Age-related changes in muscle coordination patterns of stepping responses to recover from loss of balance

Introduction: Reactive stepping capacity to recover from a loss of balance declines with aging, which increases the risk of falling. To gain insight into the underlying mechanisms, we investigated whether muscle coordination patterns of reactive stepping differed between healthy young and older individuals. Methods: We performed a cross-sectional study between 15 healthy young and 14 healthy older adults. They recovered from 200 multidirectional platform translations that evoked reactive stepping responses. We determined spatiotemporal step variables and used muscle synergy analysis to characterize stance-and swing-leg muscle coordination patterns from the start of perturbation until foot landing. Results: We observed delayed step onsets in older individuals, without further spatiotemporal differences. Muscle synergy structure was not different between young and older individuals, but age-related differences were observed in the time-varying synergy activation patterns. In anterior-posterior directions, the older individuals demonstrated significantly enhanced early swing-leg synergy activation consistent with non-stepping behavior. In addition, around step onset they demonstrated increased levels of synergy coactivation (mainly around the ankle) in lateral and anterior directions, which did not appear to hamper foot clearance. Conclusion: Although synergy structure was not affected by age, the delayed step onsets and the enhanced early synergy recruitment point at a relative bias towards non-stepping behavior in older adults. They may need more time for accumulating information on the direction of perturbation and making the corresponding sensorimotor transformations before initiating the step. Future work may investigate whether perturbation-based training improves these age-related deficits.


Introduction
Humans have a remarkable ability to robustly move in and interact with diverse and challenging environments without losing balance.However, falls become more frequent as we age, with 27.5 % of older adults reporting at least one fall a year (Moreland et al., 2020).These falls can cause injuries and, consequently, limit mobility (Komisar et al., 2022).While the etiology of falls is typically multifactorial, poor balance is known to be a key modifiable risk factor.
When confronted with a loss of balance, fast reactive stepping responses are critical to prevent ourselves from falling.Recent studies have shown that reactive balance capacity is impaired in older adults.Following balance perturbations, older adults show poorer reactive stepping performance compared to their younger counterparts, such as shorter step lengths and lower stepping thresholds (Crenshaw and Grabiner, 2014;Hsiao-Wecksler and Robinovitch, 2007).To elucidate the mechanisms underlying these age-related declines in reactive balance capacity, previous studies have focused on postural muscle recruitment and found that older adults react slower and show greater levels of co-activation of agonist and antagonist muscles, while conflicting results have been reported regarding response amplitudes (Lee et al., 2017;Phu et al., 2022).Yet, these previous analyses were largely restricted to the early stages of the response (the so-called automatic postural response, APR), while the proper execution of a reactive step requires muscle recruitment well beyond the APR time window.In addition, outcomes are commonly reported at the level of individual muscles, while stepping involves multi-segmental co-activation of synergistic groups of leg and trunk muscles in accordance with the distinct function of either leg (i.e., stance or swing leg).Hence, we lack a good understanding of whether and how aging affects the coordinated muscle recruitment that drives reactive steps.
Muscle synergy analysis is an established method for evaluating coordination patterns of muscle recruitment (Chvatal and Ting, 2013;Torres-Oviedo and Ting, 2007).Muscle synergies are represented as fixed spatial structures of groups of coactive muscles, accompanying time-varying activation coefficients.In a group of healthy young subjects, Chvatal et al. demonstrated robust muscle synergy expression of the stance leg during feet-in-place and reactive stepping responses following multidirectional balance perturbations (Chvatal et al., 2011).Other studies have used this method to elucidate differences in coordinated muscle recruitment between successful and unsuccessful recovery from laboratory-induced slips in older adults (Sawers and Bhatt, 2018) and between people after stroke and healthy controls responding to support-surface translations (de Kam et al., 2018).Interestingly, a recent study suggested that subtle differences may exist in the structure of muscle synergy recruitment between healthy young and older people in backward reactive stepping responses, yet methodological limitations (e.g.limited set of muscles, single stepping direction, n = 1 repetition per subject) precluded drawing firm conclusions (Wang et al., 2021).
Here, we applied muscle synergy analysis to investigate whether muscle coordination patterns of reactive stepping differed between healthy young and older individuals.We used an extensive experimental protocol with multidirectional support-surface translations and conducted a comprehensive analysis of muscle synergy recruitment during the entire stepping response (i.e., from the onset of perturbation until step landing) in both the stance and the swing leg.We hypothesized that compared to the young, older adults would demonstrate modestly greater levels of synergy co-activation (i.e., 'stiffening up') in response to a balance perturbation.

Participants
Fifteen young (age: 23.7 ± 2 yr, male/female: 9/6) and 14 healthy older adults (age: 63.5 ± 8 yr, male/female: 12/2) who were naïve to platform perturbations participated in this study.Participants were included if they had no history of neurological or musculoskeletal disorders or other conditions that may affect balance.All participants provided written informed consent.Experimental procedures were approved by the Research Ethics Committee of the Radboud University Medical Center (Nijmegen, The Netherlands, NL67690.091.18).

Experimental paradigm & data acquisition
Participants stood barefoot on a movable platform (Radboud Fall Simulator(RFS) (Nonnekes et al., 2013)), with their feet 5 cm apart and arms alongside the body.This narrow stance width forced the participants to use stepping strategies in lateral directions (de Kam et al., 2017).The RFS translated unexpectedly with at 1.5 m/s 2 (300 ms acceleration; followed by 500 ms constant platform velocity and 300 ms deceleration).Previous studies in our lab have shown that this acceleration allows older adults to recover balance with a single step, it was high enough to consistently evoke stepping responses in the young when instructed to respond naturally (de Kam et al., 2017;de Kam et al., 2016).Participants were instructed which leg they had to use for stepping, whenever they felt the need to take a step to recover balance.This instruction was used to acquire equal numbers of steps taken with each leg and to enforce taking side steps (and avoid cross steps) in any lateral perturbation direction.For each leg, perturbations were delivered in five directions: posterior (POST), 45 • posterolateral (PostLat), lateral (LAT), 45 • anterolateral (AntLat) and anterior (ANT) (Fig. 1).Prior to data collection, the participants were familiarized with the experimental procedures.Specifically, participants were exposed to perturbations of increasing intensity up to 1.5 m/s 2 across the five perturbation directions for each leg.They practiced instructed stepping with either leg and performed approximately 20 practice trials in total.The actual experiment involved eight series of 25 trials (four with either leg), with five repetitions per perturbation direction (in random order), thus resulting in 200 trials.Please note that perturbation direction is defined with respect to the direction of stepping (i.e., forward perturbation is a backward platform translation evoking a forward step).

Fig. 1.
A represents the perturbations directions (i.e., stepping direction) for each leg, whereas B shows the waveform of platform displacement (upper) and velocity (lower).

Data processing & analysis 2.3.1. Spatiotemporal step variables
Force plate data was low-pass filtered at 80 Hz (5th order Butterworth IIR filters, zero-phase shift) to remove platform motor noise.
Step onset was defined as the first moment when the vertical ground reaction force (GRF) fell below a threshold of 20 N. Similarly, foot strike was determined as the first moment when the selected force plate was reloaded with at least 50 N.
Step calculated as the displacement of the mid-foot coordinate (i.e., mid-point between calcaneus and 2nd metatarsal marker) in the direction of perturbation between step onset and foot strike of the instructed leg.
Step duration was the time between these events.In addition, step clearance was defined as the maximum vertical displacement of the 2nd metatarsal marker between step onset and foot strike.

EMG pre-processing
For each individual, EMG data of correctly performed trials (i.e., steps taken with instructed leg) was band-pass filtered (20-450 Hz), rectified and low-pass filtered at 40 Hz.To account for temporal differences in muscle behavior due to stepping events, EMG activity of each individual trial was time-normalized according to perturbation onset and stepping events.Specifically, perturbation onset was the starting point (0 %), step onset was defined as 50 % and foot strike as 100 %.For each leg, EMG data from all trials was concatenated into 2 matrices of 8 (number of muscles) by n (number of trials).Thus, in total we composed 4 matrices (right/left, swing/support leg) per participant.Subsequently, each muscle was scaled to unit variance (across trials) for equal weighting in the muscle synergy extraction (Zandvoort et al., 2019).

Muscle synergy analysis
Muscle synergies are defined as coordinated patterns of muscle activity to produce functional movement (Safavynia et al., 2011).To extract muscle synergies we performed nonnegative matrix factorization (NNMF), which computational decomposition technique has been used extensively for characterization of muscle recruitment (Torres-Oviedo and Ting, 2007; Lee and Seung, 1999).NMMF was conducted on each of the 4 matrices according to NNMF decomposes the matrix M into time-invariant structures W (muscles*s, the number of synergies) and time-varying coefficients C (synergies *n samples) + e, the residual error matrix.To determine the number of synergies required to account for the EMG data, we quantified the goodness of fit for each number of synergies by the variance accounted for (VAF) using.
We selected the number of synergies per leg as the lowest number that could account for an overall VAF >80 % (Chvatal and Ting, 2013;Torres-Oviedo et al., 2006).To allow for comparison among participants, the muscle synergy structures were normalized by the maximum of each structure.The time-varying activation coefficients were scaled by the same quantity (Barroso et al., 2014).
We determined Pearson correlations (r) between the time-invariant structures (muscle synergies) to compare similarity across individuals (r > 0.707, based on alpha level of 0.05 and 8 muscles included per synergy) (Chvatal and Ting, 2012).We then pooled the synergies identified as similar into an average 'reference' synergy structure with its respective activation coefficient.

Statistical analysis
Between-group differences in step characteristics were tested using a repeated measures ANOVA with direction as within-subjects variable and group as between-subjects variable.In case of significant group*direction effect we performed independent t-tests for each separate direction with a Bonferroni adjusted alpha of 0.01 (alpha = 0.05/5 directions).
Time varying activation coefficients of each synergy were compared between the two groups using two-tailed independent t-test statistical parametric mapping (SPM) with an α level of 0.05 (Pataky, 2010).SPM allows comparing differences between groups in the temporal domain, while avoiding potential controversies associated with discrete time point analysis (mean or peak amplitudes) (Pataky et al., 2015).Temporal windows with significant between-group differences are expressed as a percentage of time between perturbation onset (0 %), step onset (50 %) and foot strike (100 %).All SPM analysis were performed within the spm1d.04toolkit.

Results
All participants were able to adhere to the study procedures and step with the instructed leg, always using a single reactive step to recover balance.Participants performed reactive steps in all trials and none of the participants fell into the safety harness during any of the trials.On average, in the younger individuals we recorded 197 ± 2 steps taken according to instruction versus 192 ± 8 in the older participants.The few trials in which participants did not step according to instructions involved stepping with the wrong leg.

Muscle synergies
Muscle synergy analysis yielded five muscle synergies for the swing leg (average VAF, young: 0.85 ± 0.02, old: 0.84 ± 0.02) and five muscle synergies for the stance leg (average VAF, young: 0.83 ± 0.03, old: 0.83 ± 0.02).The number and structure of these synergies did not differ between stepping with the right or the left leg (Appendix A).We therefore pooled the results between both legs based on their stance or swing behavior.No significant differences were observed in the presence of muscle synergies between groups (Fig. 2).
Around step onset (~40-65 %), we observed activation in SW2 in anterior directions, in SW1 and SW3 in posterior directions; and in SW3 and SW5 in lateral directions.The older participants additionally coactivated SW3 (ANT 49 %-54 %) in the anterior direction, whereas the young typically had low SW3 activation coefficients in this direction.
During the swing phase, SW4 was active in lateral and anterior directions; SW1 and SW2 in posterior directions; and SW5 in all lateral directions.Between-group differences were limited to the anterior direction, where we observed higher activation in SW1 (89 %-99 %) in older participants.

Stance leg
Stance leg synergies also demonstrated direction-dependent activation patterns (Fig. 4).Early (APR-related) activation in the anterior direction was most prominent in ST3 and ST4, whereas ST1 and ST2 were more strongly recruited in all posterior and lateral directions.This activity typically lasted during step onset and well into the swing phase (i.e., of the contralateral leg).Around step onset, we observed additional ST5 activation in posterior directions; and ST4 activation in all directions.Towards the end of the step, ST2 and ST5 were recruited across all directions.

Discussion
In this study we aimed to identify potential age-related differences in muscle coordination patterns of reactive stepping following multidirectional balance perturbations.Our results show that in both the stance and the swing leg, similar sets of five muscle synergies were consistently recruited independent of age.However, the time-varying activation coefficients of three swing leg synergies and two stance leg synergies revealed differences between young and older participants in various phases of the reactive step.In addition, the older participants initiated their steps with a general delay as compared to the young, whereas step length and duration did not differ.
The novelty of our study lies in the comprehensive analysis that elucidated the muscle coordination patterns of both the stance and the swing leg throughout the full reactive stepping response.Our finding that a similar set of five synergies was consistently expressed in both groups shows that aging alone does not substantially affect the muscle coordination patterns that drive reactive steps.This finding is consistent with previous observations that muscle synergy structure remained unaffected by aging in relatively simple locomotion tasks (Guo et al., 2022;Monaco et al., 2010).This implies that muscle coordination patterns are not strongly affected by age when performing relatively simple locomotor tasks, whereas some age-related differences emerged with increased task difficulty (e.g.beam walking) (da Silva Costa et al., 2023).Our results suggests that the presently used perturbations did not impose a major challenge to the older participants, as underscored by their ability to successfully recover balance with a single step in all trials.
The higher activation coefficients in the older participants that we observed early after perturbation in synergies SW1, SW3 and ST1 indicate enhanced APR recruitment in both forward and backward Young 43 ( 7) 41 ( 8) 41 ( 7) 41 ( 7) 43 ( 8 directions.The APR is the first line of defense to protect a human being from falling following a loss of balance.Previous studies have shown a minor (on average 14 ms) delay in these APRs with aging, but findings regarding age-related differences in APR magnitude are inconsistent (Tokuno et al., 2010).As APR magnitudes strongly depend on (un)certainty of the intensity and direction of the perturbation and also on the task instructions, the observed age-related differences have to be interpreted against this background.In our study, participants were aware of the uniform intensity (but not direction) of all perturbations, which could have allowed them to respond with a (direction-specific) default magnitude.All perturbations required stepping for successful recovery and the participants were instructed to take a step as needed.In previous studies, an instruction to step was shown to substantially reduce APR magnitudes compared to an instruction to keep the feet in place (McIlroy and Maki, 1993).Enhanced APR recruitment in the older participants may thus indicate a relative bias towards non-stepping behavior for recovering balance, which interpretation appears consistent with the observed delay in step onsets.
The finding that step onsets were delayed in the older participants is inconsistent with previous reports of similar or even advanced initiation of reactive steps (Mille et al., 2013;Mille et al., 2003).It has been postulated that older people may initiate their steps faster by using nonspecific sensory input to trigger a pre-selected stepping strategy, whereas the young may first process specific movement-related information to inform their stepping response (Rogers et al., 2003).It is suggested that such pre-selected mechanisms may reduce potential costs to stability, and may be enhanced by anxiety to fall.Yet, recruiting a stereotyped response does not appear to be beneficial when exposed to unexpected multidirectional perturbations that require directionspecific synergistic muscle responses for balance recovery, as was the case in the present study.While previous research has demonstrated equivalent perceptual acuity of perturbation direction between young and older adults (Bong et al., 2020), the presently observed delay in step onsets suggest that the older participants may have needed more time to accumulate information on the direction of perturbation and make the corresponding sensorimotor transformations for initiating a step in the correct direction (Dully et al., 2018).
Around step onset in lateral directions, the older participants demonstrated more pronounced synergy co-activation than the young did, particularly in the swing leg.We found concurrent activation of SW3 and SW4, which synergies contained antagonistic lower leg muscles.Increased co-activation of lower limb muscles promotes joint stiffness and stability (Lee et al., 2017), which mechanism has been suggested to compensate for neuromuscular impairments associated with aging.Yet, in the case of reactive balance control it appears to be rather maladaptive, as increased levels of co-activation were shown to be associated with a greater likelihood of unsuccessful recovery (Sawers et al., 2017;Falk et al., 2022).In addition, one may suggest increased coactivation of lower-leg muscles to potentially impede toe clearance and increase the risk of tripping, yet we did not observe significant betweengroup differences.Therefore, it remains elusive to what extent this coactivation may impede reactive stepping.

Limitations
While the presently observed age-related differences in muscle coordination patterns of reactive stepping appeared to be modest, it must be noted that our older participants were still relatively young with a mean age of 64 years, which may be considered a limitation.It remains to be studied whether more pronounced impairments in muscle coordination patterns due to aging may emerge in people of older age.Furthermore, it should be mentioned that our stepping leg instruction may have impeded natural reactive behavior to some extent.For instance, its combination with the perturbation directions within each trial block implicitly restricted cross-stepping strategies.We chose to focus on side steps because of consistency in step and swing leg across perturbation directions within each block of trials.This was needed for conducting the muscle synergy analysis, i.e., for obtaining symmetrical data for comparisons between leg and perturbation directions.Yet, this approach may have limited the generalizability of our results, given the greater likelihood of older individuals taking cross-over steps when behavior is not constrained (Mille et al., 2005).Investigating age-related differences in muscle coordination of cross-stepping behavior would be an interesting topic for future studies.In addition, the fixed timing of platform deceleration may have allowed our participants to modulate synergy recruitment prior to foot strike.Such anticipatory modulation in muscle activity has previously been demonstrated for feet-in-place strategies, in young and older participants alike (Tokuno et al., 2010;Carpenter et al., 2005).In future research, it may be of interest to study how the presently identified synergy recruitment is influenced by higher perturbation intensities, different perturbation profiles or by the presence of co-existing balance impairments.
In conclusion, our current findings provide novel insight into agerelated changes in muscle coordination during reactive steps.Synergy structure was not affected by age, yet our results point at a relative bias of early synergy recruitment towards non-stepping behavior in the older adults.These findings indicate that, even though the older adults managed to successfully recover balance in all trials, emerging deficits may impact their ability to recover under more challenging conditions.While there is mounting evidence that perturbation-based training (PBT) is effective for improving reactive balance capacity and preventing falls in older adults, the mechanisms underlying these beneficial effects remain elusive.The presently used methodology may help to unravel these mechanisms and provide guidance for future tailoring of PBT to individual needs.

Declaration of competing interest
None.

Fig. 2 .
Fig. 2. Representation of five (average) muscle synergies and the respective muscle weights identified in the swing and stance leg of the respective age groups.SW=Swing Leg, ST = Stance Leg.Muscle order is reflected by the first synergy and consistent across synergies.Muscles are ordered as follows: tibialis anterior (TA), peroneus longus (PER), soleus (SOL), semitendinosus (SEMT), biceps femoris (BFEM), rectus femoris (RFEM), erector spinae (ERSP), gluteus medius (GLUT) Older adults are shown in blue and younger adults in orange.Top right numbers of each synergy reflect their presence across both groups.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3 .
Fig. 3. Swing leg muscle synergies and their activation coefficients per stepping direction.Mean activation (+ SD) coefficients of elder participants are shown in blue, younger participants are shown in orange.Gray shaded areas indicate significant between group differences for each stepping condition.PO=Perturbation Onset, SO=Step Onset, FS=Foot Strike.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) Wouter H.A. Staring: Writingoriginal draft, Visualization, Methodology, Formal analysis, Data curation, Conceptualization.Sarah Zandvliet: Writingreview & editing.Digna de Kam: Writingreview & editing, Conceptualization.Teodoro Solis-Escalante: Writingreview & editing, Data curation, Conceptualization.Alexander C.H. Geurts: Writingreview & editing, Supervision.Vivian Weerdesteyn: Writingreview & editing, Supervision, Funding acquisition,

Fig. 4 .
Fig. 4. Stance leg muscle synergies and their activation coefficients.Mean activation (+ SD) coefficients of older participants are shown in blue, young participants are shown in orange.Gray shaded areas indicate significant between group differences for each stepping condition.PO=Perturbation Onset, SO=Step Onset, FS=Foot Strike.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 1
Stepping characteristics (mean and SD) across different stepping directions and age groups.