Exp Brain Res (2003) 148:62–76DOI 10.1007/s00221-002-1277-4

R E S E A R C H A R T I C L E

Oron Levin · Nicole Wenderoth · Maarten Steyvers ·Stephan P. Swinnen

Directional invariance during loading-related modulations of muscleactivity: evidence for motor equivalence

Received: 14 January 2002 / Accepted: 10 September 2002 / Published online: 9 November 2002� Springer-Verlag 2002

Abstract In the present study, we investigated theinfluence of external force manipulations on movementsin different directions, while keeping the amplitudeinvariant. Subjects (n=10) performed a series of cyclicalanteroposterior, mediolateral, and oblique line-drawingmovements (star drawing task) with their dominant limbin the horizontal plane. To dissociate kinematics from theunderlying patterns of muscle activation, spring loadingwas applied to the forearm of the moving limb. Whereasspring loading of the arm resulted in considerable changesin the overall amount of muscle activation in the elbowand shoulder muscles, invariance was largely maintainedat the kinematic level. Subjects produced the requiredmovement directions and amplitudes of the star drawinglargely successfully, irrespective of the force bias inducedby the spring. These observations demonstrate motorequivalence and strengthen the notion that the spatialrepresentation of drawing movements is encoded in thehigher brain regions in a rather abstract form that isdissociated from the concrete muscle activation patternsunderlying a particular movement direction. To achievethis goal, the central nervous system shifted between twoor more muscle grouping strategies to overcome modu-lations in the interaction among posture-dependent (jointstiffness), dynamic (inertial), and elastic (spring) torquecomponents in the joints. Spring loading induced generalchanges in the overall amount of EMG activity, whichwas largely muscle but not direction specific, presumablyto represent the posture-dependent biasing force of thespring. Loading was mainly shown to increase musclecoactivation in the elbow joint. This indicates that thesubjects tended to increase stiffness in the elbow tocompensate for changes in the spring bias forces in orderto minimize trajectory errors. Changes in muscle group-

ing of the shoulder antagonists were mainly a conse-quence of movement direction but were also affectedpartly by loading, presumably reflecting the influence ofdynamic force components. Taken together, the resultsconfirmed the hypothesis that changes of movementdirection and direction of force in the end-effectorgenerated specific sets of muscle grouping to overcomethe dynamic requirements in the joints while keeping thekinematics largely unchanged. This suggests that direc-tional tuning in muscle activity and changes in musclegrouping reflects the formation of appropriate internalmodels in the CNS that give rise to motor equivalence.

Keywords Motor control · End-effector kinematics ·Electromyography · Cyclical movements · Directionaltuning · Motor equivalence · Muscle grouping

Introduction

Recent psychophysical studies support the idea thatdetails of a movement trajectory can be scale-and plane-invariant (Gordon et al. 1994; Grasso et al. 1998;Lacquaniti et al. 1982; Lacquaniti 1989; Swinnen 1997;Wing 2000). For example, letters have the same formwhen written on a blackboard and on a paper, even thoughthe muscle groups used to implement both tasks aretotally different as are the torques required for overcom-ing gravity. Additionally, it has been shown that thekinematics of two motor tasks can be largely the same,even if the underlying muscle activation patterns differdramatically (Grasso et al. 1998; Swinnen et al. 2001;Thoroughman and Shadmehr 1999). This phenomenon,referred to as “motor equivalence” by Lashley (1930),results partly from the fact that the number of muscles issubstantially larger than the number of joints. Thisredundancy allows the CNS to adopt more effective andparsimonious control strategies (Bernstein 1967; Flashand Hogan 1985; Lacquaniti et al. 1982) and makes itpossible that different muscle groupings can still result inthe same end-effector trajectories.

O. Levin · N. Wenderoth · M. Steyvers · S.P. Swinnen ())Motor Control Laboratory, Department of Kinesiology,Group Biomedical Sciences, Katholieke Universiteit Leuven,Tervuurse Vest 101, 3001 Leuven, Belgiume-mail: Stephan.Swinnen@flok.kuleuven.ac.beTel.: +32-16-329071Fax: +32-16-329197

Motor equivalence has been observed in a wide varietyof tasks (Kelso and Tuller 1983; Kelso et al. 1984;Rijntjes et al. 1999; Swinnen 1997), suggesting thatmovements are encoded at a more abstract level thanmuscle activation patterns. Thus, hand movements maybe represented by generalized motor programs (Schmidt1988) or in terms of strokes, which are characterized byrelative positions and intended movement directions. Thedetails of motor implementation, such as stroke size andspeed, may be left unspecified until the effector is known(Grasso et al. 1998; Rijntjes et al. 1999; Wing 2000).According to this scheme, important movement parame-ters seem to be encoded within the CNS, irrespective oflimb representation itself, e.g., the activation of specificmuscle groups (Rijntjes et al. 1999).

Additionally, several studies have shown that, in orderto achieve a specific movement goal, the CNS can adaptto new environmental characteristics. For example,Shadmehr and Mussa-Ivaldi (1994) showed that, withpractice, subjects were able to overcome perturbationsinduced by an external force field, producing straightgoal-directed movements. However, they were unable togenerate the desired reaching paths whenever the per-turbing forces were unexpectedly removed. This findingsuggests that there might be different levels of motorcontrol, such that movement trajectories are representedby rather abstract parameters, which then undergo furtherprocessing at lower levels of the CNS (Swinnen 1997) toyield the specific muscle activation patterns. It can beassumed that this transformation from the intendedtrajectory to its execution may utilize internal models,which represent relations between a given environmentand the used end-effector. During adaptational processes,these internal models may undergo substantial modifica-tions while the more abstract movement parametersremain unchanged (Scheidt et al. 2000; Thoroughmanand Shadmehr 1999). Taken together, there is someevidence that goal directed movements are encoded byrather abstract motion parameters, which do not considervariations of the environment (e.g., counteracting forces)or end-effector properties (e.g., biomechanical differ-ences) and which can be largely maintained acrossdifferent requirements of speed or acceleration.

Several scientific disciplines have tried to identify suchgeneral motion parameters and have revealed evidencethat trajectories are planned in terms of movement extent(e.g., amplitude) and direction. Psychophysical studieshave suggested that movement direction and amplitudeare two important movement parameters, which arepossibly subserved by (partially) distinct neural encodingprocesses that are largely independent of the amount andpatterns of underlying muscle activation (Favilla et al.1989; Gordon et al. 1994). Neurophysiological experi-ments involving recordings from single motor corticalneurons in primates have shown to support these views(Georgopoulos et al. 1991, 1992; Scott and Kalaska 1997;Kakei et al. 1999).

The present study investigated the influence of exter-nal force manipulations on movements in different

directions, while keeping the specification of amplitudeinvariant. To achieve this goal, subjects moved theirdominant limb across eight sequential directions of a starfigure, starting with anteroposterior (AP) lines but shiftingdirection with 45� in a clockwise fashion each time aseries of five lines was completed. The force manipula-tions consisted of spring loading from different directions.Loading was accomplished by attaching a spring to themoving right forearm (proximal to the wrist). EMGactivity from the flexor and extensor muscles of theshoulder and elbow of the dominant arm were registered.The loading conditions were intended to change either orboth the quality and quantity of muscle activation patternsunderlying the different movement directions of the stardrawing. This allowed us to dissociate directional move-ment specifications from the underlying patterns ofmuscle activation because subjects were required toproduce the same kinematics while the underlying muscleactivation patterns were altered as a result of springloading. It was hypothesized that, if specifications ofmovement direction and amplitude are dissociable fromthe underlying patterns of muscle activation, the observedkinematics for the eight sequences of the star drawingwould remain largely the same across the various loadingconditions. More specifically, the question was addressedwhether the kinematic templates of a motor task areconserved across different dynamic requirements in themoving joints. In other words, is motor equivalenceevident in the control of goal-directed cyclical move-ments?

The present study also scrutinized how the CNSovercomes the biasing forces induced by spring loading.In general, motor equivalence designates the capability ofsubjects to accomplish a given motor task under varyingdynamic conditions. This adaptation could largely under-lie formation of internal models in the CNS (Thorough-man and Shadmehr 1999). In this respect, it washypothesized that manipulation of movement directionand direction of force in the end-effector generate uniquesets of muscle grouping to overcome the force require-ments while keeping the kinematics largely unchanged.More specifically, it was intended to unravel the under-lying EMG patterns associated with producing multijointcyclical drawing movements with different specificationsof movement direction and loading. Previous studies oncyclical drawing movements in monkeys have shownsystematic relations between hand trajectory and neuro-muscular activity that varied with the direction ofmovement (Moran and Schwartz 1999). Yet, there arevirtually no reports on the effect of external forcemanipulations on muscle grouping during cyclical tasks.The cyclical nature of the present movements provided aunique opportunity to study the underlying patterns ofphasing and coupling between shoulder and elbowmuscles. For this purpose, a cross-correlation functionwas used to estimate the relations and time offsetsbetween different combinations of elbow and shouldermuscles (Levin et al. 2001). The analysis was intended toassess whether the reorganization of muscle activation

63

(e.g., coupling and phasing between agonist and antag-onist pairs of the shoulder and elbow) is conserved acrossdifferent dynamic requirements in the moving joints.Accordingly, the present study sought to provide insightsinto the neural basis of the control of arm movements ingeneral and motor equivalence in particular.

Materials and methods

Subjects

Ten undergraduate students of Katholieke Universiteit Leuvenparticipated in the experiment. All subjects were right handed (fivemen and five women; mean age 21.5 years). They were notpreviously involved in a similar experiment and provided writtenconsent for participation. The experimental procedures werereviewed by the local ethical committee for biomedical research.

Apparatus and task

The apparatus consisted of an XY digitizing table (LC20-TDSTerminal Display Systems) positioned in the horizontal plane infront of the subject. The accuracy of registration was 0.25 mm.Kinematic data were digitized with respect to the X axis (parallel tothe transverse plane) and Y axis (parallel to the sagittal plane)component at a sampling frequency of 150 Hz. Subjects wereseated comfortably on a height-adjustable chair and held a stylus intheir right hand with the forearm and elbow positioned just abovethe surface of the digitizing tablet. Movements were restricted tothe shoulder and elbow joints. Wrist movements were prevented bymeans of a splint. The distance from the subject’s right shoulder(acromion process) to the center of the star figure along the Y axiswas set at 40 cm to allow subjects to produce the drawingmovements comfortably within the whole range of the workingspace. Subjects were instructed to trace a star figure imprinted onthe surface of the digitizer (template), as shown in Fig. 1a. Handmovements were always initiated with the tip of the styluspositioned at the center of the star. Subjects grasped the top halfof the stylus to support the arm against gravity. A computer-controlled electronic metronome indicated the cycling frequency(126 beats/min). Subjects were to draw a complete line (back andforth) with each pulse of the metronome, resulting in a duration of476 ms.

Subjects started off with drawing AP lines, but shiftedmovement direction with 45� after a series of five lines wascompleted within each of the eight sequences of movement, asindicated by an accent of the metronome beat. Within each trial, thesubjects completed two sets of AP, right diagonal (RD), mediolat-eral (ML), and left diagonal (LD) hand movements that werevisited in a clockwise manner. Accordingly, movement directionsof the end-effector were pair wise parallel to each other, i.e., thefirst and fifth sequences were the same, as well as the second andsixth sequences, etc. The resulting principal movement directionsof the hand are shown in Fig. 1b. Consequently, the task consistedof drawing five back and forth lines between opposite clockpositions within each of the eight sequences, i.e., the subject movedfrom one clock position to the opposite clock position, therebypassing through the center position during each movement cycle.The total duration of each trial was 21 s. The requested movementamplitudes were 18 cm peak to peak.

The star task was performed under different spring loadingconditions in which the restoring force depended on springelongation. The horizontal spring was attached to a fixed bar atone end and to a cuff, positioned around the right forearm, at theother end. The length of the spring was 1 m at rest, the springconstant was 0.130 N/mm (thickness 1.6 mm, diameter 9.4 mmheart-to-heart, DIN 17223–1.1200). Three loading protocols were

used: (1) no loading, (2) spring loading administered parallel to thefrontal plane (frontal loading), and (3) spring loading administeredparallel to the median sagittal plane (sagittal loading) (see Fig. 1c).When the stylus rested in the center position of the star drawing, theelongation of the spring was 30 cm (total length 130 cm) and therestoring force was approximately 39 N.

Surface EMG from the biceps (BIC), triceps (TRI), deltoidanterior (DA) and deltoid posterior (DP) of the dominant (right)limb were recorded by means of a four-channel surface electrodesystem (Noraxon Myosystem, 2000). EMG electrodes were placed2 cm apart, over the middle portion of the muscle belly, and alignedwith the longitudinal axis of the muscle. Preamplified signals (gain80 dB) from each channel were bandpass filtered (15–500 Hz) anddigitally sampled at 1000 Hz, parallel to the digitizer data.

Procedure

Practice and test trials were performed with full vision. Subjectswere instructed to trace the star figure as accurately as possible andto comply strictly with the temporal constraints of the metronome.Two practice trials preceded each of the two test trials saved forfurther analysis. Subjects performed six trials of star drawingmovements with their dominant limb – two unloaded trials and fourtrials in which spring loading was applied to the dominant limbalong the frontal (2) and sagittal (2) axes. This resulted in 200 linedrawing cycles performed for each direction across each of thethree loading conditions (10 subjects � 2 trials � 2 line drawingsequences along each principal movement direction � 5 drawingcycles per sequence). Following the “ready” command by theexperimenter, subjects started the movement, at which pacing ofthe metronome was initiated. No feedback on performance qualitywas given during data acquisition.

Data Analysis

Examination of representative digitizer and EMG traces (Fig. 2)revealed that muscle burst activity during the braking phase(deceleration of the end effector) is also used to initiate (accelerate)the movement of the limb in the opposite direction. Accordingly, itis impossible to discriminate within this single burst which portionis responsible for braking the action and which is responsible forreversing the movement. For that reason, analysis of kinematics andEMG data was done across the complete movement cycle without

Fig. 1 Schematic representation of the eight sequences in the stardrawing task shown here for a right handed subject (a), the fourprincipal movement directions (b), and the two spring loadingconditions (c)

64

further decomposition. Since subjects moved back and forthbetween pair-wise parallel movement directions of the star, theeight movement sequences in each trial were reduced to four levelsfor statistical purposes. EMG signals with slightly different valuesof integrated EMG (iEMG) and correlation scores for sequences 1-5, 2-6, 3-7 or 4-8 were observed occasionally across different testconditions, suggesting that subjects tended to alter overall muscleactivity and muscle synchrony within trials of the same movementcondition. It can be noticed, though, that variability of directionaltuning of muscle activity and muscle synchrony between these pair-wise parallel movement sequences was much lower than thechanges in directional tuning of muscles among the four principalmovement directions.

Kinematics: Waveforms from the digitizing table were dividedinto eight epochs of 2.38 s (476 ms�5 cycles) corresponding to theeight sequences of the star drawing. The turning points of subjects’drawing movements were determined by applying a peak-pickingalgorithm to the time course of ML hand displacement. From thesedata, the absolute orientation angle was calculated for eachmovement half cycle by determining the arctan of the ratio Dy/Dx. Here Dx denotes the difference between two subsequent MLturning points and Dy denotes the difference between twosubsequent AP turning points. Movement amplitude was calculatedby determining the square root of Dx2+Dy2. Orientation angle and

amplitude were assembled across the five drawing cycles of each ofthe eight epochs and averaged across trials with the samemovement direction and loading conditions. A 3�4 (Loading �Direction) ANOVA, with loading and movement direction asindependent variables, was applied for statistical analysis of theresults with respect to mean values and standard deviation (SD)scores of the orientation angle and amplitude. Loading consisted ofthree levels: unloaded, frontal loading, and sagittal loading.Direction consisted of four levels: AP, ML and the two diagonalmovements.

EMG data: Integrated EMG was calculated to characterize levelof muscle activity as a function of loading conditions andmovement directions. EMG signals were de-trended to compensatefor long-term drift, and DC levels were set to zero. Signals werethen divided into eight epochs of 2.38 s (476 ms�5 cycles)corresponding to the eight sequences of the star drawing. EMGsignals were rectified, and integration was performed on therectified signal over the time period of interest (2.38 s). The iEMGvalues were calculated for each of the eight epochs and averagedacross trials with the same movement direction and loadingconditions.

Cross correlations between EMG pairs were computed toquantify peak power magnitude of common myoelectric signals(designated by the peak correlation function) and to search for

Fig. 2 Displacements and EMG activity of the biceps, triceps andanterior and posterior deltoid muscles during unloaded (a), frontalloading (b), and sagittal loading (c) conditions with respect to thefour first sequences of a representative star drawing trial. Trajec-tories in the XY plane of the end effector are inserted at the top,

left-hand part of each figure. The dashed vertical bars indicate theonset time of movement or onset of transition between twoconsecutive movement directions. (ONSET movement initiation;AP, RD, ML, and LD designate the anteroposterior, right diagonal,mediolateral, and left diagonal hand displacements, respectively)

65

temporal offsets. Correlation analyses were conducted among thetested muscles to study timing and degree of coupling betweenburst activity in shoulder and elbow agonist and antagonist pairs.Cross correlations between agonist muscles were carried out, sinceboth the TRI (long head) and the BIC (short and long heads) arebiarticular muscles that act upon the elbow and shoulder. Prior tothe cross-correlation analysis, signals were full-wave rectified andband-pass filtered (zero-phase distortion, forward and backwarddigital filtering) by means of a second-order Butterworth filter witha cutoff frequency at 0.5 and 50 Hz (Grasso et al. 1998). The cross-correlation function between two myoelectric signals, xi(t) and xj(t),is given by (c.f., Winter et al. 1994):

RijðtÞ ¼

RT

0xiðtÞ � xjðt þ tÞdt

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiRT

0x2

i ðtÞdt �RT

0x2

j ðtÞdt

s ð1Þ

where T (2.38 s) is the length of the signals being cross correlatedand t is the time lag between the two signals. The latter was used toexpress the phasing of activity between the different muscles (notethat t „ 0 indicates a shift in time of the i’th waveform relative tothe j’th waveform). The numerator in Eq. 1 corresponds to thepower of the common signal in xi(t) and xj(t), and it is scaled to theproduct of total signal power (i.e., the autocovariance at time lag =0), the denominator in Eq. 1 (so that the cross-correlation functionranges from -1 to 1). A peak detection algorithm was used todetermine the highest positive correlation peak (R) and itscorresponding time lag (t). Note that the cross-correlation coeffi-cients are significantly different from zero (P<0.01) when they are>0.22, with n>100 degrees of freedom (Chatfield 1989). The cross-correlation coefficients were calculated for each of the eight epochsand averaged across trials with the same loading and directionconditions.

To test for the temporal shift in muscle burst activity betweenagonist and antagonist pairs, a distribution of time offsets wasplotted for each principal movement direction of the star, onlycontaining time lags associated with significant correlation coef-ficients. Data were grouped across subjects; each histogramspanned a single cycle duration and had 12 ms binwidth (476 msdivided into 40 bins). The number of elements (n) in eachhistogram corresponds to the number of correlation peaks thatdiffer significantly from zero (n£40). Lines were fitted to the pointsof the scatter plots of time offsets, using a locally weightedscatterplot smoother (LOWESS) curve fitting procedure.

To test for directional tuning effects on EMG characteristicsacross the tested loading conditions, the same 3�4 ANOVA wasused, with loading and movement direction as independentvariables. Since significant main effects of loading with respectto changes in EMG patterns were largely expected, they are notincluded in the statistical report. However, the significance of theLoading � Direction interaction is particularly important because itindicates that either the performance of the star drawing task orchanges in the patterns of muscle activation were not invariant withthe movement direction of the arm. In other words, this interactioncan indicate whether and to what extent the nature of directionaltuning in the upper limb was affected by loading. When significantLoading � direction effects were found, post hoc testing (i.e.,contrast analysis) was conducted to identify differential effects ofloading in relation to changes in directional tuning. Changes inEMG characteristics as function of movement direction withrespect to each loading condition will be discussed.

Modulation ratios: Changes in movement kinematics and EMGactivity as a function of loading with respect to each movementdirection were quantified. For this purpose, the modulations in lineamplitude, iEMG, and correlation scores during frontal and sagittalloading were expressed as a percentage of their unloaded valuesacross the four principal movement directions. These modulationratios were calculated for each of the eight epochs and averagedacross trials with the same loading and direction conditions. Forstatistical purposes, a 2�4 ANOVA model was used with loading

and movement direction as independent variables. Loading con-sisted of two levels (frontal and sagittal), whereas for movementdirection, the same four levels were used as before. Significanteffects were defined as those at P<0.05 probability level. Whensignificant Loading �Direction effects were found, post hoc testingwas conducted to identify the source of the differences.

Results

Example of EMG and kinematics for representative trials

Fig. 2 shows end-effector displacements and EMGactivity of the BIC, TRI, and DA and DP muscles duringAP, RD, ML, and LD first four sequences of a represen-tative right star drawing trial for unloaded (Fig. 2a),frontal (Fig. 2b) and sagittal (Fig. 2c) loading conditions.Plots of the corresponding end-effector trajectories in theXY plane are inserted at the top left-hand part of eachfigure. As can be observed, the global patterns ofmovements were preserved across loading conditions.Subjects were able to trace successfully the trajectory ofthe star figure throughout the different task conditions,suggesting that no systematic bias in the mean position ofthe shoulder and elbow occurred among trials withdifferent loading conditions. This hints at equivalence inperformance of the star task. As compared to the unloadedcondition (Fig. 2a), the amount as well as the patterns ofmuscle activity in the elbow and shoulder flexors andextensors changed dramatically during spring loading.These quantitative and qualitative changes in muscleactivity were a result of overcoming the spring’s restoringforce that pulled the limb in the lateral (frontal loading) orforward (sagittal loading) directions. Frontal loading(Fig. 2b) resulted in an increase of flexor activity (DAand BIC) and decrease of extensor activity (mainly in theDP). Sagittal loading (Fig. 2c) resulted in an increase ofextensor activity (DP and TRI) and decrease of flexoractivity (mainly in the DA) during sagittal loading. Themost apparent change was that frontal spring loadingresulted in an increase in BIC burst duration and peakactivity, while sagittal loading increased mainly theactivation of the DP. Overall, these observations demon-strate that spring loading affected the patterns of muscleactivation substantially while the patterns of the handend-effector kinematics were largely preserved.

Effect of loading on movement parameters

Orientation angle: Data with respect to mean orientationangles and orientation SD scores across the three loadingmanipulations and eight drawing sequences are shown inFig. 3. Overall, the results revealed that subjects were ableto comply successfully with the direction requirements ofthe star drawing task across all loading conditions,independent of the force requirements (as indicated inFig. 3a). The effect of spring loading was foundsignificant for mean orientation of the star (F(2,18)=11.90, P<0.005). Yet, changes in the mean orientation

66

angles among the three loading conditions were less than3� for all movement directions. The Loading � Directioninteraction was not significant (F(6,54)=1.67, P>0.05),suggesting that performance of the desired line orientationangles was invariant with respect to changes in thedirection of force. In other words, performance of thedesired movement direction in the star drawing task wasinvariant with respect to loading.

Results for variability (SD scores) of orientation anglesacross loading conditions are presented in Fig. 3b. Onlythe main effect of direction was found significant withrespect to orientation angle SD scores (F(3,27)=11.52,P<0.001). Orientation SD scores were smaller during APthan during the remaining movement directions. The maineffect of loading was not significant, nor was the Loading� Direction interaction (P>0.05), suggesting that vari-ability of the orientation angle across the four principalmovement directions was not affected by loading.

Amplitude: Data with respect to changes in meanamplitude across the three loading manipulations and thedifferent directions are presented in Fig. 4. The maineffects of both loading and direction were significant forthe mean amplitude of the movements: loadingF(2,18)=9.57, direction F(3,27)=20.62, both P<0.005(Fig. 4a). It can be observed that amplitudes were highestfor RD movements and lowest for LD movements, withthe remaining movement directions positioned in be-tween. This effect is in agreement with the phenomenonof inertial anisotropy. The Loading � Direction interac-tion was significant (F(6,54)=8.24, P<0.001), suggestingthat loading made a differential contribution to movementamplitude across the four principal movement conditions.The two diagonal movement directions demonstrated themost divergent amplitudes, irrespective of the loadingconditions, all P<0.01.

Results for variability (SD scores) of amplitudes acrossloading manipulations are presented in Fig. 4b. The maineffect of direction was significant for the amplitude SDscores (F(3,27)=10.08, P<0.001). The main effect ofloading was not significant, neither was the Loading �Direction interaction (P>0.05), suggesting that intrasub-ject variability of amplitude performance across drawingsequences was not affected by loading. Amplitude SDscores were smaller during AP direction than during theremaining directions.

Loading effects across directions

Data with respect to modulation of amplitude duringfrontal and sagittal loading (relative to unloaded values)across the four principal movement directions will bediscussed next. Results in Fig. 4c show that mean valuesacross all trials with the same loading conditions were£10% of the unloaded values for all direction conditions.

Fig. 4 Polar plot representation of the mean amplitudes of the linesin the star figure (a) and amplitude SD (b). The radius in the polarplot represents the mean amplitude, and the angle represents theeight sequences of the trial, referring to the different target(principal) movement directions that are to be produced during stardrawing. Data are presented across the three loading conditions.Changes in line amplitudes as function of loading across the fourprincipal movement directions (c). Bar plots of the amplitudemodulations are shown for frontal (black bars) and sagittal (graybars) loading

Fig. 3 Polar plot representation of the mean orientation angles ofthe lines in the star figure (a) and orientation SD (b). The radius inthe polar plot represents the actual drawing direction, and the anglerepresents the eight sequences of the trial, referring to the differenttarget (principal) movement directions that are to be producedduring star drawing

67

Relative to the unloaded trials, frontal loading wasshown to increase line amplitude in the ML and the twodiagonal movements but to decrease amplitude whensubjects performed AP movements. Significant differ-ences were observed between AP and the remainingdirections (all, P<0.01) and between ML and LDdirections (P<0.05), but not between the two diagonaldirections (P>0.05). Relative to the unloaded trials,sagittal loading was shown to increase line amplitudeacross all movement directions, with the highestmodulation ratio observed for the RD direction.Significant effects among direction conditions wereobserved between the two diagonal directions andbetween the RD and AP directions (all, P<0.01).Overall, these results suggest that subjects exploitedthe spring’s restoring force in order to regulate lineamplitude. This may have been obtained by theformation of new internal models of cyclical move-ments that embedded feed-forward signals for thedynamic (inertial) and elastic (spring) force compo-nents in the end effector.

Effect of loading and movement directionon modulation of muscle activity

The polar plots in Fig. 5 show results for mean iEMGsacross the three loading manipulations and eightdrawing sequences. It can be seen that the overalleffect of spring loading and movement direction on thetotal amount of muscle activity was more pronouncedin the BIC and DP than in the TRI and the DA. Duringfrontal loading, the overall increase in BIC and DAactivity was 126% and 20%, respectively. The decreasein TRI and DP activity was 13% and 50%, respectively.During sagittal loading, modulations in activity of flexorand extensor muscles were less specific with respectincreases/decreases in activation among pairs of antago-nistic muscles. The overall increase in TRI and DPactivity during sagittal loading was 43% and 100%,respectively. Interestingly, sagittal loading also resulted inan increase in BIC activity (15%). Decrease in DAactivity was in the order of 42%.

Directional tuning

Directional tuning effects for overall muscle activation inthe BIC, TRI, and AP and DP muscles across the threeloading manipulations and four principal directions willbe discussed next. Modulations of iEMGs among thedifferent movement directions were noticeable across allloading conditions, suggesting that muscles were moreactivated in a certain “preferred” movement directionthan in the remaining movement directions.

BIC: The Loading � Direction interaction was signif-icant (F(6,54)=9.50, P<0.001), suggesting that the decreaseor increase in BIC activation BIC among the differentmovement directions was largely associated with changes

in the direction of the spring’s restoring force. For theunloaded condition, the iEMGs were higher for the RDsequences (i.e., when the movement was accomplishedalmost entirely by rotation of the forearm about theelbow) and smaller for the LD sequences, with theremaining directions positioned in between. The differ-ence in mean iEMGs between RD and LD movementswas 22%. Relative to the unloaded condition, BICactivation was more than doubled during frontal loading,with the most prominent activation levels now observedfor the AP and RD movements. Differences in meaniEMG were 30% (RD versus LD), and 24% (AP versusLD). Relative to the unloaded condition, iEMGs werehigher for the RD sequences and smaller for the MLsequences during sagittal loading. Difference in meaniEMGs between RD and ML movements was 36% andbetween RD and LD movements was 32%.

DP: The Loading � Direction interaction was signif-icant for DP (F(6,54)=3.05, P<0.05). For the unloadedcondition, iEMGs were higher for the LD sequences (highinertial load) and smaller for the RD sequences (lowinertial load), with the remaining movement directionspositioned in between. Difference in mean iEMGsbetween RD and LD movements was 18%. This suggeststhat the DP activation level was largely associated with

Fig. 5 Polar plots representation of the integrated EMG (iEMG)levels during the star drawing task for the eight sequences of thehand. Data are presented across the three loading conditions. Theradius in the polar plot represents the mean iEMG levels, and theangle represents the eight sequences of the trial referring to thedifferent target (principal) movement directions that are to beproduced during star drawing

68

changes in the arm’s inertial anisotropy. Relative to theunloaded condition, activation in the DP was substantiallydecreased during frontal loading, with the preferreddirection now shifted from the LD to the AP direction,suggesting that subjects exploited the spring’s restoringforce to overcome inertial load. The difference in meaniEMGs between AP and ML movements was 37%, whilethe difference between the two diagonal movements wasless than 10%. Relative to the unloaded condition,activation in the DP was nearly doubled during sagittalloading. The iEMGs were higher for LD movements andsmaller for ML movements. Difference in mean iEMGsbetween LD and ML movement directions was 15%, andthe difference between the two diagonal movementdirections was 10%.

TRI and DA: The Loading � Direction interaction wasnot significant for the TRI and DA (P>0.05), suggestingthat loading manipulations had no significant effect ondirectional tuning in the activation levels of these musclesamong the different movement directions. In other words,the preferred direction for TRI and DA activation wasfound invariant across the three loading manipulations.For the TRI, the maximal iEMGs were observed for RDlines i.e., when the movement was accomplished almostentirely by rotation of the forearm about the elbow. Forthe anterior deltoid, maximal iEMGs were observed forLD lines, i.e., for movements with a high inertial load.

Loading effects across directions

We will next discuss the changes in overall muscleactivity (iEMG values), as a function of loading withrespect to each movement direction. Fig. 6 showsmodulation ratios of iEMGs as a function of frontal andsagittal loading across the four movement directions.Overall, the results indicated that loading made adifferential contribution to elbow muscle activation acrossthe four principal directions but not to shoulder muscleactivation. For the BIC, these changes mainly involvedhigher modulations in overall muscle activation duringthe AP direction as compared to those observed during theRD (P<0.001), ML (P<0.05) and LD (P<0.01) directionsduring frontal loading. Relative to the unloaded trials, anincrease in BIC activation during sagittal loading wasmore prominent with respect to the RD than with respectto the ML direction (P<0.05). No significant differenceswith respect to changes in modulations of iEMG amongthe four movement directions were observed for the TRIand AP and DP (P>0.05). This suggests that changes inthe activation level of shoulder muscles during frontal orsagittal loading were largely muscle but not directionspecific.

Effect of loading and movement direction on modulationof degree of muscle synchrony

The polar plots in Fig. 7 show the peak cross-correlationvalues across the three loading manipulations and eightdrawing sequences. Analyses were made for the mainmuscle groups in the experimental paradigm, includingantagonistic pairs, i.e., DA versus DP, and BIC versusTRI; and agonistic pairs, i.e., DA versus BIC, and DPversus TRI. We will next discuss changes in thecorrelation as a function of movement direction withrespect to each loading condition.

Directional tuning

Elbow antagonists (BIC-TRI): Relative to the unloadedcondition, we found only minor changes in the correlationscores between BIC and TRI during frontal and sagittalloading (Fig. 7a, left side). The Loading � Directioninteraction was not significant (F(6,54)=1.74, P>0.05),suggesting that loading manipulations did not significant-ly affect the coupling between these muscles as functionof movement direction.

Shoulder antagonists (DA-DP): Changes in correlationscores between the shoulder antagonists were mainly aconsequence of movement direction and were largelyassociated with changes in the direction of the dynamic(inertia) forces in the end effector (Fig. 7a, right side).The Loading � Direction interaction was significant

Fig. 6 Changes in muscle activation integrated EMG (iEMG) asfunction of loading across the four principal movement directions.Bar plots of the iEMG modulations are shown for frontal (blackbars) and sagittal (gray bars) loading

69

(F(6,54)=4.25, P<0.05), suggesting that loading manipula-tions did have a differential effect on the degree ofcoupling between the shoulder muscles as function ofdirection. For the unloaded and frontal loading conditions,the highest correlation scores were observed for theperformance of the LD lines, i.e., for movements with ahigh inertial load. Relative to unloaded and frontalloading, mean correlation scores for movements in theML direction were substantially increased during sagittalloading, with the preferred direction now shifted from LDto ML. Post hoc tests revealed that correlation scores weresignificantly increased when the ML lines were drawnduring sagittal loading (R=0.41), relative to the frontalloading (R=0.26, P<0.01) and no loading (R=0.28,P<0.05) conditions. Correlation scores in the remainingdirections were mildly affected by the different loadingconditions. For the unloaded condition, correlation scoresbetween DA and DP were larger during LD movementsand smaller during RD movements (0.36 versus 0.25,

P<0.01). For frontal loading, correlation scores werelarger during LD movements and smaller during MLmovements (0.32 versus 0.26, P<0.01). For sagittalloading, correlation scores were larger during ML move-ments and smaller during AP movements (0.41 versus0.28, P<0.01). Overall, the findings suggest that thedegree of synchrony between the antagonist actuators ofthe shoulder was directionally tuned with respect tochanges in loading conditions.

Flexors (DA-BIC): Changes in correlation scoresbetween DA and BIC among the four movement direc-tions were largely associated with changes in the upperlimb inertial properties only for the unloaded condition(Fig. 7b, left side). The Loading � Direction interactionwas significant with respect to cross correlation scoresbetween the shoulder and elbow flexors (F(6,54)=2.40,P<0.05). Correlation scores between the DA and BICwere significantly increased when the LD (high inertia)lines was drawn relative to the RD (low inertia) lines(0.37 versus 0.25, P<0.001). Significant differences incorrelation scores during the unloaded condition werealso observed between the RD and the remainingmovement directions (0.25 versus 0.34, P<0.01 in theAP, and 0.35, P<0.05 in the ML direction). During frontalloading, correlation scores varied significantly onlybetween RD and ML directions (0.25 versus 0.33,P<0.01). No significant variations in correlation scoresas a function of direction was observed during sagittalloading (P>0.05). Relative to the unloaded condition,coupling between DA and BIC was decreased duringfrontal loading when subjects performed AP (0.34 versus0.29, P<0.05) and LD (0.37 versus 0.26, P<0.01)movements. Overall, the findings suggest that the synergybetween flexor muscles of shoulder and elbow wasdirectionally tuned with respect to changes in loadingconditions.

Extensors (DP-TRI): Changes in the correlation scoresbetween DP and TRI among the four movement direc-tions were largely associated with movements in LD andML directions (Fig 7b, right side). The Loading �Direction interaction was significant with respect to crosscorrelation scores between shoulder and elbow extensors(F(6,54)=2.52, P<0.05). During the unloaded condition,highest correlation scores among the four movementdirections were observed for ML movements relative toAP and the two diagonal movements, all P<0.05. For theremaining loading conditions, changes in correlationscores across the four principal movement directionswere not significant (P>0.05). These findings suggest nochanges in the degree of synergy between the shoulderand elbow extensors across the four movement directionswith respect to loading.

Loading effects across directions

We will next discuss the changes in correlation scores as afunction of loading with respect to each movementdirection. Fig. 8 shows modulation ratios of correlation

Fig. 7 Polar plots representing mean values of the correlationscores between pairs of EMG data recorded from the shoulder andelbow muscles during the star drawing task. Correlation scoresbetween pairs of EMG data from antagonistic muscles (BIC-TRI,DA-DP) (a). Correlation scores between pairs of EMG data fromagonist muscles (DA-BIC, DP-TRI) (b). Results are shown for theeight sequences of the hand across the three loading conditions. Theradius in the polar plot represents mean correlation scores, and theangle represents the eight sequences of the trial, referring to thedifferent target (principal) movement directions that are to beproduced during star drawing. (DA anterior deltoid, DP posteriordeltoid, BIC biceps, TRI triceps)

70

scores as a function of frontal and sagittal loading acrossthe four movement directions. Relative to the unloadedtrials, frontal and sagittal loading generated no significantchanges in the correlation between BIC and TRI. On theother hand, loading made a differential contribution to thecoupling between the shoulder muscles (DA-DP) acrossthe four principal directions during sagittal but not frontalloading. These changes mainly involved an averageincrease of 50% in correlation scores during the MLdirection, as compared to a 20–30% decrease in correla-tion scores during the AP or LD directions (both P<0.01).No significant changes were found with respect tochanges in the modulation of correlation scores betweenthe two diagonal directions and between the RD and theremaining directions (P>0.05). For shoulder and elbowflexors (DA-BIC), these changes involved an averageincrease of 30% in correlation scores between the twodiagonal directions (P<0.05) during sagittal loading ascompared to relatively small modulations (<10%) for theremaining directions, relative to the unloaded condition.Relative to the unloaded condition, frontal loading largelyinvolved a decrease in correlation scores for the LDdirection, which was modulated stronger than RD or MLdirections (both P<0.05). Relative to the unloadedcondition, changes in correlation scores during sagittalloading between DP and TRI involved an averageincrease of 50% for the LD directions as compared torelatively small modulations (<15%) in the AP and MLdirections (both P<0.05), or the RD direction (P>0.05).Relative to the unloaded condition, frontal loading largelyinvolved a decrease in correlation scores between theflexors of shoulder and elbow. Changes mainly involvedslightly higher but nonsignificant modulations (P>0.05)during the ML direction as compared to those observed inthe remaining directions.

Effect of loading and movement direction on modulationof time offsets

We will next discuss the effect of loading on the timeoffset modulations of the correlation peaks as function ofmovement direction with respect to each loading condi-tion (Fig. 9).

Elbow antagonists (BIC-TRI): The Loading � Direc-tion interaction was not significant with respect to timeoffset modulation between the elbow antagonist pair(P>0.05). Nevertheless, spring loading was shown toinduce qualitative changes in time offsets (P<0.001).Relative to the unloaded condition, frontal and sagittalloading were shown to increase coactivation of the elbowmuscles, with a prominent inclination toward the left ofthe time offsets histograms observed across all fourprincipal movement directions.

Shoulder antagonists (DA-DP): Changes in temporaloffsets of burst activity among various directions andloading conditions were prominent, with a significantLoading � Direction interaction (P<0.001). Offsets weresignificantly shorter in RD movements relative to ML

movements during frontal loading (142 ms versus 241 ms)and relative to AP movements during sagittal loading(97 ms versus 213 ms), all P<0.05. Offsets did not differamong the four movement directions during the unloadedcondition (mean 188 ms). For ML movements, offsetswere significantly decreased during sagittal loading(97 ms) relative to the unloaded (164 ms, P<0.05) andfrontal loading condition (142 ms, P<0.01). In summary,loading was shown to alter burst activity of the shouldermuscles. Alterations were largely associated with changesin the line of action of the spring forces. Sagittal loadingwas shown to increase shoulder muscle coactivation, withthe most prominent inclination toward the left of the timeoffsets histograms observed during ML movements. Inaddition, shoulder muscle coactivation was also evidentwhen elbow movements were involved (i.e., arm move-ment in the RD direction).

Flexors (DA-BIC): Modulations in the temporal offsetsresulted in a significant Loading � Direction interaction(P<0.001). Relative to RD movements, offsets weresignificantly decreased for AP movements during theunloaded condition (96 ms versus 204 ms) and for MLmovements during frontal loading (86 ms versus 204 ms),all P<0.01. The remaining effects of either or bothmovement direction and loading were not significant.Overall, frontal loading was shown to enhance coactiva-tion between DA and BIC during ML movements and atthe same time to increase temporal shifts between burstsduring AP movements. On the other hand, sagittal loading

Fig. 8 Changes in correlation scores as function of loading acrossthe four principal movement directions. Bar plots of the modula-tions are shown for frontal (black bars) and sagittal (gray bars)loading

71

72

did not result in significant changes in temporal shiftsbetween bursts.

Extensors (DP-TRI): Changes in the mean temporaloffset of burst activity between DP and TRI among thevarious movement direction and loading conditions werenot significant (P>0.05). Overall, the time offsets dem-onstrated a tendency toward coactivation of the shoulderand elbow extensor muscles that was independent ofdirection and loading manipulations (mean 97 ms).

Discussion

We studied the effect of spring loading on end-effectorkinematics and EMG activity of shoulder and elbowmuscles during performance of star drawing movementsin the horizontal plane. Two major findings emerged fromthis work and will be discussed in the subsequentsections.

1. Comparing the spring-loaded to the unloaded trials,the amount as well as the pattern of muscle activity inboth shoulder and elbow muscles changed dramatically,while performance at the kinematic level remained nearlyinvariant. This observation demonstrates motor equiva-lence in the task performance and suggests that subjectswere able to dissociate the actual movement direction(resulting from net torque) from its underlying pattern ofmuscle activation (muscle torque).

2. Spring loading induced general changes in theoverall amount of EMG activity, which are largely musclebut not direction specific and, therefore, are presumed torepresent the posture-dependent biasing force of thespring. However, in addition, we found qualitativechanges in muscle grouping, presumably reflecting theinfluence of dynamic force components. The mostapparent one was that loading increased coactivation ofthe elbow muscles. This indicates that the subjects tendedto increase stiffness in the elbow to compensate forchanges in spring bias forces in order to minimizetrajectory errors. On the other hand, shoulder musclespredominantly showed antiphase EMG bursts that werelargely invariant across the four principal directionsduring unloading and frontal loading but not duringsagittal loading. Coactivation between the prime moversof the shoulder that also involved a shift towardsantiphase burst activity between the shoulder and elbowsflexors, was observed for movements in the ML direction

during sagittal loading. This indicates that subjects tendedto shift between two or more muscle grouping strategiesin order to maintain the kinematic requirements of thetask under various dynamic constraints.

Dissociation between spatial and force specifications:evidence for motor equivalence

The bias force induced by spring loading was shown tomodulate the magnitude as well as the patterns of muscleactivation in the elbow and shoulder joints, whereasinvariance was largely maintained at the kinematic level.With respect to the mean orientation angle, changesamong the three loading conditions were less than 3�across all movement directions. Amplitude performancewas slightly less stable, as subjects tended to overshootthe required amplitude during the loaded conditions.Differences were more prevalent in RD and ML move-ments (10%) than in LD and AP movements (<5%).Furthermore, it was observed that orientation angle andamplitude SD scores were susceptible to changes inmovement direction but not to changes in loading. Thelowest SD scores for either orientation angle (Fig. 3b) oramplitude (Fig. 4b) were associated with movements inthe AP direction, and values significantly increased whensubjects produced RD or ML movements. In other words,subjects were able to produce robust tracking movementswhen AP lines were drawn, and they tended to showincreased instability during the remaining movementdirections, independent of loading condition. During theunloaded condition, however, the discrepancy betweendesired and actual amplitudes was consistently associatedwith the inertial anisotropy of the arm: Movements in thehigh inertia (LD) direction were generally undershot andwere smaller in magnitude than those in the low inertia(RD) direction.

Overall, the findings support the idea that handtrajectory is processed in the CNS on the basis of anabstract representation of movement direction withouttaking into account the specific dynamic effects of themotion (Gordon et al. 1994; Karst and Hasan 1991a,1991b). These findings are also consistent with thegeneral view proposed by Ghez and his collaborators,that subjects do not fully compensate for direction-dependent differences in the arm’s inertial load (Gordonet al. 1994; Ghez et al. 1997). Preservation of kinematicsis also impressive in view of the fact that virtually nopractice was allowed. In this respect, subjects were able toadjust motor commands within different dynamic envi-ronments, but at the same time to maintain a largelyinvariant representation of end-effector movements. Thefindings provide indirect but nevertheless strong evidencethat encoding of movement direction in the star drawingtask is dissociated from commands to specific muscles(Wing, 2000; Swinnen et al. 2001).

The aforementioned findings are consistent withsubstantial evidence from single-cell recording studiesin animals, showing that direction is a primary movement

Fig. 9 Histogram, showing the distribution of the time delays ofthe cross-correlation peaks for the principal movement directions ofthe hand during the star-drawing task. Distribution of time delaysfor correlations between pairs of EMG data from antagonisticmuscles (BIC-TRI, DA-DP) (a). Distribution of time delays forcorrelations between pairs of EMG data from agonist muscles (DA-BIC, DP-TRI) (b). Results are shown for the four principalmovement directions across the three loading conditions. A grayarrow pointing to the left indicates a tendency toward coactivation.A gray arrow pointing to the right indicates a tendency towardantiphase EMG bursts. (DA anterior deltoid, DP posterior deltoid,BIC biceps, TRI triceps)

73

parameter coded in various brain structures during theproduction of goal-directed arm movements (Kakei et al.1999; Kakei et al. 2001; Moran and Schwartz 1999;Schwartz and Moran 1999; Scott and Kalaska 19971997).For example, Schwartz and Moran (1999) argued that theprocessing of drawing movements in the CNS is carriedout in segments for which the central representation of thekinematics would be kept invariant. In addition, recentstudies revealed a functional classification among primarymotor cortex neurons, which represented either force ordirection. These neurons either changed their activity withrespect to direction of the movement in space, indepen-dent of patterns of muscle activity, or exclusivelymirrored the underlying patterns of muscle activation(Scott and Kalaska 1997; Kakei et al. 1999).

Rescaling of overall muscle activation as a resultof loading

Spring loading the arm resulted in considerable changesin the underlying activation of elbow and shouldermuscles. As expected, differences in overall muscleactivation across loading conditions were quantitativelydramatic (up to 150%). During frontal loading, increasedflexor activity was associated with decreased extensoractivity in both the shoulder and elbow joints. Conversely,during sagittal loading, decreased activity of the elbowand shoulder flexors was accompanied by increasedactivity of their respective antagonists. These findingsindicate a strong synergy between agonistic flexors andextensors of the shoulder and elbow. Overall, theseobservations demonstrate that spring loading inducedquantitative as well as qualitative changes in muscleactivation patterns.

It should be mentioned, though, that adjustment ofjoint torque could not be achieved simply by rescaling theoverall amount of muscle activity. This is (in part)because the torque generation capacity of the musclethroughout the joint range of motion mainly depends onthe ratio between muscle length and muscle moment arm(Lieber 1992; Zajac 1989). In this respect, control of theongoing arm movement might well require closed-loopregulation of muscle activity on the basis of changes injoint angles and muscle lengths. The results also suggestthat the CNS may use different internal models to regulatethe dynamic (inertial) and elastic (spring) force compo-nents in the end effector. This might (partly) be theconsequence of scaling muscle forces in advance to meetkinematic and dynamic requirements of the end-effectormovements, or of changing the movement time profile, asachieved by changing the phasing between the primemovers of the joints.

Phasing and coupling between muscles withinand across joints

Cross-correlation analysis of EMG signals from antago-nistic muscle pairs showed that shoulder and elbowcontributed differently to movement of the upper limb. Itshould be mentioned, though, that the generation ofvoluntary movements in humans typically involves con-trol of a redundant musculoskeletal system with multiplemuscle combinations. This redundancy allows the CNS toadopt various control strategies when complying with thetask-specific kinetic requirements. In this respect, corre-lation feature modulations between surface EMG wave-forms of different muscles most likely reflect ongoingchanges in the interaction among posture-dependent,dynamic (inertial) and elastic (spring) torque require-ments in the joints. The present observations largelyextended previous observations on task-dependent muscleactivity regulation to adjust for intersegmental interac-tions while preserving invariant kinematics under differ-ent specifications of torque requirements in the joints(Gribble and Ostry 1998; Sainburg et al. 1999; Takahashiet al. 2001).

Patterning between antagonist shoulder muscles waspredominantly susceptible to changes in movementdirection but was also modulated by loading. Forexample, we observed an increase in coupling betweenantagonist muscles of the shoulder (i.e., DA and DP) forRD and ML movements during sagittal loading relative tounloading or frontal loading. This increase was accom-panied by a substantial coactivation of these muscles(Fig 9a). On the other hand, the increased coactivationbetween antagonistic elbow muscles during frontal andsagittal loading relative to the unloaded condition did notinteract with changes in movement direction. Thisindicates that subjects tended to increase stiffness in theelbow to compensate for changes in spring bias forces inorder to minimize trajectory errors, while shouldermuscles were largely used to produce movement in aspecific direction.

Interestingly, activity in the shoulder flexor waspredominantly in antiphase with that of the elbow flexor,with a two-fold increase in time offset between DA andBIC for ML movement during sagittal loading relative tothe frontal loading and unloaded condition (Fig. 9b). It isimportant in this respect to notice that the BIC long andshort heads span both the elbow and shoulder joint (Woodet al. 1989). As such, coactivation of the shoulder andelbow agonistic muscles perhaps suggests that simulta-neous recruitment of these muscles served (in part) tomanipulate motion of the upper limb but also to correctfor unintentional extension in the elbow caused by eitheror both shoulder motion and spring restoring forces (MLmovements during frontal loading). In other words,subjects tended to increase shoulder stiffness by coacti-vating monoarticular shoulder muscles whereas regula-tion of torque requirements at the shoulder was largelymaintained by activating the BIC for shoulder flexion andby exploiting restoring spring force for shoulder exten-

74

sion. Shoulder flexor activity during the remainingloading and direction conditions was predominantly inantiphase with that of shoulder extensors and wastherefore assumed directly responsible for generatingthe hand end-effector trajectory.

Simultaneous agonist muscle activity hinted thatcommon central descending commands may play a rolein muscle activity coordination when manipulating mus-cle torques about these joints (Grasso et al. 1998). Thismay become evident as an increase of common temporalordering of descending commands when a large numberof muscles must be controlled simultaneously to producecoordinated motion in individual joints (Weijs et al.1999). Our findings supported this hypothesis for armmovement in the ML direction in which temporal patternsof muscle activity in the shoulder and elbow (i.e., betweenDP and TRI) showed the highest correlation scores (Fig. 7)and the lowest time offsets (Fig. 9). Modulation incorrelation scores but not time offsets across differentloading conditions between DP and TRI presumably hintsthat regulation of muscle tension to meet joint torquerequirement was mainly accomplished by facilitating thesynergy between agonist muscle groups. This controlparadigm bears similarities with the concept of linearsynergy proposed by Gottlieb and colleagues (1997) tomodel dynamics in the shoulder and elbow duringreaching movements. More specifically, they suggestedthat the CNS uses a single command to produce coupledmuscle contraction in the shoulder and elbow that leads totorque of a similar shape but scaled in amplitude to thedynamical demands of the task in each joint (Gottlieb etal. 1997; Osu and Gomi 1999).

Conclusions

The influence of loading on cyclical arm movementsacross different movement directions was studied toreveal changes in spatiotemporal features of the arm end-effector trajectory, levels of muscle activation, andmuscle grouping. One major outcome emerging fromthis study is that kinematic goals were dissociated fromtheir underlying muscle activation patterns. Subjectsproduced required movement direction and amplitudesuccessfully, even though bias forces induced by springloading were shown to dramatically modulate bothmagnitude and patterning of underlying muscle activa-tions. The aforementioned observations demonstrate’equivalence’ in human motion control and strengthenthe notion about independent encoding of direction andforce in the CNS. The results also suggest that the CNSincreases the limb’s stiffness to reduce the effect of springbias forces while rescaling activity and phasing betweenjoint prime movers to regulate dynamic (inertial) andelastic (spring) force components in the end-effector. Thiswas mainly obtained by increasing elbow muscle coac-tivation during loading and by modulations in patterningbetween shoulder muscles to cope with changes in bothloading and direction. Taken together, these observations

strengthen the notion that different muscle groupings canstill result in the same end-effector trajectory simply byadopting new coordination strategies among agonistic andantagonist muscles. This suggests that motor equivalencecan be achieved by rescaling both magnitude andpatterning of underlying muscle activations in the shoul-der and elbow.

Acknowledgments Support for this study was provided through agrant from the Research Council of K.U. Leuven, Belgium(Contract number OT/99/39) and the Flanders Fund for ScientificResearch (Project G.0285.98). Dr. Oron Levin was supported by afellowship from the Research Council of K.U. Leuven (Contract #F99/113). Dr. Nicole Wenderoth was supported by a grant from theResearch Council of K.U. Leuven, Belgium (Contract number OT/99/39).

References

Bernstein N (1967) The coordination and regulation of movements.Pergamon Press, Oxford

Chatfield C (1989) The analysis of time-series: An introduction.Chapman and Hall, London

Favilla M, Hening W, Ghez C (1989) Trajectory control in targetedforce impulses. VI. Independent specification of responseamplitude and direction. Exp. Brain Res 75: 280–294

Flash T, Hogan N (1985) The coordination of arm movement: Anexperimentally confirmed mathematical model. J Neurosci 5:1688–1703

Georgopoulos AP (1991) Current issues in directional motorcontrol. Trends Neurosci 18: 506–510

Georgopoulos AP, Ashe J, Smyrnis N, Taira M (1992) The motorcortex and encoding of force. Science 256: 1692–1695

Ghez C, Favilla M, Ghilardi MF, Gordon J, Bermejo R, Pullman S(1997) Discrete and continuous planning of hand movementsand isometric force trajectories. Exp Brain Res 115: 217–233

Gordon J, Ghilardi MF, Cooper SE, Ghez C (1994) Accuracy ofplanar reaching movements II. Systematic extent errors result-ing from inertial anisotropy. Exp Brain Res 99: 112–130

Gottlieb GL, Song Q, Almeida GL, Hong D-A, Corcos D (1997)Directional control of planar human arm movement. J Neuro-physiol 78: 2985–2998

Grasso R, Bianchi L, Lacquaniti F (1998) Motor patterns of humangait: backward versus forward locomotion. J Neurophysiol 80:1868–1885

Gribble PL, Ostry DJ (1998) Independent coactivation of shoulderand elbow muscles. Exp Brain Res 123: 355–360

Kakei S, Hoffman DS, Strick PL (1999) Muscle and movementrepresentation in the primary motor cortex. Science 285: 2136–2139

Kakei S, Hoffman DS, Strick PL (2001) Direction of action isrepresented in the ventral premotor cortex. Nature Neurosci 4:1020–1025

Karst GM, Hasan Z (1991a) Initiation rules for planar, two-jointarm movements: agonist selection for movements throughoutthe work space. J Neurophysiol 66: 1579 – 1593

Karst GM, Hasan Z (1991b) Timing and magnitude of electro-myographic activity for two-joint movements in differentdirections. J Neurophysiol 66: 1594 – 1604

Kelso JAS, Tuller B (1983) Compensatory articulation underconditions of reduced afferent information: A dynamic formu-lation. J Speech and Hearing Res 26: 217–224

Kelso JAS, Tuller B, Vatikiotis-Bateson E, Fowler C (1984)Functionally specific articulatory cooperation following jawperturbations during speech: Evidence for coordinative struc-tures. J Exp Psychology: Human perception and performance10: 812–832

75

Lacquaniti F (1989) Central representations of human limbmovements as revealed by studies on drawing and handwriting.Trends Neurosci 12: 287–291

Lacquaniti F, Soechting JF, Terzuolo C (1982) Some factorspertinent to the organization and control of arm movements.Brain Res 252: 394–397

Lashley K (1930) Basic neural mechanism in behavior. PsycholRev 37: 1–24

Levin O, Ouamer M, Steyvers M, Swinnen SP (2001) Directionaltuning effects during cyclical two-joint arm movements in thehorizontal plane. Exp Brain Res 141: 471–484

Lieber RL (1992) Skeletal muscle structure and function: Impli-cations for rehabilitation and sport medicine. Williams &Wilkins, Baltimore, pp. 129–130

Moran D, Schwartz AB (1999) Motor cortical activity duringdrawing movements: population representation during spiraltracing. J Neurophysiol 82: 2693–2704

Osu R, Gomi H (1999) Multijoint Muscle regulation mechanismexamined by measured human arm stiffness and EMG signals.J Neurophysiol 81: 1458–1468

Rijntjes M, Dettmers C, Buchel C, Kieble S, Frackowiak J,Cornelius W (1999) A blueprint for movement: Functional andanatomical representations in the human motor system. J Neu-rosci 15: 8043–8048

Sainburg RL, Ghez C, Kalakanis D (1999) Intersegmental dynam-ics are controlled by sequential anticipatory error correction,and postural mechanisms. J Neurophysiol 81: 1045–1056

Scheidt RA, Reinkensmeyer DJ, Conditt MA, Rymer Z, Mussa-Ivaldi A (2000) Persistence of motor adaptation duringconstrained, multi-joint, arm movements. J Neurophysiol 84:853–862

Schmidt RA (1988) Motor control and learning. A behavioralemphasis, 2nd edn. Human Kinetics, Champaign

Schwartz AB, Moran DW (1999) Motor cortical activity duringdrawing movements: population representation during lemnis-cate tracing. J Neurophysiol 82: 2705–2718

Scott SH, Kalaska JF (1997) Reaching movements with similarhand paths but different arm orientations. I. Activity ofindividual cells in motor cortex. J Neurophysiol 77: 826–852

Shadmehr R, Mussa-Ivaldi FA (1994) Adaptive representation ofdynamics during learning of a motor task. J Neurosci 14: 3208–3224

Swinnen S (1997) Motor control. In: Dulbecco R (ed) Encyclopediaof human biology, 2nd edn, vol 5, Academic Press, San Diego:pp 767–783

Swinnen SP Dounskaia N, Levin O, Duysens J (2001) Constraintsduring bimanual coordination: the role of direction in relationto amplitude and force requirements. Behav Brain Res 123:201–218

Takahashi CD, Scheidt RA, Reinkensmeyer DJ (2001) Impedancecontrol and internal model formation when reaching inrandomly varying dynamical environment. J Neurophysiol 86:1047–1051

Thoroughman KA, Shadmehr R (1999) Electromyography corre-lates of learning an internal model of reaching movements.J Neurosci 19: 8573–8588

Weijs WA, Sugimura T, van Ruijven LJ (1999) Motor coordinationin a multi-muscle system as revealed by principle componentsanalysis of electromyographic variation. Exp. Brain Res 127:233–243

Wing AM (2000) Mechanisms of motor equivalence in handwrit-ing. Current Biol 10: R245-R248

Winter DA, Fuglevand AJ, Archer SE (1994) Crosstalk in surfaceelectromyography: theoretical and practical elements. J Elec-tromyography and Kinesiology 4: 15–26

Wood JE, Meek SG, Jacobsen SC (1989) Quantitation of humanshoulder anatomy for prosthetic arm control. II. Anatomymatrices. J Biomechanics 22: 309–325

Zajac FE (1989) Muscle and tendon: properties, models, scalingand application to biomechanics and motor control. CRC CritRev Biomed Eng 17: 359–411

76