EMG Recordings During Real Locomotion

Electromyographic activity (EMG) of nine major hindlimb muscles was chronically recorded in six adult female cats (2.5–4.0 kg) during overground and treadmill level walking. All surgical and experimental procedures were in agreement with the Guide for the Care and Use of Laboratory Animals (NIH Pub. No. 86-23, 1985) and approved by the Institutional Animal Care and Use Committees of Drexel University and Georgia Institute of Technology.

The methods of animal training, implantations of EMG electrodes, and locomotion data collection have been described elsewhere (Boyce et al. 2007; Gregor et al. 2006; Ollivier-Lanvin et al. 2011; Prilutsky et al. 2011) and are described only briefly here. Prior to surgical implantation of EMG electrodes, cats were trained to walk on a treadmill or across a walkway (3.0 × 0.4 m) with Plexiglas walls with the use of operant conditioning methods and food reward. After completion of training, the skin around the major hindlimb joints was shaved and small reflective markers (6–9 mm in diameter) were attached to bony landmarks with double-sided adhesive tape. Gait of the cat was recorded with a high-speed (120 or 300 Hz) motion capture system (Vicon) during motorized treadmill or overground locomotion, and ground reaction forces were also recorded during overground walking with small (0.16 × 0.11 m) force plates (360 Hz, Bertec) embedded in the floor of the walkway. The kinematics of markers and ground reaction forces were used to identify the timing of stance and swing phases during treadmill and overground locomotion, respectively, to measure speed of overground walking and to confirm that cats demonstrated typical level walking mechanics during EMG data collection.

After initial locomotion data collection, bipolar EMG electrodes (multistranded, Teflon-insulated, stainless steel wires, AS633 Cooner Wire) were surgically implanted in nine hindlimb muscles unilaterally to record the corresponding EMGs during locomotion (see Figs. 1 and ​and2B):2B): sartorius medialis (SartM, hip flexor, knee flexor), iliopsoas (Ip, hip flexor), anterior biceps femoris (AB, hip extensor), rectus femoris (RF, hip flexor, knee extensor), posterior biceps femoris (PB, hip extensor, knee flexor), vastus medialis or lateralis (VA, knee extensor), tibialis anterior (TA, ankle flexor), soleus (Sol, ankle extensor), and medial or lateral gastrocnemius (MG/LG, ankle extensor, knee flexor). All implanted wires were tunneled subcutaneously to either exit between the two shoulders to attach to a connector mounted on a small jacket worn by the cat or exit at a head-mounted multipin connector. Locations of the EMG electrodes were verified by mild electrical stimulations through the implanted wires.

After recovery (10–14 days), EMG activity (sampling rate 2,400 Hz, treadmill; 3,000 Hz, overground) and kinematics of walking, as well as ground reaction forces during overground walking, were recorded synchronously as previously described (Boyce et al. 2007; Gregor et al. 2006; Ollivier-Lanvin et al. 2011; Prilutsky et al. 2011; see Table 1). Cats walked overground at a self-selected speed (typically between 0.4 and 0.6 m/s) and on the motorized treadmill at a speed of 0.4 m/s.

The EMG recordings were filtered (band-pass filter 30 Hz to 1 kHz), rectified, and low-pass smoothed by using a binomial smoothing filter with 30 Hz cutoff frequency (Marchand and Marmet 1983). Examples of rectified and smoothed EMG recordings from a freely walking cat are shown in Fig. 2B. Since we did not find marked differences in EMG patterns recorded during treadmill and overground locomotion, matching similar observations in cats (Carlson-Kuhta et al. 1998; Smith et al. 1998) and humans (Arsenault et al. 1986; Murray et al. 1985; Nymark et al. 2005), data from both types of experiments were processed together. Records from all experiments were averaged and normalized with the same methods as for the processing of the fictive locomotion data (see below and in appendix).

Data Processing

Only ENG and EMG records that exhibited stable (with <20% variability in both amplitude and burst duration) activity during several consecutive locomotion cycles (not less than 10 for fictive and treadmill locomotion and 4 for overground walking) were chosen for further analysis (Table 1). In each experiment, four to nine different nerves or muscles were recorded simultaneously. Custom software developed at the Spinal Cord Research Centre, University of Manitoba and operating under a Linux operating system was used to detect the timing of burst onset and offset in the preprocessed ENGs recorded during fictive locomotion. The cycle duration in fictive locomotion was defined as the average time between two consecutive flexor or extensor bursts. The burst onset in the ENG of SartM (hip and knee flexor) was considered as the beginning of the flexion phase, and the burst onset in the ENG of semimembranosus-anterior biceps (SmAB, hip extensors) was considered as the beginning of the extension phase (see Figs. 2A and ​and3A).3A). Because the locomotion cycle and phase durations varied substantially between preparations, the durations of flexion and extension phases were normalized separately over all experiments (e.g., Berkowitz and Stein 1994; Orlovsky 1972; for details, see appendix). Correspondingly, the timing of onset and offset of ENG bursts in all nerves were represented as portions of the corresponding (flexion or extension) phase, and then motoneuronal activities normalized and averaged within each phase were connected together in the proper sequence.

Open in a separate window

Fig. 3.

Hindlimb motoneuronal/muscle activities normalized and averaged over all records during fictive and real level locomotion (see Table 1). The standard deviation (SD) for each pattern is shown as a gray corridor around the corresponding mean trace. Note that according to the normalization and averaging methods used (see appendix), the SD at each point represents deviation in time (but not in amplitude), i.e., ±SD extends left and right. A: motoneuronal ENGs normalized and averaged over all experiments during fictive locomotion. Vertical dashed lines indicate onsets of flexion and extension phases. Inset: types of PBSt and RF patterns observed during fictive locomotion and their possible combinations (see text for details). B: muscle EMGs normalized and averaged over all experiments during unrestrained locomotion. Vertical dash-dotted lines indicate onsets of swing and stance phases; vertical dashed lines show onsets of flexion and extension phases. For abbreviations see Figs. 1 and ​and22.

The locomotion cycle duration for unrestrained walking was defined as the average time between two consecutive moments of liftoff or touchdown that were considered the starting points of the swing and stance phases, respectively. Ground reaction forces were used to evaluate liftoff and touchdown for overground locomotion, while kinematic analyses of the metatarsophalangeal (MTP) and toe markers were used to evaluate liftoff and touchdown for treadmill locomotion. The normalized and averaged EMG profiles for real locomotion were processed with the same methods as for fictive locomotion records (see above).

To compare the averaged ENG and EMG profiles calculated for fictive and unrestrained locomotion, respectively (Figs. 3​3–5), the normalization coefficients representing the ratios between the average duration of swing and stance phases and the duration of the locomotion cycle during walking were applied to the fictive locomotion profiles to provide the same relative duration for the flexion and extension phases relative to the locomotor cycle. Hence in all cases the locomotion cycle duration was considered to be equal to 1 and the (relative) duration of each phase for fictive locomotion was made equal to that for real locomotion.

Open in a separate window

Fig. 4.

Comparison of normalized ENG patterns obtained during fictive locomotion with the corresponding normalized EMG patterns recorded during real locomotion. Patterns were taken from Fig. 3, A and B. The ENG profiles from fictive locomotion (fict.) are shown positive, whereas the EMG profiles from real locomotion (real) are shown negative. For abbreviations see Figs. 1 and ​and22.

Open in a separate window

Fig. 5.

Scatterplot representation (time of burst offset vs. time of burst onset) of nerve and muscle activity recorded from all cats during fictive and real locomotion, respectively (data shown in Table 1), and the corresponding results of minimum spanning tree (MST) clustering. Each symbol in each plot represents normalized averaged data from 1 analyzed record from Table 1. A: scatterplot and MST-based clusters for fictive locomotion. B: scatterplot and MST-based clusters for real locomotion. Color ellipses indicate the identified clusters shown on right. As a result of normalization, the step cycle period in both plots was equal to 1, with durations of flexion and extension equal to 0.36 and 0.64, respectively. Dashed black lines in both A and B show onsets of flexion and extension; dash-dotted blue lines in B show onsets of swing and stance phases. Identified clusters are numbered in parentheses (right). For convenience of data comparison between fictive and real locomotion the same numbering scheme is used for the clusters found in both data sets (A and B), and 2 identified clusters for fictive locomotion (in A) were split into groups [cluster (1) into groups 1 and 2 and cluster (3) into groups 3a and 3b]. Note that when the PBSt nerve was active during extension (pattern types 2 and 3 in Fig. 3A), this extensor activity (PBSt_ext indicated in A by *) was classified into the cluster containing all hip and ankle extensors [cluster (3)]. In contrast, the PBSt flexor burst (PBSt_fl, see pattern types 1 and 3 in Fig 3A) always represented a separate cluster conditionally numbered as (5). Also, each RF burst represents a separate cluster (marked in A by **) conditionally numbered as (2) and (7). Also, 2 PB EMG bursts in real locomotion (PB_fl and PB_ext in B) initially represented separate clusters (5) and (6) (marked by ***). For all other clusters identified in fictive and real locomotion, the cluster (and its number) is equal to the corresponding group (and its number). Based on the performed classification, group 1 of cluster (1) in fictive locomotion (A) corresponds to cluster (1) in real locomotion (B); group 2 of cluster (1) in fictive locomotion (A) corresponds to cluster (2) in real locomotion (B); cluster (3), including groups 3a and 3b, in fictive locomotion (A) corresponds to cluster (3) in real locomotion (B); clusters (4), (5), and (6) (groups 4, 5, and 6) in fictive locomotion (A) correspond to the corresponding clusters (groups) in real locomotion (B). For abbreviations see Figs. 1 and ​and2.2. See text for details.

Cluster Analysis

Cluster analysis was used to identify the nerves and muscles with synergistic activity during fictive (Fig. 5A) and real (Fig. 5B) locomotion. The same MST-based clustering algorithm was applied to the normalized activity profiles of different nerves or muscles. Activity of 14 nerves was analyzed during fictive locomotion: Ip, SmAB, SartM, PBSt, RF, VA, TA, Plong, extensor digitorum longus (EDL), MG, LG/Sol, plantaris (Plant), flexor hallucis longus (FHL), and flexor digitorum longus (FDL) (see Table 1 and Fig. 5A). On the basis of the cluster analysis, all hip and ankle flexor nerves (SartM, Ip, TA, and Plong) were grouped into the same cluster [cluster (1) that included groups 1 and 2; see Fig. 5A]. Also, all hip and ankle extensor nerves (SmAB, MG, LG/Sol, and Plant), FHL, and the extensor burst of PBSt fell into a single cluster [cluster (3), groups 3a and 3b; see Fig. 5A]. The VA, flexor bursts of PBSt, RF, EDL, and FDL each represented separate clusters (Fig. 5A). Note that when the PBSt was active during extension (pattern types 2 and 3 in Fig. 3A), this extensor activity (Fig. 5A) was classified into the cluster containing most of hip and ankle extensors [cluster (3)]. In contrast, the PBSt flexor burst (see pattern types 1 and 3 in Fig. 3A) always represented a separate cluster. Also, each RF burst fell into a separate cluster (Fig. 5A).

The results of cluster analysis applied to real locomotion are shown in Fig. 5B. In contrast to fictive locomotion, the hip flexors (SartM and Ip) and ankle flexor (TA) fell into different clusters [clusters (1) and (2), respectively, in Fig. 5B]. At the same time, similar to fictive locomotion, all hip and ankle extensors (AB, MG, Sol) belonged to one single cluster (Fig. 5B). Also similar to fictive locomotion, VA, PB, and RF each represented a separate cluster. Moreover, each of the two PB bursts represented a separate cluster (Fig. 5B).

Muscle Synergies Determined from Fictive Locomotion Studies

Fictive locomotion allowed for identification of the possible synergies used by the locomotor CPG for the control of coordinated motoneuronal activity in the absence of afferent feedback. To our knowledge, this study represents the first attempt to identify such synergies. It should be noted that ENG activity analyzed in the present study was evoked by brain stem (MLR) stimulation and hence may also depend on the organization of brain stem circuits. However, the similarity in fictive locomotor patterns between the decerebrate and spinal cat preparations (see, e.g., Hamm et al. 1999; Nielsen et al. 2005; Pearson and Rossignol 1991) suggests that the ENG patterns and synergies observed in the present study were generated to a large extent by spinal circuits.

Nine groups of synergistic motoneuronal pools controlling hindlimb were identified in this study (Fig. 5A), including groups for hip flexors (SartM and Ip) and ankle flexors (TA and Plong), a group for extensors (MG, LG/Sol, Plant, FHL, and extensor activity of PBSt, if present), a separate group for knee extensor VA, a separate group for hip extensor (SmAB, identified based on deletion analysis), and separate groups for EDL, FHL, PBSt (flexor burst), and RF (with 2 separate bursts).

Comparison of this classification with the classification of synergetic muscle activities during real locomotion (from both our study and that of Krouchev et al.) has shown strict similarity in the synergies uncovered, especially for motoneurons controlling one-joint muscles. The major difference occurred in the classification of EDL, FHL, and SmAB, which according to fictive locomotion data each fell in a separate group (Fig. 5A).

At the same time, ENG of EDL actually represents a two-joint motoneuronal pool (ankle flexor and digit extensor). Similar to real locomotion, the onset of EDL nerve activity during fictive locomotion coincided with the onset of flexor nerve activity (SartM, Ip, TA, Plong), but the tail of its activity was often prolonged beyond the end of bursts of other flexor nerves and varied from preparation to preparation (see Figs. 3A and ​and5A).5A). That probably was the reason why EDL fell into a separate group in our analysis of fictive locomotion data. We believe that during normal level walking the offset in EDL activity is controlled by afferent feedback.

The FDL nerve controls a two-joint FDL muscle (ankle extensor and digit flexor), which is a direct functional antagonist of EDL. Its separation from the common extensor group during fictive locomotion (to which it belongs according to the Krouchev et al. classification) was based on its delayed onset during extension resulting from its antagonism with EDL that expressed a prolonged activity during fictive locomotion (see above). Again, we believe that the onset of FDL activity during real locomotion is controlled by afferent feedback and its interaction with EDL.

The AB recorded in our study during real walking and the biceps femoris (BF) recorded by Krouchev et al. were both included in the common group of extensors (together with MG, LG, Sol, etc.). The SmAB nerve recorded in fictive locomotion usually behaves as a pure extensor (see Figs. 2A and ​and3A)3A) and, according to the MST clustering, also belongs to the extensor group (Fig. 5A). However, this classification failed to pass the deletion test (see Fig. 6D). At the same time, the sometimes inconsistent behavior of SmAB in fictive locomotion during deletions could result from the complex nature of the SmAB nerve, which consists of AB and SmA branches. This issue requires further investigation.

Motoneuronal and Muscle Synergies, Locomotor CPG, and Afferent Control of Locomotion

During locomotion, the activities of most extensor and flexor muscles controlling single joints are locked to the corresponding extensor or flexor phase. However, as seen in the present and many previous studies, muscles spanning more than one joint, such as PB, St, and RF, express more complicated activity patterns (sometimes even exhibiting 2 bursts per walking cycle), which depend on gait, speed, and/or locomotor conditions (Carlson-Kuhta et al. 1998; Halbertsma 1983; Rossignol 1996; Smith et al. 1993, 1998). The complex and nontrivial activity profiles of these muscles and corresponding motoneurons have been considered a strong argument against the bipartite organization of the locomotor CPG proposed by Graham Brown (1914) and expanded by Lundberg, Jankowska, and their colleagues (e.g., Jankowska et al. 1967a, 1967b; Lundberg 1981). Several attempts have been made to resolve this problem within the framework of a bipartite locomotor CPG by suggesting the existence of special intermediate neural networks mediating the control of motoneuronal pools by the bipartite RG (Burke et al. 2001; Orsal et al. 1986; Perret 1983, etc). McCrea, Rybak, and their collaborators (McCrea and Rybak 2007, 2008; Rybak et al. 2006a, 2006b) have recently proposed a two-level CPG organization suggesting that the spinal locomotor CPG contains a bipartite RG and specially organized PF circuitry operating between the RG and motoneurons. According to this concept, PF contains interacting neural populations that are active during a particular phase of the step cycle producing the phase-specific synchronized activation of the corresponding group of synergetic motoneuron pools. A reduced computational model of the two-level CPG model controlling only two antagonist motoneuron pools, flexor and extensor, was developed (Rybak et al. 2006a, 2006b). The model generated a realistic locomotor pattern and reproduced many characteristics of the fictive locomotor pattern including spontaneous non-resetting deletions and various effects of afferent stimulations. However, this reduced model has not explicitly simulated the activity of motoneurons controlling two-joint muscles, such as PBSt and RF, as well as the unique activity profiles of some motoneurons controlling one-joint muscles (e.g., VA). Further development of this model requires knowledge of the organization of synergetic motoneuronal activity, which is necessary for constructing (and suggesting) the neural organization of the PF circuitry allowing the RG (and afferent feedback) to control the specific muscle synergies (predefined genetically or formed during learning or development), including those responsible for generating complex patterns of two-joint muscles, such as PB, St, and RF, and other unique activity profiles, such as that of VA. The results of this work provide necessary insights for the extension of the two-level CPG model (Markin et al. 2011; Shevtsova et al. 2009).

It is necessary to note that the engaged motor synergies may or may not be specific for locomotion. Experiments in frogs suggest that synergies are conserved across a variety of motor behaviors (d'Avella and Bizzi 2005; Hart and Giszter 2004) and that many complex and automatic movements may be produced by sequential activation of the synergies by sensory feedback or supraspinal commands (d'Avella et al. 2003; Kargo and Giszter 2008). One objective of this study was to compare the profiles of ENG activities recorded during fictive locomotion (in the absence of afferent feedback) with the corresponding profiles of EMGs recorded during level walking; the latter is supposed to be regulated/adjusted by afferent feedback. The other objective was to make comparisons between the potential synergies operating during fictive versus real locomotion and to learn how afferent feedback modifies these synergies. The specific question is whether afferent feedback directly controls the PF, and hence affects motoneurons via the synergies engaged by the CPG, or uses synergies independent of the CPG or even directly controls the activity of individual motoneuronal populations.

An important characteristic of this study was that all comparisons were made after preliminary normalization of all ENG and EMG activities to the corresponding locomotion cycle duration and the corresponding durations of flexion and extension (and swing and stance) phases. We believe that this normalization does not affect our main conclusion about the influence of afferent feedback on the locomotor circuitry. If the activity profiles and groups of synergists identified during fictive and real locomotion remain the same after normalization, then a reasonable suggestion would be that afferent feedback does nothing more than control the durations of the locomotion cycle and flexion/extension phases, i.e., controls only the operation of the RG defining the timing of phase transitions and the durations of locomotor cycle and its phases. On the other hand, any differences in the activity of muscle synergist groups between fictive and real locomotion, independent of the durations of locomotor cycle and locomotor phases, allow for a number of reasonable suggestions about the possible specific and/or direct effects of afferent feedback on the activity of these synergist groups or individual motoneuronal pools.

Let us now consider the differences in the activity profiles between fictive and real locomotion (Figs. 3 and ​and4).4). The behaviors of most one-joint flexor and extensor nerves and corresponding muscles during fictive and real locomotion are very similar. In contrast, the behaviors of two-joint muscles and nerves innervating them, such as PB (PBSt), RF, and (with the account of the Krouchev et al. study) EDL and FDL, are different. If we focus for the moment on PB (PBSt) and RF, the comparison of the scatterplot for fictive locomotion (Fig. 5A) with the scatterplots for real locomotion (Fig. 5B in the present report and Fig. 7 in Krouchev et al. 2006) reveals the following. The contents of major identified groups of potential synergist are very similar if not the same. The location of clouds that correspond to major synergist groups (that are active either throughout flexion or throughout extension) is also similar. Therefore, the activity of these synergist groups is most likely controlled by the CPG and/or by afferent feedback via the CPG (synergies inherent to the CPG organization). Furthermore, the major differences between scatterplots for fictive versus real locomotion include the presence and location of the clouds corresponding to two-joint muscles and motoneurons controlling them. This specifically concerns the muscles/nerves (PB/PBSt and RF) that are usually active for a short time period around the flexion-extension (swing-stance) and/or extension-flexion (stance-swing) transitions. Therefore, the activity of these motoneuronal pools and muscles during real locomotion seems to be mostly dependent on afferent feedback and to a lesser degree on the RG.

Perret and Cabelguen (1980) also reported much more variable patterns of motoneurons innervating two-joint muscles (including PBSt and RF) as opposed to consistent activity of motoneurons controlling one-joint muscles. They explained this difference by the complex nature of connections from a half-center CPG to these motoneurons. A recent study in the isolated neonatal rat spinal cord with an attached hindlimb (Klein et al. 2010) has also found a much greater variability in activity patterns of two-joint RF and St muscles compared with the one-joint hip flexor Ip and the other thigh muscles with actions at both hip and knee joints (adductor magnus and semimembranosus). The multiple patterns of RF and St were preserved after deafferentation, suggesting that both the intrinsic CPG properties and afferent feedback may contribute to their behavior.

Activities of Two-Joint Muscles and Mechanical Demands in Different Forms of Locomotion

The above conclusion about the possible role of motion-dependent feedback in regulating activity of two-joint muscles is consistent with results of previous studies on different forms of real locomotion (see below) in which two-joint muscles of the thigh demonstrated mutable activity patterns whereas one-joint extensors and flexors typically maintained similar reciprocal activity. For example, during cat backward walking, RF and sartorius anterior (hip flexors, knee extensors) are activated prior to stance and maintain activity for most of stance (Buford and Smith 1990; Pratt et al. 1996), a pattern that was not observed in forward level walking or fictive locomotion in our study (Fig. 3) but is consistent with the required hip flexor and knee extensor moments during this task (Perell et al. 1993).

During very slow walking, the typical two-burst pattern of PB (and St) observed in forward level walking (Fig. 3B) changes to a one-burst pattern (without the precontact burst; Smith et al. 1993) and becomes similar to type 1 PBSt fictive locomotion pattern (Fig. 3A, inset). It may be speculated that during very slow walking, in which the hindlimb angular momentum is insufficient to place the leg in the appropriate position prior to paw contact (Smith et al. 1993), hip flexor and knee extensor activities are required to accomplish this task. This may explain the absence of the St burst and the occasional presence of RF activity during late swing in forward level walking (Engberg and Lundberg 1969; see also the corresponding RF ENG burst in Fig. 3A). With increasing locomotion speed from slow walking to fast galloping, the demand for deceleration of the hip flexion and knee extension in late swing is increasing and the late swing St activity grows and reaches the magnitude exceeding the late stance/early swing burst; the latter occasionally disappears at very high galloping speeds (Engberg and Lundberg 1969; Smith et al. 1993).

The activity of two-joint thigh muscles changes markedly during upslope and downslope walking. During upslope walking (27° and 45°), the St and PB pre-paw contact activity becomes a stance-related extensor activity with the burst lasting ~40% (27°) or most of the stance (45°; Figs. 8 and 10 in Carlson-Kuhta et al. 1998). This St and PB activity pattern is similar to type 3 PBSt fictive locomotion pattern found in our study (Fig. 3A, inset). The above changes in St activity during upslope walking compared with level walking can again be explained by the increased demands brought by the simultaneous development of hip extensor and knee flexor moments (Fig. 6 in Gregor et al. 2006).

During downslope walking (−27° and −45°), the demands for hip flexion and knee extension activities in stance increase, as reflected in the magnitudes of hip flexion and knee extension muscle moments (Fig. 6 in Gregor et al. 2006). As expected, the two-joint RF increases its activity substantially during stance in terms of the magnitude and duration and is active throughout the entire stance (Fig. 8 in Smith et al. 1998). We did not observe such ENG patterns of RF in fictive locomotion studies (Fig. 3A).

In all above examples, one-joint muscles typically demonstrate simple reciprocal flexion-extension synergies that were observed in both real level walking and fictive locomotion. There is just one exception. During downslope walking, the hip one-joint flexor Ip becomes active during stance with other hindlimb one-joint extensors (VA and Sol), whereas the one-joint hip extensors AB and SmA become silent (Figs. 8, 10 in Smith et al. 1998). This seemingly inconsistent Ip and AB activity corresponds to the task demand to control cat descent using hip flexor activity (Gregor et al. 2006).

It is interesting to note that the activity of two-joint muscles during different tasks in cats and humans including walking, running, jumping, cycling, and isometric exertion of external forces is closely correlated with joint moment demands (for review see Prilutsky 2000). This fact is consistent with the suggestion that the activity of two-joint muscles may be under strong influence of motion-dependent feedback.

The comparison of the normalized ENG patterns recorded during fictive locomotion and the normalized EMG patterns obtained during forward level walking in cats presented in this report is based on the assumption that the MLR stimulation in the decerebrate paralyzed cat evokes locomotor activity patterns that “by default” correspond to level forward walking and can potentially be transformed into activity patterns typical for other types of walking, such as slope forward or backward walking, in the presence of the corresponding motion-dependent afferent feedback and/or additional descending control. This suggestion, however, does not have explicit experimental support so far. We also believe that the asymmetric changes in the phase durations and duty factor (the ratio of stance duration to step cycle duration) with the speed of locomotion (Halbertsma 1983) are provided by afferent feedback regardless of the duty factor of CPG operation (Markin et al. 2010; Spardy et al. 2011a, 2011b), which is currently under debate (Frigon and Gossard 2009; Hayes et al. 2009; Juvin et al. 2007).

Authors are greatly thankful to Dr. David A. McCrea for providing direct access to the extensive database of the Spinal Cord Research Centre, University of Manitoba, Winnipeg, MB, Canada, for allowing use of the experimental data on fictive locomotion from this database, and for very useful discussions of the results obtained in this study and earlier drafts of this article.