Use of explicit processes during a visually guided locomotor learning task predicts 24-h retention after stroke

Abstract

Implicit and explicit processes can occur within a single locomotor learning task. The combination of these learning processes may impact how individuals acquire/retain the task. Because these learning processes rely on distinct neural pathways, neurological conditions may selectively impact the processes that occur, thus, impacting learning and retention. Thus, our purpose was to examine the contribution of implicit and explicit processes during a visually guided walking task and characterize the relationship between explicit processes and performance/retention in stroke survivors and age-matched healthy adults. Twenty chronic stroke survivors and twenty healthy adults participated in a 2-day treadmill study. Day 1 included baseline, acquisition1, catch, acquisition2, and immediate retention phases, and day 2 included 24-h retention. During acquisition phases, subjects learned to take a longer step with one leg through distorted visual feedback. During catch and retention phases, visual feedback was removed and subjects were instructed to walk normally (catch) or how they walked during the acquisition phases (retention). Change in step length from baseline to catch represented implicit processes. Change in step length from catch to the end of acquisition2 represented explicit processes. A mixed ANOVA found no difference in the type of learning between groups (P = 0.74). There was a significant relationship between explicit processes and 24-h retention in stroke survivors (r = 0.47, P = 0.04) but not in healthy adults (r = 0.34, P = 0.15). These results suggest that stroke may not affect the underlying learning mechanisms used during locomotor learning, but that these mechanisms impact how well stroke survivors retain the new walking pattern.

NEW & NOTEWORTHY This study found that stroke survivors used implicit and explicit processes similar to age-matched healthy adults during a visually guided locomotion learning task. The amount of explicit processes was related to how well stroke survivors retained the new walking pattern but not to how well they performed during the task. This work illustrates the importance of understanding the underlying learning mechanisms to maximize retention of a newly learned motor behavior.

INTRODUCTION

Locomotor learning has primarily been studied using the split-belt treadmill paradigm in both young healthy adults (1–4) and stroke survivors (5–8). Traditionally, changes in behavior during split-belt treadmill walking are believed to result from sensorimotor adaptation (9). Sensorimotor adaptation is a specific form of motor learning that is an implicit (i.e., automatic), primarily cerebellar process that occurs in response to sensory prediction errors (i.e., a mismatch between what is expected and what actually occurs) (10–12). However, there are other types of learning that are distinctly different from sensorimotor adaptation, such as use-dependent plasticity and strategic learning, which have rarely been studied during locomotion. For example, use-dependent plasticity is an implicit process that may involve the primary motor cortex and occurs in response to repetitive practice (13–17). Strategic learning, on the other hand, is explicit in nature and can be thought of as intentional changes to behavior that occur in response to a perturbation (18). This type of learning is thought to rely heavily on prefrontal and cognitive processes and occurs in response to task performance error (i.e., if you were successful at the task) (12, 15, 19). Despite the unique characteristics of these learning processes, they rarely occur in isolation and most often occur together during a single learning task (14, 20–24). In other words, the behavior observed during learning tasks usually reflects multiple underlying learning processes that occur in parallel. Therefore, understanding how these learning processes work together within a single task is critical to developing a comprehensive understanding of locomotor learning (i.e., both the acquisition of a new behavior and retention of that behavior). This is particularly important when developing rehabilitation interventions targeting locomotion, because damage to different neural pathways due to stroke and other neurological conditions likely impact each of these specific learning processes differently.

Despite the importance of understanding underlying learning mechanisms and how they work together during locomotion and after stroke, the learning processes that occur within a single locomotor learning task have not been examined. This is in contrast to visuomotor rotation reaching tasks for which learning processes have been well characterized, making it a useful task to illustrate the importance of understanding underlying learning mechanisms. Historically, the visuomotor rotation task was believed to involve primarily sensorimotor adaptation; however, recent work has demonstrated that explicit, strategic learning processes contribute significantly to the behavior that is observed during acquisition of the new behavior, whereas sensorimotor adaptation and use-dependent plasticity (i.e., implicit learning processes) have a smaller role (12, 19, 22, 25–29). The combination of these distinct learning processes produces the observed reaching behavior and impacts how individuals perform and retain the task. In addition, because these learning processes rely on distinct neural pathways, neurological conditions selectively impact the learning processes that occur, thus, having direct implications for learning and retention. For example, individuals with cerebellar damage have difficulty with sensorimotor adaptation, whereas individuals with damage to the prefrontal cortex have difficulty with explicit, strategic learning. Damage to these different regions of the brain (i.e., the cerebellum and prefrontal cortex) results in different observed behavior during extended practice of a visuomotor rotation task when an aiming target is provided because of impairments to specific learning processes based on lesion location (12, 29). These results suggest that in individuals with cerebellar damage, tasks that are dominated by sensorimotor adaption may be more impaired, whereas tasks that are dominated by explicit, strategic learning may be more affected in individuals with prefrontal damage. This knowledge can then be leveraged to develop effective, motor-learning based rehabilitation interventions.

In locomotor learning research, little work has sought to understand how these different learning processes work together during the same task or how stroke impacts these parallel processes. Work in young healthy adults and one study in those with stroke have used dual-task walking, often by combining split-belt treadmill walking with some form of visual feedback (2, 30–34), to provide evidence that multiple learning processes can occur during different tasks that are occurring simultaneously. Of particular interest is the work of Roemmich et al. (33) which found that young healthy adults made a conscious correction related to step length when visual feedback was provided during split-belt treadmill walking, indicating that implicit and explicit processes were occurring together during this task. This was supported by the successful application of the dual-rate state space model. In addition, Cherry-Allen et al. (30) found that stroke survivors were able to use visual feedback to learn to walk with more knee flexion while learning a new step length on a split-belt treadmill, again indicating that multiple learning processes can occur together during locomotion. Despite evidence that multiple learning processes can occur simultaneously during locomotion in healthy adults and in stroke survivors, past work has not investigated how different learning processes operate together within the same locomotor task after stroke. Furthermore, it is not known how the learning processes used during a locomotor learning task impact the acquisition of the new walking behavior and retention of that behavior. Understanding how stroke impacts parallel learning processes within a single task and how the learning processes used relate to performance and retention will guide the development of more effective locomotor rehabilitation interventions that promote efficient learning processes in individuals after stroke.

As a result, the purpose of this work is twofold. First, we wanted to understand whether the contributions of implicit and explicit processes differed between stroke and age-matched neurologically intact subjects during a locomotor task known to involve both forms of learning. To do this, we used a distorted visual feedback (DVF) locomotor learning task previously described by our laboratory (20, 24) and shown to involve both implicit and explicit processes. The contribution of each type of learning was estimated using walking behavior while visual feedback is provided (i.e., during the acquisition of the new walking pattern). Because explicit, strategic learning relies on cognitive processes (12), which are often impaired after stroke (35–37), we hypothesized that explicit, strategic processes would contribute less to the acquisition of the new walking pattern in stroke survivors than in age-matched healthy adults. We also hypothesized that implicit processes would contribute similarly in stroke survivors and age-matched healthy adults as past work has suggested that implicit processes are intact after cerebral stroke (6, 7, 38–41). Second, we wanted to understand the relationship between the amount of explicit, strategic processes and locomotor learning in stroke survivors and age-matched healthy adults. Here locomotor learning refers to the following: 1) performance (i.e., behavior while visual feedback is present), which reflects the change in behavior that occurrs during the practice session, and 2) retention (i.e., behavior when the visual feedback is removed), which reflects longer term storage of the new walking pattern. Based on upper extremity work suggesting that the amount of implicit and explicit processes do not impact performance of a task (28), we hypothesized that the contribution of explicit processes would not be related to how well stroke survivors or age-matched healthy adults performed while acquiring the new walking pattern. We also hypothesized that the amount of explicit processes would relate to how well the new walking pattern was retained (i.e., recalled at a later time when the visual feedback was removed). This hypothesis was based on past work in upper extremity reaching showing that aftereffects are smaller (i.e., retention is decreased) when a task is more explicit (28, 42).

MATERIALS AND METHODS

Subjects

Stroke survivors were recruited from local physical therapy practices, support groups, and advertisements, whereas age-matched healthy adults (±5 years) were recruited from the local community through fliers and advertisements. To be included, all subjects had to meet the following criteria: 1) age 18–85 yr, 2) able to walk 10 m without physical assistance, 3) resting heart rate within 40–100 beats/min, and 4) resting blood pressure within 90/60 mmHg and 185/100 mmHg. In addition, stroke survivors had to have MRI evidence of a single unilateral chronic (>6 mo) stroke and healthy adults needed to be in generally good health. Subjects were excluded if they reported the following: 1) visual changes post stroke, 2) history of coronary artery bypass graft or myocardial infarction within 3 mo, 3) unexplained dizziness in the last 6 mo, or 4) pain that limited walking. Stroke survivors were also excluded if any of the following criteria were met: 1) MRI evidence of a cerebellar stroke, 2) neurological conditions other than stroke, 3) neglect, or 4) inability to communicate with or understand investigators. All subjects signed an informed consent approved by the Human Subjects Review Board at the University of Delaware before participation.

Experimental Design

Subjects walked on a dual-belt treadmill instrumented with two 6 degrees of freedom force plates (Bertec Corporation, Columbus, OH) during the 2-day study. Day 1 consisted of five phases: baseline, orientation, acquisition1, catch trial, acquisition2, and immediate retention, whereas day 2 consisted of 24-h retention (Fig. 1A). During all bouts of walking, stroke survivors walked at a comfortable, self-selected walking speed with a 20-inch computer monitor positioned in front of the treadmill (Fig. 1B). As detailed below, a visual feedback display created with the Motion Monitor Toolbox (Innovative Sports Training Inc., Chicago, IL) was presented on this monitor during specific walking bouts of the study. Age-matched healthy adults walked at the same speed as their corresponding stroke survivor.

Figure 1.

Experimental setup and paradigm. A: subjects participated in a 2-day treadmill-walking study. Day 1 consisted of six phases of walking, whereas day 2 consisted of one phase of walking. During baseline, catch trial, immediate, and 24-h retention, no visual feedback was provided. During orientation, acquisition1, and acquisition2 (dotted boxes), visual feedback about step length was displayed. During catch trial, subjects were instructed to walk normally. During immediate and 24-h retention, subjects were instructed to use the walking pattern they had learned during the acquisition phases. B: during treadmill walking, a computer monitor was positioned in front of the treadmill. This display presented real-time feedback about step length during orientation, acquisition1, and acquisition2; otherwise, the display was blank. C: during acquisition1 and acquisition2, visual feedback of the manipulated leg was distorted such that the bar looked shorter than the actual step being taken (dotted line represents actual step length); thus, a longer step with the manipulated leg was needed to hit the target line. The bar of the nonmanipulated leg was accurate. Each time the bars high the target line the counter increased. ML, manipulated leg; NML, nonmanipulated leg; VF, visual feedback.

No visual feedback was displayed during baseline. During this phase, we calculated the average step length for each leg and identified the leg that took a shorter step. This leg will be called the manipulated leg (ML), whereas the other leg will be called the nonmanipulated leg (NML). This information was used during the next phases to create a visual feedback display.

During orientation, acquisition1, and acquisition2, the computer monitor had a bar graph that displayed real-time information about step length as described in French et al (20). Briefly, this display had two bars, one representing the left leg and one representing the right leg (Fig. 1B). Each bar grew in real-time during the swing phase until heel strike was identified, at which point the bar maintained its height until the next swing phase was initiated on that side (Fig. 1C). In addition, two horizontal target lines were positioned two-thirds of the way up the screen (Fig. 1, B and C). These target lines represented the subject’s baseline step lengths and were scaled such that they were positioned two-thirds of the way up the screen for all subjects (Fig. 1, B and C). The height of these line remained the same throughout these three phases. Orientation was included to ensure that subjects understood the display. Prior to the first portion of orientation, subjects were told, “You are going to see a bar graph on the screen in front of you. Each bar will represent your step length. Step length is the distance from one heel to the other when the front heel hits the ground. On the graph, the green bar will represent your right leg and the blue bar will represent your left leg. You will also see two pink horizontal lines across the screen. These lines represent your baseline step length and will serve as a target line later. Please watch the display and walk normally.” Subjects were asked to verbalize what step length meant to confirm understanding. During a portion of this phase, subjects were instructed to “take the longest possible step” with the ML to make the corresponding bar “go above the pink line.” This portion of orientation was used to ensure that subjects were physically able to change their step length enough to succeed with the task.

During acquisition1 and acquistion2, we used DVF to teach subjects a new walking pattern (20). During these phases, the target lines remained the same, but the bar representing the ML was distorted such that it appeared that subjects were taking a shorter step with the ML than they actually were (Fig. 1C). Thus, to get the bar of the ML to hit the target line, subjects needed to increase the step length on the ML. The bar was distorted by 15% of the baseline step length of the ML for all subjects (Fig. 1C). The bar representing the NML was accurate. During the learning phases, there were numbers above the bar that increased when each leg’s bar was within ±2% of the target line. Prior to acquisition1, subjects were told, “You are now going to walk while watching the same graph. This time there will be numbers above each bar. These numbers will increase each time the bar hits the pink line. Your goal is to get both of those numbers as high as possible by getting the bars to hit the pink line; not go above or below it but right on the line. The bars will not be completely accurate, so you will have to change your walking to do this.” Subjects were then specifically told which leg to take a longer step with (i.e., the ML) and which to keep the same (i.e., the NML). This was done to make the learning paradigm as explicit as possible as we were interested in explicit, strategic learning. Before acquisition2, subjects were reminded of the goal and which leg to change. They were also asked to “try and remember exactly what you are doing to get the bars to match the pink line, because later I will ask you to repeat this learned pattern without the visual feedback.” These instructions are similar to those used in previous studies (20, 43). These phases were used to teach subjects a new walking pattern and to quantify performance.

Between acquisition1 and acquisition2, subjects completed a 30 s catch trial, where visual feedback was removed (Fig. 1C). During this bout, subjects were told, “You will not see anything on the display. During this time please walk normally.” This phase will be used to quantify the contribution of each type of learning process (i.e., implicit and explicit). Specifically, the instructions to walk normally were provided to remove the contribution of explicit, strategic processes; thus, allowing us to estimate the implicit processes by observing the aftereffects. This approach is commonly used to examine storage of implicit processes in both upper extremity (23, 44, 45) and walking tasks (4, 7, 20, 32); however, it is possible that subjects continued to use some explicit strategy during the catch trial. In addition, subjects were excluded if they were not able to understand the investigators; thus, we assumed that subjects stopped using whatever strategy they developed and used the catch trial to estimate the contribution of implicit processes.

Finally, during immediate retention and 24-h retention, no visual feedback was provided and subjects were told, “You will not see anything on the display, but I want you to walk how you were walking while you watched the display. So, whatever you were doing to get the bars to hit the pink line, I want you to recreate that.” Consistent with previous work (43, 46–50), we used these phases, when visual feedback was removed and subjects are asked to recall the pattern, to quantify retention.

Data Analysis

Motion capture data were collected from six markers placed on bilateral heels, lateral malleoli, and fifth metatarsal heads. Kinematic data were collected at 100 Hz using an eight-camera Vicon Motion Capture system (Vicon MX, Los Angeles, CA), whereas kinetic data were collected at 1,000 Hz. Heel strike was determined as the first frame when the ground reaction force exceeded 20 N and was visually inspected to accuracy. After identifying gait events, we calculated step length of the ML (SL_ML) using a custom written MATLAB script (MathWorks, Natick, MA). SL_ML was then used to calculate the amount of implicit and explicit, strategic learning processes and our measures of performance and retention.

To determine the contribution of implicit and explicit, strategic processes that occurred during the task, we adapted a method used during visuomotor rotation tasks to measure contributions from specific learning mechanisms. In visuomotor rotation tasks, subjects report the specific location to which they were aiming before the reach to represent the amount of explicit, strategic learning. The difference between that reported aim and the observed behavior reflects the implicit processes (22, 23, 51). Because locomotion is a continuous task, a similar reporting method to quantify the explicit, strategic learning process is not practical. However, catch trials have been commonly used during locomotion to quantify implicit processes (7, 20, 32, 52). Thus, rather than measuring the contribution of explicit, strategic processes as done in visuomotor rotation tasks, we first estimated the amount of implicit processes and then estimated the contribution of explicit, strategic processes. The amount of implicit processes was defined as the difference between the average SL_ML during catch trial and the average SL_ML during baseline (i.e., the aftereffects) and, therefore, was measured in cm (Fig. 2). The instructions to “walk normally” were designed to remove the contribution of strategy to behavior. This approach of removing the contribution of strategy with a cue has been used previously (20, 23, 24); however, it does assume that subjects follow this instruction. The contribution of explicit, strategic processes was then estimated as the difference between the average SL_ML during late acquisition (i.e., the last 30 steps of acquition2) and the average SL_ML during the catch trial and was also measured in cm (Fig. 2). Late acquisition was used to quantify the contribution of explicit processes as past work has used the acquisition phase to understand the ongoing learning mechanisms (22, 23, 51). For these metrics of implicit and explicit processes, a higher value indicates a larger change in behavior due to that specific learning processes.

Figure 2.

Calculation of the amount of implicit and explicit, strategic processes. The amount of implicit learning was calculated as the difference between the average step length during the catch trial (dashed line) and baseline (solid line). The amount of explicit, strategic processes was calculated as the difference between the average step length at late acquisition (dotted line) and catch trial. The observed behavior at late acquisition is the sum of these two types of learning processes.

To quantify performance while visual feedback was present, we used absolute error (AE) to capture improvements in accuracy. This measure was calculated as follows:

$${AE}_n=\frac{|{ASL}_n-{SL}_{Target}|}{{SL}_{BL}}*100,$$

where ${\text{A}\mathrm{S}\mathrm{L}}_{\mathrm{n}}$ is the step length of the ML on step n during acquision1 and acquision2, ${\mathrm{S}\mathrm{L}}_{\mathrm{T}\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{e}\mathrm{t}}$ is the step length of the ML that would result from hitting the target every step (i.e., 15% longer than the baseline SL_ML), and ${\mathrm{S}\mathrm{L}}_{\mathrm{B}\mathrm{L}}$ is the average step length of the ML during baseline. This metric represents how far the subject was from the target line during acquisition phases. This metric is expressed as a percentage of the subjects’ baseline step length, because the amount of distortion was based on subjects’ baseline step length; thus, to accurately compare error between subjects, the amount of error must be relative to baseline step length. For this metric a smaller value indicates less error or better performance while acquiring the new walking pattern. We would expect performance to improve as indicated by a reduction in error if learning was occurring.

To quantify retention, we calculated two measures of forgetting, one for immediate retention (FORGET_Imm) and one for 24-h retention (FORGET_24h). These metrics were calculated as the absolute difference between SL_ML at late acquisition (i.e., the last 30 steps of acquisition2) and the average SL_ML during immediate or 24-h retention (Fig. 1A). The average SL_ML was calculated across the entire 2 min of the retention phases, as we do not expect behavior to change over this time. However, to confirm that this did not impact our results, we calculated the measures of FORGET_Imm and FORGET_24h with the average of the first 30 steps as well and found no impact of using different number of steps to represent the retention phases; thus, we only present the results using the entire retention bout here to calculate FORGET_Imm and FORGET_24h. For this metric a smaller value indicates less forgetting or better retention.

Statistical Analysis

All statistical analyses were performed in SPSS v25 (IBM, Chicago, IL) with α set at 0.05. All measures described in this section were screened for outliers (i.e., data >2 SDs above/below the mean), and, if identified, outliers were removed from the subsequent analysis. We performed Shapiro–Wilks and Levene’s tests to assess normality and homogeneity, respectively. If normality or homogeneity were violated, nonparametric tests were used, otherwise parametric tests were used as outlined below. In addition, Mauchly’s test of sphericity was used to assess sphericity of the data during ANOVAs and, if violated, a Greenhouse–Geisser correction was used. When a repeated-measures ANOVA was planned but parametric statistics could not be used, we performed Mann–Whitney U tests.

Before conducting our primary analyses, we examined if stroke survivors and healthy adults performed and retained similarly on this task, as differences in these measures could impact subsequent analyses. To do this, we compared the average AE during the first 10 steps of acquisition1 (i.e., early acquisition) and the last 30 steps of acquisition2 (i.e., late acquisition) between groups with a 2 × 2 mixed ANOVA. In addition, we used a 2 × 2 mixed ANOVA to compare FORGET_Imm and FORGET_24h between groups.

Next, we began our primary analyses by comparing the contribution of implicit and explicit processes between groups using a 2 × 2 mixed ANOVA, with learning process and group as factors. Finally, we examined the relationship between the contribution of explicit processes and performance and retention within each group. AE served as our measure of performance, whereas FORGET_IMM was our measure of immediate retention and FORGET_24h was our measure of 24-h retention. Thus, we performed Pearson’s correlations (or Spearman’s correlation, if data were not normal) between the contribution of explicit processes and 1) AE at late acquisition, 2) FORGET_Imm, and 3) FORGET_24h. All three of these correlations were performed separately in the group of stroke survivors and age-matched healthy adults.

RESULTS

Twenty stroke survivors (64.2 ± 8.9 yr; 10 women) and twenty age-matched healthy adults (64.8 ± 8.5 yr; 13 women) were included in this study. Additional characteristics of the subjects are presented in Table 1. When asked to take the longest possible step with the ML during orientation, all subjects exceeded 15% of their SL_ML during baseline (stroke survivors 26.4 ± 13.5%, healthy adults 42.4 ± 23.3%), indicating that all subjects were physically able to walk with the pattern we were trying to teach them; thus, no subjects were excluded from the analysis due to inability to complete the desired task. We identified one outlier in the age-matched healthy adult group for AE at late acquisition and one outlier in the age-matched healthy adult group for FORGET_24h; thus, data for that specific measure for these individuals were removed from the analysis.

Table 1.

Demographic and clinical characteristics of subjects

	Stroke Survivors (n = 20)	Healthy Adults (n = 20)
Age, yr; means ± SD	64.2 ± 8.9	64.8 ± 8.5
Sex	10 F; 10 M	13 F; 7 M
Manipulated leg	10 R; 10 L	10 R; 10 L
Time since stroke mo; means ± SD	51.1 ± 29.2
Side of lesion	16 R; 4 L
SSWS, m/s; means ± SD	0.93 ± 0.2
FMLE, means ± SD	25.9 ± 5.8

F, female; FMLE, Fugl–Meyer lower extremity; L, left; M, male; R, right; SSWS, self-selected walking speed.

Before performing our primary analyses, we compared performance and retention between groups. Due to non-normal data, we compared AE during early acquisition and late acquisition between groups with Mann–Whitney U tests to examine performance. This showed no significant difference between groups during both early acquisition (Z = −0.54, P = 0.60) and late acquisition (Z = −0.98, P = 0.34; Fig. 3, A and B). This result suggests that the performance of both groups improved similarly during the acquisition phases. To examine retention, we compared FORGET_Imm and FORGET_24h between groups using a 2 × 2 mixed ANOVA (Fig. 3C). We found an effect of time [F(1, 37) = 4.82, P = 0.04] but no effect of group [F(1, 37) = 0.50, P = 0.48) and no interaction [F(1, 37) = 2.03, P = 0.16). This result suggests that subjects forgot more 24 h later; however, both the groups forgot similarly. Because these metrics of performance and retention were not different between groups, we were able to examine the contribution of implicit and explicit processes between groups without the quality of performance or magnitude of retention influencing the results.

Figure 3.

Performance and retention in stroke survivors and age-matched healthy adults. A: mean absolute error for stroke survivors (dark gray) and healthy adults (light gray) during acquisition1 and acquisition2 in bins of five steps with the shaded region indicating the standard error. Data are truncated to the shortest number of steps taken by a single subject with the means and standard error of the last 30 steps for each subject shown at the end of each acquisition1 and acquisition2. A reduction in error over the course of the acquisition phases is observed in both groups, suggesting that performance improved over this time. B: absolute error, our measure of performance, was not different between groups at early acquisition (P = 0.70) or late acquisition (P = 0.34). As shown by the individual lines, most subjects demonstrated a reduction in error from early to late acquisition. C: forgetting, our measure of retention, was not significantly different between stroke survivors and healthy adults at immediate (P = 0.56) or 24-h retention (P = 0.11). The dots represent each subject and error bars indicate standard error in B and C.

Contribution of Implicit and Explicit Processes

As the first part of our primary analysis, we used the aftereffects during the catch trial to quantify implicit and explicit processes. Both groups used implicit and explicit processes. However, a 2 × 2 mixed ANOVA did not find a main effect of type of learning process (i.e., implicit or explicit) [F(1, 38) = 1.45, P = 0.24] or group {F(1, 38) = 0.11, P = 0.74; Fig. 4). In addition, we did not find a significant interaction [F(1, 38) = 0.04, P = 0.85; Fig. 4). These results suggest that the types of learning processes were not different between groups and that similar amounts of implicit and explicit processes occurred during the task in both groups. Despite these results, it is important to note the significant intersubject variability in implicit and explicit processes in both groups (Fig. 4).

Figure 4.

Amount of implicit and explicit, strategic processes in stroke survivors and age-matched healthy adults. Both implicit and explicit processes contributed to the acquisition of a new walking pattern in stroke survivors (dark gray) and age-matched healthy adults (light gray). A 2 × 2 mixed ANOVA showed no main effect of type of learning [F(1, 38) = 1.45, P = 0.24] or group [F(1, 38) = 0.11, P = 0.74] and no interaction [F(1, 38) = 0.04, P = 0.85). Individual dots represent subjects and illustrate the significant amount of variability in both groups. Error bars reflect standard error.

Relationship Between the Contribution of Explicit Processes and Performance/Retention

Last, we examined the relationship between the amount of explicit processes and 1) AE, 2) FORGET_Imm, and 3) FORGET_24h in stroke survivors and age-matched healthy adults. Because of non-normality, we used Spearman’s correlation to assess the relationship between the contribution of explicit processes and performance, as measured by AE at late acquisition, in both groups. In stroke survivors, we found no relationship between the amount of explicit processes and AE at late acquisition (ρ = 0.18, P = 0.43; Fig. 5A). Similarly, in age-matched healthy adults, we found no relationship between the contribution of explicit processes and AE at late acquisition (ρ = −0.23, P = 0.33; Fig. 5B). These results suggest that the amount of explicit, strategic processes did not have a significant impact on performance.

Figure 5.

Relationship between the amount of explicit, strategic processes and locomotor learning. There was no significant relationship between the amount of explicit processes and performance in stroke survivors (A) and healthy adults (B). There was a moderate and significant relationship between the amount of explicit processes and same-day (C) and 24-h forgetting (E) in stroke survivors. However, in age-matched healthy adults, there were no significant relationships between the amount of explicit processes and same-day (D) and 24-h forgetting (F).

In stroke survivors, we found moderate relationships between the contribution of explicit processes and FORGET_Imm (r = 0.53, P = 0.02; Fig. 5C) and FORGET_24h (r = 0.47, P = 0.04; Fig. 5E). In age-matched healthy adults, we found no relationship between the contribution of explicit processes and FORGET_Imm (r = −0.03, P = 0.91; Fig. 5D) or between the contribution of explicit processes and FORGET_24hr (r = 0.34, P = 0.15; Fig. 5F). These results suggest that stroke survivors who had higher amounts of explicit, strategic learning processes forgot more (i.e., did not retain as well); but, this is not the case in age-matched healthy adults.

As an additional analysis, we also examined the relationship between performance and immediate and 24-h retention in both groups. In stroke survivors and in age-matched subjects, there was no relationship between AE at late acquisition and FORGET_Imm (ρ = 0.07, P = 0.76; ρ = 0.15, P = 0.56, respectively) or between AE at late acquisition and FORGET_24h (ρ = −0.03, P = 0.90; ρ = 0.17, P = 0.51, respectively). This suggests that stroke survivors and healthy adults who learned the new walking pattern well did not necessarily retain the pattern well.

DISCUSSION

In this study, we sought to examine the contribution of implicit and explicit processes during a locomotor learning task and to understand how these underlying mechanisms impacted performance and retention in stroke survivors and age-matched healthy adults. We found that the amount of explicit, strategic processes was related to how well the newly learned walking pattern was retained both immediately and 24 h later in individuals after stroke. In addition, we found no relationship between the amount of explicit, strategic learning and performance, and no difference in performance or retention between stroke survivors and age-matched healthy adults.

Relationship Between Explicit, Strategic Processes and Retention

Our results suggest that stroke survivors who used a greater amount of explicit, strategic processes did not retain the walking pattern as well. It is possible that this is because explicit, strategic learning processes decay quicker than implicit processes; thus, individuals who used more explicit, strategic processes would have difficulty remembering the new walking pattern long term. This is supported by evidence from the dual-rate state space model, which suggests that explicit processes, represented by the fast phase in the dual-rate state space model, decays quickly (22, 53).

However, there are likely other factors that impact this relationship such as age and cognition. Trewartha et al. (54) suggested that forgetting of the fast learning phase (i.e., explicit, strategic learning) during a force field reaching task was impacted by age. This would suggest that the relationship between explicit processes and forgetting could be mediated by age. If this was the case, we would have expected he relationship to be similar in the group of stroke survivors and healthy adults, since they were matched on age. However, we only observed a relationship between the amount of explicit processes and forgetting in individuals with stroke, not in age-matched healthy adults. Trewartha et al. (54) also suggested that cognition, rather than age, may be important in understanding the relationship between explicit processes and retention. Similarly, cognitive abilities have been found to relate to retention of new, functional upper extremity tasks (55, 56). Thus, it is possible that the relationship between the use of explicit processes and retention is mediated by cognitive capacity. In other words, individuals who have higher cognitive capacity and use large amounts of explicit processes may retain better than individuals who use similar amounts of explicit processes but have lower cognitive capacity. Due to normal variability in cognitive capacity and potential changes to cognition due to stroke and age, we are unable to make a hypothesis regarding how cognitive capacity directly impacted our results. However, it is possible that cognition may serve as a mediator in the relationship between explicit processes and retention. Furthermore, age and cognition may both impact this relationship in complex ways; thus, future work examining the impact of cognition and age on this relationship would provide a more comprehensive understanding of the relationship between learning mechanisms and retention.

Unlike the relationship with retention, we did not find a significant relationship between the amount of explicit, strategic processes and performance during the task in either stroke survivors or age-matched healthy adults. In other words, individuals performed similarly despite the underlying learning processes. For example, two individuals may perform with the same amount of error, yet the processes used to accomplish the task may be very different (e.g., high or low contribution of explicit processes). This suggests that the strategic processes were flexibly combined with the ongoing implicit processes to successfully learn. This is consistent with upper extremity work, which has found that strategic processes are highly flexible and can be adjusted over time to accomplish the task at hand (19, 25, 51).

The discrepancy in the relationship between the amount of explicit processes and performance and retention is important to note because it suggests that underlying learning mechanisms have a greater implication for retention than for performance during visually guided locomotor learning. In addition, the analysis examining the relationship between performance and retention illustrates the importance of understanding the underlying learning mechanisms, as performance was not significantly related to immediate or 24-h retention. Because rehabilitation aims to improve long-term behavior, this work supports the need for mechanistic motor learning research to further understand how different learning processes impact retention of a newly learned locomotor task. This line of research will allow for the design of more effective rehabilitation interventions that impact long-term behavior.

Contributions of Implicit and Explicit, Strategic Processes

When quantifying the contribution of implicit and explicit, strategic processes during this task, we found that both implicit and explicit processes contributed to the acquisition of the new walking pattern. This was expected given that we have previously found that the behavioral change observed during DVF is contributed to by both implicit and explicit, strategic processes in young, healthy subjects (20, 24). Other work has also shown that implicit and explicit, strategic processes can occur in parallel during locomotion in stroke survivors (30) and healthy adults (2, 32, 33); however, this past work had utilized dual-task paradigms using a split-belt treadmill paradigm and visual feedback. In contrast, our results provide evidence that multiple learning processes can occur within a single locomotor task, as commonly occurs in everyday life, in individuals after stroke and age-matched healthy adults.

Interestingly, we found that there was no difference between the amount of explicit, strategic processes in stroke survivors and age-matched healthy adults, although there is significant variability in both stroke survivors and age-matched healthy adults. We hypothesized that explicit, strategic processes would be reduced in stroke survivors due to the high cognitive demands of this type of learning (12) and the prevalence of cognitive impairments after stroke (35–37). However, cognition also declines with age; thus, it is likely that subjects in the age-matched healthy adult group had normal age-related cognitive changes that impacted the use of explicit, strategic processes. Therefore, both of our groups may have had some level of cognitive changes, either due to age, stroke, or both, that impacted the use of explicit, strategic processes, and that may explain the significant variability observed in both groups. Several studies have examined the impact of cognitive capacity with upper extremity motor learning and found that cognitive capacity relates to motor learning in older (54, 57, 58) and younger adults (55, 59, 60). Few of these studies have examined explicit, strategic processes specifically. Christou et al. (60), however, examined the relationship between cognitive capacity and specific learning processes during an upper extremity visuomotor rotation task. Their work found that individuals with poorer visuospatial working memory used more implicit processes than individuals with higher visuospatial working memory (60). Thus, an individual’s cognitive capacity may influence the type of learning they use during specific tasks. Future work examining the role of cognition on the use of specific learning processes during this and other locomotor learning tasks would provide additional insight into the underlying mechanisms of locomotor learning and help deliver targeted motor learning-based interventions to older adults and individuals post stroke.

As we hypothesized, there was no difference in the amount of implicit processes used between groups, suggesting that implicit processes used in this task were intact in stroke survivors. Based on recent work from our laboratory, use-dependent plasticity is likely the primary contributor to the implicit process within this task despite the presence of a sensory prediction error due to the distortion of the visual feedback (24). Use-dependent plasticity is dependent on neural pathways that include the primary motor cortex (13, 16) and, therefore, could be impaired in individuals with cortical lesions. In addition, it has been suggested that age impacts use-dependent plasticity (61–65). If age does impact use-dependent plasticity, the magnitude of the implicit processes in both groups may be affected. As a result, further examination of the contribution of different learning processes in young healthy adults would provide a better understanding of the underlying implicit mechanisms occurring within this task.

Lastly, although not a direct question of this study, the amount of variability in the contribution of each learning process for both groups is notable. Although both groups used the same amounts of implicit and explicit strategic processes, some individuals used explicit processes exclusively, whereas others used implicit processes exclusively. This is particularly interesting given that we intentionally made the task as explicit as possible through the instructions provided. This suggests that there are factors other than the design of the task that impact how learning processes are used during a single task. Given the relationship between the amount of explicit processes and retention, especially in stroke survivors, it is critical to understand sources of this intersubject variability to deliver targeted rehabilitation interventions that will be effective for each specific patient. Because age and cognitive capacity may help explain this variability, work examining how these factors are related to the type of learning used is crucial for applying this work to rehabilitation practice. For example, identifying individuals who are likely to use one type of learning over another would provide rehabilitation professionals with insight into how well an individual may retain the locomotor pattern being trained. This would also allow rehabilitation professionals to select interventions that cater to the learning processes that specific patients are prone to use.

Limitations and Future Directions

Although this work provides valuable insight into the underlying learning processes occurring during a locomotor learning task with visual feedback, it is not without limitations. First, we calculated the contributions of implicit and explicit processes through a subtraction; thus, we are assuming that only implicit and explicit processes are contributing to the behavioral change that occurs over the course of acquisition1 and acquisition2. Although this assumption has been made previously (22, 23, 51), it is important to note that it is an assumption. The explicit portion of this calculation likely represents strategic aiming as well as the use of online visual feedback, whereas the implicit portion of this calculation cannot distinguish between different forms of implicit processes. As a result, future work to distinguish the unique role of each of these specific implicit and explicit mechanisms would be valuable. In addition, we did not test whether subjects were able to accurately step to a target by taking a longer step at baseline and so it is possible that some of the subjects were quite good at this task at the start, requiring limited learning. This, however, has not been found to be the case in a study using a similar paradigm in young, healthy subjects (24). Another limitation is that we used absolute value when examining performance and retention. Although this was necessary to accurately understand the group data (i.e., prevent group averages from being close to zero due to some individuals with positive values and others with negative values), it does not reflect the presence of over and under shooters in our data. Lastly, we only examined this paradigm over 2 days with 1 day of training. Previous work has found that repeated practice can make behaviors learned through explicit, strategic processes more automatic and less cognitively demanding (66, 67); thus, the relationships we found between explicit, strategic processes and retention may be different if multiple practice sessions occur.

CONCLUSIONS

This work provides evidence that the underlying learning processes were related to retention of a newly learned walking pattern even though performance was not. This relationship was particularly strong in individuals after stroke, potentially due to cognitive changes that often accompany stroke. In addition, the results highlight the significant intersubject variability in the type of learning processes used during the same task. Together this understanding of locomotor learning processes will allow for the development of rehabilitation interventions that optimize retention of the learned task and that are tailored to the specific individual being treated; however, future work is needed to understand how factors, such as cognition and age, and repetitive practice impact the use of implicit and explicit processes and how that impacts the retention of newly learned walking patterns.

GRANTS

This work was funded by the National Institutes of Health Grants 1R01-HD-078330-01A, S10RR028114-01, and F31NS111806 and The Foundation for Physical Therapy Research [2016 Florence P. Kendell Doctoral Scholarship, 2018–2019 Promotion for Doctoral Studies (PODS) Level I Award, and 2019–2020 Promotion for Doctoral Studies (PODS) Level II Award].

DISCLOSURES

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

AUTHOR CONTRIBUTIONS

M.A.F., S.M.M., and D.S.R. conceived and designed research; M.A.F. performed experiments; M.A.F. analyzed data; M.A.F., S.M.M., and D.S.R. interpreted results of experiments; M.A.F. prepared figures; M.A.F. and D.S.R. drafted manuscript; M.A.F., S.M.M., and D.S.R. edited and revised manuscript; M.A.F., S.M.M., and D.S.R. approved final version of manuscript.