Motor timing is shared between limbs during symmetric bimanual movements

Summary

The brain activity during bimanual movements exhibits a unique pattern not seen when either arm is moved alone. Existing theories suggest that the distinct pattern may come from the integration of motor timing between the hemispheres, but specifically how they combine remains unclear. Here, we measured the muscle activation timing of the left and right shoulders during a unimanual and bimanual periodic force control task and compared the empirical results with the predictions from three competing models of motor timing integration. Our results are most consistent with the view that the motor timing of the dominant hemisphere can be shared to improve the non-dominant arm’s motor timing by up to 20%. However, the sharing of motor timing occurs more readily during symmetric bimanual movements relative to antisymmetric movements. Our results highlight the unique neural processes underlying symmetric coordination and clarify the circumstances in which motor timing can be shared between the limbs.

Graphical abstract

Highlights

• •Unclear how cerebral hemispheres integrate motor timing during bimanual actions
• •Non-dominant arm’s bimanual motor timing improves by 20% and reduces force variance
• •Improvement in motor timing occurs more readily during symmetric bimanual movements
• •Dominant hemisphere’s precise motor timing can be shared to drive both arms

Biological sciences; Neural networks

Introduction

The brain’s control of motor timing has long been of interest to neuroscientists. Past studies have revealed important features of the brain’s control of motor timing and differences in the timing precision of the left and right arms. In one-handed or unimanual movements, the primary motor cortex primarily activates to move the contralateral hand,¹ particularly its distal musculature. While this contralateral control is prominent, proximal muscles in the shoulder also receive more unilateral and polysynaptic inputs.^2,3 Studies have shown how the dominant hand, or the left motor cortex in right-handed individuals, is capable of faster movements like rapid index finger tapping compared with the non-dominant hand.^4,5 This superior performance is not only attributed to inherent hemispheric dominance but also influenced by motor learning, frequent use, and the dominant arm’s higher skill level.⁶ Furthermore, the dominant brain hemisphere has superior control of motor timing as the interval between each tap is less variable in the dominant hand.^7,8 In a recent study, we showed that the precision in the motor timing between the agonist and antagonist muscles is a major determinant of the arm’s force variability during rapid periodic movements.⁹ We extended this finding in another study where we found that the motor timing of the wrist, elbow, and shoulder muscles were all more precise in the dominant right arm,¹⁰ and a linear relationship between timing precision and force variability was observed. Thus, the temporal precision of muscle activation is a critical factor during periodic movements. However, our two previous studies solely examined the control of muscle timing during unimanual movements. The purpose of this study is to investigate the control of timing during bimanual movements.

The motor timing of the left and right arms may change when movements are carried out bimanually instead of unimanually, as bimanual movements demand the coordination of motor areas in both hemispheres.^11,12 This coordination involves complex interhemispheric interactions characterized by neural crosstalk where neural activity intended for one limb unintentionally influences the other limb.¹³ This can manifest as motor interference where movements of one limb disrupt the independent control of the other,¹⁴ or neural coupling where the two hemispheres become functionally linked.¹⁵ Understanding these interactions is crucial for predicting how individual arm timings combine during bimanual movements. Some studies have reported a phenomenon known as the “bimanual advantage” where the motor timing of both hands improved when tapping with both hands simultaneously.^16,17 However, a recent study reported that the motor timing of the hands did not change during bimanual tapping.¹⁸ These conflicting results could be due to the low tapping speeds used in these studies. Force variability during periodic movements increases with speed^9,19 and the difference in the motor timing of the left and right arms becomes more evident at speeds faster than 3 Hz.¹⁰ However, previous studies used a speed of 2.5 Hz to investigate bimanual motor timing,^16,17,18 which may be too slow to observe improvements, if any, in bimanual motor timing. Furthermore, these studies relied on indirect measures of motor timing like button or lever presses that only measure the timing of the downward stroke of the movement.

To address these limitations, here, we measured the activity of the agonist and antagonist muscles in the left and right shoulders to measure their motor timing during a unimanual and bimanual isometric force control task at speeds of 3, 4 and 5 Hz. We considered three models that could explain how the bimanual motor timing of the left and right arms changed with respect to their unimanual timings (Figure 1A). First, the clock averaging model predicts that independent timers in the left and right hemispheres are averaged into a unified clock for bimanual movements.¹⁷ This model assumes that the brain combines the timing signals from both sides equally. For left and right motor timing variances ${\sigma }_L^2$ and ${\sigma }_R^2$, the average timing model predicts the variance of the bimanual motor timing to be ${\sigma }_B^2=\left({\sigma }_L^2+{\sigma }_R^2\right)/4$. The averaging timing model predicts that bimanual timing precision is superior to both the independent timers in the left and right hemispheres. In the second Bayes optimal weighting model, the left and right timers could be combined with a Bayes weighting that puts higher weight on the more precise timer, a phenomenon observed when integrating information under uncertain contexts.^20,21,22 This model predicts the bimanual timing variance to be ${\sigma }_B^2=\frac{{\sigma }_L^2{\sigma }_R^2}{{\sigma }_L^2+{\sigma }_R^2}$. The Bayes timing model also suggests that bimanual timing would be more precise than either unimanual timing, as the combined variance is always less than the minimum of the two individual timer variances. The third model assumes that the brain shares the dominant hemisphere’s timer for bimanual movements.^23,24,25 This shared timing model predicts the bimanual timing variance to be ${\sigma }_B^2={\sigma }_R^2$ for right-handed individuals. Furthermore, the shared timing model predicts ${\sigma }_B^2={\sigma }_R^2$ only for symmetric bimanual movements²⁶ while the predictions of the average and Bayes timing models remain unchanged for antisymmetric movements. Thus, we also investigated symmetric and antisymmetric bimanual movements to see if bimanual motor timing is different for these two modes.²⁷ We compared the theoretical predictions of these three models against the empirical data to see which model best predicted bimanual motor timing.

Figure 1.

Motor timing of the left and right arms during unimanual and bimanual movements

(A) Three models of motor timing during bimanual movements. Timing of the left and right hemispheres could be averaged or Bayes optimally combined. Alternatively, the dominant hemisphere’s timing could be shared.

(B) Schematic of the experimental setup. Agonist cursor size increased with shoulder flexion force and antagonist cursor size increased with shoulder extension force.

(C) Experimental protocol.

(D) Force trajectories in the x–y plane from an exemplar participant in the unimanual left, unimanual right, symmetric bimanual and antisymmetric bimanual conditions. Non-dominant arm’s force was more variable compared to the dominant arm.

Results

Force deviation varied across arms and conditions

Participants used their left and right arms to periodically exert a force against the robot handle (Figure 1B). They completed trials with either the left or right arm alone (unimanual condition), with both arms pushing and pulling simultaneously in the same direction (symmetrical bimanual condition), or with one arm pushing while the other pulled in the opposite direction (antisymmetric bimanual condition; Figure 1C). The force trajectories from the same exemplar participant completing the task at 3, 4, and 5 Hz speeds show how the trajectories were less variable with the dominant right arm (Figure 1D). Furthermore, a noticeable increase in force variability can be seen in the unimanual left condition at higher speeds.

To quantify the variability of the force trajectories, we calculated the Euclidean magnitude of the force in the x-y plane and extracted the peak values within every trial (Figure 2A). The mean force magnitude was separated by speed and condition for each arm (Figure 2B). A three-way repeated measures ANOVA with the arm (left and right), frequency (3, 4, and 5 Hz), and the condition (unimanual, symmetric bimanual, and antisymmetric bimanual) revealed a significant main effect of the frequency and condition on the mean force magnitude (see Table 1 for summary). Post-hoc tests using Tukey’s HSD to control for multiple comparisons revealed no significant differences between the left and right arms nor among the unimanual and symmetric bimanual conditions for all frequencies. The only significant differences were in the antisymmetric bimanual condition where the mean force was significantly higher on average relative to the unimanual and symmetric condition.

Figure 2.

Force deviation during unimanual and bimanual movements

(A) Force magnitude as a function of time from a sample unimanual right trial. Force deviation is the standard deviation of the peak force magnitude.

(B) Force magnitude as a function of speed for each condition per arm (mean ± SEM). No significant differences were found between the unimanual and symmetric bimanual conditions.

(C) Force deviation of the left (circle) and right (dot) arms during unimanual, symmetric bimanual and antisymmetric bimanual movements (mean ± SEM). Force deviation decreased significantly in the non-dominant left arm during symmetric bimanual movements (three-way repeated measures ANOVA, ∗∗signifies p < 0.005 and ∗∗∗signifies p < 0.001).

Table 1.

Three-way repeated measures of ANOVA to test the effect of the arm, frequency, and condition on the mean force magnitude and the force deviation

Factor	Mean force magnitude	Force deviation
Arm	F(1,14) = 4.2, p = 0.061, η2 = 0.057	F(1,14) = 45.3, p < 0.001, η2 = 0.50
Frequency	F(2,28) = 20.1, p < 0.001, η2 = 0.15	F(2,28) = 21.2, p < 0.001, η2 = 0.31
Condition	F(2,28) = 10.0, p < 0.001, η2 = 0.40	F(2,28) = 11.0, p < 0.001, η2 = 0.44
Arm × frequency	F(2,172) = 0.6, p = 0.57, η2 = 0.0065	F(2,172) = 6.1, p = 0.0027, η2 = 0.066
Arm × condition	F(2,172) = 3.8, p = 0.024, η2 = 0.042	F(2,172) = 6.5, p = 0.0018, η2 = 0.071
Frequency × condition	F(4,172) = 3.5, p = 0.009, η2 = 0.075	F(4,172) = 2.5, p = 0.043, η2 = 0.055

We then calculated the standard deviation of the peak force magnitude in every trial and separated the data to get the force deviations of the left and right arms in each condition at different speeds. A three-way repeated measures ANOVA with the arm (left and right), frequency (3, 4, and 5 Hz), and the condition (unimanual, symmetric bimanual, and antisymmetric bimanual) revealed a significant main effect of the arm, frequency and condition on the force deviation (see Table 1 for summary). We carried out post-hoc tests using Tukey’s HSD and the frequency factor was collapsed for visualization (Figure 2C). Consistent with our previous work,¹⁰ the force deviation in the unimanual left arm was significantly greater than in the unimanual right arm (p < 0.001). Furthermore, the left arm’s force deviation was smallest in the symmetric bimanual movement, greatest in the antisymmetric movement, and the unimanual condition was in between. The right arm’s force deviation was similar in both unimanual and symmetric bimanual conditions, but it was significantly greater in the antisymmetric condition (p < 0.001). To summarize, the force deviation improved during symmetric bimanual movements but only in the non-dominant left arm, and it degraded in both arms during antisymmetric movements.

Timing of peak muscle activity

Why did the left arm’s force deviation decrease during symmetric bimanual movements? To answer this question, we examined the muscle activity of the shoulder flexor and extensor from the left and right arms. The raw muscle activity from all muscles was filtered to extract the timing of their peaks (Figure 3A). Figure 3B shows the filtered muscle activities of the same exemplar participant from four sample trials (unimanual left and right, symmetric and antisymmetric bimanual), all at a speed of 5 Hz. The times of peak flexor activity within every trial were identified, then the muscle activities were centered and overlaid at the time of peak flexor activity. This allowed us to visualize the timing of the neighboring peaks in extensor activity to see how variable their timings were. For example, the peak extensor activities were all clustered around each other in the unimanual right condition, indicative of precise motor timing (Figure 3B, circles in top right). However, the peak extensor timing was variable in the unimanual left condition (Figure 3B, top left). Interestingly, the left arm’s peak extensor timing was precise during symmetric bimanual movements (Figure 3B, bottom left). In antisymmetric movements, the peak extensor timing was imprecise in both arms (Figure 3B, bottom right).

Figure 3.

Deviation in the timing of the muscle activity during unimanual and bimanual movements

(A) Raw and filtered muscle activity from a sample trial.

(B) Filtered muscle activity from the left and right flexors and extensors in each condition at 5 Hz (top panels are unimanual left and right arms, bottom left panel is symmetric bimanual and bottom right is antisymmetric bimanual). Timing was centered at each peak flexor activity. The left arm’s neighboring extensor peaks (circles) had precise timing only during symmetric bimanual movements.

(C) Mean EMG timing as a function of frequency for each condition and arm (mean ± SEM). Only the antisymmetric bimanual condition was significantly different when comparing within the same frequency (three-way repeated measures ANOVA).

(D) Standard deviation in the timing of the EMG peaks separated by arm and condition (mean ± SEM). The left arm’s EMG timing deviation decreased significantly during symmetric bimanual movements (three-way repeated measures ANOVA, ∗ signifies p < 0.05 and ∗∗∗ signifies p < 0.001).

(E) Force deviation was linearly related to the EMG timing deviation in both arms and in all conditions.

(F) Standard deviation in the EMG’s peak amplitude in each condition separated by arm (mean ± SEM). EMG amplitude deviation was significantly reduced in the left arm during both symmetric and antisymmetric bimanual conditions relative to the unimanual one (three-way repeated measures ANOVA, ∗∗∗ signifies p < 0.001).

We calculated the relative timing of the peak flexor and extensor activities in each trial to quantify the mean EMG timing in all conditions (Figure 3C). A three-way repeated measures ANOVA was carried out with the arm (left and right), frequency (3, 4, and 5 Hz), and the condition (unimanual, symmetric, and antisymmetric) as factors, which revealed a significant main effect of the frequency and condition factors on the mean EMG timing (see Table 2 for summary). We conducted post-hoc tests with Tukey’s HSD to compare the mean EMG timing within the same frequency between different conditions and between the two arms. No significant differences in the mean EMG timing were observed between either arm nor between the unimanual and symmetric bimanual conditions. However, the mean EMG timing was slower in the antisymmetric bimanual condition relative to both the unimanual (p < 0.001, frequency and hand factors collapsed) and symmetric bimanual conditions (p < 0.001).

Table 2.

Three-way repeated measures of ANOVA to test the effect of the arm, frequency, and condition on the mean EMG timing, the EMG timing deviation, and the EMG amplitude deviation

Factor	Mean EMG timing	EMG timing deviation	EMG amplitude deviation
Arm	F(1,14) = 1.3, p = 0.38, η2 = 0.013	F(1,14) = 96.6, p < 0.001, η2 = 0.55	F(1,14) = 5.3, p = 0.04, η2 = 0.16
Frequency	F(2,28) = 676.0, p < 0.001, η2 = 0.97	F(2,28) = 43.6, p < 0.001, η2 = 0.28	F(2,28) = 91.3, p < 0.001, η2 = 0.78
Condition	F(2,28) = 11.9, p < 0.001, η2 = 0.41	F(2,28) = 31.0, p < 0.001, η2 = 0.47	F(2,28) = 2.6, p = 0.09, η2 = 0.10
Arm × frequency	F(2,172) = 0.2, p = 0.78, η2 = 0.0028	F(2,172) = 2.3, p = 0.10, η2 = 0.026	F(2,172) = 0.3, p = 0.77, η2 = 0.003
Arm × Condition	F(2,172) = 3.6, p = 0.03, η2 = 0.040	F(2,172) = 32.3, p < 0.001, η2 = 0.27	F(2,172) = 11.0, p < 0.001, η2 = 0.11
Frequency × condition	F(4,172) = 0.1, p = 0.98, η2 = 0.0019	F(4,172) = 0.9, p = 0.46, η2 = 0.021	F(4,172) = 2.4, p = 0.04, η2 = 0.05

Deviation in peak muscle activation timing is related to force variability

Next, we examined the EMG timing deviation in the left and right arms (Figure 3D). A three-way repeated measures ANOVA was carried out with the arm (left and right), frequency (3, 4, and 5 Hz), and the condition (unimanual, symmetric, and antisymmetric) as factors, which revealed a significant main effect of all factors on the EMG timing deviation (see Table 2 for summary). The frequency factor was collapsed, then post-hoc tests were conducted with Tukey’s HSD to control for multiple comparisons. These tests revealed that the deviation in the left arm’s peak EMG timing decreased in the symmetric bimanual condition relative to the unimanual condition (p < 0.001). Furthermore, the EMG timing deviation was the highest in both arms during antisymmetric bimanual movements. These results resemble the findings from the force deviation analysis (Figure 2C).

Deviation in the muscle activity’s timing leads to unintended overlaps in flexor and extensor torques, which may be a major contributor toward force variability.^9,10 Figure 3E shows the force deviation charted as a function of the EMG timing deviation for each participant in the unimanual (black), symmetric bimanual (blue), and antisymmetric bimanual condition (red, data from all speeds was averaged in all conditions). A linear mixed-effects analysis revealed a significant linear relationship between the force deviation and the EMG timing deviation (${\chi }^2$ (1) = 58.84, p < 0.001). However, this relationship did not depend on the arm (${\chi }^2$ (1) = 2.45, p = 0.12) nor on the condition (${\chi }^2$ (1) = 0.14, p = 0.71). Therefore, the movement condition (unimanual, symmetric, and antisymmetric bimanual) likely changed the EMG timing deviation, which led to greater or lesser force deviation.

EMG amplitude deviation was influenced by condition

We also checked whether the deviation in the size of the muscle activity was different between the unimanual, symmetric bimanual, and antisymmetric bimanual conditions (Figure 3F). The muscle activity’s amplitude was normalized within each participant so the mean amplitude was equal to one. This normalization was done for each muscle separately. We then carried out a three-way repeated measures ANOVA with the arm (left and right), frequency (3, 4, and 5 Hz), and the condition (unimanual, symmetric, and antisymmetric) as factors, which revealed a significant main effect of the arm and frequency factors on the EMG amplitude deviation (see Table 2 for summary). The frequency factor was collapsed, then post-hoc tests were conducted with Tukey’s HSD to control for multiple comparisons. These tests revealed that the deviation in the left arm’s EMG amplitude decreased in both the symmetric and antisymmetric bimanual conditions relative to the unimanual condition (p < 0.001). The right arm’s EMG amplitude deviation was the same in all conditions. These results differ from the force deviation (Figure 2C) and EMG timing deviation (Figure 3D), which were both significantly larger in the left arm during antisymmetric bimanual movements, suggesting that bimanual movements affect the EMG’s size and timing deviation differentially.

Comparison with model predictions

Based on the timing of the unimanual left and right arm’s movements, we can use the average, Bayes, and shared timing models to predict the expected deviation in EMG timing during symmetric bimanual movements and subtract the measured timing deviation. The mean timing deviation error was negative for all three timing models, suggesting that all models overestimated the improvement in the symmetric bimanual timing deviation (Figure 4A). A one-way repeated measures ANOVA was carried out, which revealed a significant main effect of the condition on the model timing deviation error (F(2,28) = 32.3, p < 0.001). Post-hoc tests using Tukey’s HSD revealed that the average timing model was more accurate than the Bayes model (p < 0.001), and the shared timing model was more accurate than the average timing model (p = 0.02). Thus, in the order of least to most accurate were the Bayes, average, and shared timing models.

Figure 4.

Timing of the dominant arm was shared with the non-dominant left arm during symmetric bimanual movements

(A) Signed error between the left arm’s symmetric bimanual timing deviation and the deviations predicted by the average, Bayes, and shared timing models (mean ± SEM). Shared timing model was the most accurate (one-way ANOVA, ∗ signifies p < 0.05, ∗∗∗ signifies p < 0.001).

(B) Correlation of flexor-flexor and extensor-extensor activity (median ± IQR) was high during symmetric bimanual movements, but it was low between flexor-extensor activity during antisymmetric movements, suggesting that timing can only be shared during symmetric bimanual movements (Mann Whitney U test, ∗∗∗ signifies p < 0.001).

The shared timing model predicts similar timing between the left and right arms only during symmetric and not during antisymmetric bimanual movements. We carried out a correlation analysis between the flexor-flexor and extensor-extensor pairs during symmetric bimanual movements and between the flexor-extensor pairs during antisymmetric movements (Figure 4B). The correlation in the muscle activity during symmetric bimanual movements was very high with a median of 0.79 (IQR: 0.76–0.80). In contrast, the correlation was low with a value of 0.28 (IQR: 0.02–0.57) during antisymmetric bimanual movements. A Mann-Whitney U test revealed a significant difference in the correlation of the muscle activity between the two bimanual conditions (U = 345, z = 4.65, p < 0.001, r = 0.85). Consistent with the shared timing model, the non-dominant left arm could only utilize the dominant hemisphere’s precise timing during symmetric bimanual movements.

Discussion

In this study, we measured the motor timing of the left and right arms via their peak muscle activities during periodic movements. By comparing the deviation in the timing of the non-dominant arm during unimanual and bimanual movements, we tested the predictions of three timing models that claimed to explain how motor timing may be combined or shared between the left and right hemispheres. The shared timing model, which proposes that bimanual movements utilize the timing of the dominant hemisphere, was most accurate at predicting the deviation in the timing of non-dominant arm during symmetric bimanual movements.

We did not observe the bimanual advantage observed in two earlier studies comparing the deviation in the timing between unimanual and bimanual movements.^17,28 This could be due to the following two reasons. First, earlier studies measured timing via lever or button presses, which only capture the timing of the downwards motion, i.e., the flexor. Furthermore, the lever or button press does not capture the timing of the peak downwards motion as the finger may continue to push the button beyond its triggering threshold. A force sensor would be needed to estimate the peak of the downwards motion. On the other hand, our study measured the timing of both the peak flexor and extensor activities, which provides a clearer picture of motor timing control. Second, previous studies tested movements at a speed of 2.5 Hz, which may be too slow to induce an appreciable difference between the left and right arm’s motor timing. Force variability during periodic movements is known to increase with speed^9,10,19 and the difference between the timing deviation of the left and right arms only becomes evident at speeds greater than 3 Hz.¹⁰ This may explain why the two original studies reported an improvement of only 0.2–0.3% in the timing deviation during bimanual movements.^17,28 Furthermore, a more recent study examined tapping speeds of 0.8–2 Hz and did not find any significant improvement in timing deviation during symmetric bimanual movements.¹⁸ Thus, the difference between unimanual and bimanual movement timing may be difficult to discern at low movement speeds.

Our results offer insights when interpreted through the proposed clock models. The average^17,28 and Bayes^20,21,22 timing models predicted an improvement in the timing deviation of both the left and right arms during bimanual movements. The average model predicted bimanual timing variance of ${\sigma }_B^2=\left({\sigma }_L^2+{\sigma }_R^2\right)/4$, suggesting both arms would have better timing precision during bimanual movements relative to unimanual ones. The Bayes model predicted ${\sigma }_B^2=\frac{{\sigma }_L^2{\sigma }_R^2}{{\sigma }_L^2+{\sigma }_R^2}$, implying an optimal reduction in the timing variance in both arms. In contrast to these predictions, we observed an improvement of 2.3 ± 0.4% (mean ± standard error) in the non-dominant arm’s timing deviation, but the dominant arm’s timing deviation was effectively the same in both unimanual and symmetric bimanual movements (change of 0.2 ± 0.3%). These results are consistent with the shared timing model, which predicts that the bimanual timing variance for right-handed individuals would approximate the dominant arm’s unimanual variance (${\sigma }_B^2={\sigma }_R^2$), thereby explaining the dominant arm’s precise timing and the improvement in the non-dominant arm’s timing. However, the timing was only shared in select circumstances when the bimanual movement was symmetric.

Why can the left hemisphere’s precise motor timing only be leveraged during symmetric bimanual movements in our task demanding shoulder activations? The neurophysiological control of proximal shoulder muscles involves distinct pathways compared to distal muscles.^2,3 While distal muscles are primarily controlled contralaterally, proximal muscles receive more bilateral and polysynaptic input, including significant contributions from ipsilateral projections and strong interhemispheric communication²⁹ via the corpus callosum. The corpus callosum, which enables communication between the left and right hemispheres, may play a crucial role in the sharing of timing. Damage to the corpus callosum impaired the timing of finger tapping in the non-dominant but not in the dominant hand.²⁵ Furthermore, damage to the left hemisphere impairs both hands but right hemisphere damage impaired the contralateral arm only.^23,24 Additionally, patients with cerebellar lesions have impaired unimanual tapping, but their motor timing recovers to the level of the unimpaired hand during symmetric bimanual movements.³⁰ This suggests that the left hemisphere in right-handed individuals play a key role in controlling the motor timing of periodic movements, and the corpus callosum is a critical structure needed to share the left hemisphere’s motor timing with the ipsilateral arm.

This sharing of the timing with the ipsilateral non-dominant arm during symmetric bimanual movements may have been facilitated by the suppression of the right hemisphere’s activity to allow the left hemisphere’s motor timing to dictate the movement of both limbs.^26,31 This process minimizes competing neural signals, effectively reducing neural crosstalk in a way that benefits synchronous movements. The inherent ipsilateral contributions to proximal muscle control may make them particularly amenable to such shared, centrally driven timing. However, during antisymmetric bimanual movements, the left and right hemispheres must both actively control the motor timing of their respective contralateral arms. Furthermore, strong and precise transcallosal inhibition is required to prevent accidental activation of the ipsilateral arm due to interfering motor commands from the opposite hemisphere.³² Such regulation of both the contralateral arm’s motor timing and the inhibition of the contralateral motor commands makes antisymmetric bimanual movements significantly more challenging than bimanual movements,^27,33,34 preventing the sharing of timing observed in symmetric movements.

Interestingly, while both the left arm’s motor timing and its force deviation deteriorated during antisymmetric bimanual movements, the deviation in the muscle activity amplitude was smaller in both bimanual conditions relative to the unimanual one. This suggests that while the sharing of motor timing suffers during antisymmetric bimanual movements, the control of the size of the muscle activity experiences a beneficial effect so long as the movement is bimanual irrespective of its symmetry. A separation of timing and amplitude control in muscle activity has been observed in a brain stimulation study³⁵ but further research is necessary to clarify this distinction between the control of muscle activation timing and amplitude during unimanual and bimanual movements.

Our results suggest that timing precision may be supported by interhemispheric coordination mechanisms in right-handed individuals. Differences between the left and right arms’ control have been ascribed to the accuracy of the internal model³⁶ and to peripheral signal-dependent noise.³⁷ However, the improvement in the timing of the non-dominant arm’s muscle activity during single-joint symmetric bimanual movements cannot be explained by either of these theories.⁹ These results are most consistent with our hypothesis that the dominant hemisphere’s precise control of motor timing is the key to the dominant arm’s skill during rapid movements.¹⁰

Limitations of the study

Our experiments exclusively recruited right-handed individuals, which may limit the generalizability of our findings to the broader population. Left-handed individuals have been shown to exhibit less lateralized activity in the primary, premotor, and supplementary motor areas relative to right-handers,^38,39 so a study on left-handed participants is needed to test whether the observed sharing of the dominant hemisphere’s motor timing applies to both left- and right-handed individuals or is specific to the lateralization observed in right-handers.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Atsushi Takagi (atsushi.takagi@ntt.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •MATLAB data has been deposited at Figshare at https://doi.org/10.6084/m9.figshare.29068454 and is publicly available as of the date of publication.
• •MATLAB code has been deposited at Figshare at https://doi.org/10.6084/m9.figshare.29068454 and is publicly available as of the date of publication.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We would like to thank NTT, Inc. for funding our work.

Author contributions

A.T., conceptualization, data curation, formal analysis, methodology, software, and visualization. K.H., data curation and methodology. All authors contributed to writing and reviewing the manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
MATLAB data file (https://doi.org/10.6084/m9.figshare.29068454)	Figshare	RRID:SCR_004328
Software and algorithms
MATLAB (R2019b)	MathWorks	RRID:SCR_001622
MATLAB analysis code (https://doi.org/10.6084/m9.figshare.29068454)	Figshare	RRID:SCR_004328

Experimental model and study participant details

The experiment was performed at NTT Communication Science Laboratories, Japan. The study was approved by the institutional ethics committee with all participants providing written informed consent prior to participation (ID: R05-014). A total of fifteen right-handed participants (all male aged 36±4) with no known neuromuscular disorders or recent injuries took part in the experiments. Since all participants were male, it is not possible to examine the influence of sex on the results of our study. The Edinburgh Inventory⁴⁰ (0.96±0.07, mean and standard deviation) and the motor skill assessment⁴¹ were used to confirm the right-handedness of our participants.

Method details

Experimental apparatus

Participants were asked to exert a periodic force against two handles of a robotic interface (KINARM endpoint, BKIN Technologies). The positions of the left and right handles were held strongly at their origins by a strong spring-like force with stiffness $1,500$ N/m. The positions were adjusted so that the elbows were flexed by 90° and the shoulders were horizontally flexed at 30° to engage the shoulder flexor (pectoralis major) and extensor muscles (posterior deltoid). The strong spring-like force was active during both unimanual and bimanual movements. Participants pushed against these handles to produce a force with a frequency determined by a metronome. The applied force was displayed as circles on a computer monitor facing downwards, which was viewed through a film mirror placed 10 cm above the robotic planar workspace (Figure 1B). The mirror blocked direct view of the robot and the entire arm. During unimanual movements, two circular targets of radius 1 cm and two circular cursors were shown, corresponding to the flexor and extensor forces being exerted against the handle (the sign of the y-axis force was used to distinguish between them). The cursor size was proportional to the norm of the force exerted against the handle and the cursor and target radii were equal when the target force of 4N was exerted against the handle. Participants were instructed to match the cursor sizes of both the flexor and extensor targets whilst adhering to the speed set by a metronome. During bimanual trials, the targets and cursors for the left and right arms were visually only 2 cm apart from each other so participants could monitor the performance of both arms simultaneously.

A piezoelectric speaker was connected to the analog output of the robotic interface to produce beeps at a constant frequency. The metronome was placed directly behind the participant’s head at their midline.

The electromyographic (EMG) activity from the pectoralis major (flexor) and the posterior deltoid (extensor) from the left and right shoulders were measured using wireless electrodes (picoEMG, Cometa). The digital signal from the electrodes were converted to an analog signal, which were fed to the robotic interface for synchronized recording. The skin was cleaned using alcohol pads prior to the application of the electrodes. The data of the robotic handles’ positions, their respective forces, and the muscle activity from the left and right shoulders were digitized and recorded at 1,000 Hz.

Experimental protocol

Participants first familiarized themselves with the robotic interface, the visual feedback, and to time their force output with the beeps in a practice session preceding the experiment. Participants were asked to push against the handle periodically to match each cursor size with its respective target after each beep, hence the length of two beeps corresponded to one period of the arm’s periodic force output. The metronome’s frequency was set to 3, 4 or 5 Hz on each trial, whose duration lasted six seconds. These frequencies were selected as the difference between the left and right limbs becomes clearer at and above 3 Hz,¹⁰ and the maximum frequency of periodic shoulder movements is around 5 Hz.⁴² Each frequency was tested consecutively for five trials. The practice session consisted of 2 unimanual trials per arm at the lowest frequency of 3 Hz, followed by 2 symmetric and 2 antisymmetric bimanual trials.

Three conditions were tested: unimanual, symmetric bimanual and antisymmetric bimanual. In the unimanual condition, either the left or right arm alone was used to match the cursor’s size with the target’s size in time with the metronome’s beat. Visual feedback was displayed only for the tested arm in the unimanual condition. In the symmetric and antisymmetric bimanual conditions, the cursors and targets of both arms were displayed onscreen. In the symmetric bimanual condition, the participant had to hit the top and bottom targets with the left and right arms in rhythm with the beats from the metronome. During the antisymmetric bimanual condition, the top-left and bottom-right targets and the top-right and bottom-left target had to be hit together, i.e., participants had to match left shoulder flexion with right shoulder extension and vice versa.

The experimental protocol began with a unimanual block at 3 Hz with the right arm followed by the left arm, where each block contained five trials (Figure 1C). This process was repeated for 4 and 5 Hz. Then, participants completed symmetric bimanual blocks at 3, 4 and 5 Hz. Finally, they completed antisymmetric bimanual blocks at 3, 4 and 5 Hz.

Quantification and statistical analysis

All analysis was conducted on MATLAB (R2019b). Anderson-Darling normality tests were conducted prior to analysis. When violated, non-parametric tests were conducted. The comparisons of the mean force magnitude, force deviation, mean EMG timing and EMG timing deviation were carried out using a three-way repeated measures ANOVA with the arm (left, right), frequency (3, 4, 5 Hz), and the condition (unimanual, symmetric bimanual, antisymmetric bimanual) as factors. We used Tukey’s HSD for post-hoc tests to control for multiple comparisons. ∗ signifies p < 0.05, ∗∗ signifies p < 0.005, and ∗∗∗ signifies p < 0.001.

The relationship between the force deviation and the EMG timing deviation was analyzed using a linear-mixed effects analysis with the force deviation as the dependent variable and the EMG timing deviation as the independent variable with the participant as a random factor. The effect of the arm and condition factors on the slope was tested by comparing the base model with another model that included the interaction term between the factor and the EMG timing deviation.

A Mann-Whitney U test was conducted to test the difference in the correlation between the two bimanual conditions because the correlation in the muscle activity between the flexor-flexor and flexor-extensor pairs during symmetric bimanual and antisymmetric bimanual movements violated normality (Anderson-Darling tests, p=0.0099 and p=0.022 for symmetric and antisymmetric conditions respectively).

Published: July 18, 2025