Phase dependent modulation of cortical activity during action observation and motor imagery of walking: An EEG study

Action observation (AO) and motor imagery (MI) are motor simulations which induce cortical activity related to execution of observed and imagined movements. Neuroimaging studies have mainly investigated where the cortical activities during AO and MI of movements are activated and if they match those activated during execution of the movements. However, it remains unclear how cortical activity is modulated; in particular, whether activity depends on observed or imagined phases of movements. We have previously examined the neural mechanisms underlying AO and MI of walking, focusing on the combined effect of AO with MI (AO+MI) and phase dependent modulation of corticospinal and spinal reflex excitability. Here, as a continuation of our previous studies, we investigated cortical activity depending on gait phases during AO and AO+MI of walking by using electroencephalography (EEG); 64-channel EEG signals were recorded in which participants observed walking with or without imagining it, respectively. EEG source and spectral analyses showed that, in the sensorimotor cortex during AO+MI and AO, the alpha and beta power were decreased, and power spectral modulations depended on walking phases. The phase dependent modulations during AO+MI, but not during AO, were like those which occur during actual walking as reported by previous walking studies. These results suggest that combinatory effects of AO+MI could induce parts of the phase dependent activation of the sensorimotor cortex during walking even without any movements. These findings would extend understanding of the neural mechanisms underlying walking and cognitive motor processes and provide clinically beneficial information towards rehabilitation for patients with neurological gait dysfunctions.


Introduction
Action observation (AO) can be defined as "the perception of other's action " ( Fadiga et al., 1995 ;Rizzolatti et al., 2001 ), while motor imagery (MI) can be defined as "mental simulation or rehearsal of a movement without any motor output " ( Decety, 1996 ). Both AO and MI are motor simulations that recruit activity in the same cortical regions related to the planning and execution of actual movements ( Hardwick et al., 2018 ; Abbreviations: AO, action observation; MI, motor imagery; AO + MI, action observation combined with motor imagery; EEG, electroencephalography; GRF, ground reaction forces; MNI, Montreal Neurological Institute; ERSP, event-related spectral perturbations; ICA, independent component analysis. that both AO and MI of walking activate the premotor cortex and the supplementary motor area devoted to the execution of walking, suggesting overlap of cortical regions activated in the AO, MI, and actual walking conditions ( Miyai et al., 2001 ;Iseki et al., 2008 ). There are not only overlaps but also differences in cortical activity related to walking between AO and MI ( Iseki et al., 2008 ;Jahn et al., 2004 ;Maffei et al., 2015 ;Malouin et al., 2003 ;Miyai et al., 2001 ). Thus, AO + MI of walking would induce AO-and MI-related cortical activities simultaneously. Recently, it was suggested that the MI-component of AO + MI most likely works as motor simulation of own's action, while the AO-component of AO + MI likely works as an external visual scaffolding of MI ( Meers et al., 2020 ). Thus, AO + MI may also make mental simulation clearer than MI alone by visual guiding. Therefore, the present study focused on the AO + MI of walking which could induce cortical activity more related to actual walking than by either technique alone.
We also focused on modulation of cortical activity according to walking phases. During walking, muscle activation pattern and joint kinematics depend on gait phases ( Arendt-Nielsen et al., 1991 ;Neptune et al., 2009 ;Song and Hidler, 2008 ;Yokoyama et al., 2016 ). The electromyographic and kinematic patterns are related to phase dependent modulation of neural activity such as corticospinal excitability ( Capaday et al., 1999 ;Schubert et al., 1997 ), Hoffmann reflex (H-reflex) ( Capaday and Stein, 1986 ) and cortical activity ( Bradford et al., 2016 ;Bulea et al., 2015 ;Gwin et al., 2011 ;Seeber et al., 2014 ;Wagner et al., 2012 ;Yokoyama et al., 2020Yokoyama et al., , 2019. Interestingly, phase dependent modulation also has been reported to modulate corticospinal excitability during observed and imagined upper-limb movement phases ( Fadiga et al., 1998 ;Lemon et al., 1995 ) or walking ( Kaneko et al., 2018b ), even without the corresponding electromyographic and kinematic patterns. Based on the previous studies, AO + MI of walking possibly modulates cortical activity depending on the phase. Therefore, the purpose of the present study is to investigate gait phase dependent modulation of cortical activity during AO + MI of walking. In addition to AO + MI, AO of walking was examined as a control condition. The present study used electroencephalography (EEG), which has high temporal resolution, to determine cortical activity depending on walking phases. Generally, motor intention has been considered to be associated with decrease of the alpha and beta band power of cortical activity ( Pfurtscheller and Lopes Da Silva, 1999 ). Such power decrease has been observed during AO, MI and AO + MI of upper limb movements ( Berends et al., 2013 ;Eaves et al., 2016 ;Muthukumaraswamy et al., 2004 ;Pfurtscheller et al., 2006 ). Based on this knowledge, our first hypothesis was that, during AO + MI and AO of walking, alpha and beta band power would be reduced compared to resting activity in the sensorimotor, anterior cingulate and parieto-occipital cortices, which have been reported to be involved in AO, MI, and execution of walking in previous studies ( Bradford et al., 2016 ;Bulea et al., 2015 ;Gwin et al., 2011 ;Iseki et al., 2008 ;Jahn et al., 2004 ;Malouin et al., 2003 ;Miyai et al., 2001 ;Seeber et al., 2014 ). In these cortical regions during walking, in addition to the power reduction, the alpha and beta band power changes depend on the walking phase ( Bradford et al., 2016 ;Bulea et al., 2015 ;Gwin et al., 2011 ;Seeber et al., 2014 ;Wagner et al., 2012 ;Yokoyama et al., 2020 ). Therefore, our second hypothesis was that the alpha and beta band power is modulated depending on walking phases during AO + MI and AO of walking. Finally, it was also hypothesized that AO + MI induces greater power reduction and clearer phase dependent modulation than AO alone. Regarding the difference between AO + MI and AO, our previous studies found that increases in corticospinal and spinal reflex excitability during AO + MI is greater than that during AO ( Kaneko et al., , 2018a( Kaneko et al., , 2018b. Facilitation of corticospinal and spinal excitabilities was negatively correlated with alpha and beta EEG powers ( Jarjees and Vu čkovi ć, 2016 ;Lepage et al., 2008 ;Mäki and Ilmoniemi, 2010 ;Sauseng et al., 2009 ;Takemi et al., 2015 ). Thus, AO + MI may have a greater effect on cortical activity than AO because of the additive effect of the MI-component.

Participants
Twelve healthy males (mean ± standard deviation = 25.3 ± 2.5 years old, range = 22-30 years old) with no history of neurological disorders participated in the present study after providing informed consent. All experimental procedures were approved by the local ethics committee of the University of Tokyo. This study was performed in accordance with the Declaration of Helsinki (1964).

Experimental design and procedure
Participants were seated in a chair placed about 1.5 m away from a treadmill (Bertec, Columbus, OH, USA). The participants performed the following three conditions: 1) Rest; 2) AO + MI; 3) AO ( Fig. 1 A). In the Rest condition, the participants sat on a chair for one minute. The instruction given was ' please open your eyes and relax without imagining anything' in Japanese. In both AO + MI and AO conditions, another person walked on the treadmill with a fixed speed of 3.6 km/h. In the AO + MI condition, participants were instructed to observe the person's right leg from the sagittal plane (3rd person perspective) and to imagine that they were walking like the person (kinesthetic MI) for one minute, without overt action. The instruction was ' please observe his right leg and imagine that you are walking according to observed stance and swing phases of walking '. In the AO condition, the participants were instructed to simply observe the person's right leg and to not imagine anything for one minute. The instruction was ' please observe his right leg without imagining anything '. In all conditions, the participants were given the same instructions. They were asked not to move, to keep their legs relaxed, and concentrate on each task during the recording. First, the Rest condition was conducted. The AO and AO + MI conditions were performed six times each in a random order (i.e. six minutes for each condition).

Data acquisition
During each condition, EEG data were acquired with a 64-channel EEG cap (Waveguard Original, ANT Neuro b.v., Enschede, Netherlands) and a mobile EEG amplifier (eego sports, ANT Neuro b.v., Enschede, Netherlands) at a sampling rate of 500 Hz. The 64-electrode configuration was set according to the international 10-20 system ( Fig. 1 B). Reference and ground electrodes were placed on CPz and AFz, respectively. Electrode impedances were less than 30 k Ω (20 k Ω in most electrodes), which was lower than the recommended impedance (below 50 k Ω) for the high-impedance EEG amplifier.
In AO + MI and AO conditions, the vertical component of threedimensional ground reaction forces (GRF) were recorded from force plates under the right and left belts of the treadmill at a sampling rate of 1000 Hz. Furthermore, a trigger signal which starts the EEG and the GRF recording was recorded with the same analog-to-digital converter (PowerLab/16SP; AD Instruments, Castle Hill, Australia) to synchronize EEG and GRF data.

Data analysis
MATLAB 2019b (MathWorks, Natick, MA, USA) was used to perform all the post-processing analyses offline. EEG data analysis was performed using custom programs in MATLAB incorporating functions of EEGLAB 14.1b ( Delorme and Makeig, 2004 ).

EEG pre-processing
The data was band-pass filtered between 1 and 200 Hz with a fourthorder Butterworth filter. Using the "CleanLine " function in EEGLAB, the electrical line noise (50 and 100 Hz) was removed. Then, the filtered data was re-referenced to the common average reference. Artifact subspace reconstruction, which is an artifact rejection method based on were seated in a chair placed about 1.5 m away from a treadmill and performed the following three conditions: Rest, action observation combined with motor imagery (AO + MI), and action observation (AO). (B) EEG signals were recorded using a 64-electrode configuration according to the international 10-20 system. principal component analysis in EEGLAB, was used to attenuate highvariance artifact components (e.g. blink, eye movement, muscle, and heart activity) from the EEG data by comparison with resting EEG data (i.e. the Rest condition) ( Mullen et al., 2015 ). For artifact subspace reconstruction, a lax cut off criterion was set to 8 standard deviation to agree with previous studies ( Artoni et al., 2017 ;Yokoyama et al., 2020 ). The processed EEG data was segmented into epochs of 1900 ms (from 300 ms before to 1600 ms after each heel contact of the observed right foot). The timing of heel contact and toe off of the observed right foot was determined from the vertical component of GRF low-pass filtered with a fourth-order Butterworth filter using a 10-Hz cutoff frequency. A force threshold of 60 N was used. One gait cycle was defined as the heel contact to next heel contact of the same foot.

EEG cortical source localization
Independent component analysis (ICA) was used to separate EEG data into brain activity, blink, eye movement, and other artifacts. One ICA decomposition was performed for each participant over all conditions (i.e., AO + MI, AO, and Rest conditions) using the 'runica' function in EEGLAB ( Bell and Sejnowski, 1995 ;Makeig et al., 1996 ). For each participant and independent component (IC), the DIPFIT toolbox in EEGLAB ( Delorme et al., 2012 ;Oostenveld and Oostendorp, 2002 ) was used to estimate an equivalent current dipole located within a standardized three-shell boundary element head model based on the Montreal Neurological Institute (MNI) standard brain. For the source localization, a template of electrode locations based on the MNI head model was used for all participants. The ICs were used for further analysis if their best-fit equivalent current dipoles were located within the head and accounted for more than 85% of the variance seen at their scalp ( Shirazi and Huang, 2019 ), and if not, their scalp map or spectra were visually identified as components of an eye or muscle artifact ( Jung et al., 2000 ). Consequently, further analysis considered 130 ICs, which is an average ± SD of 10.8 ± 4.2 brain related ICs per participant (ranging between 5 and 18 ICs).

EEG group analyses
For the remaining ICs, event-related spectral perturbations (ERSP), which are frequency power modulations related to an observed walking phase, and power spectral density (PSD) were computed using EEGLAB functions ( Delorme and Makeig, 2004 ). The ERSPs were calculated by computing the power spectra over a sliding window. For time-frequency analyses, two-cycle standard Morlet wavelet transform was used at each frequency from 4 to 40 Hz and progressing through the 1900 ms data epochs ( − 300 − 1600 ms around each heel contact of an observed right foot) in 9.5 ms window length. Then, the ICs across the participants were clustered using EEGLAB functions ( Delorme and Makeig, 2004 ). For clustering of the ICs, feature vectors were created combining differences in dipole location, PSD (frequency range: 4 − 40 Hz), and ERSP (frequency range: 4 − 40 Hz, time range: − 300 − 1600 ms around heel contact). The total 130 ICs from 12 participants were clustered with a k-means clustering algorithm ( k = 12) in EEGLAB. The number of clusters was set to 12 to agree with previous studies ( Bradford et al., 2016 ;Gwin et al., 2011 ;Sipp et al., 2013 ). ICs further than three standard deviations from any of the resulting cluster centroids were identified as an outlier cluster and subsequently eliminated from analysis. Clusters which contain less than half of the participants ( ≤ 7) were excluded from group analysis following previous studies ( Gwin et al., 2011 ;Peterson and Ferris, 2019 ;Yokoyama et al., 2020 ). Cluster centroids located too far from the scalp were also excluded from group analysis (cluster centroid coordinates, x = − 3, y = − 24, z = − 31) because of difficulty of estimating ICs in the deep brain area from scalp EEG. Fig. 2 shows dipole locations of the clustered ICs and centroids visu-  alized in the MNI brain volume. Additionally, Table 1 displays coordinates of cluster centroids and the number of participants and sources contained in each cluster. For each cluster and condition (i.e., AO + MI, AO, and rest conditions), the average PSD was calculated (frequency range: 4 − 40 Hz). For each participant and IC, single trial ERSPs were time-warped to compute gait cycle ERSPs using a linear interpolation function, so that the time-points for an observed right heel contact and toe off occurred at the same adjusted latencies in each epoch ( Gwin et al., 2011 ;Wagner et al., 2012 ). Then, baseline was calculated as their respective mean log power spectrum for all gait cycles for each condition (i.e. AO + MI and AO). For visualization of power modulations across the frequency range, baselinenormalized ERSPs were calculated by subtracting the baseline from each log spectrogram for each individual gait cycle (i.e. gait cycle ERSPs). Finally, the baseline-normalized ERSPs were averaged for each cluster and condition.
A Kolmogorov-Smirnov test revealed the PSD and baselinenormalized ERSP were not normally distributed ( p < 0.05). Nonparametric statistical tests were thus used to evaluate significant differences between each condition. For the PSD, regions of significant difference between each condition were computed with the 2000-iteration bootstrapping method available in EEGLAB ( Delorme and Makeig, 2004 ). For the baseline-normalized ERSP, regions of significant difference from baseline frequency power across the gait cycle were computed with the 2000-iteration bootstrapping method separately for AO condition, AO + MI condition, and the difference between AO and AO + MI conditions. Additionally, false discovery rate (FDR) correction was performed for multiple comparisons of the PSD, baseline-normalized ERSP in each condition and the ERSP between AO + MI and AO. The significance level was set at p < 0.05 (FDR-corrected).
In addition to the above group analysis using all ICs in the clusters, another group analysis using only one IC from each participant in the same cluster was performed (see supplemental information). A representative IC which accounts for the largest variance of the channel data was selected for each participant ( Peterson and Ferris, 2019 ). For the cluster including only the representative ICs, cluster locations, PSDs and ERSPs were averaged and nonparametric statistical tests also were performed with the same analysis method as described above.

Independent component (IC) clusters
Six clusters containing more than half of the participants ( > 7) were located in left sensorimotor (9 participants, 16 ICs), central sensorimotor (10 participants, 15 ICs), right sensorimotor (8 participants, 13 ICs), anterior cingulate (9 participants, 17 ICs), left parieto-occipital (9 participants, 14 ICs), and parieto-occipital cortices (8 participants, 11 ICs). Fig. 2 and Table 1 , respectively, show dipole locations and coordinates of the six clusters. Fig. 1S and Table 1S also show those including only representative ICs. Fig. 3 shows the average PSD for each cluster and condition. For three clusters located in the sensorimotor cortex, the PSDs in the theta (5-7 Hz), alpha (8 − 12 Hz), and beta bands (15 − 30 Hz) significantly decreased in both AO + MI and AO conditions, compared to Rest ( Fig. 3 A-C; FDR-corrected p -value < 0.05). Also, the PSDs in the gamma band (30 − 40 Hz) significantly decreased in both AO + MI and AO conditions, compared to Rest ( Fig. 3 B and C; FDR-corrected p -value < 0.05). For clusters located in the anterior cingulate and right parieto-occipital cortices, the PSDs in the alpha and beta bands significantly decreased in both AO + MI and AO conditions, compared to Rest ( Fig. 3 D and F; FDRcorrected p -value < 0.05). For clusters located in the anterior cingulate, and left and central sensorimotor cortices, the PSDs in the alpha and beta bands during the AO + MI condition were significantly lower than those in the AO condition ( Fig. 3 A, B and D; FDR-corrected p -value < 0.05). Additionally, supplemental group analysis using the representative ICs also showed the alpha and beta power reduction for the sensorimotor, anterior cingulate, and right parieto-occipital clusters (Fig. S2). Fig. 4 shows the average ERSP plots for each cluster in the AO + MI and AO conditions, and AO + MI minus AO. The left column in Fig. 4 demonstrates alpha and beta power modulation depending on observed walking phases in the left and central sensorimotor, anterior cingulate, and parieto-occipital clusters during the AO + MI condition. For the AO, there was phase dependent modulation in the left sensorimotor cluster (the middle columns in Fig. 4 ). On the other hand, in the AO + MI, the phase-dependent modulation was clearer than that in the AO in all the clusters (the left and middle columns in Fig. 4 ). Furthermore, the left and central sensorimotor, anterior cingulate, and right parietooccipital clusters showed obvious differences in the power modulation between AO + MI and AO conditions (the right column in Fig. 4 A,  Fig. 3. Power spectral densities (PSDs) of independent component (IC) clusters . The average PSDs in all ICs per cluster (line) and their standard error (shading) are presented for each condition [black; Rest, red; action observation combined with motor imagery (AO + MI), and blue; action observation (AO)]. A, B and C: PSDs of clusters located in the left, center, and right sensorimotor cortex, respectively. D, E and F: PSDs of clusters located in the anterior cingulate, left, and right parieto-occipital cortices, respectively. Comparisons of the PSD between three conditions, AO + MI and Rest conditions, AO and Rest conditions, and AO + MI and AO conditions were respectively shown in order from the left column. The underbars (gray) denote frequency region of significant differences (the false discovery rate-corrected p value < 0.05).

Event-related spectral perturbation (ERSP)
B, D, and F). Notable power modulations were as follows: (1) for the left sensorimotor cluster, AO + MI increased both alpha and beta powers at the terminal-stance phase, and decreased alpha power after the heel contact and beta power at the early-stance and mid-swing phases ( Fig. 4 A), (2) for the left sensorimotor cluster, AO increased alpha power before the right heel contact and beta power after heel contact, and decreased alpha and beta powers in the early-swing phase ( Fig. 4 A), (3) for the central sensorimotor cluster, AO + MI increased alpha power in the early-swing phase and beta power in the mid-stance phase and decreased alpha and beta power from the swing phase to around the heel contact ( Fig. 4 B), (4) for the anterior cingulate cluster, AO + MI increased alpha and beta power in the mid-stance phase and decreased alpha and beta power in second half of the swing phase ( Fig. 4 D), and (5) for the right parieto-occipital cluster during AO + MI, alpha and beta powers were increased at the early-stance phase and decreased from the mid-swing phase to around heel contact, and alpha and low-beta power were increased at the early-swing phase and decreased at the terminal-stance phase ( Fig. 4 F).
Furthermore, supplemental group analysis showed similar phase modulations although there were slight differences in phase modulation between using all ICs and using the representative ICs (Fig. S3). Therefore, the supplemental results indicate that not only the alpha and beta power reduction but also the phase modulation during AO + MI and AO were robust and not driven by a few participants whose change in the PSD and ERSP is weighted more due to having more ICs in a cluster.

Discussion
We found that PSDs in the alpha and beta bands were decreased in the AO + MI and AO conditions compared with Rest condition in the sensorimotor, anterior cingulate and parieto-occipital cortices ( Fig. 3 ). Moreover, the decreases of the PSDs in the AO + MI were greater than those in the AO in the left and central sensorimotor and anterior cingulate cortices ( Fig. 3 A, B and D). Our results also showed clearer power spectral modulation depending on walking phases during AO + MI than AO ( Fig. 4 ). Parts of the power modulations during AO + MI as follows shared characteristics of those during real walking in previous studies: alpha and beta power reduction in the sensorimotor and anterior cingulate cortices ( Bulea et al., 2015 ;Seeber et al., 2014 ;Wagner et al., 2012 ) and phase modulation in the left sensorimotor cortex, which is beta power reduction at the early-stance and mid-swing phases and alpha and beta power increase at the terminal-stance phase ( Bradford et al., 2016 ;Bulea et al., 2015 ;Gwin et al., 2011 ;Oliveira et al., 2017 ;Yokoyama et al., 2020 ). Therefore, our results suggest that cortical ac-tivity induced by AO + MI, not AO, could better emulate that induced by actual walking.

The activation of the sensorimotor cortex in AO + MI related to actual walking
In the present study, the sensorimotor cortex showed that PSDs in the alpha and beta bands significantly decreased in both AO + MI and AO conditions, compared to Rest ( Fig. 3 A-C). Also, for the left and central sensorimotor cortices, the alpha and beta power reduction during the AO + MI was significantly greater than those in the AO ( Fig. 3 A and B). These results are also in line with previous EEG studies which have reported that the power reduction during AO + MI of hand movement was greater than that during AO ( Berends et al., 2013 ;Eaves et al., 2016 ).
Our previous study showed facilitation of the corticospinal excitability during AO + MI similar to that used in the present study, compared with during rest and AO alone ( Kaneko et al., 2018b ). The facilitation would indicate that AO + MI in the present study was more related to kinesthetic MI than visual MI because kinesthetic MI, but not visual MI, facilitates corticospinal excitability ( Stinear et al., 2006 ). This idea is supported by a previous study which showed that hand kinesthetic MI, but not visual MI, induced similar sensorimotor activation to that found during actual hand movement ( Neuper et al., 2005 ). The alpha and beta power reduction in the sensorimotor cortex was also reported during actual walking ( Seeber et al., 2014 ;Wagner et al., 2012 ). Therefore, power modulation in the sensorimotor cortex during AO + MI in our results would be induced by kinesthetic MI ( Fig. 3 A-C). A recent study suggested that, during AO + MI, kinesthetic MI mainly works as motor simulation of one's own action, while AO likely works as an external visual scaffolding of MI ( Meers et al., 2020 ). Based on this explanation, cortical activity during AO + MI would indicate effects of kinesthetic MI supported by visual information. In AO + MI and AO conditions, participants observed another walking person from a 3rd person perspective. A previous study showed that the premotor cortex was more involved in motor planning during passive robot-assisted walking in a 3D virtual environment using a 3rd person perspective than using 1st person perspective and normal passive robot-assisted walking . Therefore, the power modulation in the sensorimotor cortex during AO + MI and AO suggested that AO of walking from a 3rd person perspective would also involve the activation of the premotor cortex. Specifically, motor planning enhanced by the AO would support kinesthetic MI during AO + MI. Therefore, it was suggested that AO + MI more strongly induces sensorimotor activation than AO alone. Actually, the activation in the premotor and supplementary motor areas during AO + MI was considered to be greater than AO ( Eaves et al., 2016 ;Macuga and Frey, 2012 ;Nedelko et al., 2012 ;Villiger et al., 2013 ).
The activation of the left and central sensorimotor cortices during AO + MI also presented obvious power modulations depending on walking phases ( Fig. 4 A and B). Interestingly, the phase modulations in both the AO + MI and AO + MI minus AO were similar to those during actual walking as reported by previous walking studies ( Bradford et al., 2016 ;Bulea et al., 2015 ;Gwin et al., 2011 ;Oliveira et al., 2017 ;Yokoyama et al., 2020 ). However, the phase modulations during AO differed from those during AO + MI and actual walking. Kinesthetic MI, but not visual MI, would induce similar sensorimotor activation to that found during actual movement ( Neuper et al., 2005 ). Therefore, AO + MI would enhance motor simulation in a manner more related to walking than the AO alone due to kinesthetic MI.
Our results support previous EEG results of sensorimotor activation during walking ( Bradford et al., 2016 ;Bulea et al., 2015 ;Gwin et al., 2011 ;Oliveira et al., 2017 ;Seeber et al., 2014 ;Yokoyama et al., 2020 ). Phase modulation of cortical activity was thought to be caused by both somatosensory inputs (e.g., movement-related afferent inputs) and descending commands. In our study, however, it is unlikely that the somatosensory inputs caused phase modulation during AO + MI, because participants performed all experimental tasks without moving their bodies. Phase modulation during AO + MI in the left sensorimotor cortex, which is the beta power reduction at the early-stance and midswing phases and alpha and beta power increase at the terminal-stance phase ( Fig. 4 A), was also found during actual walking in previous studies ( Bradford et al., 2016 ;Bulea et al., 2015 ;Gwin et al., 2011 ;Yokoyama et al., 2020 ). Therefore, it was suggested that the similar parts of phase modulation between AO + MI and actual walking reflect cortical activity related to descending command but not the somatosensory inputs and motion artifacts. Conversely, differences of modulation in the central sensorimotor cortex might primarily show cortical activity induced by somatosensory inputs during walking. With respect to the medial part of the primary motor cortex, somatosensory inputs have been suggested to mainly affect phase modulation of corticospinal excitability during walking ( Kamibayashi et al., 2009 ).
The right sensorimotor cortex did not show clearer phase modulation during AO + MI and AO when the left and right sensorimotor cortices were compared. The lateralization of the sensorimotor cortices is because participants were asked to observe and imagine the right leg but not the left leg. It is also possible that the lateralization is related to that during actual walking. Previous EEG studies indicated that, during actual walking, the right sensorimotor cluster was not identified from a majority of the participants ( Bulea et al., 2015 ) and showed a weaker response than the left sensorimotor cluster ( Bruijn et al., 2015 ;Sipp et al., 2013 ;Yokoyama et al., 2020 ). The left sensorimotor activation was suggested to associate with gait stability such as foot placement and balance control ( Bruijn et al., 2015 ;Sipp et al., 2013 ). In the present study, the participants were asked to imagine according to the observed stance and swing phases, and thus would focus on foot placement in AO + MI and AO conditions. Therefore, lateralization during AO + MI and AO may be related to the association of a stronger activation in the left sensorimotor cortex with foot placement.

. The higher activation of the anterior cingulate during AO + MI than that during AO
The anterior cingulate cortex showed decreases of PSDs in the alpha and beta bands during AO + MI and AO conditions, compared to Rest ( Fig. 3 D). Motor-related functions of the anterior cingulate are known to be attention processing, action monitoring, and error detection and correction ( Bush et al., 2000 ;Pardo et al., 1990 ;Van Veen and Carter, 2002 ). Also, there are connections of the anterior cingulate with sensorimotor cortex which may mediate the understanding and imitation of the actions of others ( Rizzolatti and Craighero, 2004 ). Thus, the alpha and power reduction in the anterior cingulate cortex during AO + MI and AO would be related to attention, action monitoring, and understanding of the action.
Our results showed that AO + MI induced greater power reduction and clearer phase modulation than AO ( Figs. 3 D and 4 D), suggesting that the anterior cingulate activation during AO + MI is higher than that during AO. A previous study showed higher activation of the anterior cingulate during MI than that during AO probably because MI required the more internally focused attentional process than AO ( Munzert et al., 2008 ). Therefore, higher activation of the anterior cingulate during AO + MI than that during AO may be because AO + MI made participants internally imagine that they were walking in addition to AO. Moreover, the anterior cingulate cortex was considered to work for preventing the execution of an imagined action during MI ( Makary et al., 2017 ). Thus, the activation of the anterior cingulate cortex during AO + MI might reflect inhibition of unconscious movements imagined in order to allow the motor cortex to freely react to MI of walking without the risk of performing any movements.

Clearer phase modulation in the parieto-occipital cortex during AO + MI than during AO
The activity in the parietal-occipital cortex during AO of walking is suggested to play an important role in processing visuospatial information and detecting biological locomotion ( Iseki et al., 2008 ). Furthermore, the extrastriate body in the occipital cortex, which is adjacent to the inferior parietal lobe, was considered to specifically react to visual perception of the human body ( Downing and Kanwisher, 2001 ). In our results, the right parieto-occipital cortex showed alpha and beta power reduction during AO + MI and AO conditions, compared to Rest ( Fig. 3 F). There were also no differences in PSDs between in the AO + MI and AO in the parieto-occipital cortex ( Fig. 3 E and F). These results suggested that the parieto-occipital cortex was involved with visual processing in both AO + MI and AO of walking.
The parieto-occipital cortex showed clearer phase modulation during AO + MI than that during AO ( Fig. 4 E and F), as well as the left and central sensorimotor cortices ( Fig. 4 A and B). The parieto-occipital cortex during MI of walking has been considered to play roles in the precise predictions of the sensory consequences of the motor planning ( Bakker et al., 2008 ). A previous study also suggested that the activation of the supplementary motor area, premotor and primary motor cortices reflect temporal sensory prediction ( Bengtsson et al., 2009 ). It was suggested that somatosensory inputs mainly affect phase facilitation of corticospinal excitability during actual walking ( Kamibayashi et al., 2009 ). The instruction for AO + MI, which is ' please observe his right leg and imagine that you are walking according to observed stance and swing phases of walking ', made participants focus on observed walking phases. Furthermore, visual information by AO would support participants to clearly imagine walking and to predict somatosensory inputs. Thus, phase modulation in the sensorimotor and parieto-occipital cortices during AO + MI might be caused by sensory prediction of somatosensory inputs according to phases during walking.

. AO + MI activates the sensorimotor cortex and corticospinal and spinal excitabilities
Our recent study showed that corticospinal excitability during AO + MI were greater than that during rest and AO, especially at the terminal swing phase ( Kaneko et al., 2018b ). Previous studies showed that kinesthetic MI, but not visual MI, induces the facilitation of corticospinal excitability ( Stinear et al., 2006 ) and sensorimotor activation ( Neuper et al., 2005 ). Thus, AO + MI in the present study similar to that in our recent study would mainly include kinesthetic MI. The present results showed that, in the sensorimotor cortex, the alpha and beta power in AO + MI was lower than that in Rest and AO conditions ( Fig. 3 A-C). Moreover, ERSP of AO + MI minus AO in the left and central sensorimotor cortices showed the alpha and low-beta power reduction around the terminal swing phase ( Fig. 4 A and B). Additionally, we recently showed facilitation of spinal excitability assessed by H-reflex and transcutaneous spinal cord stimulation techniques during AO + MI ( Kaneko et al., , 2018. Therefore, it was suggested that kinesthetic MI of AO + MI would activate the sensorimotor cortex and excitability of corticospinal tract and the spinal circuit. Previous studies suggested a relationship between alpha and beta power reduction in the sensorimotor cortex and corticospinal and spinal excitabilities during rest, motor execution and preparation, but not during AO and visual MI ( Jarjees and Vu čkovi ć, 2016 ;Lepage et al., 2008 ;Mäki and Ilmoniemi, 2010 ;Sauseng et al., 2009 ;Schulz et al., 2014 ;Takemi et al., 2015 ). The relationship would be related to kinesthetic MI because kinesthetic MI, but not visual MI, induces power modulation in the sensorimotor cortex and facilitation of corticospinal excitability ( Neuper et al., 2005 ;Stinear et al., 2006 ). Thus, power modulation in the sensorimotor cortex during AO + MI in the present study might be related to excitability of corticospinal tract and the spinal circuit in our previous studies mainly due to kinesthetic MI. On the other hand, power modulation in the anterior cingulate and parieto-occipital cortices during AO + MI might be related to sensory and cognitive processes.

Clinical and technical implementation
In clinical studies, rehabilitation involving AO and MI of walking have been reported to be effective in restoring gait ability for the patients with neurological disorders such as stroke and Parkinson's disease ( Bang et al., 2013 ;Buccino, 2014 ;Dunsky et al., 2008 ;Linden et al., 1989 ;Pelosin et al., 2010 ). This study indicates that the sensorimotor activation during AO + MI was more related to actual walking than AO. Thus, rehabilitation using AO + MI might induce cortical plasticity that supports walking more than by using AO alone.
Elucidation of neural mechanisms underlying AO and MI of walking could also contribute the development of brain machine interface (BMI) for restoring or assisting the patients with neurological gait disorders. Recent studies combining MI with BMI, and exoskeleton or functional electrical stimulation for the lower limbs, reported that long-term BMI training could improve gait ability ( Donati et al., 2016 ;Selfslagh et al., 2019 ). Phase modulation of cortical activity during AO + MI would allow patients to operate mechanical devices skillfully via BMI. Thus, combination of AO + MI and BMI may strengthen the link between cortical activity and use of these devices and provide more effective rehabilitation for motor function recovery.

Limitations of the study
In the present study, our instruction of the AO + MI condition would lead participants to conduct kinesthetic MI and AO from a 3rd person perspective. However, there is a possibility that the AO might make participants unconsciously perform visual MI because it is difficult to completely separate kinesthetic and visual MI when they simultaneously conducted MI and AO from a 3rd person perspective. The activation of the parieto-occipital cortex during AO + MI might reflect unconsciously visual MI.
The present study labeled clusters that correspond to a relatively wide range of cortical areas (i.e., sensorimotor and parieto-occipital clusters) rather than specific cortical areas. EEG data were acquired with 64 electrodes which are comparable with the minimum number of electrodes recommended for the source localization with reasonable accuracy ( Lantz et al., 2003 ). Also, estimation of an equivalent current dipole was performed based on the MNI standard brain. If the present study had used individualized brains for the estimation, a higher density EEG system (e.g., 128 or 256 channel EEG system), or electrode locations digitized for each participant, clusters and ICs would show more specific cortical activity.
Our results showed lateralization of sensorimotor activation between left and right sides. However, it would be interesting to investigate if the lateralization flips accordingly in a control experiment in which participants focus on the left leg. Another control experiment would be also interesting in which participants would focus on both legs and imagine symmetrical walking.

Conclusion
In conclusion, our results showed clearer phase modulation during AO + MI than that during AO in the whole cortex. In particular, the modulation during AO + MI in the left sensorimotor cortex was similar to that which occurs during actual walking, as in a previous study ( Yokoyama et al., 2020 ), suggesting that AO + MI could induce cortical activity more related to actual walking than AO alone. These results reflect enhanced motor simulation of walking by combined use of AO and MI. Our findings reveal the neural mechanisms underlying cognitive motor processes and lead to better neurorehabilitation strategies for patients with neurological gait dysfunctions.

Data and code availability statement
Data presented in this manuscript is newly acquired for the present study. Due to privacy issues of data, it is not available to the community via open repository. The datasets generated during the present study are available from the corresponding author on reasonable request. Considerations will be made based on the review of reasons for requesting the data and the procedures for ensuring data privacy.