Interlimb Responses to Perturbations of Bilateral Movements are Asymmetric

Abstract

Previous research has revealed rapid feedback mediated responses in one arm to mechanical perturbations applied to the other arm during shared bimanual tasks. We now ask whether these interlimb responses are expressed symmetrically. We tested this question in a virtual reality environment: a cursor representing each hand was used to ‘pick up’ each end of a virtual bar and place it into a target trough. Near the onset of occasional, unpredictable trials, one arm was perturbed. Regardless of which arm was perturbed, ipsilateral responses were significant during the perturbation. However, responses in the arm contralateral to the perturbation were asymmetric. While the non-dominant arm showed a significant kinematic response to correct the bar orientation when the dominant arm was mechanically perturbed, the dominant arm did not respond when the non-dominant arm was perturbed. We also saw an asymmetric response in early EMG activity, in which only the non-dominant anterior deltoid showed a significant reflex response within 100 milliseconds of perturbation onset in response to dominant arm. This response was consistent with correcting the bar position, but not with correcting its orientation. We conclude that responses to perturbations during bilateral movements are expressed asymmetrically, such that non-dominant arm responses to perturbations to the dominant arm are stronger than dominant arm responses to non-dominant arm perturbations.

Introduction

Bilateral coordination is crucial for most activities of daily living, including preparing and eating food, donning and doffing clothing, and many other work, self-care, and leisure-related activities. Not surprisingly, bimanual coordination appears to recruit specialized neural circuits that do not appear to be a simple summation of left and right unimanual control networks (Brinkman, 1984; Sadato et al., 1997; Donchin et al., 1998; Jäncke et al., 1998, 2000; Debaere et al., 2001). For example, neuroimaging studies have indicated a specialized role of supplementary motor area in bimanual coordination, based on findings that show increased activation of the SMA during bimanual movements (Sadato et al., 1997; Jäncke et al., 1998, 2000; Debaere et al., 2001) . Although the specific role of the SMA in bimanual coordination is unclear, evidence suggests a role in planning an integration of bilateral signals (Jäncke et al., 2000). In addition to specialized cortical networks, bimanual coordination is also mediated by subcortical circuits, such rapid feedback responses including reflexes (Mutha & Sainburg, 2009; Dimitriou et al., 2012; Omrani et al., 2013). In a seminal study, Diedrichsen (2007) applied a velocity-dependent force field to one arm during bimanual forward reaching movements and reported error corrections in the non-perturbed arm that occurred late in the movement, but were only expressed when a single cursor was shared between both arms, and not when each arm carried its own cursor to its own target. This study did not examine muscle responses. Mutha and Sainburg (2009) extended these findings, reporting reflex responses at the shoulder joint that occurred within 50 milliseconds of a 40 N force pulse perturbation onset, a latency associated with the transcortical component of the stretch reflex (Kurtzer, et al., 2008; Omrani et al., 2013). This study was done by applying an unpredictable 50 Newton force perturbation of 50 millisecond duration to one of the arms, under two task conditions: 1) Two cursors, controlled independently by each arm were displayed, and the cursors were to be brought to two separate targets simultaneously. 2) A single cursor displayed between the hands that depended on the position of both hands was to be brought to a single target. Reflex responses occurred in both the perturbed dominant arm and the contralateral non-dominant arm when the task and cursor were shared between the arms, but only in the perturbed arm when the task and cursor were not shared. However, this study only examined responses to dominant arm perturbations, and not to non-dominant arm perturbations. The long-latency and task-specific nature of bilateral responses suggests the involvement of transcortical loops (Evarts & Tanji, 1976; MacKinnon, et al., 2000; Pruszynski, et al., 2014). Further, because of the asymmetry in cortical motor systems (Sainburg, 2014), we hypothesize that the expression of bimanual responses that are modulated by cortical motor areas, are also asymmetric.

We have previously reported substantial asymmetries in cortical mechanisms of control, throughout the posterior frontal and parietal lobes (Mutha, et al., 2014). We hypothesize that this asymmetry in the cortical control of movement might be reflected in long-latency reflexes. Thus, we predict that long latency bimanual responses might be expressed asymmetrically in a manner that reflects neural lateralization for motor control. Previous research has suggested that the hemisphere contralateral to the non-dominant arm is specialized for feedback mediated control mechanisms that are important when performing movements in unpredictable mechanical environments and to stabilize the limb against loads applied by the dominant arm (Duff & Sainburg, 2007; Schabowsky et al., 2007; Yadav & Sainburg, 2014; Woytowicz et al., 2018). In contrast, the hemisphere contralateral to the dominant arm appears specialized for predictive control of limb and task dynamics, which is particularly effective when performing movements in consistent environmental conditions (Sainburg, 2002; Yadav & Sainburg, 2014). As a result of this asymmetry in control, participants adapt more effectively using the dominant arm when performing in predictable and consistent force environments, but adapt better to unpredictable force fields, using the non-dominant arm (Yadav & Sainburg, 2014). A simplified view of this asymmetry can be expressed as dominant hemisphere/limb specialization for predictive control, and non-dominant hemisphere/limb specialization for reactive control mechanisms.

We now test this hypothesis in the context of bimanual movements. We predict that the non-dominant arm will be more responsive to dominant arm perturbations than dominant arm will be to non-dominant arm perturbations. We tested this hypothesis using a virtual object manipulation task in which both arms were required to work together to move a shared virtual object to a target. At the onset of occasional and unpredictable trials, a solenoid acted as a clutch preventing motion, primarily along the antero-posterior axis, which produced errors in motion of the bar that required corrections. We perturbed the dominant and non-dominant arms individually to examine whether responses to the perturbation in the non-perturbed arm depended on the side of the perturbation.

Materials and Methods

Participants

Participants were 10 healthy right-handed adults (5 male, 5 female) aged 18-25 years old. All participants were screened for handedness using the Edinburgh Inventory (Oldfield, 1971) with a mean handedness score of 68.85 across all participants, indicating moderate right-handedness. Each participant provided informed consent before participation in this study, which was approved by the institutional review board of Penn State University.

Experimental Setup

Participants were seated at a 2-D virtual-reality workspace in which stimuli from a TV screen were reflected by a mirror, with the participants’ arms under the mirror. Figure 1 shows this experimental set-up. Participants’ arm movements were tracked using 6 DOF magnetic sensors (Ascension TrackStar) placed on the hand and upper arm. All joints distal to the forearm were splinted. We digitized the location of the tip of the index finger, as well as multiple locations on the hand, and upper arm, and used custom software to estimate the locations of the wrist, elbow, and shoulder joints, relative to these digitized landmarks. Vision of the participants’ arms was occluded while position of the index finger was provided as a cursor on the screen. Participants’ arms were supported on air sleds that reduced the effects of friction, and eliminated gravitational torques at the joints. Attached to each air sled was a metal rod that glided through a low-friction vinyl sleeve on a swivel. During baseline movements of the arms, the rods glided through the low friction sleeve, allowing for unhindered motion. A friction-instantiated brake, attached to a solenoid clamped the rod when triggered, restricting motion of one arm, predominantly in the antero-posterior direction for perturbations. It is important to note that there was no mechanical connection between the arms, and thus there was no mechanical stimulus applied to the non-perturbed arm. The solenoid clamp was mounted on a low friction ball-bearing swivel, which when clamped restricted motion of the arm about an arc with a radius defined by the location of the hand along the anterior posterior axis (Y-Axis). While the perturbation occurred in time, relative to the start in movement (10 ms after movement onset), it always occurred prior to the hand traversing midway between the start and the target location. This midpoint was 50 centimeters from the center of the swivel.

FIGURE 1.

Experimental Set-up- A TV screen positioned above a mirror, creating a 2-D virtual reality workspace is shown on the left. Pictures on the right show how the subjects were required to move the virtual bar and the position of the perturbation devices

The geometry of this set-up allowed us to approximate the perturbation as arresting forward motion, but allowing perpendicular displacement. Specifically, an excursion along an arc of 10 cm is associated with a displacement parallel to the target direction of just 0.99 cm, and a displacement perpendicular to the target direction of 9.93 cm. We thus approximate the perturbation as braking motion along the target direction. The trial started when both hands, and thus sides of the virtual bar, left the start circles. The onset of movement was defined by the last minimum (below 5% maximum tangential velocity) prior to the maximum in the index finger’s tangential velocity profile. The perturbation was triggered 10 ms after the start of the trial. We used kinematic data to confirm the onset of perturbation. This was determined as the time at which the tangential acceleration of the perturbed hand deviated from the average of all unperturbed trials by 1 standard deviation. EMG activity was recorded from the biceps, triceps, and anterior and posterior deltoid of both arms with active electrodes (35 mm electrode distance, 500 Hz low pass filter, Biopac Systems inc).

Experimental Task

The experimental session consisted of 171 total bimanual movements. Participants were first required to “grab” a virtual bar (20 cm across) by moving cursors representing the position of each hand to each end of the bar. Once the cursors locked on to each end of the bar, participants controlled the movement of the bar with both hands. Next, participants moved the bar into the start position, with each end of the bar placed in the small green circles. Once in the start position, after 100 ms, participants were given an auditory start signal, and the cursors disappeared, giving subjects only visual feedback of the bar during the trial. The task required participants to move the bar with both hands quickly to a target “trough” that was 25 cm away from the start position, and stay there until the end of the trial. Points were given for accuracy, as long as participants reached the target speed of at least 0.8 m/s. After each trial, participants were shown the velocity of their movement compared to the minimum requirement on a velocity meter in order to inform them if they needed to move faster. Participants were instructed to be as accurate as possible in the final position while maintaining the velocity requirement, and to respond as quickly as possible to maintain accuracy through any perturbations.

Accuracy required displacement of the bar the correct distance, as well as stopping with the bar parallel to the trough, requiring control of both bar displacement and bar orientation. As shown in Figure 1, movement along the long axis of the bar was redundant, allowing the hands to move outside of the bar. The perturbation device acted as a clutch that prevented motion primarily along the antero-posterior axis for one of the arms, and locked in position for 200 milliseconds. There was no specific load applied to the arm, but rather an increased resistance to movement of the hand in the direction of the target for 200 milliseconds. After 20 baseline non-perturbed trials, a perturbation was applied 10 ms after movement onset approximately every 10th trial, resulting in 16 total perturbation trials (8 right, 8 left). The perturbations alternated between the dominant and non-dominant arms. The relatively low number of perturbations was done in an effort to prevent participants from anticipating perturbations and to ensure a consistent baseline. Increased muscle activity and co-contraction during adaptation to many different environments has been shown (Milner, 2002; Franklin et al., 2003), and these changes can persist for several trials even when the perturbation is removed. In addition, long-latency reflex responses can be sensitive to the predictability of the perturbation (Rothwell et al., 1980). Thus, we attempted to limit changes in muscle activity and co-contraction brought on by anticipating the perturbation during the session by having more non-perturbed trials in between perturbed trials.

Kinematic Analysis

We calculated arm segment positions and angles from digitized locations relative to the Trackstar 6-DOF sensors. Data were collected from each sensor at 116 Hz. We digitized multiple positions on the hand, wrist and upper arm. Using custom software, we calculated 10 degrees of freedom per arm, however, because this task was restricted to the horizontal plane by air sled support, and all joints distal to the forearm were splinted, we report only planar motion of the hand, as well as horizontal flexion/extension of the shoulder and elbow joint flexion/extension. All kinematic data were low-pass filtered at 8 Hz (3rd order, dual pass Butterworth) and differentiated to yield velocity and acceleration. Trials in which the subjects failed to make a corrective response were excluded. This typically meant that the participant stopped movement in response to or prior to the perturbation. We analyzed kinematic responses immediately after the perturbation to characterize the mechanical effect of the perturbation. We also analyzed the kinematic response in the recovery phase, which included 100 ms after the end of the perturbation. We chose this interval as it shows the immediate voluntary responses of each arm after the perturbation and it is consistent with the timing of previously reported bilateral kinematic responses to perturbations (Diedrichsen, 2007).

EMG Analysis

EMG data was collected at 1 KHz for the biceps, triceps, anterior deltoid, and posterior deltoid, and high-pass filtered with a cutoff of 5 Hz. Collection began one second prior to the start of the trial and ended two seconds after the trial ended. EMG signal was then full-wave rectified, low-pass filtered at 400 Hz using a third-order dual-pass Butterworth filter, and normalized to the maximum EMG recorded during the experiment for that muscle for each subject. It should be noted that EMG in the figures are low-pass filtered at 10 Hz for a clearer presentation. Trials in which there was no discernable response in perturbed arm electrodes were eliminated from analysis. Out of the 160 perturbation trials across all subjects, 13 trials were excluded. Similar to previous studies (Kurtzer et al., 2008; Omrani et al., 2013) we binned muscle activity into three intervals, relative to the initiation of the perturbation: 1) R1 (20-45 ms), 2) R2 (45-75 ms), and 3) R3 (75-105 ms) to reflect the latencies of the short-latency, early long-latency and late long-latency components of the stretch reflex. We also examined EMG activity in each muscle 100 ms prior to the perturbation (pre-perturbation interval) to insure a consistent baseline activity for each muscle, and we looked at activity in the voluntary interval (105-200 ms after perturbation onset). In order to examine the effect of the perturbation on muscle activity, for each muscle we calculated the integral of the EMG in each of these intervals. We compared muscle activity in these intervals in the arm ipsilateral and contralateral the perturbation with the corresponding activity in matched non-perturbed trials.

Statistical Analysis

Statistical analysis of the kinematic measures was done using a two-way mixed-factor ANOVA, with perturbation condition (non-perturbed, ipsilateral perturbed, contralateral perturbed) and hand (dominant and non-dominant) as the independent factors. Significant interactions and main effects were adjusted for multiple comparisons and subjected to post hoc analysis using Tukey HSD. A separate two-way ANOVA (hand by interval) was done only on the intervals during the perturbation, to assess potential asymmetries in the kinematic effect of the perturbation in the perturbed arm. For all statistics, our alpha value was set at 0.05 and only p values less than or equal to 0.1 will be reported.

For statistical analysis of the EMG within the reflex interval, we compared muscle activity for each muscle using a 3 by 3 mixed-factor ANOVA with perturbation condition (no perturbation, dominant arm perturbation, non-dominant arm perturbation) and reflex interval (R1, R2, R3) as the within subject variables. We also conducted separate one-way ANOVAs for the pre-perturbation and voluntary intervals with perturbation condition as the independent variable. We used the Shapiro-Wilks test to test for the normality of the data and the Levene’s test of unequal variances. If these criteria were not satisfied we performed Box-Cox transformations of the data. Significant interactions and main effects were subjected to post hoc analysis using the Tukey HSD test, which adjusts for family-wise multiple comparisons (JMP Statistical Software, SAS Software).

Results

Subjects were asked to quickly move the virtual bar with both hands along the anteroposterior axis and to bring the bar to a stop in a target trough. A successful trial required the participants to reach the end with the bar horizontally oriented, requiring accurate displacement and accurate orientation of the bar. The perturbation arrested motion of the hand, primarily in the anteroposterior direction for 200 ms. Figure 2 shows an example of a perturbed movement along with example velocity (center) elbow and shoulder displacement profiles (left and right columns, top) and velocity and acceleration profiles for each hand (left and right columns, bottom). The lines representing stick figures of the arm are drawn between every 2 collected data points (17.2 ms). The virtual black bar was moved from the starting location to the gray trough, and the path of the middle of the bar is shown in between the hands. Near the beginning of the movement, the perturbation arrested motion, predominantly along the axis of movement for 200 ms. Because the rod attached to hand was free to swivel, arc motion was not impeded, allowing the observed medial displacement of the dominant hand-path. Only one arm was perturbed at a time, thus the left arm was not perturbed in this trial. We identify two phases of the perturbation: 1) perturbation phase, shown in Figure 2 inside the open rectangles on the example trials and 2) Recovery phase, shown in Figure 2 within the cross-hatched rectangles, which included the 100 ms immediately following the perturbation.

FIGURE 2.

Example Perturbed Movement- Displacement of the shoulder (solid line) and elbow (dashed line) is shown in the top graphs for an example right arm perturbed trial. Bottom graphs show velocity (solid line) and acceleration (dashed line) profiles of the hand with the perturbation interval shown inside the empty box and the recovery period shown inside the hatched box. An example movement is shown in the middle with stick figures of the arms drawn between every two data points (17.2 ms)

Perturbation phase kinematics

We conducted an analysis to test for a potential confound of an asymmetry in the direct effect of the perturbation on dominant and non-dominant arm kinematics. This could have occurred due to asymmetrical variations in limb configurations and/or muscle activations at the onset of the perturbation. To do this, we separated the 200-millisecond perturbation period into four 50-millisecond intervals. We then calculated the average tangential hand acceleration, elbow joint acceleration, and shoulder joint acceleration, shown in Figure 3 as the gray bars, along with acceleration for the 200 milliseconds preceding the perturbation. Baseline trial acceleration intervals are shown as open bars. We conducted a three-way ANOVA (limb) by interval (4 50-millisecond intervals within the 200-millisecond perturbation) by perturbation condition (no perturbation, left perturbation, and right perturbation)) for hand acceleration, elbow joint acceleration, and shoulder joint acceleration during the four 50-millisecond intervals that comprised the perturbation for each perturbed and baseline arm.

FIGURE 3.

Effect of Perturbation- Mean ± SE acceleration of the hand across all subjects broken up into 50 ms intervals for the hand, elbow, and shoulder with baseline, non-perturbed trials in empty bars, overlayed with perturbed trial average in gray

As expected, ANOVA indicated an interaction between perturbation and interval for hand acceleration (F(3,27) = 4.9082, p = .0075), elbow acceleration (F(3,27) = 8.4381, p = .0004), and shoulder acceleration (F(3,27) = 4.1958, p = .0147). There was also a main effect of perturbation for hand acceleration (F(1,9) = 36.7594, p = .0002) elbow acceleration (F(1,9) = 70.4034, p < .0001), and for shoulder acceleration (F(1,9) = 14.1591, p = .0045, as well as a main effect of interval for hand acceleration (F(3,27) = 18.8494, p< .0001), elbow acceleration (F(3,27) = 20.6454, p < .0001), and shoulder acceleration (F(3,27) = 37.4683, p < .0001). However, there were no main effects nor interactions with hand (left, right) for any of these dependent variables. Thus, we conclude that the kinematic effects of the perturbations did not differ between the hands.

Response Kinematics of the Recovery Phase

Ipsilateral Arm: Intralimb Response

We examined the kinematic response that occurred in the 100 ms period immediately following the perturbation. This recovery period is shown in the trials of Figure 2 by the cross-hatched rectangles Extension of the elbow and flexion of the shoulder are impeded during the perturbation phase, requiring corrections in the recovery phase to move the hand toward the target. Kinematic analysis showed asymmetric responses of the perturbed dominant and non-dominant arms in the recovery phase.

The bar graphs in Figure 4 show these responses across trials and subjects. The bars represent the mean acceleration across the 100 ms interval, following the perturbation (ipsi perturbation) for the perturbed arm (right bars), and the same time period for the same arm, but during unperturbed trials (left bars). For the unperturbed trials (left), there were no differences between hand acceleration (a), elbow acceleration (b), or shoulder acceleration (c). Our two-way ANOVA (perturbation by hand) revealed a significant interaction between hand and perturbation (F(2,18)=11.8134, p=.0005) for hand acceleration. There was also a main effect of perturbation (F(2,18) = 23.5242, p< .0001), but no main effect of hand. Both hands accelerated after the perturbation in order to recover, but post-hoc analysis revealed that the perturbed non-dominant hand accelerated significantly more than the perturbed dominant hand immediately following the perturbation (p = .0028).

FIGURE 4.

Ipsilateral Recovery Phase Kinematics- Mean ± SE acceleration of the hand (a), elbow (b), and shoulder (c) across subjects during the 100 ms following the perturbation. Representative acceleration profiles for each joint and each arm are also shown for non-perturbed trials (solid line) and ipsilateral perturbed (dashed line).

Joint angular accelerations also showed significant differences between the hands. The ANOVA for mean elbow acceleration during the 100 ms post-perturbation (Figure 4b) showed an interaction between hand and perturbation (F(2,18)=11.4452, p=.0006), as well as a main effect of perturbation (F(2,18) = 103.3875, p<.0001), but no main effect of hand. The perturbed non-dominant arm accelerated in the extension direction significantly more than the perturbed dominant arm (p = .0117), which contributed to the rapid forward motion of the hand.

Average shoulder acceleration (Figure 4c) showed an interaction between hand and perturbation (F(2,18)=19.2873, p<.0001), and a main effect of the perturbation (F(2,18) = 9.7342, p=.0014), but no main effect of hand. Post-hoc analysis showed that the perturbed dominant shoulder accelerated significantly more into extension than the perturbed non-dominant shoulder (p=.0011). In addition, the perturbed non-dominant shoulder showed significantly less acceleration into extension than baseline trials (p = .0007), while the perturbed dominant shoulder showed no significant difference from baseline.

Overall, the kinematic data from the perturbed arms showed increased acceleration of the perturbed arms, effectively recovering both the position of the bar, as well as the orientation of the bar. This recovery response, however, was somewhat asymmetric in that the non-dominant hand showed greater forward hand acceleration, which was associated with greater elbow extensor acceleration, and lower shoulder extensor acceleration.

Contralateral Arm: Interlimb Responses

We next examined the kinematics of the unperturbed contralateral arms. Although no corrective response to the perturbation was required for the contralateral hand to complete the task, analysis of contralateral kinematics revealed bilateral responses that were asymmetric. During the recovery period, the contralateral hand decelerated in the forward direction. Figure 5a shows the peak deceleration (i.e. minimum in negative acceleration) of the hand during the 100 ms after the perturbation ended for the contralateral hands on perturbed trials (right) and for baseline (left). The ANOVA for peak hand deceleration showed an interaction between perturbation and hand (F(2,18) = 4.4871, p = .0262), a main effect of perturbation (F(2,18) = 9.5424, p = .0015), and no main effect of hand. Post-hoc analysis showed that the contralateral non-dominant hand had a significantly lower peak hand deceleration than did the contralateral dominant hand (p = .0471), and the peak deceleration for the contralateral non-dominant hand was significantly lower (greater) than the baseline non-dominant hand during unperturbed trials (p = .0252), while the contralateral dominant hand showed no difference from baseline performance.

FIGURE 5.

Contralateral Recovery Phase Kinematics- Minimum acceleration of the hand (a), maximum acceleration of the elbow (b), and minimum acceleration of the shoulder (c) during the 100 ms following the perturbation averaged across subjects. Representative acceleration profiles for each joint and each arm are also shown for non-perturbed trials (solid line) and contralateral perturbed (dashed line).

Immediately after the perturbation, the contralateral elbow showed positive, or flexor acceleration (Figure 5b). Our two-way ANOVA for maximum elbow acceleration showed an interaction between hand and perturbation (F(2,18) = 21.2084, p < .0001), a main effect of perturbation (F(2,18) = 72.4423, p < .0001), but no main effect of hand. Post-hoc analysis showed that the contralateral non-dominant elbow showed a greater maximum acceleration than the contralateral dominant elbow (p = .0146).

The shoulder accelerated in the extensor direction during the recovery period. The peak shoulder extensor acceleration (negative minimum) during the 100 ms following the perturbation (Figure 5c) showed a significant interaction between perturbation and hand (F(2,18) = 19.3707, p< .0001), and a main effect of perturbation (F(2,18) = 4.6728, p=.0232), but no main effect of hand. Post-hoc analysis indicated significantly lower peak extensor acceleration for the non-dominant, as compared to the dominant shoulder (p=.0110).

These joint kinematic data suggest that the non-dominant hand slows down significantly in response to a dominant arm perturbation. Given this finding, it would make sense that the movement duration also increases for the non-dominant hand in response to contralateral perturbation. Figure 6 shows mean movement duration for the left and right hand in baseline non-perturbed trials and in contralaterally perturbed trials. We ran two-way (hand by perturbation condition) ANOVAs as we did with the other kinematic data, and found an interaction between perturbation condition and hand (F(2,18) = 11.4265, p = .0006), and a main effect of perturbation (F(2,18) = 18.6623, p< .0001), as would be expected. Post-hoc analysis using Tukey HSD tests show that while dominant arm movement times during contralateral perturbation conditions were not significantly different from non-perturbed trials (p = .2871), non-dominant arm movement times during contralateral perturbations were significantly greater than non-dominant movement times during non-perturbed trials (p = .0181). Thus, as the joint kinematics suggest, the dominant arm movement duration is not significantly different from baseline when the non-dominant arm is perturbed, but the non-dominant arm slows down to allow the perturbed dominant arm to catch up, and to recover the horizontal orientation of the bar.

FIGURE 6.

Non-perturbed vs Contralateral Perturbed Movement Duration- Mean ± SE of total movement duration for left (white bars) and right (black bars) for non-perturbed and contralateral perturbed trials.

It should be noted that overall task success was similar between all perturbation conditions, showing that participants were able to recover from perturbations of either arm to achieve accurate final positions. The two main measures of task success were the final position error of the center of the bar and the angle of the bar at the end of movement. One-way ANOVAs comparing performance of these two measures between the three perturbation conditions showed no significant difference between conditions for either measure.

Rapid EMG Responses to the Perturbation

Pre-Perturbation Muscle Activity

In order to check that the EMG activity prior to the perturbation was consistent for all conditions and ensure that automatic gain scaling (Pruszynski et al., 2009) was not a confound, we calculated the EMG impulse 100 ms prior to the perturbation onset for all muscles. As shown in Table 1, for each of the eight muscles recorded, a one-way ANOVA confirmed that there were no significant differences in the EMG impulse prior to the perturbation between perturbation conditions, assuring that any differences in the EMG activity after the perturbation were not simply a reflection of pre-perturbation muscle activity.

TABLE 1.

Pre-perturbation Muscle Activity- Statistics are shown for one-way ANOVA with perturbation condition as the factor for all muscles.

Muscle	Effect of Perturbation	Effect of Interval	Perturbation x Interval
Right Bicep	F(2,18) = 10.1307	F(2,18) = 9480.08	F(4,36) = 2.5942
	p = .0011*	p < .0001*	p = .0526
Right Tricep	F(2,18) = 4.9477	F(2,18) = 29181.82	F(4,36) = 1.1638
	p = .0194*	p < .0001*	p = .3429
Right Ant. Delt.	F(2,18) = 20.8452	F(2,18) = 729.4724	F(4,36) = 1.2893
	p < .0001*	p < .0001*	p = .2924
Right Post. Delt.	F(2,18) = 19.6879	F(2,18) = 107.7066	F(4,36) = .7277
	P < .0001*	p < .0001*	p = .5789
Left Bicep	F(2,18) = 15.0673	F(2,18) = 68.5247	F(4,36) = 1.1580
	p = .0001*	p < .0001*	p = .3455
Left Tricep	F(2,18) = 23.9688	F(2,18) = 566.7890	F(4,36) = 4.5686
	p < .0001*	p < .0001*	p = .0044*
Left Ant. Delt.	F(2,18) = 14.9935	F(2,18) = 10153.00	F(4,36) = 2.2473
	p = .0001*	p < .0001*	p = .0832
Left Post. Delt.	F(2,18) = 24.1431	F(2,18) = 126.5930	F(4,36) = 1.9801
	p < .0001*	p < .0001*	p = .1184

IPSILATERAL ARM: Intralimb Responses.

Figures 7 and 8 show (Figure 7: non-dominant, Figure 8:dominant) example non-normalized EMG profiles for each muscle (low pass filtered at 10 Hz), with the baseline trial in gray and the perturbed trial in black. The bar graphs show the average normalized EMG impulse during the three intervals representing short, medium and long latency reflexes (R1 = 20-45 millisecond post-perturbation, R2 = 45-75 millisecond, R3 = 75-105 millisecond), with the shading of the bars corresponding to the three intervals. Statistics for the 3 × 3 (perturbation condition by interval) ANOVA for each muscle are shown in Table 2, with post-hoc results reported with Tukey’s HSD. When post-hoc analysis showed significant difference from baseline, the corresponding shading of the bar is included in the EMG profile to illustrate the timing of the intervals. The perturbation arrested forward motion of the arm, which as expected elicited short and long-latency reflex responses of the triceps and anterior deltoid of the dominant and non-dominant arm. We also saw significant short and long-latency response of the posterior deltoid in both arms. In the biceps, both arms showed significant R2 and R3 response, but only the non-dominant arm showed a significant R1 response.

FIGURE 7.

Non-Dominant Ipsilateral Perturbed EMG- Bar graphs show mean ± SE of EMG impulse across subjects during R1, R2, and R3 intervals for baseline (No Pert) muscle activity and ipsilateral perturbed (Ipsi Pert). Example EMG profiles are also shown with corresponding shading to show significant differences in each interval between baseline (gray lines) and perturbed (black lines).

FIGURE 8.

Dominant Ipsilateral Perturbed EMG- Bar graphs show mean ± SE of EMG impulse across subjects during R1, R2, and R3 intervals for baseline (No Pert) muscle activity and ipsilateral perturbed (Ipsi Pert). Example EMG profiles are also shown with corresponding shading to show significant differences in each interval between baseline (gray lines) and perturbed (black lines).

TABLE 2.

Muscle Activity in Reflex intervals-Statistics for the 3x3 (perturbation condition by interval) ANOVA for each muscle are shown.

Muscle	Effect of Perturbation
Right Biceps	F(2,18) = 1.0529, p = .3694
Right Triceps	F(2,18) = 3.0766, p = .0709
Right Anterior Deltoid	F(2,18) = 1.8233, p = .1901
Right Posterior Deltoid	F(2,18) = .0755, p = .9275
Left Biceps	F(2,18) = .7727, p = .4765
Left Triceps	F(2,18) = .6420, p = .5379
Left Anterior Deltoid	F(2,18) = .4360, p = .6532
Left Posterior Deltoid	F(2,18) = 1.3601, p = .2818

We also examined the EMG activity in the voluntary response interval (105-200 ms after perturbation onset) as shown in Figure 9. All muscles except for the dominant right anterior deltoid showed a significant main effect of perturbation in our one-way ANOVA. Post-hoc analysis using Tukey’s HSD test show that the activity in the ipsilateral perturbed condition was significantly greater than non-perturbed trials in those muscles.

FIGURE 9.

Ipsilateral Perturbed Voluntary EMG Activity- Bar graphs show mean ± SE of EMG impulse across subjects during the voluntary activity interval (105–200 ms post-perturbation) for baseline and Ipsilateral perturbed conditions.

In summary, both arms exhibited significant short latency (R1), and long latency (R2) responses in triceps brachii, the muscle undergoing a shortening contraction during the arrested movement. In addition, the biceps brachii showed significant long latency responses, reflecting co-activation during this long latency phase. At the shoulder, both arms showed significant short and long latency responses in the anterior and posterior deltoid. All muscles, except for the dominant anterior deltoid also showed increased activity in the voluntary interval.

CONTRALATERAL ARM: Interlimb Responses.

The contralateral arm did not experience any mechanical effect of the perturbation. However, previous research has indicated that reflex responses can be elicited in the unperturbed contralateral arm when the task is shared between both arms (Mutha and Sainburg, 2009). Figure 10 bar graphs shows average normalized EMG impulse area for the anterior deltoid under baseline conditions and contralateral conditions, along with example non-normalized EMG profiles. For the non-dominant anterior deltoid, there was a significant effect of perturbation condition (F(2,18) = 24.1431, p = .0001). Post-hoc analysis revealed that both the left and right arm perturbation conditions were significantly different from the baseline non-perturbed trials (left perturbation: p = .0001; right perturbation p = .0102). No other muscles recorded showed any significant difference from baseline in the contralateral perturbation condition. In the voluntary response interval, no muscles showed any significant difference from baseline in the contralateral perturbation condition. Interestingly, the contralateral reflex response in the non-dominant arm was consistent only with compensating the position of the bar, which was arrested by the perturbation, but not in compensating the perturbed orientation of the bar.

FIGURE 10.

Contralateral Perturbed EMG- Bar graphs show mean ± SE of EMG impulse across subjects during R1 and R2 intervals for baseline anterior deltoid (empty) and contralateral perturbed (filled). Example EMG profiles are also shown with corresponding shading to show significant differences between baseline (gray lines) and perturbed (black lines)

Discussion

In this study, we examined bilateral responses to unimanual perturbations during a shared bimanual task. We found apparent asymmetries in the kinematic responses between the dominant and non-dominant arms. When the dominant arm was perturbed, the contralateral non-dominant arm showed significant deceleration after the perturbation, presumably to ‘wait for’ the perturbed dominant arm. When the non-dominant arm was perturbed, the contralateral dominant arm showed no difference from baseline in kinematic measures. Likely due to the lack of contralateral response of the dominant arm, the perturbed non-dominant arm accelerated significantly more than the perturbed dominant arm after the perturbation. We also observed differences in muscle activity depending on which arm was perturbed. We observed a significant effect of perturbation condition of the contralateral non-dominant shoulder (anterior deltoid) but no significant responses in any muscles of the contralateral dominant arm. Overall, the non-dominant arm responded to a perturbation of the dominant arm, while the dominant arm showed little to no response to perturbation of the non-dominant arm.

The current findings suggest the non-dominant arm is highly responsive to error-related feedback during bimanual tasks, while the dominant is not. This is consistent with de Poel’s hypothesis that the dominant hand takes on the role of the “prime actor” during bimanual tasks (Swinnen et al., 1996 Johansson et al., 2006; de Poel et al., 2007). In a bimanual oscillatory wrist movement, De Poel and colleagues (2007) reported that patterns of stability and bimanual coupling were higher when the dominant arm was perturbed, as compared to when the non-dominant arm was perturbed, suggesting that the non-dominant arm compensates for the perturbation to a greater degree when the dominant is perturbed than vice versa. Similarly, in a bimanual task in which subjects had vision of one hand but not the other, visual perturbations to the dominant hand caused interference in the non-dominant hand (Kagerer, 2014).

Diedrichsen and Dowling (2009) also showed that responses to unpredictable force field exposure of one arm during reaching movements are asymmetrical, and more robust for the non-dominant arm, such that the participants relied more on their non-dominant hands to correct. These authors attributed the asymmetry to an asymmetrical error assignment process. According to this idea, when errors emerge during a bilateral task, and the origin of those errors is ambiguous, the controller tends to assign the task errors to movements in the less-reliable arm. In our current study, it is difficult to imagine how the asymmetry in responses could be attributed to error-assignment, because the nature of the perturbation is to arrest movement of one arm in the reaching direction rather than a gradual perturbation from viscous curl field, a salient cue that is not ambiguous with regard to which arm was perturbed. Thus, in our task, error assignment would appear to be dictated by the abrupt mechanical perturbation and when the dominant arm is perturbed, the response of the non-dominant arm should not be attributable to erroneous error assignment.

Our finding of substantial bilateral responses in the non-dominant but not in the dominant arm is consistent with previous studies showing that the non-dominant arm is more responsive to unexpected perturbations in unimanual movements (Bagesteiro 2003; Duff & Sainburg, 2007). This is likely due to the non-dominant hemisphere’s greater reliance on impedance control mechanisms that are dependent on feedback (Schabowsky et al.,2007; Yadav & Sainburg, 2014). This study extends these findings to bimanual movements, showing that the non-dominant limb is more responsive to task-related proprioceptive feedback even when that feedback originates from the contralateral limb.

Whereas stretch reflexes are typically elicited during a postural task, the ipsilateral EMG findings here are consistent with the proposition that the rapid feedback responses reported here were mediated by stretch reflexes in our movement task. The R1 response reflects spinal circuits and has a latency of about 20-25 ms in the lower arm muscles in humans (Shemmel et al., 2010). Given the short latency of the R1 component, it is reasonable to conclude that a perturbation that elicits significant response during this short latency would result from activation of the same spinal circuits. As described in the results, the effect of the perturbation was to arrest forward motion of the hand, resisting ongoing extension at the elbow. The predictions for induced reflex responses at the elbow and shoulder, however, are not straightforward. This is because of two factors: 1) During voluntary movements, gamma motor neurons are activated along with alpha motor neurons (Prochazka, 1981). Thus, resisting ongoing movement should result in continued activation of gamma motor neurons, without ongoing shortening of the muscle. This should result in increased activation of the spindle afferents. 2) During multijoint movements, heteronymous pathways link the actions of muscles spanning multiple joints (Manning & Bawa, 2011), and can result in activation of muscles that are not directly stretched by the stimulus. These heteronymous responses can occur in muscles proximal and distal to the stretched muscle, as well as in short and long-latency intervals (Manning & Bawa, 2011). We expect that our stimulus that arrests forward motion of the arm should result in stimulation of stretch reflexes in the triceps brachii and anterior deltoid, due to continued contraction of the muscle, against resistance. However, due to multi-joint effects, the posterior deltoid and biceps can be stimulated through heterogenous pathways that appear to stabilize the limb against inertial interactions produced by motion of connected segments (Shemmel et al., 2010). Stretch reflexes have also been shown to elicit short latency coactivation of muscles under conditions that warrant impedance responses in the limb (Lacquaniti et al, 1991), which is similar to our results.

Our current results from the contralateral arm are consistent with previous reports, demonstrating muscle responses within 100 ms of a perturbation in the contralateral arm. However, our study extends those findings in demonstrating that contralateral responses only occurred when the dominant arm was perturbed. In a previous study, Mutha and Sainburg (2009) reported responses of dominant arm perturbations, applied with a robotic manipulandum, during bilateral reaching movements. Long latency (50 millisecond) reflex responses occurred in the left arm in response to a perturbation applied to the dominant arm, during bilateral forward reaching movements. The current finding of significant contralateral non-dominant anterior deltoid activation is consistent with the Mutha and Sainburg finding and consistent with the asymmetric kinematic response that occurred later in movement. Dimitriou et al. (2012) did not report asymmetries in bimanual reflexes. However, this may be due to differences between postural mechanisms and voluntary movement mechanisms (Crammond & Kalaska, 1996; Kurtzer, et al., 2005; Shadmehr, 2017) and the task-dependent nature of long-latency reflexes (Kurtzer et al., 2008; Mutha & Sainburg, 2009; Dimitriou, et al., 2012).

This finding however, is difficult to interpret, since it is unclear how anterior deltoid activation should have the opposite effect of the later response shown in the kinematics to slow down movement. It is true that these responses seem to be contradictory, however, they may reflect the dual goals of the task. The anterior deltoid response is not a functional response if the goal of the subject is to limit bar tilt, however, that is only part of the task. There are two goals for the task: 1) To move the bar forward into the target area and 2) to stabilize the bar orientation. Although an anterior deltoid response does not correct for goal #2, it is consistent with goal #1. The outcome of the perturbation is to slow the forward movement of the bar toward its target and the anterior deltoid response is consistent with correcting the position of the bar. Recent studies (Pruszynski et al., 2011; Lee and Perreault, 2019) identify distinct components of rapid motor responses; one relating to the rapid release of planned movements and the other relating to stabilization. In the Lee and Perreault study (2019), the authors construct a task in which participants must stabilize the arm against a haptic field and a force perturbation to reach a target, thus forcing a stabilizing response and a goal directed response. This is similar to what is occurring in our study, as one important component is to stabilize the orientation of the bar, and another component is to move the bar forward toward the target. In the Lee and Perreault study they found that the stabilizing response and the goal directed response worked independently from each other to form a sophisticated response to the perturbation. It is plausible that the two distinct responses in our study of early left anterior deltoid response and later slowing down of the non-dominant hand reflect two distinct components of rapid motor responses. It is also plausible that the most rapid reflexive responses are less sophisticated than longer latency responses, and only respond to one aspect of the task goal.