-data set reliability of these areas, in each group t values were calculated. Areas then were tested for replication.
The reproducibility of regions identified during tracing performance in one group (dominant-hand tracing; RH performance group) was tested during comparable performance in the other group (nondominant-hand tracing; LH performance group) and vice versa. In each group (RH or LH performance) brain regions active during rest conditions were subtracted from activations acquired during the five tracing scans and were identified and tested for replication in the other group. When activations replicate in two completely different subject groups performing the same tasks but using different hands, we observe not only replication but also generalization.
A multiple-step analysis was performed consisting of two phases. During the first phase, the hypothesis-generating phase, the first step was to create two composite summed difference images, one for the RH performance group and one for the LH performance group, including images from the five experimental scans minus their selected rest conditions. Scans obtained during rest were subtracted from scans obtained during naive, prac, and novel as well as during sq. fast and sq. slow, being, respectively, scans 4, 5, 7, 8, and 10. We decided to use the data from the second sq. fast tracing (scan 7) in the analysis because tracing was done in clockwise direction in the RH performance group while in counterclockwise direction in the LH performance group. This way tracing during each of the five scans used for analysis was performed with the right hand and in clockwise direction in the RH performance group, with the left hand and in counterclockwise direction in the LH performance group. In the second step, all regions with a blood-flow change of ≥50 PET counts (increases and decreases) were identified in the summed image of each group by using an automatic maximum-detection algorithm which averaged magnitudes over a 7-voxel (14 mm) diam (Mintun et al. 1989).
During the second phase, the hypothesis-testing phase, we determined which of the identified regions showed significant blood-flow changes in the other group. This was done by calculating values of blood-flow change in the difference image for each of the 16 subjects in the hypothesis-testing groups (RH and LH performance, respectively) around the peak locations defined in the summed difference images of the hypothesis-generating groups (LH and RH performance, respectively). The magnitudes for the subjects in the hypothesis-testing groups were computed by taking all voxel values within a 7-voxel-diam sphere centered at each peak location defined in the hypothesis-generating groups and calculating the average PET counts within each sphere. Using a one-tailed, one-sample t-test with a hypothesized mean of 0, mean changes in blood flow were tested in locations corresponding to the peaks in the other group. A one-tailed test was appropriate because the generate group data specified the expected sign of blood-flow change in the test group. When the location of a peak identified in the hypothesis generating group was not adequately sampled (<80% of voxels) in a subject in the hypothesis-testing group, the data from that particular subject were excluded. Areas observed in both groups are considered to be related to tracing performance, regardless of hand of performance. Areas with high magnitudes and reliability within a single group will be considered as candidate areas related to hand of performance.
Identification of areas affected by condition. To investigate differences in magnitude between conditions in each group, the same composite summed difference images as described above (including the five conditions of interest, naive, prac, novel, sq. fast, and sq. slow, minus their rest condition) were used. The composite image was used to exclude any bias of results toward differences between conditions by defining the regions of interest on one condition. In each group, the coordinates of the regions that showed magnitudes of ≥50 PET counts were used to find the average regional value in each subject, creating per region and per subject a magnitude for each subtracted condition. For each region, a repeated-measures ANOVA was performed on the magnitudes in the five conditions for those subjects in which ≥80% of the voxels in the region was sampled. When a significant effect of condition was observed in the overall ANOVA, specific comparisons were examined post hoc to assess the pattern of significant differences between particular conditions. Probabilities were adjusted using the Greenhouse-Geisser epsilon.
Using a composite image to identify significant brain activations has the potential to miss activations in areas that are mainly activated during only one or two conditions. To make sure that such task-specific activations were not missed, in each group separate analyses were done for each of the five conditions. Each of these images then was checked to see if all areas showing activations of ≥50 PET counts and a t value of ≥3.5 were also identified in the composite images. Another source of variance could be introduced by comparing the experimental scans to a rest condition. We tried to make the rest condition as comparable with the tracing conditions as possible except for the movement. Subjects never were told what they were expected to do during the next scan to prevent preprocessing of the next task. To make sure any additional processing during the rest scan would not influence our data, each experimental condition also was compared directly with each of the other experimental conditions and again areas with ≥50 PET counts and a t value of ≥3.5 were checked for occurrence in the combined image.
EMG and EOG recordings
None of the subjects showed gross deltoid muscle EMG activity during the rest scans. If muscle activity was apparent during the other scans, it was in all cases much lower than the activity produced by the subjects during EMG testing. Furthermore, no differences in deltoid activity were noted among the five conditions mentioned above. The EOG recordings revealed that eye movements were minimal. No differences in EOG activity were noted among the five conditions.
To be able to combine the PET data from the “top” and “bottom” subjects (subjects from which either the top 2/3 of the brain was imaged or the bottom 2/3), performance for both had to be similar. The means and standard deviations for each scan position and group are given in Table 2. For each group, an ANOVA was performed, with scan position (top vs. bottom) as the between factor, which showed that there were no significant differences between top and bottom positions with respect to velocity and duration of stops (P > 0.37 for the RH and P > 0.72 for the LH performance groups, respectively). Percentage of stops and number of errors were nearly identical in both scan positions (P > 0.90 for both variables in each group).
EFFECT OF HAND OF PERFORMANCE.
Next any differences between right and left hand performance were assessed. For each dependent variable, an ANOVA was performed with group (right vs. left hand) and scan position (top vs. bottom) as between factors. As can be seen in Table 3, the main effect of group was not significant for any of the variables. Although subjects were a bit faster and stopped less when the tasks were performed with the right hand compared with the left hand, these differences did not approach significance. Number of errors and percentage of stops were nearly the same in both groups. As expected, the main effect of scan position on the above mentioned variables was not significant nor was the interaction of group and scan position.
EFFECT OF CONDITION.
Data concerning velocity, errors, and stops for the five conditions (3 conditions for the errors because “dead-end” errors were not relevant in both sq conditions) are presented in Fig. 4. Only when an increase in velocity is accompanied by a decrease in errors can an improvement in performance be surmised (Fitts 1964). Furthermore, a decrease in the number and duration of stops at corners is an indication that the movements are performed smoothly (van Mier et al. 1993). As can be seen in Fig. 4, performance after maze practice (prac) and during sq. fast clearly shows the characteristics of more skilled behavior; the movements are performed quickly with few stops or errors. The performance study by van Mier et al. (1993) has shown that the square is learned easily and needs hardly any practice to become skilled. Because we used the data from the second sq. fast condition, it seems justified to describe square performance as skilled. On the other hand, a skilled task also can be performed differently as is shown in Fig. 4 for the sq. slow condition. Subjects indeed traced the square at a low speed and stopped at each corner. Repeated measurement ANOVAs, using the Greenhouse-Geisser epsilon adjustment, showed that the mean effect of condition was highly significant (P < 0.001) for each dependent variable (see Table 4).
The most interesting post hoc comparisons between specific conditions also are shown in Table 4. It was observed that in both groups, subjects were significantly faster and stopped significantly fewer times and for shorter durations in the prac and sq. fast conditions than in the naive, novel, and sq. slow conditions (P < 0.001). A significant decrease in the number of errors was found during prac compared with naive and novel tracing (P < 0.001). As can be seen in Table 4, performance during naive was not significantly different from performance during novel and sq. slow. The only difference between prac and sq. fast was found for velocity, a difference that may be partially accounted for by mechanical aspects of performance (changing direction 8 times and turning right and left in the maze compared with changing direction only 4 times and making either right or left turns in the square).
None of the subtractions used to generate the difference images had to be excluded because of movement artifacts. As was stated before, for the composite difference images data from the following five conditions were combined in each group; naive, prac, novel, sq. fast, and sq. slow. Data from the same five conditions also were used to study conditional effects. For the sq. fast condition, the data from the second square tracing (scan 7) were used in the analysis, representing clockwise tracing in the RH performance group and counterclockwise tracing in the LH performance group.
The separate images that were generated in each group for each of the five conditions minus rest showed that significant brain activations observed in these images also were identified in the composite images. Furthermore direct comparisons between all experimental scans did not identify any additional brain activations not observed in the composite images.
First brain activations related to tracing performance in general will be presented and secondly areas that show differentiated activations related to condition.
IDENTIFICATION OF BRAIN ACTIVATIONS AND EFFECT OF HAND OF PERFORMANCE.
Blood-flow increases. As can be seen in Fig. 5, in each group the combined image of the five maze and square tracing conditions shows activations in primary motor cortex (area A), secondary motor areas [dorsal premotor cortex (area B) and SMA (area C)], inferior and superior parietal (areas D and E), somatosensory (area F), and ventral premotor (area G, BA 44/6) areas, insular cortex, thalamus, as well as several cerebellar regions [anterior (area H), right (area I), left (area J) and right posterior (area K) cerebellum]. Figure 5 shows that most of these areas were activated similarly for both right- and left-hand performance. Areas that shifted lateralization of responses between right- and left-hand performance were primary motor cortex (area A) and anterior cerebellum (area H), with right-hand performance mainly activating left primary motor cortex and right anterior cerebellum, left-hand performance activating the same areas in the opposite hemispheres. Not surprisingly, differences in magnitudes between right- and left-hand performance in right and left primary motor cortex as well as in right and left anterior cerebellum were highly significant (P < 0.001 for all 4 comparisons).
On the other hand, the other areas mentioned above were activated in the same hemispheres in both groups, regardless of which hand was used to perform the tasks. In each group, regions that corresponded to peaks of blood-flow increases with a magnitude of ≥50 PET counts were assessed in the composite image. Within-data set reliability was tested in these activated regions in each group by calculating t values for each region and a threshold of t of ≥3.5 was set. Activations in regions that significantly replicated, showing across-data set reliability and had a t value of ≥3.5, showing high within-data set reliability, can be considered dependable. On the other hand, in this study areas that didn't replicate but had a high t value might be related to hand of performance. Areas that met these criteria are given in Table 5.
For most regions, differences in magnitude measured in the RH and LH performance group were not significantly different, with the exception of the activation in left dorsal premotor area, which was significantly more activated during right-hand performance than left-hand performance (P < 0.05).
Blood-flow decreases. Figure 6 shows the areas in which reduced activity was observed during tracing performance. As can be seen, for the decreases the images of RH and LH performance also appear very similar. In both groups, the combined image of the five maze and square tracing conditions shows decreases in temporal (BA 38, 39, 21, and 22; areas W, O, U, and S), occipital (BA 19; area Y), precuneus (BA 7; area N), and frontal areas (BA 8, 9, 10, and 45/46; areas L, P, Q, and R). A decrease in activation also was found in anterior (BA 24 and 32; areas V and T) and posterior cingulate (BA 31; area M), and caudate (area X).
Areas in which blood-flow decreases were found of ≥50 PET counts and had t values of ≥3.5 are listed in Table 6. As can be seen, all of these areas significantly replicated in the other group. Differences in magnitude between right- and left-hand performance in these areas were not significant.
IDENTIFICATION OF AREAS AFFECTED BY CONDITION.
Blood-flow increases. All regions with ≥50 PET counts in the composite image (all 5 conditions-rest) of the RH and LH performance group were tested for conditional effects. As described in methods, for each region and condition, the average regional value in each subject was calculated. Activated areas located adjacent to the primary motor cortex in the hemisphere contralateral to the hand used could not be identified as significantly distinct from the nearby large primary motor response. Distinction of two separate foci required separation by ≥14 mm (FWHM). Filtering at a higher resolution (10 mm) did not separate the dorsal premotor, somatosensory, and parietal responses from the response in the primary motor area. However, the replication data showed that dorsal premotor, parietal, and somatosensory cortex were highly activated in the contralateral hemispheres. To calculate the average regional value in these areas contralateral to the hand used, we applied the coordinates of the foci that were identified in the same areas and same hemisphere in the other group (ipsilateral to the hand used). These coordinates are italicized in Tables 7 and 8. Each region included all voxels within a 7-voxel-diam (14 mm) sphere centered at each location.
For each region, a repeated-measures ANOVA was performed on the magnitudes in the five conditions for those subjects in which ≥80% of the voxels in the region was sampled. Regions that showed a significant effect of condition (P < 0.05, Greenhouse-Geisser adjusted; except for right inferior parietal cortex in the LH performance group for which a P < 0.07 was found) are presented in Table 7, showing for each group the P values for the main effect of condition as well as the magnitudes per condition. The only area in which a significant effect of group was found was the left dorsal premotor cortex (P < 0.05), which was more activated in the RH group. The interaction of group and condition was not significant in any of the other regions, suggesting that differences in magnitude between conditions showed the same pattern in each group. This also was confirmed by the similarity of the specific post hoc tests.
Post hoc comparisons between conditions showed that magnitudes between naive and novel and between prac and sq. fast conditions were not significantly different in the areas presented in Table 7 and that changes in these areas could be categorized as either velocity-, practice-, or “capacity”-related (see Table 8) with the exception of left superior parietal cortex. Figure 7 shows the cortical and cerebellar activations for each condition in each group.
1)Velocity-related changes. Activation in contralateral primary motor cortex showed the same pattern as the speed at which subjects performed the tracing tasks: low during naive and novel as well as during sq. slow, high during prac and sq. fast. During right-hand performance, the left primary motor cortex showed this pattern, whereas during left-hand performance, this activation pattern was observed in the right primary motor cortex (see Fig. 7 A; area A).
Post hoc comparisons between conditions with respect to the magnitude showed the same significance pattern as the comparisons with respect to velocity, another indication that changes in activations in the contralateral primary motor area are most likely related to changes in the velocity. Figure 8 presents the magnitudes per condition for each group in the contralateral primary motor cortex.
We do not want to exclude that areas that show changes related to velocity also could contain learning effects. However, in this study, activations in primary motor cortex track velocity irrespective of level of “skill” (e.g., activation during square slow performance is similar to activation during naive maze performance). Although it does not mean that there are no skill-related effects in primary motor cortex, in this study, the velocity-related effects dominate the activation.
2)Practice-related changes. Conditions of less skilled or unskilled performance include the naive and novel conditions, in which subjects are presented with a particular complex maze design for the first time. Those conditions that are performed at higher levels of skill include the prac, sq. fast, and sq. slow conditions. Thus areas related to more effortful unskilled performance should show greater levels of activity in the naive and novel conditions than in the other three conditions. Areas related to more skilled performance should show the opposite pattern (low activity in naive and novel and greater activity in prac and both sq. conditions). The crucial difference between these patterns and velocity-related changes is the behavior of the sq. slow condition. If activity is tracking velocity, the activity of sq. slow and sq. fast should be very different. If activity is tracking level of skill, then the activity for these conditions should be very similar.
A)Areas related to unskilled performance. Four areas showed blood-flow activation across the conditions that was consistent with a relationship to unskilled performance, i.e., that were significantly lower in the prac and sq. conditions compared with novel and naive (see Tables 7 and 8 and Fig. 9). These were right dorsal premotor, right inferior and superior parietal cortex, and an area in left lateral cerebellum (Fig. 7 A; areas B, D, E, and B; area J, respectively). For the cerebellar area, naive and novel differed significantly from the sq. conditions, whereas a trend was found for the comparison with the prac condition, as can be seen in Table 8.
B)Areas related to skilled performance. Blood-flow activity in SMA across the conditions was consistent with a relationship to more skilled performance, i.e., significantly higher during prac and both sq. conditions than during naive and novel (see Table 8 and Fig. 7 A, area C). As can be seen in Table 7 and Fig. 10, activations in the SMA in prac and sq. conditions are very similar, a finding that was observed in both the RH and LH performance group.
3)Other condition-related changes. A number of areas showed a significant effect of condition, but the changes were neither related to velocity nor practice. As can be seen in Table 7 and Fig. 11, several of these areas were significantly more active when the task was performed for the first time and at a relatively low speed (naive and novel conditions) or when the task was practiced or more skilled and performed at a high speed (prac and sq. fast conditions) than when the task was more skilled but performed at a low speed (sq. slow condition). The first four conditions shared at least one aspect at which they differed from sq. slow: subjects were instructed to perform the tasks as quickly and accurately as possible. So, it is possible that the instruction, in which speed and accuracy were highly stressed, might have forced subjects to perform at or near “capacity” level during those conditions. These changes will be referred to as capacity-related, although we stress that these changes might have been caused by other, at this point unknown, aspects. The capacity-related changes (displayed in Fig. 11) were found in the left dorsal premotor area (see Fig. 7 A; area BB), ipsilateral anterior cerebellum, right posterior cerebellum (see Fig. 7 B; area K), and at or near the right cerebellar dentate. As was observed in the primary motor cortex, activation in the anterior cerebellum shifted lateralization; right-hand performance activated the right anterior cerebellum, left-hand performance the left. As can be seen in Tables 7 and 8, activation in the sq. slow condition was significantly lower than in the other four conditions. Again these effects were found during right- as well as left-hand performance, suggesting that many of these areas show a capacity effect that is independent of the hand used.
A final conditional effect was observed in left superior parietal cortex (see Fig. 7 A; area EE). As can be seen in Table 7, this area was highly activated during the maze tasks but less during square tracing. Post hoc comparisons in the LH performance group showed significant differences in magnitudes between all maze and square conditions (P < 0.01). In the RH performance group, only activations during naive and novel were significantly different from activations during both sq. conditions (P < 0.05) but the pattern was essentially the same.
Blood-flow decreases. Of all the areas that showed decreases during maze and square tracing, only one area showed a significant effect of condition: left middle temporal cortex. This area showed a significant practice-related effect (see Tables 7 and 8) and was more depressed during unskilled naive and novel conditions than during more skilled performance in the prac and sq. conditions (see Fig. 12). None of the other depressed areas showed significant effects of condition.
Cortical-cerebellar connections. As can be seen in Figs. 9 and 11, magnitudes in right cortical areas (right dorsal premotor and right inferior and superior parietal cortex) and left lateral cerebellum showed a different conditional pattern than magnitudes in left dorsal premotor cortex and right cerebellar areas. Namely, right cortical and left cerebellum showed practice-related effects, whereas left cortical and right cerebellum showed capacity-related effects. This not only reveals differences between right and left dorsal premotor cortex as well as between left and right cerebellar regions but also demonstrates a “functional connection” between cortical and contralateral cerebellar areas. The consistency of the activation patterns and effects between cortical areas and contralateral cerebellar regions is demonstrated in Fig. 13.
The aim of this paper was to identify brain areas that are involved in tracing cutout designs with the eyes closed, to investigate differences and similarities related to dominant- and nondominant-hand performance and to examine effects related to practice. During the tracing tasks an increase in blood flow was observed in cortical, subcortical, and cerebellar areas, in many of these areas bilaterally.
Strong similarities were found between right- and left-hand performance both behaviorally and in regional brain activations. Almost all of the areas activated in one group generalized to the other group, showing not only that the activations were very reliable but also that right- and left-hand maze and square tracing activated many of the same areas in right-handed subjects. The only areas that changed laterality with hand of performance were primary motor cortex and anterior cerebellum.
We compared regional cerebral blood flow (rCBF) in subjects performing the maze tracing task for the first time and after extensive practice to determine to what extent the practice-related changes that were observed in performance were related to changes in brain activation. The effects of practice in the activated brain areas were the same for both right- and left-hand performance. After practice, performance was much faster and smoother and more accurate. These behavioral changes were accompanied by decreases in brain activity in right dorsal premotor, right inferior, and superior parietal cortex as well as in left lateral cerebellum, whereas activity in supplementary motor area increased as an effect of practice. These findings clearly show a shift in the pattern and areas activated during unskilled and skilled maze tracing. The same patterns of change across the conditions were found in both RH and LH performance groups.
A transition of brain activation from one or more areas to other area(s) during learning also is reported by Sakai et al. (1998). Using fMRI they showed a shift in brain activations in a visuomotor learning task. Left dorsolateral prefrontal cortex and pre-SMA were mostly activated during early learning, precuneus and right dorsolateral prefrontal cortex were most highly activated in intermediate stages, whereas involvement of the parietal cortex increased in the more advanced learning stage.
Performing the tracing task resulted in increased rCBF in SMA, in the contralateral primary motor cortex, and ipsilateral anterior cerebellum, while premotor, inferior, and superior parietal, inferior frontal, and somatosensory cortex were activated bilaterally as was the thalamus. A cerebellar activation, located in the dentate area, was observed on the right side. More posteriorly the cerebellum was activated bilaterally; however, anterior activation was only found in the left cerebellar cortex. Most of the above-mentioned areas also were activated in a learning study during which right-handed subjects performed a motor sequence with the right hand and eyes closed (Jenkins et al. 1994) as well as in other motor performance studies (Seitz et al. 1997; Stephan et al. 1995; Winstein et al. 1997).
Some of the areas, like inferior frontal and somatosensory cortex as well as the thalamus, were activated to almost the same extent during each condition and might be related to tracing or motor performance in general regardless of the level of skill. The other areas showed activations that were related to conditional differences.
Activations will be discussed separately for the different brain areas.
PRIMARY AND SOMATOSENSORY MOTOR CORTEX.
The tracing task increased rCBF significantly in the contralateral primary motor cortex. The site of the activations in primary motor cortex is consistent with the somatotopic representation of the hand area in humans (Grafton et al. 1991, 1992b, 1993). Left-hand performance activated the left primary motor cortex with a magnitude that was only 10% of that activated by right-hand performance, but this ipsilateral response was significant. In contrast, right-hand performance did not produce a significant activation in the right primary motor cortex. This hemispheric asymmetry in the primary motor cortex also was observed in another PET study (Kawashima et al. 1994) as well as in a fMRI study (Kim et al. 1993). In both studies, subjects were tested on a sequential opposition of thumb-to-finger task with the dominant and nondominant hand. We have no reason to believe that the activation in the ipsilateral motor cortex was related to inadvertent movement of the contralateral hand. Before each scan, subjects were instructed explicitly to move only the hand that performed the tracing tasks. The other hand and arm were checked visually. No movements were observed in the latter. Furthermore, inspection of the EMG recordings of the left arm in the LH performance group during tracing of the 10-segment maze, which was performed with the right hand, showed little or no EMG activity. Activation in ipsilateral primary motor cortex without contralateral hand movement also has been observed in single-unit recordings (Tanji et al. 1988) and cortical microstimulation studies (Aizawa et al. 1990) in monkeys and might be due to the fact that ∼10–15% of the fibers in the lateral cortical spinal tracts are uncrossed in humans (Nyberg-Hansen and Rinvik 1963) and monkeys (Glees and Cole 1952; Yakovlev and Rakic 1966).
As for conditional effects, activations in primary motor cortex showed the same pattern as the velocity; performance at higher speed activated the contralateral motor cortex more than slower performance. This result is consistent with findings in other imaging studies (Blinkenberg et al. 1996; Bonde et al. 1995; Roland et al. 1980a, 1989; Schlaug et al. 1996; Seitz et al. 1990, 1997). The above-mentioned increases are most likely not simply an effect of practice or skill as illustrated by the relatively low magnitude in primary motor cortex in the more skilled sq. slow condition in our experiment. Furthermore, in studies where movement rate and amplitude was kept constant in novel and practiced conditions, no differences in primary motor cortex activation was found (Friston et al. 1992; Jenkins et al. 1994; Raichle et al. 1994).
It is, however, possible that in primary motor cortex, effects related to skill are masked by the velocity-related effects. A number of studies, in which kinematics were held constant, have shown a progressive increase in motor cortex activations during motor learning (Grafton et al. 1992a; Hazeltine et al. 1997; Karni et al. 1995; Pascual-Leone et al. 1994). Although our subjects were more skilled after practice, they had not reached asymptotic performance, as was the case in the study by Karni et al. (1995).
Somatosensory cortex was activated bilaterally in both groups with no significant differences in magnitude between right- and left-hand performance. Although in both groups somatosensory cortex in both hemispheres was most activated during naive maze tracing, the effect of condition was not significant.
SECONDARY MOTOR AREAS.
Whereas the peak activation in the SMA was found in the hemisphere contralateral to the hand used, the replication data of the SMA in Table 5 show that in both groups the medial SMA activations most likely extended across both hemispheres. Because the FWHM in our experiment was 14 mm, it is hard to draw any conclusions from this. We did not replicate the findings of Grafton et al. (1992b) and Kawashima et al. (1994) that the SMA was more activated when subjects performed the task with the left hand. The former studied changes in brain activity while subjects performed visuomotor tracking tasks with the right (12 subjects) or left hand (2 subjects).
Both right- and left-hand performance activated lateral dorsal premotor cortex bilaterally, a finding also observed with fMRI by Rao et al. (1993), who studied subjects while they performed simple and complex finger tapping sequences with either the right or left hand. Although results from other PET studies (Kawashima et al. 1994; Roland et al. 1980b; Shibasaki et al. 1993) demonstrated that the dorsal premotor area contralateral to the hand used was more activated than the ipsilateral area, we found that the left dorsal premotor area was activated more than the right one in both RH and LH performance groups, suggesting lateralized differences with respect to tracing movements.
As for conditional effects, rCBF in both right dorsal premotor cortex and SMA changed as an effect of practice; however, the changes were in opposite directions (see Figs. 9 and 10). Whereas activation in right dorsal premotor cortex was significantly greater during unskilled performance, activation in SMA was significantly greater during skilled performance. These findings are consistent with other PET studies addressing motor learning (Decety et al. 1992; Grafton et al. 1992a; Jenkins et al. 1994; Jueptner et al. 1997b). The dichotomy in SMA and dorsal premotor neuronal activity often is explained by the difference between internal and external cues. It is hypothesized that during initial learning, performance is guided mainly by external sensory cues that are likely to be controlled by the dorsal premotor cortex. However, when the task is overlearned, the movements mostly are self-initiated and guided by internal cues, most likely controlled by SMA (Chen et al. 1995; Deiber et al. 1991; Demiralp et al. 1990; Halsband and Freund 1990; Jenkins et al. 1994; Kawashima et al. 1994; Mushiake et al. 1991; Passingham 1985, 1988; Praamstra et al. 1996; Rizzolatti et al. 1983; Roland et al. 1980a,b; Thaler et al. 1995; Wise 1985).
The high percentage of stops and the long stop duration at intersections during naive and novel conditions suggest that subjects planned and programmed each segment separately during initial performance on the maze. During practice, segments were grouped into chunks of increasing size (see also van Mier et al. 1993). In the prac condition, most subjects could organize and execute the whole sequence as a single unit. It seems reasonable that the programming of the single, more or less externally cued (bumping into the side at the end of the segment) components is controlled by the right dorsal premotor cortex, whereas programming of the internally cued sequence may be controlled by the SMA (Grafton et al. 1992b; Hikosaka et al. 1996; Jueptner et al. 1997a). Evidence that the SMA is likely to be involved because of the sequential character of the task comes from a PET study by Shibasaki et al. (1993), in which sequential finger movements (right thumb had to touch each of the other fingers a different number of times) were compared with repetitively touching the thumb against the tips of all other fingers. SMA activation was significantly higher in the more complex sequential task, a finding that also was reported by Rao et al. (1993) in a fMRI study and in a single cell recording study by Tanji and Shima (1994). Recently, Gerloff et al. (1997) showed that high-frequency repetitive transcranial magnetic stimulation over the supplementary motor area interfered with the organization of subsequent elements in an complex overlearned finger-movement sequence on an electronic piano, indicating a critical role of the SMA in the organization of forthcoming movements in complex motor sequences that are rehearsed from memory and fit into a precise timing plan.
It was found that activations in left dorsal premotor cortex were different from those in right dorsal premotor cortex. Activity in the left dorsal premotor area did not decrease as an effect of practice, as was found in right dorsal premotor cortex, but seemed to be related to capacity. The latter suggests a different role for the left dorsal premotor cortex, the activation pattern of which might be due to involvement of the left dorsal premotor area with the temporal aspects of movement planning (Halsband et al. 1993). The left dorsal premotor area has been postulated to be involved in the acquisition and execution of motor skills, in particular with respect to the temporal aspects of the task (Seitz et al. 1994). These temporal aspects play a role when learning the maze but also when tracing the maze and square at a high speed. The latter requires a precise timing plan as to when to decelerate and accelerate to turn smoothly at corners. When temporal aspects were less demanding, as during sq. slow, when subjects were instructed to bump into the sides, left dorsal premotor area was significantly less activated. We would have expected more or less the same activation in left dorsal premotor cortex during all conditions if this area processes movement-related rules or is involved in motor mapping (Wise 1985). When professional pianists were asked to play a musical score with the right hand, left premotor area was activated, whereas SMA was activated when they played ascending and descending scales (Sergent et al. 1992), again suggesting an involvement of the left premotor area in performance tasks with greater temporal demands.
Although the conditional effects were different in each dorsal premotor area, the effects were the same for both right- and left-hand performance in each area, strongly suggesting that the dorsal premotor cortex in involved in processes related to abstract aspects of this task rather than processes more directly related to the pure motor execution aspects of the task. This is consistent with findings from a single-cell recording study performed by Tanji et al. (1988), who concluded that “... the relation between neuronal activity and the laterality of hand movements is much more complex in the secondary motor areas (premotor cortex and SMA) than in the primary motor area. In secondary areas, the activity of a majority of neurons is not simply related to contralateral movement execution.” Shen and Alexander (1997) showed that when the direction of forelimb movement was dissociated from the spatial location of a target, by varying the spatial mappings between joystick and cursor, the activity of a majority (94%) of directionally tuned neurons in the dorsal premotor cortex was target dependent, whereas none was limb dependent. These findings suggest that the role of the dorsal premotor cortex could be more related to context-dependent processing than to pure motor processing.
A more ventrally located premotor area in inferior frontal cortex (BA 44/6) was activated bilaterally in both groups, with activations in this area being similar in both groups. No significant conditional effects were observed in these areas. Gur et al. (1983) also found bilateral activation in premotor areas 44/6 in subjects performing a spatial task; however, activation was much larger in the right compared with the left hemisphere. Although our tasks had a spatial aspect, subjects did not have to make decisions based only on spatial cues; this might explain the almost similar activations in the right and left premotor areas 44/6 in our study.
Both right- and left-hand tracing activated the ipsi- and contralateral inferior and superior parietal cortex (BA 40 and 7). In Jenkins' (1994) motor learning study, right-hand performance also activated both inferior and superior parietal cortices bilaterally. No parietal activity was reported by Rao et al. (1993), while Kawashima et al. (1994) did not measure activity in parietal areas. In our study, the posterior areas were activated similarly regardless of which hand was used. Responses in the inferior parietal cortex, however, were larger when the task was performed with the contralateral hand, a finding consistent with results reported by Grafton et al. (1992b), suggesting a hand effect without hemispheric lateralization. The inferior parietal cortex is known to be involved in spatial attention tasks (Petersen et al. 1994) and was activated when subjects attended to tactile stimulation (Pardo et al. 1991) and when subjects covertly shifted attention to different locations in the visual field (Corbetta et al. 1993).
Although superior parietal activations often are ascribed to perceptual aspects of a task and are observed in tasks in which a visual cue is used to trigger a motor response (Grafton et al. 1992b) or when subjects perform a perceptual maze task (Ghatan et al. 1995), parietal activations also are found without any visual input with imagined movements (Stephan et al. 1995), with auditory cues (Deiber et al. 1991), with performance of a prelearned saccade sequence in darkness (Petit et al. 1996), or with construction of mental three-dimensional images based on verbally presented instructions. This suggests that spatial processing demanded by the tasks requires the involvement of the posterior parietal cortex.
Whereas the parietal cortex is thought to be responsible for the sensory component during spatial processing, the premotor cortex may be involved in the motor component when a motor response has to be made using spatial information. This dorsal parietofrontal pathway also was observed in our study, demonstrating a relationship between activations in parietal and premotor cortex, as described in a variety of spatial tasks (Corbetta et al. 1993; Ghatan et al. 1995; Haxby et al. 1994; Mellet et al. 1995; Petit et al. 1996), including those not associated with movement (Mellet et al. 1996).
As for conditional effects, right inferior and superior parietal cortex showed significant practice-related decreases. These areas were less activated when the task was more skilled, (prac, sq. fast, and sq. slow) (similar to Jenkins et al. 1994; Jueptner et al. 1997b). In the naive and novel conditions, subjects most likely have to attend to the spatial aspects of the task, aspects that need less or no attention once the task can be performed more skillfully. In a study by Roland et al. (1980b) in which subjects were guided through a maze by verbal commands, inferior and superior parietal cortex also were activated bilaterally, as was the case in a study by Ghatan et al. (1995), who used a perceptual maze test. In the latter study, no activation in these areas was found when subjects randomly pressed keys to guide a cursor from bottom to top in a control condition. The practice-related effects found in right parietal cortex were again hand independent.
A significant effect of condition also was found in left superior parietal cortex. However, differences between conditions were not velocity, practice, or capacity related. As can be seen in Table 7, this area was more activated during maze tracing than during square tracing, suggesting a task-related effect, an effect that was strongest during left-hand performance. Jueptner et al. (1997b) report activation in left inferior and superior parietal cortex during both novel and practiced performance on finger movement sequences.
There is one other PET study that compared cerebellar activations during right- and left-hand performance (Grafton et al. 1992b). They measured only the ipsilateral anterior cerebellum and reported significant changes during performance with either hand. However, the changes were bigger when the task was performed with the left hand, consistent with our results. The finding that movement planning and/or execution activate the ipsilateral anterior cerebellum is supported by several other PET studies (Colebatch et al. 1991; Decety et al. 1992; Deiber et al. 1996; Grafton et al. 1991, 1992a, 1993, 1994; Seitz et al. 1994; Stephan et al. 1995). Apart from the anterior cerebellum, we also found activation bilaterally in the inferior cerebellar cortex and right cerebellar dentate.
During the last 20 yr, more and more evidence suggests that the cerebellum is not only involved in motor control but also contributes to processes including learning (see e.g., Doyon 1997; Fiez 1996; Raichle et al. 1994; Schmahmann 1996; Thach 1997) and cognitive processing (Kim et al. 1994; Middleton and Strick 1994, 1997), findings that are in line with results from studies involving cerebellar patients (Doyon 1997; Doyon et al. 1997; Fiez et al. 1992; Molinari et al. 1997; Sanes et al. 1990). Our results support this hypothesis. It is, however, still unclear which aspects of learning are controlled by the cerebellum—temporal, spatial, attentional, and/or sequential—and which areas of the cerebellum control these aspects. Recently Middleton and Strick (1997) showed evidence that different output channels in the cerebellar dentate project to different cortical areas and therefore might be concerned with distinct aspects of behavior at a motoric as well as cognitive level.
The activation that we found in the anterior cerebellum is probably related to movement execution (Allen et al. 1997). As in the primary motor cortex, this area shifted lateralization when the opposite hand was used. However, the activation did not show a velocity-related change but rather a capacity-related change. The latter is consistent with results reported by Seitz et al. (1990), who found no increase in anterior cerebellar activity from initial to skilled performance, although the frequency of finger movements almost doubled. The fact that the anterior cerebellum was significantly less activated during sq. slow than during the other conditions, but that the activation was lateralized, suggests that this area might relate to movement timing at a muscle-specific level. The other cerebellar activations showed significant conditional effects that were independent of the hand used and suggest processing at a more abstract level.
We found that activation in the left cerebellar hemisphere decreased as an effect of practice. When subjects were learning the maze (naive or novel) left cerebellum was highly activated. After practice, the activation had dropped considerably and was reduced further during performance of the squares. These practice-related changes in left cerebellum are comparable with results reported by Jenkins et al. (1994) and Jueptner et al. (1997b). Left cerebellar activations also are reported by Ghatan et al. (1995), who tested subjects on a perceptual maze with little pretraining. Furthermore Lalonde and Botez (1990) showed that animals with cerebellar damage were impaired on maze learning. Flament et al. (1996) reported a strong correlation between cerebellar activation and errors. The fact that the six (RH performance group) and eight (LH performance group) subjects in which this left cerebellar area was scanned still made some errors during the prac condition, in contrast to both sq. conditions, might have accounted for the trend effect when activations in this area were compared between naive and prac.
Two areas in the right cerebellum, one most likely located in the dentate area, the other more medial and posterior, showed capacity-related changes and may be more related to temporal aspects of the task as has been described above for the left dorsal premotor cortex. This specific result is consistent with proposals that the cerebellum is involved in the regulation of movement timing (Raymond et al. 1996) as well as issues of timing more generally (Ivry 1997; Keele and Ivry 1990).
The differential effects in the cerebellar hemispheres are consistent with data from cerebellar patients. Molinari et al. (1997) found that although patients with cerebellar lesions were severely impaired on a procedural motor learning task, patients with lesions in the left cerebellum had even more problems learning the motor task than patients with right cerebellar lesions. Furthermore, their patients were impaired during both right- and left-hand performance.
Thalamus was activated bilaterally in both groups. In each group, activations were slightly, but not significantly, larger on the contralateral side. Bilateral activation of the thalamus is not surprising because the thalamus can relay information between cerebellum and (pre)motor cortex (Shinoda et al. 1993), areas which also were activated bilaterally in our tracing tasks.
The only area that was activated in the LH performance group and not in the RH performance group was the midbrain. Jenkins et al. (1994) report activation in this area in their right-handed subjects performing the task with the right hand and suggest that this activation, which seems to be located in the red nucleus, might represent a cerebellorubral pathway. The red nucleus is known to mediate voluntary movement especially distal limb movements probably through an indirect pathway (Kuypers 1973, 1981). It is possible that this pathway is used mainly when subjects perform the tasks with their nondominant hand or when subjects, as in the Jenkins' study, have to relearn a new sequence, tasks that both might require increased attention. This is in agreement with a study by Sadato et al. (1997), who only found increased activation in the midbrain when bimanual sequential parallel movements were performed but not during bimanual sequential mirror movements, the latter requiring less attention.
No significant conditional effects were found for the brain activations in thalamus and midbrain.
As was the case with blood-flow increases, most areas showing decreases were identified during both RH and LH performance. Areas that were less activated during tracing performance in general than during the rest condition were found in temporal, occipital, cingulate, and prefrontal cortex and caudate. Results from PET studies that reported decreases suggest that when subjects attend to a specific modality, activity in modalities not needed for the task is suppressed (Buckner et al. 1995; Friston et al. 1991; Frith et al. 1991; Grafton et al. 1994; Haxby et al. 1994; Jenkins et al. 1994; O'Sullivan et al. 1994; Seitz and Roland 1992). Compared with the resting condition, attending to the somatosensory motor aspects of the tracing task depressed activity in medial occipital cortex and temporal cortex, areas involved in the processing of visual and auditory information, information that might not be relevant for this task. Most of the other areas in which the observed decreases were replicated might be areas in which activations commonly are depressed during active conditions compared with passive tasks, regardless of the task performed. In an across-study analysis, involving nine studies of human visual information processing, Shulman et al. (1997) found reduced activity during active scans in many of the same areas seen here. This suggests that these decreases generalize over tasks and might be caused by increased activation during passive conditions reflecting common ongoing processes.
As for conditional effects, only one of the areas in which reduced brain activity was observed showed a practice-related effect: left middle temporal cortex. This area processes auditory information and was significantly more depressed during unskilled naive and novel than during skilled performance in the prac and both sq. conditions. It is possible that subjects attended more to the somatosensory aspects of the motor task during naive and novel than during the other conditions and therefore were suppressing auditory information more in the former conditions.
The fact that all subjects performed the tasks in the same fixed order could be seen as a limitation of the study. A fixed-task order is inherent in the experimental design. We are, however, confident that the practice-related effects are indeed related to practice and not to consolidation or time effects. The finding that when subjects were presented after practice with a novel maze, the same areas that showed changes in activation during naive maze tracing became activated or deactivated again, almost to the same extent, supports this hypothesis. That the changes were not related to changes in movement kinematics per se is demonstrated by comparison of brain activity in areas measured during prac as well as during both sq. conditions. Several areas showed no significant differences in brain activations among these three conditions, although the velocity was three to five times higher during prac and sq. fast than during sq. slow. Furthermore although somatosensory and proprioceptive input might have been different among these three tasks, these differences did not translate to differences in activations in practice-related areas between the three conditions, again strongly suggesting that the changes were related to level of skill. A direct comparison between the first and last fast square tracing did not reveal any significant differences in brain activations, although subjects performed the second square tracing ≥45 min after the first.
The results of this study demonstrate two important principles: that practice produces a shift in activity from one set of areas to a different area and that these practice-related activations appeared in the same areas and hemisphere regardless of the hand used, suggesting that these areas code information at a level abstract from motor implementation itself.
Areas relating directly to motor performance, including primary motor cortex and anterior cerebellum, shift lateralization during the performance of maze and square tracing with the right and left hand. Activation in the other cortical and cerebellar areas was independent from the hand used to perform the tasks, suggesting that these areas code information that is abstract from the motor performance of the task itself. The representation of the movement pattern seems to be abstract rather than muscle specific (Keele 1981) and might contain sequential, temporal, and/or spatial aspects.
It was demonstrated in two separate groups that brain areas are activated differentially during naive and practiced maze tracing. Although unskilled performance involved a number of right cortical areas and left cerebellum, activation shifted to the SMA during skilled performance. The differential effects in different areas suggest that right dorsal premotor cortex and SMA might be associated with the sequential aspects of the task, whereas left dorsal premotor cortex and right cerebellum are more likely to be associated with the temporal aspects of the task. Spatial and attentional aspects might explain the involvement of parietal areas while the left cerebellum might be related to error processing and correction. These effects were the same regardless of the hand used, once more suggesting that the processes related to this motor learning task activate areas that most likely code abstract information rather than muscle specific information.
We thank D. Hunton, M. Makram, L. Lich, J. Hood, T. Videen, and the staff of the Cyclotron Unit for technical assistance.
This work was supported by National Institutes of Health Grants NS-32979, NS-06833, and HL-13851 and the McDonnell Center for the Study of Higher Brain Function.
Address for reprint requests: H. van Mier, Dept. of Neurology and Neurological Surgery, Washington University School of Medicine, 660 S. Euclid, P.O. Box 8111, St. Louis, MO 63110.