Neural coupling between homologous muscles during bimanual tasks: effects of visual and somatosensory feedback

This study investigated the effects of somatosensory feedback during bimanual tasks on the neural coupling between arm muscles, which remains largely unexplored. Somatosensory feedback using a balancing apparatus, compared with visual feedback, significantly increased neural coupling between homologous muscles (indicated by intermuscular coherence values) and improved temporal correlation of bilateral force production. Notably, feedback type modulated coherence in the α- and γ-bands (more subcortical pathways), whereas task type mainly affected β-band coherence (corticospinal pathway).

Abstract

While the effects of sensory feedback on bimanual tasks have been studied extensively at two ends of the motor control hierarchy, the cortical and behavioral levels, much less is known about how it affects the intermediate levels, including neural control of homologous muscle groups. We investigated the effects of somatosensory input on the neural coupling between homologous arm muscles during bimanual tasks. Twelve subjects performed symmetric elbow flexion/extension tasks under different types of sensory feedback. The first two types involve visual feedback, with one imposing stricter force symmetry than the other. The third incorporated somatosensory feedback via a balancing apparatus that forced the two limbs to produce equal force levels. Although the force error did not differ between feedback conditions, the somatosensory feedback significantly increased temporal coupling of bilateral force production, indicated by a high correlation between left/right force profiles (P < 0.001). More importantly, intermuscular coherence between biceps brachii muscles was significantly higher with somatosensory feedback than others (P = 0.001). Coherence values also significantly differed between tasks (flexion/extension). Notably, whereas feedback type mainly modulated coherence in the α- and γ-bands, task type only affected β-band coherence. Similar feedback effects were observed for triceps brachii muscles, but there was also a strong phase effect on the coherence values (P < 0.001) that could have diluted feedback effects. These results suggest that somatosensory feedback can significantly increase neural coupling between homologous muscles. Additionally, the between-task difference in β-band coherence may reflect different neural control strategies for the elbow flexor and extensor muscles.

NEW & NOTEWORTHY This study investigated the effects of somatosensory feedback during bimanual tasks on the neural coupling between arm muscles, which remains largely unexplored. Somatosensory feedback using a balancing apparatus, compared with visual feedback, significantly increased neural coupling between homologous muscles (indicated by intermuscular coherence values) and improved temporal correlation of bilateral force production. Notably, feedback type modulated coherence in the α- and γ-bands (more subcortical pathways), whereas task type mainly affected β-band coherence (corticospinal pathway).

the majority of human manual activities requires concurrent use of both hands (Jones and Lederman 2006), with the degree of task involvement often differing between the hands (Guiard 1987). Coordinated use of two arms/hands is a crucial aspect of performing functional tasks but is significantly degraded following hemiparesis due to stroke (Rose and Winstein 2004). Thus, proper understanding of neural processes engaged in bimanual motor activities would be crucial in developing effective treatments for functional impairment of upper extremity after neurological injuries such as stroke.

Previous investigations of the neural control of bimanual tasks have provided somewhat different explanations for how the central nervous system (CNS) controls the multiple degrees of freedom (DOFs) associated with these tasks (de Oliveira 2002). The framework of generalized motor programs (Schmidt 1975) suggests that a “general” unified motor program controls the movements of both arms and can be “fine-tuned” to produce asymmetric movements in cases where independent limb movements are required (Kelso et al. 1979). Others have postulated that movements of the two arms are planned separately but that these motor plans interact (i.e., “cross-talk” effects) at different levels of motor control (Marterniuk et al. 1984; Heuer et al. 2001). Indeed, different regions of the CNS, including both cortical and subcortical areas, have been associated with the planning and execution of bimanual tasks. These include primary motor cortex (Donchin et al. 1998), supplementary motor area (Serrien et al. 2002), cerebellum (Debaere et al. 2004), and other subcortical areas (Mayville et al. 2002).

In conjunction with motor outflow at the cortical/subcortical levels, incoming afferent information from sensory modalities such as visual (Debaere et al. 2003) or haptic (Meli et al. 2014) systems is also used and integrated during the execution of bimanual tasks. This information is typically considered as “coordination constraints” (Swinnen and Wenderoth 2004). Because sensory information is used “on-line” during task performance, performance of these tasks is generally robust to external perturbation despite the large number of DOF to be controlled. Furthermore, control of bimanual movements is flexible and can be “optimally” adjusted in response to external perturbation or task changes (Diedrichsen et al. 2004; Diedrichsen 2007; Dimitriou et al. 2012). However, some sensorimotor “perturbations,” specifically designed to disrupt bimanual coordination, have been found to be capable of degrading task performance (Peper et al. 1995) or even destabilizing the task dynamics (Swinnen et al. 1990). The nature of the motor task itself, which affects the interplay between motor and sensory information, could also be a factor that destabilizes the task (Kelso et al. 2001; Temprado et al. 2003). Taken together, this evidence emphasizes the importance of integration of sensory information for bimanual task performance and its interaction with specific task parameters.

As of yet, however, the mechanism by which sensory input influences bimanual task performance is not completely understood. The effects of sensory input on bimanual task performance have typically been examined at two ends of the motor control hierarchy, i.e., cortical [functional magnetic resonance imaging (Nair et al. 2003; Debaere et al. 2004; Sun et al. 2004), magnetoencephalography (Jirsa et al. 1998; Kilner et al. 2003); EEG (Oda and Moritani 1996; Cui and Deecke 1999)] or behavioral [task performance (Serrien et al. 2003)], but little is known about the “coupling/interaction” at the lower level of the neuromechanical system involved in the bimanual tasks, i.e., muscle level. For instance, imaging studies have provided evidence of cortical processing of sensory information during bimanual task performance (Blakemore et al. 2001; Debaere et al. 2003). However, very few studies have examined the “biomechanical substrate” of bimanual task performance, i.e., control of homologous muscles that “drives” the movements. Thus, our understanding of the neuromechanical pathway [how sensory input affects coordination of multiple muscles associated with bimanual tasks, which directly affect the task performance (and/or its degradation)] remains limited.

The goal of this study was to investigate the effects of different types of somatosensory input during bimanual tasks on the neural control of muscles. In particular, we quantified intermuscular coherence between homologous muscles during bimanual task performance under different feedback conditions. Measures of intermuscular coherence provide information on the degree of common neural input to different muscles (Farmer et al. 1993), and such motor synchrony is found to be affected by various factors such as age (Farmer et al. 2007) and training (Semmler et al. 2004; Duchateau et al. 2006). We examined the intermuscular coherence of different muscle pairs of the left and right arms, which are categorized based on their task relevance, i.e., agonist and antagonist muscle pairs. As prior studies of intermuscular coherence during bimanual tasks have shown, sensory feedback provided during task performance (Evans and Baker 2003; Stenneken et al. 2006; Boonstra et al. 2007; Boonstra et al. 2009) can increase the neural coupling/common neural drive of the two limbs. Previous studies also showed that, when complex bimanual tasks were performed, providing proper sensory information was crucial in forming cognitive strategies to successfully perform such tasks [“motor binding” (Franz et al. 2001; Mechsner et al. 2001; Wenderoth and Weigelt 2009; Beets et al. 2012)]. In this study, we further postulated that the type of sensory feedback used to constrain the limbs together may also play a significant role in the level of “binding” or neural coupling of bilaterally homologous muscles. Specifically, we tested three conditions that provided different types of sensory feedback (visual vs. somatosensory/proprioceptive) with different accuracy requirements. Based on the results from previous studies that showed the significance of somatosensory feedback, albeit in ipsilateral muscles, on intermuscular coherence (Kilner et al. 2000; Fisher et al. 2002), we hypothesize that the modality of sensorimotor feedback would have significant effects on the neural binding of bilateral muscles. We also hypothesize that the nature of the target task (i.e., flexion vs. extension) would also affect the degree of neural coupling of the muscles (Ridderikhoff et al. 2005; Boonstra et al. 2007).

METHODS

Subject characteristics.

Twelve healthy subjects, with ages ranging from 19 to 34 yr, participated in the study. All participants were right-hand dominant, as determined by their self-report. All subjects had no history of neurological or orthopedic conditions that would have affected their performance on the study tasks. All procedures were approved by the local Institutional Review Board, and all subjects provided written informed consent.

Experimental setup.

Subjects were seated at a table in a height-adjustable chair with their torso securely strapped to the chair (Fig. 1). Two test handles were positioned over the table at shoulder height. When the two hands grasped the test handles, the elbows were flexed to 90°, and the shoulder joints were abducted in the frontal plane to ∼90°. In this posture, both forearms and upper arms of the subject would lie within the same horizontal plane, and the application of horizontal forces at the handles was achieved through elbow flexion and extension. Forces at the handles were measured with force sensors (Flexiforce A401; Tekscan, Boston, MA). EMG electrodes were placed over the short head (medial) of the biceps brachii and the lateral head of the triceps brachii muscles of both arms (MyoSystem 1400A; Noraxon, Scottsdale, AZ). Maximum voluntary contraction (MVC) of the dominant limb in elbow extension was recorded with a hand-held force muscle tester (MicroFET2; Hoggan Scientific, Salt Lake City, UT), and the measured force level was used to determine the target force level for both flexion and extension contractions, which was set to 50% of the force magnitude measured during MVC.

Fig. 1.

Experimental setup. A: physical apparatus. B: user interface. For feedback 1 and 2, two handles were rigidly attached to the table (apparatus 1). For feedback 3, three pivot joints allowed free movements of the two arms, and subjects stabilized the apparatus by producing equal forces at two hands. Regarding visual feedback, for feedback 1(C), an image of a rectangular bar was provided on a computer screen, and the vertical location of its left and right ends corresponded to the force at the left and right hands; unequal forces caused the bar to tilt. Vertical location of the target window corresponded to the target force level. Feedback 2(D) was similar to feedback 1, except a ball rolled along the bar when it was tilted. For feedback 3(E), the bar did not rotate, and the vertical location of the bar was computed by the average value of the forces at the right (F_R) and left (F_L) hands.

EMG signals were sampled at 1,000 Hz and bandpass filtered between 10 and 500 Hz. Force and EMG data were synchronously recorded using the Optotrak Data Acquisition Unit II (Northern Digital, Ontario, Canada). Sampled force data were processed real time within the MATLAB xPC target toolbox environment (MathWorks, Natick, MA) to produce the different types of graphical visual interface used for the three feedback conditions (Fig. 1).

Protocol.

Subjects performed bilaterally symmetric force generation, either toward the distal direction that required bilateral elbow extension (task 1) or proximal direction that required bilateral elbow flexion (task 2), following a ramp and hold trajectory at the target force level. Three different feedback conditions (feedback 1-feedback 3), described below (Feedback conditions) in detail, were implemented for both tasks (bilateral extension/flexion), yielding 6 blocks (3 feedback conditions × 2 tasks) of 15 trials.

To prevent fatigue, subjects did not perform the same task consecutively; for instance, after a subject performed one task, either flexion or extension, under one type of feedback, he/she performed the other task under the same feedback. One-half of the subjects performed flexion first, and the other one-half started with extension. The order of the three feedback conditions followed a Latin Square design and was balanced across subjects. A rest period of 14 s was provided between trials and whenever requested by the subject. Additional rest period of 30 s was provided in between conditions to allow time for the experimenter to change over the test apparatus.

Feedback conditions.

Under feedback 1, the handles were rigidly attached to the table, and subjects generated isometric force (Fig. 1A). A computer screen provided visual feedback of the force produced at the two hands via the position and tilt of a rectangular solid bar on the screen (Fig. 1C). The vertical location of the left and right ends of the bar corresponded to the force magnitudes from the left and right hands, respectively. Thus, the bar was level when the two forces were equal. The instruction was to keep the bar inside of a slightly larger rectangle that moved vertically on the screen for 3.6 s (ramp phase) and then stayed at a position corresponding to the target force for 3.4 s (hold phase). This required the mean force of the two hands to track the movement of the target rectangle and for the forces in the two hands to be close in magnitude.

Feedback 2 was similar to feedback 1, but it imposed a stricter constraint on the force symmetry by displaying a rolling ball on the bar (Fig. 1D). Tilt of the bar caused the ball to roll toward the lower end of the bar, and the ball dropped off the bottom of the screen, when it reached the end, as if it was under the influence of gravity. Equations of motion of the rolling ball, which calculated its acceleration vector as a function of the tilt angle, were applied to provide a realistic simulation. The introduction of the rolling ball forced subjects to promptly adjust force asymmetry, since the ball would have fallen off had the asymmetry not been corrected in a timely manner. Thus feedback 2 imposed a stricter constraint on the force symmetry compared with feedback 1. During experiments, when the ball fell off, subjects stopped the task, and the trial was marked off. Data up to that point in the trial were retained in the analysis.

For feedback 3, the bolts that rigidly attached the handles to the table were removed (Fig. 1B), and the resulting structure coupled the two handles together through a linkage with three pivot points. The linkage allowed free movement of the two handles within the horizontal plane. If the forces from the two hands were not equal, the two handles would displace in opposite directions, similar to a seesaw. Thus the two hands were coupled together through the mechanical linkage, the requirement of equal forces was enforced by the apparatus itself (i.e., balancing), and isometric force was exerted on the sensor only when the forces at each hand were approximately equal. Under this condition, the force asymmetry information obtained from the force sensors was not used in the visual feedback, and the visual interface simply displayed the average magnitude of the two forces. Because the solid bar did not tilt, there was no visual feedback of force symmetry (Fig. 1E). The instruction was to keep the bar inside the target window while keeping the handles still.

Data analysis.

Force data analysis. From the force data, the following measures were computed for each feedback condition and each phase (ramp, hold): force error (ε_F), correlation between the left and right hand force (r_F), and force symmetry error (ε_FS). Force error ε_F was defined as the magnitude of the difference between target force and the average of left and right hand forces, and quantified how well subjects tracked the targets. Correlation between left and right hand forces r_F was computed by first removing trends within each trial by subtracting the least-squares best-fit regression line. The resulting detrended data from all trials of a particular task were concatenated, and a correlation coefficient was calculated between left and right hand forces. Force symmetry error ε_FS was defined as the magnitude of the difference between left and right hand forces, and quantified how well subjects kept the two forces equal.

EMG data analysis.

Muscle activation levels were estimated by the root mean square of the EMG. Intermuscular coherence between two EMG signals was computed to quantify correlations between muscle activations in the time and frequency domains. We did not rectify the EMG signals, since previous studies showed negative effects of the EMG rectification on the coherence estimation (Neto and Christou 2010; McClelland et al. 2012). EMG-EMG coherence was calculated between left and right homologous muscle pairs during both elbow extension and flexion. For each task, EMG data for each phase (ramp or hold) from all 15 trials were concatenated before analysis.

EMG-EMG coherence between muscle pairs was estimated using nonoverlapping segments (rectangular window), which resulted in a frequency resolution of 2 Hz within the MATLAB environment (MathWorks), employing a script developed by Neurospec (www.neurospec.org; Halliday 1995). Given two EMG signals x and y, let the power spectra of the two signals be denoted as f_xx(λ) and f_yy(λ), and their cross-spectrum as f_xy(λ). The coherence at frequency λ, R_xy(λ), is then computed as:

$$\left|R_{\mathrm{xy}}\left(\lambda \right)\right|=\frac{{\left|f_{\mathrm{xy}}\left(\lambda \right)\right|}^2}{f_{\mathrm{xx}}\left(\lambda \right)f_{\mathrm{yy}}\left(\lambda \right)}$$ (1)

The coherence value at a given frequency was considered only if it was greater than the 95% confidence limit. The coherence estimates were z-transformed as follows:

$$Z\left(\lambda \right)=\frac{1}{2}\ln \frac{1+R_{\mathrm{xy}}\left(\lambda \right)}{1-R_{\mathrm{xy}}\left(\lambda \right)}$$ (2)

The “z-transformed” coherence values will be normally distributed with a SD of ∼1 (Rosenberg et al. 1989). The three frequency bands of interest were α (8–12 Hz), β (13–35 Hz), and γ (36–55 Hz). The integral of z-transformed coherence was calculated within each band (CI_α, CI_β, and CI_γ).

Here, it should be acknowledged that the EMG system used for this study had a band-pass filter (10–500 Hz) embedded in the hardware (Myosystem 1400A), which we were not able to remove. Although the coherence value, theoretically, is not affected by the magnitude of two signals, this filter setup could have affected our coherence estimation, particularly the coherence in the α-band (8–12 Hz), if the power in this band was low.

In addition to the coherence integrals, contour plots of the time-dependent coherence estimates were generated using type 2 analysis of the Neurospec software, which allowed visual examination of potential interactions between the time (phase) and feedback.

Statistical test.

A univariate analysis of variance (ANOVA) was performed on each of the force variables (ε_F, r_F, and ε_FS) with task (flexion/extension), phase (ramp/hold), and feedback (1–3) as independent variables (within-subject factors) (SPSS Statistics, version 22; IBM, Armonk, NY). For coherence values, a multivariate analysis of variance (MANOVA) was performed on the coherence values (CI_α, CI_β, and CI_γ) as dependent variables and task (flexion/extension), phase (ramp/hold), and feedback (1–3) as the independent variables. First, four representative test statistics (i.e., Pillai's Trace, Wilks' Lambda, Hotelling's Trace, and Roy's Largest Root) were used to calculate P values, and Pillai's Trace was used to determine the P value for each independent variable. Here, a significance level was set to 0.05.

Additionally, as a conservative approach to avoid type 1 error (given the relatively small sample size), post hoc univariate ANOVA was performed for the independent variables when any of the four P values, calculated by the four test statistics, was found to be <0.05, since multivariate analyses (such as MANOVA) could be subject to type 1 error when the sample size is small (Wetcher-Hendricks 2011). Significance level in the univariate ANOVA was then corrected for multiple comparisons with a Bonferroni correction.

RESULTS

Force data: task performance.

All subjects were able to achieve the task goal in all three feedback conditions (Fig. 2), and there was no significant difference in task performance (i.e., force error) between these conditions. Overall, the force error ε_F was significantly smaller during the hold phase than in the ramp phase (P < 0.001; Fig. 3A).

Fig. 2.

Typical experimental data (subject 2, feedback 2). A: force profiles. B: EMG profiles. C: EMG power. D: coherence. The broken black lines show the target window provided by the visual interface. Overall, subjects were able to complete the task under all three feedback types.

Fig. 3.

Task performance measures. A: force error. B: correlation coefficient. Here, the error bars denote SD values. No significant difference was found in the force error (ε_F) between feedback types, but temporal correlation between left/right forces was significantly higher during feedback 3 (P < 0.001). The force error was significantly smaller during the hold phase (P < 0.001).

However, temporal correlation between forces at the left and right hands was found to be significantly higher under feedback 3 than feedback 1 or feedback 2, as indicated by their correlation coefficient values (r_F) (feedback 1 vs. feedback 3: P < 0.001; feedback 2 vs. feedback 3: P < 0.001) (Fig. 3B). In contrast, the force asymmetry error (ε_FS) was the highest during feedback 3, since the ε_FS values recorded during feedback 3 were significantly greater than those from feedback 1 (P < 0.001) and feedback 2 (P < 0.001) (Fig. 3C).

In addition, spectral power in the force signal did not overlap with the frequency range of the three EMG bands of the coherence values (i.e., α-, β-, and γ-band). Power spectral analysis of the force profiles (Welsh periodogram method, spectral resolution of 0.39 Hz) revealed that spectral power in the force signal was predominantly below 3 Hz; across all conditions, the power had dropped to below 1% of the peak power by 3.1 Hz. Also, there was no significant difference in the peak spectral power values across the conditions.

Electromyography data: activation level and intermuscular coherence.

Whereas there were no significant differences in the activation level of the antagonist muscles across feedback conditions (i.e., biceps brachii during elbow extension; triceps brachii during elbow flexion), there was a significant between-condition difference found for both muscles (P < 0.01) when they acted as the agonist (i.e., biceps brachii during elbow flexion; triceps brachii during elbow extension) (Table 1). The activation level of both muscles was the greatest under the somatosensory feedback (feedback 3) and the smallest under the visual feedback with less strict constraint (feedback 1).

Table 1.

Activation level of the two muscles across conditions (biceps brachii and triceps brachii)

	Biceps Brachii Activation Level, %		Triceps Brachii Activation Level, %
	Flexion**	Extension	Flexion	Extension**
Feedback 1	26.7 (3.3)	14.3 (7.9)	19.1 (10.6)	35.8 (4.4)
Feedback 2	35.8 (2.7)	14.0 (7.1)	18.7 (9.5)	47.8 (3.4)
Feedback 3	42.5 (4.6)	14.0 (7.2)	18.8 (9.7)	56.8 (6.2)

Data are expressed as means (SD).

^**P < 0.01.

For the homologous biceps brachii muscle pair, feedback type was found to have a significant effect on the intermuscular coherence values [F(6,130) = 3.843, P = 0.001; Table 2 and Figs. 4A and 5]. Post hoc pairwise comparison revealed that feedback type had significant effects mainly on the coherence values in the α-band (P = 0.006) and γ-band (P = 0.002) but not on β-band coherence (P = 0.380). The intermuscular coherence values in these bands were significantly greater under feedback 3 than feedback 1 or 2 (Fig. 4A).

Table 2.

P values for the three independent variables (within-subject factor: phase, task, feedback) from MANOVA and post hoc ANOVA

	Homologous Biceps Brachii Muscle Pair	Homologous Triceps Brachii Muscle Pair
	Phase
MANOVA	P1 = 0.568	P1 < 0.001***
ANOVA
α	NA	P < 0.001***
β	NA	P = 0.001**
γ	NA	P = 0.216
	Task
MANOVA	P1 = 0.026*	P1 = 0.226
ANOVA#
α	P = 0.617	NA
β	P = 0.011*	NA
γ	P = 0.102	NA
	Feedback
MANOVA	P1 = 0.001**	P1 = 0.099° (P2 = 0.020)
ANOVA#
α	P = 0.006**	P = 0.031*
β	P = 0.380	P = 0.315
γ	P = 0.002**	P = 0.093°
	Phase × feedback
MANOVA	P1 = 0.100° (P2 = 0.020)	P1 = 0.080° (P2 = 0.011)
ANOVA#
α	P = 0.143	P = 0.021*
β	P = 0.181	P = 0.289
γ	P = 0.082°	P = 0.454

P₁, Pillai's trace; P₂, Roy's largest root; NA, not applicable. °P < 0.1,

^*P < 0.05,

^**P < 0.01, and

^***P < 0.001.

^#Post hoc analysis was adjusted for multiple comparison (Bonferroni correction).

Fig. 4.

Coherence integral (CI) values for the two muscle pairs. A: homologous biceps brachii pair. B: homologous triceps brachii pair. Here, the error bars denote SD values. For both muscle pairs, the effects of feedback were observed (P = 0.001 for the biceps brachii pair; P = 0.099 for the triceps brachii pair). However, significant effects of task (flexion vs. extension) were observed from the biceps pair (P = 0.026), whereas the effect of phase (ramp vs. hold) was found significant for the triceps pair (P < 0.001).

Fig. 5.

Average coherence spectra between the homologous biceps brachii muscles. A: task: flexion, phase: ramp. B: task: flexion, phase: hold. C: task: extension, phase: ramp. D: task: extension, phase: hold. Coherence values were greater under feedback 3 than other feedback types. Also, a significant task effect was observed, since the coherence values were greater during extension than during flexion (A and B vs. C and D).

Additionally, the intermuscular coherence of the biceps brachii muscles was significantly affected by the task [elbow flexion vs. extension; F(3,130) = 3.202, P = 0.026; also see Fig. 4] but not by the movement phase [ramp vs. hold; F(3,130) = 0.677; P = 0.568]. In contrast to the feedback type, the task mainly affected the coherence values in the β-band (P = 0.011) but not those in the α- or γ-band. The β-band coherence was significantly higher during elbow flexion than extension.

For the homologous triceps brachii muscle pair, similar trends of the feedback effect on the intermuscular coherence were observed, albeit to a lesser degree (Table 2 and Figs. 4B and 6). Post hoc pairwise tests showed that, somewhat similar to the biceps brachii muscle pair, the feedback type had strong effects on the coherence values in the α-band (P = 0.031) but not those in the β-band (P = 0.380) or γ-band (P = 0.093).

Fig. 6.

Average coherence spectra between the homologous triceps brachii muscles. A: task: flexion, phase: ramp. B: task: flexion, phase: hold. C: task: extension, phase: ramp. D: task: extension, phase: hold. There was a significant phase effect, since the coherence values were generally smaller during the hold phase (A and C vs. B and D). Also, the between-feedback difference was larger during the ramp phase than the hold phase (phase × feedback interaction).

Unlike the biceps brachii muscle pair, there was a strong effect of phase observed on the coherence values in the homologous triceps (P < 0.001; also see Fig. 4B), since the coherence values in the α- and β-band during the ramp phase were significantly greater than those during the hold phase (Table 1 and Fig. 5). Some degree of interaction between phase and feedback was also observed (i.e., the between-phase difference was greater particularly under feedback 3), which diluted the effects of feedback type (i.e., phase × feedback interaction; P = 0.08 from Pillai's Trace; P = 0.011 from Roy's Largest Root; Fig. 7).

Fig. 7.

Representative time-dependent coherence estimate for the homologous triceps brachii muscles during elbow flexion (subject 3). A: feedback 1. B: feedback 2. C: feedback 3. For this subject, intermuscular coherence between homologous triceps brachii muscles clearly increased significantly under somatosensory feedback (feedback 3). In addition, there was some effect of the phase, since the coherence value was generally higher during the ramp period (0 < t < 3,000 ms). There was also a significant phase × feedback interaction, since the between-phase difference was much greater under feedback 3.

DISCUSSION

Task performance.

Although the feedback type did not appear to affect task outcome (i.e., no significant difference in force error across feedback types), the somatosensory feedback specifically improved temporal coordination of the bilateral forces, indicated by its significantly larger correlation coefficient values (P < 0.001; mean r_F = 0.57 for feedback 3 vs. 0.26 for feedback 1 and 0.20 for feedback 2). Although “spatial” characteristics of bimanual tasks, such as force magnitude or direction, were found to be an important aspect of these tasks (Spijkers and Heuer 1995; Swinnen et al. 2001), temporal control of the bilateral forces/movements is thought to “operate at a higher level of the control hierarchy” and to “dominate nontemporal control of bimanual force coordination” (Rinkenauer et al. 2001). Therefore, although no significant difference in task performance (defined by the force error) was detected across feedback conditions, examination of temporal correlation between bilateral forces suggests that homologous muscles were operated in a higher degree of temporal synchrony under somatosensory feedback (feedback 3) than visual feedback (feedback 1 and 2) conditions.

It should also be acknowledged that the force symmetry error was significantly larger under feedback 3 than other feedback conditions. The observed force asymmetry could be explained by small asymmetries in the device used to provide somatosensory feedback (i.e., difference in the mechanical moment arm values for the left and right hand forces). Additionally, slight asymmetries in the angles of the links in the apparatus could have required nonequal forces in the two hands to stabilize the linkage. Interestingly, the CNS appeared to adapt very quickly to these asymmetries in the linkage, producing the imbalanced forces needed to stabilize the linkage in each case, and the added complication of stabilizing the linkage during feedback 3 did not negatively affect the goal of keeping the average left-right force within the target window. Besides, it is also possible that the somatosensory feedback was not as sensitive as the visual feedback, which may have allowed small degrees of force asymmetry to remain unadjusted in the somatosensory feedback condition (feedback 3).

Effects of feedback condition and task on neural coupling of homologous muscles.

The type of feedback was found to significantly affect the activation level of the homologous agonist muscles (biceps brachii during elbow flexion; triceps brachii during elbow extension). The between-condition difference in the muscle activation level could have resulted from the between-condition difference in the force vector (i.e., greater shear force). Because the target force level was not significantly different across the conditions, it is probable that the increased level of agonist muscle activation contributed to the shear force (lateral force) at the handle (dimension irrelevant to the task goal), by which subjects attempted to “stabilize” the handle to control the apparatus more precisely. In other words, subjects tried to increase the structural impedance of the apparatus by producing a pair of lateral forces (shear force) whose kinetic effects were irrelevant to the task goal (normal force). Note that, whereas the increase in the muscle activation under somatosensory feedback (feedback 3) was the greatest among the three conditions (P < 0.01 for both feedback 1 vs. 3 and feedback 2 vs. 3), imposing a stricter constraint (achieved by displaying a rolling ball) during visual feedback also had significant effects on the muscle activation levels (P < 0.01 for feedback 1 vs. feedback 2).

However, more importantly, results from the intermuscular coherence analysis indicate that, during bimanual tasks, neural coupling between homologous muscles was significantly increased by the use of somatosensory feedback, whereas imposing stricter constraints/requirements during visual feedback did not yield significant difference in the coherence values (i.e., P values > 0.3 for all bands between feedback 1 and 2). Note that the observed between-condition difference in the coherence values is not likely to originate from the difference in the muscle activation level, since the coherence values are not affected by the magnitude of the signals.

Here, it is noteworthy that significant increase in the coherence values by somatosensory feedback was mainly observed in the α- and γ-bands. Although the origin of the β-band intermuscular coherence is considered to be cortical (Halliday et al. 1998; Grosse et al. 2002), intermuscular coherence in the other bands, the α- and γ-bands, is generally thought to arise subcortically (Farmer et al. 1993; Norton et al. 2003; Norton et al. 2004; Nishimura et al. 2009). Therefore, when interpreted in the context of the “multilevel cross talk model” (Marteniuk and MacKenzie 1980), which hypothesizes interactions between two arms result from cross talk of neural signals controlling two arms at multiple levels (see de Oliveira 2002 for review), our results suggest that feedback modality (i.e., use of somatosensory feedback) mainly influenced, or enhanced, lower-level interactions at subcortical levels that are related to execution, rather than planning stage at the cortex [such as a motor program that prespecifies the entire movement dynamics, as hypothesized in the “generalized motor program theory” (Schmidt 1975; Schmidt et al. 1979)].

Here, it should be noted that the α-band coherence values were generally much greater than the γ-band coherence values. Furthermore, for the γ-band coherence (and its distribution within the band), some degree of “between-subject” variability was observed. Although the γ-band coherence was above a significance level for many subjects, there were other subjects for whom the EMG-EMG coherence values in the γ-band were relatively small (and not significant). Additionally, between-subject variability was observed in the distribution of the coherence within the γ-band; for instance, the peak frequency of the γ-band intermuscular coherence was found in the lower γ-band range for some subjects (∼40 Hz), whereas for others it was found in the upper γ-band (∼50 Hz) or in both. Such between-subject variability could also have contributed to the small γ-band coherence values observed in the grand-average coherence spectra (Fig. 5). Regardless, given a large degree of between-subject variability in the data, the results from the γ-band coherence analysis (i.e., significant “between-condition” difference of CI_γ) need to be interpreted with caution.

Interestingly, the task type (extension vs. flexion) also had a significant effect on the intermuscular coherence between the homologous muscles. In contrast to the feedback effect, however, the task type only influenced the coherence values in the β-band. Note that previous studies have consistently demonstrated that the β-band intermuscular coherence mainly originates from the motor cortex (Conway et al. 1995; Halliday et al. 1998). The observed strong task effect on β-band intermuscular coherence, therefore, would support the notion that the type of target task (flexion vs. extension) mainly affected the planning of the bimanual movements at the motor cortex. There is some evidence that the task type could affect intermuscular coherence between homologous muscles (during bimanual task) or even between distal and proximal muscles in the ipsilateral side (during unilateral task). Previous studies (Ridderikhoff et al. 2005; Boonstra et al. 2007) found that intermuscular coherence values between homologous extensor muscles during fatiguing contraction are higher than those between flexor muscles, although the difference was observed in a different band (i.e., α-band). On the other hand, our previous study (Lee et al. 2014) found that the elbow flexion, but not the elbow extension, significantly reduces intermuscular coherence in all bands between the ipsilateral hand muscles, indicating that the elbow flexion task reduces common neural input to different muscles on the ipsilateral side; similarly, the elbow flexion may also reduce common input to the muscles in the contralateral side. Alternatively, the observed task dependency in the intermuscular coherence could simply originate the physiological difference in the elbow flexor and extensor muscles; for instance, intrinsic excitability of the triceps brachii muscle is significantly higher than that of the biceps brachii muscle (Wilson et al. 2015). Taken together, the difference in the neural control of the elbow flexor and extensor muscles and/or their physiological characteristics may account for the observed difference in the intermuscular coherence in the β-band, i.e., different motor planning strategy for these two different groups of muscles.

Dissimilarities in the neural control of the elbow flexor and extensor muscles are also indicated by their different phase effects. In addition to the aforementioned between-task differences, there was also a significant phase effect on the intermuscular coherence between the homologous triceps brachii muscles (their coherence values were significantly higher during the ramp vs. the hold phase), but not between the biceps brachii muscles (Table 1). The observed phase effects on the triceps brachii muscles indicate that common neural inputs to the homologous triceps brachii muscles increased during force development (force increase; ramp phase) and then decreased during steady-state force production (hold phase). Interestingly, the observed phase effects on coherence values were somewhat contrasting to previous studies that examined intermuscular coherence during isometric force production in which a decrease in intermuscular coherence was observed during the ramp phase (force development) (Kilner et al. 1999). In this study, the authors postulated that the observed decrease in intermuscular coherence was needed to “suppress oscillatory activity” during movements and/or during production of varying levels of force, whereas the degree of common input to multiple muscles was increased during the hold phase to simplify the control. Such contrasting results could be because of the difference between this previous study and ours; note that, in Kilner et al. (1999), intermuscular coherence was only calculated between muscles within the ipsilateral side. Also, the intermuscular coherence values were estimated for all possible muscle pairs and averaged, which makes it difficult to assess whether such phase effects were observed across all muscle pairs. Our results showed that, regardless of the task (flexion or extension), common neural input to homologous triceps brachii muscles is increased during the ramp phase; this suggests that the triceps brachii muscles may have acted as a stabilizer during production of varying levels of force (Franklin and Milner 2003).

Implications.

The findings of this study, particularly the increase in the neural coupling between homologous muscles by means of somatosensory feedback, could have direct implications on designing training for stroke survivors. In recent years, there has been considerable interest in using bilateral task training in unilaterally impaired stroke patients to enhance its rehabilitative outcome (Luft et al. 2004; Cauraugh and Summers 2005; Stoykov et al. 2009; Lewis and Perreault 2009), since bilateral training is thought to have a “facilitation effect from the nonparetic arm to the paretic arm” (Whitall et al. 2000). In theory, task conditions that maximize the coupling between homologous muscles would achieve the largest effect on the training outcome. Our results strongly support the use of somatosensory (or haptic) feedback during bilateral training, which could effectively increase the degree of neural coupling between homologous arm muscles. Previous studies have indeed demonstrated that sensory feedback greatly affects the performance of bimanual tasks for both healthy (Gooijers et al. 2011) and stroke (Torre et al. 2013) survivors. Furthermore, the observed between-task difference in the intermuscular coherence values (flexion vs. extension) also suggests that the target task itself may also affect the neural coupling between homologous muscles during training, with elbow extension tasks generating higher intermuscular coherence values.

The observed pattern of between-condition variation in coherence values, with significant differences observed in the α- and γ-bands, also strongly supports the use of somatosensory feedback in stroke rehabilitation. Brain stem pathways, such as reticulospinal pathways, were postulated to play an important role in the functional recovery of arm (Wolf et al. 1989; Schwerin et al. 2008; Zaaimi et al. 2012) and hand (Baker 2012) following stroke, and the ipsilateral (uncrossed) projection from the reticulospinal pathway following stroke was postulated to largely explain the impaired multijoint coordination of the arm (Schwerin et al. 2008). These observations also support the use of somatosensory feedback during bilateral training for stroke survivors, since it may promote the use of ipsilateral (and contralateral) brain stem pathways largely involved in the functional use of the affected arm after stroke.

GRANTS

This work was in part supported by U.S. Army Medical Research and Materiel Command Grant W81XWH-05-1-0160.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

H.B.N. and S.W.L. performed experiments; H.B.N., S.W.L., and P.S.L. analyzed data; H.B.N., S.W.L., and P.S.L. drafted manuscript; H.B.N., S.W.L., M.L.H.-L., and P.S.L. edited and revised manuscript; H.B.N., S.W.L., M.L.H.-L., and P.S.L. approved final version of manuscript; S.W.L., M.L.H.-L., and P.S.L. interpreted results of experiments; S.W.L. and P.S.L. prepared figures.