Motor Learning of Compatible and Incompatible Visuomotor Maps

Scott T. Grafton1, Joanna Salidis2, and Daniel B. Willingham2 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021

Abstract

& Brain imaging studies demonstrate increasing activity in occurring with respect to target location. Positron emission limb motor areas during early motor skill learning, consistent tomography (PET) functional imaging of compatible learning with functional reorganization occurring at the motor output identified increasing activity throughout the precentral gyrus, level. Nevertheless, behavioral studies reveal that visually maximal in the arm area. Incompatible learning also led to guided skills can also be learned with respect to target location increasing activity in the precentral gyrus, maximal in the or possibly eye movements. The current experiments exam- putative frontal eye fields. When the incompatible task was ined motor learning under compatible and incompatible switched to a compatible response and the previously learned perceptual/motor conditions to identify brain areas involved sequence was reintroduced, there was an increase in arm in different perceptual±motor transformations. Subjects motor cortex. The results show that learning-related increases tracked a continuously moving target with a joystick-controlled of brain activity are dynamic, with recruitment of multiple cursor. The target moved in a repeating sequence embedded motor output areas, contingent on task demands. Visually within random movements to block sequence awareness. guided motor sequences can be linked to either oculomotor or Psychophysical studies of behavioral transfer from incompa- arm motor areas. Rather than identifying changes of motor tible (joystick and cursor moving in opposite directions) to output maps, the data from imaging experiments may better compatible tracking established that incompatible learning was reflect modulation of inputs to multiple motor areas. &

INTRODUCTION mentary motor area. With additional practice on a Humans can acquire new skills in a matter of minutes to second day there is evidence for a further increase of hours. With sufficient practice, motor skills such as activity in contralateral putamen (Grafton, Woods, & typing or sports are retained lifelong. This ease is Tyszka, 1994). This basic observation of enhanced activ- remarkable given the complexity involved in the timing ity in the motor circuit with practice has been observed and generation of new muscle synergies. A persistent with many other procedural and implicit sequence question for functional mapping is where a new motor learning tasks when studied in the initial period of skill skill is represented in the brain. Converging evidence learning (Hazeltine, Grafton, & Ivry, 1997; Jueptner et al., from invasive and noninvasive brain mapping techniques 1997; Shadmehr & Holcomb, 1997; Doyon, Owen, Pet- have begun to identify a set of sensory and motor areas rides, Sziklas, & Evans, 1996; Flament, Ellermann, Kim, where learning-related changes might occur. Ugurbil,&Ebner,1996;Karnietal.,1995,1998;Rauchetal., Previous imaging studies of procedural motor learn- 1995; Jenkins, Brooks, Nixon, Frackowiak, & Passing- ing from our laboratory examined the pursuit rotor task ham, 1994; Schlaug, Knorr, & Seitz, 1994; Seitz et al., (Grafton et al., 1992). This is a prototypical procedural 1994; Grafton et al., 1992, 1994; Grafton, Hazeltine, & task in which subjects hold a stylus and chase a small Ivry, 1995; Haier et al., 1992; Roland, Gulyas, & Seitz, target positioned on the outer edge of a rapidly rotating 1991; Lang et al., 1988; Saint-Cyr, Taylor, & Lang, 1988). disc (Ammons, 1947). Over the course of 20±40 min the Complementary methods, such as transcranial magnetic time on target, a measure of accuracy and learning, stimulation, have identified analogous learning-related increases dramatically. The skill can be retained over increases of the relevant digit representations in human many years (Ammons et al., 1958). As subjects learn this motor cortex during implicit sequence learning (Clas- skill, imaging studies identify progressive increases of sen, Liepert, Wise, Hallett, & Cohen, 1998; Pascual- activity in contralateral motor cortex and the supple- Leone, Grafman, & Hallett, 1994). Interpretation of motor learning-related changes of brain activity requires a thorough identification of what 1 Emory University School of Medicine, 2 University of Virginia skill or knowledge is actually acquired by a subject

D 2001 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 13:2, pp. 217±231

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 (Willingham, 1998, 1999). For procedural tasks such as RESULTS the pursuit rotor task, subjects could be learning any Experiment 1a: Psychophysics of Spatially combination of the following factors: (1) spatial location Compatible Procedural Learning of the target (2); eye movements; (3) specific sequential arm movements; and (4) an internal model of abstract This experiment validated a procedure that leads to motor goals such as limb direction or gestures. There is implicit motor skill learning under spatially compatible behavioral evidence that in the pursuit rotor task more conditions with respect to limb, target, and cursor than one of these factors may contribute to performance movement directions. The task, a modification of the improvements (Adams, 1987). For example, eye tracking pursuit tracking task originally developed by Pew (1974), is more kinematically constrained than other learning without arm movements can lead to subsequent perfor- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 mance enhancement, although it is of lesser magnitude procedures such as mirror writing or pursuit rotor. than performance gains achieved by limb movement. Unlike pursuit rotor, there is no explicit representation Although procedural learning tasks such as pursuit rotor of the spatial sequence that could define movement. are typically described as implicit tasks, it is clear that From observations made during pilot experiments with rate of learning can be strongly modulated by conscious videotaping it is apparent that subjects consistently strategies (Rawlings & Rawlings, 1974; Rosenquist, follow the moving target with the eyes. Thus, there is 1965). For example, in the pursuit rotor task the target a general congruency between eye movement and target moves in a circle. Subjects could use the mental repre- movement in this task. sentation of ``circle'' to generate an internal model of Subjects tracked a target that moved across the screen the task requirements, reducing the need for visual in the horizontal direction only. There were two kinds of guidance. This effect might recruit frontal areas known trials, sequence and random. In random trials, the target to be involved in motor imagery such as the SMA or reversed direction at randomly generated screen posi- parietal cortex. Thus, the effects of mental rehearsal, tions. In sequence trials, the target reversed at random imagery, and strategy may complicate interpretation of positions for the first last and middle 10 sec of each trial; functional mapping results. during the remaining 60 sec the target turned at points The goal of the present set of experiments was to determined by a sequence. Figure 1A shows the hor- behaviorally isolate and then identify functional ana- izontal position of the target through the 90-sec trial and tomic substrates for some of these factors involved in Figure 1B shows a repeating 10-sec movement pattern. the acquisition of new visuomotor skills. We present In each trial throughout the experiment, the same 10- two experiments. In Experiment 1a we validate a continuous visually guided tracking task that can be used to assess procedural learning. The task is a major improvement over the pursuit rotor task for control- ling variability of kinematics, precision, and mental strategies. The psychophysical data establish that learning is sequence specific and implicit. In Experi- ment 1b we use this task with brain imaging to examine the functional anatomy of implicit procedural learning under spatially compatible conditions. The results show an increase of activity predominately in the limb motor cortex, in agreement with previous imaging studies of procedural learning. In Experiment 2a this tracking task is validated behaviorally under a spatially incompatible condition. The technique of behavioral transfer was then used. After learning a sequence of movements, subjects switched to a compatible tracking condition. Based on performance improvements after the sequence was reintroduced, the knowledge associated with this task is shown to be linked to target locations. In Experiment 2b the functional anatomy of implicit procedural learning under spatially incompatible conditions, followed by behavioral transfer, is examined. In the incompatible case, Figure 1. Schematic of the motor learning paradigm used in learning-related increases develop predominately in Experiment 1. Subjects tracked a target that moved horizontally on a computer screen for 90 sec (A). A repeating pattern was intermixed oculomotor areas rather than limb areas. With transfer with brief random movements present in the initial, middle, and final to the compatible condition there is a shift of activity 10 sec of the block. Performance (B) was assessed by comparing target into the arm motor cortex. (heavy line) and cursor position (thin line).

218 Journal of Cognitive Neuroscience Volume 13, Number 2

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 learned the specific sequence along with general parameters of the task. First, when the sequence trials were introduced, participants improved further [F(17,102) = 4.44, p<.01, MSE = 10.73]. Secondly, the RMSE increased when participants were switched back to random tracking in the last trial [F(1,6) = 10.42, p<.01, MSE = 2.59]. Rating and recognition scores for explicit knowledge indicated that some participants may have been aware of

the sequence, but their recognition scores were quite Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 low. The mean rating for the `Which group do you think you were in?' question was 4.14 (SE = 0.63), where 5 was ``no idea.'' Only two participants gave ratings above 5. Participants rated the sequences in the recognition test on a scale of 1 to 99, where 1 indicated certainty that the present sequence was not the practiced sequence and 99 indicated certainty that it was. The recognition score Figure 2. Behavioral results of Experiment 1a. During the initial nine was calculated as the rating given to the sequence minus trials the target moved randomly. This was accompanied by a rapid the mean of the ratings given to the distracters. This improvement of performance that stabilized prior to sequence score (M = 15.29, SE = 7.88) was significantly greater learning. During sequence learning (Trials 10 through 27) there was a than 0 [t(6) = 1.94, p < .05], indicating that participants subsequent performance gain. The loss of performance gains at Trial could recognize the pattern in a recognition perfor- 28 when the sequence was replaced with a random target confirms that this later learning was sequence specific. mance task. This experiment established that a spatially compatible visually guided learning task could be used to study sec sequential pattern of target movement was repeated motor skill learning. The test of explicit recognition three times in a row, followed by the middle 10 sec of suggests that subjects were unaware of a sequence and random turns, followed by the remaining three repeti- the task was successful at minimizing strategic learning. tions of the sequence. The contribution of ``top-down'' mental processes to After each of 28 trials, the screen displayed the skill acquisition could be treated as temporally constant participant's root mean square error (RMSEÐa measure in the subsequent imaging study. The significant recog- of the average distance of the cursor from the target) for nition scores suggest the presence of some form of that trial. Participants were told to try to minimize the implicit recognition or perceptual fluency. Furthermore, RMSE. the leveling off of performance during the early all- Figure 2 shows the mean RMSE for each trial. The random trials establishes that nonspecific learning of figure indicates that RMSEs decreased over the course of the apparatus and general task requirements had the experiment, as expected with learning. Some of this reached a plateau prior to the subsequent sequence- learning may be due to learning the parameters of the related learning. task, such as the target speed and the relationship between joystick movement and cursor movement, Experiment 1b: Functional Imaging of Spatially rather than the sequence per se. Indeed, learning was Compatible Procedural Learning reliable over the first nine random trials [F(8,48) = 5.04, p < .01, MSE = 16.90]. As Figure 2 indicates, this result is The spatially compatible learning task of Experiment largely due to learning in the initial four trials; perfor- 1a was examined during positron emission tomogra- mance was constant for the rest of the random trials. phy (PET) imaging to identify longitudinal changes of Two aspects of the data suggest that participants also brain activity associated with the acquisition of a new

Table 1. Experiment 1b: Schedule of Compatible Motor Pattern Learning Block 123456789101112131415161718192021222324252627282930 PET scan 1 2 3 4 5 678910 Stimulus RRRRRRRRRS S S S S S S S S S S S S S S S S S RRR Mapping CCCCCCCCCCCCCCCCCCCCCCCCCCCCCC

R = stimuli presented as random pattern; S = stimuli presented as a repeating spatial pattern; C = compatible mapping between cursor and joy-stick direction.

Grafton, Salidis, and Willingham 219

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 Subject performance while tracking in the PET scan- ner for six subjects is shown in Figure 3. Mean RMSEs decreased over the course of the experiment when the sequence was presented. Improvement over the sequence trials with PET scanning was marginally significant [F(1,5) = 5.686, p < .06]. More importantly, this behavioral change was specific to sequence learning because the RMSE increased when participants were switched back to random tracking in the last trials: Scan

9 vs. Scan 10 [F(1,5) = 20.744, p < .006]. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 The ratings scores were all low, indicating that participants were not aware of the presence of a sequence. As in Experiment 1a, there was evidence for recognition when the sequence was presented to the subjects [M = 35.66, SE = 10.36, t(6) = 3.44]. This mixed pattern of results suggests a form of awareness Figure 3. Behavioral results of Experiment 1b. Subjects practiced the that may be associated with implicit rather than explicit task prior to brain scanning, so that there was no dramatic recognition. performance gain observed during the initial random blocks. During sequence learning there was a late gain of performance. The loss of As summarized in Table 2, the functional imaging performance gains in the final trial confirms that this later learning was data revealed significant longitudinal increases of activ- sequence specific. ity during skill acquisition in the superior aspect of the contralateral (left) sensorimotor cortex and adjacent motor sequence. The primary prediction was an precentral and dorsal frontal gyrus. The latter two increase of activity in the motor cortex during learn- areas encompassed a portion of the premotor cortex ing, analogous to results obtained with the pursuit and supplementary motor area, respectively. The site rotor task. The experiment structure is shown in of motor cortex activation shown in Figure 4A was Table 1. superior, in a region associated with proximal arm

Table 2. Experiment 1b: Localization of Compatible Pattern Encoding Talairach coordinates (mm) Percent change of activity

Random Pattern Pattern to Anatomic location x y z trials trials random

Increasing activity Right globus pallidus 22 À16 3 À0.3 7.7 À2.0 Left thalamus À12 À22 4 1.3 6.6 À0.6** Left parieto-occipital fissure (18/31) À16 À61 12 À1.9 6.4 À2.2** Left postcentral gyrus (40) À57 À19 22 À0.7 4.8 À2.2 Left precuneate cortex (7) À4 À43 31 1.3 7.5 À3.9** Right anterior cingulate gyrus (32) 3 13 43 1.5 5.2 À2.6 Left central sulcus/precentral gyrus À13 À30 64 1.5 3.7 À1.3** Dorsal frontal gyrus

Decreasing activity Right posterior cerebellum* 7 À67 À24 À0.8 À4.8 3.3 Right lateral cerebellum* 34 À60 À18 À1.7 À2.2 3.6 Right fusiform gyrus (20)* 45 À30 À16 À2.6 À4.7 2.3**

Locations were determined by repeated measures ANOVA and linear contrasts on CBF values, corrected for multiple comparisons. Anatomic locations in parenthesis are Brodmann's areas according to the atlas of Talairach and Tournoux (1988). *Significant at p < .001, uncorrected for multiple comparisons. **Significant (p < .05) changes at this site were also observed in subjects of Experiment 2b.

220 Journal of Cognitive Neuroscience Volume 13, Number 2

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 1996; Petit & Haxby, 1999; Sweeney et al., 1996; O'Sullivan, Jenkins, Henderson, Kennard, & Brooks, 1995). The next most prominent area showing learning-related changes was the left postcentral sulcus of the inferior parietal lobule. Figure 4A shows this area extending into the postcentral gyrus. The time course of activity in these two areas is shown in Figure 4B. The most significant changes were near the end of the study, when most of the performance gains occurred.

Table 2 shows that the percent change of blood flow Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 in all areas during the early random blocks was incon- sequential whereas it increased substantially with the sequence trials. At the end of the experiment, all of the sites showed a drop of activity when the random task was reintroduced. After correcting for multiple comparisons there were no areas demonstrating a progressive decrease of activity during learning. When the data were reexamined without correction for multiple comparisons at a threshold of p < .001, there were two sites

Figure 4. (A) Learning-related increases of CBF during spatially compatible sequential arm movements. Areas of significance (p < .05, after correcting for multiple comparisons; in blue) are projected onto an MRI scan from a normal subject. The left hemisphere is displayed in a superior oblique view with a cutout through the arm area of the sensorimotor cortex. The most superior site is located in the arm area of sensorimotor cortex. This activation extends continuously into the left supplementary motor area (partially seen in the interhemispheric fissure). The center of this distributed activation (À13, À30, 64) is summarized in Table 2. The more inferior site is located in the rostral inferior parietal lobule. The area in red is the frontal eye field, which was significant at p < .01 without correction for multiple comparisons). (B) CBF plotted as a function of time at the sensorimotor and parietal cortex.

movements (Colebatch et al., 1991). The joystick task we used required more proximal arm movements as well. When the changes in the sensorimotor cortex were examined without correction for multiple comparisons, the activation extended inferiorly to encom- Figure 5. (A) Learning-related decreases of CBF during spatially pass the precentral gyrus in the region of the putative compatible sequential arm movements. Areas of significance (p < .001, frontal eye fields as defined by functional imaging uncorrected for multiple comparisons; in white) are projected onto an MRI scan from a normal subject. Image right is the left hemisphere. Two studies of saccade or smooth pursuit (Dejardin et al., sites in the right cerebellum and an area fusiform gyrus are shown. (B) 1998; Law, Svarer, Holm, & Paulson, 1997; Petit et al., CBF plotted as a function of time at these three sites.

Grafton, Salidis, and Willingham 221

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 Table 3. Experiment 2a: Mean Rating and Recall (Explicit Knowledge) Rating Recall

Condition M SE M SE

Perceptual 5.83 0.44 3.06 0.53 Motor 5.78 0.48 2.89 0.54 Random 5.17 0.39 3 0.4

Identical 5.83 0.36 2.72 0.42 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021

in the motor cortex and SMA. This drop does not occur in the first block for the discrete case. This could be related to a more rigid internal model for sequences linked to discrete responses, i.e., it simply takes more trials for the activity associated with a sequential repre- Figure 6. Behavioral properties of spatially incompatible motor sentation to attenuate. For continuous movements the learning. Three groups of subjects were tested during spatially error associated with random tracking might be de- incompatible learning and one group, ``identical,'' was tested under tected sooner. spatially compatible learning throughout. After transfer to a spatially compatible condition, only the ``perceptual'' group (solid diamonds) Increasing activity in the arm motor cortex are demonstrates evidence of significant performance improvement when probably not related to increases of force, velocity, the previously learned sequence was reintroduced in the perceptual or or amplitude. Although limb kinematics were not oculomotor reference frame. measured, the pattern of gross limb movements observed visually were relatively constant. If anything, subjects make less forceful movements with less fre- identified in the right cerebellum and a third site in quent and smoother corrections over time as they the right fusiform gyrus, shown in Figure 5A. The learn this task. The current results also establish that time course of activity in these areas is shown in involvement of the SMA can occur with motor learning Figure 5B. of continuous sequential movements as has been The primary observation of Experiment 1b was a shown previously with discrete movements. The re- progressive increase of activity in the proximal arm area sults also show decreasing activity in the cerebellar of the primary sensorimotor cortex and SMA with the cortex ipsilateral to the moving limb that is propor- learning of a sequential arm movement. Less significant tionate to the amount of performance error. This is changes were also observed in the region of the frontal consistent with a previous fMRI study showing de- eye fields. Under spatially compatible conditions the creasing error-related activity in the cerebellum during sequence appears to be represented at the level of a task in which subjects must learn to move a joystick motor effector (eye and arm) with a predominance of in new directions to control a cursor (Flament et al., changes occurring in the cortex associated with arm 1996). Although it is tempting to infer that this movements. There is an interesting difference between cerebellar decrease could be a direct correlate of an this study and previous imaging studies of sequencing error signal, there are alternative explanations includ- discrete finger movements using the SRT task. When ing the coding of a new ``internal model'' (Imamizu the subjects switch from a sequence to all random et al., 2000). In the latter case the cerebellum would block in the present study, there is a drop of activity be more active early by facilitating the development of

Table 4. Experiment 2b: Schedule of Incompatible Motor Pattern Learning

Block Practice 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 PET scan 1 2 3 4 5 6 7 8 9 10 Stimulus R R S S S S S S S S S R R S R R Sp Sp Sm Sm R R Mapping C I I I I I I I I I I I I I CCCCCCCC

R = stimuli presented as random pattern; S = stimuli presented as a repeating spatial pattern; C = compatible mapping between cursor and joystick direction; I = incompatible mapping between cursor and joystick direction; Sp = same pattern reintroduced in perceptual dimension; Sm = same pattern reintroduced in motor response dimension.

222 Journal of Cognitive Neuroscience Volume 13, Number 2

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 Experiment 2a: Psychophysics of Spatially Incompatible Procedural Learning Are continuous visuomotor skills represented in a com- mon set of brain structures? This could be investigated using the technique of behavioral transfer between incompatible and compatible visuomotor mappings. To do this, subjects first performed tracking with the joystick reversed. Thus, target and arm moved in opposite direction. A sequence was learned in this condition. Then, the joystick was returned to a standard configura- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 tion so that target and arm movements were compatible. The previously learned sequence was reintroduced with respect to the pattern of target movements (perceptual group, n = 18) or with respect to hand movements Figure 7. Behavioral results of spatially incompatible motor learning (motor group, n = 18). If learning occurs with respect to in subjects participating in imaging study. Data are from scan blocks target location (or perhaps eye movements, which are only. During the first five PET scans and presentation of the spatially approximately congruent with target movement), only incompatible sequence (IS) there is significant improvement of the perceptual group will show performance gains at performance. Deterioration of performance with an all random block transfer. If movements are related to motor effector, at Scan 6 (IR) confirms this performance improvement was sequence specific. After transfer to a spatially compatible condition there is a then only the motor group will show performance gains significant performance improvement when the sequence is reintro- at transfer. Two additional control groups performed (1) duced with respect to learned target (perceptual) movements (Cp) and random movements for all trials (``random'' n = 18) and no additional improvement when it is presented with respect to limb (2) all compatible trials (``identical'' n = 18). (motor) movements (Cm). The mean RMSE across trials for each condition are summarized in Figure 6. All participants (except those in a new model. With time the model might be repre- the Random condition) learned the sequence in the sented in other structures and cerebellar involvement learning phase. The perceptual group demonstrates would decrease. transfer, as indicated by the similarity between their

Table 5. Experiment 2b: Localization of Incompatible Pattern Encoding Talairach coordinates (mm) Percent change of activity

Anatomic location x y z Pattern trials Pattern to random

Increasing activity L thalamus/caudate À9 À15 21 4.2 À2.1 Right precentral gyrus (6) 43 À7 36 4.3 0.9* Left precuneus (7/24) À12 À24 43 5.6 0.1* Left precentral gyrus (4/6) À42 À15 46 5.4 À1.3* Right precentral gyrus (4/6) 42 À22 52 3.8 À0.9* Left dorsal frontal gyrus (6) À1 À13 69 4.2 À1.6 Right dorsal frontal gyrus (6) 7 À30 70 4.2 1.0*

Decreasing activity Left cerebellar nuclei and cortex À28 À60 À37 À6.0 2.1 Right lateral cerebellum 37 À54 À19 À5.3 1.4* Left middle temporal gyrus (21) À49 À54 6 À3.3 0.0 Right middle temporal gyrus (39) 46 À72 16 À4.7 1.6

Locations were determined by repeated measures ANOVA and linear contrasts on CBF values, corrected for multiple comparisons. Anatomic locations in parenthesis are Brodmann's areas according to the atlas of Talairach and Tournoux (1988). *Significant (p < .05) changes at this site were also observed in subjects of Experiment 1b.

Grafton, Salidis, and Willingham 223

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 performance and the identical group at transfer. In sequence conditions differed from the random condi- contrast, the motor group does not look different from tion (all Fs < 1). the random group after transfer. The main finding indicates that participants learned a Initial sequence learning was confirmed with a re- visually guided, sequential movement in relationship to peated measures ANOVA on the RMSEs on the first nine blocks with trial as a within- and condition as a between- subjects factor. Most importantly, a significant trial by condition interaction indicated that the learning rates differed between the conditions, as expected if there

was sequence-specific learning [F(24,544) = 4.36, p < Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 .01, MSE = 109.41]. Planned contrasts confirmed that the motor and perceptual groups' rate of learning exceeded that of the random group's [F(8,544) = 2.18, p < .05]. The level of RMSE differed among the four conditions [F(3,68) = 17.56, p < .01, MSE = 1494.79], at least partially because of the greater difficulty of reversed tracking. The effect of trial was also significant [F(8,544) = 74.08, p < .01, MSE = 109.41]. An ANOVA on the difference in RMSE between Trials 9 (sequence) and 10 (random) was also conducted to assess sequence-specific learning. As expected, this score differed among the conditions [F(3,68) = 10.85, p < .01, MSE = 42.64]. Planned contrasts showed that the motor, perceptual, and identical difference scores were greater than the random difference scores, (F values > 17), and did not differ from each other (all F scores < 1.1). This analysis indicates that the initial level of sequence-specific learning among the three sequence learning groups was comparable. Of course, the purpose of the experiment was not to demonstrate that participants can learn the sequence, but rather to specify exactly what type of sequence they learn. Do they learn the perceptual sequence, the motor sequence, or both? An ANOVA on the difference in RMSE between Trials 12 (sequence) and 13 (random) was conducted to answer this question. Again, this difference score differed among the four conditions [F(3,68) = 41.26, p <.01,MSE = 2805.37]. Planned contrasts revealed that, although the perceptual group differed from the random group [F(1,68) = 6.25, p < .05], the motor group did not [F(1,68) < 1]. Furthermore, as expected if participants were entirely relying on a perceptual representation, the perceptual group's scores did not differ from the identical group's scores [F(1,68) < 1]. Table 3 shows the mean rating and recall scores for each condition. As in Experiment 1, participants were not aware of the sequence. Participants' ratings of whether they tracked a sequence or not did not differ by condition [F(3,68) < 1]. More focused contrasts comparing each of the sequence groups to the random Figure 8. (A) Learning-related increases of CBF during spatially group also did not indicate any significant differences incompatible sequential arm movements. Areas of significance (p < (all Fs < 2). .05, after correcting for multiple comparisons; in gray) are projected The recall task yielded very similar results. The num- onto an MRI scan from a normal subject. The left hemisphere is ber of segments correctly sequenced, with three in a row displayed on image left in a superior view. The sites in the correct being the minimum chunk required for scoring, interhemispheric fissure encompass bilateral supplementary motor areas. The bilateral foci in the lateral precentral gyri are located at the was the measure of awareness. This score did not differ probable location of the frontal eye fields. (B) CBF plotted as a by condition [F(3,68) < 1]. Furthermore, none of the function of time in supplementary motor area and frontal eye fields.

224 Journal of Cognitive Neuroscience Volume 13, Number 2

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 target location. Because eye movements were probably Subject performance (Figure 7) reveals that RMSEs congruous with target movements across transfer, we decreased significantly over the initial learning phase cannot determine if the representation is at the level of of the experiment [F(8,32) = 5.75, p < .0005]. Most target location, specific eye movements, or both. The importantly, this improvement can be attributed to results provide evidence that the motor sequence is sequence learning because the RMSE increased when probably learned in a perceptual domain. As with com- participants were switched back to random tracking in patible learning, subjects were unaware of a sequence. Trial 10 [F(1,4) = 16.233, p < .02] (missing performance Even when told there was a sequence, they could not data for one subject). recall any of the sequential movement better than a After transfer there was a significant reduction of the

random group. Because a recognition test was not RMSE after the previously learned sequence was reintro- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 performed, we cannot rule out the possibility of implicit duced at the perceptual level. This is reflected in the perceptual learning. Nevertheless, the tests of explicit difference between Trials 14 and 15 [F(1,5) = 47.45, knowledge, mixed as they are, argue against any sort of p < .001]. The subsequent transfer to the motor condi- strategic learning. tion did not lead to a significant difference of performance. From Experiment 2a we can deduce that subjects are most like-ly relearning the sequence as a Experiment 2b: Functional Imaging of Spatially spatially compatible task during the ``Sm'' condition. Incompatible Procedural Learning This is supported by the observation that performance The last experiment identified longitudinal changes of deteriorates in the final all random block compared to brain activity associated with learning the incompatible the ``Sm'' condition in which the target was presented task. The incompatible task of Experiment 2a was used. with respect to movements. Because perceptual transfer was more robust than mo- The mean rating (5.33 1.2) and recognition (6.00 tor transfer, it was tested first after transfer in all 10.5) scores indicated that participants were not aware subjects. The experiment timeline is shown in Table 4. of the sequence.

Table 6. Experiment 2b: Localization of Pattern Transfer Talairach coordinates (mm) Percent change of activity, Anatomic location x y z random to Sp

Increasing activity L inferior temporal gyrus À46 À25 À7 4.3 L middle temporal gyrus À42 À79 10 5.1 R postcentral gyrus 58 À24 18 4.0 L anterior cingulate À15 43 22 3.3 L central sulcus À16 À31 67 3.4

Decreasing activity L middle frontal gyrus À37 39 12 À4.8 L posterior parietal cortex À43 À45 39 À3.9 R middle frontal gyrus 24 22 37 À4.2 L posterior parietal cortex À57 À43 36 À4.0 L inferior precentral gyrus À31 À37 42 À4.1 L precentral gyrus À43 15 43 À4.4 L superior parietal À42 À42 54 À4.3 R superior frontal sulcus 21 À15 57 À2.1 L superior parietal À25 À57 61 À2.8 L superior frontal sulcus À16 À664 À4.4

Locations were determined by a comparison of CBF values for Scans 7 and 8 using t test corrected for multiple comparisons. Anatomic locations in parenthesis are Brodmann's areas according to the atlas of Talairach and Tournoux (1988).

Grafton, Salidis, and Willingham 225

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 Changes of brain activity during the initial sequence Figure 9 shows that the location of the sensorimotor acquisition are summarized in Table 5. Increases of site overlaps with the location of maximally increasing activity were most prominent in the bilateral precentral relative cerebral blood flow (rCBF) under compatible gyrus, in the probable location of the frontal eye fields. conditions observed in Experiment 1b, i.e., the arm area. Bilateral increases were also present in the bilateral Numerous sites decreased in activity with the reintro- mesial frontal cortex within the supplementary motor duction of the sequence under compatible condition, area. The location of these sites with respect to local including the left frontal eye field and bilateral dorsal gyral anatomy and their time activity profiles are shown premotor cortex. in Figure 8. Learning effects in the sensorimotor cortex We did not test for transfer effects between the Sp and

were also examined without correction for multiple Sm conditions because of the lack of significant beha- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 comparisons. In this case, significant increases were also vioral differences between these two tasks and the observed in the arm area of the sensorimotor cortex ambiguity in interpreting any transfer effects at this during incompatible learning, although they were of a point of the experiment. lesser magnitude than the changes in the frontal eye fields. Relative decreases of activity during sequence GENERAL DISCUSSION learning were observed in the bilateral cerebellum and temporal cortex. The primary conclusion of these experiments is that the To examine the effect of behavioral transfer, we functional anatomic correlate for learning continuous compared changes of brain activity between Trials 14 sequential patterned limb movements depends critically and 16. In this case, all subjects had already transferred on the specific sensory to motor transformation. to trials where the joystick and cursor moved in the Although skill learning in the imaging experiments was same direction. From Experiment 2a it was known that accompanied by progressive increases of activity in both the sequence was initially learned with respect to target the arm region of the sensorimotor cortex and the movement. When the target sequence was reintroduced frontal eye fields, the relative contribution of these two in the same way that it was learned (Sp), with respect to areas depended on the spatial congruency of the task. the direction of target movement, there was a significant From the present data it could argued that there is a increase of activity in the anterior cingulate, temporal functional hierarchy in this process. In the congruent and sensorimotor cortex, as summarized in Table 6. case the arm (and cursor) movement is directly tracking target location. Thus, a representation of the sequence linked to the arm cortical area takes precedence. For the incongruent case the arm area is inversely related to target (and cursor) direction. Sequential eye movements, on the other hand, could be congruent to cursor and target and form the next best solution for tracking. Consistent with this hypothesis our results showed maximal learning-related changes in the frontal eye fields and weak activation in the arm area. Although we did not measure eye movements, from pilot studies with videotaping of subject eye movements we found consistent tracking movements of the eyes with the target. As a corollary, we found subjects could not reliably perform this task with the eyes held in fixation. This hierarchical model is also supported by CBF changes that occurred after transfer. With the reintroduction of the sequence under spatially compatible conditions, there was an associated increase of activity in the arm motor area of the sensorimotor cortex and decrease of activity in the frontal eye fields. A potential pitfall of our study was the use of multiple Figure 9. (A) Location of increasing blood flow during compatible sequence learning (Experiment 1b). The increase encompasses the study populations and experiments to identify areas supplementary motor area and sensorimotor cortex. (B) Location of involved in different forms of motor learning. Never- increasing blood flow during incompatible sequence learning (Experi- theless, we have been cautious in not pooling our data ment 2b). The increase encompasses the supplementary motor area. inappropriately and the main conclusions of this study (C) Increase of CBF after transfer from a spatially incompatible to do not rely on negative inferences between populations. compatible condition and reintroduction of a just learned sequence. Note the increase of activity in the motor cortex (white arrow) that It is emphasized that the findings of Experiment 1b, overlaps in location with the increase observed in the group of subjects (reorganization in the limb area of the motor cortex) who only learned under spatially compatible conditions. strongly recapitulate what has been observed in many

226 Journal of Cognitive Neuroscience Volume 13, Number 2

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 previous studies of procedural learning. The results from tory presynaptic activity (Nudo & Masterton, 1986). Experiment 2b are strikingly different. It is unlikely that What is remarkable from the current study is that the these group differences are related to experimental representation of previously learned sequences can error. Another potential problem of this study was the ``relocate'' across motor output areas. This effect begs rather complex result of testing for explicit recall and the question of whether there are unique sites in the recognition in the spatially compatible task (Experi- cortex or basal ganglia that might be a source for this ments 1a and 1b). In this task, subjects had no explicit sequential information. We have previously suspected a knowledge of the presence of a short, simple repeating critical role for the rostral inferior parietal lobule in sequential pattern. In Experiment 2, none of the sub- representing sequences (Grafton et al., 1998). Learning-

jects, even when instructed there was a sequence, could related changes were again observed in this area in Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 reliably recall a portion of the sequence. This rules out Experiment 1b, under spatially compatible learning the presence of conscious awareness in the learning conditions. However, changes were not significant un- task. Thus, learning-related changes during functional der spatially incompatible conditions. Alternatively, imaging are unlikely to be related to ``top-down,'' men- there is remarkable consistency with which the SMA tal, or strategic operations. On the other hand, some of and/or pre-SMA are active during sequence learning, the subjects, after being told of a sequence, had the suggesting a critical role for this premotor area. In ability to correctly recognize the movement pattern and contrast, we found no learning-related changes in the only if they actually performed the movements. This is basal ganglia. From other studies of continuous visually consistent with implicit perceptual (visual or somato- guided tracking with similar methods we can consis- sensory) learning. This may represent the main substrate tently detect significant changes of activity in the globus for skill learning in this task. pallidus, thalamus, and posterior putamen as a function The dynamic property of learning-related changes of movement (Winstein, Grafton, & Pohl, 1997), move- observed in the current study is not without precedent ment velocity (Turner, Grafton, Votaw, Delong, & Hoff- (Grafton, Hazeltine, & Ivry, 1998). In a previous intralimb man, 1998), and movement amplitude (unpublished transfer study we observed a transfer of learning-related data). Thus, the lack of learning-related changes in the activity from the finger area to the arm area of the motor present study is unlikely to be due to a general inability cortex when a sequence that was learned with respect to to detect activation of subcortical structures. Finally, it is finger movements was reintroduced after subjects had possible that a widespread set of cortical areas may transferred to a large keyboard requiring whole-arm share information used to generate sequential informa- movements. The results suggest that functional imaging tion. In such a widely distributed process, no single area may yield a fundamentally different measure of learning may uniquely demonstrate significant increases of blood than is provided by experiments employing transcortical flow during skill acquisition. magnetic stimulation (TMS) or direct physiologic recording (Pascual-Leone et al., 1994). With the TMS method, the size of the representation for individual muscles of METHODS the hand can be estimated. With learning, the size of the Experiment 1a representation for relevant fingers enlarged whereas it remained constant for irrelevant fingers. In this case the Subjects size of the representation is presumed to be linked to Seven undergraduate students (four male) from the the motor cortex output map. A logical interpretation of University of Virginia participated to partially fulfill a the TMS data is that the change of activity or size of course requirement. representation is related to a remodeling of the output map in the motor cortex. This interpretation is sup- Stimuli and Apparatus ported by physiological studies of rat and monkey showing use dependent, rapid reorganization of fore- Participants used an Advanced Gravis optical joystick limb motor output maps, mediated by shifts of inhibitory (Advanced Gravis Computer Technology, Burnaby, Ca- inputs to the motor cortex (Donoghue, 1995; Jacobs & nada) to track a moving target, a 1.06-cm diameter black Donoghue, 1991; Nudo, Jenkins, & Merzenich, 1990; circle, across a white background (the computer Nudo, Milliken, Jenkins, & Merzenich, 1996). screen). The experiment was conducted on a Macintosh The functional imaging data identify a transitive pro- IIci computer. cess, with shifts of activity across motor output areas Throughout each 90-sec trial, the target moved that may occur more rapidly than what is observed with smoothly at a rate of 105.6 mm/sec across the screen TMS or neural recording. These shifts of activity mea- in the horizontal direction only. The vertical position sured by CBF may represent changes of afferent activity was the midpoint of the screen. There were two kinds of projecting into the motor areas. Indeed, it has been trials, sequence and random. In random trials, the target shown that metabolic or blood flow activation most reversed direction at randomly generated screen posi- strongly reflects a combination of excitatory and inhibi- tions. In sequence trials, the target reversed at random

Grafton, Salidis, and Willingham 227

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 positions for the first last and middle 10 sec of each trial; and no person was on psychoactive medications. All during the remaining 60 sec the target turned at points subjects were strongly right handed (decile range 89± determined by a sequence. Figure 1A shows the hor- 100; Oldfield, 1971). izontal position of the target through the 90-sec trial and Figure 1B shows a repeating 10-sec movement pattern. Behavioral Tasks and Performance Measures In each trial throughout the experiment, the same 10- sec sequential pattern of target movement was repeated The pursuit task defined in Experiment 1a was used. A three times in a row, followed by the middle 10 sec of total of 30 trials were presented over the PET scanning random turns, followed by the remaining three repeti- session. As shown in Table 1, there were two practice

tions of the sequence. There were no breaks in target trials between each scan trial and there were three Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 movement between these segments of the trial. Cursor random blocks at the end of the session. location (controlled by the subject) and target position were compared by the program 5 times/sec. The fre- Imaging quency with which the target changed directions was the same for the random and sequence trials. rCBF was determined using the PET autoradiographic method (Herscovitch, Markham, & Raichle, 1983; Raichle, Martin, & Herscovitch, 1983). For each scan, a Procedure 15 bolus of 35 mCi of H2 O was injected intravenously Each participant performed a total of 28 trials. The first commensurate with the start of the behavioral task. A nine trials and the last trial were entirely random. All 90-sec scan was acquired in ``2-D mode'' beginning 10 other trials contained the sequence. Participants were sec after tracer administration. Attenuation correction not informed about the repeating nature of the se- was based on a calculated method using boundaries quence trials. To simulate the body position to be used defined separately on each emission scan coupled with a for PET imaging, a participant tracked the target while transmission scan of the PET head-holder. After recon- lying on the floor on a mat. Their head was supported struction by filtered back-projection, image resolution with a pillow. Each participant was asked to position the was 11.8 mm full width at half maximum (FWHM) as joystick so that their grasp was comfortable. The moni- verified by a line source. Blood samples were not tor was approximately 75 cm above the participant's acquired. Images of radioactive counts were used to knees, with the screen tilted towards their face. estimate rCBF as described previously (Mazziotta et al., After each trial, the screen displayed the participant's 1985; Fox, Mintun, Raichle, & Herscovitch, 1984). RMSE for that trial. Participants were told to try to minimize the RMSE. Rest breaks, provided after every Image Analysis trial, were 2 min long. In addition, after the ninth and eighteenth trials, participants were given 10 min rest For each subject, all PET scans were coregistered to each breaks during which they were asked to get up, stretch, other and the mean PET (Woods et al., 1998a) was then get a drink of water, etc. coregistered using affine and then nonlinear algorithms Following the 28 trials, participants were told that the to a PET target atlas centered and rescaled to the target's movement was always random for some people Talairach atlas (Talairach and Tournoux, 1988; Woods (although it was not), but it often moved in a sequence et al., 1998b). The target was comprised of PET scans for others. They were asked to indicate which group from 20 normal adult subjects. All PET studies were they believed they were in on a scale of 1 (definitely in smoothed with a Gaussian filter to a final image resolu- the random group) to 9 (definitely in the sequence tion of 14.8 mm FWHM and globally normalized to each group). Participants were also asked to track five 10- other by proportional rescaling. Application of the gen- sec segments of target movement (the sequence and eral linear model of analysis of variance (ANOVA) were four distracters), and rate the likelihood that each was used to calculate task differences on a pixel by pixel the segment that was repeated in the earlier task. basis without global pooling of image variance (Woods et al., 1996). For the given experimental paradigm there were Experiment 1b several possible approaches for identifying learning-related changes of brain activity. We used a model of Subjects learning predicated on the notion that areas involved Six normal young adult subjects (three men and three in the initial encoding of a movement sequence should women), mean age 32 14 years, volunteered for this demonstrate progressive increases of brain activity as study under informed consent in accordance with the learning takes place. This was tested with an ANOVA Emory University Human Investigations Committee. design using weighted linear contrasts comparing regio- Subjects were judged to be normal by excluding any nal CBF values. In this case, the weights were defined by prior neurologic, psychiatric, or major medical history the mean performance for each of the 10 scan trials. This

228 Journal of Cognitive Neuroscience Volume 13, Number 2

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 statistic would identify areas where blood flow increased seated, with their arm resting on a counter with the at a rate similar to changes of performance. This ap- monitor and joystick. proach was used previously in studies of pursuit rotor There were two phases to the procedure, learning and learning (Grafton et al., 1992, 1994). The resultant t transfer. During the learning phase, participants learned a statistic image was masked at a threshold of p < .001. repeating sequence. During the transfer phase, some Areas achieving this threshold were further evaluated for aspects of the sequence were kept constant and some significance after correcting for multiple comparisons changed. Aspects of the sequence were dissociated by (Friston, Worsley, Frackowiak, Mazziotta, & Evans, changing the stimulus±response mapping so that some 1994). Areas showing an inverse relationship between participants saw the same perceptual sequence, but made

performance and rCBF were similarly identified. Areas different motoric responses to it (perceptual condition) Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 reaching significance were superimposed on a reference whereas other participants saw a different perceptual atlas of MRI anatomy centered in Talairach coordinates sequence at transfer than they had seen during the with 2-D and 3-D renderings using the Advanced Visua- learning phase, but made the same sequence of motoric lization Systems (AVS) software package. rCBF values responses that they had during the learning phase (motor showing a significant change at all sites, for all scans and condition). Another group saw a different perceptual subjects, were obtained and percent change of CBF sequence and made different motor movements (random between tasks were calculated (see Tables 2, 5, and 6). condition) and for a final group of participants, both the perceptual and motor sequence at transfer matched those seen during training (identical condition). The goal of Experiment 2a these transfer conditions was to examine which aspect of the sequence (perceptual or motor) could support good Subjects performance during the transfer phase. Seventy-two undergraduate students (21 male) from the The dissociation of the perceptual and motor se- University of Virginia participated to partially fulfill a quences was made possible by changing the stimulus± course requirement. Participants were randomly as- response mapping at transfer. During the learning signed to the ``random,'' ``perceptual,'' ``motor,'' or phase, participants tracked the target with an incompa- ``identical'' conditions with the constraint that an equal tible stimulus±response mapping. When the joystick was number of participants were in each condition. moved to the left, the cursor moved to the right, and vice versa. At transfer, the mapping was made compatible. Thus in the perceptual condition participants saw Stimuli and Apparatus the same perceptual sequence, but a different motor The stimuli and apparatus were identical to those used sequence was necessary to track the target. In the motor in Experiment 1 except as follows. (1) The sequence condition, a difference perceptual sequence was used at involved 5 sec, rather than 10 sec, of target movement. transfer, constructed so that the motor responses ne- Therefore, each 10-sec sequence segment consisted of cessary to track it were identical to those used during two seamless repetitions of the sequence. (2) After the learning phase. In the random condition, the se- behavioral transfer the target sequence was presented quence used at transfer had not been seen during the in reverse order. A reversed sequence was generated learning phase. The identical condition was slightly from the original sequence by translation and rotation: different; they used the compatible mapping during 250 pixels were added to each target position and then both the learning and the transfer phase so that the the direction of movement was reversed. The translation perceptual and motoric sequences would be same in the was necessary to keep the sequence on the screen. two phases. Subjects in all conditions saw the identical Participants in the perceptual, motor, and random stimuli during the transfer phase. The difference among groups tracked with a reversed joystick during the first the conditions lay in the training during learning phase portion of the experiment. The reversed joystick was a and how it related to the transfer phase. joystick with its base turned 1808 from the participant. The learning phase consisted of nine sequence trials With a reversed joystick there is a spatial 1808 incompat- followed by one random trail. The transfer phase was ibility between joystick and cursor directions. Moving composed of a random trial, a sequenced trial, and then the handle to the right results in the cursor moving to a final random trial. the left, and vice-versa. The ``identical'' participants used As in Experiment 1, participant's awareness of the a joystick in standard orientation with compatible move- sequence was assessed with a rating scale. In addition, a ments of joystick and cursor. recall rather than a recognition test was used. This change was made to avoid the possibility of recognition failure due to idiosyncratic starting points. If participants Procedure think the sequence starts in its actual middle, for Each participant performed a total of 13 trials, separated example, they may fail to recognize it when they see it by 30-sec rest breaks. All participants were comfortably starting at its beginning. For the recall test, participants

Grafton, Salidis, and Willingham 229

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 were asked to reconstruct the sequence out of card- test comparing Scans 7 and 8. Results were assessed for board segments. The pieces were three different lengths significance, corrected for multiple comparisons and with arrows drawn on them. The pieces corresponded rendered using the same methods as Experiment 1b. to the different segment lengths (the distance the target would travel before turning) of the sequence. Partici- Acknowledgments pants were asked to place 10 pieces in order to indicate the direction and distance of the target's movement in This work was supported by US Public Health Service Grants the sequence. They were to use the piece that most NS 33504 to S.T.G. and National Science Foundation Grant SBR-9905342 to D.W. closely matched the desired length.

Reprint requests should be sent to Scott T. Grafton, MD, Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 Center for Cognitive Neuroscience, 6162 Moore Hall, Dart- Experiment 2b mouth College, Hanover, NH 03755. Subjects Six normal young adult subjects (two men and four women), mean age 30.0 4.2 years, volunteered for REFERENCES this study under informed consent in accordance with Adams, J. A. (1987). Historical review and appraisal of research the Emory University Human Investigations Committee. on the learning, retention and transfer of human motor Subjects were judged to be normal by excluding any skills. Psychology Bulletin, 101, 41±74. prior neurologic, psychiatric, or major medical history Ammons, R. B. (1947). Acquisition of motor skills: II. Rotary pursuit performance with continuous practice before and after a and no person was on psychoactive medications. All single rest. Journal of Experimental Psychology, 37, 393±411. subjects were strongly right-handed (decile range 83± Ammons, R. B., Farr, R. G., Bloch, E., Neumannn, E., Dey, M., 100; Oldfield, 1971). Marion, R., & Ammons, C. H. (1958). Long-term retention of perceptual±motor skills. Journal of Experimental Psychol- ogy, 55, 3318±3328. Behavioral Tasks and Performance Measures Classen, J., Liepert, J., Wise, S. P., Hallett, M., & Cohen, L. G. (1998). Rapid plasticity of human cortical movement repre- The task defined in Experiment 2a was used. Each trial sentation induced by practice. Journal of Neurophysiology, lasted 90 sec. A total of 20 trials were presented over the 79, 1117±1123. PET scanning session. As shown in Table 4, there was Colebatch, J. G., Adams, L., Murphy, K., Martin, A. J., Lam- one practice trial between each scan trial. The session merstsma, A. A., Tochon-Danguy, H. J., Cark, J. C., Friston, K. J., & Guz, A. (1991). Regional cerebral blood flow during vo- began with two trials each of random practice in the litional breathing in man. Journal of Physiology, 443, 91±103. compatible and incompatible conditions to familiarize Dejardin, S., Dubois, S., Bodart, J. M., Schiltz, C., Delinte, A., subjects with the apparatus. Then, all subjects learned Michel, C., Roucoux, A., & Crommelinck, M. (1998). PET the sequence with the joystick reversed such that cursor study of human voluntary saccadic eye movements in dark- and joystick moved in opposite directions. At transfer, ness: effect of task repetition on the activation pattern. European Journal of Neuroscience, 10, 2328±2336. the joystick was rotated 1808 so that the cursor and Donoghue, J. P. (1995). Plasticity of adult sensorimotor repre- joystick moved in the same direction. After transfer the sentations. Current Opinion in Neurobiology, 5, 749±754. previously learned target was reintroduced with the Doyon, J., Owen, A. M., Petrides, M., Sziklas, V., & Evans, A. C. target moving in the same sequence as it had prior to (1996). Functional anatomy of visuomotor skill learning in transfer (``perceptual transfer, Sp''). Then, the target human subjects examined with positron emission tomography. European Journal of Neuroscience, 8, 637±648. sequence was reversed leading to identical movements Flament, D., Ellermann, J. M., Kim, S.-G., Ugurbil, K., & Ebner, as had been used prior to transfer (``motor transfer, T. J. (1996). Functional magnetic resonance imaging of cere Sm''). Perceptual transfer was tested first as this manip- bellar activation during the learning of a visuomotor disso- ulation demonstrated better evidence for a transfer ciation task. Human Brain Mapping, 4, 210±226. effect in the psychophysical study 2a. Fox, P. T., Mintun, M. A., Raichle, M. E., & Herscovitch, P. (1984). A non-invasive approach to quantitative functional 15 brain mapping with H2 O and positron emission tomogra- Imaging phy. Journal of Cerebral Blood Flow and Metabolism, 4, 329±333. Imaging procedures and spatial normalization were Friston, K. J., Worsley, K. J., Frackowiak, R. S. J., Mazziotta, J. C., identical to Experiment 1b. Learning-related increases & Evans, A. C. (1994). Assessing the significance of focal activations using their spatial extent. Human Brain Map- of activity were identified using an ANOVA design using ping, 1, 210±220. weighted linear contrasts. In this case, the weights were Grafton, S. T., Hazeltine, E., & Ivry, R. (1995). Functional defined by the mean performance for the six scan trials anatomy of sequence learning in normal humans. Journal of prior to behavioral transfer. This statistic would identify Cognitive Neuroscience, 7, 497±510. areas where blood flow increased at a rate similar to Grafton, S. T., Hazeltine, E., & Ivry, R. B. (1998). Abstract and effector-specific representations of motor sequences identi- changes of performance. After transfer, areas showing fied with PET. Journal of Neuroscience, 18, 9420±9428. changes of activity in association with the reintroduction Grafton, S. T., Mazziotta, J. C., Presty, S., Friston, K. J., Frack- of the sequence were identified using a pixel by pixel t owiak, R. S. J., & Phelps, M. E. (1992). Functional anatomy of

230 Journal of Cognitive Neuroscience Volume 13, Number 2

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021 human procedural learning determined with regional cere- tations in primary motor cortex of adult squirrel monkeys. bral blood flow and PET. Journal of Neuroscience, 12, Journal of Neuroscience, 16, 785±807. 2542±2548. Oldfield, R. C. (1971). The assessment and analysis of handed- Grafton, S. T., Woods, R. P., & Tyszka, J. M. (1994). Functional ness: The Edinburgh inventory. Neuropsychologia, 9, 97±113. imaging of procedural motor learning: relating cerebral O'Sullivan, E. P., Jenkins, I. H., Henderson, L., Kennard, C., & blood flow with individual subject performance. Human Brooks, D. J. (1995). The functional anatomy of remembered Brain Mapping, 1, 221±234. saccades: a PET study. NeuroReport, 6, 2141±2144. Haier, R. J., Siegel, B. V., MacLachlan, A., Soderling, E., Lot- Pascual-Leone, A., Grafman, J., & Hallett, M. (1994). Modula- tenberg, S., & Buchsbaum, M. S. (1992). Regional glucose tion of cortical motor output maps during development of metabolic changes after learning a complex visuospatial/ implicit and explicit knowledge. Science, 263, 1287±1289. motor task: a positron emission tomographic study. Brain Petit, L., & Haxby, J. V. (1999). Functional anatomy of pursuit

Research, 570, 133±143. eye movements in humans as revealed by fMRI. Journal of Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/2/217/1759041/089892901564270.pdf by guest on 18 May 2021 Hazeltine, E., Grafton, S. T., & Ivry, R. (1997). Attention and Neurophysiology, 81, 463±471. stimulus characteristics determine the locus of motor-se- Petit, L., Orssaud, C., Tzourio, N., Crivello, F., Berthoz, A., & quence encoding. A PET study. Brain, 120, 123±140. Mazoyer, B. (1996). Functional anatomy of a prelearned se- Herscovitch, P., Markham, J., & Raichle, M. E. (1983). Brain quence of horizontal saccades in humans. Journal of Neu- 15 blood flow measured with intravenous H2 O: I. Theory roscience, 16, 3714±3726. and error analysis. Journal of Nuclear Medicine, 24, Pew, R. W. (1974). Levels of analysis in motor control. Brain 782±789. Research, 71, 393±400. Imamizu, H., Miyauchi, S., Tamada, T., Sasaki, Y., Takino, R., Raichle, M. E., Martin, W. R. W., & Herscovitch, P. (1983). Brain 15 Putz, B., Yoshioka, T., & Kawato, M. (2000). Human cere- blood flow measured with intravenous H2 O: II. Imple- bellar activity reflecting an acquired internal model of a new mentation and validation. Journal of Nuclear Medicine, 24, tool [see comments]. Nature, 403, 192±195. 790±798. Jacobs, K. M., & Donoghue, J. P. (1991). Reshaping the cortical Rauch, S. L., Savage, C. R., Brown, H. D., Curran, T., Alpert, N. motor map by unmasking latent intracortical connections. M., Kendrick, A., Fischman, A. J., & M., K. S. (1995). A PET Science, 251, 944±947. investigation of implicit and explicit sequence learning. Hu- Jenkins, I. H., Brooks, D. J., Nixon, P. D., Frackowiak, R. S. J., & man Brain Mapping, 3, 271±286. Passingham, R. E. (1994). Motor sequence learning: a study Rawlings, E. I., & Rawlings, I. I. (1974). Rotary pursuit tracking with positron emission tomography. Journal of Neuro- following mental rehearsal as a function of voluntary control science, 14, 3775±3790. of visual imagery. Perception and Motor Skills, 38, 302. Jueptner, M., Stephan, K.-M., Frith, C. D., Brooks, D. J., Roland, P. E., Gulyas, B., & Seitz, R. J. (1991). Structures in the Frackowiak, R. S. J., & Passingham, R. E. (1997). Anatomy of human brain participating in visual learning, tactile learning, motor learning: frontal cortex and attention to action. and motor learning. In N. M. W. Larry, R. Squire, G. Lynch, & Journal of Neurophysiology, 77, 1313±1324. J. L. McGaugh (Eds.), Memory: Organization and locus of Karni, A., Meyer, G., Jezzard, P., Adams, M. M., Turner, R., & change (pp. 95±113). New York: Oxford University Press. Ungerleider, L. G. (1995). Functional MRI evidence for adult Rosenquist, H. S. (1965). The visual response of component motor cortex plasticity during motor skill learning. Nature, rotary pursuit tracking. Perception and Motor Skills, 21, 377, 155±157. 555±560. Karni, A., Meyer, G., Rey-Hipolito, C., Jezzard, P., Adams, M. M., Saint-Cyr, J. A., Taylor, A. E., & Lang, A. E. (1988). Procedural Turner, R., & Ungerleider, L. G. (1998). The acquisition of learning and neostriatal dysfunction in man. Brain, 111, skilled motor performance: fast and slow experience-driven 941±959. changes in primary motor cortex. Proceedings of the Na- Schlaug, G., Knorr, U., & Seitz, R. (1994). Inter-subject varia- tional Academy of Sciences, U.S.A., 95, 861±868. bility of cerebral activations in acquiring a motor skill: a Lang, W., Lang, M., Podreka, I., Steiner, M., Uhl, F., Suess, E., study with positron emission tomography. Experimental Muller, C., & Deecke, L. (1988). DC-potential shifts and re- Brain Research, 98, 523±534. gional cerebral blood flow reveal frontal cortex involvement Seitz, R. J., Canavan, A. G. M., Yaguez, L., Herzog, H., Tellmann, in human visuomotor learning. Experimental Brain Re- L., Knorr, U., Huang, Y., & Homberg, V. (1994). Successive search, 71, 353±364. roles of the cerebellum and premotor cortices in trajectorial Law, I., Svarer, C., Holm, S., & Paulson, O. B. (1997). The learning. 5, 2541±2544. activation pattern in normal humans during suppression, Shadmehr, R., & Holcomb, H. H. (1997). Neural correlates of imagination and performance of saccadic eye movements. motor memory consolidation. Science, 277, 821±825. Acta Physiologica Scandinavica, 161, 419±434. Sweeney, J. A., Mintun, M. A., Kwee, S., Wiseman, M. B., Brown, Mazziotta, J. C., Huang, S.-C., Phelps, M. E., Carson, R. E., D. L., Rosenberg, D. R., & Carl, J. R. (1996). Positron emis- MacDonald, N. S., & Mahoney, K. (1985). A noninvasive sion tomography study of voluntary saccadic eye movements positron computed tomography technique using oxygen-15- and spatial working memory. Journal of Neurophysiology, labeled water for the evaluation of neurobehavioral task 75, 454±468. batteries. Journal of Cerebral Blood Flow and Metabolism, Turner, R. S., Grafton, S. T., Votaw, J. R., Delong, M. R., & 5, 70±78. Hoffman, J. M. (1998). Motor subcircuits mediating the Nudo, R. J., Jenkins, W. M., & Merzenich, M. M. (1990). Repe- control of movement velocity: A PET study. Journal of titive microstimulation alters the cortical representation of Neurophysiology, 80, 2162±2176. movements in adult rats. Somatosensory and Motor Re- Willingham, D. B. (1998). A neuropsychological theory of mo- search, 7, 463±483. tor skill learning. Psychology Review, 105, 558±584. Nudo, R. J., & Masterton, R. B. (1986). Stimulation-induced Willingham, D. B. (1999). Implicit motor sequence learning is [14C]2-deoxyglucose labeling of synaptic activity in the cen- not purely perceptual. Memory and Cognition, 27, 561±572. tral auditory system. Journal of Comparative Neurology, Winstein, C. J., Grafton, S. T., & Pohl, P. S. (1997). Motor task 245, 553±565. difficulty and brain activity: investigation of goal-directed Nudo, R. J., Milliken, G. W., Jenkins, W. M., & Merzenich, M. M. reciprocal aiming using positron emission tomography. (1996). Use-dependent alterations of movement represen- Journal of Neurophysiology, 77, 1581±1594.

Grafton, Salidis, and Willingham 231

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901564270 by guest on 02 October 2021