Greater reliance on impedance control in the nondominant arm ...

1 downloads 0 Views 534KB Size Report
Christopher N. Schabowsky ф Joseph M. Hidler ф. Peter S. Lum. Received: 30 November 2006 .... (Volpe et al. 2002; Hidler et al. 2005; Lum et al. 2002) and.
Exp Brain Res DOI 10.1007/s00221-007-1017-x

RESEARCH ARTICLE

Greater reliance on impedance control in the nondominant arm compared with the dominant arm when adapting to a novel dynamic environment Christopher N. Schabowsky Æ Joseph M. Hidler Æ Peter S. Lum

Received: 30 November 2006 / Accepted: 1 June 2007  Springer-Verlag 2007

Abstract This study investigated differences in adaptation to a novel dynamic environment between the dominant and nondominant arms in 16 naive, right-handed, neurologically intact subjects. Subjects held onto the handle of a robotic manipulandum and executed reaching movements within a horizontal plane following a pseudo-random sequence of targets. Curl field perturbations were imposed by the robot motors, and we compared the rate and quality of adaptation between dominant and nondominant arms. During the early phase of the adaptation time course, the rate of motor adaptation between both arms was similar, but the mean peak and figural error of the nondominant arm were significantly smaller than those of the dominant arm. Also, the nondominant limb’s aftereffects were significantly smaller than in the dominant arm. Thus, the controller of the nondominant limb appears to have relied on impedance control to a greater degree than the dominant limb when adapting to a novel dynamic environment. The results of this study imply that there are differences in dynamic adaptation between an individual’s two arms. Keywords Motor control  Motor adaptation  Handedness  Impedance control  Reaching movements

C. N. Schabowsky  J. M. Hidler  P. S. Lum Center for Applied Biomechanics and Rehabilitation Research (CABRR), National Rehabilitation Hospital, 102 Irving Street, NW, Washington, DC 20010, USA C. N. Schabowsky  J. M. Hidler  P. S. Lum (&) Department of Biomedical Engineering, The Catholic University of America, Pangborn Hall, room 131, 620 Michigan Ave., NE, Washington, DC 20064, USA e-mail: [email protected]

Introduction Motor adaptation studies have focused on the use of robotics to examine the ability to adapt to novel dynamic environments. In this setting, subjects perform reaching tasks in force fields applied to the hand via a robotic manipulandum. These forces alter the dynamics of a reaching task, greatly distorting previously straight hand trajectories. With practice, the central nervous system learns to compensate for the perturbing forces by altering the feedforward commands associated with the movements. These feedforward commands are the result of an internal model of the force field that develops during repeated exposure to the novel environment (Shadmehr and Mussa-Ivaldi 1994; Conditt et al. 1997). This is achieved by recalling the perturbation strengths and resulting hand-path errors of the preceding reaching movement(s) to update the internal model and eventually straighten subsequent reaching movements (Thoroughman and Shadmehr 2000; Scheidt et al. 2001). Though impedance control strategies (stiffening the arm) could explain this phenomenon, the existence of aftereffects supports the idea of internal models. An aftereffect occurs when the perturbing field is unexpectedly removed during a ‘‘catch trial.’’ The hand path of an aftereffect tends to resemble a mirror image of the distorted hand path caused by initial force field exposure. This implies that subjects are actively predicting and compensating for the novel environment rather than simply stiffening their arm (Flanagan and Wing 1997). It is also important to note that as the internal model develops aftereffects become larger (Conditt et al. 1997). Thus, practice leads to improved adaptation to the novel force field. Currently, the process of memory formation is under debate. Past work claims that motor adaptation occurs in

123

Exp Brain Res

two stages. Initially, internal models are developed in fragile form as short-term memory and a recently developed internal model can interfere with the development of a second internal model if the two models are learned within a short time interval (approximately 5 h). Following consolidation (maturation of the memory into a long-lasting stable form), the internal model can be recalled up to 5 months after training (Brashers-Krug et al. 1996; Shadmehr and Brashers-Krug 1997). A more recent study challenges the consolidation theory and supports the idea that memories can shift between active and inactive states (Caithness et al. 2004). Not only did this study fail to show independent learning of opposing force fields separated by up to a week, but also it showed that an extended session of null trials can eliminate previous learning. It is important to consider these findings when designing protocols that investigate differences in motor adaptation strategies of both arms. Adequate time must be allowed between test sessions to limit anterograde facilitation or interference and a set of null trials may abolish previous learning. The majority of motor adaptation studies have focused on the dominant arm. However, recent handedness investigations have shown that there are differences in control strategies between the dominant and nondominant arm leading to a hypothesis that these arms may have different capabilities (Sainburg and Kalakanis 2000; Sainburg 2002). The dynamic dominance hypothesis states that the factor that distinguishes dominant from nondominant arm performance is the ability to control limb dynamics (Sainburg 2002). Where the dominant arm is more capable of skillful reaching tasks, the nondominant arm is adept in posturing (Sainburg and Kalakanis 2000). If this hypothesis is correct, one would expect different control strategies for the two arms when adapting to novel dynamic environments, with a greater reliance on impedance control in the nondominant limb. Motor adaptation studies concerning both arm controllers have focused on interlimb transfer, the phenomenon that training in one limb affects the subsequent performance of the other arm (Malfait and Ostry 2004). For instance, reaching tasks within a velocity dependent force field (Criscimagna-Hemminger et al. 2003) and within altered inertial dynamics (Wang and Sainburg 2004a, b) both showed interlimb transfer from dominant to nondominant limb, but not vice versa. However, the method of interlimb transfer seems to depend on the types of forces applied to the arm. Interlimb transfer following adaptation to different inertial loads occurred within instrinsic (jointbased) coordinate system. Whereas, interlimb transfer following the velocity dependent force field protocol occurred within an extrinsic (global) coordinate system. In the velocity dependent force field experiment, transfer was only evident after exposing the nondominant limb to

123

identical forces as the previously trained dominant limb. Therefore, when adapting to a velocity dependent force field, transfer effects from one arm to the other can be minimized by using mirror image curl fields. Though there is evidence of interlimb transfer in adaptation to dynamic environments, there has been little focus on the possible differences in adaptive abilities between dominant and nondominant arm controllers. Recent studies of interlimb differences in adaptation to novel inertial dynamics have shown different adaptive mechanisms. During adaptation, the dominant arm controller adapted slightly slower, but more completely (larger aftereffects) than its nondominant counterpart (Sainburg 2002; Duff and Sainburg 2006). This was attributed to the fact that the dominant arm is more capable of coordinating shoulder and elbow interaction torques. This study aims to further investigate the differences in adaptation strategies between the dominant and nondominant arm controllers in naı¨ve subjects by comparing performance in a velocity dependent force field. There is growing interest in the use of novel dynamic environments to study the adaptation ability of subjects with neurological injuries such as stroke (Takahashi and Reinkensmeyer 2003; Patton et al. 2006; Scheidt and Stoeckmann 2007). Since a unilateral stroke can greatly impair the contralateral limb, stroke victims heavily rely on their less affected arm, regardless of dominance. Currently, clinical rehabilitation of both more-affected and less-affected limbs neglect motor lateralization and the different control strategies between the dominant and nondominant arm. Further understanding the differences between dominant and nondominant arm controllers may have important implications on strategies for motor rehabilitation of both arms after stroke. For example, it has been hypothesized that training the nondominant arm controller to act as a dominant arm controller will require specific training paradigms such as adapting to novel dynamic conditions (Sainburg and Duff 2006). Robot-assisted therapy has shown therapeutic potential (Volpe et al. 2002; Hidler et al. 2005; Lum et al. 2002) and is capable of creating such novel dynamic environments. Therefore, the data collected in this study would serve as a baseline for development of therapeutic dynamic environments for retraining of arm function after stroke.

Materials and methods Experimental setup Sixteen right-handed, neurologically intact subjects, aged 19–45 (mean = 23 ± 6), participated in this experiment. Handedness was assessed with the ten-item Edinburgh inventory (Oldfield 1971). Only subjects that received a

Exp Brain Res

laterality quotient of 80% or greater were admitted into this study. All subjects signed an informed consent form prior to admission to the study. All protocols were approved by the National Rehabilitation Hospital’s Internal Review Board for Protection of Human Subjects. Both arms of all subjects were tested and none of the subjects were previously exposed to robotic interactions of the type used in this study. This study used the planar, 2-degree of freedom robot designed at Massachusetts Institute of Technology and made commercially available by Interactive Motion Technologies, Inc. (InMotion2, Cambridge, MA). Reaching movements occurred within a 70 · 40 cm workspace. As an added safety measure, LED sensors (World Beam Q12 series, Banner Engineering Corp.) were positioned about the perimeter of the workspace. If the handle of the robot moved outside of the designated workspace, the sensors would trigger an emergency stop and turn off the motors. Custom programs (Matlab 7.1, XPCtarget 2.8; The MathWorks Inc., Natick, MA) were used to control the robot. Also, an inverse dynamics algorithm was used to partially compensate for the inertia of the robot links. This program dramatically decreased the intrinsic anisotropy of the robot. The forces applied to the hand by the robot were measured with a force sensor during the movements used in this study. With inertial compensation, the peak resistance force imposed by the robot along the direction of movement was reduced from 13.5 to 6.5 N. Importantly, the range of peak resistance forces across the four tested movement directions was reduced from 9 to 3 N, and lateral forces never exceeded ±1 N. Subjects seated in an adjustable chair held onto the endeffecter (handle) of the robot and executed reaching movements within a horizontal plane. For comfort, the handle was positioned slightly below shoulder height. All participants were fitted with a shoulder harness to minimize torso movement. A splint supported the subject’s forearm and restricted wrist rotation. Each subject was positioned with the shoulder in 60 degrees of horizontal adduction (ºS) and the elbow flexed 60 (ºE) as the subject held the handle in the start position. Subjects performed reaching movements while viewing an eye level computer screen and could see their arm throughout this experiment. Under some conditions, the robot created a velocity dependent force field, or curl field. To produce this field, the robot applied forces (Fx, Fy) to the handle that were orthogonal to its velocity (vx, vy) (Eq. 1). To limit interlimb transfer and facilitation/interference of adaptation, we tested the two arms on separate days, used mirror-image curl fields in the two arms, and an extended null field set was completed before exposure to the second curl field. The dominant right arm experienced a clockwise (CW) field, while the nondominant arm experienced a counterclockwise (CCW) field (see Fig. 1b, c).

"

#





 vx CW curl field ) ¼ vy 20 0 Fy " #    Fx 0 20 vx CCW curl field ) ¼ vy 20 0 Fy Fx

0

20

ð1Þ

Experimental tasks Subjects performed consecutive ‘‘center out and back’’ reaching movements following a pseudorandom sequence of targets. Throughout this experiment, handle position and the desired targets were displayed (real time) as colored circles with diameters of 1 cm. Four targets were positioned 10 cm radial to a start position that was oriented along the subject’s midline. Target 1, target 2, target 3 and target 4 (T1, T2, T3, T4) were positioned at 45, 135, 225 and 315 relative to the medial-lateral axis (Fig. 1a). Each subject was instructed to perform rapid, ballistic reaching movements to the target. Only the ‘‘center out’’ reaching movements were analyzed and the return movements were not considered. Feedback was provided to encourage subjects to perform reaching tasks with consistent peak tangential velocities. After a trial, the target circle changed color: white signaled that the reaching movement fell within the desired peak tangential velocity range (45– 55 cm/s); green and red signaled that movements were too slow or fast, respectively. Every subject performed reaching movements under three conditions. Subjects executed 60 reaching movements within the null field to become familiarized with the experimental setup and practice reaching within the desired peak tangential velocity range. Under this condition, the motors were active but only provided compensation for the inertia of the robot links. After familiarization, subjects completed 240 trials within the curl field. During this training condition, the curl field was active for both center out and return movements and the direction of the curl field depended on the tested arm. Dominant and nondominant arms experienced CW and CCW fields, respectively. Finally, subjects performed 160 reaching movements in the third condition. This final condition produced the same curl field, but introduced catch trials. Occurring pseudorandomly (mean = 1 of 8 trials), the catch trials removed the curl field and induced aftereffects by unexpectedly reintroducing subjects to the null environment. Subject groups To determine whether the order of target exposure affects adaptation, subjects were evenly divided into two groups of eight. Each group performed all three conditions with both

123

Exp Brain Res Dominant

B

1 0.5

Vy(m/s)

Fig. 1 a Top view of experimental setup: testing began with subjects holding a robotic manipulandum (not shown) at a central start point oriented along the subject’s midline. Subjects were positioned so that ºS and ºE were 60. Target 1, target 2, target 3 and target 4 (T1, T2, T3, T4) were positioned at 45, 135, 225 and 315 relative to the medial-lateral axis. b Forces exerted by robot while testing the dominant arm (clockwise field). c Forces exerted by robot while testing the nondominant arm (counterclockwise field)

A

0 -0.5 -1 -1

C

-0.5

0 Vx(m/s)

0.5

1

Nondominant 1

Vy(m/s)

0.5 0 -0.5 -1 -1

arms; however the pseudorandom target sequence differed between groups. For all subjects in both groups, the same pseudorandom sequence was used in the dominant limb. In the identical target sequence (ITS) group, the nondominant limbs received the same pseudorandom sequence; while in the symmetrical target sequence (STS) group, the nondominant limb received a mirror-image sequence. For example, in the ITS group, if a specific trial prompted the dominant limb to reach in the anterior lateral direction to T1, the same trial for the nondominant arm would be in the anterior medial direction (T1). The corresponding target in the nondominant limb of subjects in the STS group would be in the anterior lateral direction (T2). Separating subjects into these two groups allowed for a more complete analysis, particularly concerning any factors that may have directional dependency. Within both STS and ITS groups, the limb order assignment was balanced so that four subjects were tested with the dominant arm first and the remaining four subjects were tested with the nondominant arm first.

-0.5

0 Vx(m/s)

0.5

1

cessation of movement was defined as 5% of the maximum tangential velocity. The aftereffects were characterized by a double peak velocity profile. At times, the trough between these two peaks was lower than 5% of the maximum tangential velocity. For these specific trials, the end point was manually adjusted to ensure that the entire trajectory was considered. Adaptation was quantitatively assessed using two metrics. Peak error was defined as the maximum orthogonal distance between an observed trajectory and an ideal straight trajectory. Typically, the peak error occurred at maximum tangential velocity and within the first 200 ms of the movement. The figural error between reaching movements performed within the curl field and the mean null field trajectory was also calculated (Conditt et al. 1997). Conditt et al. defined this metric as the repeated measure of the Euclidean distance between each point in one trajectory and all of the points in the other trajectory. Given a trajectory A with n points {A(1), A(2), ... , A(n)} and another trajectory B with m points {B(1), B(2), ... , B(m)} one can calculate two vectors of dimension n and m.

Data analysis distAB ðiÞ ¼ min½jjAðiÞ  BðjÞjj ð1  i  nÞ

ð2Þ

distBA ðjÞ ¼ min½jjBðjÞ  AðiÞjj ð1  j  mÞ

ð3Þ

j

For each reaching task, custom software recorded the position of the handle as measured by the robot’s encoders. These signals were digitally differentiated and low pass Butterworth-filtered (fS = 1 kHz, fC = 30 Hz, second order) to yield hand tangential velocity. The initiation and

123

i

Vector distA–B consists of the distances between trajectory B and each point in trajectory A, and vector distB–A consists

Exp Brain Res

of the distances between A and each point in B. Figural distance is then defined as

i¼1

distAB ðiÞ þ

m P i¼1

ðm þ nÞ



Dominant Arm Late Training

AL

5

5

distBA ðjÞ ð4Þ

0

Results Figure 2 illustrates the movement kinematics for early and late learning in a single subject. During early learning, the errors in both limbs were much larger when the arm moved in the anterior-medial and posterior-lateral directions. This pattern was observed in all subjects. Figure 3 displays the average adaptation patterns separated by group (STS, ITS). ANOVA analysis of the time constants of exponential fits to the adaptation patterns indicated no between-limb differences for both peak error (P = 0.807) and figural error (P = 0.095) metrics. Additional ANOVA analysis of the entire adaptation time course was performed by grouping consecutive trials into bins of 10. For both metrics, no significant between-limb differences were found. This is because later learning dominated the adaptation time course and both arms asymptoted to equivalent, small errors within the first 25 trials. By the 25th trial, more than 80% of the total error correction (e) had been completed (ep = 84.3 ± 2.9%, ef = 83.1 ± 14.7%). When this early learning phase was isolated, significant between-limb differences arose and these findings are discussed below. The target peak velocity range for this study was 45– 55 cm/s. The peak velocities for the dominant and nondominant arms for all subjects were 53.7 ± 1.2 and 52.9 ± 1.3 cm/s (mean ± standard error), respectively. Over the first 25 trials, peak velocity was not influenced by the factors of limb (P = 0.447), limb order (P = 0.771), trial (P = 0.695), target sequence (P = 0.418) and target

-10

PM -10

-10

PL 0 cm

AL

10

PL

PM -10

10

Nondominant Arm Early Training

0 cm

10

Nondominant Late Training

AM

AM

AL

10

5

5 cm

cm

0 -5

-5

It is important to note that figural error has two advantageous characteristics; the metric considers the entire shape of the observed trajectory and is insensitive to velocity (Conditt et al. 1997). Throughout this paper, measurements concerning the peak error and the figural error will be denoted with a subscript of p or f, respectively. Both metrics were analyzed with repeated-measures ANOVA. The within-subject factors were limb (dominant, nondominant) and trial number, and the between-subject factors were group (STS, ITS) and limb order (dominant first, dominant second). Since the majority of the adaptation occurred during the first 25 trials in the curl field, the analysis focused on these trials. The errors of the first three catch trials in each movement direction was calculated for each limb and analyzed with repeated-measures ANOVA.

AL

AM

10

cm

eðA; BÞ ¼

n P

AM

10

cm



Dominant Arm Early Training

0 -5

0 -5

-10

PL -10

PM 0 cm

10

-10

PM

PL -10

0 cm

10

Fig. 2 Typical adaptation to the curl field. Early learning displays the first 30 movements performed in the curl field. Late learning shows the last 30 movements in the curl field. Initially, the curl field caused drastic, directionally dependent peak errors. For both arms, movements in the anterior-medial (AM) and posterior-lateral (PL) direction resulted in much larger errors than the anterior-lateral (AL) and posterior-medial (PM) directions. Learning occurred quickly and by late learning errors were minimal

direction (P = 0.124). This indicated that under all conditions, the peak velocities of subjects’ reaching tasks were consistent. Because the curl field was velocity dependent, all subjects’ reaching movements within the curl field were exposed to equivalent perturbations. Over the first 25 trials, the time of peak error for the dominant and nondominant arms of all subjects was 198 ± 6 and 209 ± 9 ms, and fell within the time before voluntary reaction could cause significant error correction. The time of peak error was unaffected by limb (P = 0.232), limb order (P = 0.534), trial (P = 0.362), target sequence (P = 0.931) and target direction (P = 0.425). Figure 4 depicts the mean early learning patterns separated by group (STS, ITS). In the STS group, Pearson correlation coefficients between right and left arm error indicate that the two limbs displayed remarkably similar adaptation patterns (Rp = 0.926, Rf = 0.858). For the ITS group, the adaptation patterns between left and right arms were dissimilar and highly affected by target location (Rp = 0.002, Rf = 0.521). If data for all four targets were analyzed together, the adaptation rates would be highly influenced by target sequence. Therefore, subsequent analysis separated the movements into two groups:

123

Exp Brain Res ITS Peak Error 4 2 0

1

35

69

103 137 171 205 239 trial

4 2 0

1

35

69

ITS Figural Error

2.5 2 1.5 1 0.5 0

1

35

D 2.5

STS Peak Error 6

error (cm)

error (cm)

B

C

ND D

6

error (cm)

error (cm)

A

103 137 171 205 239 trial

2 1.5 1 0.5 0

69 103 137 171 205 239 trial

STS Figural Error

1

35

69 103 137 171 205 239 trial

Fig. 3 Mean learning patterns separated by target sequence group (STS, ITS). a Peak error and c figural error adaptation time course for the ITS pattern and b peak error and d figural error for the STS pattern show that most of the learning occurred in the first 25 trials. For both

arms, errors asymptote to equivalent small magnitudes. For both metrics, there were no significant between-limb differences. However, when the early learning phase was isolated, there were significant differences between limb performance

anterior-medial, posterior-lateral (AM–PL group, large errors) and anterior-lateral, posterior-medial (AL–PM, small errors). Figure 5 depicts the mean adaptation patterns separated by movement direction (AM-PL, AL–PM) and collapsed across subject groups (ITS, STS). When the data is separated in this fashion, early learning is characterized by a steady decrease in error, but quickly asymptotes. In the case of the AM–PL data, the majority of motor adaptation occurred within the first 6 AM–PL target exposures to the curl field. For both peak and figural error, we examined the first six AM–PL movements across all subjects with repeated-measures ANOVA. The limb effect was significant (Fp(1,12) = 13.7, pp = 0.003; Ff(1,12) = 6.44, pf = 0.026) indicating that overall, errors were smaller in the non-

dominant limb (mean peak error = 3.5 ± 0.4 cm, mean figural error = 1.4 ± 0.18 cm) compared to the dominant limb (mean peak error = 4.3 ± 0.5 cm, mean figural error = 1.7 ± 0.2 cm). The effect of trial number was highly significant (Fp(5,8) = 62.0, pp < 0.001; Ff(5,8) = 24.30, pf < 0.001), indicating motor adaptation was taking place and errors were decreasing with each additional exposure to the curl field. The limb · trial interaction was not significant (pp = 0.11, pf = 0.371). This indicates that the pattern of change of kinematic error with increasing trial number was no different between limbs. Correlation measures further show that the rate of adaptation for both arms was comparable in both STS (Rp = 0.866, Rf = 0.918) and ITS (Rp = 0.876, Rf = 0.865) groups. The effects of target sequence and limb order were not significant (pp ‡ 0.13,

ITS Peak Error

A8

Nondominant

ITS Figural Error

C3 error (cm)

erorr (cm)

Dominant

6 4 2

2 1 0

0 0

5

10

15

20

0

25

5

10

STS Peak Error

6 4 2 0

25

20

25

2 1 0

0

5

10

15

20

25

trial

Fig. 4 Mean learning patterns separated by target sequence group (STS, ITS). Solid markers indicate movements in the AM or PL direction (large errors), clear markers indicate movements in the AL or PM direction (small errors). a Peak error and c figural error for the

123

20

STS Figural Error

D3 error (cm)

erorr (cm)

B8

15 trial

trial

0

5

10

15 trial

ITS pattern do not correlate because each trial resulted in different reaching directions for the two arms. b Peak error and d figural error for the STS patterns were highly correlated because each trial invoked equivalent reaching movements in intrinsic coordinates

Exp Brain Res

A

Peak Error Early Learning

8

DAM-PL NDAM-PL

error (cm) error (cm)

error (cm)

6 6

DAL-PM NDAL-PM

6

Peak Error Peak Error

A

4

*

*

**

Dominant Arm Dominant Arm Nondominant Arm

4 4 2 2

2

0 0 0 0

2

3 trial

4

5

AM

6

B

Figural Error Early Learning

3

PL AL movement direction PL AL

PM PM

Figural Error

2.5 **

*

**

2 error (cm)

error (cm)

B

1

AM

2

1.5 1 0.5 0

1

0 0

1

2

3 trial

4

5

6

Fig. 5 a Peak error ± SE and b figural error ± SE of the dominant and nondominant arms during the early learning phase separated by direction and averaged across all subjects. When reaching in the AM– PL directions, nondominant arm movements resulted in significantly smaller kinematic error than the dominant arm. There is no significant difference between arms when reaching in the AL–PM directions. Note that the trial numbers refer to consecutive trials in a given direction (AM–PL or AL–PM) and are different from the trial numbers in Fig. 4

pf ‡ 0.10). The limb · limb order interaction was not significant (pp = 0.067, pf = 0.573). This indicates that no order effects occurred. A complementary analysis was performed on the first six movements in the AL–PM directions. The limb effect was not significant (pp = 0.930, pf = 0.71) indicating that the magnitude of errors was comparable between dominant and nondominant arms. The effect of trial number was significant for peak error (F(5, 8) = 6.48, pp = 0.011), but not for the figural error (pf = 0.096). The effect of target sequence was not significant (pp = 0.15, pf = 0.07). The limb · trial interaction was not significant (pp = 0.12, pf = 0.55), indicating the adaptation rates of the two limbs were not different. The limb order main effect and the limb · limb order interaction were not significant (pp ‡ 0.091, pf ‡ 0.083). This indicates that no order effect occurred. Aftereffect analysis was consistent with the analysis of kinematic error during curl field adaptation. Figure 6

AM

PL AL movement direction

PM

Fig. 6 Mean aftereffects ± SE averaged across all subjects and separated into the four reaching directions. For both arms, catch trials for reaching movements in directions that invoked large errors (AM–PL) produced large aftereffect peak errors. The nondominant arm had significantly smaller aftereffects than the dominant arm, especially in the AM–PL directions. Significant differences between dominant and nondominant arms are marked with an asterisk (P < 0.05)

displays group means in aftereffects for the four movement directions. Repeated-measures ANOVA found that there was a significant limb effect (Fp(1,12) = 12.3, pp = 0.004; Ff(1,12) = 4.87, pf = 0.048), indicating the aftereffects during catch trials were larger in the dominant arm than the nondominant arm. Limb order and target sequence were not significant (pp > 0.46, pf > 0.39). The movement direction factor was also significant for peak error (F(3,36) = 40.7, pp < 0.001), indicating that aftereffect peak errors varied with movement direction. However, figural error in the aftereffects showed no directional dependency. We also calculated the ratio of aftereffect errors (mean of first three catch trials) to direct errors during early learning (mean of first three exposures to the curl field). For both arms, this ratio was approximately 1 for all targets and no significant differences were found between limbs. The lack of directional dependence in the figural errors of aftereffects was further investigated. Figure 7 illustrates the aftereffect trajectories for a representative subject. Through visual inspection, one notices that the shape of these error trajectories were directionally dependent. In directions that elicited large errors (AM-PL) during training, the aftereffect trajectories were a mirror image of the

123

Exp Brain Res

error trajectories observed during initial adaptation to the curl field. However, in the small-error directions (AL–PM), the aftereffect trajectories were characterized by a large extent error, or ‘‘overshoot’’, followed by a feedback driven correction. Therefore, figural errors in the AL–PM directions were heavily influenced by overshoot errors, while in the AM–PL directions, figural errors were mostly from feedforward directional errors. Though the peak error was smaller in the AL–PM directions, because of the large extent error in these directions, the figural errors in all directions were comparable.

Dominant arm curl field and catch trials

A AM

AL

0.1

Discussion This study investigated differences in adaptation to a novel dynamic environment between the dominant and nondominant arms in naive subjects. Our results indicate that the rate of motor adaptation between both arms was similar. Yet, the peak and figural errors of the nondominant arm were significantly less than those of the dominant arm in early adaptation. Also, the nondominant limb’s aftereffects were significantly smaller than the dominant arm’s aftereffects. It is important to note that the ratio of aftereffects to direct effect errors during early learning was approximately 1 for both arms. This means that the nondominant arm controller did not increase impedance control during adaptation. Rather, the controller of the nondominant limb appears to have relied on a consistent default impedance control strategy to a greater degree than the dominant limb when adapting to a novel dynamic environment.

0.1

Further support that kinematic error drives motor adaptation -0.1

m

-0.1

PM

0

0

0

0

0.1 0.1

-0.1 -0.1

PL

m

B

Nondominant arm curl field and catch trials AL

AM

0.1

m

-0.1 -0.1

PL

0.1

0

0

0

0

-0.1 -0.1

0.1 0.1

PM

m Fig. 7 Reaching movements during the third condition with the curl field on (grey) and unexpectedly turned off (black) for a dominant and b nondominant arms. All aftereffect errors are in the opposite direction of errors during the learning phase. Although the shape of aftereffects in the AM–PL directions mirror the shape of errors seen during early learning, the shape of aftereffects in the AL–PM directions were characterized by a large extent error (overshoot)

123

In both limbs, peak and figural errors were clearly larger in the AM–PL directions compared to the AL–PM directions. This was likely due to the biomechanics of the movements. Stiffness ellipses and EMG recordings have been used to investigate the effects of external forces on arm postural error under static conditions and during multi-joint planar reaching movements. These studies show that the arm is least stiff when perturbed in the axis orthogonal to the forearm (Mussa-Ivaldi et al. 1985; Flash and Mussa-Ivaldi 1990; Gomi and Osu 1998, 1999). Based on the chosen curl fields, this corresponds to movements toward the AM and PL targets in our paradigm. For the ITS group, our initial analysis suggested that the adaptation rates were very different between limbs. However, this disparity was anticipated. Recall that the magnitude of kinematic error is a function of movement direction, with AM–PL movements resulting in larger errors. The target sequence presented to the ITS group resulted in a nonsymmetrical sequence of reaching movements between dominant and nondominant arms. This resulted in seemingly different adaptation rates between dominant and nondominant arms. When presented with a symmetric target sequence (STS group) the adaptation rates of both arms were very similar. When only AM–PL movements were considered, adaptation rates for the two limbs were similar for both the ITS and STS groups. This implies that AM–PL movements dominated motor adaptation for both arms and that the small kinematic errors from AL–PM movements had little effect on adaptation in the AM–PL movements. For example, in the ITS group, the adaptation rates during the

Exp Brain Res

first six AM–PL movements were no different between dominant and nondominant arms. However, for the nondominant arm, the first 6 AM–PL movements were completed in the first 7 movements, while the first 6 AM–PL dominant limb movements were completed in the first 18 movements. Thus, 11 more AL–PM movements were experienced by the dominant limb before the sixth AM–PL movement, yet the adaptation rates of AM–PL movements were identical between limbs. Consequently, the AL–PM movements had little effect on adaptation in the AM–PL movements. This result is consistent with theories that the magnitude of the kinematic error plays a dominant role in adaptation (Scheidt et al. 2000, 2001; Milner and Hinder 2006). Differences in mean kinematic error between dominant and nondominant arms During early adaptation and in movement directions where kinematic error was largest (AM–PL), subjects performed reaching tasks with the nondominant arm that had significantly smaller errors than the dominant arm. This implies that the dominant and nondominant arms have different adaptation strategies. Smaller kinematic errors in nondominant arm reaching tasks were likely due to a greater reliance on impedance control to reduce errors from external forces. Several results support this hypothesis. Peak errors were approximately 1 cm smaller in the nondominant limb during early adaptation, and nearly 1 cm smaller in the nondominant limb during the catch trials after motor adaptation had stabilized. If the nondominant arm was relying on feedforward control to the same degree as the dominant limb, the errors in the catch trials would be equivalent. Smaller aftereffects in catch trials are evidence for a greater reliance on impedance control in the nondominant limb. Furthermore, smaller errors were evident in the very first exposure to the curl field, indicating that greater reliance on impedance control is a default control strategy in the nondominant limb, independent of motor adaptation processes. Further examination of the aftereffects leads to an estimate of the relative magnitude of the between-arm differences. While 1 cm of difference between arms in the AM–PL movements might seem like a small difference, when one normalizes this difference to the size of the errors in the nondominant limb, one sees that the errors in the dominant arm were between 21 and 24% larger than in the nondominant limb. We consider this an important difference in the context of motor laterality studies. Differences between arms in the other two directions (AL–PM) were not significant, but this was likely due to the fact that errors in the curl field were much smaller than in the other directions. In fact, the average amount of adaptation during

early learning in the AL–PM directions was only 0.84 cm compared to 4.3 cm for the AM–PL directions. This small amount of adaptation in the AL–PM directions likely obscured any differences between the arms. A number of studies suggest that the strategy of optimal impedance control is employed by the dominant arm under certain conditions. For instance, when exposed to a noisy force field that hinders the ability of the CNS to predict appropriate feedforward compensatory forces, the dominant arm relies on impedance control to regain stability (Takahashi et al. 2001; Franklin et al. 2003; Osu et al. 2003). As destabilizing endpoint loads increase, the central nervous system selectively co-contracts the biarticular muscles to maintain stability (Franklin and Milner 2003). Further investigation has shown that adaptation to a destabilizing dynamic environment involves a combination of both feedforward internal model development to anticipate perturbations and impedance control to maintain stability (Milner and Franklin 2005). Recall that handedness studies show that the dominant arm is more adept at skillful reaching movements than the nondominant arm (Sainburg and Kalakanis 2000). However, when stability is needed, the dominant arm compensates by increasing impedance. Interestingly, in our data the nondominant arm shows this type of behavior when exposed to a predictable curl field and during the very first exposure to the field. It may very well be the nondominant arm’s intrinsic bias toward tasks requiring posture and stability that leads to increased impedance as a default mechanism when exposed to a novel dynamic environment. Whether this increased reliance on impedance control is a compensatory mechanism for reduced ability to formulate internal models remains to be determined. The greater reliance on impedance control in the nondominant limb prohibits a direct comparison of internal model formation ability between the limbs, but the fact that the rate of early adaptation was comparable between limbs shows that the ability to adapt is in tact in the nondominant limb. One alternative explanation is that the error magnitudes have been influenced by the relatively different sizes and strength levels of each subject’s dominant and nondominant arm. However, this is unlikely since the nondominant limb would be expected to have less mass and less strength, which would lead to larger errors due to the perturbations. We observed the opposite trend, smaller errors in the nondominant limb. It is also possible that the increased reliance on impedance control in the nondominant limb is unrelated to coordination ability, but instead is a compensatory mechanism because the limb has less mass and strength, which may reduce the limb’s ability to resist the perturbations. Since we did not record strength levels in the limbs we tested, it is not possible to test this hypothesis with data from this study.

123

Exp Brain Res

Differences in intrinsic muscle stiffness between limbs may have been a contributing factor. Muscles of larger size would be expected to have higher levels of intrinsic stiffness due to larger amounts of muscle fibers and connective tissue in parallel. However, this also would not explain our results, as the nondominant limb would have the smaller muscles, yet displayed more stiffness compared to the dominant limb. Since muscle intrinsic stiffness increases with activation, this was likely a mechanism by which the increased impedance in the nondominant limb was achieved. By co-contracting muscles, the stiffness of the limb was increased and the errors due to perturbations were smaller.

tance when investigating the motor adaptation ability of subjects with neurologic impairments, such as stroke. For instance, contemporary studies that have investigated rehabilitation and motor adaptation of stroke patients focus on the paretic limb (Patton et al. 2006). However, the dominant or the nondominant limb can be affected by stroke and currently the different motor adaptation strategies used by these limbs are neglected. It may prove beneficial to consider the different adaptation strategies dictated by arm dominance in future studies. Furthermore, understanding the differences between limbs may have important benefits when developing intervention strategies for motor rehabilitation after stroke.

Conclusions

Acknowledgments The authors would like to show our appreciation to Lindsay DiRomualdo, Daniela Monterrubio and Shannon O’Brien for assisting with subject recruitment, testing and analysis. We also acknowledge the Imaging Science and Information Systems (ISIS) Center at Georgetown University for providing the InMotion2 robot.

It was previously thought that motor laterality was simply a byproduct of a more developed left hemisphere/right arm control system that demonstrated greater proficiency in controlling many factors required for upper limb motor tasks. Accordingly, individuals would heavily rely on this system and generally neglect the underdeveloped right hemisphere/left arm system (Taylor and Heilman 1980; Parlow and Kingsbourne 1989). However, recent work has shown that each hemisphere/limb system is specialized for different tasks (Sainburg 2002; Sainburg and Wang 2002). While the dominant arm is more effective at coordinating interaction joint torques and controlling speed, direction and curvature of movement (Sainburg 2002; Bagesteiro and Sainburg 2002), the nondominant arm is more capable of controlling steady state stability (Sainburg and Kalakanis 2000). Consistent with these studies, we found that the dominant arm controller shows evidence of greater reliance on the development of an accurate internal model to compensate for the forces experienced within the novel dynamic environment. The nondominant arm controller relied on impedance control more than the dominant arm controller. Impedance control is energetically inefficient compared with a feedforward response that precisely counters the perturbation. However, it is not possible to determine from our data if use of impedance control is a use-dependent default choice, a compensatory mechanism for reduced ability to formulate internal models, or because of reduced ability to adapt these models to different contexts (Nozaki et al. 2006). With the proper experimental conditions, it is possible the nondominant limb would have relied more on an accurate internal model. For example, impedance control results in limitations in peak speed. Performing movements fast as possible might have resulted in identical internal models in the two arms. This study shows that there are differences in dynamic adaptation to velocity dependent force fields between an individual’s two arms. This may be of particular impor-

123

References Bagesteiro LB, Sainburg RL (2002) Handedness: Dominant arm advantages in control of limb dynamics. J Neurophysiol 88:2408–2421 Brashers-Krug T, Shadmehr R, Bizzi E (1996) Consolidation in human motor memory. Nature 382:252–255 Caithness G, Osu R, Chase H, Klassen J, Kawato M, Wolpert DM, Flanagan JR (2004) Failure to consolidate the consolidation theory of learning for sensorimotor adaptation tasks. J Neurosci 24(40):8662–8671 Conditt MA, Gandolfo F, Mussa-Ivaldi FA (1997) The motor system does not learn the dynamics of the arm by rote memorization of past experience. J Neurophysiol 74:2174–2178 Criscimagna-Hemminger SE, Donchin O, Gazzaniga MS, Shadmehr R (2003) Learned dynamics of reaching movements generalize from dominant to nondominant arm. J Neurophysiol 89:168–176 Duff SV, Sainburg RL (2006) Lateralization of motor adaptation reveals independence in control of trajectory and steady-state position. Exp Brain Res Dec 15 (Epub ahead of print) Flanagan JR, Wing AM (1997) The role of internal models in motor planning and control: evidence from grip force adjustments during movements of hand-held loads. J Neurosci 17:1519–1528 Flash T, Mussa-Ivaldi F (1990) Human arm stiffness characteristics during the maintenance of posture. Exp Brain Res 82(2):315–326 Franklin DW, Milner TE (2003) Adaptive control of stiffness to stabilize hand position with large loads. Exp Brain Res 152:211– 220 Franklin DW, Osu R, Burdet E, Kawato M, Milner TE (2003) Adaptation to stable and unstable dynamics achieved by combined impedance control and inverse dynamics model. J Neurophysiol 90:3270–3282 Gomi J, Osu R (1998) Task-dependent viscoelasticity of human multijoint arm and its spatial characteristics for interaction with environments. J Neurosci 18:8965–8978 Gomi J, Osu R (1999) Multijoint muscle regulation mechanisms examined by measured human arm stiffness and EMG signals. J Neurophysiol 81:1458–1468 Hidler J, Nichols D, Pelliccio M, Brady K (2005) Advances in the understanding and treatment of stroke impairment using robotic devices. Top Stroke Rehabil 12(2):22–35

Exp Brain Res Lum PS, Reinkensmeyer DJ, Mahoney R, Rymer WZ, Burgar CG (2002) Clinical considerations in the use of robotic devices for movement therapy following stroke. Top Stroke Rehabil 8(4):40–53 Malfait N, Ostry DJ (2004) Is interlimb transfer of force-field adaptation a cognitive response to the sudden introduction of load? J Neurosci 24:8084–8089 Milner TE, Franklin DW (2005) Impedance control and internal model use during the initial stage of adaptation to novel dynamics in humans. J Physiol 567:651–664 Milner TE, Hinder MR (2006) Position information but not force information is used in adapting to changes in environmental dynamics. J Neurophysiol 96:526–534 Mussa-Ivaldi FA, Hogan N, Bizzi E (1985) Neural, mechanical, and geometric factors subserving arm posture in humans. J Neurosci 5:2732–2743 Nozaki D, Kurtzer I, Scott SH (2006) Limited transfer of learning between unimanual, bimanual skills within the same limb. Nat Neurosci 9(11):1364–1366 Oldfield RC (1971) The assessment and analysis of handedness: the Edinburgh Inventory. Neuropsychologia 1:97–113 Osu R, Burdet E, Franklin DW, Milner TE, Kawato M (2003) Different mechanisms involved in adaptation to stable and unstable dynamics. J Neurophysiol 90:3255–3269 Parlow SE, Kingsbourne M (1989) Asymmetrical transfer of training between hands: Implications for interhemispheric communication in normal brain. Brain Cogn 11:98–113 Patton JL, Stoykov ME, Kovic M, Mussa-Ivaldi FA (2006) Evaluation of robotic training forces that either enhance or reduce error in chronic hemiparetic stroke survivors. Exp Brain Res 168:368–383 Sainburg RL (2002) Evidence for a dynamic-dominance hypothesis of handedness. Exp Brain Res 142:241–258 Sainburg RL, Duff SV (2006) Does motor lateralization have implications for stroke rehabilitation? JRRD 43:311–322 Sainburg RL, Kalakanis D (2000) Differences in control of limb dynamics during dominant and nondominant arm reaching. J Neurophysiol 83:2661–2675

Sainburg RL, Wang J (2002) Interlimb transfer of visuomotor rotations: independence of direction and final position information. Exp Brain Res 145:437–447 Scheidt RA, Stoeckmann T (2007) Reach Adaptation and Final Position Control Amid Environmental Uncertainty Following Stroke. J Neurophysiol Jan 31 (Epub ahead of print) Scheidt RA, Reinkensmeyer DJ, Conditt MA, Rymer WZ, MussaIvaldi FA (2000) Persistence of motor adaptation during constrained, multi-joint arm movements. J Neurophysiol 84:853–862 Scheidt RA, Dingwell J, Mussa-Ivaldi FA (2001) Learning to move amid uncertainty. J Neurophysiol 86:971–985 Shadmehr R, Brashers-Krug T (1997) Functional stages in the formation of human long-term motor memory. J Neurosci 17:409–419 Shadmehr R, Mussa-Ivaldi FA (1994) Adaptive representation of dynamics during learning of a motor task. J Neurosci 14:3208– 3224 Takahashi C, Reinkensmeyer D (2003) Hemiparetic stroke impairs anticipatory control of arm movement. Exp Brain Res 149:131– 140 Takahashi C, Scheidt R, Reinkensmeyer D (2001) Impedance control and internal model formation when reaching in a randomly varying dynamical environment. J Neurophysiol 86:1047–1051 Taylor HG, Heilman KM (1980) Left-hemisphere motor dominance in righthanders. Cortex 16:587–603 Thoroughman KA, Shadmehr R (2000) Learning of action through adaptive combination of motor primitives. Nature 407:742–746 Volpe BT, Ferraro M, Krebs HI, Hogan N (2002) Robotics in rehabilitation treatment of patients with stroke. Curr Atheroscler Rep 4(4):270–276 Wang J, Sainburg RL (2004a) Limitations in interlimb transfer of visuomotor rotations. Exp Brain Res 155:1–8 Wang J, Sainburg RL (2004b) Interlimb transfer of novel inertial dynamics is asymmetrical. J Neurophysiol 92:349–360

123