Repeated adaptation and de-adaptation to the pelvis resistance force facilitate retention of motor learning in stroke survivors

Abstract

Locomotor adaptation to novel walking patterns induced by external perturbation has been tested to enhance motor learning for improving gait parameters in individuals poststroke. However, little is known regarding whether repeated adaptation and de-adaptation to the externally perturbed walking pattern may facilitate or degrade the retention of locomotor learning. In this study, we examined whether the intermittent adaptation to novel walking patterns elicited by external perturbation induces greater retention of the adapted locomotion in stroke survivors, compared with effects of the continuous adaptation. Fifteen individuals poststroke participated in two experimental conditions consisting of 1) treadmill walking with intermittent (i.e., interspersed 2 intervals of no perturbation) or continuous (no interval) adaptation to externally perturbed walking patterns and 2) overground walking before, immediately, and 10 min after treadmill walking. During the treadmill walking, we applied a laterally pulling force to the pelvis toward the nonparetic side during the stance phase of the paretic leg to disturb weight shifts toward the paretic side. Participants showed improved weight shift toward the paretic side and enhanced muscle activation of hip abductor/adductors immediately after the removal of the pelvis perturbation for both intermittent and continuous conditions (P < 0.05) and showed longer retention of the improved weight shift and enhanced muscle activation for the intermittent condition, which transferred from treadmill to overground walking (P < 0.05). In conclusion, repeated motor adaptation and de-adaptation to the pelvis resistance force during walking may promote the retention of error-based motor learning for improving weight shift toward the paretic side in individuals poststroke.

NEW & NOTEWORTHY We examined whether the intermittent versus the continuous adaptation to external perturbation induces greater retention of the adapted locomotion in stroke survivors. We found that participants showed longer retention of the improved weight shift and enhanced muscle activation for the intermittent versus the continuous conditions, suggesting that repeated motor adaptation and de-adaptation to the pelvis perturbation may promote the retention of error-based motor learning for improving weight shift toward the paretic side in individuals poststroke.

INTRODUCTION

Many stroke survivors suffering from impaired motor control and/or reduced strength of the paretic leg typically rely less on the paretic leg and more on the nonparetic leg during walking (1, 2). As a consequence, stroke survivors often demonstrate difficulty in bearing weight on the paretic side (3–5). Reduced weight bearing on the paretic side has been associated with functional gait deficits (6, 7). Thus, restoring the capacity to shift weight toward the paretic side may be an important goal for rehabilitation poststroke (8).

Forceful weight shift toward the paretic side through the application of pelvis assistance force may enhance muscle activity of the paretic leg (9). Our previous studies also indicated that the application of gradually increased or varied pelvis assistance force may facilitate weight shift toward the paretic side (10, 11). Furthermore, the application of the pelvis resistance force to enhance errors may also induce improved weight shift toward the paretic side, through error-based motor learning mechanisms (12, 13). In line with this, a previous study also indicated that the application of a resistance force to the paretic leg may increase gait symmetry (14). However, the error-based motor learning is generally short-lived. It remains unclear which approaches are more effective to increase retention of the error-based motor learning in individuals poststroke.

Introducing intermittent intervals in-between repetitions of locomotor adaptation may improve the retention of aftereffects in individuals poststroke. One of the fundamental principles in motor learning is that the amount of practice determines the degree of improvement in motor performance (15). Though repetitions of the same movement may be an effective way to improve motor performance within a session, it may not be the optimal way to retain the learning effect over a longer period of time (16). In the literature, intermittent repetitions of practice have been considered as an important factor in motor learning particularly in healthy individuals (17–19). Specifically, retention of adapted motor skills can be better facilitated by intermittent versus continuous practice of a motor task (18). Both intermittent and continuous practice paradigms have been applied to locomotor training for people with stroke to improve their fitness, cardiorespiratory and/or vascular functions, and aerobic capacity (20–23). The resultant outcomes from the intermittent practice paradigms have been better than those derived from the continuous paradigm. However, little is known about whether intermittent versus continuous adaptation to an external perturbation force may enhance the retention of error-based motor learning, i.e., the retention of aftereffects, in individuals poststroke.

The purpose of this study, therefore, was to determine whether intermittent versus continuous adaptation to the pelvis resistance force would facilitate retention of motor learning in stroke survivors. We hypothesized that intermittent adaptation to the pelvic resistance would result in better retention of its aftereffects (i.e., increased weight shifts toward the paretic side), compared with the result of the continuous adaptation.

MATERIALS AND METHODS

Participants

We recruited 15 individuals (58.5 ± 6.0 yr, 6 females) with chronic stroke (>6 mo) to participate in this study from the outpatient clinics of the Shirley Ryan AbilityLab or from its stroke patient registry (see Table 1). The inclusion criteria were as follows: 1) 21–75 yr old; 2) a single unilateral, supratentorial stroke of either ischemic or hemorrhagic etiology, confirmed by radiographic imaging; 3) weakness/paresis of the affected leg; and 4) ability to stand and walk (>10 m) without external physical assistance and with the use of assistive devices or ankle-foot orthoses, if needed. Exclusion criteria were as follows: 1) brainstem or cerebellar stroke; 2) a score of <24 on the Mini Mental State Examination (24); 3) other neurological conditions, cardiorespiratory/metabolic disorders, or orthopedic conditions affecting ambulatory ability; 4) uncontrolled hypertension (systolic >200 mmHg, diastolic >110 mmHg during sitting); 5) botulinum toxin injection within the prior 6 mo; 6) inability to tolerate 30 min of treadmill walking with sitting breaks as necessary. The Northwestern University Medical School Institutional Review Board approved all procedures, and all participants signed written informed consent before data collection.

Table 1.

Demographic information for the participants

P	Sex	Age, yr	Weight, kg	Height, cm	Postinjury, yr	Paretic Side	Brace	Assistive Device	Self-Selected Comfortable Speed, m/s	Pelvis Resistance Force, N
1	M	67	71	173	5	L	AFO	None	0.50	49
2	F	57	61	168	7	R	AFO	SPC	0.60	36
3	F	60	94	163	1	R	None	None	0.55	65
4	M	63	102	185	17	R	None	SPC	0.36	50
5	F	48	102	157	4	L	None	None	0.67	80
6	M	60	129	191	5	R	None	WBQC	0.54	101
7	F	65	70	163	7	L	AFO	SPC	0.82	55
8	F	60	73	163	10	L	AFO	SPC	0.26	36
9	M	50	111	168	12	L	AFO	SPC	0.87	54
10	M	53	88	170	20	R	AFO	None	0.60	60
11	F	61	97	165	12	R	None	None	0.56	47
12	M	56	92	178	5	R	AFO	None	0.46	54
13	M	61	112	175	10	L	None	SPC	0.36	55
14	M	50	98	170	10	L	AFO	None	0.52	48
15	M	66	75	180	12	L	AFO	None	0.70	59

AFO, ankle foot orthosis; F, female; L, left; M, male; m/s, meter per second; N, newton; P, participant; R, right; SPC, single point cane; WBQC, wide based quad cane; yr, year.

Apparatus

We used a custom-designed cable-driven robotic system, which was mounted over a treadmill, to apply lateral pulling force at the pelvis toward the nonparetic side during treadmill walking (Fig. 1, A and B). Specifically, the robotic system was composed of two nylon-coated stainless-steel cables driven by two motors (AKM 33H, Kollmorgen, Radford, VA) through cable spools located at both sides of the treadmill (25). In this study, we used one set of motor and a cable pulley system on one side, depending on which side of each participant was affected, to apply a controlled pulling force to the pelvis, which was resistant to the mediolateral pelvis movement. Specifically, we attached one cable to the participant’s pelvis through a waist belt and used a pulley system to deliver resistance force to the pelvis toward the nonparetic side. Furthermore, a set of customized three-dimensional position sensors was attached to participants’ ankles and pelvis to collect position signals of the pelvis and ankles (Fig. 1A) (26). The robotic system was controlled by a personal computer through a custom-written program in LabVIEW (National Instrument, Austin, TX). The controller used ankle position signals to trigger the resistance force at a targeted phase of gait, i.e., early-to-mid stance phase of the paretic leg (Fig. 1B).

Figure 1.

Experimental setup and protocol. A: participants walked on a treadmill with laterally pulling force toward the nonparetic side (red arrow; pelvis resistance force), applied by a cable-driven robotic system. While walking on the treadmill, muscle activation from the paretic leg and position signals were recorded. B: the timing of applying laterally pulling force toward the nonparetic side (NP; red arrow) was the early-to-mid stance phase of the paretic leg. C: during a single visit, all participants completed two sessions (i.e., intermittent vs. continuous condition), which were randomly ordered (n = 6, intermittent → continuous; n = 9, continuous → intermittent). D: each session contained the treadmill walking task that consisted of the 1-min baseline, 8-min adaptation, 2-min postadaptation, 2-min standing break, and 5-min readaptation period. Specifically, the treadmill walking task for the intermittent condition consisted of: 1-min baseline treadmill walking; 2-min treadmill walking with lateral resistance force (adaptation 1); 1-min treadmill walking (de-adaptation 1); 2-min treadmill walking with lateral resistance force (adaptation 2); 1-min treadmill walking (de-adaptation 2); 2-min treadmill walking with lateral resistance force (adaptation 3); 2-min treadmill walking (postadaptation); 2-min standing break; 2-min treadmill walking with lateral resistance force; 1-min treadmill walking; 2-min treadmill walking with lateral resistance force. The treadmill-walking task for the continuous condition consisted of: 1-min baseline treadmill walking; 8-min treadmill walking with lateral resistance force (adaptation); 2-min treadmill walking (postadaptation); 2-min standing break; 5-min treadmill walking with lateral resistance force (readaptation). Also, participants performed the overground walking task before treadmill walking, immediately after treadmill walking, and 10 min after treadmill walking. EMG, electromyography.

Experimental Protocol

All participants were tested under two conditions (i.e., intermittent vs. continuous) within a single visit. Each condition was conducted in a separate experimental session, and the order of the two conditions was randomized across participants (Fig. 1C; n = 6, from the intermittent to the continuous condition; n = 9, from the continuous to the intermittent conditions). A lateral pulling force was applied to the pelvis toward the nonparetic side during the time period when the paretic leg was in contact with the treadmill belt (i.e., starting from ∼100 ms following the initial contact of the paretic leg, for 400 ms). We selected this time period because this was the time when the nonparetic leg was primarily in the swing phase, and therefore the pulling force could induce participants to forcibly use their paretic leg (27). The magnitude of force was set at 8% of body weight, which was determined based on previous studies (9, 28) and was adjusted depending on each participant’s tolerance.

For the continuous condition, each session included the following procedures (Fig. 1D): 1) baseline overground walking; 2) 1 min of treadmill walking with no resistance (baseline); 3) 8 min of treadmill walking with resistance (adaptation); 4) 2 min of treadmill walking with no resistance (postadaptation); 5) 2-min standing break; 6) 5 min of treadmill walking with resistance (readaptation), which was used to test the transfer of motor adaptation from treadmill to overground walking; 7) overground walking immediately after treadmill walking; 8) 10-min seated break; 9) overground walking. For the intermittent condition, a protocol that was comparable with the continuous condition was used, but two 1-min de-adaptation walking periods were interspersed in-between during the adaptation period, and one 1-min de-adaptation walking period was included in the middle of the readaptation period, see Fig. 1D.

The treadmill-walking speed was set at a self-selected comfortable speed for each participant, which was determined at the beginning of the testing session. This selection remained the same for all testing sessions. An overhead harness was used for the purpose of protection only (no body weight support provided). All participants were allowed to hold onto the front handrail using their hands during treadmill walking for safety. Before data collection, participants walked on the treadmill at their maximum speed for 30 strides and we used electromyography (EMG) activity during this session to normalize the integrated EMGs during the other tasks. Participants walked overground on the GAITRite mat (CIR Systems, Inc., Franklin, NJ) at their normal comfortable speed for three trials per task in each condition. We transported participants from the treadmill to the GAITRite mat using a wheelchair to reduce the potential washout of motor skills during the transition period.

Data Collection

We recorded muscle activity of tibialis anterior (TA), medial gastrocnemius (MG), soleus (SOL), rectus femoris (RF), vastus medialis (VM), medial hamstring (MH), hip abductor (ABD; gluteus medius), and hip adductor (ADD; adductor magnus) in the paretic leg using surface EMG electrodes (BagnoliTM, Delsys, Boston, MA). EMG signals were amplified (1,000 times) and bandpass filtered (20–450 Hz) using Bagnoli-16 Amplifier (Delsys, MA), and then sampled with an A/D board (National Instruments, Austin, TX) at 500 Hz using a custom-written program in LabVIEW. Ankle and pelvis positions were sampled at 500 Hz with customized three-dimensional position sensors (Fig. 1A) using the customized program in LabVIEW. Gait speed and other spatiotemporal gait parameters during overground walking were calculated using the software from the GAITRite.

Data Analysis

All kinematic and EMG data were analyzed using custom-written programs in MATLAB (MathWorks, MA). Kinematic data were lowpass filtered using fourth-order Butterworth filter with cut-off frequency at 10 Hz. Due to step-to-step variation of foot position in the mediolateral direction during treadmill walking, we defined the midpoint of two ankle positions at the initial contact as a reference point for each gait cycle. To calculate pelvis mediolateral displacement of each gait cycle, we subtracted the midpoint between two ankle positions at the initial contact of each leg from the midpoint between paretic and nonparetic pelvis positions, which was used to estimate the center of mass (i.e., center of mass; CoM) (29). Then, we quantified the weight shift toward the paretic side using the peak of pelvis lateral displacement during stance phase of the paretic leg (9).

EMG data were high-pass filtered at 10 Hz, band-stop filtered from 59 to 61 and 119 to 121 Hz (to remove electrical power-line noise), rectified, and smoothed using a lowpass filter at 20 Hz (fourth-order Butterworth) (30). The EMG data were segmented into different gait cycles based on ankle position data and were normalized to peak values of each muscle activity while participants walked at their maximal walking speed. Then, the integrals of muscle activities during stance phase of gait were calculated.

Weight shift and EMG integral during the following six subintervals of treadmill walking were calculated (Fig. 2). The subintervals were: the last 30 steps during baseline (B); the first 20 steps (early adaptation, EA) and the last 20 steps (late adaptation, LA) during the adaptation period; the first five steps (early postadaptation, EPA), the middle five steps (middle postadaptation, MPA), and the last five steps (late postadaptation, LPA) during the postadaptation period. Also, in the intermittent condition, we included four additional subintervals (Fig. 2A): the first five steps (early de-adaptation 1, EDA1) and the last five steps (late de-adaptation 1, LDA1) during the first de-adaptation period; and the first five steps (EDA2) and the last five steps (LDA2) during the second de-adaptation period.

Figure 2.

Stride-by-stride lateral weight shift toward the paretic side from one representative participant for the intermittent (A) and continuous (B) conditions. The following steps were selected (blue dots) for the statistical analysis of the two testing conditions: last 30 steps during the baseline period (B); first and last 20 steps of the adaptation period (EA and LA); first, middle, and last 5 steps of the postadaptation period (EPA, MPA, and LPA). First and last 5 steps of each de-adaptation period (EDA1–2 and LDA1–2) were additionally selected for only the intermittent condition. B, baseline; EA, early adaptation; EDA, early de-adaptation; EPA, early post-adaptation; LA, late adaptation; LDA, late de-adaptation; LPA, late post-adaptation; MPA, middle post-adaptation.

We also examined whether the retention of aftereffects during the postadaptation period differed for the intermittent and continuous conditions. We assumed that the aftereffect would be washed out if the outcome measure re-entered the 95% confidence interval (CI) of the baseline value for three consecutive steps (31). The CI was computed based on the participant’s performance (i.e., weight shift in this case) during the last 30 steps of baseline period. Thus, to quantify the retention of aftereffect during the postadaptation period, we counted the number of steps starting from load release until the outcome measure (i.e., weight shift) re-entered the CI of the baseline period for three consecutive steps.

Statistical Analysis

A one-way ANOVA with repeated measures was used to compare 1) weight shift, 2) integrated EMG, 3) overground walking speed, 4) step length and its asymmetry during overground walking, and 5) cadence during overground walking at different time periods (4 time points for weight shift and integrated EMG; 3 time points for overground walking) within each experimental condition. If a significant main effect was identified, post hoc tests with the Bonferroni adjustment were further conducted to identify specific differences. In addition, a two-way ANOVA with repeated measures was used to identify whether there was an interaction between condition (intermittent and continuous conditions) and time point (i.e., 4 time points: B, EPA, MPA, and LPA) on weight shift.

Dependent t tests were used to compare 1) differences in weight shift, integrated EMG, overground walking speed, overground step length, overground step length asymmetry, and overground cadence between the two conditions, and 2) weight shift between the early and the late moment of each of the de-adaptation and postadaptation periods.

To compute the number of steps with improved weight shift toward the paretic side while the force was applied (intermittent: adaptation + de-adaptation periods; continuous: adaptation period), we counted the number of steps that the peak pelvis displacement exceeded at least 1 standard deviation above the mean peak pelvis displacement during baseline period for each condition. A dependent t test was used to compare the number of steps with improved weight shift toward the paretic side between the two testing conditions. A linear regression model was used to determine the association between the number of steps with improved weight shift toward the paretic side and the change in weight shift toward the paretic side from baseline to the late postadaptation period.

To assess participant’s motor performance during the postadaptation period, we selected the time point EPA (early postadaptation) to examine the immediate aftereffect of the motor adaptation induced by the resistance force (i.e., intermittent or continuous) on the weight shift and integrated EMG, and chose the time points MPA (middle postadaptation) and LPA (late postadaptation) to examine the retention of the aftereffect (Fig. 2). In addition, for the intermittent condition, we also selected the time points EDA (early de-adaptation) 1 and LDA (late de-adaptation) 1 during the first de-adaptation period and the time points EDA 2 and LDA 2 during the second de-adaptation period (see Fig. 2A). These time points were used to assess the changes in weight shift and EMG integrals during this short period of time (i.e., 1 min) when resistance force was removed.

The IBM SPSS Statistics 20.0 statistical package (IBM Corp., Armonk, NY) was used for analyses. The α level for all statistical tests was set at 0.05. Data are reported as means ± standard deviation in the text and means ± standard error in the figures.

RESULTS

Weight Shift toward the Paretic Side during Treadmill Walking

Weight shifts toward the paretic side during treadmill walking from one typical participant for the intermittent and continuous conditions are shown in Fig. 2. For the intermittent condition, the application of lateral pulling force toward the nonparetic side induced an abrupt deviation in weight shift toward the nonparetic side during the early adaptation period (EA) but it gradually returned to a level that was comparable with baseline during the remainder of the first adaptation period (adaptation 1). Then, the participant showed an increase in weight shift toward the paretic side following the force release during the early de-adaptation period (EDA 1) of the first de-adaptation period (de-adaptation 1), and it returned to a level that was comparable with baseline again at the end of this period (LDA 1; late de-adaptation). The participant also showed an increase in weight shift toward the nonparetic side with the application of the resistance force during the early adaptation 2 period, and it returned to a level that was comparable with baseline during the late adaptation 2 period. Following the force release, the participant demonstrated an aftereffect consisting of further increased weight shift toward the paretic side during the early post adaptation 2 period, with gradual decay, although it was partially retained during the late post adaptation 2 period. During the adaptation 3 period, the participant showed a motor adaptation pattern that was comparable with that during the adaptation 2 period. Following the release of force, the participant showed an aftereffect consisting of improved weight shift toward the paretic side during early postadaptation period (i.e., EPA). The improved weight shift toward the paretic side was partially retained during the late postadaptation period (LPA), and the level of the retention was greater than that during the LDA1 and LDA2. On the other hand, for the continuous condition, the application of lateral pulling force toward the nonparetic side induced a sudden deviation in weight shift toward the nonparetic side during the early adaptation period, but afterward it gradually returned to a level that was comparable with baseline, and then this level was retained during the remainder of the adaptation period. When the force was released during the postadaptation period, the participant showed an increase in weight shift toward the paretic side (i.e., an aftereffect) and it gradually returned to a level that was comparable with baseline during the early-to-mid postadaptation period, which was maintained during the rest of the postadaptation period.

The group average of weight shifts toward the paretic side (i.e., peak of lateral pelvis displacement during stance phase of the paretic leg) is shown in Fig. 3. The group average results showed that participants showed an aftereffect, i.e., increased weight shift toward the paretic side during the early postadaptation period, for both conditions (Fig. 3A). The size of aftereffect was comparable across two conditions, but the retention of aftereffect was longer for the intermittent condition than that for the continuous condition. Specifically, weight shift toward the paretic side during treadmill walking significantly differed between time points (i.e., B, EPA, MPA, and LPA) for the intermittent (F = 10.43, P < 0.0001) and continuous (F = 8.14, P < 0.001) conditions (Fig. 3B). For the intermittent condition, post hoc analysis using the Bonferroni correction (6 comparisons) showed significant increase in weight shift toward the paretic side from baseline to that during the early, the mid, and the late postadaptation period (P < 0.03). For the continuous condition, post hoc analysis also revealed significant increase in weight shift toward the paretic side from baseline to that during the early post adaptation period (P = 0.02), but the increased weight shift decreased from that during the early postadaptation period to that during the mid- and the late postadaptation periods (P < 0.02). The aftereffect (i.e., increased weight shift toward the paretic side) was retained significantly longer for the intermittent condition than for the continuous condition (t = 2.4, P = 0.032, Fig. 3C).

Figure 3.

Lateral weight shift toward the paretic side during treadmill walking. A: group average of lateral weight shift toward the paretic side. Dashed lines indicate the baseline average. B: within-condition comparisons of lateral weight shift. C: between-condition comparison of retention of aftereffect. B, baseline; CoM, center of mass; EPA, early post-adaptation; LPA, late post-adaptation; MPA, middle postadaptation. *Significant difference.

Weight shift toward the paretic side during treadmill walking significantly differed between the time points of baseline (B), early de-adaptation 1 (EDA1), early de-adaptation 2 (EDA2), and early postadaptation (EPA) (F = 8.73, P < 0.001, ANOVA), and between the time points of baseline (B), late de-adaptation 1 (LDA1), late de-adaptation 2 (LDA2), and late postadaptation (LPA) (F = 9.56, P < 0.001) for the intermittent condition (Fig. 4A). Post hoc analysis with the Bonferroni correction (6 comparisons) showed significant increases in weight shift from B to EDA2 (P = 0. 008) and from B to EPA (P = 0. 006), from LDA1 to LDA2 (P = 0.01), and from LDA1 to LPA (P = 0.03). In addition, weight shift toward the paretic side significantly decreased from EDA1 to LDA1 (t = 2.46, P = 0.03, Fig. 4B, t test), but showed no significant difference between EDA2 and LDA2 (t = 1.63, P = 0.13) and between EPA and LPA (t = 1.51, P = 0.16).

Figure 4.

A: weight shift toward the paretic side became greater during the early moment of the second de-adaptation (EDA2) and postadaptation (EPA) periods, compared with the baseline level (B). Also, the weight shift toward the paretic side was still greater until the late moment of the second de-adaptation (LDA2) and postadaptation (LPA) periods, compared with the baseline level and the level of the late moment of the first de-adaptation period (LDA1). B: effect immediately after the load release during the first de-adaptation period (EDA1) appeared to be reduced at the end of this period (LDA1). However, the aftereffects during the second de-adaptation (EDA2) and postadaptation (EPA) periods remained similar until the end of each period (LDA2/LPA). C: participants exhibited the greater number of steps with weight shift toward the paretic side exceeding the baseline level for the intermittent condition than for the continuous condition. D: for the intermittent condition, participants who had the greater number of steps with weight shift toward the paretic side exceeding the baseline level showed greater improvements in weight shift toward the paretic side from baseline to late postadaptation period. CoM, center of mass; EDA, early de-adaptation; EPA, early postadaptation; LDA, late de-adaptation; LPA, late postadaptation. *Significant difference. #Significant level between 0.5 and 0.7.

In addition, a two-way ANOVA with repeated measures revealed that the condition (intermittent vs. continuous) × time (B, EPA, MPA, and LPA) interaction was significant for weight shift toward the paretic side (F = 3.77, P < 0.02). Post hoc tests did not show a significant difference between the two testing conditions at any time points (P > 0.05).

The number of steps with improved weight shift toward the paretic side, i.e., steps with peak pelvis displacement exceeding 1 standard deviation above the baseline mean peak pelvis displacement, during the adaptation period (the intermittent condition included adaptation + de-adaptation periods) was greater for the intermittent condition than for the continuous condition (t = 2.57, P = 0.02; Fig. 4C). Furthermore, for the intermittent condition, participants who had the greater number of steps with improved weight shift toward the paretic side demonstrated greater improvement in weight shift from baseline to the late postadaptation period (R² = 0.52, Durbin–Watson = 2.37, P = 0.004; Fig. 4D).

Neural Activation of the Paretic-Leg Muscles during Treadmill Walking

A one-way ANOVA with repeated measures revealed significant differences, across different time points (i.e., B, EPA, MPA, and LPA), in muscle activities of TA (F = 5.45, P = 0.003), SOL (F = 10.84, P < 0.001), VM (F = 3.88, P = 0.02), RF (F = 3.56, P = 0.02), ADD (F = 9.71, P < 0.001), and ABD (F = 9.60, P < 0.001) for the intermittent condition, and in muscle activities of TA (F = 4.54, P = 0.01), SOL (F = 7.28, P < 0.001), ADD (F = 9.24, P < 0.001), and ABD (F = 9.15, P < 0.001) for the continuous condition, Fig. 5. Post hoc analysis using the Bonferroni correction (6 comparisons) revealed the followings: 1) SOL EMG increased from baseline to the early, the mid, and the late postadaptation periods (P < 0.05; Fig. 5C, middle) in both conditions; 2) ADD EMG increased from baseline to the early, the mid, and the late postadaptation periods (P < 0.05) for the intermittent condition, and increased from baseline to the early and the mid postadaptation periods, but it decreased from EPA to MPA and to LPA period (P < 0.05; Fig. 5G, middle) for the continuous condition; 3) ABD EMG increased from baseline to EPA, MPA, and LPA periods, and from MPA to LPA periods (P < 0.05) for the intermittent condition, and increased from baseline to EPA, but it decreased from EPA to MPA and to LPA periods (P < 0.02; Fig. 5H, middle) for the continuous condition.

Figure 5.

Muscle activation of the paretic leg during the treadmill walking for the intermittent and continuous conditions. A: TA EMG. B: MG EMG. C: SOL EMG. D: VM EMG. E: RF EMG. F: MH EMG. G: ADD EMG. H: ABD EMG. ABD, hip abductor; ADD, hip adductor; B, baseline; EA, early adaptation; EPA, early postadaptation; LA, late adaptation; LPA, late postadaptation; MG, medial gastrocnemius; MH, medial hamstring; MPA, middle post-adaptation; RF, rectus femoris; SOL, soleus; TA, tibialis anterior; VM, vastus medialis. *Significant difference. #Significant level of 0.07.

Change in ABD muscle activity from baseline to LPA was significantly greater for the intermittent condition than for the continuous condition (t = 2.85, P = 0.01; Fig. 5H, right). Changes in ADD muscle activation from baseline to LPA was not significant (t = 1.97, P = 0.069; Fig. 5G, right) between the two conditions. Changes in other muscle activities from baseline to EPA, MPA, and LPA periods were not significant between the two conditions (P > 0.07).

Overground Walking Performance after Treadmill Walking

Overground walking speed was significantly different across time points (i.e., B, post, and 10-min post) for the intermittent condition (F = 5.25, P = 0.01; ANOVA, Fig. 6B) but this was not the case for the continuous condition (F = 0.85, P = 0.44). Post hoc analysis with the Bonferroni correction (3 comparisons conducted) showed that overground walking speed significantly increased from baseline to the time point 10-min after treadmill walking (P = 0.04), and from the time point immediately after treadmill walking to the time point 10-min after treadmill walking (P = 0.03) for the intermittent condition.

Figure 6.

Overground walking speed. A: comparison of overground walking speed between two conditions. B: within-condition comparison of overground walking speed. C: comparison of change in overground walking speed between the two conditions. B, baseline; Post, immediately post treadmill walking; 10 min post; 10-min post-treadmill walking. *Significant difference. #Significant level of 0.06.

Change in overground walking speed from baseline to the time point 10-min after treadmill walking tended to be greater for the intermittent condition than for the continuous condition, although this was not significant (t = −2.02, P = 0.06; Fig. 6C). For the intermittent condition (Table 2), participants with shorter step length on the nonparetic leg at baseline (n = 11) showed significant increase in step length on the nonparetic leg (P = 0.008), which resulted in a slightly decreased step length asymmetry, at 10 min after treadmill walking. Participants with longer step length on the nonparetic leg at baseline (n = 4) also exhibited an increase in step length on the nonparetic leg (P = 0.046) at 10 min after treadmill walking. Walking cadence was significantly different between time points (F = 13.98, P < 0.001). Post hoc analysis with the Bonferroni correction revealed that cadence significantly increased from baseline to the time point 10 min after treadmill walking (P = 0.002). For the continuous condition, step length (P > 0.15) and cadence (F = 2.08, P = 0.14) showed no significant difference across time points.

Table 2.

Step length and cadence during overground walking

	B	Post	10-min Post	Δ From B to 10-min Post
Step length (Paretic > Nonparetic; 11 participants)
Intermittent
Paretic, cm	51.3 (3.6)	50.5 (3.0)	51.4 (3.1)	0.0 (1.1)
Nonparetic, cm	41.1 (3.4)	40.8 (3.2)	44.1 (3.5)*	3.0 (0.8)
Asymmetry, %	128.9 (7.4)	127.8 (7.1)	119.5 (5.5)	−9.4 (4.6)
Continuous
Paretic, cm	51.3 (3.6)	49.3 (2.9)	50.4 (2.8)	−1.0 (1.7)
Nonparetic, cm	41.1 (3.4)	40.5 (3.0)	41.9 (3.1)	0.8 (1.1)
Asymmetry, %	128.9 (7.4)	124.6 (5.7)	123.2 (5.8)	−5.6 (3.7)
Step length (Paretic < Nonparetic; 4 participants)
Intermittent
Paretic, cm	48.3 (1.8)	52.6 (0.9)	52.8 (0.7)	4.5 (1.7)
Nonparetic, cm	50.8 (2.5)	53.6 (2.2)	54.9 (2.1)*	4.1 (0.8)
Asymmetry, %	95.3 (2.1)	98.6 (4.4)	96.6 (4.0)	1.3 (2.8)
Continuous
Paretic, cm	48.3 (1.8)	49.9 (1.1)	52.3 (1.9)	4.0 (1.6)
Nonparetic, cm	50.8 (2.5)	53.1 (2.7)	53.4 (1.6)	2.6 (1.4)
Asymmetry, %	95.3 (2.1)	94.5 (3.8)	98.0 (2.5)	2.7 (0.7)
Cadence
Intermittent	76.9 (2.9)	76.2 (3.1)	80.4 (3.1)*	3.5 (0.8)
Continuous	76.9 (2.9)	76.9 (3.3)	79.1 (3.2)	2.2 (0.8)

B, baseline; Post, immediately post-treadmill walking; 10-min post, 10-minutes post-treadmill walking; Δ, change.

^*Significant difference from baseline.

DISCUSSION

In this study, we aimed to determine whether intermittent versus continuous adaptation to a novel walking pattern induced by the application of lateral pelvis pulling force during a single session of walking practice would induce better retention of locomotor adaptation. We found that both intermittent and continuous adaptation to the external pelvis resistance induced improved weight shift toward the paretic side and enhanced muscle activation of hip abductors and adductors immediately after the force was removed. Furthermore, we found that the intermittent adaptation exhibited a greater retention of the improved weight shift toward the paretic side and enhanced muscle activation of the paretic leg, compared with the results of the continuous adaptation. In addition, 10 min after the treadmill walking, participants showed an increase in overground walking speed for the intermittent adaptation condition, compared with the continuous condition, in which participants showed no significant change in overground walking speed. These findings suggest that intermittent versus continuous locomotor adaptation to externally applied perturbation force may promote retention and generalization of motor learning in individuals poststroke.

Repeated adaptation and de-adaptation may enhance retention of motor learning. A major finding of this study was that the locomotor adaptation induced by the application of intermittent versus continuous external pelvis resistance load resulted in greater retention of aftereffect, i.e., improved weight shift toward the paretic side, suggesting that repeated adaptation and de-adaptation processes rather than a continuous massive motor adaptation may be more effective in inducing a longer retention of motor learning. One possible reason is that repeated adaptation and de-adaptation to the external pelvis resistance force in the intermittent condition may induce repeated unlearning and relearning process, which may solidify an internal model and may promote the retention of the aftereffect. Spec?>ifically, sensory prediction errors induced by the pelvis resistance force have been suggested to be the key factor that contributes to the development of an internal model (32), which has been demonstrated by an aftereffect consisting of increased weight shift toward the paretic side during the de-adaptation period. Over time, the aftereffect was washed out during the de-adaptation period. However, when the pelvis resistance force was applied again during the adaptation period 2, the previously experienced errors, i.e., the errors that were experienced during the early adaptation period 1, were revisited. Prior experience with similar sensory prediction errors may increase the sensitivity to errors encountered during relearning, which may induce savings (33, 34). Thus, participants may exhibit faster readaptation to the perturbation force (33). On the other hand, although the aftereffects appear to disappear during the washout period, the extinction of the aftereffect does not mean a complete removal of the short-term motor memory (35–38). The lingering short-term motor memory may promote the motor relearning during the subsequent adaptation period (39) and consequently reinforce the internal model, which may be shown as longer retention of the aftereffects (35, 38).

On the other hand, both the fast and slow motor learning processes might be involved during the motor adaptation to the pelvis resistance force (40). Specifically, in the intermittent condition, although the fast motor learning process may respond strongly to errors to reduce them during the early adaptation periods 1, 2, and 3, it also may induce poor retention. In contrast, the slow motor learning process may respond weakly to errors, but it may enable better retention. Thus, it is possible that the internal model developed by the slow motor learning process may not be completely washed out during the de-adaptation period 1 and may be reinforced during the adaptation period 2, which may be partially washed out during the de-adaptation period 2 and may be further reinforced during the adaptation period 3. This may have resulted in participants showing a prolonged retention of the aftereffect during the postadaptation period.

In the continuous condition, participants who adapted to the lateral pelvis resistance force during the adaptation period showed an aftereffect that consisted of improved weight shift toward the paretic side after force release during the postadaptation period, suggesting that the central nervous system of those patients was able to develop an internal model in response to the external perturbation while walking (32). This is also consistent with previous studies in healthy individuals with the resistance force was applied to the leg (41) or individuals poststroke with the resistance force applied to the paretic leg (14). However, the aftereffects are generally short lived, which may be washed out within a short period of time (i.e., extinction of aftereffects) when they walk in the absence of exposure to the perturbation (i.e., de-adaptation). It is possible that both the fast and slow motor learning processes might be involved to reduce errors during the early motor adaptation period, until the pelvis displacement of the paretic side attained to the level that was comparable with baseline. However, after the participants attained the adapted walking pattern, the sensory prediction errors were diminishing during the late adaptation period, which might have little effect on updating the internal model through the slow motor learning process (40). This may have resulted in participants showing a shorter retention of the aftereffect for the continuous condition.

Alternatively, the prolonged retention of the aftereffect in the intermittent condition may have been due to repeated walking practice with improved weight shift during the de-adaptation periods 1 and 2. Specifically, during the two de-adaptation periods, participants showed an aftereffect consisting of improved weight shift toward the paretic side, although it was gradually decayed over steps. However, overall participants still experienced greater numbers of steps with improved weights shift toward the paretic side in the intermittent condition than in the continuous condition (Fig. 4C). The walking experience with improved weight shift toward the paretic side during the two de-adaptation periods may induce the improved motor performance during the postadaptation period. For instance, in the intermittent condition, participants who had more steps with improved weight shift toward the paretic side during the de-adaptation periods also exhibited greater improvements in weight shift toward the paretic side from the baseline to the late postadaptation period (Fig. 4D). This is consistent with findings from previous studies showing that repetition of a specific movement or muscle activation induced by external perturbation (28, 42, 43) or obstacle (44) influenced movements or patterns of muscle activation after the perturbation/obstacle was removed.

Another possible mechanism is that walking with intervals of no force perturbation in-between the intermittent adaptation periods during locomotor practice may alleviate fatigue and therefore enhance the learning effect. Walking itself is a complex task requiring the coordination of numerous muscles acting on a number of different joints (45). Thus, the continuous adaptation to pelvis perturbation while walking on a treadmill may be physically too demanding for those who suffer from hemiparesis following stroke. In the intermittent condition, which provided periodic intervals of walking with no pelvis perturbation, participants may have been less likely to experience physical fatigue than in the continuous condition. Results from previous studies suggest that fatigue is detrimental to motor learning in healthy individuals (46–48) and to ambulation in people with stroke (49). Thus, less fatigue during locomotor training in the intermittent condition might improve the retention of the aftereffect.

A different pattern of washout of aftereffects for each condition might lead to only a difference in mid postadaptation period between conditions for change in weight shift (Fig. 3C). Specifically, the intermittent adaptation induced slow decline of aftereffects and the continuous adaptation induced immediate, fast decline of aftereffects. Therefore, during the middle postadaptation period, aftereffects appeared to mostly remain for the intermittent condition but to be mostly washed out for the continuous condition, which might lead to a significant group difference during this period. Afterward, however, the remained aftereffects for the intermittent condition were also further declined (although the rate of washout appeared to be smaller), which might lead to no condition difference during the late postadaptation period.

Improvements in weight shift toward the paretic side may induce enhanced muscle activity of hip abductors and adductors of the paretic leg during the stance phase of the paretic leg. Results from previous studies demonstrate that hip abductor and adductor muscles are key muscles for the control of mediolateral weight shift during walking (50–52). Specifically, the hip abductors and adductors are antagonistic muscle pairs and enhanced co-activation of those muscles might improve modulation of mechanical impedance of hip joints (53), contributing to greater stability of the paretic hip joint. Thus, the enhanced muscle activation of hip abductors and adductors during the early postadaptation period for the continuous condition and during both the early and late postadaptation for the intermittent condition might be due to the improvement in weight shift toward the paretic side.

Improved weight shift toward the paretic side may be generalized from treadmill to overground walking, which is supported by the observation that participants showed improvements in step length of the nonparetic leg, cadence (Table 2), and walking speed 10 min after treadmill walking for the intermittent condition, but this was not the case for the continuous condition. This is consistent with the results from previous studies showing that adapted motor control that was achieved during locomotor practice on a treadmill was generalized to overground walking in patients with stroke (28, 41, 54, 55). In particular, a longer retention of the improved weight shifting toward the paretic side for the intermittent condition compared with the continuous condition may be the reason why we observed more improvement in gait symmetry during overground walking for the intermittent condition. Specifically, for the intermittent condition, improved weight shift toward the paretic side might be transferred from treadmill to overground walking, which might facilitate leg swing of the nonparetic leg and consequently result in a longer step length. Increased step length along with improved cadence might result in increased overground walking speed.

Results from this study may have potential applications for gait rehabilitation of patients with chronic stroke. In clinics, physical therapists may apply manual pelvis assistance to facilitate weight shift during treadmill training. Results from this study indicate that applying intermittent pelvis resistance to intermittently increase errors while walking can be used as an approach to improve weight shifting toward the paretic side for a longer retention. The findings from this study can be used by physical therapists or clinicians to develop a more effective intervention approach to further improve weight shifting toward the paretic side, and to enhance retention of error-based motor learning in locomotion in individuals poststroke.

Limitations

In this study, the “dose” of pelvis perturbation was different between the intermittent and continuous condition, i.e., perturbation for 6 versus 8 min. It should be noted that although the intermittent condition used a lower dose of pelvis perturbation, this condition still was more effective in promoting retention of motor learning, compared with results of the continuous condition. Future studies are necessary to test the effectiveness of those two conditions under a same dose of perturbation. Furthermore, in this study, participants walked without pelvis perturbation load during de-adaptation periods for the intermittent condition. We do not know the effect of having sitting/standing breaks instead of walking without the perturbation during de-adaptation periods on retention of motor learning in stroke locomotion. Thus, additional studies are necessary to determine whether it is more effective in facilitating retention of error-based motor learning between having rests or walking without pelvis perturbation during interval periods. In addition, all participants held onto the front handrail for safety, which might have allowed participants to compensate for the pelvis perturbation by using the nonparetic arm. However, we believe that using the handrail did not systematically impact our results because participants held onto the handrail for both testing conditions. Furthermore, due to the technical limitation, we were not able to measure weight shift toward the paretic side while participants walked overground and we were not able to measure ground reaction force during treadmill walking. In addition, we focused on short-term changes in behavioral and neural changes. Future studies are needed to determine the effect of retention or transfer of walking performance using the same paradigm that was used in this study over a longer period of time.

Conclusions

Repeated motor adaptation and de-adaptation to the pelvis resistance force during walking may promote the retention of error-based motor learning for improving weight shift toward the paretic side in individuals poststroke. Although both fast and slow motor learning processes may contribute to the error-based motor learning, repeated unlearning and relearning through the slow motor learning process may promote a longer retention of motor learning and facilitate transfer of motor skills from treadmill to overground walking in individuals poststroke. Knowledge acquired from this study may provide insights into development of error-based locomotor training protocols to promote retention and generalization of motor skills in people poststroke.

GRANTS

This work was supported by the National Institute of Health Grant R01HD082216.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

S.H.P. and M.W. conceived and designed research; S.H.P., S.Y., W.D., and R.R. performed experiments; S.H.P. analyzed data; S.H.P., E.J.R., W.Z.R., and M.W. interpreted results of experiments; S.H.P. prepared figures; S.H.P. drafted manuscript; S.H.P., E.J.R., W.Z.R., and M.W. edited and revised manuscript; M.W. approved final version of manuscript.