Corrective Muscle Activity Reveals Subject-Specific Sensorimotor Recalibration

Abstract

Recent studies suggest that planned and corrective actions are recalibrated during some forms of motor adaptation. However, corrective (also known as reactive) movements in human locomotion are thought to simply reflect sudden environmental changes independently from sensorimotor recalibration. Thus, we asked whether corrective responses can indicate the motor system's adapted state following prolonged exposure to a novel walking situation inducing sensorimotor adaptation. We recorded electromyographic (EMG) signals bilaterally on 15 leg muscles before, during, and after split-belts walking (i.e., novel walking situation), in which the legs move at different speeds. We exploited the rapid temporal dynamics of corrective responses upon unexpected speed transitions to isolate them from the overall motor output. We found that corrective muscle activity was structurally different following short versus long exposures to split-belts walking. Only after a long exposure, removal of the novel environment elicited corrective muscle patterns that matched those expected in response to a perturbation opposite to the one originally experienced. This indicated that individuals who recalibrated their motor system adopted split-belts environment as their new "normal" and transitioning back to the original walking environment causes subjects to react as if it was novel to them. Interestingly, this learning declined with age, but steady state modulation of muscle activity during split-belts walking did not, suggesting potentially different neural mechanisms underlying these motor patterns. Taken together, our results show that corrective motor commands reflect the adapted state of the motor system, which is less flexible as we age.

Keywords: EMG; locomotion; motor learning; split-belt walking.

Copyright © 2019 Iturralde and Torres-Oviedo.

Figures

Figure 1.

Summary of methods used in this study. A, schedule of belt speeds experienced by all subjects. B, middle column, sample EMG traces of one muscle (LG) during baseline (solid) and early long exposure (dashed) for a representative subject (S14). Median activity across strides (lines), and the 16th to 84th percentile range (shaded). Data in traces was processed as described in Materials and Methods and further lowpass filtered solely for visualization purposes. Color bars below the traces represent averaged normalized values during 12 kinematically-aligned phases of the gait cycle (two for DS, four for SINGLE, two for second DS, four for SWING; see Materials and Methods) for baseline and early adaptation (both gray), and the difference between the two (ΔEMGon (+), red indicates increase, blue decrease). The activity of each muscle is aligned to start at ipsilateral heel-strike. Top panels, Data for non-dominant/slow leg. Bottom panels, Dominant/fast leg. Left column, summary of muscle activity during baseline walking for all muscles. Median across subjects. Because of the alignment procedure, each column of muscle activity variables is synchronous for all muscles in the non-dominant (top panel) and dominant (bottom panel) legs separately, but not across legs. Right column, summary of change in muscle activity from baseline to early adaptation (ΔEMGon (+)). Red colors indicate higher levels of activity during early adaptation, while blue colors indicate lower values. Median across subjects. Black dots indicate significant differences from 0; p value threshold: p = 0.035.

Figure 2.

Schematic of the expected corrective responses following short and long exposures to split-belts walking. A, schematic of the input-output relation (system) of study. We consider belt-speeds (the walking environment) as the input to the system and motor commands (as measured by EMG activity) as its output. We specifically proved rapid EMG changes in response to sudden transitions in the walking environment. Hypothetical EMG patterns in response to an on (+) and on (–) transition are illustrated in yellow and purple, respectively. In the on (+) transition, the dominant leg walks unexpectedly faster than the non-dominant one and vice versa for the on (–) transition. B, we contrasted corrective responses on removal of the (+) environment following either a short (ΔEMGoff (+)short) or a long (ΔEMGoff (+)long) exposure duration. In the case of a short exposure (top), we expect muscle activity to return to the activation pattern before the introduction of the split-belts environment. In the case of a long exposure (bottom), corrective responses could either remain the same as those following a short exposure (i.e., non-adaptive, O1); or be adaptive, exhibiting a structure similar to that in the “off (–)” transition (O2). C, schematics of hypothetical patterns of activity in muscle space under the environment-based (O1) and adaptive (O2) alternative outcomes. We present a two-dimensional muscle space for illustration purposes, but we characterized muscle patterns in a 360-dimensional muscle space. A point in this space represents a pattern of activity across all muscles, whereas colored arrows represent changes in muscle activity from one activation pattern to another on an environmental transition. EMG changes over the course of adaptation and washout periods (gray) were not investigated. Under O1, ΔEMGoff (+)long (black) and ΔEMGon (+) (yellow) are expected to be numerically opposite. Under O2, ΔEMGoff (+)long is expected to be equal to ΔEMGon (−) (purple). D, schematic and equation of the regression model used to quantify the structure of ΔEMGoff (+)long. βadapt quantifies the similarity to responses expected under O2, whereas βno-adapt to those under O1.

Figure 3.

Steady-state muscle activity during slow walking, late split-belts walking, and aftereffects. Panels reflect differences in muscle activity the three epochs with respect to the reference (baseline) condition. Colormap reflects effect size and dots indicate FDR controlled significant differences (see Materials and Methods). Muscle activation variables were displayed starting with the ipsilateral heel-strike. A, muscle activity modulation during slow (tied-belts) walking. Most muscle-phases show reduction of activity, consistent with a monotonic link between walking speed and muscle activity amplitude; p value threshold: p = 0.022. B, muscle activity modulation during late adaptation. Broadly, patterns of activity are anti-symmetric, with groups of muscles increasing activity in one leg and decreasing contralaterally; p value threshold: p = 0.022. Differences between the slow leg’s activity at late long exposure (B, top) and slow (tied) baseline (A, top) illustrate that split-belts patterns do not match the expectation from simple ipsilateral speed modulation. C, muscle activity modulation during early washout with respect to baseline. Black dots indicate significance; p value threshold: p = 0.026. Few similarities are found between the steady-state activity during late long exposure and observed aftereffects. Black outlines indicate the muscle-intervals with activity changes of at least 10% and the same sign (i.e., increase or decrease with respect to baseline) for both late long exposure and early washout.

Figure 4.

Corrective responses were adapted following a long exposure to a split-belts walking environment. A, B, expected corrective responses elicited by the off transition under the environment-based (O1; A) and adaptive (O2; B) cases. Data (in color) and significance (black dots) were derived from the observed corrective responses on the introduction of the (+) walking environment (Fig. 1B, right column), by either taking the numerical opposite (O1) or by transposing leg activity (O2). For more details, see Figure 2. C, actual corrective muscle activity responses on removal of the (+) environment (i.e., off transition). Black dots indicate significant changes in activity following FDR correction; p value threshold: p = 0.035. D, quantification of corrective responses’ structure on removal of the (+) environment following long (black) and short (gray) exposure durations. As expected, responses following the short exposure displayed environment-based structure. In contrast, those following a long exposure appeared as if removal of the novel (+) environment was equivalent to introducing a novel (–) environment. E, changes in anterior-posterior hip position (with respect to stance foot) following the long exposure (black). Expected changes under O1 (yellow) and O2 (purple) are also illustrated. These were computed in the same way as EMG factors displayed in A, B. The similarity across all traces indicates that it is impossible to characterize the adaptive and environment-based (non-adaptive) nature of corrective responses solely from a global measure of body position.

Figure 5.

Age modulates EMG-based learning measures but not kinematic ones. Single dots represent values for one subject. Spearman’s correlation coefficients (ρ) and p values are presented on the legend. The best line fit of the dependent variable onto age is only displayed when significant (p ≤ 0.05). A, regression coefficients from model quantifying the structure of EMG corrective responses (purple, βadapt; yellow, βno-adapt). Both regression coefficients were significantly correlated with age. B, magnitude of corrective activity following the introduction (‖ΔEMGon​(+)‖) and removal (‖ΔEMGoff​(+)long‖) of the (+) environment. We observed a significant effect of age at both transitions. C, magnitude of steady-state changes in muscle activity during late long exposure. No correlation to age was found. This confirms that older subjects were able to modulate muscle activity as much as healthy subjects. D, step-length asymmetry aftereffects were also not correlated with age. E, size of muscle activity modulation during early washout (aftereffects) were correlated with age. This suggests EMG-based measures of learning were more sensitive than kinematic-based ones.