Gait variability following abrupt removal of external stabilization decreases with practice in incomplete spinal cord injury but increases in non-impaired individuals

We have previously observed that when a laterally stabilizing external force field is abruptly removed during treadmill walking that individuals with and without incomplete spinal cord injury (iSCI) exhibit aftereffects of decreased step width that required 14 steps to return to steady state [1]. During this locomotor transition from a stabilizing environment to unassisted walking, individuals with iSCI also demonstrated a substantial increase in variability in center of mass (COM) speed and step width when compared to baseline performance [1]. For individuals with iSCI, this substantial locomotor variability during the transition may have posed a safety risk, and was thus likely a result of a temporary inability to control one’s mediolateral COM movement rather than a decision to not control variability. No change in locomotor variability during the transition was observed in non-impaired participants. This contrasting result suggests that in the absence of a wide base of support, non-impaired individuals were able to successfully utilize alternative methods to control mediolateral COM movement (e.g. potentially increased muscle co-activation).

The ability to rapidly and appropriately transition gait patterns to match changes in the external environment is an important skill that may improve with practice. Several studies have demonstrated savings – the phenomenon that previous learning speeds subsequent relearning – in a variety of motor tasks, including saccades [2], upper-limb reaching [3], and locomotion [4]. Several factors may enhance savings. In particular, abrupt changes in the external environment have been shown to result in more accurate perception and recall of the environment than gradual changes [4]. The positive effects of abrupt transitions on savings has been attributed to greater cognitive demand [5] and/or reliance on feedforward vs. feedback processes [4, 6]. Here we investigate the effects of practicing abrupt walking transitions from a highly stable environment to an unassisted environment in individuals with and without iSCI. We anticipate that the two groups will prioritize different control objectives during subsequent transitions based on the very different initial responses to this abrupt transition observed in our previous study [1].

We hypothesize that individuals with iSCI will reduce COM and step width variability with practice transitioning from a stabilizing environment to an unassisted environment. Reductions in variability with practice would suggest a prioritization of control performance. Such a prioritization can be observed in individuals with iSCI during steady-state walking conditions, where they exhibit persistent cautious gait patterns, including wide steps [7] and increased double-support time [8], that reflect a general strategy to improve their ability to resist and recover from perturbations [1]. These behaviors increase control performance but come with a significant associated metabolic [7, 9] and cognitive [10] cost.

In contrast, we hypothesize that non-impaired individuals will reduce muscle activity with practice transitioning from the stabilizing environment to an unassisted environment. As observed previously [1], the control methods used by non-impaired participants during a novel environment transition were successful for controlling mediolateral COM movement, but they may not have been efficient. Reductions in muscle activity with practice would suggest a prioritization of control effort. Franklin et al. have suggested a motor learning model in which the central nervous system first adapts control parameters to reduce error, and once errors are below a threshold, the nervous system will gradually reduce excessive muscle activity [11]. This motor “slacking” model in which people reduce control effort with practice has been observed in non-impaired individuals walking with a powered exoskeleton [12]. In accordance with Franklin’s model, we hypothesize that non-impaired individuals will prioritize reductions in control effort with practice because movement errors were substantially small during the locomotor transition from a stabilizing environment to unassisted walking [1, 11].

In support of our first hypothesis, with practice, all participants with iSCI, including participants who were highly challenged during Transition 1 (see Additional file 4), exhibited savings, demonstrating improved COM control performance during transitions between environments. Studies examining locomotor adaptation using a split-belt treadmill paradigm have also identified savings during transitions between walking environments [21, 22]. Several factors are thought to contribute to savings, including experience making abrupt changes between environments [4]. The savings observed in the current study may have been a result of error-based learning, which would have benefited from the relatively large movement errors [23] created by the abrupt changes in external stabilization. There is considerable evidence that the cerebellum plays an important role in the error-based learning that occurs with walking practice [24, 25]. Animal research also suggests that the lumber spinal cord in isolation can support locomotor adaptations to repeated external perturbations [26]. Participants in the current experiment had incomplete injuries, so it is possible that both spinal and supraspinal structures contributed to the observed improvements in locomotor transitions that occurred with practice.

In contrast to participants with iSCI, non-impaired participants did not show savings in control performance with repetitions of transitions. This is consistent with a previous experiment that also found no evidence of motor savings in non-impaired individuals following practice walking in a force field applied to the single limb [27]. The absence of savings in these experiments suggests that savings might only occur when people perceive sufficient challenge or risk. For learning to occur, it may not be enough that a kinematic error occurs, but rather that the central nervous system interprets the error as meaningful.

In support of our second hypothesis, non-impaired individuals adopted a “slacking” control strategy that reduces effort in response to repetitions of transitions from a stabilizing field to unassisted walking. This reduction in muscle activity during the transition was paralleled by decreased muscle activity in the stabilization force field with repeated exposures. Interestingly, the reduction in muscle activity with practice was accompanied by an unexpected increase in movement variability during the transition with practice. Franklin’s model of motor learning suggests that muscle activity should decrease only when movement errors are small [11]; movement variability levels likely remained subthreshold throughout all transition repetitions. That movement variability increased with practice may further suggest that non-impaired individuals were prioritizing control effort over control performance. This finding is consistent with past data suggesting that people can select more efficient walking patterns at the expense of stability [28, 29].

In addition, the greater time length of Transition 1 for non-impaired individuals (22 steps on average) vs. individuals with iSCI (9 steps on average) could also be explained by a greater tendency for non-impaired individuals to prioritize control effort at the expense of control performance (defined in terms of variability). While a more delayed reaction to a perturbation may result in less bounded behavior (less orbital stability), it may conserve energy and be beneficial to maintaining balance - the initial reaction to a perturbation may amplify the perturbation and a time delay may allow for more context-specific voluntary responses [30].

Hip abductor activity in participants with iSCI also showed a non-significant decreasing trend with repetitions of transitions. The lack of statistical significance could be attributed to the small number of participants with iSCI from whom clean EMG signals were collected, and, thus, this population may have also been trying to reduce control effort in parallel with efforts to improve control performance.

Surprisingly, there were large differences in balance ability within each group, which were sometimes larger than the average differences between groups, which complicated our statistical analysis. For example, a highly-challenged non-impaired participant (Fig. 6a) maintained narrow step widths in all three transitions and exhibited significant variability in COM motion during the first two transitions. This multi-step persistence of behavior during Transition 3 was not observed in participants with iSCI (Figs. 2 and 6b). Further investigation is needed to identify if these within-group differences in balance ability can be predicted a priori. Given the large within-group differences and lack of age/gender matching, we did not directly compare the two groups statistically, but we did perform an additional analysis to examine the effect of age across all participants (iSCI and non-impaired). With all participants combined, we identified an average transition period of 15 steps during Transition 1 (using the same exponential fitting method described earlier) and found a significant correlation between age and step width variability during this period (ρ = 0.414, p < 0.05) (Additional file 5). The correlation suggests that age may be an important factor independent from spinal cord injury for identifying individuals who may experience challenges in walking transitions from stabilizing to unassisted environments.

There were several limitations in the current study. First, the method of using the transition length calculated from Transition 1 for all three transitions likely impacted the results, but was the most robust method for dealing with high inter-subject and inter-trial variability in step width data. The transition data was highly variable across trials, especially for Transition 2 and 3, where the participant likely learned from Transition 1 and explored strategies of responding to the transition. Based on visual inspection of data, we found it to be a fair assumption that the longest period of transition occurred during Transition 1, so we used that same length of data to analyze all subsequent transitions for consistency. Another factor contributing to difficulties in fitting a model to all three transitions was that the experimental protocol assumed that all participants would return to steady-state walking kinematics within the allotted 100 steps of the second unassisted period after removal of the stabilizing field. However, as shown in the video for the example participant with iSCI (Additional file 4), some participants had such a strong aftereffect that they did not reach steady state during this period. Another limitation was that some participants were provided brief instances of physical support from experimenters for safety. Physical support likely improved COM control performance and decreased control effort. Although limited, this physical support may have reduced our estimates of savings in participants with iSCI. A final limitation of this study relates to assumptions about control effort and kinematic variability. While we instructed participants to try and center their midline over a target location during walking, individuals’ interpretation of this task and level of motivation may have affected their effort to control variability. One way to address this issue in future studies would be to design more constrained and challenging walking tasks with clear measures of and feedback on error. Such tasks would help identify if changes in movement variability are directly related to changes in ability vs. preference in movement control.

Ambulatory individuals with iSCI showed rapid and substantial improvements in their ability to reduce COM movement variability during a challenging transition from a stabilizing environment to unassisted walking. In contrast to participants with iSCI, non-impaired participants’ movement variability actually increased with practice, and was accompanied by reductions in muscle activation. The strategy observed in non-impaired participants possibly reflected a trade-off in control performance for lower control effort. Thus, control priorities of different populations may influence what aspect of performance change with practice of a locomotor balance task.