Humans with spinal cord injury (SCI) modulate locomotor output in response to limb load. Understanding the neural control mechanisms responsible for locomotor adaptation could provide a framework for selecting effective interventions. We quantified feedback and feedforward locomotor adaptations to limb load modulations in people with incomplete SCI. While subjects airstepped (stepping performed with kinematic assistance and 100% bodyweight support), a powered-orthosis created a dorisflexor torque during the “stance phase” of select steps producing highly controlled ankle-load perturbations. When given repetitive, stance phase ankle-load, the increase in hip extension work, 0.27 J/kg above baseline (no ankle-load airstepping), was greater than the response to ankle-load applied during a single step, 0.14 J/kg (P = 0.029). This finding suggests that, at the hip, subjects produced both feedforward and feedback locomotor modulations. We estimate that, at the hip, the locomotor response to repetitive ankle-load was modulated almost equally by ongoing feedback and feedforward adaptations. The majority of subjects also showed after-effects in hip kinetic patterns that lasted 3 min in response to repetitive loading, providing additional evidence of feedforward locomotor adaptations. The magnitude of the after-effect was proportional to the response to repetitive ankle-foot load (R2 = 0.92). In contrast, increases in soleus EMG amplitude were not different during repetitive and single-step ankle-load exposure, suggesting that ankle locomotor modulations were predominately feedback-based. Although subjects made both feedback and feedforward locomotor adaptations to changes in ankle-load, between-subject variations suggest that walking function may be related to the ability to make feedforward adaptations.
Although people with incomplete spinal cord injury (SCI) can significantly improve walking ability with intense locomotor practice (Behrman and Harkema 2000; Dietz et al. 1995; Wernig and Muller 1992; Wernig et al. 1995, 1998), the factors influencing training outcomes are debated (Barbeau et al. 1998, 2006a; Dietz 2006; Hidler 2005; Nooijen et al. 2009; Wernig 2005, 2006; Wolpaw 2006). Developing an accurate model of the variables affecting locomotor learning could provide a conceptual framework for selecting clinical interventions that most effectively improve walking function. However, identifying the ongoing factors that influence learning is difficult because walking outcomes are measured on large time scales. For example, on a step-to-step basis do small modulations in limb load or gait speed affect the acquisition of locomotor learning occurring over weeks and years?
One way to gain insight into the motor learning process is by studying locomotor adaptations occurring on the scale of minutes. It is suggested that short-term adaptations are the basis for learning permanent motor patterns (Bastian 2008; Reisman et al. 2010; Shadmehr and Wise 2005) and as such can serve as a platform for examining the role isolated factors play in locomotor learning. In this study, we examine whether individuals with incomplete SCI make feedforward locomotor adaptations to changes in ankle-foot load. The ability to modulate feedforward locomotor output can improve gait efficiency (Verdaasdonk et al. 2009) by minimizing sensory-motor delays (Kawato 1999; Shimansky 2000) and compensating for imperfect and/or incomplete sensory feedback (Kuo 2002; Shadmehr and Wise 2005). Thus the nervous system's capability to adapt locomotor patterns allows it the flexibility to effectively control ambulation in a multitude of environments (Bastian 2008). For rehabilitation purposes, the presence of feedforward locomotor adaptations indicates that the nervous system has temporarily recalibrated motor output. Although brief, these short-term adaptations engage neural pathways that with repeated exposure have the potential to produce clinically significant performance changes by training the nervous system to effectively make centrally mediated modifications in locomotor output.
Mechanoreceptors stimulated during the weight-bearing phase of locomotion, in particular load-sensitive Golgi tendon organs (Donelan and Pearson 2004a; Grey et al. 2004) and length-sensitive secondary muscle spindles (Grey et al. 2001) of the ankle extensor muscles, play a critical role in sculpting ongoing locomotor output (for review, see Dietz and Duysens 2000; Donelan and Pearson 2004b). People with SCI show a positive relationship between limb load magnitude and stance phase, lower limb extensor muscle activity (Dietz et al. 2002; Gordon et al. 2009; Harkema et al. 1997). In humans, this load response has been identified during steady-state stepping (i.e., locomotor output is assessed while subjects step with a constant limb load). Although steady-state experiments are valuable for studying locomotor adaptation, they are not designed to evaluate the individual contributions of feedback and feedforward control mechanisms.
To quantify feedback and feedforward contributions to load-mediated locomotor adaptations, dynamic changes in efferent output must be examined. For example, the dynamic response to limb load (e.g., immediate EMG changes to unanticipated limb load modulations during stepping; Hiebert and Pearson 1999) has been used to identify the ongoing role sensory feedback plays in modulating locomotor output. Research suggests that potentially one half of the extensor muscle activity during gait is mediated directly by limb-load afferent feedback (Donelan and Pearson 2004a; Hiebert and Pearson 1999). To date, human experiments attempting to quantify the contributions of short-term (within minutes) feedforward locomotor adaptations to changes in limb load magnitude have not been performed. In humans, changes to feedforward locomotor output have previously been quantified by examining after-effects, the persistent production of motor patterns appropriate for a previous state (Choi and Bastian 2007; Earhart et al. 2002b; Emken and Reinkensmeyer 2005; Gordon and Ferris 2007; Gordon et al. 1995; Lam et al. 2006; Noel et al. 2009; Reisman et al. 2005; Weber et al. 1998). In this study, we use this methodology to examine human feedforward locomotor adaptations to limb-load modulations.
In people with incomplete SCI, it is not clear if the locomotor response to stance phase, ankle-foot load is mediated strictly through ongoing feedback mechanisms or if the response is controlled through a combination of feedback and feedforward adaptive systems. Consequently, our purpose was to identify the contribution that both feedback and feedforward mechanisms play in modulating locomotor output in response to changes in ankle-foot load. Previous research suggests that the human spinal cord can make ongoing feedback-driven adaptations to locomotor output (Field-Fote and Dietz 2007) but that feedforward adaptations require supraspinal input (Morton and Bastian 2006). Given a loss of descending supraspinal input, we hypothesized that ambulatory individuals with incomplete SCI would be capable of making feedback modulated responses to changes in limb load but would have limited capacity to make feedforward locomotor adaptations. An alternative hypothesis is that ambulatory individuals with incomplete SCI have sufficient sparing of ascending and descending spinal pathways to effectively perform both feedback and feedforward locomotor adaptations in response to modulations in limb load.
To test these alternative hypotheses, we measured the locomotor responses to both single and multiple step changes in ankle-foot load during stepping. Our approach consisted of an airstepping paradigm with the controlled addition of ankle-foot load using a powered orthosis (Fig. 1). Previously, we used this experimental set-up to examine the locomotor response in lower limb muscle activity and hip joint torque to different steady-state ankle-foot loading conditions in individuals with complete and incomplete SCI (Gordon et al. 2009). In this study, we investigate the locomotor response to dynamic changes in ankle-foot load to identify the individual contributions of feedback and feedforward adaptive mechanisms. To quantify the feedback response and to probe underlying feedforward locomotor output, we examined immediate changes in locomotor patterns (hip joint torque and soleus EMG during stepping) when subjects experienced catch trials during which ankle-foot load was unexpectedly altered for a single step. In addition, to further examine feedforward adaptations we quantified after-effects in locomotor patterns when ankle-foot load was removed entirely after a conditioning period of repetitive ankle-foot loading.
Ten incomplete SCI subjects (40 ± 10.4 yr of age, 79.8 ± 16.8 kg, 1.78 ± 0.09 m, 2 females; Table 1) gave written informed consent and participated in the study. The Northwestern University Institutional Review Board approved the protocol. All subjects had a spinal cord lesion occurring between C1 and C7 caused by a nonprogressive etiology, were >1 yr after injury, were classified as clinically incomplete (American Spinal Injury Association Impairment Scale C or D; Ditunno et al. 1994), and had the ability to perform overground ambulation with the use of assistive devices. Exclusion criteria included concurrent severe medical illness, unable to tolerate 30 min of standing without orthostasis, or history of lower-limb peripheral nerve injury, traumatic head injury, cardiovascular/pulmonary complications, or metabolic (endocrine, hepatic)/renal dysfunction. Subjects did not alter medications for this study. One subject was prescribed antispasticity medications (baclofen, 20 mg/day) to reduce the intensity and frequency of spasms.
We constructed bilateral ankle-foot loading devices to provide clear and overt sensory stimuli to the mechanoreceptors of the ankle and foot (i.e., group I and II ankle plantar flexor muscle afferents and cutaneous pressure receptors on the sole of the feet) during stepping (Fig. 1) (Gordon et al. 2009). Each 4.3-kg loading device consisted of a powered, ankle-foot orthosis that allowed free sagittal plane rotation about the ankle joint. The orthosis was rigidly attached to the distal end of a commercial robotic device (Lokomat, Hocoma, Zurich, Switzerland) used to assist walking. A pneumatic cylinder was used to create a dorsiflexor torque about the ankle joint. The air pressure sent to the cylinder was adjusted for each subject to provide a peak torque during stepping of ∼0.6 Nm/kg. This load is similar in magnitude to the peak ankle moment experienced during slow walking, 0.55 m/s (Stoquart et al. 2008), and approximately one half of the peak moment occurring at normal walking speed, 1.6 m/s (Eng and Winter 1995). The applied load was controlled in an “all or none” manner by a solenoid value regulating air flow to the pneumatic cylinder. The solenoid value was regulated by a laptop computer equipped with a data acquisition card (National Instruments, Austin, TX) running custom control software (National Instruments). The dynamics of the ankle joint emerged from the applied dorsiflexor torque, the subjects' lower-leg muscular activation, and the system mechanics (i.e., gravity and inertia). Thus the device controlled the magnitude of the applied ankle torque and not ankle kinematics. When the solenoid valve was opened to release air pressure, the rapid expulsion of air from the cylinder created a brief auditory hiss. Pressurizing the cylinder did not provide auditory cues detectable over the background noise of the Lokomat actuators and treadmill. Pressurized air was delivered via copper pipe at 300 psi from a source external to the laboratory (a.k.a. shop air). This setup allowed pressurization of the pneumatic cylinders to occur in a relatively quiet environment.
A driven gait orthosis, Lokomat (Hocoma), provided assistance to subjects during stepping and produced a consistent kinematic gait pattern. DC motors aligned at each hip and knee joint moved the legs through a controlled sagittal plane trajectory. Subjects wore a harness attached to an overhead lift that supported 100% of the subjects' body weight during airstepping. Using a Lokomat provided several benefits. First, the Lokomat delivered consistent and measurable levels of assistance. Variability in assistance could alter sensory feedback (cutaneous at the robotic/subject interface and proprioceptive if gait kinematics change) and obscure results. Second, we instrumented the Lokomat with load cells (JR3, Woodland, CA) to calculate joint torques during stepping (Hidler 2004). Finally, the Lokomat motors and scaffolding supported the ankle-foot loading devices, meaning subjects did not have to adjust their stepping patterns to accommodate for the added mass of the device.
We recorded lower limb joint kinematics, kinetics, and EMG activity during stepping. Hip and knee joint kinematics were measured using the joint position sensors of the Lokomat. Sagittal plane motion about each ankle joint was measured with a potentiometer rigidly attached to the “ankle” joint of the ankle-foot loading device. The subjects' thigh and shank kinetics were measured directly from 6-df load cells (JR3) attached in series to all six standard leg attachment cuffs of the Lokomat. EMG signals were recorded from the soleus muscles (Motion Lab Systems, Baton Rouge, LA) using preamplifed surface EMG electrodes. EMG signals were amplified (×10,000) and filtered (band-pass at 10–500 Hz) in hardware. All analog signals were sampled at 1,000 Hz using a data acquisition card (National Instruments) on a PC running custom Matlab software (The Mathworks, Natick, MA).
Subjects were secured within the robotic assistive orthoses (Lokomat and bilateral ankle-foot loading devices), given 100% body weight support and elevated ∼25 cm above the treadmill surface. First, we recorded 60 s of airstepping data at 0.55 m/s. During this initial bout, subjects were encouraged to completely relax their lower limbs and allow the Lokomat to fully actuate their hip and knee joint movements. As described in detail later, this passive airstepping condition was used to estimate the contribution of nontissue components (i.e., inertia and gravity) and passive tissue (i.e., muscle and tendon) to net joint torques. After the passive bout, subjects were given a 2 min rest.
Next, we recorded data during three consecutive periods of airstepping: BASELINE, CONDITIONING, and WASH-OUT (Fig. 2). Each period consisted of four min of airstepping at 0.55 m/s. Subjects were verbally encouraged to actively move their legs with the Lokomat. Subjects received 2 min of rest between the BASELINE and CONDITIONING periods. No rest was provided during the remainder of testing. During BASELINE (except for 4 catch trials as described later) and WASH-OUT, no dorsiflexor torque was applied using the ankle-foot loading devices (Fig. 3). Thus subjects' feet and ankles were allowed unrestricted movement in the sagittal plane that was not resisted by the pneumatic cylinder; however, the foot was strapped to the loading device which effectively changed the mass of the foot and the effects of gravity and inertial accelerations compared with stepping without the device. The added mass from the ankle-foot loading device was consistent across all conditions. During the CONDITIONING period, dorsiflexor torque was applied to the ankle-foot bilaterally (1.16-s duration) during the period of the gait cycle that corresponds approximately to the stance phase of walking (Fig. 3). For the remainder of the gait cycle (i.e., swing phase), the ankle-foot loading device was disengaged, allowing unrestricted sagittal plane ankle movement. During each step of the CONDITIONING period, real-time hip position data triggered the onset of the applied dorsiflexor torque. After the WASH-OUT period, subjects rested for 2 min and performed another 60 s of passive airstepping.
During BASELINE and CONDITIONING, catch trials were performed at the 30 s mark of each minute of stepping, resulting in a total of 4 catch trials/period (Fig. 2). LOAD catch trials during BASELINE consisted of a unilateral, ankle-foot load applied to the right lower limb during the stance phase of a single step. During the CONDITIONING period, the NO-LOAD catch trials consisted of no ankle-foot load applied to the right lower limb during the stance phase of a single step.
We examined data from the right lower limb (the leg subjected to catch trials). EMG data were high-pass filtered (cut-off frequency: 20 Hz) to remove motion artifact (Basmajian and De Luca 1985) and rectified in software. Root mean square (RMS) EMG values were calculated for each subject during the stance phase of each gait cycle from the filtered, rectified EMG signals. For illustrative purposes only, RMS EMG values were normalized to the average RMS EMG value of the first 10 steps occurring during the BASELINE condition. We normalized to the first 10 steps of BASELINE because these steps occurred before any catch trials. However, all statistical tests were performed on non-normalized EMG. All kinematic and kinetic data were smoothed using a fourth-order Butterworth low-pass filter (cut-off frequency: 7 Hz) with zero lag. We calculated the applied ankle dorsiflexor torque created by the ankle-foot loading device by multiplying force measured from a tension-compression load cell (Transducer Techniques, Temecula, CA) attached in series with the pneumatic cylinder by the moment arm of the cylinder about the ankle joint.
We calculated the subject's active contribution to sagittal plane hip joint torques during airstepping as previously described (Hidler 2004; Hidler and Neckel 2006). Three-dimensional forces recorded from the load cell attached in series between the Lokomat thigh cuff and the Lokomat upper leg were multiplied by their respective moment arm distance about the hip joint and summed with the load cell torque measurements to calculate hip joint torque during stepping. For each subject, we created an average hip joint torque profile for a complete gait cycle using data from ∼20 steps recorded during the two passive airstepping trials. We assumed that during passive airstepping, hip joint torque was created entirely by passive tissue (i.e., muscle and tendon) and nontissue components (i.e., inertia and gravity) and that the contributions from these components were consistent between steps because the Lokomat constrained kinematic trajectories. Thus we estimated the subjects' active hip moments (i.e., muscular contribution) by subtracting the mean passive airstepping hip torque profile from the hip torque profile recorded for each individual gait cycle occurring during all other airstepping conditions (Hidler 2004). Henceforth, when we refer to hip moments, we are referring only to the active component.
We separated each gait cycle into extensor and flexor hip moment regions. For the hip extensor moment regions, we created corresponding hip moment-angle plots. Calculating the area of the positive and negative regions of the moment-angle plots yielded positive and negative work, respectively. We calculated total hip extensor work by summing the absolute value of the positive and negative work performed.
We quantified ankle kinematics and applied dorsiflexion torque between conditions and subjects. We compared changes in total ankle range of motion and peak ankle dorsiflexion between the last 10 steps of the BASELINE period, the first 10 steps of the CONDITIONING period, and the first 10 steps of the WASH-OUT period using a repeated-measures ANOVA to look for differences between conditions and subjects. Because of instrument failure, two subjects did not have complete ankle angle data for all three periods, so their data were excluded from this analysis. Next, we used a repeated-measures ANOVA to look for differences between amplitude of applied dorsiflexion torque during the first and last 10 steps of the CONDITIONING period (BASELINE and WASH-OUT were not included because dorsiflexor torque was not applied during these periods). Finally, we compared the rate of applied dorsiflexor torque between the first and last 10 steps of the CONDITIONING period and between subjects using a repeated-measures ANOVA. The significance level was set at α = 0.05 for all ANOVA tests. When appropriate, a Bonferroni/Dunn test with a family-wise error rate of α = 0.05 was used to check for differences.
We selected two main outcome variables as representative of motor performance: total hip extensor work and soleus stance phase RMS EMG. Results from the study confirmed these two parameters were sensitive to the ankle load perturbations, but it is important to explain the rationale for selecting these parameters a priori. We chose to examine the hip and ankle joints because these joints have been shown to generate the majority of sagittal plane positive work performed by the lower limb joints during gait (Eng and Winter 1995). In incomplete SCI subjects, we have found the major response to stance phase, ankle-foot load is lower limb extension (Gordon et al. 2009). Thus we selected measures that target hip and ankle extension.
At the hip joint, we examined total hip extensor work for two reasons. First, total hip extensor work gave a quantitative representation of the subjects' hip kinetic profile over the entire extensor phase of the gait cycle. Second, total hip extensor work is a function of the net activity of all muscles crossing the hip joint. Although this is an indirect measure of muscle activity, we have previously shown that incomplete SCI subjects increase both total hip extensor work and gluteus maximus EMG activation with stance phase ankle-foot load (Gordon et al. 2009).
At the ankle, we examined stance phase soleus RMS EMG. Unlike the hip, we did not analyze ankle joint torques because the applied ankle-foot loads directly influenced this measure. The soleus muscle, in particular, was valuable for assessment of plantar flexor activity because it is a single joint muscle. Examining stance phase RMS EMG activity provided a measure of overall recruitment of the soleus motoneurons during the period of the gait cycle when humans typically create a plantar flexor torque (Perry 1992).
To quantify the role feedback mechanisms play in modulating locomotor output and to probe underlying feedforward output, the response to unexpected changes in ankle-foot load was examined using catch trials. For each subject, we calculated the mean total hip extensor work and mean stance phase soleus RMS EMG from the four LOAD catch trial steps occurring during the BASELINE period and the four NO-LOAD catch trials occurring during the CONDITIONING period. We also calculated the mean response for the four steps immediately preceding and the four steps immediately following each catch trial. For each variable, total hip extensor work and soleus RMS EMG, we ran a repeated-measures ANOVA with the significance level set at α = 0.05 to look for differences between the mean catch trial step and the steps immediately preceding and following. When appropriate, a Bonferroni/Dunn test with a family-wise error rate of α = 0.05 was used to check for differences between the preceding step (this value represented steady-state performance before the catch trial perturbation) and both the catch trial step and the step immediately following the catch trial.
To quantify the role feedforward mechanisms play in modulating locomotor output, the response to lasting changes in ankle-foot load was examined using after-effects. The magnitude of after-effects was quantified as the difference in total hip extensor work and stance phase soleus RMS EMG between the two no-load periods: BASELINE and WASH-OUT. For each subject, we calculated mean values for both variables occurring during the ∼25 steps of the final minute of the BASELINE period and for all 4 individual min of the WASH-OUT period. All catch trials and steps immediately following the catch trials were excluded from these calculations. For each variable, we ran a repeated-measures ANOVA to look for differences between the final minute of BASELINE and each of the 4 min of the WASH-OUT period with significance set at α = 0.05. A Bonferroni/Dunn test with a family-wise error rate of α = 0.05 was used to check for differences where appropriate.
Examining differences in locomotor patterns during steps with similar ankle-foot loading profiles that occurred during different airstepping periods provided a second method of quantifying feedforward locomotor adaptations. To do so, we examined performance during the no ankle-foot load steps occurring during BASELINE and the NO-LOAD catch trials steps that occurred during the CONDITIONING period. We calculated the mean values of total hip extensor work and stance phase, soleus RMS EMG occurring during the ∼100 steps of the entire BASELINE period. We excluded all catch trials and steps immediately following the catch trials from this calculation. As described earlier, we also calculated mean values for the four NO-LOAD catch trial steps occurring during the CONDITIONING period. We ran paired, one tailed, t-tests with significance set at α = 0.05 to compare both total hip extensor work and soleus RMS EMG between the BASELINE period and the NO-LOAD catch trials occurring during the CONDITIONING period.
Similarly, we also examined locomotor output during the ankle-foot load steps occurring during the CONDITIONING period with the LOAD catch trial steps occurring during BASELINE. We calculated the mean value for total hip extensor work and stance phase, soleus RMS EMG occurring during the ∼100 steps of the entire CONDITIONING period. We excluded all catch trials and steps immediately following the catch trials from this calculation. The statistical analysis was similar to that used for the unloaded conditions.
Ankle-foot load intervention
Subjects' ankle range of motion (ROM) was 25 ± 11° (mean ± SD) during passive stepping, 28 ± 13° during BASELINE, 35 ± 13° during CONDITIONING, and 31 ± 12° during WASH-OUT (Fig. 3). Significant differences were found in ankle ROM between the three active stepping periods: BASELINE, CONDITIONING, and WASH-OUT (ANOVA: F(7,2) = 5.51, P = 0.017). Post hoc testing showed that ankle ROM during CONDITIONING was significantly greater than BASELINE (Bonferroni/Dunn: P = 0.005). No differences in ankle ROM were found between BASELINE and WASH-OUT (Bonferroni/Dunn: P = 0.095). During CONDITIONING, subjects had increased ankle dorsiflexion during the period of applied load. Peak dorsiflexion angle during CONDITIONING was 81 ± 11° (90° = neutral, smaller angles indicate greater dorsiflexion) compared with BASELINE (104 ± 9°) and WASH-OUT (97 ± 8°). Peak dorsiflexion angle was significantly different between the three active stepping periods (ANOVA: F(7,2) = 14.56, P = 0.0004). Post hoc tests showed that subjects had greater dorsiflexion during CONDITIONING than during BASELINE (Bonferroni/Dunn: P = 0.0001) or WASH-OUT (Bonferroni/Dunn: P = 0.0026). No significant differences in dorsiflexion were found between BASELINE and WASH-OUT (Bonferroni/Dunn: P = 0.131). As well, no significant differences in peak dorsiflexion were found between subjects during BASELINE, CONDITIONING, and WASH-OUT (ANOVA: F(2,7) = 0.52, P = 0.809).
Peak applied dorsiflexor torque was not significantly different between the first 10 (0.65 ± 0.09 Nm/kg) and last 10 steps (0.66 ± 0.11 Nm/kg) of CONDITIONING (ANOVA: F(9,1) = 1.21, P = 0.301). The average loading rate during the first 10 steps of CONDITIONING was 151 ± 41 Nm/s during 222 ± 102 ms. This rate was not significantly different from during the last 10 steps of the CONDITIONING period when the loading rate was 154 ± 46 Nm/s during 202 ± 95 ms and was not significantly different between subjects (ANOVA: F(9,1) = 0.99, P = 0.346).
Subjects' ankle kinematics were similar during BASELINE and WASH-OUT periods and were similar between subjects. As expected, subjects increased stance phase ankle dorsiflexion when a dorsiflexor torque was applied to the ankle. The peak torque and loading rate of the applied ankle-foot load was consistent across the CONDITIONING period and between subjects.
Feedback response to unexpected load
Subjects had a substantial within-step increase in hip extension moments in response to the unexpected LOAD catch trials occurring during BASELINE (Fig. 4A). For the group, in the perturbed leg, we observed significant differences in total hip extension work between the LOAD catch trial steps and the no-load BASELINE steps immediately preceding and following (ANOVA: F(9,2) = 20.17, P < 0.0001; Fig. 5A). Post hoc tests confirmed that total hip extension work performed during the LOAD catch trials (0.35 ± 0.18 J/kg) was significantly greater than during the preceding BASELINE step (0.21 ± 0.14 J/kg; Bonferroni/Dunn: P < 0.0001). There were no significant differences in total hip extension work between the BASELINE steps immediately preceding and following (0.21 ± 0.14 J/kg) the LOAD catch trials (Bonferroni/Dunn: P = 0.9605).
ANOVA results also indicated significant differences in stance phase soleus EMG RMS amplitude between the LOAD catch trials and the no-load BASELINE steps immediately preceding and following (F(9,2) = 14.12, P = 0.0002; Figs. 4C and 5C). Post hoc tests confirmed observations that soleus EMG amplitude occurring during the LOAD catch trials (8.2 ± 11.0, normalized value) was significantly greater than the preceding BASELINE step (1.2 ± 0.3, normalized value; Bonferroni/Dunn: P = 0.0.0002). No significant differences in soleus EMG RMS amplitude between the BASELINE steps preceding and following (1.2 ± 0.3, normalized value) the catch trial were found (Bonferroni/Dunn: P = 0.996). In the majority of the subjects (7 of 10), the applied dorsiflexor torque elicited large motor unit bursts in the soleus (Fig. 4, C and D).
Overall, subjects showed a significant within-step increase in hip extension work and soleus EMG amplitude when they received an unexpected, single step, stance phase, ankle dorsiflexor torque applied that occurred during a bout of no-load airstepping. These increases in hip extension work and soleus EMG did not persist during the step immediately following the isolated, single step load application.
Feedback response to loss of ankle-foot load
Subjects also had a substantial within-step decrease in hip extension moments in response to the unexpected absence of ankle-foot load during NO-LOAD catch trials occurring within the CONDITIONING period (Fig. 4B). For the group, in the perturbed leg, significant differences were observed in total hip extension work between the NO-LOAD catch trials and the CONDITIONING steps immediately preceding and following the catch trials (ANOVA: F(9,2) = 7.58, P = 0.004; Fig. 5B). Total hip extension work performed during the NO-LOAD catch trials (0.33 ± 0.24 J/kg) was significantly less than during the preceding CONDITIONING step (0.50 ± 0.29 J/kg; Bonferroni/Dunn: P = 0.0001). No significant differences were found between the preceding and following (0.47 ± 0.28 J/kg) CONDITIONING steps (Bonferroni/Dunn: P = 0.579).
In addition, substantial within-step decreases in soleus EMG amplitude were observed during unexpected absence of ankle-foot load occurring during the NO-LOAD catch trials within the CONDITIONING period (Fig. 4D). Subjects showed significant differences in stance phase soleus EMG amplitude between the NO-LOAD catch trials and the CONDITIONING steps immediately preceding and following (ANOVA: F(9,2) = 36.761, P < 0.0001; Fig. 5D). Post hoc tests confirmed observations that soleus EMG amplitude occurring during the NO-LOAD catch trials (1.2 ± 0.4, normalized value) were significantly less than the preceding CONDITIONING step (l6.8 ± 6.0, normalized value; Bonferroni/Dunn: P < 0.0001). Again no significant differences between the preceding and following (7.4 ± 7.3, normalized value) CONDITIONING steps were observed (Bonferroni/Dunn: P = 0.743).
Overall, subjects showed a significant within-step decrease in hip extension work and soleus EMG amplitude when they experienced an unexpected, single step with no applied ankle dorsiflexor torque that occurred during a bout of airstepping with stance phase ankle dorsiflexor torque normally applied every step. These decreases in hip extension work and soleus EMG did not persist during the step immediately following the isolated, no-load step.
Feedforward locomotor adaptations: after-effects
There were no significant differences in the group data between total hip extensor work performed during the final minute of BASELINE (minute 4 BASELINE: 0.22 ± 0.16 J/kg) and any individual minutes of WASH-OUT (i.e., minute 1 WASH-OUT: 0.26 ± 0.18 J/kg; ANOVA: F(9,4) = 0.98, P = 0.43). However, examination of the individual hip kinetic data suggested that three subjects performed in a manner inconsistent with the other subjects. One subject (subject 5) did not show reactive increases in hip extensor moments when given ankle-foot load during CONDITIONING compared with BASELINE (Table 2). This null response differed from the other nine subjects, all of whom increased hip extension moments when receiving ankle-foot load. Two additional subjects (subjects 8 and 9) stood out from the group when we examined hip kinetic after-effects. To examine differences in after-effects for all subjects, we normalized changes in total hip extensor work to BASELINE performance. We plotted the normalized changes in total hip extensor work during CONDITIONING against the normalized changes in total hip extensor work during WASH-OUT (Fig. 6). We ran a linear regression on this data. Two subjects had standardized residual values, ⊻RStudent⊻, >2, indicating their data were outliers (RStudent = −2.90 and −5.53). These two subjects produced the least hip extension moments during BASELINE stepping of the entire subject pool (Table 2). The remaining subjects (n = 8) showed a strong linear relationship between the observed changes in hip extensor work during the CONDITIONING and WASH-OUT periods (R2 = 0.92).
Based on the examination of individual data, we classified seven subjects as being both responsive to ankle-foot load (shown by increases >5% in total hip extension work when receiving ankle-foot load) and able to produce total hip extension work >0.05 J/kg during BASELINE stepping (Table 2). In subjects meeting these performance criteria (n = 7), there were significant differences in total hip extensor work performed during the final minute of BASELINE and the 4 individual min of WASH-OUT (ANOVA: F(6,4) = 4.492, P = 0.0075; Fig. 7A). Post hoc tests found that total hip extensor work was significantly higher during minutes 1 (0.28 ± 0.16 J/kg; Bonferroni/Dunn: P = 0.0031), 2 (0.30 ± 0.17 J/kg; Bonferroni/Dunn: P = 0.0009), and 3 (0.28 ± 0.18 J/kg; Bonferroni/Dunn: P = 0.0055) of the WASH-OUT period compared with minute 4 of BASELINE (minute 4 BASELINE: 0.20 ± 0.14 J/kg).
There were no significant differences in stance phase soleus RMS EMG occurring during the final minute of the BASELINE condition (minute 4 BASELINE: 1.23 ± 0.40, normalized) and any minute of the WASH-OUT period (i.e., minute 1 WASH-OUT: 1.51 ± 1.31, normalized; ANOVA: F(9,4) = 0.71, P = 0.591; Fig. 7B). Unlike the response in hip extensor moments, an examination of the individual soleus EMG data did not show any outliers (Table 2; Fig. 6B). However, removing the three subjects who did not meet the performance criteria listed previously for hip kinetic performance did not change the results (ANOVA: F(6,4) = 0.15, P = 0.963).
Immediately after a 4-min bout of airstepping with stance phase, ankle-foot load applied every step, 7 of 10 subjects showed a lasting increase in hip extension work when the ankle-foot load was removed. This increased hip extension work lasted for 3 min. The amplitude of the hip extension work after-effect was proportional to the amplitude of hip extension work subjects showed during the CONIDITIONING period. Two subjects who did not show hip extensor work after-effects had the greatest clinically assessed locomotor impairments. None of the subjects showed after-effects in soleus EMG amplitude after the CONDITIONING period.
Feedforward locomotor adaptations: catch trials
Subjects took steps with no ankle-foot load throughout BASELINE and during the NO-LOAD catch trials of the CONDITIONING period. During no ankle-foot load steps, total hip extensor work was significantly greater during the NO-LOAD catch trials (0.33 ± 0.24 J/kg) than during BASELINE (0.21 ± 0.14 J/kg; paired t-test; P = 0.024; Figs. 8A and 9A). No significant differences were found in stance phase soleus EMG between these two periods (paired t-test; P = 0.26; Figs. 8B and 9C).
Subjects took steps with ankle-foot load throughout the CONDITIONING period and during the LOAD catch trials occurring during BASELINE. Total hip extensor work was found to be significantly greater during the CONDITIONING period (0.48 ± 0.28 J/kg) than during the LOAD catch trials (0.35 ± 0.18 J/kg; paired t-test: P = 0.027; Figs. 8A and 9B). No significant differences were found in stance phase soleus EMG between these two periods (paired t-test: P = 0.29; Figs. 8B and 9D).
Comparing single step and multiple step exposures to similar ankle-foot load conditions provided a measure for assessing changes in feedforward locomotor output. Hip extensor work was significantly higher during a single, unexpected, no-load step occurring during the CONDITIONING period when ankle-foot load was normally applied every step than during BASELINE when subjects experienced no ankle-foot load every step. Hip extensor work was also significantly higher during an ongoing bout of airstepping with stance phase, ankle dorsiflexor torque applied every step (CONDITIONING) than during a single unexpected step with stance phase, ankle dorsiflexor torque occurring in the midst of a bout of no-load airstepping (BASELINE). In contrast, there were no differences in soleus EMG amplitude between single and multiple-step exposures.
This experiment studied the adaptive control mechanisms used to modulate locomotor output in response to limb load changes. Our results indicated that ambulatory individuals with incomplete SCI have the capability to tune locomotor output to load-related walking demands using a combination of feedback and feedforward control strategies. In reaction to acute changes in ankle-foot load, subjects made substantial within-step modulations to locomotor output that did not persist for future steps, providing evidence that ongoing feedback is used to rapidly adjust efferent patterns. The majority of subjects also made changes to their underlying locomotor patterns in response to repetitive ankle-foot load, suggesting that feedforward output is modulated in response to predictable ankle-foot load demands. These feedforward adaptations occurred over a relatively short conditioning period and persisted for several minutes. Together, the feedback and feedforward adaptation processes had a combined effect on ongoing locomotor output. Additional findings indicated that locomotor adaptation strategies are joint-specific, that short-term adaptive capability may be related to walking function after SCI, and that the magnitude of feedforward adaptations at the hip joint are proportional to the magnitude of the ongoing response to repetitive ankle-foot loading.
Observations of feedback and feedforward locomotor adaptations
Incomplete SCI subjects made immediate reactive responses at both the hip and ankle joints when presented with acute changes in ankle-foot load during stepping. Specifically, during catch trials, subjects made within-step locomotor responses to acute modulations in ankle-foot load that did not persist when loading patterns returned to their original state. Substantial increases in hip extension moments and soleus muscle activation accompanied LOAD catch trials and conversely decreased during NO-LOAD catch trials. These findings indicate that individuals with incomplete SCI make reactive feedback-modulated adaptations to locomotor output in response to changes in limb load. Given that animal models clearly show the ability of spinal networks to make feedback-driven locomotor responses to limb-load (Conway et al. 1987; Gossard et al. 1994; Hiebert et al. 1994) and that humans with complete SCI have been shown to make feedback-driven locomotor responses to single joint perturbations (Field-Fote and Dietz 2007), this finding is not unexpected. However, quantification of the feedback-modulated response in this study is valuable for understanding and assessing any feedforward adaptations that subjects potentially might produce in response to changes in repetitive lower limb loading.
Results from this study also indicate that ambulatory incomplete SCI subjects are able to modulate feedforward locomotor output at the hip joint. Two findings support this statement. First, when rhythmic ankle-foot loading was removed at the end of the CONDITIONING period, the majority of subjects continued to step with increased hip extension moments compared with BASELINE. This after-effect persisted for multiple minutes, and its magnitude was proportional to the ongoing response to repetitive ankle-foot loading. Second, during steps with similar ankle-foot load profiles, subjects produced greater hip extension moments when the nervous system anticipated ankle-foot load. This was evidenced by observations that subjects 1) produced greater amounts of hip extension work when they repetitively stepped with ankle-foot load (CONDITIONING period) than during single step unexpected LOAD catch trials (occurring during BASELINE) and 2) produced greater hip extension work during the NO-LOAD catch trials (occurring during CONDITIONING) than during the repetitive no-load steps (BASELINE). Overall, locomotor output during steps with similar loading profiles was dependent on both the current loading state and that occurring during previous steps. Collectively, these results suggest that ambulatory incomplete SCI subjects are capable of making modifications to feedforward locomotor output in response to repetitive and predictable limb load perturbations.
These results are consistent with our previous work (Gordon et al. 2009) and that of others showing that individuals with SCI will modulate locomotor patterns in response to limb load (Dietz et al. 2002; Harkema et al. 1997). This research builds on previous work by examining the dynamic response to ankle-foot load rather than examining steady-state performance. Our findings show that the magnitude of the hip extensor response to ankle-foot load is time dependent. The amplitude of hip extension work that subjects produce is greater after several minutes of stepping with ankle-foot load than in response to a single step load application. This time-dependent change in locomotor output provides insight into the individual contributions of feedback and feedforward adaptation. Specifically, the feedback-driven response to a single, unanticipated ankle-foot load step was an increase in hip extension work of 0.14 J/kg compared with BASELINE (no-load) stepping. However, when subjects received repetitive ankle foot loading, the increase in hip extension work was 0.27 J/kg greater than BASELINE stepping. This additional increase in hip extension work indicates that the nervous system responds differently to isolated and repetitive load perturbations. One explanation for the increase in hip extension work to repetitive loading is that the nervous system has adjusted feedforward locomotor output. If the feedback driven response to ankle-foot load was consistent across time, then the increase in hip extensor torque with repetitive ankle-foot loading was caused almost equally by feedback-mediated locomotor changes (∼52%) and feedforward adaptations (∼48%).
Pearson suggests that, for rhythmic behaviors, successful adaptation is a twofold process comprised of 1) the ability of the nervous system to precisely regulate complex movement patterns dictated by task demand and system mechanics and 2) the capacity of pattern-generating networks to adjust to persistent conditions (Pearson 2000). In the context of this study, execution of this first step in the adaptation process was shown by subjects' ability to continually update locomotor patterns in response to acute changes in ankle-foot load feedback. The capacity of the nervous system to use sensory feedback associated with the weight-bearing phase of gait to regulate walking patterns has been shown in numerous human (Dietz et al. 2002; Gordon et al. 2009; Harkema et al. 1997; Mazzaro et al. 2005; Pang and Yang 2000; Sinkjaer et al. 2000; Stephens and Yang 1999; Yang et al. 1991, 1998) and animal (Gorassini et al. 1994; Guertin et al. 1995; Hiebert and Pearson 1999; Pearson et al. 1992; Whelan et al. 1995) models. This previous research indicates that load feedback arising from group I (Conway et al. 1987; Duysens and Pearson 1980; Guertin et al. 1995; McCrea et al. 1995; Pearson et al. 1992; Whelan et al. 1995) and stretch-sensitive feedback arising from group II (Grey et al. 2001) ankle extensor muscle afferents and cutaneous receptors of the sole of the foot (Duysens and Pearson 1976; Duysens et al. 1990; Yang and Stein 1990) can act through spinal pathways (Conway et al. 1987; Gossard et al. 1994) to enhance ongoing ankle (Grey et al. 2007; Sinkjaer et al. 2000; Yang et al. 1991) and hip (Guertin et al. 1995) extensor muscle activity during the stance phase of gait. The reactive modulation observed at the hip joint in response to transient changes in ankle-foot load in this study further supports that, in humans, the excitatory reflex response to stimulation of the ankle extensor afferents and cutaneous mechanoreceptors during gait act to shape the amplitude and duration of ipsilateral extensor activity throughout the lower limb.
In contrast, the capability of the CNS to update central or feedforward commands in response to persistent changes in limb load has not been widely studied. Increasing evidence suggests that, during locomotion, the CNS develops an internal model of limb dynamics and of the environment, which is used to appropriately scale feedforward muscle activation levels (Emken and Reinkensmeyer 2005; Kriellaars et al. 1994; Lam et al. 2006; Pearson et al. 1999). Motor learning theory proposes that error-feedback signals act to gradually update the internal model and recalibrate feedforward output (Kawato et al. 1987; Reinkensmeyer et al. 2009). Although limited, there is evidence that load-feedback is used to recalibrate locomotor feedforward locomotor output. Cats exposed to chronic medial gastrocnemius loading show differential adaptation rates of initial and late bursting patterns during gait (Pearson et al. 1999). The increase in the late burst occurs rapidly and is attributed to modulations in gain of afferent pathways. In contrast, the increase in the initial burst (which is responsible for setting the stiffness of the ankle before ground contact) occurs gradually. It was originally proposed that error-feedback from the abnormally stretched muscle was used to scale the feedforward locomotor output, controlling the initial burst (Pearson et al. 1999). However, more recent findings suggest a limited involvement of stretch receptors in the adaptation process (Bouyer et al. 2001; Gritsenko et al. 2001) implying that other pathways [e.g., load (Cote et al. 2003) or pressure (Frigon and Rossignol 2007) sensitive mechanoreceptors] are likely involved in the recalibration process. It is possible that in this study, load-related error signals arising from the ankle were used in a similar manner to adjust feedforward locomotor patterns at the hip.
Neural structures involved in adaptation
Motor adaptation involves many structures of the CNS (Kawato 1999; Shadmehr and Wise 2005). The contribution of specific neural structures as they relate to locomotor adaptation has been thoroughly reviewed (Lam and Pearson 2002; Pearson 2000; Reisman et al. 2010). The cerebellum, which is important for detecting movement errors (Kitazawa et al. 1998; Seidler et al. 2004), is believed to play a major role in mediating feedforward locomotor adaptations. Observations that cats lose their ability to make locomotor adaptations when nitric oxide deprivation blocks cerebellum function provides strong evidence of cerebellum involvement (Yanagihara and Kondo 1996). Humans with cerebellar damage also show limited ability to make predictive adaptations during locomotion (Earhart et al. 2002a; Morton and Bastian 2006), giving additional evidence that the cerebellum is important for mediating locomotor adaptation. It is also possible that spinal networks play a role in locomotor adaptation. There is evidence that spinal networks are capable of forming a neural representation of limb dynamics (Shimansky 2000) and that spinal pattern generators can be entrained using sensory feedback (Kiemel and Cohen 2001; McClellan and Jang 1993; Vogelstein et al. 2006). Together these capabilities could provide the basis for performing feedback-driven predictive movement adaptations. The involvement of spinal circuitry in controlling feedforward locomotor adaptations is supported by research showing that spinalized rats make modulations to locomotor output that persists for several steps after repetitive perturbation by a viscous force field (Heng and de Leon 2007). However, in humans, the ability of the isolated spinal cord to make predictive adaptations has not been clearly observed. The motor cortex may also contribute to locomotor adaptation. In individuals with incomplete SCI, improvements in locomotor ability after treadmill training is associated with enhanced corticospinal connectivity (Norton and Gorassini 2006; Thomas and Gorassini 2005). Improved connectivity could be related to plastic changes in the cortex or associated descending pathways, suggesting that these sites may also play an important role in controlling locomotor adaptations. However, individuals with chronic cerebral stroke and associated hemiparesis show an ability to make locomotor adaptation during split-belt treadmill walking, indicating that the cerebrum may have limited involvement in locomotor adaptation (Reisman et al. 2007).
We found no evidence that incomplete SCI subjects make adaptations in feedforward locomotor patterns at the ankle joint. At the ankle, subjects made substantial feedback-mediated adaptations, implying that ankle extensor activation patterns are primarily driven by on-line feedback. Supporting this theory, experiments examining decerebrate cat stepping have found that up to 70% of ankle extensor muscle activity is directly modulated by limb load feedback (Hiebert and Pearson 1999). However, feedforward locomotor adaptation at the ankle joint has been observed in cat models after chronic muscle loading lasting several days (Pearson et al. 1999). Major differences between this study and previous work were the length of time the load conditioning was administered and the manner in which load was applied. Thus it is possible that extending the conditioning period beyond the 4 min used in this study could have allowed subjects to produce feedforward adaptations at the ankle joint. Also, there may be a difference in how the nervous system interrupts chronic loading provided through a mobilizing splint and phasic load applied during specific periods of individual gaits cycles. The former might be interpreted by the nervous system as a change in the musculoskeletal system and the latter as afferent feedback about the walking environment. Differences in these internal and external perturbations to the nervous system could result in different locomotor adaptation strategies (Cain et al. 2007).
At the hip, subjects made both feedback and feedforward modulations. In individuals with incomplete SCI, hip flexion, extension, and abduction strength have been identified as the most likely determinant of walking ability (Kim et al. 2004). Given that the hip and ankle joints generate the majority of sagittal plane positive work performed during gait (Eng and Winter 1995), our results suggest that hip strength may best predict walking ability because incomplete SCI subjects have a greater ability to modulate predictive components of gait at the hip joint than the ankle. Other research has found that individuals with motor-incomplete SCI in the cervical region show greater volitional and transcranial magnetic stimulation (TMS)-evoked contractions of foot and ankle muscles than the thigh muscle (Calancie et al. 1999), likely because of the greater number of corticomotoneuronal innervations in distal muscle groups before injury. If this finding is true in our subject population, it might suggest that the ability to produce feedforward locomotor adaptations is not a cortically controlled response.
There was variability in the capability of the incomplete SCI subjects to produce lasting feedforward locomotor adaptations. Two subjects, who increased hip extension work during the CONDITIONING period, did not show proportional after-effects. These two subjects stood out in terms of their functional walking ability. Comparatively, these subjects produced the least hip extension torque during BASELINE stepping, had the lowest 6-min walk scores, and had the slowest over ground walking speed of the group. Results from other clinical tests did not provide any additional distinguishing characteristics. A recent study found that individuals with incomplete SCI who responded to body weight–supported treadmill training showed greater levels of muscle activity during walking before training compared with individuals who did not respond to training (Gorassini et al. 2009). As well, examination of the capability of individuals with incomplete SCI to perform incline walking suggests that level-ground walking speed is a good predictor of ability to make locomotor adaptations (Leroux et al. 1999). These findings are consistent with our results indicating that walking ability is related to the ability of incomplete SCI subjects to produce feedforward locomotor adaptations. Differences in initial walking ability are likely because of differences in the extent of SCI, which can be quite broad in the ASIA C classification. Although we made an effort to recruit subjects of similar injury and movement ability, individual differences in the neural damage between subjects was inevitable. Because participants in our study had incomplete SCI, this study cannot isolate the specific neural structures and spinal pathways necessary for feedforward adaptations. Future studies targeting individuals with complete SCI or lesions occurring in specific regions of the brain and brain stem might provide greater insight into the neural structures responsible for controlling feedforward locomotor adaptations to limb load.
Implications for rehabilitation
In the United States, there are between 259,000 (National Spinal Cord Injury Statistical Center 2009) and 1,275,000 (Christopher and Dana Reeve Foundation 2009) people living with SCI, approximately one half of which have motor incomplete lesions (National Spinal Cord Injury Statistical Center 2009). Depending on the severity of the initial injury, the prognosis for ambulation after incomplete SCI is between 50 and 95% (Consortium for Spinal Cord Medicine 1999). Intense locomotor practice that emphasizes active participation in task-specific motions has been shown to increase walking ability after incomplete SCI. Progressively increasing limb load is an integral part of this gait rehabilitation (Barbeau et al. 2006b). Loading the limb is used to facilitate muscle activity, which is believed to be an important component for activity-dependent plasticity (Lynskey et al. 2008). Results from this study support integrating ankle-foot loading into training. Our findings showed that load feedback can drive short-term feedforward locomotor adaptations. These adaptations indicate that the nervous system is capable of recalibrating centrally driven locomotor output. It is suggested that these short-term adaptations are the basis of motor learning (Bastian 2008; Reisman et al. 2010; Shadmehr and Wise 2005) and that repeated exposure could result in more permanent locomotor improvements (Reisman et al. 2010).
The differences between hip extension moments observed when comparing catch trials to repetitive stepping performance could have been influenced by feedback from the contralateral leg. Specifically, subjects produced greater hip extension work during the CONDITIONING period than during the LOAD catch trials occurring during BASELINE. Whereas ankle-foot load in each of these conditions was similar in the leg directly subjected to catch trials, load profiles on the contralateral limb were not. In the contralateral limb during periods of double support, subjects experienced ankle-foot load during the CONDITIONING period but not during the LOAD catch trials occurring during BASELINE. The total period of double support was ∼300 ms, occurring at the beginning and end of the 1.16-s stance phase (period of ankle-foot loading). Although it is possible that contralateral limb loading could have influenced subject responses, our previous research has found that incomplete SCI subjects do not exhibit differences in hip extension work in the contralateral limb when the ipsilateral limb steps either with or without stance phase ankle-foot loading (Gordon et al. 2009).
The robotic devices we selected to isolate specific locomotor variables restricted lower limb movements and produced sensory stimulation that differed from normal walking. Using these robotic devices increased our ability to create a controlled environment while also potentially limiting the applicability of our findings to unobstructed over ground walking. We used a robotic device to load the ankle and foot independent of other joints during stepping, which is an experience different from the total limb-loading that happens during the stance phase of walking. Thus our findings can provide insight on the isolated effect of load signals arising from the ankle and foot, which may be enhanced or reduced when combined with additional limb loading signals arising from multiple locations during normal walking.
The method we used to calculate hip joint moments may have limitations when comparing load and no-load steps. We calculated hip muscle moments by subtracting the net hip torque created during passive no-load stepping from that produced during the other trials. Errors could occur if the lower limb dynamics were not consistent from step to step. This was unlikely at the hip and knee joints because kinematics were controlled. At the ankle, kinematics did differ between the load and no-load steps. This potential source of error should not influence conclusions made when examining feedforward adaptations because all these comparisons were made on steps with similar ankle loading profiles. Errors could occur during feedback comparisons between load and no-load steps. However, given the relatively small mass of the foot compared with the rest of the lower limb (Zatsiorsky and Seluyanov 1983) and the small changes in displacement during the loaded steps, this effect on our calculations was probably minor and unlikely to account for the substantial modulations in hip torque we observed between load and no-load trials.
Results from this study indicate that repetitive limb load feedback can induce residual enhancement of locomotor output in ambulatory individuals with incomplete SCI. The observed short-term feedforward locomotor adaptations indicate the capacity of the nervous system to recalibrate efferent patterns to load feedback. After repeated exposure, these locomotor adaptations may be valuable for inducing lasting improvements in locomotor performance. We also found that subjects used different control strategies at the hip and ankle joint to modulate locomotor patterns. Adaptations at the hip were controlled through a combination of feedback and feedforward strategies, whereas modulations in ankle patterns were primarily feedback driven. Finally, between-subject variation suggests that walking function after SCI may be related to the ability to produce feedforward locomotor adaptations.
This research was supported by Craig H. Neilsen Foundation Grant 2787 and the Searle Research Fund.
No conflicts of interest, financial or otherwise, are declared by the authors.
We thank Y. Dhaher for support on this project.
- Copyright © 2010 the American Physiological Society