Spatiotemporal modulation of a fixed set of muscle synergies during unpredictable and predictable gait perturbations in older adults

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 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 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 facilitate the effectiveness to deal with perturbations and promote the transfer of adaptation between different kinds of perturbations. Summary statement The flexible recruitment of the four and well-known muscle synergies responsible for unperturbed walking during unpredictable and predictable gait perturbations indicates an effective way to counteract locomotor perturbations, where fast reactive responses are necessary to maintain postural stability.


Introduction
for unperturbed walking. For all analyzed parameters, differences between unpredictable and predictable 262 cycles were investigated by pairwise comparisons of the matching cycles (e.g. cycle 1 in unpredictable 263 vs. cycle 1 in predictable). The p-values for all post-hoc tests were adjusted for multiple comparisons 264 according to Benjamini-Hochberg. All these analyses were conducted in R v4.1.0. 265 To check for a proactive modulation of muscle activity in the predictable trials, statistical parametric 266 mapping (SPM) paired two-tailed t-tests (Pataky, 2010)

277
We found a significant effect of predictability (p < 0.001) and gait cycle (p < 0.001) on cadence, stance 278 time, swing time and duty factor in both mediolateral and anteroposterior perturbations (table 1). The 279 post-hoc analysis showed that all four temporal gait parameters were significantly affected by the 280 perturbation in at least one and up to seven cycles (p < 0.05) in each of the four perturbed conditions 281 compared to the respective unperturbed gait cycle (table 1). The three cycles before the perturbation did 282 not show any statistically significant differences (p = 0.228 to 0.998) in the temporal parameters in 283 either condition (table 1). Cadence generally increased, while stance and swing durations decreased due 284 to the perturbation. The duty factor increased in the perturbed cycle but decreased in the following 285 recovery cycles compared to unperturbed walking. The observed alterations in the gait parameters 286 during the unpredictable condition were significantly higher in both perturbation directions in two and 287 up to seven recovery gait cycles (p < 0.05) compared to the predictable one (table 1). 288 An example of the EMG-activity in the unperturbed, perturbed and recovery gait cycles of every 289 recorded muscle is presented in figure 2. In all gait cycles of all conditions, we identified four muscle 290 synergies that were able to sufficiently reconstruct the recorded EMG data (R² was 0.89 ± 0.03 for 291 mediolateral unpredictable, 0.88 ± 0.03 for mediolateral predictable, 0.90 ± 0.03 for anteroposterior 292 unpredictable and 0.89 ± 0.02 for anteroposterior predictable). The four synergies are functionally 293 associated with the phases of the gait, namely the weight acceptance synergy with high activation of 294 knee extensors and glutei, the propulsion synergy with a main contribution of plantar flexors, the early 295 swing synergy mainly represented by the foot dorsiflexors and the late swing synergy highly related to 296 the knee flexors ( figure 3). 297 We found a statistically significant effect of gait cycle (p < 0.007) on the CS of the activation patterns 298 (figure 4) and muscle weights (figure 5) in all synergies, which shows a modulation of the 299 spatiotemporal structure of the muscle synergies during and after the perturbation (figures 6). The CS 300 of the activation patterns decreased (p < 0.001) in all synergies and conditions during the perturbation 301 cycle compared to the unperturbed one (figure 4). The CS of the muscle weights was also significantly 302 decreased (p < 0.04) in most of the synergies during the perturbation and the first recovery cycle (figure 303 5). Furthermore, there was a statistically significant effect of predictability (p < 0.05) on almost all CS 304 of the activation patterns and muscle weights with lower values in the unpredictable condition in both 305 the mediolateral and anteroposterior direction (figures 4 and 5). Only in the mediolateral weight 306 acceptance synergy of the muscle weights, the decrease of CS in the predictable condition was not 307 statistically significant (p = 0.12), while a significant (p < 0.001) interaction effect of predictability by 308 gait cycle was found in that synergy. The significant effect of predictability on the similarity of the 309 activation patterns was present until the third-fourth recovery cycle (figure 4). 310 We found a significant effect of cycle (p < 0.02) and predictability (p < 0.02) on the CoA for nearly all 311 synergies in both mediolateral and anteroposterior direction (table 2). Only the early swing synergy 312 showed no significant effect of predictability (p < 0.349), but an interaction effect of cycle by 313 predictability (p < 0.001). In the perturbation and the first recovery cycle, the CoA of the activation 314 patterns shifted either earlier or later within the gait cycle compared to the unperturbed walking with 315 greater changes in the unpredictable condition (table 2). The shift of the CoA was statistically significant 316 in all synergies and conditions (p < 0.05) except for the late swing synergy in the predictable 317 anteroposterior condition. The alterations in the CoA lasted until the 5th recovery cycle (table 2). There 318 was an effect of gait cycle (p < 0.005) and a cycle by predictability interaction (p < 0.005) on the FWHM 319 of nearly all synergies in both the mediolateral and anteroposterior direction. In most cases, the FWHM 320 of the activation patterns became wider (table 3). The significant differences compared to the 321 unperturbed walking were mainly observed in the perturbation and in the first recovery cycle for all 322 investigated conditions. The increased widening of the activation patterns during the perturbation and 323 first recovery cycle was in most cases larger in the unpredictable condition (table 3). There was a 324 decrease in FWHM of the propulsion synergy in the first recovery cycle after the unpredictable 325 anteroposterior condition (table 3). 326 The SPM analysis showed significant differences (p < 0.05) of EMG-activity before the initiation of the 327 perturbation between the predictable and unpredictable conditions (figure 7). In both mediolateral and 328 anteroposterior perturbations, rectus femoris, vastus medialis and vastus lateralis showed a significantly 329 higher EMG-activity (p ≤ 0.002) before the perturbation in the predictable compared to the 330 unpredictable condition. During the mediolateral perturbations, biceps femoris (p = 0.049), peroneus longus (p < 0.001) and gastrocnemius lateralis (p < 0.001) showed a significantly higher EMG-activity 332 in the predictable condition, whereas in the anteroposterior perturbations, the EMG-activity of 333 semitendinosus (p = 0.006), tibialis anterior (p = 0.003) and gastrocnemius medialis (p = 0.033) was 334 higher (figure 7). 335 336

337
In the current study, we investigated the effects of predictable and unpredictable perturbations on the 338 modular organization of locomotion in older adults. For both the muscle weights and activation patterns, 339 we found a significant decrease in CS between the perturbed cycle and the following recovery cycles 340 compared to unperturbed walking, indicating a modulation of the spatiotemporal structure of muscle 341 synergies by the sensorimotor system in order to maintain gait stability after the perturbations. The The recruitment of muscle synergies in the perturbed gait cycles was associated with specific functional 391 phases of walking and their modulation corresponded to the perturbation-specific stability requirements. to a longer activation and thus longer force generation of the plantar flexor muscles relative to the stance 412 phase during propulsion, indicating a mechanism for an effective push-off and initiation of the swing 413 phase. It is to mention that muscle force generation is also dependent on muscle contractile conditions 414 like muscle force potentials due to the force-length and force-velocity relationships. Nevertheless, a 415 widening of the activation of the triceps surae muscles in spite of varying force-length-velocity 416 potentials will introduce a longer force generation and thus a more effective push-off. 417 The widening of the basic activation patterns of the muscle synergies during the perturbed and the first 418 recovery steps lead to an increased overlap of activation between chronologically adjacent synergies 419 and can be interpreted as a compensatory mechanism adopted by the central nervous system to deal with    Cosine similarity (CS) of the activation patterns of the muscle synergies for the two unperturbed cycles (U2, U3), the perturbation cycle (P) and the following 8 recovery cycles (R1 to R8) with regard to the first unperturbed cycle. Statistically significant differences to the CS of the unperturbed cycle (U2) (*) and between the unpredictable and predictable condition ( †) are highlighted (p<0.05).

Figure 5.
Cosine similarity (CS) of the muscle weights of the muscle synergies for the two unperturbed cycles (U2, U3), the perturbation cycle (P) and the following 8 recovery cycles (R1 to R8) with regard to the first unperturbed cycle. Statistically significant differences to the CS of the unperturbed cycle (U2) (*) and between unpredictable and predictable condition ( †) are highlighted (p<0.05).

Figure 6.
Mean curves of the activation patterns of the muscle synergies for the four different conditions during the unperturbed walking (U1), the perturbation cycle (P) and the following three recovery cycles (R1 to R3). All amplitudes of the activation patterns were normalized to 1. The stance and swing phases were normalized to an equal distribution of data points.          Fig. 7 Table 1. Means ± SD values of cadence, stance time, swing time and duty factor during the three unperturbed cycles (U1 to U3), the perturbation cycle (P) and the following 8 recovery cycles (R1 to R8) for the four different conditions. Statistically significant differences to the first unperturbed cycle (*) and between the unpredictable and predictable condition ( †) are highlighted (p<0.05).  Table 2. Means ± SD values of the center of activity (CoA) of the activation patterns of the muscle synergies during the three unperturbed cycles (U1 to U3), the perturbation cycle (P) and the following 8 recovery cycles (R1 to R8) for the four different conditions. Statistically significant differences to the first unperturbed cycle (*) and between the unpredictable and predictable condition ( †) are highlighted (p<0.05).  Table 3. Means ± SD values of the full width at half maximum (FWHM) of the activation patterns of the muscle synergies during the three unperturbed cycles (U1 to U3), the perturbation cycle (P) and the following 8 recovery cycles (R1 to R8) for the four different conditions. Statistically significant differences to the first unperturbed cycle (*) between the unpredictable and predictable condition ( †) are highlighted (p<0.05).