Age-Related Di ﬀ erences in Muscle Synergy Organization during Step Ascent at Di ﬀ erent Heights and Directions

: The aim of this study was to explore the underlying age-related di ﬀ erences in dynamic motor control during di ﬀ erent step ascent conditions using muscle synergy analysis. Eleven older women (67.0 y ± 2.5) and ten young women (22.5 y ± 1.6) performed stepping in forward and lateral directions at step heights of 10, 20 and 30 cm. Surface electromyography was obtained from 10 lower limb and torso muscles. Non-negative matrix factorization was used to identify sets of ( n ) synergies across age groups and stepping conditions. In addition, variance accounted for (VAF) by the detected number of synergies was compared to assess complexity of motor control. Finally, correlation coe ﬃ cients of muscle weightings and between-subject variability of the temporal activation patterns were calculated and compared between age groups and stepping conditions. Four synergies accounted for > 85% VAF across age groups and stepping conditions. Age and step height showed a signiﬁcant negative correlation with VAF during forward stepping but not lateral stepping, with lower VAF indicating higher synergy complexity. Muscle weightings showed higher similarity across step heights in older compared to young women. Neuromuscular control of young and community-dwelling older women could not be di ﬀ erentiated based on the number of synergies extracted. Additional analyses of synergy structure and complexity revealed subtle age- and step-height-related di ﬀ erences, indicating that older women rely on more complex neuromuscular control strategies.


Introduction
Aging is associated with gradual changes in neuromuscular control [1,2]. Eventually, these changes can have a major impact on fall risk, mobility and independence in older adults [3][4][5], which may be exacerbated in post-menopausal women due to accelerated muscle wasting [5]. One major challenge with assessing changes in neuromuscular control in healthy community-dwelling older adults is that, due to the gradual nature of age-related neuromuscular deterioration, pre-clinical changes in neuromuscular control of everyday tasks may go undetected [6]. In healthy older adults, changes in neuromuscular control can be revealed by increasing task challenge. For example, by increasing gait speed, cadence, step length, and step height [7][8][9].
Step ascent, a functional task in daily life, provides a challenge that can easily be modified by increasing step height and has been demonstrated to be an effective training stimulus to improve muscle volume and functional ability in older women [10,11]. However, the effects of aging and task challenge on the neuromuscular control strategies behind step ascent have not yet been thoroughly explored. In a previous study conducted with older women, we found that peak activation of several major lower limb muscles occurred during the ascent phase of stepping and that there is a positive dose-response relationship between step height and peak muscle activation [10]. However, the increase in step height was also accompanied by increased between-subject variance of peak activation magnitudes [10]. This could be attributable to a tendency of older adults to modulate their motor strategies (e.g., shifting joint moment generation from the knee to the ankle), in order to operate within the limits of their physical capacity when ascending steps [12,13]. Additionally, it is unknown if other observed age-related changes in neuromuscular control strategies, such as increased antagonistic co-contraction of quadriceps and hamstrings to help maintain postural control during dynamic tasks [14][15][16][17], may interact with increased task challenge.
One way to assess neuromuscular control is through muscle synergy analysis. Previous studies have found that low-dimensional sets of motor modules, also known as muscle synergies, can be used to reconstruct muscle activation patterns during various motor tasks [18][19][20][21][22]. These synergies are composed of groups of muscles that are assumed to be activated by a single neural command [23]. It is thought that the central nervous system employs this modular organization to reduce the large number of degrees of freedom inherent to the redundancy of the human musculoskeletal system [24], and to allow for flexible but accurate response selection during motor tasks [25]. However, some researchers have argued that modular recruitment of muscles might represent predetermined control strategies and could merely be an effect of task constraints or optimized performance criteria, rather than reflecting neural control strategies employed by the central nervous system [23,26]. Regardless of the mechanisms underlying modular organization of muscle activation, extracting muscle synergies from electromyographic (EMG) signals can provide important insights about neuromuscular control strategies used to perform functional tasks [27]. In older adults with a history of falls, declines in neuromuscular control are reflected in a decreased number of extracted synergies from walking tasks that challenge dynamic balance [6]. A decreased number of synergies is indicative of decreased complexity of motor control or a decreased motor repertoire. These underlying changes in synergy complexity can be quantified using variance account for (VAF) by a given set of extracted synergies and defining the number of synergies required to adequately reconstruct the original EMG signals (indicated by a-priori or a-posteriori set thresholds for VAF) [28]. For example, high VAF by a limited number of synergies represents decreased complexity of motor control, which is often associated with neuromuscular pathologies such as cerebral palsy and stroke [6,27,29] and characterized by increased levels of co-activation between individual muscles [30]. However, healthy aging is associated with a more gradual process of physical decline and thus changes in complexity of motor control may not manifest in a decreased number of synergies [8]. Consequently, age-related changes in motor control may be better reflected by comparisons of VAF for a fixed number of synergies [29,31,32] or by assessing changes in spatio-temporal organization of muscle synergies, such as altered module composition and shifts in activation patterns [2,6]. Differences in spatio-temporal organization within a stable number of synergies can arguably be considered more subtle than differences in the total number of synergies as they tend to reflect compensatory or alternative motor strategies in order to overcome increased or altered task challenges, whereas a decreased number of synergies is usually used as an indication of neural impairments. Although it is currently unknown how aging and task intensity affect muscle synergy organization during stair ascent, analyses of underlying differences could provide a basis to improve detection of pre-clinical age-related deterioration in neuromuscular control and more effectively target fall prevention programs at individuals most at risk. Therefore, the purpose of this study was to explore muscle synergy recruitment during step ascent in forward and lateral directions and with incremental step heights in young and older adults. Specifically, we aimed to assess the effects of age-related changes, task intensity, and their interaction on complexity and organization of motor control by comparing the number of extracted synergies, variance accounted for (VAF) by a fixed number of synergies, and spatio-temporal characteristics of the extracted synergies across conditions.

Participants
Eleven older women (67.0 y ± 2.5, 161.3 cm ± 4.9, 64.4 kg ± 6.8) and ten young women (22.5 y ± 1.6, 168.9 cm ± 1.7, 64.2 kg ± 7.9) were recruited for this study. Potential participants were excluded if they suffered from neurological or motor disorders, impaired balance control, or if they had been involved in a structured training program in the last 6 months prior to participation in the study. This study was approved by the Human Ethics Committee of KU Leuven in accordance with the Declaration of Helsinki and registered with the Clinical Trial Center UZ Leuven (S56405). All participants signed informed consent prior to participation in the study.

Experimental Protocol
Participants performed a series of stepping tasks consisting of stepping onto a wooden block in forward direction (Fstep) and in lateral direction (Lstep). Task intensity was determined by the height of the block (10, 20 and 30 cm) [10]. A previous study by Singh and Latash revealed that muscle fatigue can cause higher variability of muscle activation patterns and composition of synergy components [33]. In contrast with more common tasks used for muscle synergy analyses (e.g., gait and perturbation trials), step ascent with incremental heights presents a relatively high physical challenge for older adults [10], thus increasing the risk of fatigue with a high number of repetitions. Pilot testing showed that, despite regular breaks, execution of more than three repetitions per trial, combined with the number of trials, was already quite fatiguing for the older women in this study. Therefore, each trial consisted of three repetitions to avoid confounding effects of fatigue on synergy composition. Every repetition was performed with the dominant leg first, followed by the trailing leg, and ending in double support on top of the block. Left-right dominance was determined during familiarization by noting with which foot participants preferred to take the first step. As a control question, participants were asked with which foot they would prefer to kick a ball [34]. Only two participants (both young women) showed left-side dominance.
Step ascent was assessed in both forward and lateral directions, as these are functional tasks that require simultaneous coordination of the hip, knee and ankle musculature and can be challenging for older adults [12,13].
Stepping up in forward direction shows close functional resemblance to stair-climbing [34], which is an activity of daily life associated with high fall risk in older adults [35]. Stepping up in lateral direction is a less common task for older adults and can provide an additional challenge to the hip abductors, which play an important role in medio-lateral balance control [36]. The speed of task execution was controlled by a metronome at 1 second for ascent, 1 second stance, and 1 second descent to avoid differences in muscle activation due to explosive movements.

Kinematics
Kinematics were recorded at a sampling rate of 100 samples/s by means of 3D motion capture cameras (Vicon ® , Oxford Metrics, Oxford, UK). Infrared reflective markers (diameter 14 mm) were placed on both heels and the sacrum. Only data from the ascent phase were used for analysis. Based on the kinematic data, the start of the ascent phase was defined at 200 ms prior to initial vertical displacement of the leading heel marker beyond 2× the standard deviation obtained during normal stance. The end of the ascent phase was defined at 500 ms after maximum knee extension, defined as the maximum relative distance between the heel and sacrum. Two sub-phases (foot clearance and pull-up) of step ascent were defined using the dominant heel marker trajectories. The shift from foot clearance to pull-up phase was defined at 100 ms after vertical displacement of the leading heel marker dropped below 2× the standard deviation obtained during normal stance. Kinematic data from the trailing heel were included to detect and eliminate trials with undesirable events, such as the toe getting caught on the edge of the step. No such events were detected.

Electromyography
Muscular activation was collected unilaterally from ten lower limb and trunk muscles on the dominant side using surface electromyography (EMG) (Aurion ® , ZeroWire, Milan, Italy) sampled at 1000 samples/s. Activation was recorded from the following 10 muscles: the tibialis anterior (TA), the lateral head of the gastrocnemius (GL), soleus (SOL), vastus lateralis (VL), rectus femoris (RF), biceps femoris (BF), semitendinosus (ST), gluteus maximus (GMAX), gluteus medius (GMED), and the erector spinae (ERS), in accordance with SENIAM guidelines [37]. The skin was shaved and thoroughly rubbed with an alcohol swab to ensure optimal conductivity. Bi-polar Ag/Ag-Cl electrodes (Ambu ® BlueSensor P, Ballerup, Denmark) were then placed on the belly of the muscles with an inter-electrode distance of 25 mm. Sampling of kinematic and EMG data was synchronized.

Synergy Extraction and Data Analyses
All EMG and kinematic data were processed using custom MATLAB scripts (MATLAB R2014b, MathWorks ® , Natick, MA, USA). The EMG signals were high-pass filtered with a 1 st order Butterworth filter with a cut-off at 20 Hz, full-wave rectified and smoothed with a 0.1-s moving average window [10,38]. EMG signals from forward and lateral stepping were normalized to the respective maximum activation obtained over all trials performed in the congruent direction so that activation could not exceed 100% [39,40]. The EMG signals were time-synchronized with the kinematically defined start and end points and subsequently normalized over time to define 0%-100% of the step cycle. Finally, because EMG data were only collected during step ascent and therefore represented intervals, rather than continuous activation patterns (as would be the case during gait trials), signals were averaged over the three repetitions performed in each condition. The choice to average signals rather than concatenating them was made in order to obtain the best reconstruction quality for our relatively short intervals, at the risk of losing information on step-to-step variability [40].
Muscle synergies were extracted from the individual average EMG data matrix using non-negative matrix factorization (NNMF). NNMF calculates muscle synergies (W) and their relative temporal activation patterns (C), resulting in muscle activations being represented as W × C + e. W represents the relative muscle co-activation, defined as the relative weight of each muscle per synergy, and is constructed as an m × n matrix where m is the total number of muscles and n is the selected number of synergies. C represents the temporal activation patterns and is constructed as an n × t matrix where t represents the number of data points over normalized time (100 per individual trial) and e is the residual error matrix [22,41]. The algorithm was repeated 1000 times for each subject to avoid local minima. The appropriate number of synergies was defined using two criteria. First, using an iterative process where the number of synergies varied between 1 and 10, the minimum number of synergies was selected based on the number required to reach ≥ 85% of group-averaged variance accounted for (VAF). As an additional local criterion, synergies had to account for ≥75% VAF for each individual muscle [42]. This double criterion approach was selected in order to adequately reproduce relevant features of the synergy compositions. VAF was defined as the uncentered Pearson correlation coefficient between W × C and the EMG amplitude time series. To compare spatio-temporal characteristics, individual synergies obtained from different subjects were pooled and matched based on the correlation of their structure (muscle weightings in each synergy of W) using a custom cluster analysis algorithm [43]. If a synergy showed equal correlation to more clusters, that synergy remained in the pool it was initially assigned to. Each synergy of W and C was subsequently averaged over all participants in that age group. For comparisons between age groups, the group-averaged synergies were also matched based on their structure using cluster analysis. Finally, we computed time-averaged standard deviations of the synergy activation patterns, with a fixed number of synergies, to assess if age-related differences in group-averaged VAF could be affected by between-subject variability of temporal activation patterns.

Statistical Analyses
Statistical analyses were performed with SPSS (IBM ® SPSS v23 Statistics for Windows, Armonk, NY, USA). For both step directions, two-way repeated measures ANOVA (age × synergy number) for VAF was used to assess the interaction effect of age and number (n) synergies on VAF [39]. The number of synergies needed to adequately reconstruct the EMG signals was then determined using the iterative process described in the previous paragraph. Subsequently, two-way repeated measures ANOVA (age × step height) was used to assess the main and interaction effects of age and step height on synergy complexity, which was defined as VAF obtained with n synergies [29] fixed to four. Data were tested for normality with a Kolmogorov-Smirnov test. Sphericity was checked using Mauchly's test for sphericity. If a significant main effect was found, post-hoc tests comparing differences between age groups were performed using independent samples t-tests, while related-samples t-tests were used to compare differences per step height and synergy number. Alpha was set to 0.05 for all statistical tests.
Similarity of muscle synergies (based on muscle weightings, W) was quantified based on Pearson's correlation coefficients where r > 0.7 represented significant similarity and r > 0.45 represented marginal similarity [20,44]. Correlated synergies within age groups between step heights, and between age groups for each step height, were considered to be shared synergies, while non-correlated synergies were considered task-specific or age-related synergies [44]. Differences in muscle contributions to each synergy (W) between age groups were checked using Mann-Whitney U tests.

Results
Time-normalized kinematic data from the heel and pelvic markers showed high similarity (r > 0.9) in averaged vertical displacement over time between young and older women for all step heights. An example of the averaged vertical displacement patterns and standard deviations at 30 cm step height is provided in Figure 1.
Two-way ANOVA (age × synergy number) of VAF revealed significant main effects of synergy number for all step directions and heights (p < 0.001), but no interaction effects with age (p ≥ 0.05). A significant main effect of age (p = 0.028) was detected only for lateral stepping at 30 cm. For the group-averaged VAF, four muscle synergies were required to achieve a threshold level of 85% VAF for reconstructed signals across both age groups, step directions and step heights ( Figure 2). This indicates that age, step direction and step height did not affect the number of synergies needed to reconstruct the EMG data. Consequently, the following results were obtained assuming n = 4 synergies. For forward stepping, four synergies accounted for 90.5%, 89.8% and 91.8% of variance in young women and 88.5%, 87.3% and 87.4% in older women for step heights of 10, 20 and 30 cm respectively. In lateral stepping, VAF by four synergies was 90.3%, 90.0% and 91.7% in young women and 88.2%, 88.0% and 87.4% in older women for step heights of 10, 20 and 30 cm respectively. Two-way ANOVA (age × step height) on VAF obtained from n = 4 synergies revealed a significant main effect of step height (p = 0.002) and age (p = 0.026) in forward direction, but not in lateral direction (p = 0.187 and p = 0.138 respectively). No significant age × step height interaction effect was found for either step direction (p > 0.05). For forward stepping, related-samples t-tests revealed a significant difference between step heights of 10 cm versus 20 and 30 cm (p = 0.009 and 0.014 respectively), but not between 20 and 30 cm (p > 0.05). Independent samples t-tests revealed a significant difference between age groups for each step height (p = 0.005, p = 0.041 and p = 0.019 for 10, 20 and 30 cm respectively).  Comparisons between muscle weightings (Table 1, Figures 3 and 4) showed that synergy 2 and 4 had high inter-step height similarity for both age groups and step directions. Synergy 4 also appeared to be highly similar between age groups. In synergy 2, a lower similarity between age groups was found, probably due to a difference in quadriceps/hamstring co-activation, as characterized by a significantly decreased contribution of the quadriceps and increased contribution of the hamstrings for most stepping conditions in older women. The composition of synergy 3 appeared to be the most variable. In contrast with forward stepping, which showed a robust synergy organization within and between age groups, lateral stepping resulted in lower correlations between step heights for the young group when compared to the older group.  Figure 3. Muscle weightings and temporal activation patterns for four synergies (S1-S4) extracted during forward stepping (Fstep). Each set of two columns represents muscle synergies extracted from step heights of 10, 20 or 30 cm per age group with young women represented in the left column and older women in the right column. * indicates a significant difference between contributions of individual muscles between age groups. Time-averaged foot clearance (white bars) and pull-up (black bars) phases of ascent are depicted at the bottom. TA = tibialis anterior, GL = gastrocnemius lateralis, SOL = soleus, VL = vastus lateralis, RF = rectus femoris, BF = biceps femoris, ST = semitendinosus, GMAX = gluteus maximus, GMED = gluteus medius, ERS = erector Spinae. Analyses of the temporal activation patterns (Figures 3 and 4) and the time-averaged standard deviation of these temporal activation patterns ( Figure 5) showed that the between-subject variability of activation timing for most step heights and directions was significantly higher in the older cohort.

Discussion
The purpose of this study was to investigate the effects of age and task challenge on neuromuscular control and resultant complexity of neuromuscular activation patterns of women during step ascent by examining muscle synergy organization during stepping tasks with incremental step heights in both forward and lateral directions.
Our results show that complexity of motor control is quite robust across step heights and age groups for stepping in forward and lateral directions. We found no differences in the number of synergies between age groups and step heights for either age group. Further analyses of VAF by four synergies revealed main effects, but no interaction effect, of age and step height on complexity of motor control during forward stepping, indicating subtle differences that could not be detected by analyses of the number of extracted synergies. These analyses revealed that older women actually exhibited more complex motor control strategies, indicated by lower VAF [29], for each step height. Additionally, forward step heights of 20 and 30 cm yielded a significantly lower VAF at n = 4 synergies compared to 10 cm. Comparisons of muscle synergy organization revealed that muscle contributions to individual synergies (e.g., muscle weightings) during forward stepping were quite robust between step heights for both age groups. Finally, age-related differences in synergy organization were characterized by a notably lower similarity index between step heights for lateral stepping in young women and increased between-subject variability of the temporal activation patterns in older women.
The extraction of four synergies from step ascent is in agreement with previous studies of locomotion in healthy adults that included a maximum of ten lower limb muscles [24,40]. An important aspect of this study is the inclusion of EMG signals obtained from 10 lower limb muscles, including the gluteus medius, gluteus maximus and the erector spinae. Our results were in line with a previous study by Oliveira et al., who also found four synergies were sufficient to reconstruct the EMG signals of ten lower limb muscles. However, they also proposed that the addition of EMG measurements from hip extensors and abductors during gait would likely increase dimensionality [40]. Our data show that this is not necessarily true for step ascent. For example, during forward stepping, gluteus medius activation coincided with activation of the triple extensors (gluteus maximus, plantar flexors and rectus femoris). During lateral stepping, inclusion of the gluteus medius and erector spinae did not increase dimensionality in the form of an additional synergy as their activation coincided mainly with tibialis anterior activation during lift-off of the trailing foot and trunk stabilization prior to the double support phase after ascent. In line with previous studies including healthy older adults, our results indicate that age did not affect the number of synergies required to reconstruct the muscle activation, which implies that the complexity of motor control (or motor repertoire) of our healthy older cohort was not reduced [6,8,17,45]. However, we were surprised to find that the VAF at a fixed number of synergies, which can be used as an alternative way to assess complexity of motor control [29], was decreased in older women and with step height, indicating increased rather than decreased complexity. We propose that this may be due to the increased challenge posed by step ascent for older adults, forcing some to adopt different motor strategies to compensate for decreased physical capacity [15,46]. As a consequence, the organization or timing of motor modules may be altered, inevitably leading to higher between-subject variability in older adults compared young adults. The assumption that shifts in muscle synergy organization are attributable to (relative) increases in task challenge is in line with previous findings by Routson et al., showing that changes in speed, cadence, step length, and step height during gait can lead to altered temporal activation patterns [7]. Additionally, other studies have explored the interaction between age and task challenge during gait and found that walking at a higher than preferred cadence revealed small differences in spatio-temporal characteristics of neuromuscular control in older but not in young adults [8,9], although this proposed interaction effect was not confirmed by our results.
More detailed analyses of synergy organization in our older cohort revealed a trend towards increased contribution of the hamstrings and decreased contribution of the quadriceps in synergy 2, indicating an increase in quadriceps/hamstring co-activation during the initial foot clearance phase for both step directions compared to young women. These results are in line with findings from previous studies that have shown elevated muscle co-activation in older adults to increase joint stiffness and enhance stability during activities of daily life, such as stair climbing and single step descent [17,[47][48][49], and directly affect muscle synergy organization [8,17]. Additionally, comparisons of synergy organization across step heights revealed that, in young women, increasing the step height from 20 to 30 cm was associated with an increased contribution of the gluteus medius and maximus to synergy 1 and of the calf muscles to synergy 3 during foot placement and the initial pull-up phase of ascent in lateral direction, while the composition of the remaining synergies remained similar. These changes indicate that some synergies reflect basic motor patterns which are activated during a variety of tasks, whereas other synergies can be flexibly recruited to match task-specific demands [20,44], such as the increased challenge to medio-lateral stability imposed by increased step heights. This is reflected by kinematic analyses of motor strategies for stair negotiation, showing both common and variable patterns, with the highest variability often seen at the hip joint [50]. The age-related differences in neuromotor strategies found in this study were reflected by subtle differences in kinematic profiles of the heel and sacrum. For example, visual inspection ( Figure 1) revealed that, for older women, average vertical displacement of the sacrum was more linear in both directions with distinctly higher between-subject variability during the pull-up phase of forward stepping and higher between-subject variability of heel peak height during the foot clearance phase. However, these kinematic profiles were primarily included to define start and end points of each step cycle. As such, they provide limited information about how muscle synergy organization affects strategies such as total lower limb extension patterns and joint kinematics and kinetics and should be a focus of future studies [50].
Additional analyses were included to detect possible age-related differences of temporal activation patterns. Our results did reveal higher between-subject variability of temporal activation patterns, indicating increased heterogeneity of motor control strategies within the older cohort. This may reflect a relative increase in functional demand imposed by step heights of 20-30 cm in this age group, necessitating individual modulations of synergy timing to compensate for decreased physical capacity [7,12]. Higher between-subject variability of temporal activation patterns associated with increased task challenge in young women indicates that synergy organization is likely also associated with differences in motor skill levels [27]. This is illustrated by notable changes in between-subject variability of activation patterns in synergies 1 and 3 between step heights of 20 and 30 cm in lateral direction. Finally, although not the primary focus of this study, the differences in synergy organization between forward and lateral stepping are in line with findings from previous studies, showing that EMG recruitment patterns are task-specific for forward and lateral stepping [34,51].
Some limitations of this study have to be recognized. We chose to include only women in this study. As such, additional research is required to assess if these findings also apply to older males. EMG was only collected from the dominant leg. For this reason, differences in motor strategies involving additional push-off force of the trailing leg could not be analyzed [25]. Additionally, due to the technical limitations of surface EMG recording, no data were collected from the deeper thigh muscles such as the hip adductors. Future studies involving step ascent should consider including the hip adductors in order to increase dimensionality and provide information regarding the effects of antagonistic co-contraction of the hip ab-and adductors. Finally, a possible limitation lies with averaging EMG signals of individual participants over three repetitions of the same trial, rather than concatenating them prior to running the factorization algorithm. This may have led to a decrease of detail in the data [40].

Conclusions
Neuromuscular control of young and community-dwelling older women in stepping up in forward and lateral direction could not be differentiated based on the number of synergies. However, additional analyses of synergy complexity, such as VAF by the given number of synergies, and synergy structure revealed several age-and step-height related differences. These findings show that the ability to modulate synergy composition is well preserved in healthy older women and that they respond to more challenging physical tasks by adapting basic muscle recruitment patterns. This results in more complex motor control patterns despite evidence of increased antagonistic co-activation, likely indicating increased involvement of mechanisms for balance control.