The neuromuscular control of human movement can be described by a set of muscle synergies factorized from myoelectric signals. There is some evidence that the selection, activation and flexible combination of these basic activation patterns are of a neural origin. We investigated the muscle synergies during incline and level walking to evaluate changes in the modular organization of neuromuscular control related to changes in the mechanical demands. Our results revealed five fundamental (not further factorizable) synergies for both walking conditions but with different frequencies of appearance of the respective synergies during incline compared with level walking. Low similarities across conditions were observed in the timing of the activation patterns (motor primitives) and the weightings of the muscles within the respective elements (motor modules) for the synergies associated with the touchdown, mid-stance and early push-off phase. The changes in neuromuscular control could be attributed to changes in the mechanical demands in support, propulsion and medio-lateral stabilization of the body during incline compared with level walking. Our findings provide further evidence that the central nervous system flexibly uses a consistent set of neural control elements with a flexible temporal recruitment and modifications of the relative muscle weightings within each element to provide stable locomotion under varying mechanical demands during walking.
Based on the high redundancy of the motor system, a single motor task can be performed in multiple ways with a similar end result (Bernsteĭn, 1967). The individual control of every single muscle to perform each specific motor task or its variations would imply a very high complexity of the human brain. For this so-called ‘degrees of freedom problem’ Bernsteĭn (1967) developed a well-accepted theory of motor control simplification, in which the complexity of the high movement diversity is reduced by combining a lower number of basic muscle activation patterns to accomplish a specific motor task (Chhabra and Jacobs, 2006). There is some evidence that the selection, activation and flexible combination of these basic patterns, called muscle synergies, are of a neural origin and follow a modular organization (Bizzi and Cheung, 2013; Bizzi et al., 2008; Cheung et al., 2009b; d'Avella et al., 2003; Roh et al., 2011). These basic muscular activation patterns are proposed to be adjusted in their specific timing and force generation (Latash et al., 2010; Martin et al., 2009). It is commonly accepted that motor behaviors with a similar task complexity demonstrate the same number and congruent shapes of muscle synergies (Chhabra and Jacobs, 2006). From this, a changing number of synergies may be related to changes in the biomechanical demands of the motor tasks (Clark et al., 2010). Assuming a neural origin of muscle synergies, changes in the dynamic demands might be represented in the modular organization of the neuromuscular system. This could probably result in changes of the number of muscle synergies and/or the basic shape of the activation patterns as well as in alterations of the specific muscle contributions to the respective synergies.
Previous studies on level walking using this approach report highly consistent results for four (Oliveira et al., 2012) to five synergies and their time structure across individuals (Cappellini et al., 2006; Chvatal and Ting, 2013; Ivanenko et al., 2006a; Sasaki and Neptune, 2006). The time structure of the synergies, also referred to as ‘motor primitives’ (Dominici et al., 2011) is also consistent across different walking speeds (Ivanenko et al., 2004, 2006a,b). Furthermore, the specific muscle contributions (motor modules) to the respective synergies have been found to change with varying loading conditions (Ivanenko et al., 2004; McGowan et al., 2010). In their experimental study, Ivanenko et al. (2004) show that, although the individual muscle activation differed with changes in the velocity and in the gravitational load, the modular control of the muscles has more or less similar basic patterns. Using a musculoskeletal model, McGowan et al. (2010) provide evidence showing that the timing of the modular control of human walking is unaffected by different mechanical demands and that a flexible modulation of motor modules may robustly reproduce human walking.
In everyday life, changes in the environment, such as slope, require an appropriate recruitment and coordination of muscles and therefore are important factors in human locomotion. For walking on inclined surfaces, changes in the resultant joint torque characteristics compared with level walking have been demonstrated, especially for the touchdown and push-off phases (Devita et al., 2008; Lay et al., 2006; McIntosh et al., 2006; Silder et al., 2012). These mechanical differences result in changes in the demands on the activation magnitude and/or durations of the respective muscles (Fischer et al., 2015; Franz and Kram, 2012; Lay et al., 2007). Analyzing potential differences in muscle synergies during incline versus level walking could provide further insights in motor strategies applied under different environmental walking conditions. In particular, it might help to understand modular control strategies and adjustments during human locomotion in relation to changes of the dynamic demand in everyday life.
The purpose of the current study was to investigate the modular control of human locomotion during incline and level walking using the muscle synergies concept. To compare both movement conditions more quantitatively, we first calculated the similarities using the coefficient of determination (R2) and then set their repeatability (intraday) and reproducibility (interday) thresholds. We hypothesize differences in the modular organization of the neuromuscular system between incline and level walking, primarily based on changes in the specific muscle contributions to the respective synergies. By contrast, we expect the same number of synergies as well as rather consistent time structures for both walking tasks, based on the previously reported high robustness of these parameters across various walking conditions (Cappellini et al., 2006; Chvatal and Ting, 2013; Ivanenko et al., 2004, 2006a,b; Sasaki and Neptune, 2006).
MATERIALS AND METHODS
Ten female and 10 male young adults (Table 1) volunteered for the experiments. All participants were right dominant and physically active. They did not use orthotic insoles and had no known history of neurological or motor disorders or injuries over the last 6 months prior to the measurements. All participants had common but no advanced experience in incline walking. Written informed consent was given before participation and the local ethical committee approved the procedures.
To obtain the individual preferred walking speed for each participant, we used the method of limits (Treutwein, 1995) performed on a treadmill. Starting at 0.8 m s−1, the speed was randomly increased by 0.02 to 0.05 m s−1 at varying time intervals of 5–10 s. After the participant confirmed the preferred speed, the procedure was repeated by decreasing the speed starting from a pace 0.5 to 1.0 m s−1 higher than the preferred speed. The preferred speed was calculated as the mean speed of both tests. If the difference between tests was larger than 10%, the test was repeated. Thereafter all participants randomly performed two trials of incline and two of level walking on the treadmill on each of two measurement days resulting in eight trials per participant. Time between trials on the same day was 10 min, time between measurement days was at least 48 h (137±92 h). During this time, the participants were asked not to change their daily routine and not to have any hard training sessions on the day prior to the measurements. Level walking was performed at the individual preferred speed (1.4±0.2 m s−1). Incline walking was performed at a slope of 10 deg (Devita et al., 2008) and 85% of the preferred speed. In each trial, recordings began after a preliminary movement stabilization phase of around 60 s.
The activity of 24 muscles on the right-hand side of the body was recorded by surface EMG. The pairs of Ag/AgCl electrodes (N-00-S, Ambu, Denmark) for bipolar derivation were applied according to SENIAM (surface electromyography for the non-invasive assessment of muscles) standards (Hermens et al., 2000). Signals were recorded at 1000 Hz and 16 bit A/D converter resolution using two synchronized wireless EMG systems (one 16-channel and one 8-channel, Myon AG; Switzerland). The EMG systems had a built-in band-pass filter (5–500 Hz, 3 dB/oct, 4th order). Lower leg muscles recorded were gastrocnemius medialis (GM) and lateralis (GL), soleus (SO), peroneus longus (PL) and tibialis anterior (TA). Upper leg and hip muscles recorded were vastus medialis (VM) and lateralis (VL), rectus femoris (RF), semitendinosus (ST), tensor fasciae latae (TFL), adductor longus (AL), the long head of the biceps femoris (BFS) and gluteus medius (GMED) and maximus (GMAX). Trunk EMG signals were obtained from the lower part (L2) of the erector spinae (ESL2), the rectus abdominis (RA), anterior external obliquii (AEO), latissimus dorsi (LD) and trapezius (TRS). In addition, biceps brachii (BB), the long head of triceps brachii (TB), the anterior (DA) and posterior (DP) head of the deltoideus and the splenius capitis (SPL) muscles were recorded. During the walking tasks 50 (49±4) consecutive gait cycles were measured (Oliveira et al., 2014). Plantar pressure distributions were measured at 120 Hz by a pressure plate (FDM-THM-S, Zebris Medical, Germany) integrated in the treadmill (Mercury, H-p-cosmos Sports & Medical, Germany). The pressure plate data were acquired using the proprietary software (WinFDM-T v2.5.1, Zebris Medical) and then extracted in a raw format for autonomous post-processing using a validated custom algorithm (Santuz et al., 2016). The pressure plate was synchronized with the EMG data using an analogue signal.
EMG signals were high-pass filtered, fully rectified and low-pass filtered at cut-off frequencies of 50 and 20 Hz respectively using 4th order zero-lag IIR Butterworth filters. Each EMG envelope was time-normalized to 200 data points per gait cycle (Cheung et al., 2009a). The 50 cycles of each trial were averaged (Oliveira et al., 2014) and the amplitude was normalized by setting the highest peak for each muscle of each participant for each measurement day to one (Bizzi et al., 2008; Devarajan and Cheung, 2014; Karamanidis et al., 2004). The gait cycles (i.e. stance and swing phase) were obtained from the pressure plate. The touchdown and toe-off were determined from the ground reaction forces using a validated custom post-processing algorithm (Santuz et al., 2017), where the touchdown was identified as the first non-zero pressure matrix after the last toe-off. In addition to the contact times, the pressure data were used to quantify step lengths, cadence, vertical ground reaction forces (VGRFs) and total impulse of the stance phase for the gait cycles. For each step, the mean VGRF of the first and second half of the stance phase was calculated and the total impulse was obtained by integrating the VGRFs over the contact time. Forces and impulses were then related to body weight (BW). Again, the data of 50 gait cycles for each trial were averaged.
The limit of convergence was reached when a change in the calculated R2 between the original matrix V and the reconstructed matrix VR was smaller than 0.01% in the last 20 iterations (Cheung et al., 2005). This was done for a number of 1 to 10 synergies. The computation was repeated 10 times for each synergy, each time creating new randomized initial matrices, in order to avoid local minima (d'Avella and Bizzi, 2005). These results were plotted as a curve of R2 versus the number of synergies. To determine the number of synergies required to represent the original EMG signals, we applied a widely used method based on the cross-validation of the R2 values obtained for each number of synergies (Cheung et al., 2005). The curve of R2 versus the number of synergies was fitted by a simple linear regression model that initially uses all 10 synergies. The mean squared error was then calculated. This procedure was repeated, each time removing the lowest point of the R2 curve for the applied linear fitting until the amount of the mean squared error was less than 10−5. The procedure described above resulted in a specific number of synergies for each of the total 160 trials (20 participants, 2 days, 4 trials per day) necessary to sufficiently reconstruct the respective original EMG dataset.
Based on the shape of the motor primitives, we classified fundamental and combined synergies. A fundamental synergy can be defined as an activation pattern that cannot be factorized any further. When two (or more) fundamental synergies are blended into one, a combined synergy appears. An example of combined synergies is shown in Fig. 1.
The fundamental synergies recognition was implemented using a custom learning algorithm based on a curve-fitting model. In the first implementation step, examples of single-peak activation patterns, which might represent a fundamental primitive, are chosen. The code is then fed by these manually selected examples of fundamental primitives and a search of similar shapes is done across the whole dataset of factorized curves. With a first iteration, the primitives that have a high similarity (R2>0.95) with the ones present in the manually created database are added to the set. The number of fundamental primitives is then selected by looking at the motor modules and merging possible repetitions. After updating the database, the code starts the recognition across the entire dataset searching, synergy by synergy, for similar primitives (we found R2>0.5 to be a good threshold in this phase). Non-recognized curves can then be visually inspected with an interactive routine or automatically identified as new fundamental or combined primitives. This approach, validated in a pilot study, can reproduce the results of a completely manual selection of the curves with a margin of error of ±5%.
R2 was calculated to assess the similarity of the motor primitives and motor modules, respectively, between incline and level walking. In our analysis, we only evaluated the fundamental synergies, leaving out the combined synergies (Fig. 1). The outcomes are the mean R2 values of day 1 and day 2 obtained by averaging the two trials from day 1 and the two trials from day 2, and comparing incline (first condition) and level walking (second condition). Assuming that muscle synergies are neurophysiological entities (Bizzi and Cheung, 2013), this analysis was performed separately for each synergy identified by the NMF. Since we recorded two trials per day on two different days for each condition, we could investigate the repeatability and reproducibility of the intraday and interday measurements. The intraday similarities were computed for both conditions by comparing the two trials of each measurement day. The interday similarities were found by comparing the mean of the first with the mean of the second day's two trials. For classification purposes, we calculated a similarity threshold defined by the means of four intraday R2 similarities (both measurement days and both walking conditions). The difference between incline and level walking was classified as relevant when the R2 similarity between the walking conditions for the respective synergies was lower than the similarity threshold defined above. This allowed for the identification of global differences between walking conditions in the motor primitives and motor modules within a specific synergy.
Wilcoxon signed-rank tests were used to identify differences in the mean number of muscle synergies (factorization rank) and the frequencies of appearance of each of the synergies observed during incline versus level walking. An ANOVA for repeated measures was performed for the similarities of the motor primitives and motor modules, respectively. If the normality assumption necessary for the validity of the ANOVA was not satisfied, the Wilcoxon test was used.
For any other comparison, for example, the pressure plate data, we used a paired samples t-test or a Wilcoxon signed-rank test when the data did not satisfy the normality assumptions. All levels of significance were set to α=0.05. All the statistical analyses were conducted using R v3.2.2 (R Foundation).
All participants demonstrated significantly (P<0.01) longer contact time (0.564±0.442 s versus 0.531±0.031 s), shorter (P<0.001) step length (0.64±0.07 m versus 0.71±0.07 m) and lower (P<0.01) cadence (step frequency) of 106±8 to 114±6 steps min−1 when comparing incline with level walking. The mean values of the left and right VGRFs in the first half of the stance phase were significantly lower (P<0.001) in the incline condition. Expressing the mean VGRF per body weight (BW), participants showed 0.92±0.1 BW during incline compared with 0.99±0.08 BW level walking. The total impulse related to body weight was significantly (P=0.001) higher during incline (0.52±0.04 BW s) compared with level walking (0.50±0.04 BW s).
No significant differences for the number of synergies were observed across walking conditions in the NMF and reconstruction quality criteria (P=0.30). During incline walking, 5.0±0.6 synergies were required, and during level walking, 4.8±0.8 synergies were necessary to reconstruct the original dataset. The mean reconstruction quality of the NMF represented by the coefficient of determination (R²) was 0.90 and 0.92 for incline and level walking, respectively. Across all 160 trials, we also observed no significant differences (P=0.22) in the number of recognized fundamental synergies during incline and level walking. In both conditions, up to five fundamental synergies were observed (Figs 2 and 3).
The first synergy (Syn 1, peak at ∼4% of the gait cycle) was strongly related to foot contact and the early stance phase (Figs 2 and 3). As shown by the motor modules, Syn 1 mainly represented the activation of the hip extensors (GMED, GMAX) and knee extensors (VM, VL, RF). The second synergy (Syn 2, peak at ∼40% of the gait cycle) is associated with the mid-stance and the beginning of the push-off phase, with large contributions of the plantar flexor muscles (GM, GL, SOL and PL). The third synergy (Syn 3, peak at ∼50% of the gait cycle) occurred at the push-off phase. Dominated by the back extensor (ESL1) it mainly accounts for the upper body muscles (LD, DP, TB). The fourth synergy (Syn 4, peak at ∼60% of the gait cycle) mainly appeared during the transition from stance to swing and ranged until the mid-swing phase. During incline walking, Syn 4 represented the knee extensors and abductors (RF, TFL, AL) as well as the LD muscle. During level walking, the contributions additionally focused on the trunk muscles (SPL, RA, AEO). Syn 5 (peak at ∼90% gait cycle) was related to the late swing phase including the preparation of the next foot contact and represented the activation of the knee adductors and flexors (ST, BFL) and the dorsi flexors (TA). Fig. 2 also shows the motor primitives of the fundamental synergies for each trial of the two walking conditions separately, including the specific frequencies of appearance. Significantly higher frequencies of appearance (P=0.02) were observed for Syn 1 and Syn 2, with values of 99% and 98% during incline compared with 70% and 86% during level walking, respectively (Fig. 2). Syn 3 was observed in 72% and 74%, respectively. Syn 4 demonstrated a significantly higher (P=0.04) frequency of appearance of 21% during incline compared with 5% during level walking, whereas Syn 5 was present in 73% and 71% of all incline and level walking tasks, respectively. The combined synergies, which are not presented in these graphs, demonstrated no significant differences in the frequency of appearance (i.e. 29% for incline and 30% for level walking, P=0.67) across both walking conditions.
Muscle synergies similarities
The R2 was used to assess the similarities between incline and level walking for the motor primitives and motor modules. Based on the intraday and interday R2 similarities (Table 2), we defined synergy-specific thresholds for the motor primitives and motor modules. These thresholds incorporate any bias included in the process of measuring, pre-processing and factorizing the EMG signals.
Table 3 reports the similarity values when comparing the incline and level walking conditions. The similarities of the motor primitives (H) for the synergies one, two, three and five are lower than the respective thresholds. For the motor modules, values lower than the thresholds were observed for the synergies one, two and five. The number of recognized fundamental primitives for synergy four was so small that the similarity analysis could not be performed because of missing data (indicated as NA in Tables 2 and 3).
As expected, we observed a consistent number of fundamental synergies for each of both walking tasks, indicating a convergent global activation pattern within the general movement type of locomotion. In both walking tasks, we observed up to five fundamental synergies in young adults. However, in many walking trials, not all of the fundamental synergies could be observed, thus resulting in different frequencies of appearance. For instance, Syn 4 was only present in 5% of the level and 21% of the incline walking conditions. In these cases, the minimum number of fundamental synergies needed to reconstruct the original EMG signals (i.e. the factorization rank) was four or less. In addition, the frequencies of appearance of the fundamental synergies partly resulted from the presence of combined synergies. During incline walking, the frequencies of appearance of the muscle synergies Syn 1, 2 and 4 were significantly increased compared with level walking.
To date, there is no explicit interpretation regarding changes in the frequency of appearance of muscle synergies. However, motor activity that is critical to successfully achieve a motor task has been found to be less variable compared with motor activity that does not directly contribute or interfere with the task (Bernsteĭn, 1967). Therefore, it can be argued that the nervous system focuses on the regulation of motor outputs that are directly related to task goals (Ting et al., 2015). From the neural perspective of modularity, an increase in the appearance of fundamental synergies may represent a more differentiated control of specific movement tasks (Clark et al., 2010; Ting et al., 2015). Our findings support the idea of a flexible use of five fundamental synergies during incline and level walking. The higher frequencies of appearance of Syn 1, 2 and 4 during incline walking could reflect a more constrained locomotor drive when dealing with activities that include higher mechanical demands. In fact, during incline walking, more muscle work has to be done to lift-up and to stabilize the body (Fischer et al., 2015; Franz and Kram, 2012; Lay et al., 2007). The greater resultant joint torques during incline walking (Devita et al., 2008; Silder et al., 2012), especially in the touchdown and push-off phase, correspond to the significantly larger vertical impulses in incline walking observed in our data. Furthermore, the four-times higher frequency of appearance of Syn 4 during incline walking might be associated with the higher mechanical demand to stabilize the trunk because of its increased forward lean (Prentice et al., 2004). The higher demand to stabilize the trunk during incline walking may increase the need to more distinctly attribute the required muscle activation to a specific synergy compared with level walking, where the activation is distributed across or included in other synergies.
While walking, the leg muscles can be characterized by three major functions: (1) generation of support during touchdown by counteracting gravity, (2) generation of propulsion, accelerating the body forward and (3) control of the mediolateral balance during each step (Perry, 1967; Winter, 1995). In the comparison of incline and level walking, the motor primitives (time structure of muscle activation patterns) and the motor modules (muscle weightings) of the first, second and fifth synergy demonstrated relevant dissimilarities across both walking conditions. The differences found in the muscle synergies in the early and mid-stance phase, as well as during the push-off phase between incline and level walking, provide evidence for adaptive modifications of the neuromuscular control to account for increased mechanical demands during incline walking. It has been shown that muscles generating vertical support and forward propulsion, such as the gluteus medius, vasti medialis and lateralis, soleus and gastrocnemii, also substantially contribute to the mediolateral stability (Pandy et al., 2010). Furthermore, the hip adductors and the plantarflexor everters accelerate the center of mass laterally, controlling the mediolateral balance during the stance phase of walking (Perry and Burnfield, 2010). Comparing joint kinetics during incline and level walking, Lay et al. (2006) reported increased knee and hip extensor moments in the support phase, as well as increased plantar flexor moments in the propulsion phase. The substantially higher muscular demand in incline compared with level walking also increases the external hip and knee adduction moments (Karamanidis and Arampatzis, 2009), indicating a greater need to stabilize the body in the mediolateral direction during incline walking.
The main changes in the modular control between incline and level walking have been found during the stance phase (Syn 1 and Syn 2) and in the preparation for the stance phase (Syn 5), indicating that the increased muscular demand was the origin for the alteration in the neural control elements. The alterations were observed in the upper leg muscles for Syn 1 and in the lower leg muscles for Syn 2, probably to generate support and control balance during touchdown and to generate propulsion during the mid-stance and the beginning of the push-off phase. The modifications in Syn 5 indicate a preparation for the stance phase, which is specific to incline walking, where the larger motor module in TA presumably reflects the increased activity to lift the tip of the foot higher at the end of the swing phase.
The aim of our study was to investigate the modular control of human locomotion during incline and level walking using the muscle synergy concept. As hypothesized, our results revealed changes in the modular organization of muscle activity during incline compared with level walking. The different modular organization was displayed in changes in the frequency of occurrence of specific muscle synergies, as well as changes in the shape of the motor primitives (time-dependent activation patterns) and the motor modules (specific muscle contribution within the respective synergies), which are both represented by lower R2 similarities with respect to the found thresholds. The observed changes are associated with the increased mechanical demand in incline compared with level walking. Our findings provide evidence showing that the central nervous system flexibly uses a consistent set of neural control elements (e.g. up to five fundamental synergies). A flexible temporal recruitment of motor primitives, as well as a flexible modification of the motor modules within each element, is used to achieve stable locomotion adapted to the mechanical demand during walking. Although there are several studies investigating clinical outcomes using the muscle synergies concept in the diagnosis and rehabilitation of pathologies, for example, in stroke (Cheung et al., 2012) and Parkinson's disease (Falaki et al., 2016), a direct translation to training and rehabilitation procedures is still limited. In the future, a better understanding of neuromuscular mechanisms through the analysis of muscle synergies may help to reliably identify motor control strategies, which is essential to improve the specificity of training and rehabilitation protocols in healthy and pathological population.
The authors are grateful to all participants, who always showed great commitment and interest during the measurements.
Conceptualization: L.J., A.S., A.E. and A.A.; Methodology: L.J., A.S. and A.A.; Investigation: L.J., A.S. and A.E.; Formal Analysis: L.J. and A.S.; Writing – Original Draft: L.J.; Writing – Review & Editing: L.J., A.S., A.E. and A.A.; Visualization: L.J. and A.S.; Supervision: A.A.
This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
The authors declare no competing or financial interests.