Augmenting propulsion demands during split-belt walking increases locomotor adaptation of asymmetric step lengths

All subjects experienced a split-belt protocol while either walking flat or incline throughout two separate experimental sessions (Fig. 1a). The flat session was always performed first. The protocol was tailored (i.e., slope, duration, and speed) to each individual’s ability so that each subject could complete both testing sessions at the same walking speed, given that walking speed directly affects propulsion forces. The subject-specific walking speed on the treadmill (i.e., the mid speed, which is reported in Table 1) was determined by subtracting 0.35 m/s from each subject’s overground walking speed during a Six-Minute Walking Test [64]. We selected this procedure to ensure all individuals completed the entire split-belt walking protocol [29]. The speeds experienced during split-belt walking were selected based on subject’s mid walking speed. The slow speed was defined as 66.6% of the mid speed and the fast speed as 133.3% of the mid speed. These percentages were selected for two reasons: 1) to have a 2:1 split-belt ratio and 2) to have the same averaged speed (i.e., (slow + fast) / 2 = mid speed) throughout all experimental epochs (i.e., baseline, adaptation, and post-adaptation). We selected an inclination of either 5° or 8.5° based on the level of physical fitness of each patient, which was qualitatively assessed with subjects’ self-reported fatigue during the flat session. More specifically, participants reporting fatigue at the end of the flat session were set to walk with a 5° slope during the inclined session, whereas those reporting continued energy at the end of the flat session were set to walk with an 8.5° slope. This was done to ensure that all participants could complete the incline session. Both 5° (e.g., [28, 37, 54, 76]) and 8.5° (e.g., [42, 70]) have been utilized in previous work on the effect of sloped walking on human gait. The flat and incline sessions were separated by 204–360 days without split-belt walking to minimize the effect of multiple exposures to the split-belt environment [45].

Experimental protocols for both sessions consisted of three epochs (i.e., Baseline, Adaptation, and Post-Adaptation). These epochs were used to assess subjects’ baseline walking characteristics and subjects’ ability to adjust and recalibrate their gait for each session-specific slope. Subjects first experienced a baseline epoch, lasting at least 50 strides, to characterize their baseline gait at the specific slope used throughout each session. Subjects walked with both belts moving at the same mid speed (Table 1) in both the flat and incline sessions. A baseline epoch with the belts moving at the slow walking speed (i.e., 66.6% of the mid speed) was also measured during the flat session. However, the slow baseline epoch was removed in the incline session to ensure that all subjects could complete the entire protocol. Next, the Adaptation epoch of at least 300 strides (Table 1) was used to assess subjects’ ability to adjust their gait in response to a split-belt perturbation. During this epoch, the non-paretic leg walked twice as fast as the paretic leg. The paretic leg was defined as contralateral to the lesion site (which was visualized with MRI). The paretic leg always walked on the slow belt in the split-belt condition, even if this experimental design does not always reduce step length asymmetry post-stroke [12, 40, 41, 61, 62]. We chose this experimental design, rather than placing the paretic leg on the fast belt, because inclination augmented the slow leg’s propulsion in young, intact subjects following the split-belt condition [70]. Thus, we focused our study on testing if incline split-belt walking would also augment the paretic propulsion in survivors of a stroke, who have known deficits generating paretic propulsion forces [4, 9]. The duration of the adaptation epoch for each individual is presented in Table 1. We had to tailor the duration of the adaptation epoch based on subject-specific abilities. More specifically, we stopped the adaptation epoch if participants expressed or showed signs of fatigue. That is, we ended the adaptation epoch if participants either had difficulty maintaining their body position in the central region of the treadmill, if their heart rate reached 80% of their maximum heart rate (Max Heart Rate = 220-subject’s age [22, 23]) for 50 consecutive strides, or if participants indicated the desired to stop walking. Despite this variation, all subjects exhibited remarkably similar results. Finally, The Post-Adaptation epoch, lasting at least 10 strides, was used to assess the after-effects when the split-belt condition was removed. Both belts moved at the same mid speed as in the Baseline epoch. We counted the number of strides in real-time to regulate the duration for each epoch, where a stride was defined as the period between two consecutive heel-strikes (i.e., foot landings) of the same leg. All participants took resting breaks as requested, except for the transition from split-to-tied walking where the belts were stopped and restarted as quickly as possible. No steps were taken during the resting breaks; subjects stood or sat still. Also, all subjects wore a safety harness to prevent falls. In addition, there was an instrumented handrail in front of the treadmill for balance support. The handrail data was used to quantify potential differences in handrail holding between the incline and flat sessions. Interestingly, the average force magnitude applied at the handrail was not different between the incline and flat sessions for Baseline (p = 0.75), Late Adaptation (p = 0.66), or After-Effects (p = 0.97). Also, 4 out of 12 individuals that held on to the handrail throughout each experimental session did not qualitatively exhibit smaller changes in adaptation or after-effects across sessions than their counterparts and were, therefore, included in the study.

Kinematic and kinetic data were used to characterize subjects’ ability to adapt their gait during Adaptation, and retain the learned motor pattern during Post-Adaptation. Kinematic data were collected with a passive motion analysis system at 100 Hz (Vicon Motion Systems, Oxford, UK). Subjects’ behavior was characterized with passive reflective markers placed symmetrically on the ankles (i.e., lateral malleolus) and the hips (i.e., greater trochanter) and asymmetrically on the shanks and thighs (to differentiate the legs). The origin of the kinematic data was rotated with the treadmill in the incline conditions such that the z-axis (‘vertical’ in the flat condition) was always orthogonal to the surface of the treadmill (Fig. 1b). Gaps in raw kinematic data were filled with a quintic spline interpolation (Woltring; Vicon Nexus Software, Oxford Uk). Kinetic data were collected with an instrumented split-belt treadmill at 1000 Hz (Bertec, Columbus, OH). Force plates were zeroed prior to each testing session so that each force plate’s weight did not affect the kinetic measurements. In addition, the reference frame was rotated at the session-specific inclination such that the anterior-posterior forces were aligned with the surface on which the subjects walked. A heel-strike was identified in real-time when the raw normal force under each foot reached a threshold of 30 N. This threshold was chosen to ensure accurate counting of strides at all slopes. During data processing we used a threshold of 10 N on median filtered data (with a 5 ms window) to detect the timing of heel strikes more precisely.

A significance level of α = 0.05 was used for all statistical tests, which were performed either with Stata (StataCorp LP, College Station, TX) or with MATLAB (The MathWorks, Inc., Natick, Massachusetts, United States). Normality was assessed with the Lilliefors Test. Except for the step length asymmetries during baseline, all of our step length asymmetry and propulsion outcome measures were normally distributed, thus parametric testing was utilized.

Step length asymmetry adaptation and recalibration were augmented by incline walking compared to flat walking. Figure 2a illustrates the evolution of step length asymmetry throughout the flat and incline sessions. Fig. 2b indicates that there was a wide range of individual Baseline step length asymmetries (colored lines, positive values indicate that the non-paretic step length is longer than the paretic step length) and slope did not change the group average biases (p = 0.30, t (11) = 1.08, d = 0.15). During Adaptation, participants exhibited similar changes in step length asymmetry from early to late Adaptation (Fig. 2d, p = 0.75, t (11) = 0.33, d = 0.08), but they were more symmetric in the incline than the flat session in Late Adaptation (Fig. 2c, p = 0.004, t (11) = − 3.68, d = 0.98). Furthermore, the incline session had larger magnitudes of After-Effects during early Post-Adaptation relative to the flat session (Fig. 2e, p = 0.008, t (11) = − 3.21, d = 1.08). Thus, incline walking augmented step length symmetry during Late Adaptation and the magnitude of After-Effects.

Survivors of a stroke adjusted both step lengths during split-belt walking. The survivors of a stroke modulate both their slow (paretic) and fast (non-paretic) step lengths during Adaptation and have After-Effects during Post-Adaptation (Fig. 3a). The change of each step length during the Adaptation and Post-Adaptation epochs are quantified in Fig. 3b. There was a significant effect of epoch (p_epoch = 0.001, F_epoch (1, 22) = 13.54, η² =0.38) and interaction between leg and epoch (p_leg#epoch < 0.001, F_leg#epoch (1, 22) = 94.23, η² =0.81) indicating that the step length with the paretic leg is reduced during Adaptation, but increased during Post-Adaptation and vice versa for the non-paretic leg. Overall, slope did not alter step length changes (p_slope = 0.16, F_slope (1, 22) = 2.14, η² =0.09, p_leg = 0.44, F_leg (1, 22) = 0.61, η² =0.03, p_slope#leg = 0.18, F_slope#leg (1, 22) = 1.88, η² =0.08, p_slope#epoch = 0.17, F_slope#epoch (1, 22) = 1.97, η² =0.08), except for the paretic leg’s de-adaptation as quantified by ΔPost (p_{leg#epoch#slope} = 0.016, F_{leg#epoch#slope} (1, 22) = 6.86, η² =0.24). More specifically, the paretic step lengths did not exhibit de-adaptation in the flat session (i.e., ΔPost is not different from zero, p_Corrected = 1.00, t (11) = 0.92), whereas step lengths for both legs had significant de-adaptation in the incline session (i.e., non-zero ΔPost, p_Corrected ≤ 0.005, t (11) ≥ ∣ 4.3∣). Survivors of a stroke use both their paretic and non-paretic leg to counteract the split-belt perturbation and both legs are recalibrated following incline adaptation.

Speed and slope-specific leg orientations mediated the distinct step length asymmetries selected during Late Adaptation and early Post-Adaptation. Figure 4a illustrates a top-down view of the baseline leg orientations that contribute to each step length relative to the hips. While we found significantly different leg orientations across individuals (colored lines, p_Individual = 0.002, F_Individual (11, 11) = 6.99, η² =0.87), we observed that the trailing leg position, X, was larger in the incline than flat condition for both legs (p_Slope = 0.042, F_Slope (1, 11) = 5.32, η² =0.33, p_Leg = 0.22, F_Leg (11, 11) = 1.72, η² =0.14, p_Slope#Leg = 0.76, F_Slope#Leg (1, 11) = 0.10, η² =0.01). The schematic in Fig. 4b illustrates the relation between baseline speed-specific leg orientations and late Adaptation [70]. Figure 4c indicates that the participants’ leg orientations during slow Baseline walking predict well those achieved during late Adaptation (solid cyan line; |y| = a*|x|; 95% confidence interval of a = [0.92, 1.13], t_slope (95) = 20.0, R² = 0.76, p < 0.001, r = 0.84). We also show as a reference, the relation between (recorded) Baseline and (predicted) Late Adaptation leg orientation values for both legs and both inclinations in young unimpaired individuals [70] (magenta dashed line; (|y| = a*|x|; 95% confidence interval of a = [0.91, 0.96], t_slope (95) = 73.04, R² = 0.89, p < 0.001, r = 0.94). Note the similarity between the intact and lesioned behavior (cyan vs. magenta lines).

Moreover, we found that the leg orientations achieved during Late Adaptation were predictive of subjects’ Post-Adaptation behavior (Fig. 4d). Specifically, the leading leg’s orientations were similar before and after removal of the split-belt perturbation (i.e., Late Adaptation α_Paretic = Post-Adaptation α_Paretic and vice versa) whereas the trailing legs’ orientations were swapped between the legs (i.e. Late Adaptation X_Paretic = Post-Adaptation X_Non-Paretic and vice versa). This is supported by the significant relationship between Late Adaptation and Post-Adaptation leg orientations observed when individual subjects’ values for each leg and both sloped sessions are regressed (Fig. 4e; solid cyan line; |y| = a*|x|; 95% confidence interval of a = [0.94, 1.02], t_slope (95) = 47.5, R² = 0.79, p < 0.001, r = 0.86). We also show as a reference, the relation between (recorded) Late Adaptation and (predicted) Post-Adaptation leg orientation values for both legs and both sloped conditions in young, intact individuals (magenta dashed line; (|y| = a*|x|; 95% confidence interval of a = [0.95, 1.03], t (95) = 47.54, R² = 0.78, p < 0.001, r = 0.87). Note the similarity between the intact and lesioned behavior (cyan vs. magenta lines). Similar to the intact motor system, the lesioned motor system is able to recover speed and slope-specific leg orientations during Late Adaptation, which predict after-effects during Post-Adaptation.

Sloped walking influenced the extent of recalibration of the non-paretic propulsion forces. Figure 5a shows that propulsion forces were altered during the Adaptation epochs. These data are plotted relative to Baseline propulsion forces (i.e., mid speed), which were larger in the incline condition and the non-paretic leg for both sloped conditions (Fig. 5b: p_Individual = 0.007, F_Individual (11, 10) = 4.84, η² =0.83, p_Slope < 0.0001, F_Slope (1, 10) = 4.84, η² =0.85, p_Leg = 0.040, F_Leg (1, 10) = 5.42, η² =0.33, p_Slope#Leg = 0.43, F_Slope#Leg (1, 10) = 0.69, η² =0.06). Note that subjects were closer to generating Baseline-like propulsion forces during Late Adaptation in the incline session compared to the flat session for both legs, resulting in larger Late Adaptation paretic propulsion forces in the incline session (Fig. 5c). Even though the Late Adaptation behavior was different across sessions (Fig. 5c; non-paretic propulsion: p = 0.032, t (11) = 2.46, d = 0.98, paretic propulsion: p = 0.015, t (10) = − 2.94, d = 1.23), the changes in propulsion forces from early to late Adaptation were similar across sloped conditions (Fig. 5d; non-paretic propulsion: p = 0.92, t (11) = 0.10, d = 0.02, paretic propulsion: p = 0.33, t (10) = − 1.04, d = 0.33). While paretic propulsion After-Effects are similar in either sloped conditions (Fig. 5e, p = 0.43, t (10) = .82, d = 0.17), the non-paretic After-Effects are larger in magnitude following incline adaptation (p = 0.015, t (11) = 2.90, d = 0.82). Note that the paretic propulsion forces change the most during Adaptation (Fig. 5d), whereas the non-paretic propulsion forces are the ones exhibiting after-effects during Post-Adaptation (Fig. 5e). In summary, incline walking results in augmented paretic propulsion forces during Adaptation and reduced non-paretic propulsion force After-Effects.

We found that survivors of a stroke behaved similarly to young, intact adults in their response to sloped split-belt walking [70]. Specifically, survivors of a stroke were able to augment their propulsion forces in response to incline split-belt walking as observed in young, healthy adults [70]. It should be noted that all our analysis was done with peak forces, but we found similar effects with other metrics to quantify propulsion, such as impulse or mean force (data not shown). Our observation is consistent with previous literature indicating that patients in the chronic phase post-stroke can modulate paretic propulsion forces in response to task demands [3, 24‐26, 34]. In particular, we observed during split-belt walking (i.e., adaptation period) that individuals exhibited larger non-paretic and paretic propulsion forces relative to baseline and early adaptation, respectively. These increments are possibly enabled by augmenting latent central drive to plantarflexors [2]. Interestingly, we find significant reduction in the fast (i.e. non-paretic) leg’s propulsion forces post-adaptation (i.e., after-effects) relative to baseline walking. This finding is consistent with our previous study in young, unimpaired adults [70], but not with other studies reporting no significant changes in propulsion forces following split-belt walking in the flat condition [57, 66]. We speculate that this might be because we use more naturalistic walking speeds than in previous experimental designs [57, 66].

Our results provide further evidence that the adaptation of step length asymmetry can be predicted from Baseline walking. Notably, it has been previously suggested that patients’ gait asymmetries during baseline walking determine the extent to which they can adapt their movements in the unusual split condition [50]. We observed that survivors of a stroke reach distinct asymmetry levels across the incline and flat conditions. Thus, our results support previous findings [40, 41] indicating that it is not baseline gait asymmetry, but kinetic demands that govern the degree to which patients adapt their motor patterns in the split-belt task. More specifically, survivors of a stroke adjusted their leg orientations to augment the propulsion forces required for walking in the incline split condition as observed in young, healthy adults [70]. In other words, the forces generated to propel one’s body forward constitute an important control variable regulating the adaptation of movements in the intact and asymmetric motor systems. Further, a recent study suggests that individuals adjust leg orientations to harness energy from the treadmill in the split condition [67]. While this theory explains well the orientation of the leading leg, it does not match well the observed orientation of the trailing leg (Additional file). Perhaps, other factors, such as stability [10], also contribute to the control of leg orientation in walking. It should be pointed out that we only tested the relation between leg orientations during baseline and adaptation in the paretic leg. We speculate that the same would have been observed in both legs and sloped conditions like young adults [70]. This is a reasonable expectation given that survivors of a stroke exhibited similar control of leg orientations to young adults during Late Adaptation and early Post-Adaptation for both legs and sloped conditions. Nonetheless, future work is needed to verify that the relation between baseline and adaptation is observed for distinct inclination conditions, or if the legs were walking at different belt-speed ratios than the ones we used.

Survivors of a stroke recruited both legs in order to adapt their gait, whereas young adults primarily adapted one leg. Notably, we observed that survivors of a stroke adapted both the paretic (slow belt) and non-paretic (fast belt) step lengths, whereas we previously found that young individuals predominantly adjusted the fast belt step length [60, 70]. This could be because survivors of a stroke may require more repetitions in the altered environment [75] to recover their baseline leg orientation with their paretic leg, whereas intact subjects can do so immediately after the split condition is introduced. Alternatively, it could be that the larger neural coupling post-stroke [35] enhances bilateral adaptation. In other words, it might not be possible for individuals who had experienced a stroke to adapt one leg in isolation due to neural drive sent to both limbs. Regarding post-adaptation, paretic after-effects were only observed in the incline condition. More specifically, paretic step lengths become longer than in baseline walking, which may be beneficial for survivors of a stroke who take short paretic step lengths [4]. On the other hand, non-paretic after-effects were observed regardless of the sloped condition. This was atypical since the non-paretic leg walked fast in the split condition and young adults only exhibit after-effects in the leg that walked slow [70]. Thus, it was unexpected to observe non-paretic step lengths shorter than those taken during baseline. This shortening of non-paretic step lengths might be a strategy to recover balance (e.g., [18]), which is challenged upon removal of the split condition [10, 29]. In summary, survivors of a stroke adapt both legs during split-belt walking, but paretic step length after-effects are only observed following incline split-belt walking.

The long-term therapeutic effect of locomotor adaptation with split-belt treadmills may be due to walking with the motor demands of the split-belt task, rather than the adaptation effects observed post-adaptation. Split-belt walking has been shown to reduce long term gait asymmetry [7, 47, 61]. Little is known about the aspects of split-belt walking that underlie these long-term changes: the cumulative effect of brief after-effects post-adaptation or the repeated exposure to motor demands specific to the split condition during the adaptation period. After-Effects could lead to motor improvements [5] such as temporarily reduced gait asymmetry [12, 62]. However, these after-effects are short lived and decrease as individuals experience multiple days of practicing the split-belt condition [38, 45, 71]. It is known that regular treadmill walking cannot modify gait asymmetries post-stroke [33, 58, 69], suggesting that the specific motor demands of the split-belt task might be important for neurorehabilitation. For example, we observe that the split condition (during the adaptation period) forces patients to take longer paretic step lengths and generate greater paretic propulsion forces. Perhaps practice of these gait features through multiple exposures to the split situation reinforces those patterns and ultimately leads to long-term reductions of gait symmetry overground. It is also possible that the strenuous nature of split-belt walking increases neural plasticity, as shown with other high-intensity exercises [1]. Thus, incline split-belt walking may be beneficial not only for inducing greater paretic propulsion, but also because it is more demanding than level walking [30]. Lastly, the initial disruption of step length asymmetry might trigger exploration of new locomotor patterns [52] that could converge to more metabolically efficient gait than their baseline walking pattern [67, 68]. In summary, the long-term benefit of split-belt walking may originate from practicing and reinforcing motor patterns imposed by the split condition, rather than repeating the briefly lived after-effects.

Split-belt walking has been shown to induce long term changes that could improve the mobility of those who have had a stroke [7, 47, 61]. Previous studies have investigated the impact of training design (e.g., paretic leg on the slow belt vs. fast belt) on the therapeutic effect of split-belt walking [40, 61]. Our work contributes to this literature by indicating that this choice depends on the rehabilitation outcome of interest. For instance, our results suggest that placing the paretic leg on the fast belt would force subjects to augment their paretic propulsion forces and lengthen their paretic steps during split-belt walking, which could be advantageous to some patients. This idea is consistent with a pilot study showing that platarflexor moments trained on the fast side increase following multiple split-belt walking sessions [8]. This therapeutic effect is beneficial to individuals with baseline asymmetries due to shorter paretic step lengths coupled with plantar flexor weakness [6]. It remains, however, an open question the extent to which incline split-belt walking could augment paretic propulsion when the paretic leg is placed on the fast belt. Future studies are needed to test this given the limited changes in paretic propulsion that we observed post-adaptation compared to controls [70]. In sum, our study provides greater understanding of the motor demands associated to the split-belt task, which could be harnessed for gait neurorehabilitation.

Incline split-belt training may be a promising way to augment locomotor and adaptation and recalibration in the lesioned motor system. Not everyone who has had a stroke re-learns to walk symmetrically following several weeks of flat split-belt training [7, 47, 61]. Thus, it is clinically relevant to explore alternative strategies to augment adaptation in survivors of stroke other than increasing the speed difference [7, 77] since not all patients can walk with large speed differences. While this work indicates that adaptation can be augmented in patients, previous work indicates that overground walking post-stroke is most improved following decline, rather than incline, interventions [11]. Moreover, it has been previously observed that motor patterns observed on the treadmill do not fully transfer to overground walking [63, 71, 73, 74]. Thus, future studies are needed to determine if the augmented adaptation in the incline environment transfers to flat overground walking. Furthermore, we designed our study to evaluate changes in step length asymmetry relative to baseline walking. It is, however, clinically relevant to determine if the augmented adaptation and after-effects lead to reduced step length asymmetry in absolute terms. Finally, it must be noted that participants were moderately to mildly impaired (i.,e, 21 < Fugl-Meyer Assessment leg motor score ≤ 34). Therefore, it remains an open question if similar results could be observed in more severely affected individuals or with different demographics than our cohort of participants.