Cerebral Cortex, Vol. 12, No. 11, 1157-1170,
November 2002
© 2002 Oxford University Press
The Role of Rostral Brodmann Area 6 in Mental-operation Tasks: an Integrative Neuroimaging Approach
1 Human Brain Research Center, Kyoto University Graduate School of Medicine, Kyoto, Japan, , 2 Human Motor Control Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, USA, , 3 PRESTO, Japan Science and Technology Corporation, Kawaguchi, , 4 Laboratory of Cerebral Integration, National Institute for Physiological Sciences, Okazaki, , 5 Department of Nuclear Medicine, Kyoto University Graduate School of Medicine, Kyoto, , 6 Biomedical Imaging Research Center, Fukui Medical University, Fukui and , 7 Autonomic Research Project, National Institute for Physiological Sciences, Okazaki, Japan
Address correspondence to Manabu Honda, Laboratory of Cerebral Integration, National Institute for Physiological Sciences, 38 Nishigonaka, Myodaiji-cho, Okazaki 444-8585, Japan. Email: honda{at}nips.ac.jp.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Recent evidence indicates that classical ‘motor’ areas may also have cognitive functions. We performed three neuroimaging experiments to investigate the functional neuroanatomy underlying three types of nonmotor mental-operation tasks: numerical, verbal, and spatial. (i) Positron emission tomography showed that parts of the posterior frontal cortex, which are consistent with the pre-supplementary motor area (pre-SMA) and the rostral part of the dorsolateral premotor cortex (PMdr), were active during all three tasks. We also observed activity in the posterior parietal cortex and cerebellar hemispheres during all three tasks. Electrophysiological monitoring confirmed that there were no skeletomotor, oculomotor or articulatory movements during task performance. (ii) Functional magnetic resonance imaging (fMRI) showed that PMdr activity during the mental-operation tasks was localized in the depths of the superior precentral sulcus, which substantially overlapped the region active during complex finger movements and was located dorsomedial to the presumptive frontal eye fields. (iii) Single-trial fMRI showed a transient increase in activity time-locked to the performance of mental operations in the pre-SMA and PMdr. The results of the present study suggest that the PMdr is important in the rule-based association of symbolic cues and responses in both motor and nonmotor behaviors.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The ability to alter our environment by manipulating objects has rarely been considered to have a close functional relationship with our ability to alter our mental contents by manipulating information in memory. Consider, however, the development of mental abacus skills. After being fully trained on an abacus and then on a mental simulation of abacus use, some people can execute complex mental calculations without any accompanying overt movement (Hatano and Osawa, 1983
; Hanakawa et al., 1999b). The development of this ability is an example of a natural transition from a motor skill to a nonmotor cognitive skill, as observed also for other cognitive skills such as visual shape discrimination (Parsons et al., 1995). Indeed, the distinction between motor and cognitive skills has recently blurred, especially regarding the subserving brain structures. The cerebellum and basal ganglia, once thought to be involved exclusively in motor control, are now recognized to have cognitive functions (Alexander et al., 1986; Schmahmann, 1997).
One candidate cortical area that may mediate a transition from motor to cognitive functions is the ‘non-primary’ motor cortex or Brodmann area (BA) 6, which is part of the frontal agranular cortex rostral to the primary motor cortex (M1) or BA4 (Geyer et al., 2000). BA6 can be subdivided into at least three functional subdivisions: the supplementary motor areas (SMA) on the medial hemispheric wall, and the dorsal and ventral parts of the lateral premotor cortex (PMd and PMv, respectively). More recent studies in animals show that the SMA, PMd and PMv each have distinct rostral and caudal subdivisions (Matelli et al., 1985; Matsuzaka et al., 1992; Luppino et al., 1993; Lu et al., 1994; Preuss et al., 1996; Geyer et al., 1998). Neuroanatomically, the caudal parts of the SMA (SMA proper), PMd (PMdc) and PMv (PMvc) send substantial projections to M1 and the spinal cord. Notably, it has been previously suggested that only those areas plus the cingulate motor areas, which also have direct connections to M1 or the spinal cord (Picard and Strick, 1996), be termed ‘true’ motor areas. In contrast, the pre-SMA and the rostral part of PMd (PMdr) closely interconnect with the prefrontal cortex (PFC) rather than M1 (Barbas and Pandya, 1987; Luppino et al., 1993; Lu et al., 1994) and lack a direct projection to the spinal cord (He et al., 1993, 1995). Physiologically, the pre-SMA and PMdr have been shown to be involved in the sensory aspects of sensorimotor integration more than the caudal counterparts (Weinrich and Wise, 1982; Johnson et al., 1996; Shen and Alexander, 1997). These findings suggest that the function of the pre-SMA and PMdr is independent of immediate movement and more closely related to the function of the PFC. Such a rostrocaudal gradient of BA6 functions likely exists in humans too (Rizzolatti et al., 1998; Geyer et al., 2000), which prompted us to re-examine the role of BA6 in human behavior. In the present study, we investigated possible non-motor functions of the human homologues of the pre-SMA and PMdr, for which motor functions in humans have been well established (Foerster, 1936; Freund and Hummelsheim, 1985; Luders et al., 1995).
There is some previous evidence in support of cognitive functions of BA6. An earlier neurophysiological observation in monkeys identified a role of the PMd in nonmotor and motor information processing (Vaadia et al., 1988). In more recent neuroimaging studies, various kinds of cognitive tasks induced activity in parts of BA6 (Jonides et al., 1993; Paulesu et al., 1993; Dehaene et al., 1996; Fiez et al., 1996; Mellet et al., 1996). One concern, however, is that many of these neuroimaging studies utilized tasks that required motor responses and then employed a cognitive subtraction method, which assumes that the motor component of the cognitive tasks (i.e. preparation for and execution of motor responses) can be corrected for by subtracting out activity evoked by a control motor task. This approach carries the assumption that there are no interactions between cognitive information processing and motor control (Friston et al., 1996). This is problematic in evaluating areas that may have both motor and cognitive functions. It is likely that the BA6 activity observed in the previous neuroimaging studies includes a conjunction of or interaction between cognitive information processing and motor control. In the present study, therefore, a series of neuroimaging studies were conducted to characterize the cognitive aspects of pre-SMA and PMdr functions in humans, using a task that did not involve overt motor responses during the acquisition of functional brain images.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Study Design
In the present study, three neuroimaging experiments were conducted to examine neural activity during mental-operation tasks that did not involve immediate motor responses. The mental-operation tasks were designed such that no motor responses were required during image acquisition, but there was still objective confirmation of the task performance. The tasks involved serial manipulation of mental representations in three different information modalities: numerical, verbal and spatial (Fig. 1). First, a global pattern of brain activity was examined using positron emission tomography (PET) during performance of these tasks and control cognitive tasks. The subjects were electrophysiologically monitored to confirm the absence of any overt movement during the task performance. Secondly, a functional magnetic resonance imaging (fMRI) experiment was performed to localize activity in BA6 evoked during the mental-operation tasks compared with activity evoked during control skeletomotor and oculomotor tasks (fMRI localization experiment). Finally, a modified version of the numerical mental-operation task was used to investigate the temporal pattern of brain activity evoked during different components of the numerical mental-operation task (single-trial fMRI experiment). We then analyzed the portions of BA6 activated during the mental-operation tasks. Although the functional neuroanatomy of the human homologue of the monkey PMvr is an interesting issue (Aboitiz and Garcia, 1997; Rizzolatti and Arbib, 1998), we focused our analysis on the dorsal and medial parts of BA6, the pre-SMA and PMdr, because the spatial mental-operation task did not activate the portion of BA6 corresponding to PMv in the present study.
|
Subjects
Ten young healthy adults were studied in each of the experiments: PET, fMRI localization, and single-trial fMRI. There were 17 subjects in total (29 ± 4.9 years old; 15 men, two women). They were recruited from the students or graduates of Kyoto University. All were right-handed, as assessed by the Edinburgh inventory (Oldfield, 1971). All subjects gave written, informed consent to participate according to the study protocol approved by the institutional ethics committee.
Mental-operation Tasks
All of the mental-operation tasks began with the visual presentation of a prime stimulus, followed by a series of visual instruction stimuli, presented at a steady rate. All stimuli subtended a visual angle of 
2.5° at the center of view so that subjects easily identified each stimulus without moving their eyes. We used SuperLab (Cedrus, Phoenix, AZ) on a Power MacIntosh computer (Apple Computer Inc, Cupertino, CA) to control visual stimulus presentation. In the numerical task, each stimulus was a single numeral selected randomly from 1 through 9. Subjects were asked to silently add the numerals and then verbally report the final sum. In the verbal task, the prime stimulus was a kanji character indicating one of the days of the week, and each instruction stimulus was a numeral 1 through 3. The subjects silently and serially advanced the day of the week according to the instruction stimulus. For example, the day was advanced from Wednesday to Friday with an instruction stimulus of 2, and from Friday to Monday with an instruction stimulus of 3. They reported the day they eventually reached after each task period. Although it is theoretically possible to add up all the presented numbers and advance the day according to the sum, the subjects were explicitly instructed to avoid this strategy. In the spatial task, the prime stimulus was a marker presented in one square of a grid subdivided into nine small squares, and each instruction stimulus was either an arrow or a pair of tandem arrows pointing in one of four directions (up, down, right or left). Subjects were requested to mentally move the marker, according to the instruction stimuli, on the imagined grid that was no longer visible on the screen. They reported the final location of the marker after each task period. Instruction stimuli that would put the marker outside of the grid were not used. In summary, the numerical and verbal tasks required mathematical operations on numerals and word strings, respectively, in one-dimensional space, while the spatial task required vector-type operations on two-dimensional spatial locations. Aside from a common feature, namely involvement of rule-based mental operations, the strategies used to complete the tasks were expected to be different: a verbal strategy for the numerical and verbal tasks (Dehaene et al., 1999) and a visuospatial strategy for the spatial task. The numerical task was modified for the single-trial fMRI experiment as described below.
Control Tasks
Visual-fixation Task
The visual-fixation task served as a baseline task for all of the tasks. The subjects were asked to look straight ahead and fixate on a cross flashing on and off in the center of view at the same rate as the instruction stimuli for the mental-operation tasks.
Verbal-rehearsal Task
The verbal-rehearsal task was used as a control cognitive task in the PET experiment. The verbal-rehearsal task began with visual presentation of a prime stimulus, which was 10 numerals presented simultaneously and which subtended a visual angle of 6° x 2.5°. The prime stimulus was presented for 18 s and then removed before tracer injection. During the task period, subjects were asked to mentally rehearse the numerals without using any mouthing movement. The rehearsal was explicitly paced with a cross flashing on and off in the center of view at the same rate as the mental-operation tasks. The subjects reported the memorized 10 numerals after the scan. This task was designed to investigate brain activity related to subvocalization, or inner speech, which would be involved in the numerical and verbal mental-operation tasks. The verbal-rehearsal task differed from the mental-operation tasks, however, in that it did not involve rule-based serial information updating in response to instruction stimuli.
Complex and Simple Finger-tapping Tasks
In the fMRI localization experiment, complex and simple finger tapping were used as control skeletomotor tasks to operationally subdivide the lateral part of BA6. For both tasks, finger tapping was visually paced by a circle (size, 0.3°) flashing on and off in the center of view at a rate of 1 Hz. The complex finger-tapping task required an execution of brisk, discrete finger-to-thumb opposition of the right hand, with a pre-determined sequence comprising twelve taps (I-II-III-IV-I-III-II-IV-IV-II-III-I, where I, II, III and IV denote the index, middle, ring and little finger, respectively). The subjects practiced this tapping task sufficiently enough to be able to perform it semi-automatically. Previous neuroimaging studies using this particular type of finger-tapping task showed that the complexity of the task correlates with activity in a portion of the PMd (Sadato et al., 1996; Catalan et al., 1998). Therefore, the simple finger-tapping task was employed as a control task for the complex finger-tapping task to highlight the PMd subdivisions relevant to higher-level motor control. The simple finger-tapping task required a simultaneous opposition of the four fingers against the thumb of the right hand. The complex finger-tapping task was considered more demanding than the simple finger-tapping task, because the former involved a more complex spatial pattern of movement than the latter.
Oculomotor Task
In the fMRI localization experiment, a control oculomotor task was used to functionally define the frontal eye fields (FEF), which are most likely represented in part of BA6 in humans (Paus, 1996). In this task, the subjects made a horizontal saccade to a cross (size, 0.3°), presented alternately to each visual hemi-field at a rate of 1 Hz (mean amplitude, 15°; range, 10–20°).
Image Acquisition and Analysis
PET Experiment
Ten men, aged 20–33 years, participated in the PET experiment, in which the relative regional cerebral blood flow (rCBF) was measured as a marker of neuronal activity. Thirty-five axial slices with interslice spacing of 4.25 mm were acquired using an Advance PET scanner (GE/Yokogawa, Tokyo, Japan) with the interslice septa retracted. A 10 min transmission scan was performed with two rotating 68Ge sources for attenuation correction. The subjects lay supine on the scanner bed in a dimly lit, sound-attenuated room. The visual stimuli were presented on a video-computer screen placed 55 cm from the subjects’ eyes. Head motion was minimized using an elastic band and sponge cushions. A 10 mCi bolus of H215O was injected into the left cubital vein 18 s after presentation of the prime stimulus, coinciding with the presentation of the first instruction stimulus (Fig. 2A). Forty instruction stimuli were serially presented at a rate of 0.33 Hz. PET image collection was started when the initial increase in radioactivity was observed, which occurred 20–25 s after the tracer injection. The PET image collection lasted for 60 s. The subjects continued the task for at least 30 s after the PET image acquisition had finished, and then verbally reported the task solutions (Fig. 2A). Because the rCBF measurements were completed long before the time for the final verbal responses, the preparation for or the generation of the verbal responses were assumed to contribute minimally, if at all, to the imaging results. Additionally, the subjects were instructed to remain completely still, except for natural blinks, until the final verbal report. To confirm the absence of overt movements during the task performance, electrooculograms (EOG), surface electromyograms (EMG) and glossokinetic potentials were monitored and stored in a digital electroencephalograph (Nihon-Koden Co., Tokyo, Japan) for subsequent review. The EOGs were recorded using two pairs of electrodes: one horizontal and one vertical. Surface EMGs were recorded using five pairs of electrodes placed over the bilateral extensor and flexor digitorum muscles and the right tibialis anterior muscle. The glossokinetic potentials were recorded using a pair of electrodes, one placed at the right oral angle and the other placed below the chin. The impedance of all electrodes was <5 k
. All signals were amplified 10 000 times and filtered (0.016–60 Hz for the EOGs and glossokinetic potentials, and 20–60 Hz for the surface EMGs). To minimize learning effects, the experiments were conducted after the subjects had fully learned the tasks, and the task order was pseudo-randomized across subjects. The subjects underwent two PET scans for each of the five tasks (numerical, spatial and verbal mental-operation; visual-fixation; and verbal-rehearsal tasks). The interscan interval was 10 min.
|
The images were reconstructed using Hanning filters, giving transaxial and axial resolutions of 6 and 10 mm full-width at half-maximum (FWHM), respectively. The field of view and pixel size of the reconstructed images were 256 and 2 mm, respectively. No arterial blood sampling was performed, thus the images collected were those of tissue activity. For image analysis, we used a statistical parametric mapping software package (SPM96, http://www.fil.ion.ucl.ac.uk/spm) implemented in MATLAB (MathWorks Inc., Natick, MA). PET images were realigned using the first image of the session as a reference (Friston et al., 1995a
). The images were spatially normalized to fit the Montreal Neurological Institute template (Evans et al., 1993) based on a stereotaxic coordinate system (Talairach and Tournoux, 1988), and resampled into voxels that were 2 x 2 x 2 mm in the x (right–left), y (rostral–caudal) and z (dorsal–ventral) directions, respectively. All images were smoothed with an isotropic Gaussian filter of 15 mm FWHM to account for variations in normal gyral anatomy. Using a general linear model, regionally specific effects of the task conditions were tested on the basis of group-averaged data. Five factors of interest, each corresponding to each task condition, were modeled. A systemic difference among subjects was removed as a confounding block effect. To account for global changes of activity among scans, the global activity in each scan was modeled as a confounding factor for each subject separately in the model (subject-specific analysis of covariance with 10 confounding effects). Consequently, the degree of freedom for the residual errors was 76. The estimates were compared by the use of linear contrasts between task pairs. The resulting set of voxel values for each contrast constituted a statistical parametric map of t-statistics. The t-values were transformed into the unit normal distribution (Z-value), which was independent of the degree of freedom of the error. Statistical parametric maps were first thresholded at a Z-value of 3.09. To account for multiple nonindependent comparisons, the significance of the activity detected in each brain region was estimated using distributional approximations from the theory of Gaussian fields in terms of spatial extent or peak height, or both (Friston et al., 1995b). An estimated P < 0.05 was used as the final threshold for significance. fMRI localization experiment
In nine men and one woman, aged 21–33 years, blood-oxygenation-level-dependent (BOLD) signals were measured as a marker of neuronal activity using a 1.5 T whole-body GE/Signa scanner with a standard head coil (Milwaukee, WI). Visual stimuli were mirror-reversed and back-projected onto a screen placed 55 cm away from the subjects’ eyes. The subjects lay supine on the scanner bed and viewed the stimuli through a mirror built into the head coil. They performed the numerical and spatial mental-operation tasks as well as the control skeletomotor and oculomotor tasks. The mental-operation tasks were slightly modified for this fMRI experiment. The verbal mental-operation task was not included in this experiment because the results from the PET experiment revealed a similar pattern of activity in the rostral BA6 for both the numerical and verbal mental-operation tasks. The prime stimulus was presented for 2 s at the beginning of each task period, followed by 15 instruction stimuli presented at a rate of 0.5 Hz. Immediately after each task period the subjects reported the task solutions in response to two visual stimuli presented side by side (numerals or grids), subtending a visual angle of 2.5° x 5°. The task solution was reported via a wrist extension movement: a left wrist extension indicated that the left figure was correct, a right wrist extension indicated that the right figure was correct, and a bilateral wrist extension indicated that neither was correct. Thus, the probability of a randomly occurring correct response was 33%. Each mental-operation task was alternated with the visual-fixation task four times in 30 s blocks. The complex finger-tapping task was also alternated with the simple finger-tapping task in separate sessions. The subjects underwent two scanning sessions for each of the five condition pairs (numerical and spatial mental-operation tasks, complex finger-tapping task, and oculomotor task each versus a visual-fixation task; and complex finger-tapping task versus simple finger-tapping task) in a pseudo-randomized order.
Using BOLD-sensitive, single-shot, gradient-echo echo-planar images, 38 contiguous axial slices were acquired. The scan parameters were as follows: repetition time (TR) = 6 s, echo time (TE) = 42 ms, flip angle (FA) = 90°, FOV = 22 cm, 64 x 64 matrix, 3.44 mm cubic voxel. For anatomical coregistration, three-dimensional (3-D), fast spoiled, gradient-recalled-at-steady-state images were acquired as follows: TR = 11.2 ms, TE = 2.1 ms, inversion time = 300 ms, FA = 30°, FOV = 22 cm, 256 x 256 matrix, 124 slices, voxel size = 0.86 x 0.86 x 1.5 mm. During the scans, the task performance was visually monitored by one of the authors (T.H. or M.H.).
The initial two images of each session (12 s) were discarded to allow for any possible magnetic instability during the first 10 s of MRI scanning. To reduce head-motion effects, all functional images were spatially aligned to the remaining first one, and then were resliced using a least sum-of-squares method with 3-D sinc interpolation (Friston et al., 1995a). Estimated head motion was <0.6 mm for all subjects. The functional images were spatially normalized into the standard stereotaxic space based on the Montreal Neurological Institute template (Evans et al., 1993), using the coregistered anatomical MRI as a reference. In this process the images were resampled into 2 mm cubic voxels.
In this experiment, we analyzed the functional images primarily on a subject-by-subject basis to localize the activity in BA6 as precisely as possible. For this individual analysis, the functional images were smoothed using an isotropic Gaussian kernel of 4 mm FWHM (i.e. twice the final voxel size). Linear detrending was applied to remove gradual signal drifts, and differences in global signal intensity were removed via proportional scaling. A general linear model was used to estimate significant correlation between boxcar functions optimally convolved with a hemodynamic response function and BOLD signal changes. A systemic difference among different imaging sessions was removed as a confounding block effect. Linear contrasts to the parameter estimates were used to test the task effects, yielding Z-value maps. The estimated spatial resolution of the individual analysis was 6.2 x 5.7 x 5.7 mm FWHM, on average. Group-averaged data were also analyzed to test the consistency of activation across subjects and to compare the results with those obtained in the PET experiment. In the group analysis, smoothing was accomplished using a 7 mm FWHM Gaussian kernel (i.e. 3 times the final voxel size) to compensate for normal variation of gyral anatomy across subjects. Repeated measures (scans) were collapsed within subjects via proportional scaling, which yielded one image per condition per subject. The statistical analysis for group inference was performed as in the PET experiment by removing a systemic difference among subjects as confounds. The final spatial resolution of the group-averaged analysis was 9.1 x 8.6 x 8.7 mm FWHM. For both the individual and the group analyses, the height threshold and spatial-extent threshold were set at Z > 3.09 and P < 0.05 with correction for multiple comparisons, respectively.
Single-trial fMRI Experiment
A total of eight men and two women, aged 20–32 years, participated in the single-trial fMRI experiment. The experimental set-up was almost the same as in the fMRI localization experiment. The numerical mental-operation task was modified for this experiment (delayed mental-addition task). Two single-digit or double-digit numbers were presented serially, with a 15 s delay. Subjects were asked to memorize the first number stimulus (S1), and add the two numbers in response to the second stimulus (S2), which always required carrying to the 10’s place. The subjects completed 20 trials for the delayed mental-addition task with an intertrial interval of 15 s, during which they reported the sums by responding with wrist extension movements, as described in the fMRI localization experiment.
The delayed mental-addition task required recognition and registration of numbers, phonological rehearsal, and increased attention at the presentation of both stimuli. It was assumed that only the response to S2 would involve mathematical mental operations, associated with the retrieval of S1 from short-term memory as well as the arithmetic tables from long-term memory. The S2-specific cognitive operation involved a crucial component of the mental-operation tasks, which is rule-based updating of memory contents according to a new sensory stimulus. Therefore, activity similarly following both S1 and S2 was considered to reflect relatively non-specific cognitive components, and activity specifically following S2 was assumed to represent a core cognitive component related to the numerical mental-operation task. The presence of transient activity was also expected to help exclude motor inhibition from a list of putative roles of BA6 because motor inhibition was required throughout the task periods.
Five axial slices were acquired for the dorsal and medial parts of BA6 above the cingulate sulcus, by using single-shot, gradient-echo echoplanar images (TR = 1 s, TE = 43 ms, FA = 60°, 64 x 64 matrix, FOV = 22 cm, 5 mm thickness with 1 mm gap). T1-weighted images (TR = 600 ms, TE = 17 ms, FA = 30°) were obtained for the corresponding area for anatomic coregistration. Whole-brain 3-D images were also acquired.
The initial nine scans (9 s) were excluded from the analysis to allow for possible magnetic instability. The resultant images were realigned to the first remaining image and resampled using 3-D sinc interpolation (i.e. size of voxels entering into statistical estimation = 3.44 x 3.44 x 5 mm). The images were then smoothed using a 10 mm FWHM Gaussian kernel (i.e. 3 times the final voxel size). Linear detrending and temporal smoothing using a 4.7 s temporal FWHM Gaussian kernel were applied for the autocorrelation of the fMRI time-series (Friston et al., 1994). Global normalization was accomplished via linear scaling. Statistical analysis was performed on a subject-by-subject basis. Boxcar functions were used to model the experimental conditions and served as regressors for multiple regression analysis after convolution with the canonical hemodynamic response function. Using a general linear model, a weighting coefficient was calculated for each regressor. We focused on transient signal changes that were time-locked to the visual stimulus presentation. For S1-related activity and S2-related activity, the t-deviate at each voxel was calculated using a linear contrast of [–1, 2, –1, 0, 0] and [0, 0, –1, 2, –1], respectively. The stimulus-related activity was defined activity that occurred equally during both stimuli (conjunction analysis). The conjunction analysis yielded t-statistics maps, which reflect conjoint testing for multiple task effects (main effects) and, at the same time, eliminate voxels where differences among those effects (interactions) are significant (Price and Friston, 1997). The S2-specific activity was defined as the transient activity that was greater in response to S2 than to S1. More specifically, a linear contrast of [0, – 1, 0, 1, 0] was given, and the voxels that remained after this analysis were masked by the S2-related activity (uncorrected P < 0.05) to minimize the confounding effects of substantial deactivation time-locked to the S1 presentation. The height threshold and spatial extent threshold was set at Z > 3.09 and P < 0.05 with correction for multiple comparisons. The location of the activated foci was first visualized by overlaying the Z-value map onto the partial anatomic images corresponding to the functional images. To obtain stereotaxic coordinates of the activated areas, the Z-value map was transformed into the standard stereotaxic space (Montreal Neurological Institute template) (Evans et al., 1993), by applying the parameters computed from the partial anatomic images coregistered onto the whole-brain images (Toma et al., 1999). Estimated spatial resolution of the subject-by-subject analysis was 12.8 x 12.0 x 8.1 mm FWHM, on average.
Anatomical Nomenclature
In the present report, we use the term BA6 to indicate the agranular frontal cortex situated between M1 and PFC. The subdivisions of BA6 and its border with M1 were determined on the basis of the recent proposals for the anatomy of human frontal agranular cortex (Zilles et al., 1995; Picard and Strick, 1996; Preuss et al., 1996; Wise et al., 1997; Rizzolatti et al., 1998; Geyer et al., 2000). BA6 was subdivided into medial, dorsolateral and ventrolateral parts. Medial BA6, which is above the cingulate sulcus on the medial surface of the hemisphere, was subdivided into the rostral part (pre-SMA) and the caudal part (SMA proper). The vertical anterior-commissural plane (Talairach and Tournoux, 1988) was used as a landmark to differentiate between pre-SMA and SMA proper (Deiber et al., 1991; Picard and Strick, 1996). The functionally determined FEF served as a landmark to differentiate between the dorsolateral BA6 (PMd) and ventrolateral BA6 (PMv); that is, the lateral part of BA6 superior to the FEF was regarded as PMd and the lateral part of BA6 inferior to the FEF as PMv. The representation of the hand in M1 (BA4) is located in the central sulcus (Yousry et al., 1997). Therefore, the convolution of the precentral gyrus corresponds mostly to PMdc at the level of the hand M1. The PMdr is likely located anterior to the superior precentral sulcus (SPcS) (Rizzolatti et al., 1998). In addition, the vertical anterior-commissural plane was used as a landmark to estimate the border between the PMdc and PMdr (Deiber et al., 1991).
In the present study, we used spatially normalized brain images; therefore, the description of the rostral–caudal and dorsal–ventral directions was based on the stereotaxic coordinate system of Talairach and Tournoux (Talairach and Tournoux, 1988). BA was determined for each activation cluster by converting the Montreal Neurological Institute coordinates (Evans et al., 1993) to Talairach and Tournoux coordinates, using a linear transformation matrix.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
PET Experiment
The primary purpose of the PET experiment was to examine the neural substrates underlying the mental-operation tasks under conditions minimizing the effects of motor preparation and execution. The electrophysiological monitoring (EOGs, surface EMGs and glossokinetic potentials) confirmed that none of the subjects moved during the PET scans except for natural eye blinks, which did not differ in frequency between the tasks [overall mean: 4.4 blinks per block, F(4,95) = 0.28, P = 0.89 by analysis of variance (ANOVA)]. The task performance was satisfactory on average; the accuracy was 80% for the numerical, 95% for the verbal and 100% for the spatial mental-operation tasks, and 85% for the verbal-rehearsal task.
Each of the three mental-operation tasks, when compared with the visual-fixation task, evoked a significant increase in rCBF in the pre-SMA and left PMdr, straddling or anterior to the vertical anterior-commissure (VAC) plane (Fig. 2B–D
, Table 1). Activity common to the three mental-operation tasks was also observed in the right cerebellar hemisphere and the left posterior parietal cortex (BA40 and 7) involving the intraparietal sulcus. Brain areas active during all three mental-operation tasks were the left frontoparietal cortical areas (PMdr, pre-SMA and posterior parietal cortex) and the right cerebellar hemisphere (overlap map; Fig. 2E).
|
The numerical and verbal mental-operation tasks evoked activity predominantly in the left frontoparietal cortex and right cerebellar hemisphere, while the spatial mental-operation task evoked bilateral and symmetrical activity in the frontoparietal cortex and cerebellum. In particular, during the numerical and verbal mental-operation tasks, activity was observed in the left frontal operculum–prefrontal cortex (BA6, 9, 44, 45). This frontal operculum–prefrontal activity, including the frontal language areas and probably part of PMv in its caudal part, was not observed during the spatial mental-operation task. In addition, the numerical and verbal mental-operation tasks primarily involved the pre-SMA, while the spatial mental-operation task primarily involved the PMdr bilaterally and the posterior superior parietal cortex (BA7). The basal ganglia were also active during the numerical and verbal mental-operation tasks. Activity was observed bilaterally in the basal ganglia, but activity on the left side did not reach statistical significance (Z = 2.11, where x = –28, y = –14, z = 6) during the numerical mental-operation task. These differences, especially the involvement of left frontal language areas in the numerical and verbal mental-operation tasks, supported the a priori assumption that verbal strategy would be used to a greater extent in the verbal and numerical mental-operation tasks than in the spatial mental-operation task.
There was a significant increase in rCBF in the left frontal operculum (BA6, 9) and bilateral cerebellar hemispheres during the verbal-rehearsal task compared with the visual-fixation task (Fig. 3A,B). The rCBF during the three mental-operation tasks and the verbal-rehearsal task (each relative to the visual-fixation task) was examined in the pre-SMA and PMdr (Fig. 3C). rCBF was markedly increased in the PMdr during the three mental-operation tasks but not during the verbal-rehearsal task (Z = 1.32, where x = –32, y = –2, z = 68, left side; Z = –0.94, where x = 20, y = 0, z = 56, right side). During the verbal-rehearsal task, a non-negligible trend toward activation was observed in the pre-SMA (Z = 2.79, where x = –4, y = 2, z = 68). Repeated-measures ANOVA showed that there was a significant task-effect bilaterally in the PMdr (P = 0.0003, left; P = 0.0001, right) but not in the pre-SMA (P = 0.2). The mental-operation tasks generally evoked a greater rCBF increase bilaterally in the PMdr than did the verbal-rehearsal task (P < 0.05), except that the numerical mental-operation task and the verbal-rehearsal task comparison did not reach statistical significance in the left PMdr (P = 0.083).
|
In summary, the three mental-operation tasks, supposedly involving different cognitive strategies, all induced a significant increase in rCBF in the pre-SMA and PMdr. Electrophysiological monitoring confirmed that the subjects did not move during the rCBF measurement. The verbal-rehearsal task involving sub-vocalization induced some activity in the pre-SMA but not the PMdr. These results indicate that performance of the mental-operation tasks (not overt movement or subvocalization) evoked activity in the PMdr.
fMRI Localization Experiment
The fMRI localization experiment was designed to gain more precise spatial information about activity evoked in the PMdr during the mental-operation tasks, by comparing it with activity evoked during the control skeletomotor and oculomotor tasks. The task performance was sufficient for both the numerical and spatial mental-operation tasks (mean accuracy, 93% in both tasks), and no tapping errors were detected visually for any subject in either the complex or simple finger-tapping tasks.
The lateral part of BA6 was operationally subdivided on a subject-by-subject basis, according to the findings from the control skeletomotor and oculomotor tasks (Fig. 4A,B). Compared with the visual-fixation task, the complex finger-tapping task activated widely distributed sensorimotor networks that typically included three activity loci at the level of the omega-shaped structure of the precentral gyrus (Yousry et al., 1997): PMdr in the depths of the SPcS, PMdc on the convexity of the precentral gyrus extending into the SPcS, and M1 on the omega-shaped structure of the precentral gyrus. PMdr activity was located in the depths of SPcS or in the caudal part of the superior frontal sulcus. These two sulci often connect to each other, forming a useful anatomical landmark (Ono et al., 1990; Erbil et al., 1998). The PMdr was dorsomedial to the functionally determined FEF (Fig. 4C). Comparison of activity evoked during the complex finger-tapping task with that during the simple finger-tapping task, showed that the PMdr (bilaterally) and posterior parietal cortex were the only cortical areas that showed greater activity for the complex finger-tapping task. PMdr activity was consistent across subjects as supported by the group analysis (Fig. 4D). The location of PMdr activity was consistent with activity associated with the complexity of finger-tapping tasks as described in previous reports using the same task (Sadato et al., 1996; Catalan et al., 1998). In contrast, PMdc and M1 activity did not differ between the complex and simple finger-tapping tasks, suggesting that these two areas were involved primarily in motor execution.
|
After the lateral part of BA6 was subdivided as above, the activated zones during the mental-operation tasks were investigated on a subject-by-subject basis in relation to those subdivisions. During the numerical and spatial mental-operation tasks, all subjects exhibited several loci of activity in the lateral BA6, among which the most conspicuous was consistent with the PMdr (Fig. 4B,D
). In seven of the 10 subjects, the activity in the PMdr evoked during the numerical and spatial mental-operation tasks and the complex finger-tapping task overlapped substantially. In three subjects, however, PMdr activity evoked during the spatial mental-operation task consisted of two small subzones: rostral activity evoked during the numerical mental-operation task and caudal activity evoked during the complex finger-tapping task. This intersubject variability may be due to the calculation strategy employed by each subject or to a variability of functional representation within this zone. Consistent with the results from the PET experiment, activity in the PMdr was observed predominantly in the left hemisphere during the numerical mental-operation task, but was bilaterally symmetrical during the spatial mental-operation task (Fig. 4B,D). Significant activity during each task at a group level is summarized in Tables 2–4.
|
|
|
The spatial relationship between the PMdr and the FEF is summarized in Figure 5
and Table 5. The location of peak activity (determined by Z-value) was identified on a subject-by-subject basis within the activation clusters in the SPcS. Activity in the right PMdr evoked during the numerical mental-operation task was excluded from this analysis because only a limited number of the subjects exhibited such activity. The PMdr was distinct from the FEF in both hemispheres [multivariate ANOVA; F (9,83) = 4.794, P < 0.0001, left; F (6,46) = 4.487, P < 0.0005, right). The PMdr was located medial and dorsal to the FEF in both hemispheres (P < 0.005), except that the difference in dorsal–ventral direction did not reach statistical significance in the left hemisphere (P = 0.19). On the other hand, activity in the PMdr during the numerical and spatial mental-operation tasks and the complex finger-tapping task was not distinguishable between tasks (P > 0.8 for all comparisons).
|
46 mm) is shown on the same axial slice, for convenience. See Table 5
|
In summary, the PMdr, as defined on a subject-by-subject basis by activity during the complex finger-tapping task, was located in the depths of the SPcS and dorsomedial to the FEF. This operationally defined PMdr corresponded to the area activated during the numerical and spatial mental-operation tasks. Furthermore, activity in the depths of the SPcS and caudal part of the superior frontal sulcus was consistent with PMdr activity observed via PET during the three mental-operation tasks.
Single-trial fMRI Experiment
Using a single-trial fMRI design, we investigated the temporal characteristics of brain activity associated with the delayed mental-addition task, which contained a crucial component of the numerical mental-operation task. Two patterns of transient BOLD signal changes were observed in the PMdr and pre-SMA (Fig. 6B). Figure 6C shows the peak activity in the PMdr and pre-SMA, which was determined on a subject-by-subject basis using Z-values (Fig. 6C). Stimulus-related activity was observed in the pre-SMA and bilaterally in the PMdr. S2-specific activity was observed in the pre-SMA and left PMdr. This left-lateralization of PMdr activity during the numerical mental-operation task was consistent with our results from the PET and fMRI localization experiments, further supporting a role of the PMdr in nonmotor mental operations.
|
Five subjects (four male, one female) participated in both the fMRI localization and single-trial fMRI experiments. We combined information from the two experiments to assess intersubject variability in activity in the PMdr and pre-SMA associated with nonmotor numerical operations. All five subjects exhibited activity in the pre-SMA and left PMdr, and three subjects also exhibited activity in the right PMdr during the numerical mental-operation task in the localization experiment. For these three areas, we explored S2-specific activity subject-by-subject in the single-trial fMRI experiment, by using a liberal statistical threshold (Z > 2.33). Four subjects exhibited moderate S2-specific activity in both the pre-SMA and left PMdr, and one subject exhibited S2-specific activity only in the right PMdr. Again, these results support the idea that S2-specific activity is a component of the activity observed in the pre-SMA and left PMdr during the numerical mental-operation task.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The present results show that rostral parts of non-primary motor cortex or BA6 in humans, presumably homologous to the pre-SMA and PMdr in monkeys, were active during mental-operation tasks that did not involve any immediate overt movement. This suggests that parts of this classical motor area have nonmotor cognitive functions in addition to the wellknown motor functions. This is similar to the idea that subzones of the basal ganglia and cerebellum, previously regarded as pure motor areas, have cognitive functions (Schmahmann, 1997
; Middleton and Strick, 2000). However, the present results indicate that the BA6 is not simply divided into cognitive and motor subzones, but that some regions of BA6 have both cognitive and motor functions. In this regard, the present conclusions are consistent with the idea that even M1, the cortical motor area most directly controlling movement, is important in the sensory aspects of motor control (Georgopoulos, 2000). The goal of the present study was to examine whether the sensory processing capabilities of motor-related areas function during behavioral processes not directing consequent movement. We speculate that the computational strategy for sensorimotor control may give rise to the strategy for sensory control of cognition. Methodological Limitations
There are methodological limitations to the present study. First, conclusions from neuroimaging studies critically depend on the selection of control tasks. It is possible that the BA6 activity occurred merely due to the complex pattern of visual inputs used in the mental-operation tasks because we used a low-level baseline (visual-fixation task) to minimize the effect of possible interactions between cognitive and motor components. The S2-specific activity in the single-trial fMRI experiment, however, argues against this possibility. Moreover, activity during mental-operation tasks was also observed in the PMdr in response to auditory stimuli (M. Honda, unpublished observation). Secondly, the present conclusion relies, in part, on the spatial resolution of the fMRI technique to discern activity evoked during cognitive versus motor tasks. In the present fMRI experiment, the spatial resolution was at the level of gyrus or sulcus anatomy. This resolution is sufficient to discuss the functional subdivision of BA6, where gyri or sulci have served as anatomical landmarks to subdivide it in monkey brain. Thirdly, activity in a structure during a task does not necessarily mean that the particular area is necessary to perform the task (Price et al., 1999). There are several patient reports, however, showing that lesions involving parts of BA6 impair arithmetic ability that cannot be attributed to motor disturbance or aphasia (Lucchelli and De Renzi, 1993; Tohgi et al., 1995). Thus, parts of BA6 possibly have roles in mental operations that include serial mental calculations. Future local inactivation studies as primates perform cognitive tasks may provide further characterization of the nonmotor functions of the PMdr and pre-SMA.
Neuroanatomical Substrates
The mental-operation tasks evoked activity in the medial part of BA6, which was defined as the pre-SMA. We previously showed in an epileptic patient that this area corresponds to part of the pre-SMA, rather than to the SMA-proper (Hanakawa et al., 2001). The present study also showed that activity in the lateral part of BA6, termed the PMdr, was consistent during the three mental-operation tasks. This area was located in the SPcS adjoining the superior frontal sulcus, and was dorsomedial to the functionally defined FEF. The PMdr, as characterized in the present study, is likely homologous to areas PMDr (Wise et al., 1997), F7 (Rizzolatti et al., 1998), 6aß (Vogt and Vogt, 1919) or 6DR (Barbas and Pandya, 1987) in nonhuman primates, according to the current views on the configuration of human motor cortex (Zilles et al., 1995; Preuss et al., 1996; Rizzolatti et al., 1998; Geyer et al., 2000). Likewise, PMdc, located on the convex portion of the precentral gyrus extending into the SPcS, probably corresponds, at least in part, to areas PMDc (Wise et al., 1997), F2 (Rizzolatti et al., 1998), 6a
(Vogt and Vogt, 1919) or 6DC (Barbas and Pandya, 1987). It has been suggested that there is functional segregation or gradients within BA6 in a rostrocaudal direction in primates (Geyer et al., 2000). The PMdc and SMA proper in the caudal BA6 have a closer relationship with M1 and have direct corticospinal projections (Murray and Coulter, 1981; Dum and Strick, 1991; He et al., 1993, 1995). These anatomical connections suggest that these areas are involved primarily in motor execution (Murray and Coulter, 1981; He et al., 1993). In contrast, the PMdr and pre-SMA in the rostral part of BA6 closely interconnect with the PFC rather than with M1 and the spinal cord (Barbas and Pandya, 1987; Luppino et al., 1993; Lu et al., 1994). Physiologically, sensory information processing during visuomotor tasks evokes more prominent neural activity in the PMdr than PMdc (Weinrich and Wise, 1982; Johnson et al., 1996; Shen and Alexander, 1997). Such rostrocaudal functional gradients are consistent with neuroimaging findings using various motor tasks in humans (Deiber et al., 1991, 1997; Kawashima et al., 1994; Grafton et al., 1998; Rijntjes et al., 1999; Toni et al., 1999). These studies suggest that the PMdr and pre-SMA are more involved with the sensory components of motor tasks, while the PMdc and SMA proper more closely relate to motor execution. These new lines of evidence suggest that the PMdr and pre-SMA have more important roles in cognitive-sensory control of behavior rather than in movement execution. There is another line of evidence, however, that the PMdr and pre-SMA also have motor functions: (i) electric stimulation of the PMdr evokes gross movement of proximal body parts in humans (Foerster, 1936) and monkeys (Preuss et al., 1996; Wu et al., 2000), and electric stimulation of part of the pre-SMA ceases ongoing movement (Luders et al., 1995); (ii) there are slow negative potentials preceding simple movements in part of the pre-SMA (Kunieda et al., 2000); and (iii) lesions involving the PMdr (lateral part of BA6 anterior to the SPcS) cause motor deficits in humans (Freund and Hummelsheim, 1985). Projections from the pre-SMA or PMdr to other motor-related areas (Barbas and Pandya, 1987; Luppino et al., 1993; Petrides and Pandya, 1999) and the brainstem reticular formation (Keizer and Kuypers, 1989) likely mediate these motor functions.
The posterior parietal cortex (BA7/40), particularly the intraparietal sulcus, was also active during all three mental-operation tasks. The posterior parietal cortex also consists of multiple subdivisions, each of which is involved in particular aspects of visual or somatosensory information processing. The posterior parietal cortex and BA6 are connected in a specific pattern, thus forming several frontoparietal circuits (Rizzolatti et al., 1998; Geyer et al., 2000). These two cortical areas function in concert during cognitive operations and motor control (Deiber et al., 1997). We also observed activity in the cerebellum and basal ganglia, which is consistent with the idea of cognitive functions in parts of these subcortical structures (Schmahmann, 1997; Middleton and Strick, 2000). The cerebellar activation was observed predominantly in the hemisphere contralateral to the frontoparietal activation, consistent with the crossed cerebrocerebellar interconnection (Middleton and Strick, 1994). The functional connectivity of the PMdr and pre-SMA with the parietal cortex and subcortical areas requires further exploration, but their close interaction appears likely.
Interpretations of Activity Evoked in the PMdr and pre-SMA during Mental Operations
The cognitive tasks previously shown to evoke activity in parts of BA6 involve mental arithmetic (Dehaene et al., 1996; de Jong et al., 1996; Rueckert et al., 1996), spatial and nonspatial working memory (Petrides, 1986; Jonides et al., 1993; Paulesu et al., 1993; Fiez et al., 1996; Owen et al., 1996; Courtney et al., 1998), attentional control (Hopfinger et al., 2000; Boussaoud, 2001), silent word production (Wise et al., 1991), conceptual reasoning (Rao et al., 1997) and mental imagery (Mellet et al., 1996; Ghaem et al., 1997; Richter et al., 2000). PMdr activity during these tasks has sometimes been attributed to the PFC or FEF rathe