Ten healthy volunteers (S1–S10, five female, five male, 25.6 ± 3.5 years) participated in this study. The experimental procedure was approved by the ethical committee of the Medical University Graz. Each subject gave informed consent before the experiment. Participants completed 4 runs (6 min each) of active walking and 3 runs of upright standing (3 min each) in a robotic gait orthosis (Lokomat, Hocoma, Switzerland). Walking speed was constant and adjusted to the participants’ leg length ranging from 1.8 to 2.2 km per hour. Body weight support was adjusted with the help of experienced physical therapists to less than 30% in every participant. The Lokomat was operated with 100% guidance force. This set-up was chosen to ensure a well-controlled and steady gait pattern during the experiment. Participants were trained to walk in a natural way in the Lokomat and were asked look straight ahead and to blink normally to avoid eye artifacts during the experiment.

120 EEG channels were recorded by combining four 32-channel amplifiers (BrainAmp, Brainproducts, Munich, Germany). To determine the electrode positions and anatomical landmarks (nasion, vertex, left- and right pre-auricular points) for each subject, we used a 3D localizer (Zebris Elpos system, USA). Structural T1 MRI scans were recorded in a post-screening session using a 3.0 Tesla (Tim Trio/Skyra, Siemens, Erlangen, Germany) scanner. EEG was sampled to 2.5 kHz, high pass filtered at 0.1 Hz and low pass filtered at 1 kHz. The electrode montage was in accordance with the 5% international 10/20 EEG system (EasyCap, Germany; Oostenveld and Praamstra, 2001). Reference and ground electrodes were placed on the left and right mastoids respectively. Electrode impedances were <10 kΩ. Foot contact was measured by electro-mechanical switches placed over the calcaneus bone at the heel of both feet. We defined one gait cycle as the interval between two right leg heel contacts. More detailed information about the experimental set-up and procedure can be found in Wagner et al. (2012).

Event-related desynchronization and synchronization topographies showed significant ERD in sensorimotor areas for the μ and β rhythm. Significant β ERD was visible focused in central sensorimotor areas (Figure 2A) in 8/10 Subjects. S6 showed weak, but still significant activity in central sensorimotor areas, while S10 showed a different pattern in the sensorimotor area. We also observed significant μ ERD for most of the subjects (Figure 2B), which were spatially less consistent across subjects than the β activity patterns. S6 and S8 showed fewer than 150 sources with significant activity in the sensorimotor cortex for the β topography. So, we used a smaller cluster of 69 sources for S8. The focal μ cluster in the central sensorimotor areas was downsized to a size of 150 sources and used as ROI for S6. Activities in Figure 2 illustrated in red are likely to be caused by EMG artifacts and not by ERS brain activities.

Amplitude changes were found in central sensorimotor ROI, both from walking to standing and across the gait cycle (Figure 3). ERD/ERS activities during the gait cycle show sustained μ and β ERD during the entire gait cycle (Figure 3A). There were pronounced amplitude modulations relative to the mean gait cycle activity at 25–40 Hz with similar temporal dynamics locked to the gait cycle in 9/10 Subjects (Figure 3C). Subject 10, who already showed unusual μ and β brain topographies, also featured different TF patterns. A comparison of the temporal dynamics of TF magnitudes at μ and β frequencies compared to GPM center frequencies revealed that the GPM magnitudes were significantly higher (Wilcoxon signed rank) at the GPM peak frequency than at the μ (p = 0.0049) and the β (p = 0.002) center frequencies (Table 1; Figure 3D). While sustained ERD is by definition maximal at β center frequencies, the temporal amplitude modulation is bigger at the peak GPM frequencies (Figures 3A,D). Moreover, the GPM frequencies were significantly different (Wilcoxon signed rank) from the μ (p = 0.002) and β (p = 0.0137) center frequencies (Figure 3B). This indicates that sustained μ and β ERD and GPM occur in different neuronal rhythms. The GPM activities were localized to the central sensorimotor areas for 8/10 Subjects.

The ERD/ERS brain topographies showed β ERD patterns in central sensorimotor areas. These spatial patterns are consistent with the classical somatotopic location of lower extremities in the human motor cortex (Jasper and Penfield, 1949). Furthermore, these patterns coincide with results for invasive electrocorticographic (ECoG) recording studies that showed β spectral power decreased in central sensorimotor areas during isolated leg movements (Crone et al., 1998; Miller et al., 2007). The variability of functional somatotopy across individuals has been reported before (Crone et al., 1998; Miller et al., 2007). Yet, the inter-subject consistent β ERD pattern located in central sensorimotor areas was a robust feature of our study. ERD/ERS topographies illustrate general, conditional spectral changes between active walking and upright standing, since there was no temporal information considered in this measure. ERD in the μ and β rhythm consequently describe general state changes for these oscillations during walking relative to a non-movement condition. ERD was interpreted as an electrophysiological correlate of activated cortical areas that are involved in sensory or cognitive processing, or in the production of motor behavior (Pfurtscheller and Lopes da Silva, 1999). Following this theory the sustained ERD reflects an active state of the sensorimotor areas during walking.

In addition to sustained μ and β ERD, during the whole gait cycle, oscillatory amplitudes are modulated relative to the gait phase in the low γ band. These two phenomena occur simultaneously during walking and are superimposed, both in spatial location and frequency range. The GPM values are significantly larger in the low γ than at μ and β center frequencies (Figure 3, Table 1). This finding shows that GPM values, which are largest for frequencies between 28 and 36 Hz, cannot be explained exclusively by modulations of the μ and β rhythm. These modulations in the low γ are strongly linked to the gait cycle phase (Figure 3C, Table 1). The GPM maxima are clustered and located isolated in central sensorimotor areas (Figure 4). This localization suggests that the GPM are caused by brain signals and not by EMG activities, considering the location of head muscles and electrical volume conduction. Moreover, large GPM is present in a limited frequency band (24–40 Hz), not as broadband activity, which was reported to be associated with motion or muscular artifacts during walking (Castermans et al., 2014). The gait phase related rhythm in the low γ band was previously reported by our group (Wagner et al., 2012) using ICA and dipole reconstruction for localization. Again, the location of particular independent components is not necessarily congruent with the location of ERD/ERS activity or the GPM measures we introduced in the present work. Here, we directly localized μ and β ERD and low γ GPM using inverse modeling in a distributed source model. Moreover, our results show that μ and β ERD occur at different frequencies than the GPM, which are driven by a different rhythm in the low γ range.

The superposition of μ and β ERD and GPM is shown in Figure 3, where the overlay of these two phenomena can be determined to be dependent upon the frequency overlap (Figure 3B, Table 1) of the β ERD and the GPM center frequencies. In some subjects, the β ERD “covers” the GPM in these plots. The isolated gait phase related modulations are illustrated in Figure 3C. Because gait cycle mean activities were used as a reference, sustained ERD/ERS activities disappear in these plots. The time courses in Figure 3D additionally outline the coexistence of sustained ERD and GPM. Depending on the difference of the β and GPM center frequencies, these time courses differs in terms of their offset (ERD/ERS) or their temporal modulation (related to GPM).

We investigated robot-assisted walking in the Lokomat for several reasons. First, the robotic gait orthosis ensures a very steady gait pattern during the experiment. Second, the participants’ trunk can be stabilized to some extent using body weight support, which reduces movement artifacts. Third, we plan to study electrophysiology during stroke rehabilitation in future studies using the Lokomat. Nonetheless, Lokomat walking is similar but still different from treadmill walking (Hidler and Wall, 2005; Hidler et al., 2008). The generalizability of our findings therefore has to be investigated for everyday walking tasks in the future. However, our findings may be helpful both for analysis of EEG data in more naturalistic human walking tasks on one hand as well as ultimately for clinical research on the other hand.

Lower extremities are both represented close to the midline in the human motor cortex (Jasper and Penfield, 1949). Due to this cortical location and the low spatial resolution of the EEG, we are limited to reporting summed activity from both feet. In this study, it is not possible to discuss EEG activity in relation to the movement of a particular foot. For example, increased amplitudes during 10–30% of the gait cycle could be related ether to the initial and midswing phases (10–37%) of the left foot or to the mid stance phase (10–30%) of the right foot. We are therefore careful in discussing TF activity in relation to certain phases of a particular foot.

In this section, we discuss our findings in relation to previous studies in terms of frequency ranges of the gait modulated oscillations and their cortical origin. The frequency ranges we report in this study for sustained β ERD (18–30 Hz) and the GPM (24–40 Hz) are very similar to corticomuscular coherence (CMC) peak frequencies from a recent study. In Petersen et al. (2012), the synchrony of EEG and EMG signals were investigated during treadmill walking. The authors reported CMC peaks located at the vertex (at the Cz electrode) for frequencies at 15–30 Hz for static contraction of the anterior tibial muscle, but for frequencies of 24–40 Hz for slow and normal walking. The drift of CMC peak frequencies from the β range during isometric movement tasks toward the low γ range during phasic movements has been previously discussed for upper limbs (Marsden et al., 2000; Omlor et al., 2007) and lower limbs (Gwin and Ferris, 2012). Considering these findings and assuming that different neuronal oscillations reflect different functional networks (Buzsáki and Draguhn, 2004; Schnitzler and Gross, 2005), the CMC peak drift from β frequencies during isometric movement toward low γ frequencies during phasic movements could be explained by different cerebral networks associated with the β rhythm and the low γ we identified and distinguished in this work.

The presence of β oscillations is related to maintenance of the current motor set (Engel and Fries, 2010) and promotes tonic activity at the expense of voluntary movement (Gilbertson et al., 2005; Brown, 2007; Pogosyan et al., 2009; Jenkinson and Brown, 2011). Thus, β ERD may reflect the suppression of an inhibitory network and signifies a neuronal state during walking that enables voluntary movement. We argue that β ERD during walking reflects a sustained active, movement related neuronal state, which is present during the whole gait cycle.

The most recent work from our group investigated different interactive virtual environment (VE) tasks during robot-assisted walking (Wagner et al., 2014). The authors reported low γ (23–40 Hz) gait cycle related modulations in the premotor cortex to be dependent on the VE task. This dependence was greatest during 10–30 and 60–80% of the gait cycle and was suggested to represent processes involved in motor planning. In local field potential (LFP) recordings, low γ frequencies at 25–40 Hz have been shown to modulate the firing rate of macaque primary motor neurons during a center-out brain-machine interface task (Canolty et al., 2010). γ synchronization is a fundamental process in cortical computation (Fries et al., 2007; Fries, 2009) and facilitates the coordination of distributed functional cell assemblies (Canolty et al., 2010). Consequently, the gait phase modulated low γ oscillations we report could be involved in gait phase dependent local synchronization of neuronal populations linked to sensorimotor processing or integration. The particular involvement of low γ oscillations in motor control will be investigated in future work.

μ and β ERD and gait phase related amplitude modulations are simultaneously present during walking. Following the view that neuronal oscillations at different frequencies are involved in different cortical networks (Buzsáki and Draguhn, 2004; Schnitzler and Gross, 2005; Siegel et al., 2012), our findings suggest that the sustained μ and β ERD reflect altered states of the associated networks during walking. Furthermore, another network related to the low γ rhythm may be modulated dynamically locked to the gait cycle phase. Gait phase related low γ amplitude modulation and sustained μ and β suppression may therefore be organized in different neuronal networks.