Muscle Synergies in People With Chronic Ankle Instability During Anticipated and Unanticipated Landing-Cutting Tasks

Participants

A total of 11 people with CAI (CAI group; 5 men, 6 women; age = 22 ± 3 years, height = 1.68 ± 0.11 m, mass = 69.0 ± 19.1 kg) and 11 people without CAI serving as a healthy control group (CON group; 5 men, 6 women; age = 23 ± 4 years, height = 1.74 ± 0.11 m, mass = 66.8 ± 15.5 kg) were recruited to participate in the study. Initial screening and inclusion in the CAI group was based on a questionnaire (Modified Ankle Instability Instrument) that established the history and severity of previous ankle sprains.^20,21 Recruits were excluded if they had a history of bilateral ankle sprains, lower extremity fractures or knee injuries, or both. In addition, the Foot and Ankle Disability Index (FADI; CAI group: 90.3 ± 0.4; CON group: 100 ± 0.0) and FADI-Sport (CAI group: 88.6 ± 9.1; CON group: 100 ± 0.0) questionnaires were used to assess ankle joint function,²² and Tegner scores (CAI group: 5.3 ± 1.2; CON group: 5.3 ± 1.0) were used to quantify the physical activity levels of participants. All individuals provided written informed consent, and the study was approved by the Human Research Protection Program of the University of Michigan.

Procedures

We recorded ground reaction forces (GRFs) using a force platform (AMTI) and muscle activations using a desktop surface EMG system (model Bagnoli; Delsys). Muscle EMG and GRF data were collected at sampling frequencies of 1200 Hz. Five single-differential EMG electrodes (model DE-2.1; Delsys) were attached to the soleus (SL), MG, lateral gastrocnemius (LG), TA, and FL muscles. The bars on the electrodes were 99.9% silver, 10 mm in length, and 1 mm in width and were separated by 10 mm. The skin was cleaned using alcohol swabs, and EMG sensors were placed according to standard recommendations.²³ To limit cross-talk, we determined the final location of the EMG sensors by asking participants to perform isolated muscle contractions. After a brief warm-up, they were instructed to execute up to 5 trials each of an anticipated landing-cutting task (Ant) and an unanticipated landing-cutting task (Unant). For both tasks, they performed a forward jump over a 15-cm box and onto a force plate. The distance between the initial position and the force plate was normalized to each person's leg length, which was defined as the distance between the anterior-superior iliac spine and the medial malleolus of the same limb. Individuals were instructed to land on their involved limb and perform a 90° cutaway from their landing limb as quickly as possible after landing on the force plate. For the CON group, the involved limb was defined as the dominant limb, which was the limb they used to kick a ball. During Ant, the direction of the cut was given to the participant before each jump. During Unant, the direction of the cut was indicated via an electronic signal displayed on a computer screen, which was positioned at waist height in front of the force plate. The signal was triggered when the person broke a light beam that was projected from a light gate, which was positioned halfway between the initial start position and the force plate. Successful trials were defined as those that were correctly performed according to the instructions and cutting direction and in which the foot was placed entirely on the force plate. Data from 1 participant in the CON group were excluded because of problems with the GRF data. In addition, 2 people in the CAI group were unable to perform Unant. Based on these exclusions, we analyzed a total of 123 trials (ie, 22 participants × 2 tasks × 3 trials − 9 bad trials).

Data Processing

Data were analyzed from the stance phase of each task. The stance phase of the landing-cutting tasks was based on GRF thresholds of 10 N for both touchdown and takeoff. The EMG data were low-pass filtered using a cutoff frequency of 450 Hz and high-pass filtered using a cutoff frequency of 20 Hz. The filtered EMG data were rectified and smoothed using a low-pass filter at a cutoff frequency of 10 Hz. The smoothed activation data for each muscle were normalized to the maximum activation observed during all trials and time normalized to 101 data points such that 0% represented touchdown and 100% represented takeoff (Figure 1).²⁴ Each muscle's activation was further divided by its SD to obtain the unit variance so that NMF could extract equally weighted muscle synergies.²⁵ The smoothed and normalized muscle activation data from each participant and each task were organized into a 5 × 303 input matrix (ie, 5 rows for all 5 muscles and 303 columns for 3 trials of 101 data points). The NMF algorithm decomposed the 5 × 303 input matrix (E) into the synergy vector (W) and activation coefficient (C): \(\def\upalpha{\unicode[Times]{x3B1}}\)\(\def\upbeta{\unicode[Times]{x3B2}}\)\(\def\upgamma{\unicode[Times]{x3B3}}\)\(\def\updelta{\unicode[Times]{x3B4}}\)\(\def\upvarepsilon{\unicode[Times]{x3B5}}\)\(\def\upzeta{\unicode[Times]{x3B6}}\)\(\def\upeta{\unicode[Times]{x3B7}}\)\(\def\uptheta{\unicode[Times]{x3B8}}\)\(\def\upiota{\unicode[Times]{x3B9}}\)\(\def\upkappa{\unicode[Times]{x3BA}}\)\(\def\uplambda{\unicode[Times]{x3BB}}\)\(\def\upmu{\unicode[Times]{x3BC}}\)\(\def\upnu{\unicode[Times]{x3BD}}\)\(\def\upxi{\unicode[Times]{x3BE}}\)\(\def\upomicron{\unicode[Times]{x3BF}}\)\(\def\uppi{\unicode[Times]{x3C0}}\)\(\def\uprho{\unicode[Times]{x3C1}}\)\(\def\upsigma{\unicode[Times]{x3C3}}\)\(\def\uptau{\unicode[Times]{x3C4}}\)\(\def\upupsilon{\unicode[Times]{x3C5}}\)\(\def\upphi{\unicode[Times]{x3C6}}\)\(\def\upchi{\unicode[Times]{x3C7}}\)\(\def\uppsy{\unicode[Times]{x3C8}}\)\(\def\upomega{\unicode[Times]{x3C9}}\)\(\def\bialpha{\boldsymbol{\alpha}}\)\(\def\bibeta{\boldsymbol{\beta}}\)\(\def\bigamma{\boldsymbol{\gamma}}\)\(\def\bidelta{\boldsymbol{\delta}}\)\(\def\bivarepsilon{\boldsymbol{\varepsilon}}\)\(\def\bizeta{\boldsymbol{\zeta}}\)\(\def\bieta{\boldsymbol{\eta}}\)\(\def\bitheta{\boldsymbol{\theta}}\)\(\def\biiota{\boldsymbol{\iota}}\)\(\def\bikappa{\boldsymbol{\kappa}}\)\(\def\bilambda{\boldsymbol{\lambda}}\)\(\def\bimu{\boldsymbol{\mu}}\)\(\def\binu{\boldsymbol{\nu}}\)\(\def\bixi{\boldsymbol{\xi}}\)\(\def\biomicron{\boldsymbol{\micron}}\)\(\def\bipi{\boldsymbol{\pi}}\)\(\def\birho{\boldsymbol{\rho}}\)\(\def\bisigma{\boldsymbol{\sigma}}\)\(\def\bitau{\boldsymbol{\tau}}\)\(\def\biupsilon{\boldsymbol{\upsilon}}\)\(\def\biphi{\boldsymbol{\phi}}\)\(\def\bichi{\boldsymbol{\chi}}\)\(\def\bipsy{\boldsymbol{\psy}}\)\(\def\biomega{\boldsymbol{\omega}}\)\(\def\bupalpha{\bf{\alpha}}\)\(\def\bupbeta{\bf{\beta}}\)\(\def\bupgamma{\bf{\gamma}}\)\(\def\bupdelta{\bf{\delta}}\)\(\def\bupvarepsilon{\bf{\varepsilon}}\)\(\def\bupzeta{\bf{\zeta}}\)\(\def\bupeta{\bf{\eta}}\)\(\def\buptheta{\bf{\theta}}\)\(\def\bupiota{\bf{\iota}}\)\(\def\bupkappa{\bf{\kappa}}\)\(\def\buplambda{\bf{\lambda}}\)\(\def\bupmu{\bf{\mu}}\)\(\def\bupnu{\bf{\nu}}\)\(\def\bupxi{\bf{\xi}}\)\(\def\bupomicron{\bf{\micron}}\)\(\def\buppi{\bf{\pi}}\)\(\def\buprho{\bf{\rho}}\)\(\def\bupsigma{\bf{\sigma}}\)\(\def\buptau{\bf{\tau}}\)\(\def\bupupsilon{\bf{\upsilon}}\)\(\def\bupphi{\bf{\phi}}\)\(\def\bupchi{\bf{\chi}}\)\(\def\buppsy{\bf{\psy}}\)\(\def\bupomega{\bf{\omega}}\)\(\def\bGamma{\bf{\Gamma}}\)\(\def\bDelta{\bf{\Delta}}\)\(\def\bTheta{\bf{\Theta}}\)\(\def\bLambda{\bf{\Lambda}}\)\(\def\bXi{\bf{\Xi}}\)\(\def\bPi{\bf{\Pi}}\)\(\def\bSigma{\bf{\Sigma}}\)\(\def\bPhi{\bf{\Phi}}\)\(\def\bPsi{\bf{\Psi}}\)\(\def\bOmega{\bf{\Omega}}\)\begin{equation}\tag{1}E = \mathop \sum \limits_{i = 1}^{{N_{{\rm{synergy}}}}} {W_i}{C_i} + \varepsilon ,{\rm }\end{equation}

View Figure

• Download as Powerpoint
• Download Figure

where N_synergy is the number of muscle synergies and ε is the residual error.²⁵

The NMF algorithm was then applied iteratively with an increasing number of muscle synergies (eg, in the case of 1 synergy, the algorithm would extract 1 set of synergy vectors and 1 activation coefficient). The NMF algorithm and output with 3 synergies, 3 sets of synergy vectors, and 3 activation coefficients are depicted in Figure 2. The total variance accounted for (VAF_Total; Equation 2) and variance accounted for by each muscle (VAF_Each; Equation 3) were calculated iteratively with continuously larger numbers of synergies until the following criteria were met: (1) VAF_Total ≥ 90% and (2) VAF_Each ≥ 75%.^25,26 In most datasets, 3 synergies were the appropriate number of muscle synergies for the criteria. In addition, the VAF values indicated the extent to which the EMG profile of all muscles (VAF_Total) and each individual muscle (VAF_Each) could be explained by the respective number of synergies. Muscle synergies were sorted based on timing of the peak activation coefficient and cosine similarity of synergy vectors²⁷: \begin{equation}\tag{2}{\rm{VA}}{{\rm{F}}_{{\rm{Total}}}} = 1 - {{\mathop \sum \nolimits_{i = 1}^p \mathop \sum \nolimits_{j = 1}^n {{\left( {{{\rm{\varepsilon }}_{i,j}}} \right)}^2}} \over {\mathop \sum \nolimits_{i = 1}^p \mathop \sum \nolimits_{j = 1}^n {{({E_{i,j}})}^2}}},{\rm }\end{equation}where p is the number of muscles and n is the number of time points, and \begin{equation}\tag{3}{\rm{VA}}{{\rm{F}}_{{\rm{Eac}}{{\rm{h}}_m}}} = 1 - {{\mathop \sum \nolimits_{j = 1}^n {{\left( {{{\rm{\varepsilon }}_{m,j}}} \right)}^2}} \over {\mathop \sum \nolimits_{j = 1}^n {{\left( {{E_{m,j}}} \right)}^2}}},{\rm }\end{equation}where m is each muscle.

View Figure

• Download as Powerpoint
• Download Figure

Statistical Analysis

The independent variables were group (CAI and CON) and task (Ant and Unant). The dependent variables were the number of muscle synergies, the VAF_Total of each synergy, and the muscle-specific weightings from each of the extracted synergies. In addition, we used the cosine-similarity values and zero-lag cross-correlation coefficients to assess whether muscle synergy data from each group and task were similar (ie, CON Ant versus Unant, CAI Ant versus Unant, CON versus CAI Ant, CON versus CAI Unant) and thus appropriate for statistical comparison. Specifically, the zero-lag cross-correlation coefficients were used to evaluate the similarity of each synergy's activation coefficient and ensure that the activation pattern of each synergy (ie, time-variant components) captured similar bursts of muscle activity. Similarly, we used the cosine-similarity values to determine the similarity of each synergy vector (ie, weighting coefficients) and ensure that the muscles that were active in each synergy (ie, time-invariant components) were the same. Cross-correlation and cosine-similarity values >0.80 were considered to exhibit high similarity.²⁷ Normality of all dependent variables was checked using the Jarque-Bera test.²⁸ The Wilcoxon rank sum test was conducted to compare the number of synergies between the CAI and CON groups separately for each task. Separate 2 × 2 analyses of variance were calculated to compare the VAF_Total and muscle-specific weightings for each of the extracted synergies between groups (CAI and CON) and across tasks (Ant and Unant). We used MATLAB (version 2019a; The MathWorks, Inc) and set the α level at .05.

Dimensionality of Muscle Synergies

Based on the VAF results, we identified 2 to 4 synergies as the appropriate number to represent the EMG data of each trial. In most cases, 3 synergies were sufficient to reconstruct the EMG data (Figure 3), and the average VAF_Total for 3 synergies was approximately 93%. We observed no difference in the number of synergies expressed by the CAI and CON groups during either task (CON Unant: 3.1 ± 0.6; CAI Unant: 2.8 ± 0.4; CON Ant: 3.1 ± 0.6; CAI Ant: 3.1 ± 0.5; Figure 4). In addition, no interaction or main effects were found for VAF_Total with 1 synergy (CON Unant: 46.6% ± 15.9%; CAI Unant: 51.0% ± 6.9%; CON Ant: 55.4% ± 13.7%; CAI Ant: 51.2% ± 7.6%), 2 synergies (CON Unant: 81.2% ± 9.3%; CAI Unant: 83.5% ± 4.6%; CON Ant: 84.6% ± 7.1%; CAI Ant: 85.4% ± 5.1%), 3 synergies (CON Unant: 92.0% ± 3.6%; CAI Unant: 93.3% ± 3.2%; CON Ant: 94.0% ± 2.1%; CAI Ant: 94.1% ± 2.7%), or 4 synergies (CON Unant: 97.4% ± 1.3%; CAI Unant: 97.8% ± 1.3%; CON Ant: 97.8% ± 0.9%; CAI Ant: 97.9% ± 1.2%; Figure 5).

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

Similarity of Muscle Synergies

The cosine-similarity values for the synergy vectors of synergies 1 and 2 were >0.80 for all group and task comparisons (Table 1). In contrast, the cosine-similarity values for the synergy vectors of synergy 3 were <0.80 for 3 of 4 group comparisons (Table 1). Specifically, the cosine-similarity value for the synergy vectors of the comparison between CON and CAI for Unant was 0.85. The zero-lag cross-correlation coefficients of all activation coefficients from all respective group and task comparisons were >0.80 (Table 2).

Table 1 Cosine-Similarity Coefficients for Each Synergy Vector Comparison (Mean ± SD)

View Fullscreen

Table 2 Zero-Lag Cross-Correlation Coefficients for Each Activation Coefficient Comparison (Mean ± SD)

View Fullscreen

Functional Interpretation of Muscle Synergies

Muscles were considered part of a specific synergy (ie, activated or recruited by that synergy) if their weighting coefficients were >0.3.²⁹ The activation coefficient of synergy 1 reflected muscle activation during the early (approximately 10% to 20%)- and late (approximately 80% to 100%)-stance phases of the landing-cutting tasks. The muscle-specific weightings in the synergy vector of synergy 1 reflected mainly TA activation. Therefore, it seems that synergy 1 functions to control the ankle angle at ground contact and during late stance. The activation coefficient of synergy 2 captured muscle activation during the middle-stance phase (approximately 30% to 80%) of the landing-cutting tasks, and the muscle-specific weightings in the synergy vector associated with this synergy indicated that it reflected primarily FL activation. Given that the FL acts mainly to evert the ankle in the frontal plane, the function of synergy 2 thus seems to be related to the transition from forward to lateral motion during the midportion of the landing-cutting tasks. Similarly, the activation coefficient of synergy 3 captured muscle activation in the middle-stance phase of the landing-cutting tasks. Unlike synergy 2, however, the muscle-specific weightings in the vector for synergy 3 were associated with activation of the SL, MG, and LG. Hence, synergy 3 seems to play a role in propulsion and helps accelerate the body toward the new cutting direction.

Group and Task Differences in Muscle Synergies

We observed no group × task interactions for any of the individual muscle weightings for any of the synergy vectors and synergies. However, main effects existed for group in the weightings for the SL and TA in the synergy vector of synergy 1 (Figures 6 and 7). Specifically, the task-averaged weightings of the TA were larger (P = .02) in the CAI than the CON group, whereas the task-averaged weightings of the SL were smaller (P = .03) in the CAI than the CON group. We also noted main effects for task in the individual muscle weightings of the MG and FL in the synergy vector of synergy 2 (Figures 6 and 7). The group-averaged weightings of MG were larger (P = .03) during Ant than Unant, and the group-averaged weightings of FL were smaller (P = .03) during Ant than Unant.

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

Dimensionality of Muscle Synergies

The dimensionality of muscle synergies did not differ between the CAI and CON groups. This finding did not agree with our hypothesis that people with CAI would exhibit fewer synergies or that each muscle synergy would exhibit a greater VAF_Total. These hypotheses were based on the work of researchers¹⁵ who suggested that people with certain neurologic conditions (eg, stroke, cerebral palsy) exhibit fewer muscle synergies and simpler neuromuscular control strategies. Fewer muscle synergies, as observed in people with neurologic conditions, may be due to greater cocontractions and result in less efficient movement.^15,30 However, given that people in the CAI group exhibited the same number of muscle synergies during both landing-cutting tasks, it appears that they used a similar synergy-based neuromuscular control strategy during landing-cutting tasks regardless of the associated cognitive load.

Different Weightings of Muscle Synergies

Analysis of the individual muscle weightings in each synergy reflected both group and task main effects. Specifically, in synergy 1, the weighting of the TA muscle was greater for the CAI than the CON group, which indicated that people with CAI recruited the TA muscle to a greater extent than people in the CON group. Given that the activation coefficient of synergy 1 captured muscle activity during the early phase of stance during the landing-cutting tasks, individuals with CAI may emphasize sagittal-plane positioning of the ankle around touchdown to a greater extent, which is clinically important because ankle positioning affects foot-ground clearance and mitigates the risk of unanticipated contact.^31,32 In addition, computer simulations have suggested that greater ankle dorsiflexion at the instant of foot contact is associated with a smaller external moment arm of GRFs about the subtalar joint, which could lessen the risk of subsequent ankle sprains.³³ Moreover, a dorsiflexed position increases stability of the ankle joint because the surfaces of the ankle joint become more congruent as dorsiflexion increases. People with CAI are thought to compensate for their lack of stability by dorsiflexing the ankle joint to achieve a more close-packed and stable position.⁶ Lastly, greater emphasis on TA activation during the early stance phase may also enhance coactivation and ankle-joint stability. For example, using a musculoskeletal model, DeMers et al³⁴ showed that increased cocontraction of the ankle invertors and evertors might improve ankle joint stability and decrease ankle injury risk. However, the interpretation related to the greater weighting of the TA by the CAI group could be deemed less protective with respect to the control of frontal-plane motion because the TA also inverts the foot at the ankle. This interpretation therefore provides an alternative explanation of the results and may indicate that the CAI group also exhibited altered frontal-plane control. Thus, it may be necessary to investigate ankle joint kinematics and kinetics in future studies to support or refute this interpretation. Although group differences existed in the weighting of the SL muscle for synergy 1, the magnitudes of this weighting were very small (ie, well below the 0.3 threshold) and hence may not reflect clinically or functionally important differences in muscle coordination between the CON and CAI groups.²⁹

In addition to the group difference between CON and CAI in individual muscle weightings for synergy 1, we found a task difference between Ant and Unant for synergy 2. Specifically, both groups displayed greater weighting of the FL muscle and less weighting of the MG muscle during Unant than Ant. Based on the timing of muscle activity, the activation coefficient for synergy 2 suggests that people used this synergy to control muscle activation during the middle phase of stance during landing-cutting tasks. The increase in FL weighting, therefore, may reflect greater emphasis on transitioning from forward to lateral motion via greater activation of frontal-plane muscles. Furthermore, greater FL weighting may also suggest an attempt to maintain balance and ankle joint stability in the frontal plane when the landing-cutting task is performed with more uncertainty and without knowing the direction of movement.³⁵ Interestingly, the increase in FL weighting was accompanied by a decrease in MG weighting; although based on the thresholds, the changes in MG weighting in synergy 2 indicate that the MG muscle would be considered active during Ant but not during Unant. Because the MG muscle is an important contributor to propulsive forces during landing-cutting tasks,³⁶ this change in neuromuscular control may mean that both groups emphasized frontal-plane stability over propulsion during Unant.

Limitations and Future Directions

Our study had several limitations. First, the sample recruited was relatively small (only 22 participants), which may have affected the results, and the procedures may warrant replication in future studies with a larger sample. Second, we recorded EMG from only 5 lower limb muscles. Using a small number of muscles in the NMF methods may have led to an overestimation of the VAF by the muscle synergies, which may in turn have resulted in the extraction of fewer muscle synergies and subsequently affected the interpretation of their dimensionality.³⁷ However, the minimum number of muscles for appropriate implementation of the NMF methods is considered 4.³⁷ Moreover, the 5 muscles in our study captured the major kinesiologic functions of the ankle joint (ie, plantar flexion, dorsiflexion, inversion, and eversion) and thus likely still adequately represented neuromuscular strategies of ankle motions from a muscle synergy-based control perspective. Considering that researchers³⁸ have reported that people with CAI exhibited functional deficits in the proximal muscles, it would be valuable to include these muscles (eg, gluteus maximus) in future studies to better characterize and more comprehensively understand the functional role of muscle synergies in people with CAI. Third, the muscles that were investigated are located relatively close to each other, which may have led to cross-talk and affected the EMG signal. Using indwelling electrodes in future research may help reduce the potential for cross-talk via more direct measurement of a muscle's activation. Fourth, the muscle activation magnitude was normalized based on the maximum value observed across all trials and tasks. Whereas the use of maximal voluntary isometric contractions represents another normalization method, the maximum observed value is commonly used in NMF research^29,39 and may be acceptable in our study because dynamic landing and cutting tasks impose high task demands that may actually elicit near-maximum muscle activation levels.⁴⁰ Similarly, normalizing the EMG to values obtained from a quasistatic position (eg, a standing trial) may be another option that is less sensitive to group differences in activation than values obtained during dynamic movements. Fifth, we analyzed only landing-cutting tasks in a laboratory setting. Given that ankle sprains also occur during other movements (eg, walking) and in other environments (eg, uneven or inclined surfaces), assessing muscle synergies across a variety of tasks and conditions may provide additional information that could hold important clinical implications.