Subjects

Nine male subjects (mean ± SD age, 25.7 ± 3.1 years old; height, 172.9 ± 4.9 cm; weight, 63.2 ± 7.3 kg) with no known history of neurological and orthopedic disorders volunteered for the study. Prior to participation, they gave written informed consent. The study was approved by the local ethics committee of the National Rehabilitation Center of Persons with Disabilities, Japan, and was conducted according to the Declaration of Helsinki.

Experimental protocols

In the experiment, subjects walked and ran on a split-belt treadmill (Bertec, Columbus, OH, USA) in which two belts (one underneath each foot) can be operated at separate velocities. The treadmill was operated in either “tied” (two belts at the same velocity) or “split” (at different velocities) [12] modes depending on the time period. The fast moving side and the slow moving side during the adaptation period are referred to as the “fast side” and “slow side”, respectively.

Fig. 1(A) illustrates the experimental protocol. Each subject participated in two experiments, which were assigned in random order. Experiments 1 and 2 consisted of walking and running, respectively. Basically, subjects went through 10-minutes adaptation periods with the treadmill driven in split (split-belt exposure) and the behaviors in the respective gaits before and after the split-belt exposure were compared in tied. The velocities were determined on the basis of our preceding study [10], in which subjects could both walk and run comfortably. During the baseline, the velocities were 1.0 (slow), 2.0 (fast), and 1.5 m s⁻¹ (middle) (1 minute each in duration). During the split-belt exposure, velocity of one belt (left) was adjusted to 2.0 m s⁻¹ and the other (right) to 1.0 m s⁻¹ (therefore, 2:1 in ratio). Subjects walked or ran under the novel split-belt physical constraint for 10 minutes. Following the 10-minute split-belt exposure, the belts were returned to the tied mode (at 1.5 m s⁻¹). This period persisted for 6 minutes. Between each period, the velocities were changed continuously (without stopping) with acceleration and deceleration of 0.5 m s⁻². The subjects were informed beforehand of these changes. The subjects were instructed to walk or run normally and to refrain from looking down at the treadmill belts to avoid receiving any visual information about the velocity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental protocols and representative waveforms of the anterior-posterior ground reaction force (GRF).

(A) In the experiments, subjects walk (experiment 1) or run (experiment 2) in the given speeds of treadmill. (B), (C) Each frame represents the GRF waveforms for five consecutive stride cycles at different time points in a single subject while walking (B) and running (C). The solid lines represent the fast side and the dashed lines are those of the slow side during the split-belt exposure.

https://doi.org/10.1371/journal.pone.0121951.g001

For safety, one experimenter stood by the treadmill during the experiment. The treadmill had handrails to minimize the risk if the subjects lost their balance, but all of the subjects were able to satisfactorily complete the experiments without using the handrails.

Data recordings and analysis

Three orthogonal components (mediolateral (Fx), antero-posterior (Fy), and vertical (Fz)) of the ground reaction force (GRF) were measured using two force sensors mounted underneath each of the treadmill belts. The force data were low-pass filtered at 8 Hz and digitized at a sampling frequency of 1 kHz using an analog-to-digital converter (Power Lab; AD Instruments). A custom-written program (VEE pro 9.0, Agilent Technologies) was used to analyze the force data. Using the vertical Fz component, stride cycles were determined on the basis of the moments of foot contact and toe off. For each stride cycle, peak absolute values in the antero-posterior Fy component were calculated for both the anterior (braking) and the posterior (propulsive) components (see Fig. 1(B)-(C) for representative waveforms). Differences in the peak forces between the sides (hereafter, termed as the degree of asymmetry) were calculated for both the anterior and the posterior components. For comparison between walking and running, since the cadence (therefore, the stride cycles taken) is different between the gaits along with possible differences between subjects, the data were averaged over stride cycles in 10-second bins. The averaged data were then normalized relative to the respective mean value under the middle velocity in the baseline period. Since the gaits could be perturbed by the acceleration or deceleration that occurred when the velocity changed, data for the first stride cycle after the velocity changed was excluded from later analysis.

To address rates of adaptation, the number of stride cycles that were required to adapt and de-adapt were calculated. On an individual basis, the stride to stride values for the degree of asymmetry (for both anterior and posterior components) were smoothed by using a 5-point moving average; the number of stride cycles the subjects took to reach plateau were determined for each parameter. The plateau was defined as the mean ± SD for each parameter over the last 50 stride cycles of the respective epoch (adaptation or de-adaptation).

Statistical analyses were performed for the degree of asymmetry in the anterior and the posterior forces at different time points (average taken over the 10 sec window) during the post-exposure periods (0–10 sec, 10–20 sec, 20–30 sec, 1 min (50 sec-1 min), 1 min 30 sec (1 min 20 sec-1 min 30 sec), 2 min (1 min 50 sec-2 min), 4 min (3 min 50 sec-4 min), and 6 min (5 min 50 sec-6 min)). A two-way analysis of variance (ANOVA) with repeated measures (factors: gait (walk or run) and time points) was used to test for statistically significant differences. When the ANOVA gave significant results, Bonferroni’s post-hoc comparisons were performed to test for differences between gaits. For the number of stride cycles taken to adapt and de-adapt, a paired Student’s t test was used to compare the differences between walking and running for each parameter. Data are presented as the mean and standard error of the mean (mean ± SEM). Stride-by-stride data are presented as the mean and standard deviation (mean ± SD). Statistical differences were accepted as significant when P < 0.05.

Locomotor adaptation and the underlying neural mechanisms

In locomotion, humans and other animals can transiently adjust their movement patterns to meet changing task or environmental demands. Moreover, repetition of the movement tasks or those under constant environmental constraints over minutes results in more persisting adjustments in motor output and hence an emergence of movement error upon return to normal condition without constraints (aftereffect) [12,18,19]. Such aspects have been investigated using particular kinds of physical constraint such as a curved path [20], a robotically- induced resistance [19], an elastic force field [18], and split-belt treadmills [12,21]. The emergence of these phenomena can be explained by the formation of a cerebellar “internal model”. This model involves the recalibration of the motor output on the basis of trial-and-error processes which occurs via a comparison between the expected and the resultant movements [1,22]. Indeed, the importance of cerebellar function on locomotor adaptation has been demonstrated both in animal [23] and human studies [8]. Long lasting motor adaptability upon walking on split-belt treadmill is impaired in both cats [23] and humans [8] with cerebellar deficits. In healthy subjects, Jarayam and colleagues [5], using a transcranial magnetic stimulation (TMS), demonstrated an involvement of cerebellar function in split-belt treadmill locomotor adaptations for human walking. They particularly revealed that the cerebellar involvement was specific to walking on the split-belt. This cerebellar involvement could not be simply attributed to the task difficulty resulting from the changing velocities of the treadmill.

Distribution of reactive feedback and adaptive feedforward strategies in locomotor adaptation

Even more importantly, the emergence of the adaptive and de-adaptive behaviors is specific to the phases of locomotion. It has been proposed that both reactive feedback and adaptive feedforward strategies are used depending on gait phases where the timing is dependent on the modality of the given physical constraint [18,19]. Locomotor adaptation in split-belt treadmill walking also exhibits a distribution of both control strategies throughout the stride cycle. Based on measurement of the GRF while walking on a split-belt treadmill, Ogawa et al. [11] revealed that feedforward control were merged to the earlier part of the stance phase (around heel contact), while feedback strategies occurred during the latter phase of the stance. They also noted that in contrast to the anterior braking component that exhibited slow adaptive and de-adaptive changes upon changing velocities, the posterior propulsive component showed only a fast reactive modification during walking [11]. To note, the velocities used in the study [11] were 0.5 and 1.0 m s⁻² for “tied” condition, and 0.5–1.0 m s⁻² for “split”, therefore different from those used in the present study. Another group has addressed gait adaptations by focusing predominantly on the temporal and spatial aspects of locomotion [12]. In a similar gait adaptation task, certain parameters such as step time, double support time, and the relative timing of two leg movement showed adaptive behavior and a subsequent aftereffect, while others, including modified stride length (anterior-posterior distance traveled by the foot during the stance phase) and stance time were reactive with only transient modifications [12]. With the possible variation in gait strategies (that is, relative contribution of both temporal and spatial adjustments to a given gait speed) both within and among subjects, use of these parameters may potentially underestimate the emergence of the adaptive phenomena. We have therefore chosen to focus on the analysis of the ground reaction forces [10,11]. However, this approach does create the potential limitation of not being able to address the swing-related aspects of locomotor adaptation.

Given the results of the preceding studies, though speculative, the difference in the adaptive and de-adaptive behaviors between walking and running may reflect different strategies by the CNS in terms of the feedback and feedforward control. Despite the similarity with well-coordinated joint movements both within and between the two multisegmental limbs, what functional differences exist between the two gaits? Possible differences in motor strategies underlying this adaptation are discussed below, based on known neural mechanisms of walking and running.

Neural mechanisms underlying walking and running

In the literature, locomotive movements are understood to be generated by specialized neural mechanisms in the CNS (for a review, see [24]). Moreover, recent studies have demonstrated independence of the neural mechanisms underlying different locomotive modes. Both direct recordings of neural activity in larval zebrafish [25] and experimental lesions of specific groups of neurons in mice [26] reveal that different groups of spinal interneurons underlie locomotion that occurs with different movement frequencies. In the supraspinal mechanisms, it has been demonstrated that stimulation to the mesencephalic locomotor region results in the emergence of different gait modes (walk, trot and gallop in cats; stepping and swimming in salamander) depending on the stimulation frequency [27–29].

In humans, a study of motor adaptation revealed that walking and running are controlled by distinct functional units being capable of each mode rather than sharing a common neural mechanism [10]. Regarding the basic motor patterns of walking and running, a number of studies have provided clear descriptions on the basis of muscle use [16,17,30,31]. The studies of Cappellini et al. [16] and Ivanenko et al. [17] analyzed activation patterns of 32 muscles during walking and running over a range of speeds. These studies revealed that the timing of activation in one out of five basic components distributed differently through the movement cycles depending on the gait mode and irrespective of the speed. Particularly, since the timing shift occurred within stance phase, this may account for one possible explanation for the present results in which the GRF component showed a different profile between the two gaits.

Kinetically, walking and running during the stance phases are characterized as “inverted” pendulum and “bouncing gait” [32]. In the inverted pendulum gait (walking), the center of mass (COM) is assumed to vault over a stiff leg. The bouncing gait (running), on the other hand, utilizes an elastic spring-like behavior of the limb. Walking and running therefore, use different forms of energy transformation, especially for horizontal braking upon foot contact and for propulsion at toe off. Through the stance phases, the amplitude of spinally-mediated monosynaptic H-reflex of the soleus muscle is known to be more suppressed in running than in walking, despite a greater level of background muscle activity in running [33]. These results indicate a greater contribution of peripherally-mediated reactive control in walking as compared with running. The soleus muscle has been shown to be an important contributor to forward propulsion in both walking [34] and running [35]. This raises the possibility that there is an increased contribution of peripherally-mediated reactive control in walking and thus a relatively greater contribution of the higher centers in running. It is possible that these differences in the neural strategies between the two gaits may have accounted for the emergence of the different behavior in the propulsive component of the GRF.

To summarize, with the similarity in the adaptation and de-adaptation of the anterior braking component, walking and running showed prominent difference in the posterior propulsive GRF component. In running, therefore, the factors contributing to the production of the posterior GRF component may additionally account for a key element for acquisition of motor patterns. This knowledge provides basic information to further studies, such as those focusing on task- or context- specificity of human gait and those aiming at performance enhancement in athletic training and/or in gait rehabilitation after particular motor dysfunction.