Older but not younger adults rely on multijoint coordination to stabilize the swinging limb when performing a novel cued walking task

Abstract

Motor flexibility - the ability to employ multiple motor strategies to meet task demands - may facilitate ambulation in complex environments that constrain movements; loss of motor flexibility may impair mobility. The purpose of this study was to determine the effects of obesity (a specific model of mobility impairment) and advanced age on motor flexibility during a task that constrained foot placement while walking. Twenty-one community-dwelling obese (OB) and 25 normal weight (NW) older adults (46 total older adults - OA) and 10 younger adults (YA) walked normally on a treadmill (baseline) then walked while stepping on lighted cues projected onto the treadmill at locations corresponding to average foot placement during normal walking (cued). The uncontrolled manifold (UCM) analysis was used to partition total variance in a set of seven lower-limb segment angles into components that did (“bad” variance) and did not (“good” variance) affect step-to-step variance in the trajectory of the swing foot. Motor flexibility was operationalized as an increase (baseline to cued) in total variance with an increase in good variance that exceeded changes in bad variance. There was no significant group x walking task interaction for total and good variance for OB vs NW, but there was a strong and significant effect for OA vs YA (p<.01; Cohen’s d>1.0). Whereas YA reduced both good and bad variance, OA increased good variance beyond the change in bad variance. In OA, these changes were associated with several functional measures of mobility. Cued walking may place greater demands on OA requiring greater reliance on motor flexibility, although functional and cognitive declines associated with obesity in otherwise healthy OA may be able to be compensated for by increasing motor flexibility under such tasks. The extent to which motor flexibility is employed during novel or constrained tasks may be a biomarker of healthy aging and a target for (re)habilitation.

INTRODUCTION

The problem of motor redundancy, whereby the number of motor components, i.e., degrees of freedom (DoFs), involved in an action exceeds those absolutely necessary to perform the action (Bernstein 1966), is central to the production of human movement. Motor redundancy applies at all levels of the musculoskeletal system: muscle forces develop from coordinated firing of multiple motor units within the muscle; joint moments develop from coordinated production of force by multiple muscles crossing the joint; during reaching, the trajectory of the hand results from coordinated joint motions of the wrist, elbow, and shoulder. As a result of motor redundancy, for any given motor problem multiple motor solutions exist - no single combination of motor unit actions, muscle forces, or joint rotations leads to a given muscle force, joint moment, or hand position. Some posit that individuals address the “problem” of motor redundancy by minimizing solutions to only those that optimize certain functions. However, such approaches ignore the possibility that the motor system may benefit from exploring the multitude of solutions offered by redundancy. Exploiting motor redundancy may facilitate performance of secondary actions that concurrently rely upon the same DoF involved in the primary task (Zhang et al. 2008) and allows the system to appropriately react to perturbations (e.g., system noise or movement errors) and altered task demands (Hsu and Scholz 2012; Latash et al. 2007; Mattos et al. 2011). In this sense, exploiting motor redundancy provides flexibility to the motor system (Latash 2018; 2012), where motor flexibility is broadly defined as the ability to employ a range of motor solutions that ensure stable (invariant) performance of a task

Motor flexibility in the lower limbs is critical when faced with novel conditions that constrain performance. For example, to accommodate constraints in foot placement during hopping (Auyang and Chang 2013), younger adults increase variance in joint angles but in a manner that does not increase variance in (stabilizes) leg orientation. Similarly, when foot placement is constrained during gait by having participants walk on a set of narrow beams, younger adults increase variance in segment angles in a manner that stabilizes the swing foot trajectory resulting in reduced step-to-step variance in step width (Rosenblatt et al. 2014b).

The extent to which older adults are able to employ motor flexibility in the lower limbs during conditions that constrain the system may be critical to maintaining mobility. Indeed, environments that place constraints on foot placement are commonly encountered during community ambulation, e.g., avoiding a pothole or icy patch. An older adult who lacks the motor flexibility required to accommodate such constraints may have an increased likelihood of balance-loss. Despite the possibility that motor flexibility may facilitate mobility in older adults, ageing is often thought to reduce motor flexibility. Indeed, motor inflexibility is traditionally thought to be a hallmark of aging, and has been reported across a variety of non-locomotor tasks (Dutta et al. 2013; Hsu et al. 2013; Hsu et al. 2014). This is not to say that older adults are unable to employ motor flexibility across locomotor tasks. For example, to increase walking speed heathy OA preferentially increase the extent to which the hip joint (relative to knee and ankle) contributes to total positive work (Kuhman et al. 2018a). Similarly, OA preferentially increase the peak positive power and positive work in the hip extensors compared to the ankle plantarflexors when walking on a 10° incline (Kuhman et al. 2018b). Nonetheless, motor flexibility within these studies does not necessitate exploitation of motor abundance; redistribution of joint kinetics to meet tasks demands need not rely on exploration of multiple motor solutions.

Operationalizing motor flexibility within the framework of motor abundance is particularly relevant when considering motor performance under environmental constraints. Neuromuscular control strategies that exploit motor redundancy offer an advantage of facilitating performance in the presence of multiple task constraints, or multiple ongoing motor tasks that rely on similar DoFs (Gera et al. 2010; Hsu et al. 2014). Moreover, such strategies may provide a means to compensate for age-related neuromusculoskeletal changes, e.g., increased motor noise and variability (Christou 2011; Goble et al. 2009; Kang and Dingwell 2009), which may amplify perturbations to the system imposed by a given task. For example, older adults exploit motor redundancy to compensate for reduced strength during sit-to-stand (Greve et al. 2013) and to compensate for age related motor noise when walking across uneven surfaces (Eckardt and Rosenblatt 2018). However, uneven walking surfaces do not place specific constraints on movement. Demonstration of motor flexibility in older adults during a locomotor task that does so would provide additional support that exploiting motor redundancy provides distinct advantages during gait and is thus important for mobility in older adults.

If, to some extent, mobility relies on an ability to exploit motor redundancy, then populations with impaired mobility may be less able to do so. Initial evidence for a generalized relationship between mobility and motor flexibility may be best obtained by considering a population with a range of mobility levels but no specific neuromuscular impairments; obese older adults may represent an ideal such population. Obesity is known to impair balance, strength, and function (De Stefano et al. 2015; Madigan et al. 2014; Teasdale et al. 2013), and considerable evidence suggests that obesity impairs mobility in youth (Wang et al. 2016) as well as older adults (Gonzalez et al. 2020). Although obesity may not hinder local dynamic stability when younger (Liu and Yang 2017) and older adults (Gonzalez et al. 2020) are able to self-regulate their walking pattern, there are reasons to believe that obesity would impair the ability to exploit motor redundancy, particularly under challenging conditions. For example, obese individuals have difficulty utilizing feedback to make adjustments during ongoing movements (Gaul et al. 2018) and are less capable of integrating sensory stimuli from multiple sources (Scarpina et al. 2016). The idea that increased body mass alone can impact one’s ability to exploit motor abundance is best demonstrated by the fact that healthy young adults alter components of movement variability related to motor flexibility simply in response to the application of external loads during gait (Qu 2012).

The primary purpose of this study was to determine the effects of obesity and age on motor flexibility as it relates to foot placement during a locomotor task that specifically constrains foot placement. A secondary purpose was to explore the extent to which reliance on motor flexibility during cued walking was related to measures of functional mobility in older adults. While motor flexibility can be quantified using a number of techniques and metrics, we chose to use the uncontrolled manifold (UCM) analysis. Within the UCM analysis, the use of motor flexibility is evaluated relative to a specific performance variable important to the task; it can be operationalized as the coordination among redundant DoFs (those that contribute to performance) in a manner that increases the number of motor solutions used without increasing step-to-step variance in performance. Accordingly, by this particular definition, motor flexibility can be dichotomized and an individual can be described as having (or not having) relied on flexibility to perform a task. We performed UCM analysis on lower limb segmental angles (our redundant DoFs) during a task that required participants to continuously step on lighted cues corresponding to average foot placement during normal gait. We hypothesized that 1) normal weight, but not obese older adults would rely on motor flexibility during cued walking and that the extent to which they did so would relate to functional mobility; and 2) both younger and older adults would rely on motor flexibility during cued walking to stabilize the trajectory of the foot during the swing phase of the gait cycle.

METHODS

Participants

The current study is part of a larger study, approved by the Rosalind Franklin University IRB, quantifying the effects of obesity on fall-risk in older adults. Initial inclusion criteria for the larger study included: age 65 years or older, self-reported ability to walk a mile at any pace with minimum rest if needed and being normal weight (NW) or obese (OB) based on body mass index (BMI=18.5-24.9 kg/m² or BMI≥30 kg/m², respectively). Exclusion criteria included: use of an assistive device for walking, artificial joint replacement, self-reported history of neurological conditions that interfere with gait (e.g., Parkinsons’ disease, multiple sclerosis), self-reported prior diagnosis of diabetic neuropathy, self-reported history of osteoporosis (verified with a study-provided DEXA scan and T-score of ≤2.5 for the femoral neck) , or any of the following determined during a physical exam with a study clinician: compromised range of motion in the lower limb or trunk, untreated hypertension, cardiovascular abnormalities, any other pathophysiology that could compromise participant safety. All eligible participants provided written informed consent to participate. Of the 57 participants who qualified (29 NW and 28 OB), the current study includes data from 25 NW participants and 21 OB participants (see Table 1 for demographics); technical issues and failure to return to the lab for the second day of testing when this protocol took place explain the difference

Table 1.

The effects of obesity on measure from the UCM analysis, on performance during cued walking task and on measures of functional mobility.

		OB (N=21)	NW (N=25)	p-value; F-value (df)	d
	Age (years)	72.0 (8.5)	68.0 (6.0)	0.06	0.59
	BMI (kg/m2)	33.7 (5.6)	23.1 (2.1)	<0.001	3.14
	Sex (male/female)	9/12	12/13	0.72	0.10
UCM measures	Vz baseline	1.35±0.33 [1.36±0.34]	1.30±0.33 [1.29±0.27]	group x task: 0.71; 0.15 (1,43)	0.12
	Vz cue	1.71±0.14 [1.76±0.34]	1.35±0.29 [1.58±0.21]	group: 0.43; 0.64 (1,43)	0.24
	log VUCM baseline	−7.80±0.51 [−7.74±0.58]	−7.88±0.50 [−7.92±0.37]	group x task: 0.67; 0.19 (1,43)	0.13
	log VUCM cue	−7.53±0.46 [−7.56±0.42]	−7.68±0.46 [−7.65±0.43]	group: 0.39; 0.75 (1,43)	0.25
	log VORT baseline	−8.60±0.71 [−8.58±0.72]	−8.50±0.70 [−8.51±0.58]	group x task: 0.98; 0.001 (1,43)	0.01
	log VORT cue	−9.11±0.65 [−9.22±0.73]	−9.02±0.65 [−8.93±0.54]	group: 0.66; 0.20 (1,43)	0.14
	log VTOT baseline	−7.87±0.50 [−7.81±0.47]	−7.94±0.50 [−7.98±0.47]	group x task: 0.63; 0.24 (1,43)	0.15
	log VTOT cue	−7.64±0.46 [−7.67±0.43]	−7.79±0.46 [−7.77±0.43]	group: 0.42; 0.65 (1,43)	0.24
Performance measures	Accuracy (%)	94.6 (8.0)	95.6 (12.7)	0.77	0.09
	SSV (m/s)	0.76±0.20	1.00±0.22	0.001	1.06
	SW baseline (mm)	150.5±39.2 [162.0±43.9]	126.5±38.8 [116.9±38.7]	group x task: 0.37; 0.82 (1,43)	0.27
	SW cue (mm)	144.0±36.1 [153.4±40.0]	115.4±35.7 [107.5±34.2]	group: 0.03; 5.21 (1,43)	0.69
	Log SWV baseline	3.25±0.37 [3.19±0.39]	3.32±0.36 [3.37±0.31]	group x task: 0.43; 0.63 (1,43)	0.24
	Log SWV cue	3.05±0.38 [3.00±0.43]	3.02±0.38 [3.06±0.28]	group: 0.80; 0.06 (1,43)	0.08
	SL baseline (mm)	430.6±45.8 [386.6±86.7]	433.9±45.3 [470.9±87.5]	group x task: 0.09; 3.03 (1,43) group: 0.58;0.31 (1,43) [<0.01; 11.58 (1,44)]	0.53 0.17 [1.03]
	SL cue (mm)	432.5±54.1 [390.0±90.7]	446.4±53.5 [482.0±87.2]
	Log SLV baseline	3.58±0.40 [3.49±0.29]	3.39±0.40 [3.50±0.37]	group x task: <0.01*; 10.19 (1,43) [0.54; 0.38 (1,44)] group: 0.77; 0.09 (1,43)	0.96 [0.19] 0.09
	Log SLV cue (mm)	3.41±0.34 [3.51±0.50]	3.55±0.34 [3.44±0.28]
Function	10 MW (m/s)	1.71±0.41	1.99±0.27	0.01	0.79
	TUG (s)	9.60 (2.41)	7.52 (1.04)	0.001	1.07
	4SS (s)	8.88 (3.38)	7.94 (2.83)	0.03	0.66
	SLS (s)	10.90 (26.39)	30.00 (3.66)	0.001	1.00
	F8W (s)	8.32 (1.27)	6.66 (1.58)	<0.01	1.09

Descriptive data are presented as either mean± standard deviation for normally-distributed variables or median (interquartile range) for non-normally distributed variables. Descriptive data is shown for speed adjusted values with raw values in square brackets, except for accuracy, SSV and all functional measures which are not adjusted. All p-values and Cohen’s d values are shown for adjusted data except when statistical findings differed between adjusted and unadjusted data (i.e. p-value <0.05 for only one), in which case p-values and Cohen’s d for unadjusted data are shown in square brackets.

^*NW significantly increase SLV (p=0.15; d = 0.77); OB significantly decrease SLV (p=0.03; d = 0.68)

In addition to NW and OB older participants, we recruited 10 younger adults (7F/3M; age: 21.1±2.6 years; BMI: 23.9±3.1 kg/m²). These younger adults (YA) were all university students who self-reported no history of neuromusculoskeletal conditions that would interfere with walking.

Functional measures

The following five tests of functional mobility were assessed in older adults only, in the stated order: 10-meter Walk test (10mW), Timed Up and Go (TUG), Figure of Eight Walk test (F8W), Four Square Step test (4SS) and single leg stance (SLS). For 10mW, participants walked 10m at their maximum speed and the time to walk the middle 6m was recorded with a stopwatch, from which speed (in m/s) was calculated. The average of 3 trials was used in analyses. For the TUG, participants rose from a chair, walked 3 meters, turned and then sat down again, and the time from initial rise to final seating was recorded. For F8W, participants stood between two cones separated by five feet and then completed a figure-8 walking pattern around the cones while walking at a comfortable speed. The time to complete the test was recorded. The F8W is a valid measure of “walking skill” in older adults (Hess et al. 2010). The 4SS is a test of dynamic balance in which participants must quickly step forwards, backwards, and sideways to complete a square-shaped pattern in the clockwise and counterclockwise directions (Dite and Temple 2002). The test was performed twice, after a practice trial, and the fastest time was used for analysis. Finally, in SLS, participants stand on one limb with hands on the hips for up to 30 seconds. This was performed three times with the maximum time recorded was entered into analyses.

Walking Protocol

As part of the larger study, participants first donned a safety harness and walked on a calibrated, motorized treadmill (C-Mill; Motek, Amsterdam) for 10 minutes at self-selected velocity (SSV), where SSV was determined using an iterative process (Rosenblatt et al. 2014a). Participants then rested while instructions were presented regarding the cued walking tasks. Participants were informed that while they walked on the treadmill, lighted cues would appear on the treadmill and that they should: “try to get the center of your foot as close to the center of the white lighted box as possible” (see Figure 1). This task was chosen as it provided a controlled means to place constraints on motor solutions by increasing the need for precision of foot placement on a step-by-step basis, and thus the need for control of the swinging foot. Moreover, a previous study showed that by simply constraining medial/lateral foot placement to the preferred step width, young healthy adults increased motor flexibility with a concomitant decrease in step width variability (increased precision) (Rosenblatt et al. 2014b). Participants then performed the baseline task, which involved walking on the treadmill for 50 s at 80% of SSV. Ground reaction force data measured during the last 10 s of the task was used to calculate average step width (SW) and step length (SL) of each participant in real time. Immediately following the 50 s of baseline walking, rectangular lighted cues were projected onto the treadmill surface for 50 s at locations such that the center to center distance corresponded to the participants average SW and SL. The length and width of the rectangular cues were based on the participants shoe size. The primary measure to quantify performance on the cued walking task was accuracy, i.e., the percent of cues for which the center of pressure at the time of contralateral mid-swing fell within the cue, as pre-defined by the treadmill software.

Figure 1.

Cued walking task. During the cued walking task, participants had to step to lighted cues projected onto the treadmill surface at locations corresponding to the average step length and step width determined during normal walking

Throughout both walking tasks, an 8-camera motion capture system (Vicon; Oxford, UK) tracked the motions of passive reflective markers placed on body landmarks according to the full body plug in gait model, or in the case of the OB older adults an obesity-specific marker set (Lerner et al. 2014). Marker data was initially processed using commercial software (Nexus; Oxford, UK). For NW, Nexus software also provided locations of the lower extremity joint centers which were needed for UCM analysis. For OB participants, custom-built code (Matlab; Mathworks, Cambridge, MA) was used to obtain locations of the lower extremity joint centers. Specifically, the ankle and knee joint centers were calculated as the midpoint of markers placed on the medial and lateral aspects of each joint, the hip joint centers were determined by virtually recreating anterior superior iliac spine (ASIS) markers based on the position of the sacral marker cluster and a digitized pointer placed over the ASIS then using well-defined relationships of the hip joint center relative to these pelvic landmarks (Davis III et al. 1991). After calculating joint centers, these positions were used to create a four-segment geometric model (Krishnan et al. 2013) of the stance leg, pelvis, swing-limb thigh and swing-limb shank and segment angles were calculated. For this analysis, data corresponding to the swing phase only was normalized from 0-100% corresponding to each left-leg swing (toe off to ipsilateral heel strike) and then entered into custom software to conduct the UCM analysis. Toe and heel marker data were also used to calculate SW, SL and their respective variabilities (standard deviations) separately during the baseline and cued task, all of which were used as additional measures of performance.

UCM analysis:

The UCM analysis quantifies how trial-to-trial variance in a set of redundant DoFs that contribute to performance (i.e. elemental variables) acts to facilitate consistent trial-to-trial output in a task-relevant variable that defines performance (Scholz and Schoner 1999). For the current analysis, the set of elemental variables were seven segmental angles of the lower limbs and the performance variable of interest was the mediolateral (ML) trajectory of the swing limb foot (Krishnan et al. 2013; Rosenblatt and Hurt 2019; Rosenblatt et al. 2015) defined as the mediolateral position of the swing limb ankle joint center relative to that of the stance limb (AJC_ML). This performance variable was chosen due to the importance of mediolateral foot placement in the control of mechanical stability while walking (Bauby and Kuo 2000; Donelan et al. 2004) and the role of interlimb coordination in dictating adjustments in foot placement (Arvin et al. 2018; Bruijn and Van Dieën 2018; Kubinski et al. 2015; Rankin et al. 2014). Moreover, the constraints of the cued walking task specifically sought to limit step-to-step variance in swing foot placement, making this an inherently important performance variable. The following four step process was used to perform the analysis (see Scholz and Schoner, 1999 for additional details):

1. A previously described geometric model was created to express the performance variables as functions of the elemental variables (Krishnan et al. 2013).

2. A Jacobian matrix was derived from the model to relate changes in elemental variables to changes in the performance variable. At every percent of the normalized swing phase, the Jacobian was evaluated at the mean values of the elemental variables. The null space of the evaluated Jacobian defined the UCM.

3. Following prior analyses, for each left-step, deviation vectors were independently calculated at every percent of swing as the difference between the elemental variables at that point and their respective means. These vectors were projected onto the UCM and a space orthogonal to it. The squared length of the projected vectors in each space, relative to the number of DoFs in that space, was averaged across multiple steps to define two variance components: “good” variance, or the variance within the UCM (V_UCM) which does not cause step-to-step variance in performance; and “bad” variance, or the variance orthogonal to UCM (V_ORT) which causes step-to-step variance in performance. The total variance in elemental space (V_TOT) per DoF was calculated as the sum of the two variance components after multiplying each by the number of DoFs within and orthogonal to the UCM (here, 6 and 1, respectively) and dividing by the total number of DoFs.

4. A synergy index was calculated as (V_UCM - V_ORT)/V_TOT and then z-transformed (ΔV_Z) (Robert et al. 2008). Based on the DOF considered herein, if ΔV_Z > 0.89 then the swing limb foot was stabilized by a kinematic synergy, i.e., covariation among segmental angles, or multijoint coordination, acted to limit variance in the ML trajectory of the swing limb foot. If ΔV_Z falls below this threshold then a synergy is absent, the magnitude has no meaning. ΔV_Z was averaged across the swing phase as were variance components and V_TOT. As suggested by Verrel (2010), variance components and V_TOT were log transformed before conducting analyses (Verrel 2010).

Statistical analysis

Effect of obesity on UCM-derived variables and performance measures

Since the study did not include an obese YA group, we could not run a two factor analysis to test for age and obesity effects. Instead, we ran separate tests to independently evaluate each effect, first testing for a difference between NW and OB older adults then testing for an effect of age on UCM-derived measures. The rationale was that if our hypothesis regarding obesity was supported, then in the second test we would compare YA and NW only, otherwise we would combine NW and OB and treat them as a single group of older adults (OA) in the analysis.

For each of the four UCM-derived variables (synergy index and log-transformed variance components), and for each walking task and group, we first checked the data for normality using Shapiro-Wilk tests. If a variable was found to be non-normal for a given task and if an extreme outlier existed (a point that was more/less than three times the interquartile range above/below the upper/lower quartile) for that variable, then the subject associated with that outlier was removed from all ensuing analyses (1 NW subject). The variables were then all normally distributed. Cued walking performance variables were also checked for normality (and outliers) using Shapiro-Wilk tests. If performance variables data for either condition was non-normal, then the variable was log-transformed for both conditions and again checked for normality.

After confirming normality, we ran separate two-factor repeated-measures ANOVAs (between subject factor of group: NW vs. OB; within subject factor of task: baseline vs. cued) to test for between-group differences in the four UCM-related measures and performance measures. Given that the amount of V_ORT relative to V_UCM may be modulated with velocity of the swing limb (the performance variable) (Friedman et al. 2009; Goodman et al. 2005) and that speed can effect gait variability in general (Chien et al. 2015; Kang and Dingwell 2008), in the event of a significant effect of BMI on SSV, the latter would be considered as a covariate in the ANOVAs. A significant task x group interaction for any of the UCM variables would indicate that one of the OA groups responded to the cued task differently than the other. The only variables that were not compared using ANOVAs were accuracy and SSV, for which Mann-Whitney U tests and independent samples t-test were used, respectively, to compare single values across groups. For all comparisons between groups and conditions, effect sizes were expressed as Cohen’s d to simplify interpretation and comparisons.

Effect of age on Performance data & UCM variables

To test for an effect of age on task performance and motor flexibility we planned to run a two factor repeated-measures ANOVAs (between-subject factor of group: YA vs. OA, or YA vs NW depending on the initial analysis; within-subject factor of task: baseline vs. cued) on performance measures and UCM-derived variables. In the event of a significant effect of age on SSV, the latter would be considered as a covariate in the ANOVAs. Significant interactions would be explored with LSD corrected post-hoc comparisons. Mann-Whitney U tests and independent samples t-test were used to compare accuracy and SSV between groups, respectively.

The association of motor flexibility and functional tests

Subjects were coded as relying on motor flexibility if all of the following criteria were met: 1) V_TOT was higher during cued task; 2) a synergy was present in both conditions and 3) the synergy index was greater during the cued task (i.e., V_UCM increased more than V_ORT). Motor flexibility was operationalized in this manner to ensure that any change in the synergy index between tasks was the result of an increase, at least in part, in the stabilizing variance (V_UCM) that also resulted in an increase in V_TOT. Based on this specific definition subjects could be dichotomized as either utilizing or not utilizing flexibility. Before evaluating whether functional measures, (i.e., five functional tests and SSV) differed between groups that did and did not rely on motor flexibility, functional measures were checked for normality and extreme outliers were removed (1 value of F8W). We then planned to use independent samples t-tests or Mann-Whitney U-tests to compare functional measures between coded groups. Prior to comparing groups, the proportion of OB and NW who used motor flexibility to complete the cued task were compared with a Chi-squared test. In the event that BMI affected the use of motor flexibility (i.e., a group effect was observed between NW and OB) then BMI group would be considered as a covariate when comparing subjects who did and did not use motor flexibility.

To better understand the extent to which individual functional abilities affected strategies used, i.e. reliance on motor flexibility to complete cued walking, we planned to perform correlational analyses between functional measures and changes in V_UCM, V_ORT and V_TOT (δV_UCM, δV_ORT and δV_TOT). The latter set of variables was calculated as the difference in each measure between the cued and baseline task (positive values indicated cued> baseline). We used changes in UCM measures within the correlations as an indicator of motor flexibility given our operationalization of the construct (see Background). We planned to use Pearson’s correlations for normal data and Spearman’s correlations for non-normal data. Before conducting correlations, we compared δV_UCM, δV_QRT and δV_TOT between OB and NW; in the event of a significant group effect, BMI group would be considered as a covariate in establishing the relationship between functional measures and UCM change variables.

RESULTS

Effect of obesity on UCM-derived variables and performance measures

There were no significant differences between OB and NW with regard to accuracy or any of the UCM variables (Table 1). As such, data from the two groups were combined when assessing age effects. Although OB subjects walked at a significantly slower SSV and with significantly shorter (non-speed adjusted) and wider steps than NW during baseline walking, these differences persisted during cued walking; the group x task interaction was not significant for SW, SL nor for SW variability (SWV). However, after accounting for speed, NW significantly increased SL variability (SLV) during the cued task whereas OB decreased SLV.

Effects of age on cued walking performance

Whereas YA performed the cued task with ≥96% accuracy, OA performed significantly worse on the cued walking task, with one-third performing at <90% accuracy (p<0.01; Table 2). Regardless of age, participants walked with steps that were, on average, 4% more narrow during the cued task, although differences were not significant after accounting for speed (p=0.09 for task). SWV was significantly reduced during cued walking, but the effects were stronger in YA, who reduced SWV by 40%, compared to a 22% reduction for OA (p=0.04 for task x group interaction). Age did not alter the effects of task on SL or SLV, although the latter was significantly greater during cued vs baseline walking (p<0.01 for task).

Table 2.

The effects of age on performance during cued-stepping task.

	YA (N=10)	OA (N=46)	P – value; F-value (df)	d
Accuracy (%)	98.7 (2.8)	94.9 (9.6)	<0.01	0.91
SSV (m/s)	1.11±0.13	0.89±0.25	0.01	0.94
SW baseline (mm)	133.3±37.3 [108.1 ±21.3]	133.8±38.8 [137.5±46.6]	group x task: 0.43; 0.63 (1,53) task: 0.09; 2.95 (1,53) [0.04; 4.41 (1,54)]	0.28 0.61 [0.75]
SW cue (mm)	124.7±33.8 [108.8±11.5]	129.8±35.2 [128.4±43.2]
Log SWV baseline	3.25±0.35 [3.28±0.25]	3.31±0.33 [3.29±0.35]	group x task: 0.04*; 4,36 (1,53)	0.74
Log SWV cue	2.69±0.36 [2.74±0.33]	3.05±0.35 [3.03±0.35]
SL baseline (mm)	446.0±39.7 [519.6±45.3]	449.8±41.3 [432.4±96.1]	group x task: 0.14; 2.29 (1,53)	0.60
SL cue (mm	453.6±47.0 [514.6±51.3]	446.6±48.3 [440.0±99.3]	task: 0.80; 0.07 (1,53)	0.09
Log SLV baseline	3.21±0.39 [3.25±0.38]	3.49±0.37 [3.49±0.33]	group x task: 0.20; 1.65 (1,53) task: <0.01; 10.16 (1,53) [0.68; 0.167 (1,54)]	0.46 1.13 [0.15]
Log SLV cue (mm)	3.37±0.36 [3.28±0.33]	3.48±0.35 [3.48±0.39]

Descriptive data are presented as either mean± standard deviation for normally-distributed variables or median (interquartile range) for non-normally distributed variables. Descriptive data is shown for speed adjusted values with raw values in square brackets, except for accuracy and SSV. All p-values and Cohen’s d values are shown for adjusted data except when statistical findings differed between adjusted and unadjusted data (i.e. p-value <0.05 for only one), in which case p-values and Cohen’s d values for unadjusted data are shown in square brackets.

^*Both groups significantly reduce SWV (YA: p<0.001, Cohen’s d=1.54; OA: p=<0.001; Cohen’s d=1.58) but YA reduce to a greater extent;*OA and YA are not different during baseline (p=0.65, d=0.16) but are significantly different during cued walking (p<0.01, d=1.00)

Effects of age on UCM derived measures

Without adjusting for speed, OA but not YA walked with stronger synergies during cued compared to baseline walking (p=0.02, F(1,54)=5.57 for task x group interaction). However, a significant interaction was absent after adjusting for SSV (p=0.10, F(1,53)=2.87, Cohen’s d =0.60). In the speed adjusted data (Figure 2), there was a main effect of task (p<0.01, F(1,53)=9.37, Cohen’s d = 1.09). These effects were mirrored in V_TOT and V_UCM.

Figure 2.

The effects of age on the synergy index. After accounting for speed, there is a main effect of task on the synergy index, however OA and YA increase the synergy using different strategies, as reflected in changes in the individual variance components

OA but not YA walked with greater variance in elemental space (V_TOT) during cued walking (p=0.001 F(1,53)=13.63, d=1.02 for task x group interaction; Figure 3). Whereas V_TOT was not significantly different between OA and YA during baseline walking (p=0.63, d=0.13), OA significantly increased V_TOT during cued walking (p=0.01, d=0.74) and YA significantly reduced V_TOT (p<0.01, d=.80). Similarly, OA but not YA walked with greater Vucm during cued walking (p<0.001, F(1,53)=8.82, d=1.19 for task x group interaction; Figure 3). Whereas V_UCM was not significantly different between OA and YA during baseline walking (p=0.36; d=0.26), OA significantly increased V_UCM during cued walking (p=0.01, d=0.74) and YA significantly reduced V_UCM (p<0.01, d=0.90). Regardless of age, V_ORT was significantly lower during cued walking (p<0.001, F(1,53)=15.36, d=1.08 for main effect of group; Figure 3). The group x task interaction was not significant for V_ORT (p=0.07, d=0.54).

Figure 3.

The effects of age and walking condition on measures derived from the UCM analysis. There was a strong, significant interaction between group and walking task for the total variance in elemental variable space (top) as well as for good variance (middle) with OA significantly increasing both while YA significantly decreased both. Regardless of age, bad variance was significantly lower during cued walking compared to baseline walking (bottom)

Association between motor flexibility and function

Regardless of BMI group, ~60% of OA participants employed motor flexibility during the cued task. BMI group was not included as a factor when comparing functional measures between subjects who did and did not use motor flexibility (12/21 OB used motor flexibility vs. 15/25 NW; p=0.84, d=0.06). Similarly, neither δV_UCM, nor δV_ORT nor δV_TOT differed between BMI group (p=0.43, 0.54, 0.59, respectively; d=0.24, 0.18, 0.16, respectively). Therefore, BMI group was not included as a factor in correlational analyses.

Of the six functional measures considered, only 10mW was different between OA who did and did not employ motor flexibility (Table 3). There was a medium effect (0.5<d<0.7) of motor flexibility use on SLS performance as well. Despite the fact that, on average, 10mW was faster in those who employed motor flexibility, the inter-subject variance in 10mW was such that it was not significantly correlated with any UCM change measures (δV_UCM, δV_ORT and δV_TOT) (Table 4). In contrast, greater δV_UCM was associated with faster F8W times, SSV and longer SLS; the strengths of all associations were medium (0.3 ≥ r or p ≥ 0.5) (9). Similar associations were seen between these functional measures and δV_TOT.

Table 3.

Comparing functional measures bewteen older adults who do and do not rely on motor flexibility during cued walking task.

Functional measure	OA not using flexibility (N=19) OA using flexibility (N=27)	p	d
10mW (m/s)	1.77±0.30 1.96±0.36	0.03	0.65 (r=0.31a)
SLS (s)	18.48 (26.39) 30.00 (18.29)	0.06	0.57
F8W (s)	7.68 (1.77) 6.95 (2.71)	0.15	0.44
4SS (s)	8.87 (3.05) 8.15 (2.66)	0.21	0.37
SSV (m/s)	0.86±0.29 0.93±0.22	0.26	0.36
TUG (s)	8.11 (1.99) 7.72 (2.761	0.35	0.28

Descriptive data are presented as either mean± standard deviation for normally-distributed variables or median (interquartile range) for non-normally distributed variables

^abased on conversion of Cohesn’s d to Pearson r for comparission with Table 4

Table 4.

Correlations between functiona measures and measures of motor flexibiltiy

	SSV	TUG	10mW	F8W	4SS	SLS
δ VORT	0.26 (0.08)	−0.05 (0.76)	0.19 (0.21)	−0.17 (0.27)	−0.04 (0.81)	0.31 (0.04)
δ VUCM	0.31 (0.04)	−0.12 (0.43)	0.24 (0.11)	−0.31 (0.04)	−0.28 (0.06)	0.44 (<0.01)
δ VTOT	0.32 (0.03)	−0.12 (0.43)	0.25 (0.09)	−0.31 (0.04)	−0.27 (0.07)	0.45 (<0.01)

Data is presented as r-value (p-value) for normally-distributed data and as p-value (p-value) for non-normally distributed data.

DISCUSSION

The primary purpose of this study was to determine the effects of obesity and age on motor flexibility as it relates to foot placement during a locomotor task that specifically constrains foot placement. We hypothesized that: 1) normal weight, but not obese older adults would rely on motor flexibility during cued walking and that the extent to which they did so would relate to functional mobility; and 2) both younger and older adults would rely on motor flexibility during cued walking to stabilize the trajectory of the foot during the swing phase of the gait cycle supported. Both NW and OB relied on motor flexibility In addition, while older relied on motor flexibility for the cued stepping task, YA did not (figure 3). Previous studies have used the UCM method to analyze reliance on motor synergies to stabilize task performance (i.e., ΔV_Z) whereby an increase in ΔV_Z suggests an increase in motor synergies and exploiting of motor redundancy. In the current investigation we showed an effect of task on ΔV_Z regardless of age group (figure 2). However, when the individual components of the index were evaluated to consider reliance on motor flexibility, age effects emerged.

The effects of obesity in motor flexibility during cued walking

We expected an effect of obesity on motor flexibility in part due to added mass as well as impairments in other process that may contribute to kinematic synergies. In addition, while obesity is not generally associated with specific neuromuscular changes, it is known to impact brain structure and function, which is important as certain neurophysiological structures and processes may play a key role in the the emergence of kinematic synergies that contribute to motor flexibility (Latash et al. 2005; Martin et al. 2019; Reimann and Schoner 2017). for example, there is considerable evidence that obesity is associated with grey matter atrophy throughout the brain, including to regions of the brain involved in motor control (Hamer and Batty 2019; Herrmann et al. 2019; Raji et al. 2010). More broadly, obesity is reported to impair cognitive function, including executive functioning (Bischof and Park 2015), which, through the sharing of resources, may also impact sensorimotor integration (Li and Lindenberger 2002). When asked to oscillate a wrist pendulum in sync with a visual stimulus, obese adults had poorer intersegmental coordination and variability, suggesting impaired ability to synchronize movement with visual stimuli, i.e., an overall deficiency in visual sensory integration (Gaul et al. 2016). Such a deficiency would be expected to interfere with the visually-cued stepping task. To some extent, obesity-related changes to neural structures and function may be attributed to common comorbidities of obesity including metabolic syndrome and hypertension that promote inflammation and brain atrophy through, for example, inflammation and reduced vascularization (see review in (Raji et al. 2010)). However, in our cohort of OB, blood pressure was medically controlled and those with diabetes were, to some degree, controlling the disease as evidence by absence of peripheral vascular neuropathy, finally, the fact that OB and NW have similar changes in UCM components and synergies may reflect compensation for obesity-related instability (Madigan et al. 2014). Indeed, our OB had reduced functional mobility compared to NW. Moreover, obese older adults are at greater risk of falls then their normal weight peers (Madigan et al. 2014), and fallers have been reported to walk with greater synergies and V_UCM then non-fallers to compensate for instability (Yamagata et al. 2019).

The effects of age on motor flexibility during cued walking

Younger adults may rely more on motor flexibility during gait only when extrinsic factors present a specific threat to balance. For example, younger adults rely on motor flexibility when walking on uneven surfaces (Eckardt and Rosenblatt 2018) that necessitate step-to-step adjustments in response to unexpected initial conditions at toe off, or when walking on raised beams (Rosenblatt et al. 2014b) in which a penalty exists for inaccurate foot placement. In contrast, when foot placement is constrained by walking in a narrow (30.5 cm wide), non-elevated path (Rosenblatt et al. 2015) that is sufficiently wide to accommodate an average step, motor flexibility is not utilized – V_UCM does not change and V_ORT decreases. While both cued and narrow path walking place constraints on foot placement, during cued walking both SW and SL are specified and unvaried, which forces participants to reduce step-to step variability in these parameters. Indeed, SWV was significantly lower during cued compared to baseline walking (Table 2). Surprisingly SLV was greater during cued compared to baseline walking, which supports the proposition that SL is less important for functional mobility (Owings and Grabiner 2004). Although a strategy that decreased total variance and variance components was not hypothesized, in light of the task requirements this strategy seems logical for YA. Moreover, increasing good variance, and reliance on motor flexibility, is thought to be appropriate for accommodating unexpected conditions (as in beam or uneven surface walking but not cued walking) and/or effectively operating under varying contexts (Freitas et al. 2010; Poston et al. 2008; Stergiou et al. 2006), which are absent in the invariant cued stepping task .

Older adults relied on a different strategy than YA to accomplish cued walking, which may reflect disproportionately larger physical demands of the task for OA versus YA. Older adults performed cued walking with significantly lower accuracy in hitting step cues compared to YA, suggesting poorer task performance (Table 2). Additionally, most OA required verbal feedback to avoid falling off the back of the treadmill during cued walking and several infrequently grasped the handrail. In a prior study of reaching to targets with the upper extremities, OA utilized motor flexibility to respond to increasing physical demands (resistive weight)(Greve et al. 2017). Similar strategies were observed in response to increasing demands of walking, i.e. walking on uneven surfaces (Eckardt and Rosenblatt 2018). In these studies, increasing task demand led to increasing destabilizing variance (V_ORT), which was countered with an even greater increase in stabilizing variance (V_UCM). However, in the current study, we observed an increase in V_UCM but a decrease in V_ORT., which could reflect task differences. For example, cued walking does not present a direct threat to balance, as does uneven surface walking, nor does cued walking directly increase physical (e.g., muscular) demands as does weighting the arm during reaching. Additionally, the cued stepping task took place on a treadmill where sensory information, e.g. optic flow (Durgin et al. 2005) and perception of self-movement (Yabe and Taga 2008), differs from walking overground. Finally ,the cued walking task includes a specific visual component which may heighten the attentional/cognitive demand of walking (Peper et al. 2012; Yamada et al. 2011), making the task disproportionately more challenging for OA compared to YA (Berg and Murdock 2011). Additional work is needed to understand how different sensory information and cognitive loads impact measures of motor flexibility within and between age groups.

OA may rely on motor flexibility not only due to greater task demands but also as an adaption to aging. The aging motor system has greater intrinsic system noise (Christou 2011; Goble et al. 2009; Kang and Dingwell 2009), which may prevent OA from systematically reducing variance to the same extent as YA. This would explain why OA were unable to reduce SWV as much as YA during cued walking (Table 2). The fact that OA were less accurate at cued walking further supports the presence of greater motor noise. (Christou et al. 2003; Wolpert and Ghahramani 2000). Alternatively, reduced accuracy may reflect a “posture first” strategy whereby upright posture (avoiding a fall) was prioritized over targeting cues (Shumway-Cook et al. 1997).

The association between motor flexibility and functional mobility

The ability to employ motor flexibility may facilitate mobility and thus be associated with higher performance on functional tests, particularly those that present greater challenge to posture and balance. For example, Hsu et al (2013) reported that faster Timed Up and Go was associated with a greater index of flexibility used to stabilize the center of mass following postural perturbations (Hsu et al. 2013). However, to our knowledge, this is the only prior study to have demonstrated a relationship between motor flexibility and mobility in OA.

Given that the kinematic synergies considered here were evaluated during walking, we expected F8W, a functional measure that directly challenges stability during gait, to be most strongly related to UCM change measures. While δV_UCM and δV_TOT were significantly associated with F8W, we were surprised to find that the relationship was just as strong for both self-selected and maximum walking speed (SSV and 10mW in Table 4). Individuals with faster comfortable and maximal walking speeds may require a more flexible motor system to contend with the instability associated with faster speeds (McCrum et al. 2019). Similarly, individuals with less flexible motor systems may be limited in their ability to contend with unstable conditions and are thus forced to self-select slower speeds. Significant associations may also reflect the overall utility of walking speed as an indicator of physical functioning and health. The ability to walk faster in older age predicts functional independence (Suzuki et al. 2003),

We were surprised to find that the strongest associations between UCM change measures during gait and SLS, a test of static stability that does not involve gait (Table 4). Performance of SLS involves two distinct phases - an initial weight shift with compensatory adjustments to ensure postural stability followed by a quasi-static phase, which, to some extent, is analogous to transitioning from double-to-single support. Increasing SLS may necessitate minimizing variability (noise) in muscle force production (Oshita and Yano 2010) and, relatedly, in the ground reaction forces (Jonsson et al. 2004). It is not unreasonable to assume that this would be achieved through synergies that manifest at the kinematic level (Park et al. 2016). Given the similarities between SLS and requirements of the stance limb during early swing, it would be pragmatic to rely on similar neural control signals for both, which may explain the observed associations between SLS and UCM change measures.

If control strategies that facilitate flexibility during gait also facilitate performance of functional tasks, then we would expect lower functioning older adults to be less capable of employing motor flexibility during challenging tasks (Table 3). However, this notion was not fully supported. Even though OB performed significantly worse on all functional tests (Figure 1) there was no effect of BMI group on UCM-derived measures (Table 1) or UCM change measures. On the other hand, the effects of obesity on function, while significant, may not have been meaningful enough to effect UCM measures. For example, the median TUG of 9.6 s for OB is far below the cut-off of 13.5 s that is associated with increased fall risk (Shumway-Cook et al. 2000) and the average 4SS times for OB tend to fall within the range reported for healthy, non-fallers (Dite and Temple 2002). Thus, obesity in absence of large functional declines may not have an effect on motor flexibility. Future work including groups with known balance deficits, including fallers versus non fallers, may provide better insight into the extent to which motor flexibility and function are related. If greater motor flexibility enables functional mobility then training paradigms that manipulate stability to facilitate exploration of motor solution and promote large amounts of good variance may be beneficial for increasing function and possibly preventing falls. Indeed, resistance training on unstable surfaces has been shown to promote motor flexibility (increase ΔV_Z and V_UCM related to the same synergies considered in the current study) and to improve functional measures related to proactive balance (Eckardt and Rosenblatt 2019) in heathy OA.

Limitations

Several study limitations are worth noting. First, the cued walking task was a novel task with which participants had no prior experience. The extent to which results reflect the novelty of the task versus the demands of the task itself are not known. Second, because only ~20 steps per participant and condition were evaluated, the values reported here for the UCM-derived measures may not reflect the true values that would be obtained with additional steps (Rosenblatt and Hurt 2019). However, the results from the repeated-measures designs would not be expected to change with additional steps. Third, we acknowledge that multiple statistical tests were conducted, particularly given the independent tests for obesity and age effects and correlations with functional measures. While this may increase the chances of type 1 error, the primary analyses involved only 4 ANOVAs and the analyses allowed us to evaluate multiple effects. Lastly, whereas the cued walking task places constrains on SW and SL, we only evaluated kinematic synergies and motor flexibility related to stabilization of the ML swing limb foot trajectory. However, the elemental variables considered here may simultaneously contribute to the ML and anterior-posterior swing limb foot trajectory. Similarly, the fact that younger adults did not employ motor flexibility to stabilize the swing limb foot does not imply the complete absence of motor flexibility, e.g., motor flexibility may have been exploited to stabilize another performance variable such as the center of mass, or other forms of flexibility such as joint work redistribution may be relied upon.

CONCLUSION

Strategies to perform a novel task that constrained foot placement during gait were significantly different between younger and older adults, but not between obese and normal weight older adults despite significant (though perhaps not clinically meaningful) differences in functional abilities. Our OB cohort may have been minimally impacted by obesity-related changes in cognition and neural structure and function that contribute to kinematic synergies or may have been compensating for instability by increasing reliance on motor flexibility. Whereas younger adults decreased total variance in segment angle space, older adults increased total variance and did so by increasing good variance more so than bad variance (which actually decreased). It is possible that older adults were compensating for age-related neuromuscular changes that increased the physical and/or cognitive demands of the novel cued waking task. In the older adult group, the use of motor flexibility was significantly associated with certain functional measures of mobility. Taken together, this study provides additional evidence that older adults retain the ability to employ motor flexibility during tasks that place constraints on motor solutions and supports the idea that motor flexibility in older adults may be a biomarker of healthy aging. If this idea is confirmed then exercises fostering exploration of motor solutions during complex tasks should be introduced to improve existing (re)habilitation programs.