gyrus in monkey. Brain 103:525-554. Bergstr6m M, Boethius J, Eriksson L, Greitz T, Wid6n L (1981). Head fixation device for reproducible position alignement in.
Exp Brain Res (1994) 98:523-534
9 Springer-Verlag 1994
O R I G I N A L PAPER
Gottfried Schlaug 9 Uwe Knorr 9 Riidiger J. Seitz
Inter-subject variability of cerebral activations in acquiring a motor skill: a study with positron emission tomography
Received: 21 February 1993 /Accepted: 29 November 1993
Abstract Cerebral structures activated during sequential right-hand finger movements were mapped with regional cerebral blood flow (rCBF) measurements by positron emission tomography (PET) in individual subjects. Nine healthy volunteers were examined twice; after initial learning and after practicing the finger movement sequence for more than 1 h. Task-specific activation sites were identified by statistical distributions of maximal activity and region size in rCBF subtraction images. A consistent task-specific activation in all nine subjects was detected in the contralateral sensorimotor cortex at an average movement rate of 3.2 Hz reached after practice. This corresponded to a significant increase of the mean rCBF in the left primary sensorimotor cortex in spatially standardised and averaged PET images. Additional task-specific activation sites detected by individual analysis were found in the lateral and medial premotor, parietal, and cingulate areas, and in subcortical structures including the basal ganglia of both cerebral hemispheres. These activations showed no or little spatial overlap from subject to subject, thus being obscured in the analysis of pooled data. The observed activity patterns were related to movement rate and accuracy in individual subjects. It is suggested that the rCBF changes associated with acquisition of a motor skill in individual humans may correspond to plasticity of sensorimotor representations reported in monkeys. Key words Positron emission tomography Regional cerebral blood flow 9 Motor learning Brain mapping 9 Human
G. Schlaug ([~) 9 U. Knorr 9 R. J. Seitz Department of Neurology, Heinrich-Heine-University Diisseldorf, P.O. Box 101007, D-40001 Dfisseldorf, Germany, FAX no : + 49(0)211-342229
Introduction Humans can acquire a wide range of motor skills and perform them almost automatically. The ability to store and retrieve motor skills has been termed procedural memory and has been differentiated from episodic or semantic memory (Squire 1986). Procedural memory operates independently of the hippocampal system, being fundamentally dedicated and inflexible (Eichenbaum et al. 1992). This means that its representations are accessible only to the processing modules that are engaged during learning and retrieving (Eichenbaum et al. 1992). Thus, motor skills become evident in procedures in which their knowledge has been embedded during acquisition. The acquisition of a motor skill can proceed in two ways: in a repetitive manner which develops slowly by continuous practice and in an abrupt manner which initiates the assured use of predictive movements (Brooks 1990). A motor skill is acquired when movements can be performed with both high accuracy and high velocity (Kihlstrom 1987). The capacity to acquire a motor skill appears to represent an individual characteristic, since learning curves of movement activity have been shown to vary considerably amongst subjects (Brooks et al. 1992). Some subjects predominantly increase the frequency of their movements, while others reduce erroneous movements or do not succeed in learning the task at all desspite practicing. Group analysis of movement activity across different subjects may therefore obscure individual learning effects. A similar situation can be expected for cerebral activation patterns measured with positron emission tomography (PET) during the acquisition of a motor skill. Regional cerebral metabolic activity and regional cerebral blood flow (rCBF) have been shown to be related to synaptic activity within populations of neurons (Raichle 1987). Currently, methods for PET image analysis are directed towards the establishment of group mean rCBF changes using pixel-by-pixel t-statistics in spatially standardised PET images (Fox et al. 1985,
524
1988; Evans et al. 1988; Seitz et al. 1990a; Friston et al. 1991a). This approach is biased towards regions showing rCBF changes of similar magnitude among subjects and towards regions with large spatial overlap in different subjects. Inter-subject pixel-by-pixel analysis of spatially standardized PET images disregards the individual variability of human brain anatomy at the macroscopic and microscopic levels (Filimonoff 1932; Stensaas et al. 1974; Eidelberg and Galaburda 1984; Steinmetz and Seitz 1991; Grafton et al. 1991; Nudo et al. 1992; Walter et al. 1992). In addition, local heterogeneities of activation patterns - possibly reflecting the use of different functional loops between different subjects will not be detected. A remarkable inter-individual variability in the activation pattern was recently demonstrated by using statistical parametric mapping in repetitively performed, visual stimulation paradigms (Watson et al. 1992). This observation is of considerable importance for our study, since visual stimulation was identical for all subjects, in contrast to the individual task performance in this study. Thus, group analysis of PET images may obscure rCBF changes occurring in individual subjects in relation to different movement rates, differences in eliminating erroneous movements, and possibly in relation to different task-solving strategies. Recently, a new method for PET image analysis has been developed which allows the identification of taskspecific rCBF changes in activation studies on individual subjects (Knorr et al. 1993). In this method, task specific rCBF changes are identified by means of their maximal activity and spatial extent, which separates them statistically from background noise in rCBF subtraction images. In a subsequent step, the rCBF changes are superimposed on anatomical structures after spatial alignment with the corresponding M R images of the same individuals, as described by Steinmetz et al. (1992). To obtain insight into the acquisition of a finger movement sequence in individual subjects, we used this new approach to re-evaluate a PET study on a model of human motor learning (Seitz and Roland 1992a). This enabled us to relate the individual activation patterns with the individual task performance, as assessed by video monitoring and EMG recordings. Some of the results have been presented in abstract form (Schlaug et al. 1992a,b).
Materials and methods Subjects and task The PET images of nine healthy right-handed male volunteers (age range 19-29 years) without any history of neurological, psychiatric or medical disorders were used for evaluation. As reported in detail in the preceeding paper (Seitz and Roland 1992a), handedness was assessed by the Edinburgh questionnaire, and informed consent was obtained in accordance with guidelines approved by the ethics committee and the radiation safety committee of the Karolinska Institute and the Declaration of Human Rights, Helsinki 1975.
The blindfolded subjects had to perform a complicated finger movement sequence with their right hand. They had to touch the thumb once with the middle finger, twice with the index finger, three times with the ring finger, four times with the little finger, three times with the ring finger, once with the middle finger, and twice with the index finger. The first rCBF measurement was taken after careful instruction (initial phase). After 1 h of training the rCBF was measured again (performance phase). A rCBF measurement at rest served as a control condition. This was randomly either the first or the last PET measurement. The task performance was monitored by video taping with a frame size of 40 ms. In addition, electromyographic activity was recorded with surface electrodes from the musculus extensor digitorum communis, the musculus flexor digitorum communis, the thenar and the hypothenar of the right forelimb. PET scanning The rCBF was measured with a bolus inhalation of [l'C]fiuoromethane, as described by Roland et al. (1987). The PC384-TB PET camera (Scanditronix) had a spatial resolution of 7.6 mm, and a slice thickness for direct slices of 11.6 mm and for cross slices of 8.0 mm in the centre of the field of view (Litton et al. 1984). The global CBF of the different PET scans calculated from manually drawn regions of interest covering the whole slice of each of the seven PET image slices was adjusted to the PaCO2 at rest according to the methods of Olesen (1974). Only absolute values of rCBF in ml/100 g/min were analysed (Seitz and Roland 1992b). The changes of the rCBF were calculated using pixel-bypixel subtractions, so that each subject served as his own control. In each subject three subtraction images were calculated: one showing the rCBF differences compared to rest in the initial phase, the same in the performance phase, and one showing the rCBF differences between the initial and performance phase. In addition, 16 transaxial magnetic resonance (MR) images were recorded in each subject with a 1.5 T Siemens Magnetom in the inversion-recovery mode. Head movements of the subjects were prevented in the PET and MR scanners by the individually moulded fixation helmet (Bergstr6m et al. 1981). Using the computerised brain atlas program (CBA), the PET and MR images were spatially standardised according to transformation parameters individually determined in MR images of each subject (Seitz et al. 1990a). This procedure produced 14 image slices of each data set with identical spatial orientation. Significant mean rCBF changes as determined by t-map analysis in these spatially standardised PET images were compared directly to significant mean rCBF changes as determined by our new method for image analysis described in detail below.
Determination of task-specific rCBF changes The mean and the individual rCBF subtraction images were analysed by the individual response identification statistics (IRIS) developed by Knorr et al. (1993). In short, significant rCBF changes are estimated from quantitative (maximal activity) and spatial (size) information on regions of interest (ROIs) present in rCBF subtraction images. ROIs were defined by a 30% isocontour of the maximal pixel value in the 14 subtraction images and gamma distributions fitted to their cumulated distribution functions. Then, a null hypothesis - that the regions in rCBF subtraction images are noise - was formulated. At a fixed probability of error (alpha = 0.01) a critical value was calculated from the gamma distribution of each parameter (e.g. maximal activity and size). Taking the dimension of the PET image matrix and the resolution characteristics of the PET camera, this error of probability corresponded to P < 0.004, ensuring that not a single ROI was accepted in the set of 14 image slices by chance. The null hypothesis was rejected if the values of size and maximal activity of a given ROI were above the calculated critical values. These ROIs were considered to represent task-specific signals. Our method enabled us to
525 Table 1 Task performance as assessed by video and EMG monitoring (I initial phase, P performance phase)
Case
Phase
Rate (Hz)
1
I P I P I P I P I P I P I P I P I P
1.9 3.5 1.2 2.7 2.5 2.7 1.6 3.6 1.9 4.0 1.6 31 1.4 2.9 1.7 3.1 2.0 3.3
I
1.7
2 3 4 5 6 7 8 9 Mean
Increase in rate (%) 84 125 8 125 111 94 107 82 65
Finger switch latency
Faults per sequence
184 28 286 58 352 418 481 150 158 7 167 61 459 151 474 172 250 82
0.7 0.3 5.3 0.5 3.3 1.1 2.9 3.7 0.3 4.5 2.9 2.9 4.9 2.4 2.7 1.9 0.8 0.7
305
2.6
Faults change (%) - 57 - 91 -67 28 1400 2 - 51 - 30 - 13
Bursts per move
Bursts duration (ms)
3.1 3.8 1.4 1.6 3.4 3.2 3.3 2.5 2.6 2.4 2.1 2.7 5.3 3.4 2.8 3 3.1 2.2
276 149 872 403 265 217 388 373 750 198 253 186 268 134 190 133 278 157
3.0
393
Bursts duration change (%) - 46 - 54 -18 -4 - 74 -27 - 50 - 30 -44
SD
0.4
126
1.8
1.1
244
P
3.2
89
114
2.0
125
2.8
217
-- 39
SD
0.4
37
96
1.5
535
0.7
102
21
determine increases and decreases in the subtraction images separately. Subtractions of the first PET scan (initial phase) from the second (performance phase) resulted in cerebral areas showing significant positive (augmentations) as well as negative (reductions) changes during the learning process. Localisation of task-induced rCBF changes For accurate anatomical localisation, the significant rCBF changes were superimposed on corresponding high-resolution MR images after spatial alignment, using the computerised brain atlas program (Seitz et al. 1990a). The anatomical identification of activation sites was achieved by an overlay of the brain structures from the data base of the computerised brain atlas (Greitz et al. 1991).
Results G r o u p m e a n images Task-specific m e a n r C B F increases as d e m o n s t r a t e d by I R I S occurred during the initial phase and after l h of practicing (performance phase) in the left s e n s o r i m o t o r h a n d area and the left p r e m o t o r cortex (Fig. 1). These regions increased in size by 2.4% from the initial phase to the p e r f o r m a n c e phase, while the m e a n r C B F was a u g m e n t e d by 8.1% in the p e r f o r m a n c e phase as comp a r e d to the initial phase (Table 2). These changes reflected the higher finger m o v e m e n t rate of 3.2 H z ( • 0.4) in the p e r f o r m a n c e phase c o m p a r e d to 1.7 H z ( • 0.4) in the initial phase (Table 1). The subtraction of the initial scan from the perform a n c e scan revealed cerebral structures showing a significant r C B F change during the learning process. Posi-
tive changes during the learning process - termed augm e n t a t i o n s - were detected in the left p u t a m e n / g l o b u s pallidus and the left h a n d area. Negative changes of the r C B F during the learning process - termed reductions were found in the right superior a n d anterior parietal areas and in the right Broca homologue. These augmentations and reductions of the m e a n r C B F during the learning process were identical with those described by Seitz and R o l a n d (1992a).
Individual r C B F patterns T h e contralateral p r i m a r y s e n s o r i m o t o r area and sites in the lateral p r e m o t o r cortex were activated in the performance phase in all subjects (Table 2). Contrastingly, in the initial phase, only three out of nine subjects showed a small area of significant r C B F changes in the p r i m a r y s e n s o r i m o t o r cortex (Figs. 4, 5). Other areas representing significant r C B F changes varied a m o n g subjects showing an individual pattern of task-specific increases and decreases (Appendix). The inter-subject variability of the task-specific activations became obvious by superimposing the activation areas of the different subjects. This is d e m o n s t r a t e d in a pixel-by-pixel display in Figs. 2 and 3. By this a p p r o a c h , it was proven that the area of m a x i m a l spatial overlap of the taskspecific activation areas was the left s e n s o r i m o t o r cortex. In addition, spatial overlap of activated areas also occurred in p e r f o r m a n c e across some subjects in the mesial frontal cortex and the right anterior cerebellum, whereas the other areas had no or little spatial overlap a m o n g the subjects (Fig. 2). Even in the s e n s o r i m o t o r
526 Table
2 Group mean and individual rCBF changes of primary sensorimotor cortex
Group a
Phase
Size (pixel)
Volume (mm 3)
Mean change (ml/100 g/min)
(Activition- rest) x 100/ rest (%)
Maximal change (ml/100 g/rain)
I P
42 43
1829 1873
17.6 22.4
29.3 37.4
31.1 41.4
I P I P I P I P I P I P I P I P I P
54 91 37 210 25 27 131 137 28 49 29 43 18 14 38 36 28 14
2352 3963 1611 9154 1089 1176 5705 5966 1219 2134 1263 1873 784 610 1655 1568 1219 610
40.7 40.7 29.4 29.7 34.0 43.1 32.9 34.6 28.0 37.8 34.8 39.8 47.0 46.0 34.7 44.8 35.6 35.1
65.8 63.7 69.8 93.8 64.0 43.1 56.8 59.4 84.3 79.3 64.7 64.8 62.4 62.6 67.0 97.1 61.5 57.2
81.8 72.5 39.8 65.1 52.6 86.9 56.8 70.9 48.4 73.8 68.0 67.4 79.7 77.4 64.6 83.5 56.2 58.4
** 14.1 72.9 ** 8.3
Case
1 2 3 4 5 6 7 8 9 Mean
I
4 3 **
1 8 7 7 **
35.2 **
66.2 **
SD P SD
35 69 ** 66
1502 3006 ** 2882
5.7 39.1 ** 5.3
7.7 74.2 ** 15.3
* * * * *
* *
61.1
a Data taken from Seitz and Roland (1992 a) * Not significant according to IRIS ** No significant differences between initial phase (I) and performance phase (P), (two-tailed t-test, P > 0.05) c o r t e x the s p a t i a l o v e r l a p of t h e specifically a c t i v a t e d a r e a s was s u c h t h a t t h e a r e a s o f e i g h t o f n i n e s u b j e c t s h a d o n l y o n e pixel in c o m m o n (Fig. 3). It s h o u l d be p o i n t e d o u t t h a t this c o m p a r i s o n of t h e i n d i v i d u a l s u b jects was performed on spatially standardised PET images. T h e i n d i v i d u a l t o p o g r a p h y of the r C B F i n c r e a s e s inc l u d e d p r e d o m i n a n t l y t h e m e s i a l f r o n t a l c o r t e x (three
cases) p r o b a b l y c o r r e s p o n d i n g to the p r e - S M A a n d S M A p r o p e r ( L u p p i n o et al. 1990, 1991), a n d the s u p e r i o r p a r i e t a l c o r t e x (three cases), w i t h a m o r e a n t e r i o r ( a r e a 5) a n d a m o r e p o s t e r i o r p a r t ( a r e a 7). A l s o , in s o m e cases t h e r e w a s a r C B F i n c r e a s e in B r o c a ' s a r e a o r in t h e r i g h t B r o c a ' s h o m o l o g u e . I n t e r e s t i n g l y , different a c t i v a t i o n a r e a s c o u l d be d i f f e r e n t i a t e d in t h e c i n g u l a t e cortex, p r o b a b l y c o r r e s p o n d i n g to c i n g u l a t e m o t o r a r e a s
Figs. 1-3 Significant mean rCBF increases induced by performing a right-hand finger movement sequence are superimposed on corresponding mean high-resolution MR images after spatial standardisation and alignment using the computerised brain atlas program (Seitz et al. 1990). According to neuroradiological convention, right in the images corresponds to the left side of the brain. Fig. 1 The mean rCBF increase in the primary sensorimotor ( x = - 2 7 , y = - 1 7 , z=48") and premotor cortex ( x = - 2 2 , y = - 5 , z=50*) compared to rest after 1 h of practicing the sequential finger movement sequence. Fig. 2 The spatial variability of the activation areas among subjects in the performance phase. It is evident that the highest degree of spatial overlap occurred in the left primary sensorimotor cortex followed by a far less pronounced spatial overlap of the activation areas in the frontomesial cortex. Spatial overlap of the activated regions in the right premotor cortex occurred in two subjects. Fig. 3 High-power magnification: spatial overlap in the specifically activated left sensorimotor cortex was only present in one pixel among eight subjects
PRM: x = - 2 3 , y = 6 , z = 50*). Fig. 6 A large activation in the performance phase of case 4 involving the primary sensorimotor cortex ( x = - 2 8 , y = - 1 7 , z=48"), the anterior parietal cortex ( x = - 3 0 , y = - 3 0 , z=48") and the premotor cortex (L PRM: x = - 2 3 , y = - 5 , z=48"; R PRM: x=28, y = - l , z=51*) on both sides. Fig. 7 Significant rCBF increases in the left cingulate cortex of case 2, probably corresponding to anatomical sites of cingulate motor areas (CMAr: x = - 5 , y = 3 , z=43"; CMAc: x = - 5 , y = - 1 1 , z=43"). Figs. 8, 9 Significantly activated areas in the performance phase of case 3 probably correspond to anatomical sites of the SMA proper (x = - 3, y = - 1, z-- 46*) and the pre-SMA (x = - 3 , y = 5, z = 46*) (8), and the CMAr (x = --7, y = 1, z = 42*) (9) in the left hemisphere. The anatomical structures in the difference images have been retrieved from the database of a computerised brain atlas (Greitz et al. 1991). Besides the mesial brain surface (yellow), the cingulate sulcus and precentral sulcus (red) are displayed in 8, while the outer contours of the cingulate gyrus are displayed in 9. Augmentations (left premotor cortex (x = 31, y = 30, z = 9*)) and reductions (right homologue of Broca's area (x = 31, y = 30, z = 9*)) of the rCBF during the learning process of case 2 in Figs. 10, 11. Augmentation of the rCBF in the left putamen/globus pallidus ( x = - 2 3 , y=4, z=10*) in case 4 is shown in Fig. 12 * coordinates [mm] in the stereotactic space of Talairach and Tournoux (1988)
Figs. 4-12 Integrated individual PET/MR images showing rCBF changes while acquiring a motor skill. Fig. 4 Activated region in the primary sensorimotor cortex ( x = - 2 7 , y = - 4 , z=41*) and premotor cortex ( x = - 2 6 , y = 6 , z=43") in the initial phase of case 1 (not significant according to IRIS). Fig. 5 Significant rCBF increases in the performance phase of case 1 showing an in-plane increase of the activated region (prSM: x = --27, y = --5, z =48*;
527
528
(CMAr and CMAc), the anterior cingulate area 24b and the posterior cingulate areas 23b and 31 (four cases). Even ipsilateral and contralateral activations of subcortical structures such as the basal ganglia (three cases), thalamus (three cases), and the lateral cerebellum (2 cases) occurred in individual subjects. As seen from Figs. 5 and 6, some of the significant rCBF changes extended over adjacent cytoarchitectonic areas. However, peaks of the rCBF changes (maximal activation sites) could clearly be visualised within regions of significant rCBF changes corresponding to different but closely adjacent anatomical locations (i.e. primary sensorimotor and premotor cortex, anterior parietal cortex and SMA).
istics, by substantially increasing the finger movement rate and decreasing the faults per sequence. One subject (case 9) showed neither a significant rCBF change in the performance phase nor a significant augmentation of the rCBF from the initial phase to the performance phase. It should be emphasised that no mirror movements were detected by two independent observers in any of the subjects with bilateral activations of the cerebral cortex or subcortical structures. Further, no vocalisation of probable internal counting was detected in subjects showing activation of Broca's area.
Discussion Individual relation of task performance and rCBF changes For analysis of the individual activations, it was examined whether subjects showed different degrees of improvement in the task performance, either as an increase of movement rate or as a decrease in the number of erroneous movements, or as a combination of both (Table 1). It becomes apparent that the majority of subjects increased the finger movement rate dramatically. The increase in finger movement rate of all subjects was inversely related to the latency between the individual finger movements (finger switch latency). This latency varied considerably among subjects. Even more heterogeneous were the subjects in their ability to diminish erroneous finger movements. Comparing the EMG data shown in Table 1 with the rCBF data in the primary sensorimotor cortex (Table 2), it turned out that, in spite of the prominent reduction of the mean burst duration recorded from the right forearm muscles from the initial to the performance phase, the mean rCBF tended to rise. Also, the mean number of bursts per movement were fewer in the performance phase than in the initial phase. Using the non-parametric Spearman-rank test, it turned out that in the performance phase the rCBF in the left sensorimotor cortex correlated inversely with the E M G burst duration (r=~0.77, P
