ABSTRACT
Muscle synergies as functional low-dimensional building blocks of the neuromotor system regulate the activation patterns of muscle groups in a modular structure during locomotion. The purpose of the current study was to explore how older adults organize locomotor muscle synergies to counteract unpredictable and predictable gait perturbations during the perturbed steps and the recovery steps. Sixty-three healthy older adults (71.2±5.2 years) participated in the study. Mediolateral and anteroposterior unpredictable and predictable perturbations during walking were introduced using a treadmill. Muscle synergies were extracted from the electromyographic activity of 13 lower limb muscles using Gaussian non-negative matrix factorization. The four basic synergies responsible for unperturbed walking (weight acceptance, propulsion, early swing and late swing) were preserved in all applied gait perturbations, yet their temporal recruitment and muscle contribution in each synergy were modified (P<0.05). These modifications were observed for up to four recovery steps and were more pronounced (P<0.05) following unpredictable perturbations. The recruitment of the four basic walking synergies in the perturbed and recovery gait cycles indicates a robust neuromotor control of locomotion by using activation patterns of a few and well-known muscle synergies with specific adjustments within the synergies. The selection of pre-existing muscle synergies while adjusting the time of their recruitment during challenging locomotor conditions may improve the effectiveness to deal with perturbations and promote the transfer of adaptation between different kinds of perturbations.
INTRODUCTION
Investigating neuromotor control during gait perturbations in older adults might help us to better understand real-life mobility in this age group. After an unpredictable gait perturbation, more than five recovery steps can be required to return to baseline walking stability (Epro et al., 2018; Oddsson et al., 2004). Fast and effective motor responses in order to maintain stability are of high importance after an unpredictable perturbation (Karamanidis et al., 2020; Maki and McIlroy, 2006), as beneficial mechanisms modulated by anticipation and prediction (Patla, 2003) are not available. Short latency reflexes, 45–70 ms, possibly mediated by subcortical centers (McDonagh and Duncan, 2002; Quintern et al., 1985; Van Der Linden et al., 2007), rapidly generate neuromuscular responses after a locomotor perturbation, triggering postural corrections (Cronin et al., 2009; Nieuwenhuijzen and Duysens, 2007). Prior experience with a certain type of perturbation enhances the opportunity to proactively plan the desired movement with ongoing adaptative adjustments in order to reduce the consequences of the perturbation (Bierbaum et al., 2010; Marigold and Patla, 2002; Pai et al., 2003). However, the short monosynaptic spinal feedback loops such as the stretch and the Golgi tendon reflexes are relatively unaffected by prior experience (Nieuwenhuijzen and Duysens, 2007; Rothwell et al., 1986). In contrast, longer feedback loops such as long latency reflexes which involve supraspinal mechanisms (Taube et al., 2006) can be modified by prior experience to enhance the effectiveness of the task-specific neuromuscular output (Wolpert and Landy, 2012). Experience-based learning leads to sensorimotor adaptations, which are supposedly mediated by the cerebellum (Diedrichsen et al., 2005; Morton and Bastian, 2006).
Muscle synergies (Bizzi et al., 1991) have been frequently used to investigate the neuromotor control of human locomotion in unperturbed and perturbed conditions (Cappellini et al., 2006; Dominici et al., 2011; Santuz et al., 2018a; Ting et al., 2015). Muscle synergies as functional low-dimensional building blocks of the neuromotor system regulate the activation patterns of muscle groups and coordinate the complex behavior of human locomotion in a modular structure (Dominici et al., 2011; Ivanenko et al., 2004). There is a linkage between muscle synergies and the connectivity of spinal premotor interneurons as well as motor cortical neurons (Amundsen Huffmaster et al., 2018; Hart and Giszter, 2010; Takei et al., 2017), which provides evidence that muscle synergies are important neurophysiological entities for motor control and learning (Bizzi and Cheung, 2013). Muscle synergies as neural control modules may be shared across a large repertoire of motor tasks (Chvatal and Ting, 2013; D'Avella and Bizzi, 2005) and flexibly recruited by descending neural pathways from supraspinal regions (Roh et al., 2011), which would be an efficient strategy to generate complex motor behavior. The modular architecture of muscle synergies simplifies motor control and facilitates an effective execution of multiple motor tasks (D'Avella et al., 2003; D'Avella and Bizzi, 2005). It has been found that the spatiotemporal structure and composition of a small number of muscle synergies is generated in a robust manner to account for the changing biomechanical demands of the task and alterations of the neural and musculoskeletal systems during locomotion (Cheung et al., 2020; Yang et al., 2019).
Four basic locomotor muscle synergies associated with distinct gait phases (i.e. weight acceptance, propulsion, early swing and late swing) were reported for steady-state unperturbed walking (Clark et al., 2010; Dominici et al., 2011; Lacquaniti et al., 2012; Santuz et al., 2018a). In the presence of consecutive external mechanical perturbations, as for example during walking on slippery or uneven terrains, the structure and number of the muscle synergies is retained, while a widening and variable recruitment of their basic activation patterns serves as a compensatory strategy to deal with the perturbations (Martino et al., 2015; Santuz et al., 2018a). Chvatal and Ting (2012) found that a common set of muscle synergies could explain the muscle activation patterns of the perturbed step during both anteroposterior and mediolateral gait perturbations. In contrast, Oliveira et al. (2012) and Nazifi et al. (2017) reported that unperturbed and perturbed walking have only a fraction of shared synergies and that the muscle weights of unperturbed walking only marginally agreed with the perturbed ones. All the above-mentioned studies investigated young participants with prior experience with the induced gait perturbations. Little is known about the spatiotemporal structure and composition of the muscle synergies and the influence of prior experience in older adults. Gait stability after unpredictable and novel perturbations is deteriorated to a greater degree in older adults than in young ones (Bierbaum et al., 2010; Maki and McIlroy, 2006). Furthermore, older adults usually need more recovery steps until complete recovery (Süptitz et al., 2013). Nevertheless, the potential to adapt based on prior experience is still maintained in older adults (Bierbaum et al., 2011; Bohm et al., 2015; Marigold and Patla, 2002).
A fast recruitment of muscle synergies related to specific functional phases of walking after a gait perturbation might be crucial for a successful recovery. The activation of a common set of well-known locomotor muscle synergies while adjusting the time of their recruitment may provide a more effective way to counteract locomotor perturbations than the generation of new muscle synergies. The purpose of the current study was to explore how older adults organize locomotor muscle synergies to counteract predictable and unpredictable gait perturbations during the perturbed steps and the recovery steps. For this purpose, we investigated their responses to anteroposterior and mediolateral perturbations with and without prior experience during walking on a treadmill and we measured the electromyographic (EMG) activity of 13 muscles of the perturbed leg. We hypothesized that older adults would use the four basic muscle synergies of unperturbed walking (weight acceptance, propulsion, early swing and late swing) and adjust the time of their recruitment during the perturbed and recovery steps. We further hypothesized that the spatiotemporal adjustments in muscle synergies would be less pronounced during predictable perturbations with prior experience than during the unpredictable perturbations without prior experience.
MATERIALS AND METHODS
Participants
A sample size calculation was conducted in G*Power 3.1 (Faul et al., 2007) for a repeated-measure within-factors ANOVA, with a power of 0.9, a moderate effect size (f=0.25), 11 groups (number of comparisons between the included gait cycles), two measurements (predictable versus unpredictable) and an α-level of 0.05. The calculation resulted in a required sample size of 55 participants. To take into account potential dropouts during the demanding measurements, we considered 63 healthy older adults in this study (33 females, 30 males; age 71.2±5.2 years, height 171.3±8.9 cm, body mass 71.6±11.9 kg, means±s.d.). Inclusion criteria were an age of 65–80 years, home dwelling and the ability to walk independently on a treadmill for at least 20 min. Participants were excluded if they had neurological, musculoskeletal or cardiovascular disorders, uncorrected visual impairments, severe vertigo or a body mass index ≥30 because of potential issues with the EMG recordings (Nordander et al., 2003). All participants gave their written consent to the experimental procedure. This study was reviewed and approved by the Ethics Committee of Heidelberg University (AZ Schw 2018 1/2).
Experimental protocol
In the current study, all experiments were conducted during walking on a particular perturbation treadmill (BalanceTutor™, MediTouch LTD, Netanya, Israel; Fig. 1). The treadmill can introduce two types of perturbations: anteroposterior and mediolateral. Mediolateral perturbations always consisted of a fast shift of 11.7 cm of the treadmill platform to the left (acceleration: 1.3 m s−2, duration of acceleration: 0.3 s, duration of deceleration: 0.3 s). Anteroposterior perturbations consisted of an alteration of the speed of the belt of 0.82 m s−1 (acceleration: 4.1 m s−2, duration of acceleration: 0.2 s, duration of deceleration: 0.2 s). For safety reasons, participants were strapped into a harness system, which was set to prevent the knees from touching the ground in case of a fall. The participants were randomly divided into two groups (A and B) by coin toss, where each group experienced the perturbations in a different order with regard to the directions.
After a familiarization session, which included two sets of 10 min unperturbed treadmill walking, the participants were invited to the main experiment on a separate day. The experiment started with a warm-up of 2–5 min walking with a gradual speed progression until the target speed of 1.1 m s−1 was reached. This speed was chosen based on normative data of the included age group (Hollman et al., 2011). In the case of an induced perturbation, the participants were requested to regain balance without stopping walking. In the first perturbation protocol, we introduced an unpredictable perturbation (mediolateral for group A and anteroposterior for group B), which was unknown in time and type (Fig. 1). We continued the protocol with three perturbations of the same type that were also unannounced. Subsequently, an unpredictable perturbation for each group in the other direction occurred (anteroposterior for group A and mediolateral for group B). We finalized the protocol with three predictable anteroposterior and three predictable mediolateral perturbations for group A, and accordingly three predictable mediolateral and three predictable anteroposterior perturbations for group B. In the predictable perturbations, the time and direction of the perturbation were announced on a screen in front of the participants. In the analysis, we selected the first unknown perturbation per direction for the unpredictable condition (trials 1 and 5) and the third announced perturbation per direction for the predictable condition (trials 8 and 11) as the participants already had experience with this kind of perturbation. All perturbations were induced in mid-stance of the right leg, as detected by automatic elaboration of the data of four built-in force sensors (sampling frequency: 60 Hz) integrated in the perturbation treadmill.
For each of the four perturbed conditions (unpredictable mediolateral, predictable mediolateral, unpredictable anteroposterior, predictable anteroposterior), 12 consecutive gait cycles surrounding the selected perturbations were included in the analysis. This series of cycles included three cycles before the perturbation, the perturbed cycle and eight cycles after the perturbed cycle. This specific number of recovery cycles was selected based on a preliminary analysis of the gait parameters (cadence, stance duration, swing duration and duty factor), which revealed that up to seven recovery cycles were significantly affected by the perturbation.
Measurements and data analysis
We recorded the EMG activity of 13 ipsilateral (right side) muscles of the lower limb at 2 kHz using a 16-channel wireless bipolar surface EMG system (aktos, myon AG, Schwarzenberg, Switzerland), with foam-hydrogel electrodes with snap connector (H124SG, Medtronic PLC, Dublin, Ireland). The following muscles were included: gluteus medius, gluteus maximus, tensor fasciae latae, rectus femoris, vastus medialis, vastus lateralis, semitendinosus, biceps femoris, tibialis anterior, peroneus longus, gastrocnemius medialis, gastrocnemius lateralis and soleus. To reduce cross-talk between EMG signals, we positioned the electrodes on the muscle belly (De Luca et al., 2012; Hermens et al., 2000). Additionally, we checked every signal individually before the measurements by asking the participants to perform movements related to the function of the specific muscle. If the plausibility or quality of the signals was not sufficient, electrodes were repositioned before another check procedure was performed.
As a measure of variability, we further calculated the coefficient of variance, defined as the quotient of the standard deviation to the mean of the group for each gait cycle in every condition, for all gait parameters (i.e. time of the stance and swing phase, cadence and duty factor).
From the EMG data, we extracted the muscle synergies for the investigated trials using a custom-written R script (http://www.R-project.org/) that utilizes the classical Gaussian non-negative matrix factorization (NMF) algorithm (Lee and Seung, 1999; Santuz et al., 2017). The raw EMG signals were filtered in the acquisition device by a band-pass filter with cut-off frequencies of 10 and 500 Hz. Then, the signals were first high-pass filtered using a 4th-order IIR Butterworth zero-phase filter with a cut-off frequency of 50 Hz, and subsequently the signals were full-wave rectified before a 4th-order IIR Butterworth zero-phase low-pass filter with a cut-off frequency of 20 Hz was applied. The amplitude of the EMG signals for the extraction of the muscle synergies was normalized to the maximum EMG value determined separately for each muscle and each trial (i.e. max. of the 12 consecutive gait cycles) for each participant, after subtracting the minimum. Each single gait cycle of each condition (12 cycles×4 conditions) was analyzed separately for each participant. To increase the robustness of the factorization, the NMF was repeated 20 times for each cycle and then averaged. Each gait cycle was then normalized in time to 200 points, where stance and swing phase each got an equal distribution of 100 points. This transformation was conducted to increase comprehension and interpretation as a result of improved comparability between different participants, conditions and gait phases (Santuz et al., 2018b).
The extraction of the muscle synergies through the NMF was conducted as previously reported by our group (Santuz et al., 2020, 2018a). At first, a matrix V=m×n was created from the original data, where m is the 13 recorded muscle activities and n is the number of normalized time points (i.e. 200×1 cycle). This matrix was factorized utilizing the NMF so that V≈VR=MPT, where VR is a new approximation of the matrix V, which is reconstructed through multiplication of the two matrices M and P. P contains the time-dependent activation patterns of a synergy (Dominici et al., 2011; Santuz et al., 2017) and has the dimensions p×n, where the number of rows p represents the minimum number of synergies necessary to satisfactorily reconstruct the original set of signals V. The matrix M contains the muscle weights, which are the time-invariant weights of the synergies (Gizzi et al., 2011; Santuz et al., 2017), and has the dimensions m×p. Muscle synergies were extracted separately by cycle and condition, so that activation patterns, muscle weights and the factorization rank (i.e. the number of extracted synergies) were independently generated for each single gait cycle. The quality of the reconstruction of the original data matrix through the extracted synergies was assessed utilizing the coefficient of determination R2 as described earlier by Santuz et al. (2017).
Fundamental synergies, defined as activation patterns with a single main activation peak (Santuz et al., 2018a), were classified as described in detail by Munoz-Martel et al. (2021). In short, muscle synergies were classified using a method based on k-means clustering. Thirteen iterations (number of included muscles) were computed building 1–13 clusters from the activation patterns of all participants for every single cycle of each condition separately. For each iteration, the sum of squares within the contained clusters was determined. Then, a curve of the sum of squares versus the number of clusters was built. The minimum number of clusters before the curve fits a linear interpolation with a mean squared error below 10−3, was used as a criterion to determine the classification. Muscle synergies that were not classified were excluded from further analysis.
To compare the activation patterns of the muscle synergies between conditions, we additionally used the full width at half-maximum (FWHM) and the center of activity (CoA), which are established metrics for the width and timing of the basic activation patterns (Cappellini et al., 2016; Martino et al., 2014; Santuz et al., 2020). The FWHM was calculated for each cycle, adding up all points exceeding the half-maximum of the activation patterns, after subtracting the minimum (Martino et al., 2014). The CoA was also calculated cycle by cycle as the angle of the vector in polar coordinates that points to the center of mass of that circular distribution (Cappellini et al., 2016). The polar direction represented the cycle's phase, with angle 0≤θ≤2π.
Statistics
For the parameters cadence, stance time, swing time, duty factor, FWHM, CoA and CS, a linear mixed-effects model was used with predictability (unpredictable, predictable) and gait cycle as fixed factors and participant as a random factor. Normal distribution of the normalized residuals was examined by visual inspection of the respective histograms, demonstrating no severe violations of the normality assumption. The modeling was conducted using the function ‘lme’ of the R package ‘nlme’ v.3.1-152 (https://CRAN.R-project.org/package=nlme), with maximizing the restricted log-likelihood and for both perturbation directions (mediolateral or anteroposterior) separately. In case of significant effects of a factor or an interaction between factors predictability and gait cycle, a post hoc analysis was performed comparing all included cycles with the third cycle prior to the perturbation as the one with the higher temporal distance with the upcoming perturbation. For the CS, the post hoc comparisons were performed with regard to the CS of the first and second unperturbed cycles as a representative metric for unperturbed walking. For all analyzed parameters, differences between unpredictable and predictable cycles were investigated by pairwise comparisons of the matching cycles (e.g. cycle 1 in unpredictable versus cycle 1 in predictable). The P-values for all post hoc tests were adjusted for multiple comparisons according to Benjamini–Hochberg. Effect sizes are provided for statistically significant differences as Cohen's d calculated based on the pooled standard deviation, where 0.2≤d<0.5 indicates small, 0.5≤d<0.8 moderate and d≥0.8 large effects. All these analyses were conducted in R v.4.1.0.
To check for a proactive modulation of muscle activity in the predictable trials, statistical parametric mapping (SPM) paired two-tailed t-tests (Pataky, 2010) were performed on the muscle's EMG activity during the stance phase comparing the predictable and unpredictable conditions for both mediolateral and anteroposterior perturbations. The initiation of the perturbation occurred in the middle of the stance phase; thus, any proactive adjustments of muscle activity in the predictable trials might be visible in the EMG activity during the first half of the stance phase. For the SPM, the EMG activity of each muscle was normalized to the maximum activity during unperturbed walking (specifically to the third cycle prior to the perturbation). The SPM was conducted in MATLAB R2017b (The MathWorks Inc., Natick, MA, USA) using the spm1d-package (v.M.0.4.10). The level of significance was set at α=0.05 for all analyses.
RESULTS
Detailed results of the linear mixed-effect models and all post hoc tests are provided in the Tables S1 and S2. We found a significant effect of predictability (P<0.001) and gait cycle (P<0.001) on cadence, stance time, swing time and duty factor in both mediolateral and anteroposterior perturbations (Table 1). The post hoc analysis showed that all four temporal gait parameters were significantly affected by the perturbation in at least one and up to seven cycles (P<0.05, d=0.31 to 4.21) in each of the four perturbed conditions compared with the respective unperturbed gait cycle (Table 1). The three cycles before the perturbation did not show any statistically significant differences (P=0.228 to 0.998, d=0.00 to 0.15) in the temporal parameters in either condition (Table 1). Cadence generally increased, while stance and swing durations decreased as a result of the perturbation. The duty factor increased in the perturbed cycle but decreased in the following recovery cycles compared with unperturbed walking. The observed alterations in the gait parameters during the unpredictable condition were significantly higher in both perturbation directions in two and up to seven recovery gait cycles (P<0.05, d=0.27 to 3.46) compared with the predictable one (Table 1). The coefficient of variation was increased in all conditions in all gait parameters during and after the perturbation.
An example of the EMG activity in the unperturbed, perturbed and recovery gait cycles of every recorded muscle is presented in Fig. 2. In all gait cycles of all conditions, we identified four muscle synergies that were able to sufficiently reconstruct the recorded EMG data (R2=0.89±0.03 for mediolateral unpredictable, R2=0.88±0.03 for mediolateral predictable, R2=0.90±0.03 for anteroposterior unpredictable and R2=0.89±0.02 for anteroposterior predictable). The four synergies are functionally associated with the phases of the gait, namely the weight acceptance synergy with high activation of knee extensors and glutei, the propulsion synergy with a main contribution of plantar flexors, the early swing synergy mainly represented by the foot dorsiflexors and the late swing synergy highly related to the knee flexors (Fig. 3). From all muscle synergies extracted in the current study, 18.7% were non-classifiable.
We found a statistically significant effect of gait cycle (P<0.007) on the CS of the activation patterns (Fig. 4) and muscle weights (Fig. 5) in all synergies, which shows a modulation of the spatiotemporal structure of the muscle synergies during and after the perturbation (Fig. 6). The CS of the activation patterns decreased (P<0.001, d=0.87 to 4.10) in all synergies and conditions during the perturbed cycle compared with the unperturbed one (Fig. 4). The CS of the muscle weights was also significantly decreased (P<0.04, d=0.47 to 2.12) in most of the synergies during the perturbation and the first recovery cycle (Fig. 5). Furthermore, there was a statistically significant effect of predictability (P<0.05, d=0.49 to 1.91) on almost all CS of the activation patterns and muscle weights, with lower values in the unpredictable condition in both the mediolateral and anteroposterior direction (Figs 4 and 5). Only in the mediolateral weight acceptance synergy of the muscle weights was the decrease of CS in the predictable condition not statistically significant (P=0.12), while a significant (P<0.001) interaction effect of predictability by gait cycle was found in that synergy. The significant effect of predictability on the similarity of the activation patterns was present until the third–fourth recovery cycle (Fig. 4).
We found a significant effect of cycle (P<0.02) and predictability (P<0.02) on the CoA for nearly all synergies in both the mediolateral and anteroposterior direction (Table 2). Only the early swing synergy showed no significant effect of predictability (P<0.349), but an interaction effect of cycle by predictability (P<0.001). In the perturbed cycle and the first recovery cycle, the CoA of the activation patterns shifted either earlier or later within the gait cycle compared with the unperturbed walking, with greater changes in the unpredictable condition (Table 2). The shift of the CoA was statistically significant in all synergies and conditions (P<0.05, d=0.53 to 4.80) except for the late swing synergy in the predictable anteroposterior condition. The alterations in the CoA lasted until the fifth recovery cycle (Table 2). There was an effect of gait cycle (P<0.005) and a cycle by predictability interaction (P<0.005) on the FWHM of nearly all synergies in both the mediolateral and anteroposterior direction. In most cases, the FWHM of the activation patterns became wider (Table 3). The significant differences compared with unperturbed walking were mainly observed in the perturbed cycle and in the first recovery cycle for all investigated conditions. The increased widening of the activation patterns during the perturbed cycle and first recovery cycle was in most cases larger in the unpredictable condition (Table 3). There was a decrease in FWHM of the propulsion synergy in the first recovery cycle after the unpredictable anteroposterior condition (Table 3).
The SPM analysis showed significant differences (P<0.05) of EMG activity before the initiation of the perturbation between the predictable and unpredictable conditions (Fig. 7). In both mediolateral and anteroposterior perturbations, the rectus femoris, vastus medialis and vastus lateralis showed a significantly higher EMG activity (P≤0.002) before the perturbation in the predictable compared with the unpredictable condition. During mediolateral perturbations, the biceps femoris (P=0.049), peroneus longus (P<0.001) and gastrocnemius lateralis (P<0.001) showed a significantly higher EMG activity in the predictable condition, whereas in the anteroposterior perturbations, the EMG activity of the semitendinosus (P=0.006), tibialis anterior (P=0.003) and gastrocnemius medialis (P=0.033) was higher (Fig. 7).
DISCUSSION
In the current study, we investigated the effects of predictable and unpredictable perturbations on the modular organization of locomotion in older adults. For both the muscle weights and activation patterns, we found a significant decrease in CS between the perturbed cycle and the following recovery cycles compared with unperturbed walking, indicating a modulation of the spatiotemporal structure of muscle synergies by the sensorimotor system in order to maintain gait stability after the perturbations. The number of muscle synergies remained unchanged in both predictable and unpredictable perturbations and the four basic walking synergies (weight acceptance, propulsion, early swing and late swing) were preserved, confirming our first hypothesis. Further, we found a smaller decrease in CS during the predictable compared with the unpredictable perturbations, indicating a less pronounced modulation of the spatiotemporal structure of muscle synergies, confirming our second hypothesis. The findings were generally consistent for both anteroposterior and mediolateral perturbations and thus invariant to the type of perturbation.
The reduced CS in all activation patterns in the perturbed gait cycle evidenced a clear modification of the temporal components of the muscle synergies. Similar to the activation patterns, there was a modification of the muscle weights during the perturbed cycle in most cases, indicating a flexible organization of the spatial components of muscle synergies during the perturbed gait. Although the spatiotemporal structure of muscle synergies was modified during gait perturbations, the basic structure and composition were preserved and the four basic synergies found in unperturbed walking were enough to counteract the challenging perturbations in both predictable and unpredictable trials. The common set of muscle synergies was also maintained during the mediolateral perturbations, whereas the gait biomechanics deviated from the unperturbed fore–aft walking. During the mediolateral gait perturbations, a fast lateral step or a crossing step is needed to counteract a sideways fall (Madehkhaksar et al., 2018). Our results showed that the participants' acute lateral steps were performed by modulating the basic activation patterns of the four synergies for unperturbed walking, leaving the number of synergies unchanged. These findings are in line with previous studies investigating continuous and less intense perturbations during walking (Martino et al., 2015; Santuz et al., 2018a), where preservation of the number of synergies was observed. In more intensive anteroposterior and mediolateral gait perturbations (i.e. platform translation of 12.4 cm with 40 cm s−1 velocity), a preservation of the number of muscle synergies has also been reported (Chvatal and Ting, 2012). In our applied perturbations, the mediolateral translation of the treadmill platform was 11.7 cm with a velocity of 39 cm s−1, and the anteroposterior perturbations consisted of an alteration of the belt velocity of 82 cm s−1, which indicates perturbation intensities comparable to the last mentioned study (Chvatal and Ting, 2012). Nevertheless, the perturbations did not introduce falls and all participants were able to govern their stability by modulating the spatiotemporal components of the unperturbed walking synergies. In more intensive walking perturbations sufficient to reproduce falls, some studies found a decrease in the number of muscle synergies, particularly in the participants who fell (Sawers et al., 2017; Sawers and Bhatt, 2018). The authors interpreted the finding as a deficient neuromotor control for the recruitment of an appropriate number of muscle synergies to avoid the fall (Sawers et al., 2017; Sawers and Bhatt, 2018).
The recruitment of the four basic walking synergies in the perturbed and recovery gait cycles indicates a robust synergy structure and a simplification of postural stability control by using activation patterns of a few, well-known muscle synergies with specific adjustments within the synergies. Stability control via recruitment of existing muscle synergies during perturbed walking with the integration of somatosensory information may reduce the need for higher-level control and promote the transfer of adaptation between different kinds of perturbations (König et al., 2022). A recent study in rats reported a robust spinal structure of muscle synergies during walking, which was preserved despite different developmental and learning processes (Yang et al., 2019). Takei et al. (2017) showed that spinal premotor interneurons generate the spatiotemporal components of muscle synergies during a precision motor task in monkeys and suggested that superimposed corticomotoneuronal pathways are responsible for more fractionated motor control. The gait perturbations applied in the current study increased the risk of falling and triggered fast reactive recovery responses, which might require less involvement of the corticomotoneuronal system. The recruitment of shared synergies for the control of different motor tasks has been reported previously (D'Avella and Bizzi, 2005) and was suggested to be a general strategy of the neuromotor system to flexibly and efficiently create complex motor behavior. Using a simulation approach, d'Avella and Pai (2010) found that a modular controller was able to adapt faster to applied force perturbations if the neuromotor control system used a set of pre-existing muscle synergies. Berger et al. (2013) reported higher adaptation rates in a novel motor task that can be executed with pre-existing synergies compared with a task that required new muscle synergies. Chvatal and Ting (2012) found modifications in the temporal recruitment of common locomotor muscle synergies during experienced gait perturbations in young participants. Our study expands those findings showing that older adults used a common number of muscle synergies and adjusted their temporal recruitments to maintain stability in both predictable and unpredictable gait perturbations with and without prior experience. The activation of pre-existing muscle synergies and adjustment of the time of their recruitment may therefore provide an effective way to counteract locomotor perturbations, where fast reactive responses are necessary to maintain postural stability (Bohm et al., 2015; Karamanidis et al., 2020; Zhang et al., 2020).
The recruitment of muscle synergies in the perturbed gait cycles was associated with specific functional phases of walking and their modulation corresponded to the perturbation-specific stability requirements. In our experimental design, the perturbations were introduced in the middle of the stance phase with a quick shift of the treadmill belt in an anteroposterior or mediolateral direction. The activation patterns of the early swing synergy showed a steep increase with a higher contribution of the tibialis anterior and rectus femoris in the perturbed cycles. Further, the shift of the CoA to an earlier time point following the perturbation in this synergy suggests a faster recruitment of the early swing synergy as a reactive control mechanism in order to increase the efficacy of execution of the following step (König et al., 2022). The first step after a gait perturbation is crucial for balance recovery and fall prevention, especially in older individuals where reactive responses have deteriorated (Arampatzis et al., 2008; Bierbaum et al., 2011; Mersmann et al., 2013). A fast and long step after the perturbation increases the base of support and results in an improvement of stability (Bierbaum et al., 2010, 2011; Maki and McIlroy, 2006). Similar to the early swing synergy, the activation pattern of the propulsion synergy showed a rapid increase in the second phase of stance in the perturbed cycles. Taking into consideration that in this synergy the plantar flexor muscles are the main contributors and achieved high values in their muscle weights compared with other muscles, the steep increase of the activation pattern in the propulsion synergy indicates a rapid force development of the plantar flexor muscles after the initiation of the perturbation, which is functionally relevant to prevent a fall (Pijnappels et al., 2005; Robinovitch et al., 2002). The rapid increase in the activation of the propulsion synergy is functionally associated with the earlier plantar flexion after anteroposterior and mediolateral gait perturbations compared to unperturbed walking (Taborri et al., 2022). Further, the widening of the propulsion synergy corresponds to a longer activation and thus longer force generation of the plantar flexor muscles relative to the stance phase during propulsion, indicating a mechanism for an effective push-off and initiation of the swing phase. It is worth mentioning that muscle force generation is also dependent on muscle contractile conditions such as muscle force potentials because of the force–length and force–velocity relationships. Nevertheless, a widening of the activation of the triceps surae muscles in spite of varying force–length–velocity potentials will introduce a longer force generation and thus a more effective push-off (Bohm et al., 2023).
The widening of the basic activation patterns of the muscle synergies during the perturbed and the first recovery steps led to an increased overlap of activation between chronologically adjacent synergies and can be interpreted as a compensatory mechanism adopted by the central nervous system to deal with the perturbation and promote robust postural control (Janshen et al., 2020; Santuz et al., 2018a). Perturbations clearly affected gait stability, as shown by the modified temporal gait parameters (i.e. cadence, stance and swing times). Cadence was increased and stance and swing times were decreased in the perturbed and in the following recovery steps, demonstrating biomechanical adjustments to regain stability after the perturbation. Maintenance of stability is a primary task-relevant goal during locomotion (Cajigas et al., 2017) and, therefore, corrective locomotor actions are important after perturbations (Bohm et al., 2015; Iturralde and Torres-Oviedo, 2019). Widening of the basic activation patterns of the muscle synergies following perturbations as a strategy to provide robust motor control seems to be a common mechanism in biological systems. Widening of activation patterns during challenging locomotor tasks has been found in healthy individuals (Martino et al., 2015; Santuz et al., 2018a), patients with multiple sclerosis (Janshen et al., 2020, 2021) and mice (Santuz et al., 2022).
The observed modulations in the spatiotemporal structure of muscle synergies during the perturbed cycle were less pronounced in the predictable condition. Prior experience with specific perturbations can promote proactive adjustments of motor control (Bohm et al., 2015; Patla, 2003) and reduce the consequences of a perturbation (Bierbaum et al., 2010; Kanekar and Aruin, 2015; Marigold and Patla, 2002). The sensorimotor system can prepare a specific motor action in a predictive manner before its execution by building up an expectation of sensory inputs and then comparing the expected and actual inputs in order to successfully drive the desired movement (Wolpert, 1997). Prior experience with a perturbation can also improve the corrective responses after the perturbation (Bierbaum et al., 2011; Van Der Linden et al., 2007), possibly through a supraspinal mediation of sensory inputs. We found an increased EMG activity in several lower leg muscles before the initiation of the perturbation (i.e. middle of the stance phase) in the predictable compared with the unpredictable condition, which indicates proactive adjustments in the activation patterns. During the predictable condition, all knee extensors showed increased activity within the last ∼100 ms prior to the perturbation in both anteroposterior and mediolateral directions. This might have contributed to an improved limb support of the stance leg increasing hip height (Pai et al., 2010), as an important strategy of the neuromotor system to avoid falling (Pavol and Pai, 2007). Increased limb support could further benefit a fast initiation of the recovery step to improve postural stability (Bierbaum et al., 2010; Maki and McIlroy, 2006).
We investigated the EMG activity of 13 muscles of the right perturbed leg. Although the responses of the perturbed leg during walking are very important for stability performance (Pijnappels et al., 2005; Robinovitch et al., 2002; van Dieën et al., 2007), the first reactive responses of the contralateral leg after the perturbation contribute significantly to the human's stability state (König et al., 2022; Van Der Linden et al., 2007; Werth et al., 2022). Possible effects of the contralateral leg on the EMG activity of the perturbed leg muscles, mediated by spinal and supraspinal interlimb reflexes (Mrachacz-Kersting et al., 2004; Stevenson et al., 2015), may influence the spatiotemporal structure of the muscle synergies. However, during perturbations in humans, the unperturbed leg showed a quite similar modular organization to that for normal walking (König et al., 2022; Oliveira et al., 2012) and the agreement of the interlimb coordination patterns was low, indicating mainly an independent leg control (Sawers et al., 2017). Nevertheless, and besides the debate about the degree of (in)dependency in the control of each leg, the non-consideration of the contralateral leg in the current study, particularly for the perturbed cycles, is a limitation.
In the current study, we did not use a specific measure in order to assess the stability state of the participants after the perturbations. In our protocols, we found clear perturbation effects in several temporal gait variables (i.e. cadence, swing time, stance time, duty factor), muscle EMG activity patterns and spatiotemporal components of the muscle synergies which were observed until the fourth recovery step. After the fourth recovery step, all these investigated variables returned to the baseline patterns. The observed temporal and neuromuscular patterns during the perturbed and the following recovery steps demonstrate an acute disturbance in the participant's movement patterns during a periodic task (i.e. walking) and a return to their original state after the perturbations. Considering that, in general, dynamic stability during gait is defined as the ability of the system to return to its original periodic patterns following a disturbance (Birn-Jeffery and Daley, 2012), we can argue that the applied perturbations modified the stability state of the participants. An increase in cadence after challenging mediolateral perturbations has been also reported previously (Hak et al., 2012; Madehkhaksar et al., 2018) and was interpreted as a mechanism to increase the safety features of the system (Reimann et al., 2018). Furthermore, the coefficient of variation in all investigated temporal parameters increased in the perturbation and recovery cycles compared with unperturbed walking, indicating an increased variability during and after the perturbations. Several studies reported an increased variability of temporal and spatial parameters during walking in unstable conditions and in humans with increased instability (Guimaraes and Isaacs, 1980; Hausdorff et al., 1997; Maki, 1997; Thies et al., 2005). Finally, Madehkhaksar et al. (2018) using an experimental design to introduce perturbations similar to the current study, found significant perturbation effects in the stability state of the participants (i.e. a decrease in the margin of stability) and increased variability in temporal and kinematic variables such as step length, step width and cadence in response to perturbations. Therefore, and although we did not directly measure the stability state in our experimental design, we are confident that in our study, the applied perturbations affected the stability state of the participants. In the current study, we did not include a matched control group of young health participants. However, the main purpose was to investigate how healthy older adults organize locomotor muscle synergies to counteract unpredictable and predictable gait perturbations, and therefore we included only older healthy participants. Chvatal and Ting (2012) investigating comparable anteroposterior and mediolateral gait perturbations (i.e. platform translation of 12.4 cm with 40 cm s−1 velocity) also found a common set of muscle synergies in young healthy participants.
In conclusion, our findings demonstrated that older adults used the same basic structure and composition of the four walking muscle synergies during unperturbed and perturbed walking, yet modulated their spatiotemporal components in order to counteract gait perturbations. The above behavior was consistent during the perturbed and the recovery steps and in all applied walking perturbations (i.e. anteroposterior, mediolateral, unpredictable, predictable). The selection of pre-existing muscle synergies and adjustment of the timing of their recruitment indicates a flexible postural control in human gait during challenging locomotor conditions, which may facilitate the effectiveness in coping with perturbations and promote the transfer of adaptation between different types of perturbations.
Acknowledgements
We like to thank all participants for their commitment to this study. We further thank Carlotta Körbi, Jaqueline Martini and Thomas Gerhardy for their assistance in data collection and Arno Schroll for statistical consultation. We acknowledge support by the Open Access Publication Fund of Humboldt-Universität zu Berlin.
Footnotes
Author contributions
Conceptualization: L.B., A.A.; Methodology: L.B., A.S., A.A.; Software: L.B., A.S.; Validation: L.B., A.S.; Formal analysis: L.B., A.S., A.A.; Investigation: L.B.; Resources: M.S., A.A.; Data curation: L.B., A.S.; Writing - original draft: L.B., A.A.; Writing - review & editing: L.B., A.S., F.M., S.B., M.S., A.A.; Visualization: L.B.; Supervision: A.A.; Project administration: L.B., M.S., A.A.; Funding acquisition: M.S., A.A.
Funding
Open access funding provided by Humboldt-Universität zu Berlin. Deposited in PMC for immediate release.
Data availability
The dataset analyzed in this article is available from Zenodo: https://zenodo.org/doi/10.5281/zenodo.10794581.
References
Competing interests
The authors declare no competing or financial interests.