INTRODUCTION

TOP ABSTRACT INTRODUCTION Methods RESULTS DISCUSSION REFERENCES

A number of methods can be employed to elucidate the relative contributions of different cortical areas in the execution of cognitive and motor tasks. Depending on the physiological phenomena that these methods measure, different levels of spatiotemporal resolution can be achieved. From the array of methods available, only a limited number can be applied to the study of physiological processes in an awake human volunteer.

Although methods like electroencephalography (EEG) and magnetoencephalography (MEG) are capable of measuring neuronal activation at a high temporal resolutions, they are subject to limitations in spatial resolution imposed by the unsolved inverse transform of electrical and magnetic flux propagation (see, for example, Cui et al. 1999). On the other hand, tomographic methods like positron-emission tomography and single photon computed emission tomography, which allow quantitative analysis of neuronal activation through the study of perfusion, metabolism, or ligand binding, are limited in their time scale to the range of tens of seconds for perfusion studies and minutes for ligand studies.

Functional magnetic resonance imaging (fMRI) (Ogawa et al. 1990), allows neuronal systems to be studied at high spatial and/or temporal resolutions (Kim et al. 1997; Menon et al. 1998). The signal measured in fMRI originates from activation-induced changes in local blood oxygenation that can be measured through MRI pulse sequences that are sensitive to changes in T2* relaxation time [blood oxygen level dependent (BOLD) contrast]. Through serial measurements of T2*-weighted images, information about the temporal properties of activation in different functional systems can be obtained. Echo planar imaging (EPI)the acquisition method most used for fMRIon experimental scanners operating at high field-strength is capable of acquiring T2*-weighted images at 30 ms per single slice, but even standard clinical equipment can generate images that are sensitive to oxygenation changes at a speed of 100 ms per slice or less.

The changes in blood oxygenation and blood flow underlying the BOLD effect are known to occur with a temporal delay relative to the onset of neuronal activation. As no absolute characterization of the nature of this hemodynamic response function is available for the human brain, only relative measurements between different areas of the brain can be performed. A constraint encountered with this approach is the fact that the assumption of an identical hemodynamic response function in all parts of the brain might not hold true. Therefore, the observation of temporal differences in BOLD signal changes between different areas in one task alone might not be sufficient to infer timing differences in actual neuronal activation. To overcome this limitation, it has been suggested that tasks be used that can be expected to exert a differential influence on activation in different structures of the neuronal network studied (Kim et al. 1997).

The cortical network that subserves the execution of voluntary movement consists of (among others) the primary motor area (M1), the lateral premotor cortex (PM), and the midline supplementary motor area (SMA). Neuroanatomical studies (Luppino et al. 1993), cortical recordings in primates (Halsband et al. 1994; Matsuzaka et al. 1992), and functional neuroimaging studies (Boecker et al. 1998; Humberstone et al. 1997) indicate functional segregation within the SMA, with a more rostral pre-SMA and a caudally situated SMA proper. From these studies, converging evidence exists for the participation of the pre-SMA in planning movement as well as in translating intent to move into an actual movement program. SMA proper, on the other hand, is more closely related to the actual execution of a movement program (for reviews see Tanji 1994 and Picard and Strick 1996).

The temporal sequence of activation in the different structures has been the subject of some discussion. However, attempts to map the neural generator by analyzing EEG or MEG data have yielded divergent results (see DISCUSSION).

Because fMRI has the capability to gather information at a time scale that is of interest in this context, it lends itself to the study of these temporal relationships. In the present work, we used single-slice EPI and simultaneous acquisition of behavioral data to study a well-established motor paradigm, namely the spontaneous execution of a prelearned motor sequence (Roland et al. 1980). To circumvent the limitations incurred through the possible variability of hemodynamic response in different parts of the brain, we also employed an auditorily triggered execution of the same sequence as a control task.

	Methods

TOP ABSTRACT INTRODUCTION Methods RESULTS DISCUSSION REFERENCES

Subjects

Eleven healthy subjects (9 female and 2 male) were recruited for the study through postings on the university premises. Because of an excessive motion artifact in two scan sessions and technical failure in another, only eight of the subjects could be included in the final analysis. All examinations were undertaken following a protocol approved by the ethics board of the faculty of medicine at the Technische Universität München. Written informed consent was obtained from all participants. All subjects were 18-45 years old (average 24.5 yr, SD 2.6 yr), right-handed, as determined by the Edinburgh (Oldfield 1971) handedness inventory, not on any medication, and free of a history of neurological disorders.

Paradigm

All experiments were performed by using a complex sequence of finger movements, as described by Roland et al. (1980), as an activation paradigm. The sequence comprises 16 flexions of digits 1-4 against the thumb (1-1-2-3-3-3-4-4-4-4-3-3-3-2-1-1). Subjects were provided with a visual representation of the sequence several days prior to the examination and were given ample opportunity to practice. In the baseline condition, subjects were instructed to keep thumb and index finger loosely opposed.

Data acquisition

Time-resolved fMRI was performed using a 1.5 T Philips Gyroscan ACS-NT scanner (Best, The Netherlands) equipped with the PT-6000 gradient set, which is capable of delivering a maximum gradient strength of 23 mT/m at a slew rate of 105 mT/m/ms. The built-in body coil was used for radio frequency (RF) excitation and a standard manufacturer-supplied quadrature-birdcage head coil with circular polarization was used for RF detection.

Auditory cues were delivered from a personal computer sound card through the communication system of the scanner via earplugs with a tubing connection (EAR-Link 3a, Cabot Safety, Indianapolis, IN).

Finger motion was recorded by a homemade contact glove connected to an electronic readout assembly. Behavioral data and a scanner synchronization signal were recorded at a sample rate of 1,000/s on a Multi I/O board using LabView data acquisition software (National Instruments, Austin, TX).

The functional imaging experiments were divided into two subsections: 1) a functional scout scan for localizing cortical motor areas and 2) multiple dedicated single-slice studies for temporally characterizing cortical responses in two different conditions.

For functional localization of cortical motor areas, a conventional block-design experiment with a total duration of 80 s was employed. Subjects were instructed to perform three epochs, 10 s each, of the complex finger movement sequence upon auditory cues. A 3-cm slab of axial T2*-weighted images positioned between the vertex and the corpus callosum was acquired with a 2D multi-slice EPI sequence [repetition time (TR) 1,000 ms, echo time (TE) 50 ms,  = 80°, matrix 128 × 128, field of view (FOV) 230 mm, thickness (th) 3 mm, 10 slices, 80 images].

Subsequently, an online analysis of the functional scout scan was performed by calculating a t-test of the images in an activated versus a nonactivated state by using the STIMULATE program (Strupp 1996). A confidence threshold of 0.99 with a minimum cluster size of 4 provided activation maps of sufficient quality to determine the areas of cortical activation.

In the second part of the examination, scans with high temporal resolution were performed in a single slice. The position for this slice was selected using the 3D activation map from the functional scout scan such that it contained the activation foci for the M1 and the SMA. For the purpose of this study, the SMA was defined as the area of activity on the mesial surface of the frontal cortex rostral to the central sulcus.

A single-shot blipped EPI sequence (TR/TE 100/50 ms,  = 60°, matrix 64 × 64, FOV = 230 mm, th = 6 mm) was used to acquire image data. Each single scan lasted 100 s with 1,000 images acquired. The first 100 images were discarded to allow for equilibrium in magnetization and RF saturation.

Two different conditions were employed for the time-resolved fMRI experiments. In the externally triggered (EXT) condition, subjects were instructed to perform a single execution of the complex finger sequence upon each of the auditory cues, which were delivered at intervals of 20-40 s throughout the scan. In the second condition, subjects were instructed to perform single executions of the same finger sequence at their own timing during the scan's duration (SELF). Through online analysis of the imaging data, the quality of the scans, with particular focus on the degree of task-correlated motion, was monitored permanently. A total of 15-25 of these scans was performed in an alternating pattern for each subject.

Data analysis

Offline data analysis was performed using the STIMULATE program and routines programmed in Interactive Data Language (IDL^TM) (RSI, Boulder, CO). Based on behavioral data from the contact glove, sections of 270 images covering the time period from 7 s before the onset of a complex finger sequence to 20 s after the onset were selected for further analysis. Sections displaying a motion artifact, defined as a center of mass shift in the x or y axis exceeding 10% of the image resolution (=0.35 mm), were discarded. Motion correction of the 2D data was not attempted because of its inability to correct for the predominant z component of motion that was previously observed in our laboratory during this type of motor experiment.

Raw data were scaled linearly to a common mean and were spatially filtered with a digital filter to reduce high-frequency noise. Shifting time course cross-correlation (ml) analysis with a set of model functions for signal increase (Richter et al. 1997b) was performed for each of the individual sections. The kernels' modeling slopes of different steepness were allowed to shift within a time window of 2,000 to +7,000 ms from the onset of finger motion as recorded by the glove assembly.

Regions of interest were delineated on a cross-correlation map that was calculated from a set of unsmoothed sum-images of all of the scan sections included in the analysis of a particular subject (ml coefficients 0.75-0.85, min cluster = 2). Based on the a priori hypothesis, regions of interest covering the precentral gyrus (M1) as well as the mesial frontal motor areas were defined. If two areas of activation were found on the mesial surface they were treated as being separate; if one contiguous area was present, it was divided into equal rostral and caudal sections and labeled rostral and caudal supplementary motor area (r-SMA and c-SMA) (see Fig. 1).

View larger version (148K): [in this window] [in a new window]   Fig. 1. Activation map generated through cross-correlation (correlation coefficient > 0.8) overlaid onto a 64 × 64 echo planar raw image (repetition time 100 ms, echo time 50 ms,  = 60°, slice thickness 5 mm). White, activated voxels; black, activated voxels within the regions of interest.

Signal time courses from volume elements in all sections belonging to one condition were averaged if they met the following criteria: 1) an ml-coefficient larger than 0.5 in the individual section and 2) spatial coordinates within the region of interest as defined on the unsmoothed images.

Time to half-maximum served as a temporal reference. To determine these values, the averaged time courses were zero-shifted, scaled to unit range, and filtered with a fast Fourier transformation-based digital filter (Gaussian kernel width equivalent to a frequency of 0.3 Hz). A first-order polynomial was fitted to the ascending slope of the signal time course between 1/4 and 3/4 of the peak value. The intersection of this line with half-maximum was used to define time offset in relation to movement onset for the different areas studied (see Fig. 2, A and B, for subject 8; Table 1).

View larger version (12K): [in this window] [in a new window]   Fig. 2. Signal time course for the three areas studied for subject 8. Dash-dotted line, rostral supplementary motor area (r-SMA); dashed line, caudal supplementary motor area (c-SMA); solid line, primary motor cortex (M1). Scaled to unit range, filtered with fast Fourier transformation (FFT)-based digital filter (Gaussian kernel, radius = 0.3 Hz). Horizontal marks on the ascending slope indicate measurement points. A: externally triggered (EXT) condition. B: self-initiated (SELF) condition.

For illustration purposes, signal time courses for all subjects were averaged and filtered with a digital filter (Gaussian kernel radius = 1 Hz; see Fig. 3, A and B).

View larger version (12K): [in this window] [in a new window]   Fig. 3. Signal time course for the three areas studied averaged over all 8 subjects. Dash-dotted line, r-SMA; dashed line, c-SMA; solid line, M1. Scaled to unit range, filtered with FFT-based digital filter (Gaussian kernel, radius = 1 Hz). A: EXT condition. B: SELF condition.

For the externally triggered condition, reaction times (RT) were extracted from the behavioral data. For all movement sequences, the average execution time (ET) and the percentage of incorrectly executed movement sequences were determined from the timing data.

	RESULTS

TOP ABSTRACT INTRODUCTION Methods RESULTS DISCUSSION REFERENCES

In Fig. 1, an activation map for the summation image as it was typically obtained is shown (subject 4). To preserve the impression of the spatial resolution obtainable with a 64 × 64 matrix, the activation map is overlaid onto a slice of averaged EPI images. Foci of activation could be delineated for the M1 and the SMA in all subjects. For six of the eight subjects, two distinct foci of activation could be discerned on the medial aspect of the hemisphere; in two subjects, one contiguous area was divided into a rostral and a caudal segment (r-SMA and c-SMA; see Fig. 1). In some of the subjects, a focus of activation lateral to the SMA and rostral to the M1, most likely representing the lateral premotor cortex, could be observed.

Of the 288 sections cut from the original data, 199 passed the 10% motion threshold and were retained for further analysis. The fraction of usable data for the individual subjects showed a large degree of intersubject variation. Data from two subjects had to be excluded from the analysis because not enough time course sections fulfilled the quality criteria. For each of the remaining subjects (n = 8), a minimum of five scan sections (average 14) was included in the generation of time courses for each of the conditions.

Times to half-maximum in relation to movement onset for M1 are in good agreement with values found in the literature for primary sensory and motor paradigms (Table 1, "abs" columns).

The degree of temporal shift between M1 and subregions of the SMA is of particular interest in our study. To obtain these differences, values for M1 were subtracted from the respective values for r-SMA and c-SMA (Table 1, "rel" columns). Averaged over all subjects, a mean difference of 0.7 versus 2.0 s can be seen for the rostral SMA in the EXT versus the SELF condition (significant at the P < 0.01 level, paired two-sided t-test). For the caudal SMA the shift is 0.5 s for EXT and 0.8 s for SELF (not significant).

Average signal increase was different for the three areas studied (c-SMA = 2.93%, r-SMA = 2.14%, M1 = 4.06% in the EXT condition; c-SMA = 2.64%, r-SMA = 1.86%, M1 = 4.09% in the SELF condition). The intra-individual differences for the two conditions proved to be nonsignificant (paired two-sided t-test, P < 0.1%).

RT in the externally cued execution had a mean of 563 ms and an SD of 121 ms (see Table 2) in the seven studies where reliable RT measurements could be obtained. No significant correlation between individual RTs and the timing information for the different areas studied could be found in the studies reported here.

The average time required to complete the task (ET) was 6,075 ms for the EXT condition and 6,442 ms for the SELF condition, a difference that proved to be significant at the P < 0.1% level (see Table 2). Although the percentage of sequences with errors was different for the two conditions studied (2.7% for EXT, 4.0% for SELF), this difference proved to be not significant.

	DISCUSSION

TOP ABSTRACT INTRODUCTION Methods RESULTS DISCUSSION REFERENCES

Through a combination of multi-slice scout imaging for localization and ultra-fast single-slice acquisition for temporal characterization, modern clinical MRI equipment allows a resolution down to the subsecond range for temporal relationships in cortical networks. In a finger tapping task, we observed a temporal shift in the BOLD response between the SMA and the M1 of 0.7 s and 2 s in the externally cued and the self-initiated conditions, respectively.

A problem encountered in time-resolved fMRI is the question as to whether differences in signal time-course are truly an expression of underlying differences in the timing of neuronal activity. Observed differences between cortical areas could be explained by synchronous neuronal activity dispersed by differences in the hemodynamic response. Given the data presented in this work, this might mean that the shift of 0.7 s between the M1 and the rostral SMA in the externally cued control condition could at least partially be explained by this phenomenon. Even if this holds true, the main finding is the longer lead time of 2.0 s for the rostral SMA in the self-paced condition. If, hypothetically, the entire difference in the control condition was caused by hemodynamic factors, this would leave a minimum of 1.3 s difference for the underlying neuronal activation in the internally controlled condition. This is an indication for earlier onset of processing in the SMA compared with the motor executive cortex.

The observed timing difference between the rostral part of the supplementary motor area and the primary sensorimotor cortex for externally triggered versus internally triggered movements is of interest because the higher spatial resolution of fMRI extends previous data obtained from EEG and MEG. Although caution is warranted when inferring from a method that observes activation-induced changes in cortical perfusion and oxygenation to methods examining electrical brain activity, our results indicate that activity in the rostral SMA precedes activity in the M1.

The difference in time shift of 1.3 s for externally and internally triggered movement is in good agreement with the early components of the readiness potential (Bereitschaftspotential, BP), a negative direct current (DC) variation observable in EEG prior to the onset of spontaneous movement (Kornhuber and Deeke 1965). This DC shift is present bilaterally in the EEG leads with a maximum amplitude at the vertex. It can be divided into an earlier component (BP1), which is present between 2.2 and 1.8 s before movement onset, and a later component (BP2), which is present between 1.0 and 0.4 s before movement onset. BP1 has a maximum in the midline leads whereas BP2 has a maximum over the contralateral hemisphere (Shibasaki et al. 1980).

The underlying neuronal generators of the readiness potential have been the subject of considerable discussion in the clinical neurophysiology literature. Studies performed with different methods and processing strategies have yielded diverging results. Originally, the supplementary motor area was discussed as the primary source of the early component (Deecke and Kornhuber 1978), a view that was challenged when new methods of analyzing EEG data evolved. Two groups using dipole source analysis acquired results that were in contradiction to a dominant role of the SMA in BP generation (Böcker et al. 1994; Bötzel et al. 1993), whereas others, using the same method with different initial settings for the algorithm, localized the generator of the readiness potential in the SMA (Knösche et al. 1996; Praamstra et al. 1996). Recent work using multichannel EEG found generators for the early components in rostral midline structures (Cui et al. 1999) that become active at different points in time (Ball et al. 1999). Furthermore, subdural recordings in patients undergoing epilepsy surgery located BP-like activity in the SMA, the pre-SMA, and the M1 (Ikeda et al. 1992, 1999; Neshige et al. 1988).

A number of fMRI studies examined CNS activation in movement control on a time scale shorter than the hemodynamic response function (time-resolved fMRI). Two of these studies used a delayed cued motor paradigm to separate different stages of motor control (Lee et al. 1999; Richter et al. 1997a). Here, a first stimulus indicates a certain property of the movement to be made; a second stimulus is given to initiate the execution of the movement. Both studies managed to demonstrate a temporal sequence in the activation of the SMA and the M1. One of the studies indicated temporal segregation within the supplementary motor area itself (Lee et al. 1999), with earlier activation of the more rostral part and activation of the caudal part simultaneously with the M1. In the study of Richter et al. (1997a), which was conducted using a 4T experimental scanner, no temporal shift for the SMA and the M1 was evident in a control task involving immediate execution of a complex sequence of finger movements. The major differences between the paradigm used in both of these studies and our design is the completely external cueing and the focus on the planning component of such a task.

Another study using an experimental high-field scanner (Kansaku et al. 1998) used a paradigm involving visually cued sequential finger to thumb opposition to demonstrate a delay of 0.47 s for M1 and SMA signal changes. In many respects, the paradigm used in Kansaku et al. (1998) is comparable with the EXT (control) condition in our experiment. Therefore, the agreement between the delay reported by Kansaku et al. (1998) for an aggregate SMA (0.47 s) and the data for caudal SMA (0.5 s) and rostral SMA (0.7 s) in the EXT condition of our study is noteworthy.

To our knowledge, the only other study directly comparable with ours in its use of a self-initiated movement (Wildgruber et al. 1997) demonstrated a delay of 0.8 s for the SMA and the M1. A self-paced repetitive button press served as the activating paradigm; a control condition (to account for the aforementioned hemodynamic factors) was not employed. The temporal shift between the SMA and the M1 reported in Wildgruber et al. (1998) is considerably less than the time delay we report for the rostral SMA (2.0 s), but is in good agreement with our data for the caudal SMA (0.8 s) in the SELF condition.

In conclusion, this is the first fMRI study that we are aware of that demonstrates the sequential activation of motor areas in self-initiated complex movement. This was achieved through a combination of different imaging techniques and a paradigm that exerts a differential influence on activation in the structures studied. The different timing behavior within the SMA further supports the functional and anatomical segregation within this cortical area in humans into a rostral pre-SMA and a caudal SMA proper.