Locomotor patterns are adapted on a trial-and-error basis to account for predictable dynamics. Once a walking pattern is adapted, the new calibration is stored and must be actively de-adapted. Here, we tested the hypothesis that storage of newly acquired ankle adaptation in walking is dependent on corticospinal mechanisms. Subjects were exposed to an elastic force that resisted ankle dorsiflexion during treadmill walking. Ankle movement was adapted in <30 strides, leading to after-effects on removal of the force. We used a crossover design to study the effects of repetitive transcranial magnetic stimulation (TMS) over the primary motor cortex (M1), compared with normal adaptation without TMS. In addition, we tested the effects of TMS over the primary sensory cortex (S1) and premotor cortex (PMC) during adaptation. We found that M1 TMS, but not S1 TMS and PMC TMS, reduced the size of ankle dorsiflexion after-effects. The results suggest that suprathreshold M1 TMS disrupted the initial processes underlying locomotor adaptation. These results are consistent with the hypothesis that corticospinal mechanisms underlie storage of ankle adaptation in walking.
The neural control of walking is highly automatic in adapting to changes in environmental dynamics. Locomotor adaptation, like throwing or reaching adaptation, is acquired through trial and error in the face of predictable perturbations (Lam et al. 2006; Choi and Bastian 2007; Emken et al. 2007; Blanchette and Bouyer 2009). Once a movement pattern has been adapted (i.e., recalibrated), it must be actively de-adapted (i.e., washed out) when the normal conditions are restored (Martin et al. 1996). It is thought that multiple processes with different timescales contribute to motor adaptation (Smith et al. 2006). Thus, fast processes may drive behavioral changes during short periods of exposure, whereas slow processes may be more engaged during longer exposure to a given perturbation.
Studies of patients with specific neurological damage suggest that an intact cerebellum is required for locomotor adaptation (Morton and Bastian 2006)—the cerebellum is thought to calibrate internal models used for the predictive control of movement (Bastian 2006). On the other hand, focal cerebral damage and hemispherectomy have less dramatic effects on locomotor adaptation (Reisman et al. 2007; Choi et al. 2009). As functional changes may be masked by compensatory mechanisms, and considering the complex nature of the walking movement, the role of the cerebral cortex in locomotion is less clear.
While the corticospinal tract is not necessary for generating the basic locomotor rhythm in most animals (Brown 1911; Grillner and Wallen 1985), more recent work in both cat (Drew et al. 2002) and human (Petersen et al. 2001) suggests that corticospinal neurons are active during steady over ground and treadmill walking. Corticospinal neurons are thus an integrated part of the network that is responsible for driving the spinal motoneurons during walking (e.g., Nielsen et al. 2003). Corticospinal activity can quickly correct the muscle activity according to visual (Widajewicz et al. 1994; Armstrong and Marple-Horvat 1996; Drew et al. 1996) and somatosensory information (Amos et al. 1989; Christensen et al. 1999; Christensen et al. 2001). As visuo-somatosensory–motor integration likely represents an important role for the motor cortex in the control of gait (Drew, Andujar, et al. 2008), it should be expected that the corticospinal tract would also participate in the gait modifications required to adapt to a force field applied during walking. This hypothesis was recently supported by Barthélemy et al. (2012), who have investigated changes in corticospinal excitability during force field adaptation of human gait using transcranial magnetic stimulation (TMS). Motor-evoked potentials (MEPs) in the tibialis anterior (TA) muscle were rapidly modulated during force field adaptation, being either upregulated for a resistive force adaptation, or downregulated during assistive force adaptation. We hypothesized that the previously reported changes in corticospinal activity might be associated with cerebral mechanisms contributing to fast adaptation processes, and therefore predict that disrupting cerebral circuits during short periods of force field exposure would impair locomotor adaptation.
Repetitive TMS (rTMS) has been used as a noninvasive tool to examine the cortical mechanisms of human motor learning (reviewed in Censor and Cohen 2011). A “virtual lesion,” or a temporary disruption of cortical processing, can be induced by applying rTMS offline with low-frequency trains (Chen et al. 1997). Applying 15–20 min of 1-Hz rTMS over the primary motor cortex (M1) immediately following practice of a ballistic finger (Muellbacher et al. 2002) or ankle (Lundbye-Jensen et al. 2011) task have been shown to disrupt the retention of motor learning. A virtual lesion can also be created on-line where TMS pulses coincide with the execution of the motor tasks (Hadipour-Niktarash et al. 2007). It has been demonstrated that on-line TMS reduced the after-effects of reaching adaptation, without affecting movement execution or performance (Hadipour-Niktarash et al. 2007). Thus, there is increasing evidence that M1 is involved in storing newly acquired motor patterns.
In this study, we tested the hypothesis that storage of walking adaptation also depends on M1. Walking adaptation to resistance against ankle dorsiflexion leads to increased activation of the TA muscle as previously demonstrated (Barthélemy et al. 2012). Adaptation to elastic force acts on the timescale of minutes and results in after-effects on removal of the elastic force (Fortin et al. 2009). Here, we investigated the role of M1, sensory cortex (S1), and premotor cortex (PMC), by applying on-line TMS over these areas during walking adaptation. We showed that suprathreshold TMS reduced the size of after-effects, suggesting that corticospinal mechanisms may be involved in the initial processes underlying locomotor adaptation.
Thirty-eight healthy volunteers (21 females and 17 males; 24 ± 4 years) with no known neurological disorder participated in this study. The study was approved by the local ethics committee (Protocol # H-A-2008-029). All methods conformed to the Declaration of Helsinki. All subjects gave written informed consent prior to participation.
The treadmill was set to a comfortable walking speed for each subject (3.2 ± 0.3 km/h). Subjects walked with an elastic band (Thera-band®; Akron, OH, USA; Gray tubing) attached to the back of a modified Klenzak ankle–foot orthosis (AFO), between the shoe and the calf band (Fig. 1A). The AFO was worn on the right leg. When under tension, the elastic band resisted ankle dorsiflexion, with a maximal effect during the swing phase (Barthélemy et al. 2012). Elastic bands of 2 different lengths, long (13 cm) and short (6.5 cm), were used to change the magnitude of the elastic force. The long elastic produced minimal torque about the ankle, and therefore considered a null field. The short elastic produced a torque around the ankle of approximately 5 Nm. This torque level was chosen and applied to all participants, because it produced a significant initial perturbation and yet required only a small to moderate amount of effort to be compensated, thereby avoiding fatigue or force saturation issues. The AFO provided a slight stabilization of the ankle joint in the coronal plane, but had a sufficient range of motion to allow completely free movement in the sagittal plane. Participants did not report any movement limitations during walking. The experimenter switched elastic bands in between testing periods, but the subjects were not told which elastic band was used.
Each block of testing consisted of a baseline, adaptation, and de-adaptation period of walking (Fig. 1B). In the baseline period, the long elastic was used and subjects walked for 30 s in the null field. In the adaptation period, the short elastic was used and subjects were instructed to “resist the force and walk normally” for 30 s. In the de-adaptation period, the long elastic was used and subjects walked for 90 s in the null field to wash out any after-effects.
We used a crossover design to determine the effects of rTMS over M1. Each subject completed 2 testing blocks, one with rTMS and one without. rTMS was applied over M1 during the adaptation period. No stimulation was applied during the baseline and de-adaptation periods. The order of testing was randomized, some subjects received TMS in Block 1 (n = 9), and others received TMS in Block 2 (n = 14). Both blocks of testing were performed on the same day. To check whether rTMS only (i.e., no force adaptation) changed walking patterns, subjects walked for 30 s without TMS, followed by 30 s with rTMS, followed by 30 s with no stimulation, all in the null field. The null block was performed at the end of each session.
A second experiment was performed to determine the effects of moving the coil away from M1. We hypothesized that motor adaptation would be affected by stimulation over M1, but not over S1 and PMC. Therefore, S1 and PMC served as control sites of stimulation. Each subject completed 3 blocks of testing: no TMS, TMS over S1 (i.e., 2 cm posterior to the vertex), and TMS over PMC (i.e., 2 cm anterior to the vertex). TMS was applied over S1 or PMC during the adaptation period only. The order of the blocks was randomized across subjects (n = 14). All 3 blocks were performed on the same day.
Ankle angle was recorded using a custom-built electrogoniometer (Axel Scherle, Freiburg, Germany) attached to the AFO. Electromyographic (EMG) activity from the right TA and soleus muscles were amplified (Zerowire, Aurion, Italy) and sampled at 2 kHz (Micro 1401 and Spike2, Cambridge Electronic Design, UK). Footswitch sensors were placed under the heel and great toe bilaterally to record the time of heel strike and toe off.
Transcranial Magnetic Stimulation
A Magstim Rapid 2 stimulator (Magstim, Whitland, UK) and a figure-of-eight coil (9-cm loop diameter) were used. The coil was held in a fixed position in relation to the head by a custom-designed harness during walking. At the beginning of each experiment, the coil was placed slightly left of the vertex in order to determine the motor threshold of the right TA muscle. The threshold was defined as the intensity at which a MEP was clearly distinguishable from the background activity in 3 of 5 trials. For the main experiment, the coil was either kept at the same place (Exp 1) or moved to S1 (i.e., 2 cm posterior to the vertex) or PMC (i.e., 2 cm anterior to the vertex) (Exp 2).
Subthreshold Single-Pulse TMS
We used the lowest level of single-pulse TMS that altered corticospinal output to the TA muscle. Suppression of EMG activity could be elicited by subthreshold TMS (Davey et al. 1994). We estimated the threshold for the TA muscle by applying single-pulse TMS during the early part of swing phase in walking (Petersen et al. 2001). TMS pulse was triggered at 100 ms delay after left heel contact (∼55% of the right stride cycle), given randomly once every 3 strides. Raw EMG data were rectified and averaged online using Signal (Cambridge Electronic Design).
At higher intensities, MEP was clear in the averaged EMG over relatively few strides. TMS intensity was decreased until there was no short latency (<40 ms) facilitation in the averaged TA EMG over 50 strides. When that intensity was reached, we recorded another 25–50 strides with TMS (for a total of 75–100 strides) to determine whether there was suppression of TA EMG activity. TMS suppression during walking is not obvious in single trials, but can be seen by averaging over a relatively large number of strides (Petersen et al. 2001). The lowest level of single-pulse TMS that produced suppression of TA EMG activity with no or minimal short latency facilitation was used as the TMS intensity for subsequent TMS interventions.
Repetitive Transcranial Magnetic Stimulation
TMS was applied during the adaptation period of the main experiment, with the coil place over either M1 (Exp 1) or S1/PMC (Exp 2). rTMS consisted of a train of pulses that lasted for 700 ms (10 Hz), triggered by left foot contact (∼50% of the right stride cycle) on each stride. The timing of TMS coincided with perturbation to ankle movement caused by the elastic force (Fig. 2C). With this setting, TMS was delivered during the swing phase on the right leg, beginning around the time of TA onset and ending around the time of TA offset. Each subject took different number of strides (27 ± 2 strides) during the 30-s adaptation period. Therefore, the total number of TMS pulses delivered varied across subjects. The stimulation intensity was predetermined for each subject based on suppression of EMG during single-pulse TMS over M1 (see above). The presence of MEPs was used to determine whether the rTMS was suprathreshold or subthreshold.
Ankle angle data were separated into individual strides, aligned on the time of right heel strike, and time normalized to 100% stride cycle. Stride-by-stride changes in kinematics were quantified by measuring the peak ankle dorsiflexion during the swing phase. Data were binned by 3 strides, and statistics were performed on the first baseline bin, the first and last adaptation bins, and the first and last post-adaptation bins. Repeated-measures analysis of variance (ANOVA) was used to test the effects of TMS, time, and TMS × time. Post hoc analysis was performed using the Tukey's HSD test. The alpha level was set at 0.05 for all statistical comparisons.
Motor-Evoked Potentials During Force Adaptation
Based on our previous study (Barthélemy et al. 2012), we expected corticospinal excitability to increase during walking adaptation. We therefore examined whether TMS during force adaptation evoked MEPs in individual subjects. Raw EMG data were high-pass filtered at 35 Hz and rectified for offline analysis. Processed EMG data were separated into individual strides, aligned on the time of right heel strike. For each subject, EMG amplitude was normalized to the peak averaged EMG during baseline walking. Background EMG was determined as the mean normalized TA EMG over a 15-ms window that ended 5 ms before each TMS pulse. MEP was determined as the mean normalized TA EMG over a 15-ms window that began 30 ms after each TMS pulse. The ratio of MEP to background EMG was used to determine whether TMS elicited a response (>130%) in the TA muscle during the adaptation period in individual subjects.
Applying resistance against ankle dorsiflexion during walking caused changes in ankle joint kinematics as previously demonstrated (Barthélemy et al. 2012). Ankle dorsiflexion during swing phase was initially reduced in the presence of the elastic force (Fig. 2A). After a short period of force exposure, subjects showed increase ankle dorsiflexion (negative after-effect) on removal of the force perturbation, indicating storage of locomotor adaptation (Fig. 2B).
Effects of TMS Over M1
We tested whether TMS over M1 during walking disrupted ankle adaptation to elastic resistance. TMS was delivered during the swing phase of the right leg during force adaptation (Fig. 2C). The results showed that average after-effects in ankle movement were reduced when subjects received TMS over M1 during the adaptation period (Fig. 2D).
Group-averaged peak ankle dorsiflexion was referenced to each subject's average during the baseline period (zero). Figure 3 shows group-averaged peak ankle dorsiflexion plotted across strides during baseline, adaptation, and post-adaptation from Exp 1. We found that applying rTMS over M1 lead to reduced after-effects on removal of the force resistance. An ANOVA showed a significant effect of time (F4,88 = 89.2, P < 0.001) and TMS × time (F4,88 = 3.9, P < 0.01).
We checked whether subjects were similarly perturbed in early adaptation, and found no significant difference in the first adaptation bin across conditions [t(1,22) = −0.7, P = 0.5]. We also checked whether the number of exposure strides were the same across conditions, and found no significant difference in the number of strides taken during the adaptation period [t(1,22) = −1.6, P = 0.1]. Subjects reached the same peak ankle dorsiflexion in the last stride bin during adaptation [t(1,22) = −0.6, P = 0.5]. Thus, we believe that reduced ankle after-effects were due to the effects of applying TMS over M1.
We realize that applying rTMS could induce suppressive or excitatory effects that outlast the stimulation period (Chen et al. 1997). We therefore verified that applying 30 s of rTMS without force adaptation did not result in after-effects in ankle dorsiflexion. A repeated-measures ANOVA was used to compare the first pre-TMS bin, the first and last TMS bin, and the first and last post-TMS bins. The ANOVA showed nonsignificant effect of time (F4,100 = 0.1, P = 0.9). This suggests that applying TMS alone did not change ankle control in general, but that applying TMS during training impaired storage of ankle adaptation.
The results suggest that applying TMS over M1 during training reduced after-effects associated with the adaptation. This is consistent with the hypothesis that the corticospinal system contributes to the storage of error-driven ankle adaptation.
Effects of TMS Over S1 or PMC
Figure 4 shows group-averaged peak ankle dorsiflexion during baseline, adaptation, and post-adaptation from Exp 2. We found that TMS over S1 or PMC did not affect group-averaged peak ankle over the course of walking adaptation. The ANOVA showed a significant effect of time (F4,48 = 68.4, P < 0.001) and no significant effect of stimulation (F2,24 = 0.3, P = 0.7) or stimulation × time (F8,96 = 1.7, P = 0.1).
Suprathreshold Versus Subthreshold TMS
Given the close proximity of M1, S1, and PMC, we examined the presence of MEPs to determine whether corticospinal pathways were stimulated in each block. We combined all data from Exp 1 and Exp 2, and sorted the testing blocks based on whether TMS elicited MEPs during the adaptation period. For each block, we calculated the ratio of the size of MEPs to background EMG. We found that TMS elicited significant MEPs (>130% background) in 9/26 blocks for M1, 2/14 blocks for S1, and 0/14 blocks for PMC.
Figure 5 shows the post-adaptation peak ankle dorsiflexion averaged over all blocks with no TMS, subthreshold TMS, or suprathreshold TMS. The ANOVAs show a significant effect of group (F2,84 = 3.5, P = 0.03) and group × time (F2,84 = 3.2, P < 0.05). Tukey's post hoc test indicates a significant difference between no TMS versus suprathreshold TMS blocks (P = 0.03). Therefore, the results suggest that suprathreshold rTMS, but not subthreshold rTMS, disrupted storage of walking adaptation to resistance.
In this study, we demonstrated that applying TMS over M1 impaired walking adaptation to ankle force perturbation. Motor adaptation is a type of motor learning acquired on a trial-by-trial basis to account for predictable perturbations. The recalibration process is reflected in the size and duration (# of strides) of after-effects on removal of the perturbation, which increases gradually with the number of strides exposed to the perturbation (Lam et al. 2006; Fortin et al. 2009). Here, we used a short period of adaptation (30 s) in order to study the neural mechanisms involved in the early phase of adaptation. We believe that the behavioral changes during the first period of exposure were driven mainly by fast adaptation processes (Smith et al. 2006). The results suggest that M1 and/or subcortical structures activated by M1 contribute to the fast processes underlying locomotor adaptation.
Our results are consistent with previous studies supporting a role of M1 during the initial stages of motor learning (Muellbacher et al. 2002; Richardson et al. 2006; Hadipour-Niktarash et al. 2007; Lundbye-Jensen et al. 2011). However, we found that only suprathreshold, but not subthreshold, TMS had an effect in ankle dorsiflexion post-adaptation. Therefore, we cannot rule out the possibility that subcortical systems are involved. The results are also consistent with other studies showing that suprathreshold, but not subthreshold, rTMS disrupted retention of motor learning (Lundbye-Jensen et al. 2011).
We showed that the control sites (i.e., S1 and PMC) are less effective in disrupting the adaptation process. To ensure that the stimulation was applied either over M1 or the control sites, the coil position was marked on the scalp and regularly inspected throughout the experiment. We did not measure the coil position across trials using a brain navigation system; however, any variability in the coil position should affect all 3 stimulation conditions equally.
Different cortical areas are involved in the planning and execution of movements during walking. Experimental evidence from cat locomotion suggests that the motor cortex is not strongly modulated during unperturbed level walking, but becomes active when precise adjustments in paw placement, or end-point control is required (Beloozerova and Sirota 1993; Drew 1993; Drew et al. 2002). Lesions of the corticospinal tract can cause marked deficits in voluntary modifications to the walking pattern, such as in obstacle avoidance (Drew et al. 2002). The motor cortex may act to regulate the duration, level, and timing of small groups of synergistic muscles, active at different times during the gait modification (Drew, Kalaska, et al. 2008). Corticospinal control may be more important in human locomotion, compared with that of other animals. This could be due to several factors, including that humans are bipedal and have greater equilibrium demands. This is supported by recent experimental findings, showing that the corticospinal tract in human drives leg muscles during normal steady-state walking (Petersen et al. 2001; Petersen et al. 2012).
Our studies suggest that corticospinal mechanisms may also be involved in peripherally driven walking adaptation. In a previous study, we have shown that adaptation is associated with task-specific changes in the corticospinal control of ankle muscles during walking (Barthélemy et al. 2012). Here, we showed that TMS disruption over M1 during walking adaptation reduces the after-effects from training. While there are synapses between the corticospinal cells and the spinal motoneurons (direct monosynaptic pathway), TMS over M1 can also influence transmission in indirect polysynaptic pathways via spinal interneurons and premotoneurons (Nielsen et al. 1993; Pierrot-Deseilligny 2002; reviewed in Barthélemy et al. 2011). Descending cortical signals can therefore interact both with motoneurons via the direct pathway and with the locomotor networks in the spinal cord via the indirect pathway to ensure that the changes are appropriately incorporated into the basic locomotor pattern. Both of these pathways could therefore be involved in the storage of walking adaptation to force field exposure. Results from the animal literature have shown the potential for adaptive capacity of subcortical neural elements. For example, completely spinalized cats walking on a treadmill can modify their muscle activation pattern in an adaptive manner after muscle nerve axotomy (Bouyer et al. 2001). They achieved this without a cerebellum or a motor cortex. While the relative contribution of spinal and supraspinal control during walking may be different in humans and animals as stated above, the subcortical system should not be considered as hardwired. Thus, storage of locomotor adaptation to a force field may involve cortical, and/or subcortical, structures with access to spinal locomotor networks. Further studies will be necessary to clarify cortical versus subcortical involvement in locomotor adaptation.
Perturbing foot dorsiflexion during the swing phase of gait can have important consequences (e.g., forefoot comes into contact with the treadmill and causes a stumble or fall). When walking with the elastic perturbation, subjects have to rapidly learn to compensate for the external perturbation to produce a minimally safe toe clearance. This could be achieved either by reducing the elastic-induced ankle plantarflexion or by increasing hip vertical displacement during swing. Previous work using this model of adaptation has shown that participants tend to use the former (Barthelemy et al. 2012). While this remains speculative at the moment, we believe that when participants are exposed to an external force during walking, they learn to increase their joint angle, and not to produce a given amount of force, as the former is more functionally relevant to the situation.
In conclusion, the results suggest that storage of walking adaptation, like reaching adaptation, depends on the corticospinal system. This is true for the fast adaptation processes studied here. This information may be critical for understanding, evaluating, and treating gait deficits after neurological damage. For instance, fast adaptation processes may be more impaired after cerebral stroke, and these patients may benefit from training strategies that promote slow adaptation processes. Moreover, adaptation during walking and reaching may be subject to the same general principles that govern motor adaptation across the upper and lower limbs.
Danish Medical Research Council (11-107721/FSS) and the Whitaker International Program.
Conflict of Interest: None declared.