All experimental procedures were approved by the Seoul National University Hospital Animal Care and Use Committee (IACUC No.13-0314). Two adult male rhesus monkeys (Macaca mulatta, Monkey 23 and Monkey 24) were included for the experiments. Two 32-channel platinum ECoG electrode arrays (Neuronexus, USA) were implanted in each hemisphere of the monkeys, covering from the PMd to part of the PPC (Figure 1A; Markov et al., 2012). The diameter of the ECoG electrodes was 300 μm and the inter-electrode distance was 3 mm. All electrodes were implanted in the epidural space. ECoG electrodes were connected to connectors (Omnetics, USA) affixed to the skull with dental cement and titanium screws (Vet implants, USA). Customized titanium head holders were also implanted to fix the monkey's head in place.

The monkeys were trained to perform a simple visual guided reaching task paradigm (Figure 1B). In each experiment, the monkey was seated in a customized primate-chair, and head movement was restrained by a head holder. We installed a total of five buttons in front of the monkey for the behavioral task. Two buttons, which were placed close to the chair, were used for ready indicators and three target buttons, located around 30 cm away from the monkey, were used to indicate the target. The trial began when the animal pushed the ready buttons. After a few, randomly spaced seconds (3–5 s), while the monkey was holding the ready buttons, the target button light was then lit. The target button could flash in two colors, i.e., red or blue, each color was associated with the hand that the monkey should move. For example, if the target button flashed red, the monkey should react using its right hand and, accordingly, its left hand for the blue light. The monkeys were instructed to reach and push the target button within 1 s. If the monkey reached the target on time using the correct hand, the trial was finished and a juice reward was given. The mapping between color and hand was maintained constant throughout the training and experiments and the order of the stimulus presentation was randomized. The experiment (1 session) was conducted once a week and within a session, the monkeys performed 200~400 trials, depending on the condition of the monkey on that day. There were 11 sessions for monkey 23 and 23 sessions for monkey 24, respectively.

Electrical recordings started 2 weeks after recovery from the surgery. ECoG signals were digitized at a sampling rate of 1 kHz (EEG 1200, Nihon Kohden, Japan). Each button's events (press or release) were stored in the system for triggers. The movement of the monkey was also simultaneously captured at a sampling rate of 20 Hz by a wireless motion tracking system (Xsens, The Netherlands). For all experiments, each monkey wore a custom-made jacket with motion trackers affixed to the left and right shoulders, elbows, and wrists.

The whole data were analyzed offline using a customized version of Matlab (The Mathworks, Natick, MA, USA) scripts using functions from the EEGLab open-source toolbox. The data were cleaned from possible line noise artifacts with a notch filter at 60 Hz and higher harmonics (120 and 180 Hz) and a band-pass filtered from 1 to 200 Hz. Using visual inspection, channels that did not clearly contain ECoG signals (e.g., such as channels that contained flat signals or noise due to broken connections) were removed prior to the analysis. Overall, these procedures reduced the total number of channels to 63 and 58 in monkey 23 and 24, respectively (Figure 1A). We then divided the continuously recorded data into non-overlapping analysis epochs with a 2,000 ms length from 1,000 ms before the target cue light to 1,000 ms after. Bad trials containing excessive noise and artifacts were manually excluded. Moreover, we only included the correct trials in the further analyses.

To define the connectivity between separate electrodes, we adopted the weighted phase lag index (wPLI) (Vinck et al., 2011). The wPLI is an extension of the phase lag index (PLI) which defines connectivity as an absolute value of the average sign of phase angle differences between two channels. The wPLI modified the PLI by deweighting vectors using the imaginary component of the cross-spectrum as a factor. Using this procedure, phase differences around zero or 180° (closer to the real axis) contribute only marginally to the final connectivity estimate. The wPLI can reduce the probability of detecting “false positive” connectivity where volume conducted noise sources have near zero and 180 phase lags and increases the sensitivity in detecting interactions when the interacting sources are spatially close (Vinck et al., 2011; Ewald et al., 2012; Haufe et al., 2013). wPLI=|∑i = 1NIm(Xi)∑i = 1N|Im(Xi)||= |1n∑i = 1N|Im(Xi)|sgn(Im(Xi))|1n∑i = 1N|Im(Xi)| where X_i is the cross spectrum of two signals in trial i of total trial N, sgn is the sign of variables and Im is the imaginary part of complex value.

To estimate the wPLI, the data were converted to the time–frequency domain via convolution with a family of complex Morlet wavelets, defined as Gaussian-tapered complex sine waves. We used 39 linearly spaced frequencies between 2 and 40 Hz, and the number of cycles was increased from 3 to 12 in logarithmic steps (Cohen, 2015). Convolution was performed via frequency-domain multiplication, in which the Fourier-derived spectrum of the EEG data was multiplied by the spectrum of the wavelet, and the inverse Fourier transform was taken. Power and phase were defined as the squared magnitude of the complex result and the angle relative to the positive real axis respectively.

To test whether the phase synchronization was significant under individual conditions, we applied jack-knife leave-one-out estimates of the standard error of wPLI and used the estimated standard errors to calculate significance with regard to the normal distribution of a mean of 0 and a standard deviation of 1 (Efron and Efron, 1982; Efron and Tibshirani, 1998). The p-values were then corrected for false discovery rate (FDR) in the time-frequency dimensions using parameters proposed by previous study (Genovese et al., 2002) with a Q = 0.2. Statistical masks were generated for corrected values of ^*p < 0.05, a one tailed (which is appropriate for a one-sample-location test used for the wPLI that ranges from zero to one). For further analysis, we only included the significant values based on statistical mask.

We firstly collected behavioral data on the monkeys. During the 34 recording sessions (11 for monkey 23 and 23 for monkey 24), the number of correct trials was 1,445/1,597 (left and right) in monkey 23 and 2,748/2,893 in monkey 24. Response time was measured as the time elapse from the cue presentation to the time when the monkey executed movement. As shown in Figure 1C, both the right hand responses (260.54 ± 40.43 and 317.85 ± 98.11 ms) were faster than the movement of the left hand (375.14 ± 64.97 and 343.55 ± 90.40 ms) (paired t-test, p < 0.005). These asymmetrical aspects can also be seen in the results for correct rate. Correct rates were calculated as an average of the performance of each session. As shown in Figure 1D, both monkeys showed a higher performance for the right hand (93.76 ± 7.60 and 88.53 ± 6.90% in monkey 23 and monkey 24, respectively) than the left hand (88.24 ± 7.10 and 78.20 ± 9.75% in monkey 23 and monkey 24, respectively) (paired t-test, p < 0.05). These results showed that, in terms of task complexity or demand, the right condition appeared to be easier for these monkeys compared with other condition.

In order to represent local neuronal activity during visually guided reaching movement, we calculated the event-related spectral perturbation (ERSP) across trials. Figure 2A shows an example of raw signal trace of the single trial in monkey 23 on the specific channel of the bilateral PMd and PPC shown on the left. Then, Figures 2B–I represent the ERSP results which aligned to the movement cue presentation for the channels in Figure 2A. The results showed that activation of the beta and gamma frequency band increased in both animals immediately after the movement cue was presented. Notably it was observed in both areas and in both hemisphere. We then attempted to determine whether the movement cue might modulate the functional connectivity between the epidural ECoG electrodes. To this end, we extracted the Fourier spectra from 2 to 40 Hz for each condition, and computed phase coherence activation using the wPLI. The wPLI represents the strength of phase coupling between two channels, similar to the phase locking value (PLV) or coherence (where 0 indicates no phase coupling, 1 indicates maximum phase coupling), but is not spuriously increased by volume conduction of single sources to 2 sensors or common references.

Figure 3 shows an example of the coherence between the bilateral PMd and bilateral PPC for both monkeys. For visualization, we selected representative channel pairs over each region (indicated in right panel in Figure 3). Consistent with previous findings (Pesaran et al., 2008), both animals showed a strong coherence at the low beta frequency band (around 10–20 Hz) between right PMd and right PPC (Figures 3B,D). The coherence emerged immediately after the cue presentation (time = 0, indicated by a solid white line) as a prominent peak near 100 ms and disappeared before the actual movement response. In contrast, the coherence in low-beta band between left PMd and left PPC was weaker than the coherence of the right pair in monkey 23 or barely observed in monkey 24 (Figures 3A,C, respectively). In monkey 23, a strong coherence of higher frequency (around 25–35 Hz) was also observed prior to the onset of the task-relevant cue and offset immediately after cue presentation, however these data were not included in further analysis due to inconsistency across animals. To test whether the coherence is related to the decisional stage or motor process, we tried to calculate the wPLI coherence by aligning the same data on the latency of movement onset. As a result, we observed that the peak value decreased compared to when the data were time-locked to the cue onset (Supplementary Figure 1). Therefore, we speculated that beta-coherence might be closer to the decisional or perceptual process of task-related cue.

To summarize the overall coherence patterns across bilateral fronto-parietal network and to specify where the coherence emerged in the spectro-temporal domain, we then calculated the average montage of wPLI values along frequency and time axes between every possible pairs over each region (Figures 4, 5). Figure 4 represents the averaged wPLI in the spectral domain which was calculated from 0 to 0.2 s along time axes. Consistent with Figure 3, we confirmed that the coherence of the low beta with a prominent peak near 15 Hz was strongly enhanced in both monkeys. Moreover, we also found that the coherence of the right fronto-parietal pair was stronger than that of left fronto-parietal pair. As illustrated in Figure 4, right PMd was strongly coordinated with right PPC (red line) while the coherence between left PMd and left PPC (blue line) was relatively low in both monkeys. Interestingly, such asymmetry remained regardless of hand movement side. Whether the monkeys moved right hand or left hand, the coherence within right hemisphere was stronger than left hemisphere.

The temporal profile of 10–20 Hz coherence, illustrated in Figure 5, suggests that the temporal evolution of coherence emerged shortly after cue presentation (time = 0). Note that both monkey's mean movement execution time (indicated by the dotted white line in Figure 3) was ~300 ms; thus the coherences mostly disappeared before the actual onset of movement. Both monkey's mean response took 299.99 and 313.03 ms in monkey 23 and monkey 24, respectively (Figure 1C). This result suggests that the phase coherence in this low beta band may signify a visual perception and decision related network that is functionally activated immediately posterior to the onset of task-relevant stimuli in both animals.

To illustrate a topographical overview of the low beta coherent network, we calculated the adjacency matrices during the post-cue period for all conditions (Figure 6). Each matrix was calculated by averaging the wPLI coherence values across corresponding electrode pair within the specific brain regions where the averaging window was defined as 0–0.2 s and 10–20 Hz. Only the statistically significant values (inside black contour line in Figure 3, Adjusted p < 0.05) were included for average calculation. We employed the Wilcoxon signed-rank test to determine whether each of the values were significantly greater than their median value to exclude the case where the whole matrix value increases due to the volume conduction effect. As result, we found a right lateralized fronto-parietal network pattern in both monkeys. Consistent across animals, the right PMd was most strongly connected with bilateral PPC, including S1 and M1 (Wilcoxon signed-rank, p < 0.001), while the left PMd was weakly or barely coordinated with the bilateral parietal area in both animals for all conditions (Figure 6). Additionally, to quantify the similarity of the low beta network across animals, we calculated the Pearson correlation coefficient for the adjacency coherence matrices of both monkeys. We found a strong significant correlation between the two matrices for every condition (across cells in the matrices) (r = 0.7539, p < 0.0001 and r = 0.6678, p < 0.0001 for left hand and right hand, respectively).

We further summarized the averaged wPLI between areas using bar plots. Figure 7 shows the distribution of the averaged coherence of electrodes from each PMd to other areas. Consistent with the adjacency matrix (Figure 6), Figure 7 demonstrates the right PMd was strongly coordinated with the bilateral PPC area across both animals, while the coherence between the left PMd and bilateral PPC was not enhanced. We also compared the averaged wPLI value between the effector's side (left hand or right hand). There was a statistically significant difference for the overall coherence being higher in the right hand condition than left hand movement in both monkeys.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.