The sensorimotor beta rhythm (∼13–30 Hz) is commonly seen in relation to movement. It is important to understand its functional/behavioral significance in both health and disease. Sorting out competing theories of sensorimotor beta is hampered by a paucity of experimental protocols in humans that manipulate/induce beta oscillations and test their putative effects on concurrent behavior. Here, we developed a novel behavioral paradigm to generate beta and then test its functional relevance. In two human experiments with scalp EEG (n = 11 and 15), we show that a movement instruction generates a high beta state (postmovement beta rebound), which then slows down subsequent movements required during that state. We also show that this high initial beta rebound related to reduced mu–beta desynchronization for the subsequent movement and, further, that the temporal features of the beta state, that is, the beta bursts, related to the degree of slowing. These results suggest that increased sensorimotor beta in the postmovement period corresponds to an inhibitory state—insofar as it retards subsequent movement. By demonstrating a behavioral method by which people can proactively create a high beta state, our paradigm provides opportunities to test the effect of this state on sensations and affordances. It also suggests related experiments using motor imagery rather than actual movement, and this could later be clinically relevant, for example, in tic disorder.
Beta oscillations (13–30 Hz) are commonly observed in sensorimotor (SM) areas of the human brain. For example, when people make a movement, the amplitude of beta oscillations decreases, referred to as the “event-related desynchronization”, followed by a subsequent above-baseline increase in beta power referred to as the “event-related synchronization” (or) postmovement beta rebound (PMBR; Pfurtscheller & Lopes da Silva, 1999). There is considerable interest in the neural origins and functional role of oscillations, in general, and in the motor domain beta, in particular (reviewed by Schmidt et al., 2019; Spitzer & Haegens, 2017; Kilavik, Zaepffel, Brovelli, MacKay, & Riehle, 2013; Jenkinson & Brown, 2011). There are different theories about SM beta, including those proposing that it reflects the “status quo” of maintaining the current SM set (Engel & Fries, 2010; perhaps compatible with an “inhibitory” function of preventing new motor programs from being executed), the sensory evaluation of feed-forward motor commands versus feedback reafference (Baker, 2007; Androulidakis, Doyle, Gilbertson, & Brown, 2006), the top–down processing of sensory predictions (Arnal & Giraud, 2012), and the clearing out of current information (Schmidt et al., 2019). Moreover, premovement and postmovement beta appear to have different neural substrates of generation (Alayrangues, Torrecillos, Jahani, & Malfait, 2019), and there are also apparent differences between “low” and “high” beta (Kilavik et al., 2012, 2013).
Testing competing theories of beta and understanding its functional significance would benefit from a clear-cut paradigm in humans for demonstrating how beta influences behavior and cognition. One way to do this is to induce a beta state and then “embed” behavior in it. Various studies have tried to do this in different ways. For example, researchers have used real-time neurofeedback to guide monkeys to raise their SM beta to a specific level and subsequently tested its effect on movement (Khanna & Carmena, 2017). Another example is the use of entrainment of beta oscillations via noninvasive brain stimulation to test if this leads to slowing in people (Romei et al., 2016; Pogosyan, Gaynor, Eusebio, & Brown, 2009). A final example is the examination of how the occurrence/timing of endogenous beta oscillations (i.e., preevent natural variation) affects perception in people (Shin, Law, Tsutsui, Moore, & Jones, 2017; Sherman et al., 2016). Although these interesting and challenging studies have shed light on the potential roles of beta, we sought a more straightforward way to test the functional relevance of beta on behavior.
In our novel behavioral paradigm, participants made a primary movement, which led to a strong PMBR in the contralateral SM cortex; we then embedded a cue in the time frame of PMBR—a cue that signaled them to perform a rapid secondary movement. Our rationale for using postmovement beta was, first, that it is a robust and large increase of beta power that confers a high probability of detecting transient beta events in single trials (Feingold, Gibson, DePasquale, & Graybiel, 2015), plausibly even in scalp EEG, and, second, there is strong lateralization of the rebound to the hemisphere contralateral to the movement (Little, Bonaiuto, Barnes, & Bestmann, 2019; Jurkiewicz, Gaetz, Bostan, & Cheyne, 2006). This lateralization effect was useful in allowing us to test the difference of a primary right-hand movement on a subsequent right side movement, versus a primary right-hand movement on a subsequent left side movement. We specifically predicted that beta rebound would slow the subsequent right versus subsequent left side movement. Such a result might be expected on the theory that postmovement beta reflects a return to the “status quo” or an inhibitory state as indicated by the earlier result that the time period of PMBR coincides with decreased corticospinal excitability (Chen, Yaseen, Cohen, & Hallett, 1998) and that, across participants, those with higher levels of PMBR had higher GABAergic concentrations in M1 (measured via magnetic resonance spectroscopy; Gaetz, Edgar, Wang, & Roberts, 2011).
In brief, we developed a new behavioral paradigm to test if PMBR retards subsequent movement, specifically in relation to the side of the body on which it is stronger. We also planned to test how the dynamics of the beta rebound, in terms of the parameters of beta bursts (Tinkhauser et al., 2017, 2020; Little et al., 2019; Shin et al., 2017; Feingold et al., 2015), relates to behavior on the same trial.
Experiment 1: Participants
There were 11 participants in Experiment 1 (nine women, mean age = 20.9 ± 0.8 years). All participants were right-handed and provided written informed consent according to a University of California, San Diego institutional review board protocol and were compensated at $20/hr.
Experiment 1: Flex and Move Task
The experiment was coded in MATLAB 2016b (The MathWorks), and stimuli were presented using Psychtoolbox (Brainard, 1997). The task required participants to perform a brisk wrist flexion with their right hand on every trial, and then, on a subset of trials (20%), they had to make a rapid subsequent movement—press a button with either the right or the left hand (Figure 1A). The right-hand wrist flexion was used to generate postmovement beta in the left SM cortex.
Participants sat in front of a monitor with their arms resting on the armrests in a pronated position. They grasped two buttons, one in each hand. The button was on the hilt of a cylindrical base such that, upon grasping the base, they could place their thumb on the button. Each trial began with a cue (white square) in the center of the screen. This instructed the participants to perform a brisk (rapid) wrist flexion using only their right hand. The square remained on the screen for 0.5 sec. Participants tried to finish the movement by the time the cue disappeared. On 80% of the trials, there was a 4.5-sec relaxation period once the square disappeared, followed by a jittered intertrial interval (ITI) of 3 ± 0.5 sec. On the remaining 20% of trials, at 2 sec from the onset of the square (a time period we expected to overlap with postmovement beta), an arrow pointing to either the left or the right was presented on the screen. Participants responded to this arrow as quickly as possible with a left or right thumb press. The arrow stimulus remained on the screen for a maximum of 1.5 sec.
Each participant performed 14 blocks of 30 trials (total: 420 trials), except for one participant who performed 10 blocks of 30 trials.
Experiment 1: EEG
We recorded 64-channel scalp EEG in the standard 10/20 configuration using an Easycap system (Easycap and BrainVision actiCHamp amplifier, Brain Products GmbH). The EEG signals were digitized at 1000 Hz.
Experiment 1: EMG
We recorded surface EMG only in Experiment 1, as there was no response recorded for the right-hand wrist flexion and we needed to identify whether the participants had successfully executed it. We obtained EMG from the right flexor carpi radialis muscle to monitor the right wrist flexion and also from right and left flexor policis brevis muscles for the button press responses using the thumb. EMG signals were amplified ×5000 between 30 and 1000 Hz (Grass QP511 AC amplifier, Grass Instruments), digitized at 2000 Hz (Micro 1401 mk II, Cambridge Electronic Design), and recorded via data acquisition software (Signal Version 4, Cambridge Electronic Design).
Experiment 1: Data Analysis
All the data analyses were performed in MATLAB (R2016b) using custom-made scripts.
Behavior and EMG
We determined the RTs of the button press responses for both the right and the left thumb. We rejected RT outliers by removing trials with RTs lower than the 25th percentile −1.5 × interquartile range (IQR) and higher than 75th percentile +1.5 × IQR. This removed, on average, only 4.1 ± 0.8% of the total secondary movement trials. Then, to see if the participants performed the right wrist flexion, we analyzed the EMG data. The method described here is similar to our recent paper (Jana, Hannah, Muralidharan, & Aron, 2020), the code for which is openly accessible (see https://osf.io/b2ng5). The EMG data were filtered using a fourth-order Butterworth filter to remove 60-Hz line noise. We then computed the root-mean-square of the signal using a centered window of 50 msec. On a trial-by-trial basis, we then estimated the peaks of EMG activity. We used a threshold of mean baseline EMG + 8 SD, where the baseline EMG was determined from a period before the cue to flex (i.e., the white square). The presence or absence of a peak aided us in marking trials where participants successfully performed the flexion or not. We performed a similar analysis and estimated the peaks of EMG activity for the left and right button press as well. We then estimated the EMG onset times (with respect to the secondary movement cue) by first starting at this estimated peak and backtracking to the point where the activity dropped below 20% of the peak for five consecutive milliseconds.
We used EEGLAB (Delorme & Makeig, 2004) and custom-made scripts to analyze the data. The data were down-sampled to 500 Hz. Finite impulse response notch filters were applied to remove line noise (60 Hz) and its harmonics (120 and 180 Hz). The data were then band-pass filtered between 2 and 55 Hz. We then removed channels that were noisy or flatline using channel correlation. This removed, on average, 1.4 ± 0.5 channels. We then employed the artifact subspace reconstruction method to remove high-amplitude bursts in the EEG data because of eyeblinks and muscle noise (Chang, Hsu, Pion-Tonachini, & Jung, 2020; Mullen et al., 2013, 2015). This method uses the standard deviation of a relatively artifact-free section of the EEG data (heuristically derived from a long stretch of EEG data by computing the channel-wise root-mean-square values on 1-sec windows, z-scoring them across all windows for each individual channel, and then deeming a section clean if the estimated z scores are between −3.5 and 5.5) to define the artifact rejection threshold. This threshold is then applied to the short-window principal component subspace of the EEG data to identify artifact subspace components and subtracted from the data. The remaining components are then back-projected to the channel space to get the artifact-cleaned data set. Finally, we used a threshold on the power to remove any noisy stretches in the data. The artifact-cleaned data were then re-referenced to the average.
We then performed logistic Infomax independent components analysis (ICA) on the noise-rejected data to extract independent components (ICs) for each participant separately (Bell & Sejnowski, 1995). Using the DIPFIT toolbox in EEGLAB (Delorme & Makeig, 2004; Oostenveld & Oostendorp, 2002), we then computed the dipole that best fit the IC scalp topography. From the ICs, we then identified a left and a right SM component by looking at the scalp topography (left and right centrolateralized distribution), frequency spectrum (1/f trend), and the dipole locations (SM source; see Supplementary Figure S1 for details; available at https://osf.io/qwb29/). The ICA approach gives a spatial filter over the motor area in each participant. The electrodes in the filter are weighted, and their activity is averaged. This gives a stronger signal-to-noise ratio of the underlying SM activity compared with using channel space alone. However, in one participant, we could not identify a left and right SM component and so used channels C3 and C4 as proxies for left and right SM sources, respectively. This was because, in this participant, the ICs were distributed more posteriorly and did not give us a boost in signal-to-noise ratio compared with C3/C4. We evaluated this by looking at the time–frequency plot (mu–beta dynamics, analysis explained below) of the trials locked to the primary movement response.
Time–frequency plots were used to estimate the typical event-related spectral perturbations (ERSPs) for mu–beta bands during both the primary and secondary movements. To start, we validated the SM filters obtained via ICA by looking at the ERSP maps of the 80% of primary right-hand movement trials and confirmed the typical mu–beta desynchronization in response to the cue. To do this, we epoched the data time-locked to the primary movement cue (right flex cue) from −1000 to 2000 msec. Next, to look at the PMBR generated by the primary movement, we epoched the same 80% right-hand movement trials, again in relation to the flex cue but for a longer time period (−1000 to 3000 msec). Although there were other timings (for instance, EMG onset/offset of the right wrist flexion) to which we could have epoched the data, we selected to do it with respect to the right flex cue because our secondary movement cue (right/left arrows) was given in relation to this cue, that is, 2 sec after (Figure 1A). Thus, it would be more appropriate to see changes in beta rebound in relation to the flex cue and see if the secondary movement cue overlaps with the PMBR of the primary movement. Finally, to look at changes in SM mu–beta for the secondary movement, we epoched the data from −3000 to 3000 msec wrt the secondary movement cues (right/left arrows). The time–frequency plots were estimated using Morlet wavelets from 4 to 30 Hz, with three cycles at low frequencies linearly increasing by 0.5 at higher frequencies. For group-level analysis, we averaged the ERSP maps across participants and used a nonparametric bootstrap method followed by false discovery rate (FDR) correction for multiple comparisons to estimate regions of significance. In addition, we also performed cluster analysis and report the average cluster statistic (z) for the significant regions in the ERSP plots.
Extracting Sensorimotor Beta Bursts
Because we were interested in the beta state before the secondary movement cue, we epoched the data from −3000 to 3000 msec in relation to the left/right button press cue. On the 80% of trials where there was no secondary movement cue, we epoched the trials such that they had the same total duration as that of the secondary movement epochs but then aligned to make sure that Time 0 corresponds to that time where the secondary movement cue would have occurred in those trials. Accordingly, the 80% right flex trials were epoched from −1000 to 5000 msec in relation to the flex cue. The epoched data were filtered in the beta frequency (13–30 Hz) using a sixth-order Butterworth filter. The Hilbert transform on the filtered data yielded the complex analytic signal. We took the absolute value of this signal to get the beta amplitude. We then defined the burst threshold using the beta amplitude in a baseline period before the start of the trial (−2500 to −2000 msec). From the baseline period, we computed the median and standard deviation of the beta amplitude pooled across all trial types and used it to define the burst threshold. A burst was any period of increase in beta amplitude within a trial that exceeded median + 1.75 SD. This threshold was selected because it was a good trade-off between identifying high-amplitude periods of beta and number of bursts within an epoch (see Supplementary Figure S2 for details on burst threshold selection [available at https://osf.io/qwb29/] and also see Little et al., 2019, for more details). For each detected burst, the time of the peak beta amplitude was marked as the time of the burst, and the burst width was computed using a slightly lower threshold (median + 1 SD) than that used to identify a burst. The approach is similar to the one used in Little et al. (2019) and is done to prevent underestimation of burst widths by only focusing on the central peaks. The amplitude at the peak was taken as the burst height. Furthermore, to compute burst probability/rate (burst %) across trials, we marked all the times where the beta amplitude crossed the lower threshold. The beta bursts in the period 1000 msec before the secondary movement cue were considered for trial-by-trial correlations with behavior (only trials having at least one burst were considered). If a trial contained more than one burst in this period, we took the mean of the burst parameters (time, width, and height) and then performed the correlation. We averaged the r values across the participants to get the estimate of the group-level relationship.
Experiment 2: Participants
There were 15 participants in Experiment 2 (10 women, mean age = 21.5 ± 0.6 years). Sample size was derived from effect size estimates in Experiment 1 (see the Results: Experiment 2 section). Again, like in Experiment 1, all participants were right-handed and provided written informed consent according to a University of California, San Diego institutional review board protocol and were compensated at $20/hr.
Experiment 2: Press/Ready and Move Tasks
The experiment was again coded in MATLAB 2016b (The MathWorks), and stimuli were presented using Psychtoolbox (Brainard, 1997). Experiment 2 had two tasks: a main task (press and move) and a control task (ready and move). The main task was very similar to the design in Experiment 1, wherein a primary movement was used to create a postmovement beta state in which a subsequent movement was sometimes embedded. The control task involved a similar design to the main task, except now, there was no “first” movement to create the postmovement beta state; instead, there was the “secondary” movement on each trial.
For the main task (press and move), the key difference from Experiment 1 was the nature of the primary and secondary movements and the muscle groups involved in performing them (Figure 3A). Here, participants were seated in front of a monitor with both their arms grasping a joystick (Logitech Attack 3, Logitech International S.A.) one in each right and left hand. They placed their index fingers over the trigger buttons (attached to the 2-D moving axis of the joystick). Each trial began with images of two joysticks on the screen, one on the left and one on the right. A cue appeared (“press”) on the center of the screen between the two joysticks images—upon which participants performed a brisk press of only the right-hand trigger button. The press cue remained on the screen for a max of 1 sec or until the participant made the response. On 80% of the trials, following this response, there was 4 ± 0.1 sec of relaxation period, followed by a jittered ITI of 2.5 ± 0.5 sec. On 20% of the trials, at 1 ± 0.1 sec after the press response was made, either the right or the left joystick enlarged in size cueing the participants to make a rapid center-out movement with that joystick, that is, pushing the joystick from the center in a forward direction (more trial timing information below). The participants were verbally instructed at the start of the experiment to make this center-out movement as fast as possible. The response was considered successful if the participant moved the appropriate joystick and if that joystick reached its maximum deflection angle. A green square was presented around the enlarged joystick as feedback if the movement was performed with the correct hand. If not, a red square would appear around it. Participants were given a maximum of 2 sec to perform the center-out movement. The feedback remained on the screen for the entire 2 sec even if the response was made. This was followed by a relaxation period of 1 sec where the image of the enlarged joystick returned back to its original size and a jittered ITI of 2.5 ± 0.5 sec.
The control task (ready and move) was very similar to the main task apart from two changes. Each trial began with a cue “ready,” which instructed the participants to just get ready (no movement needed). The ready cue was presented on the screen for the same time as the mean RT for pressing the right trigger button in the main task. Like before, 1 ± 0.1 sec after the ready cue disappeared, either the right or the left joystick enlarged in size cueing the participants to make a rapid center-out movement with that joystick. However, here, this cue was presented on every trial. The feedback, relaxation, and ITI time were the same as for the main task.
In the main task, participants performed 20 blocks of 30 trials each (total of 600 trials) of which 20% of trials included the secondary center-out movement (120 trials, 60 right and 60 left). In the control task, participants performed four blocks of 24 trials each (total of 96 trials), and because each trial had a center out movement, there were 48 right and 48 left trials. At the end of each block in both tasks, participants were presented with their average RTs of the all the center-out movement trials in that block. They were instructed to maintain their RTs between 400 and 700 msec.
Experiment 2: EEG
In this experiment, we recorded 64-channel scalp EEG in the standard 10/20 using the ActiveTwo system (Biosemi Instrumentation). The EEG signals were digitized at 1024 Hz.
Experiment 2: Data Analysis
We determined the RTs of the trigger press responses of the right hand. For the center-out movements, we collected the raw 2-D axis deflection trajectories from both joysticks. The movement end times were recorded as the time at which the joystick reached the maximum deflection angle in relation to the center-out movement cue. To estimate movement onset times, we processed the data as follows. We resampled the raw trajectories to 500 Hz by linear interpolation. The trajectories were then normalized by dividing them with the maximum deflection angle so as to set the range between −1 and 1. We then smoothed the displacement data using a 20-msec moving average window. From the processed 2-D displacement trajectories, we estimated the velocity by computing the derivative (using the diff function in MATLAB). We further smoothed the velocity profiles using a 20-msec centered rectangular window. Similar to the EMG analysis to get the movement onset times, we first estimated the peak velocity (i.e., the maximum velocity before the movement end time) and then backtracked in time until the velocity dropped below 10% of that peak (also see Figure 3B). Outlier trials in both movement onset times and end times were rejected using the same criteria as Experiment 1 (below 25th percentile −1.5 × IQR and above 75th percentile +1.5 × IQR). Furthermore, trials that had more than one peak in the velocity profile were not considered for further analysis. On an average across participants, this removed only 7.2 ± 0.6% of the secondary center-out movement trials in the main task and 4.0 ± 0.8% in the control task.
The EEG preprocessing was done exactly as in Experiment 1, except that the data were initially down-sampled to 512 Hz instead. Here, the channel correlation method removed, on average, only 1.4 ± 0.3 channels in this case. We then employed the same artifact subspace reconstruction method followed by power threshold to remove artifacts and noisy segments in the data, respectively (see the EEG Analyses section under the Experiment 1: Data Analysis section for more details). Finally, the artifact-cleaned data were then re-referenced to the average.
Similarly, logistic infomax ICA and dipole fitting were done to identify a left and a right SM component by looking at the scalp topography (left and right centrolateralized distribution), frequency spectrum (1/f trend), and the dipole locations (SM source; see Supplementary Figure S1 for details; available at https://osf.io/qwb29/). In this case, however, in four participants, we could not identify a left and a right SM component and so used channels C3 and C4. Three out of the four participants had no ICs with a centrolateral scalp topography, and one participant, like in Experiment 1, had posteriorly distributed ICs, which did not boost the signal-to-noise ratio in comparison to C3/C4. Time–frequency plots locked to the primary movement response (right button press) were used to validate the ICs.
Time–frequency analysis was also done exactly as in Experiment 1 using Morlet wavelets (4–30 Hz with three cycles at low frequencies linearly increasing by 0.5 at higher frequencies). ERSPs were estimated time-locked to both the primary and secondary movements. First, SM ICs were validated by looking at the ERSP maps of the 80% of primary right-hand movement trials (epoched with respect to the right button press cue from −1000 to 2000 msec). We confirmed the typical mu–beta desynchronization in response to the cue. We then analyzed the PMBR by computing ERSPs time-locked to the right button press response (−2000 to 2000 msec). Finally, for looking at the mu–beta rhythms for the secondary movement, we time-locked the EEG data to the secondary movement cues (right/left center-out movement cue) around the time period −2500 to 2500 msec. For group-level analysis, we averaged the ERSP maps across participants and used the same nonparametric bootstrap method followed by FDR correction for multiple comparisons to estimate regions of significance and also reported the average cluster statistics (z) for those regions.
Extracting Sensorimotor Beta Bursts
To look at SM beta bursts, we epoched the data in Experiment 2 in relation to the left/right center-out cues (−2500 to 2500 msec). Like in Experiment 1, on the 80% of trials where there was no secondary movement cue, we epoched the trials such that they had the same total duration as that of the secondary movement epochs, but then aligned to make sure that Time 0 corresponds to that time where the secondary movement cue would have occurred in those trials. Accordingly, the 80% right button press trials were epoched from −1500 to 3500 msec in relation to the right button press response. The remaining methods of filtering, thresholding, and extracting SM beta bursts were the same as in Experiment 1, the only difference being the timing of the baseline period for estimating the burst definition threshold. This was set to −2000 to −1500 msec with respect to the center-out movement cues, that is, before the start of the trial (see the Extracting Sensorimotor Beta Bursts section under the Experiment 1: Data Analysis section for more details). Again, the beta bursts parameters (timing, length, and height) in the period 1000 msec before the secondary movement cue were considered for trial-by-trial correlations with behavior (only trials having at least one burst were considered). We averaged the r values across the participants to get the estimate of the group-level relationship.
The majority of results were evaluated with one-sample or paired t tests. In some cases, nonparametric Wilcoxon's test was used to compute significance, especially when the data were nonnormal (Lilliefors test), for group-level analysis of the coefficients of correlations. Bayes factor (BF10) was also run to estimate effect sizes; these being interpreted as small (BF10 = 1–3), medium (BF10 = 3–10), or large (BF10 > 10). Repeated-measures ANOVA was performed when comparing across multiple levels. Effect sizes for ANOVAs were interpreted as small (ηp2 = .01–.06), medium (ηp2 = .06–.14), and large (ηp2 > .14). Post hoc t tests were used for testing specific hypotheses with Bonferroni correction for multiple comparisons (corrected p value, pB). For correlational analyses, Pearson's correlation coefficient (r) was used. All data are presented as mean ± SEM.
Experiment 1: High Beta State before the Cue Leads to Slower Movements
The motivation for introducing a right-hand wrist flexion was to create a stronger PMBR in the left SM cortex compared with the right SM. Our aim was then to test whether this laterality in the beta rebound has functional relevance by embedding a secondary movement during this period, that is, move either the right or the left hand. Our prediction was that a subsequent right-hand movement would be slower compared with a subsequent left hand movement. The movement trials were rare (only 20%), which ensured that the participants were generally unlikely to prepare the secondary movement once they performed the wrist flexion. Looking at the behavior (button press RTs), we observed that the right hand (588 ± 41 msec) was significantly slower than the left (569 ± 39 msec; ΔRTRight-Left = 19 ± 6 msec), t(10) = 3.02, p = .013, BF10 = 4.8 (Figure 1B). To further probe this, we looked at the times of EMG onset in both thumbs. There was no difference in EMG onsets times, suggesting that the EMG buildup started at the same time (right = 441 ± 34 msec, left = 441 ± 34 msec; ΔTimeRight-Left = 0.5 ± 6 msec), t(10) = 0.08, p = .940, BF10 = 0.2 (Figure 1B). One potential reason why the beta state selectively affected the button press RTs and not the EMG onsets could be that the rate of EMG rise to the peak was slower in the right hand, thus leading to slower RTs. We quantified this by measuring the time taken from the EMG onset to the time of the peak EMG in that trial and saw that the right-hand times (80 ± 8 msec) were significantly delayed compared with the left (70 ± 6 msec; ΔTimeRight-Left = 10 ± 3 msec), t(10) = 3.3, p = .008, BF10 = 7.5. This indeed indicates that the EMG rise was slower in case of the right compared with the left. This suggests that this high beta period might have some influence on the development of the EMG and thus behavior.
Experiment 1: High Beta State Influences Movement-related Desynchronization and Relates to Behavior
To look at SM beta power and beta bursts, we extracted a left SM and a right SM component for each participant using ICA (for one participant, we used C3/C4, see the Experiment 1: EEG Analyses section under the Methods section). We then performed a time–frequency analysis on the 80% of trials where the participants executed just the right wrist flexion. Group-level analysis of the ERSPs (masked for significance p < .05 and FDR corrected) showed the expected movement related neural dynamics in the left SM component, that is, time-locked to the flex cue (white square) we saw the typical mu–beta desynchronization followed by the beta rebound (Figure 2A, two significant clusters, desync cluster z = −5.06, p = .009; sync cluster z = 4.57, p = .001). The beta rebound was not as strong in the right SM component (Figure 2B, one significant cluster, desync cluster z = −4.71, p = .002), and the difference map between both hemispheres (FlexLSM − FlexRSM) confirmed a larger increase in beta power in the left SM component before the movement cue, which came 2 sec after the flex cue (Figure 2C, one significant cluster, sync cluster z = 5.17, p = .002). All the ERSP maps were normalized to a baseline −500 to 0 msec in relation to the flex cue. The timing of the PMBR validated our decision to introduce the subsequent movement (on 20% of trials) 2 sec after the flex cue.
We now looked at how the increased PMBR might relate to the mu–beta desynchronization for the subsequent movement. Comparing the ERSP maps of the left SM components, time-locked to the move right cue (Figure 2D) and right SM components time-locked to the move left cue (Figure 2E), we observed that there was lower mu–beta desynchronization for the former. The difference ERSP map (RightLSM − LeftRSM) showed a relative power increase in both mu and beta bands (Figure 2F, one significant cluster, sync cluster z = 4.19, p < .001). These ERSP maps were also normalized with respect to a baseline before the flex cue (−2500 to −2000 msec in relation to the secondary movement cues). As event-related desynchronization is well known to be modulated by movement preparation (Rhodes, Gaetz, Marsden, & Hall, 2018; Heinrichs-Graham & Wilson, 2016), these results suggest that there might be some influence on preparation by the earlier beta rebound.
We now turned to analyzing more closely the relationship between the beta dynamics and behavior within a particular hemisphere. We turned from average beta power to an analysis of beta bursts, as it has been observed recently that trial-averaged beta only provides a static representation of the underlying dynamics, whereas individual trials are made up of these transient bursts of oscillations (Little et al., 2019; Feingold et al., 2015). Furthermore, the single-trial level dynamics of beta (bursts) predict behavior better than just power: Specifically, it has been shown that burst rate and burst timing relate to behavior during sensory processing (Shin et al., 2017), movement (Little et al., 2019), and working memory (Lundqvist, Herman, Warden, Brincat, & Miller, 2018; Lundqvist et al., 2016). More recently, we have provided evidence of a functional role of beta bursts in action stopping, that is, timing of the beta bursts relates to stopping behavior (Jana et al., 2020). Given this evidence, it was natural for us to probe the relationship between the features of these transient beta events (burst time, height, and width) and behavior in our task.
Before we examined the relationship of these bursts to behavior, we first looked at the burst rates in both the left and right SM components. Time-locked to the secondary button press cues, there was a higher burst rate in the left SM compared with the right SM (Figure 2G). This further confirmed the laterality of the beta rebound and specifically that there were relatively more bursts in the left SM areas. A single-participant example in Figure 2H shows the beta bursts in the left SM component occur just before the secondary movement cue. We then specifically analyzed the influence of these left SM beta bursts occurring in the period before the movement cue (−1000 to 0 msec) on behavior on a trial-by-trial basis. Most trials had at least one burst (84 ± 5%), with a mean burst time of −454 ± 17 msec (relative to the secondary thumb movement cue), a mean burst width of 131 ± 11 msec, and a mean burst height of 1.8 ± 0.3 μV. Trial-by-trial correlations between beta burst height and right button press RT revealed a significant positive relationship at the group level (mean r = .1 ± .05, Wilcoxon's test W = 58, p = .02; Lilliefors test p = .02), but not for the left press (mean r = .02 ± .07), t(10) = 0.3, p = .898 (Figure 2I). The correlations, however, between the right and the left hand were not significant, but the positive relationship with the right hand suggests that the stronger the amplitude of the burst/s before the cue, the later the response. The same analysis using the other metric of bursts (time and width) did not reveal significant relationships with behavior.
Overall, these results clearly suggest that creating a high beta state can lead to movement slowing. However, there were some limitations to the study. First, we did not have a task condition where there was no beta difference between the hemispheres. This would have allowed us to look at differences between the right and left movements in the absence of the rebound. Second, it is possible that there was an influence of physical fatigue on the secondary movement, that is, that the right press slowing after the right flexion might have been observed as a result of the right hand moving twice (assuming that the right thumb flexor overlaps with the right wrist flexion). Third, the timing of the secondary movement could have been placed more tightly. In this experiment, we decided to place the secondary movement instruction at 2 sec after the flex cue on the basis that if the flexion takes, on average, 500 msec to execute and the beta rebound lasts for around 1–2 sec, our cue might be roughly in the window of time of this rebound (which we confirmed it was, on average, see Figure 2A). However, given variability in the flexion task and in the rebound, for example, related to the amount of force applied, our timing might not have been appropriate on every trial. Finally, our secondary movement here was a simple button press, which only provided us with RT for behavioral analysis—a task with kinematics would have provided richer parameters.
Experiment 2: High Beta State Delays Movement Onset and End
To address the shortcomings of Experiment 1, we redesigned the paradigm to look at the influence of a high SM beta state. As the effect size estimate from the behavioral results for Experiment 1 (ΔRTRight-Left = 19 ± 6 msec, Cohen's dz = 0.91) suggested that we would need n = 15 for a 95% chance to observe a significant effect, that sample size is what we chose for Experiment 2. Here, we used a right-hand button press (with the right index finger) as our primary movement to create a beta rebound in the left SM cortex compared with the right SM. On 20% of the trials, participants had to perform a right or a left center-out reaching movement, the instruction for which came ∼1 sec after the button press response (Figure 3A). These changes helped us overcome some of the limitations of Experiment 1. First, the primary movement was less variable, and consequently, we had a tighter control over the time at which the secondary movement was introduced (1 ± 0.1 sec after the button press response), allowing us to target the beta rebound period better. Second, this potentially helped us avoid a concern of physical fatigue, as the muscle groups involved were completely different: the index and forearm muscles for the primary button press and the shoulder muscles for the secondary center-out movement. Third, the center-out reaching movement had richer dynamics, allowing us to quantify movement onset/end times and velocity and to look at the relationship between the beta dynamics and these movement parameters more closely. Finally, we introduced a new control condition (referred to as the control task: ready and move) where we expected no difference in beta power between the left and right SM areas.
Our first behavioral analysis was for the velocity profiles in the main task, that is, the press and 20% move condition. Here, we saw that, following the right button press, the requirement to move the right arm was delayed compared with moving the left arm (Figure 3B). This was a similar pattern of result as observed in Experiment 1. To further quantify this, we looked at the movement onset and end times (see the Behavior section under Experiment 2: Data Analysis for details on their computations) in both the main (press and move) and control tasks (ready and move; Figure 3C). In the main task, both movement onset and end times were significantly longer for the right hand (onset: 446 ± 16 msec, end: 539 ± 19 msec) compared with the left (onset: 410 ± 12 msec, end: 513 ± 16 msec): onset, t(14) = 4.71, p < .001, BF10 > 100; end, t(14) = 3.50, p = .004, BF10 = 12.8. A two-way repeated-measures ANOVA with RT difference (ΔRTRight-Left) as the dependent measure and the Condition (main or control task) and Movement Parameter (onset and end) as independent measures revealed a significant effect of both Condition, F(1, 14) = 12.74, p = .003, ηp2 = .48, and Movement Parameter, F(1, 14) = 6.21, p = .026, ηp2 = .31, but no interaction, F(1, 14) = 0.83, p = .378, ηp2 = .06. Post hoc t tests showed that the difference between right and left is greater in the main compared with the control task in both movement onsets (MainRight-Left: 36 ± 8 msec, ControlRight-Left: 16 ± 6 msec), t(14) = 3.87, pB = .004, BF10 = 24.4, and movement end times (MainRight-Left: 26 ± 7 msec, ControlRight-Left: 10 ± 5 msec), t(14) = 2.71, pB = .034, BF10 = 3.3. This shows that the high beta state delays movement time.
Our motivation above to look at movement onset and end times separately was based on the notion that there might be some influence of the beta state on movement velocities. To investigate this, we examined peak velocity for the right and the left hand in the main task. A two-way repeated-measures ANOVA was performed with Peak Velocity as dependent variable and Condition (main and control) and Hand (right and left) as independent variables. There was no main effect of Condition, F(1, 14) = 0.65, p = .433, ηp2 = .045; Hand, F(1, 14) = 2.99, p = .106, ηp2 = .18; or an interaction, F(1, 14) = 2.44, p = .140, ηp2 = .15. This suggests that the beta state did not have an influence on the peak velocity.
Experiment 2: Sensorimotor Beta Dynamics and Its Relationship to Behavior
Given that we observed movement slowing of the right hand in the main task (press and move), we looked into the neural data to see if we saw similar changes as in Experiment 1. The laterality in the beta rebound was again validated by looking at the ERSPs of the 80% of right button press trials in both the left and right SM components obtained using the ICA approach (here, for four participants, we used C3/C4; Figure 4A and B). The difference ERSP map showed a clear increase in beta power in the left SM component compared with the right (Figure 4C, two significant clusters, one sync cluster z = 4.67, p < .001, and desync cluster z = −4.19, p = .016). In this case, the ERSP maps were aligned to the time of response of the right button press and were normalized to a baseline before the right button press cue (−1000 to −500 msec in relation to the button press response). Across hemisphere, spectral changes in mu and beta revealed similar pattern of results as in Experiment 1. In the press and move conditions, time-locked to the right center-out movement cue, we saw a lower mu–beta desynchronization in the left SM component (Figure 4D) compared with the right SM component time-locked to the left center-out movement cue (Figure 4E). The difference ERSP showed a power increase in both the mu and beta bands (Figure 4F, one significant cluster, sync cluster z = 3.72, p < .001). This shows again, like in Experiment 1, that the beta state induced by the primary movement affects the electrophysiological signature of the secondary movement (i.e., mu–beta desynchronization is lessened).
We now analyzed on our control condition, that is, the control task data (ready and move). Recall that this also requires a left or right movement to an imperative cue, but without the earlier primary movement. This showed (a) now, there was, of course, no beta rebound preceding the “secondary” movement, and (b) minimal difference between hemispheres in the mu–beta desynchronization (Figure 4G, H, and I; no significant clusters found during the mu–beta desync period in Figure 4I). These ERSPs were again normalized to a baseline before the right button press cue (−2000 to −1500 msec in relation to the center-out movement cues). These results from the control condition reinforce the above points from the main task that it is the earlier effect of the PMBR that corresponds to subsequent reductions in mu–beta desynchronization and slower movement.
We now examined beta bursts within the left SM hemisphere. We again looked at the burst parameters (burst time, width, and height) and correlated them to the movement parameters (onset time, end time, and peak velocity). As before, we saw that the burst rate was higher in the left SM component compared with the right SM, suggesting a high probability of beta bursts on the side contralateral to the initial button press movement (Figure 5A). Figure 5B shows an exemplar participant with bursts occurring predominantly before the secondary center-out movement cues (Figure 5B). We then looked into the left SM beta bursts and their relationship to the movement times and peak velocity in the 1000-msec period before the center out movement cues. On average, there were 80 ± 3% of trials, which had at least one burst. The mean burst time across participants was −411 ± 9 msec (relative to the center-out movement cue), with a mean burst width of 99 ± 6 msec and mean burst height of 1.9 ± 0.4 μV. We found that there was a significant positive relationship across participants between right-hand movement onsets and both burst time (mean r = .12 ± .04), t(14) = 2.72, p = .017, BF10 = 3.3, and burst height (mean r = .12 ± .03), t(14) = 3.42, p = .004, BF10 = 11.1 (Figure 5C and D), and a trend for burst length (mean r = .08 ± .04), t(14) = 2.05, p = .06, BF10 = 1.1. The same pattern was seen for right-hand movement end times with a significant positive relationships to both burst time (mean r = .12 ± .04), t(14) = 3.16, p = .007, BF10 = 7.1, and burst height (mean r = .11 ± .03), t(14) = 3.18, p = .007, BF10 = 7.4, but not for burst length (mean r = .07 ± .04), t(14) = 1.82, p = .09, BF10 = 0.8. The more interesting finding was that the relationship between the left SM beta bursts and the right-hand movement was stronger compared with the relationship between the left SM beta bursts and the left-hand movement, an observation that suggests selectivity toward the right hand. The relationship between burst time and left-hand movement was significantly weaker for both the onsets (right = 0.12 ± 0.04 vs. left = 0.03 ± 0.05), t(14) = 2.74, p = .016, BF10 = 3.4, and the end times (right = 0.12 ± 0.04 vs. left = 0.04 ± 0.04), t(14) = 2.84, p = .013, BF10 = 4.0. A similar trend was also observed for the burst height correlations (Figure 5D). For peak velocity, there was a significant negative relationship between right-hand movement onset and burst time (mean r = −.07 ± .03), t(14) = 2.2, p = .047, BF10 = 1.7, implying that the bursts closer to the movement cue correspond to reduced peak velocity. The burst height trended the same way. Given that averaging burst times in a trial only provides an estimate of whether a group of bursts occurred closer (or farther back in time) in relation to the secondary movement cue, we further probed this relationship observed between the burst times and movement times by specifically looking at the times of the last burst on each trial (i.e., the burst that was closest to the secondary movement cue). We stratified the behavior (movement RTs) based on where this last burst occurred in the following predefined different time windows (−600 to −300 msec, −300 to 0 msec). We selected these time windows to make sure there were at least five trials or more with bursts in each of these windows for a participant. We then looked at the movement onset and end times for the right and the left hand in these windows. A 2 × 2 repeated-measures ANOVA on the movement onset times with factors Hand (left and right) and Time (−600 to −300 msec and −300 to 0 msec in relation to the secondary movement cue) showed a main effect of Hand, F(1, 14) = 18.0, p = .001, ηp2 = .6, and a Hand × Time interaction, F(1, 14) = 4.8, p = .046, ηp2 = .3, but no main effect of Time, F(1, 14) = 3.4, p = .085, ηp2 = .2. Post hoc t test on the right-hand movement onsets showed a significant increase in RT if the last burst occurred in the −300 to 0 msec (454 ± 66 msec) window compared with the −600 to −300 msec window (441 ± 57 msec), t(14) = 2.72, pB = .034, BF10 = 3.3 (see Figure 5E). This increase was not seen for the left hand movement onsets (−600 to −300 msec: 411 ± 50 msec vs. −300 to 0 msec: 413 ± 46 msec), t(14) = 0.32, pB = 1.0, BF10 = 0.2 (see Figure 5F). We performed the same analysis on movement end times. There was a main effect of Hand, F(1, 14) = 11.4, p = .004, ηp2 = .5, but no main effect of Time, F(1, 14) = 3.4, p = .087, ηp2 = .2, or a Hand × Time interaction, F(1, 4) = 3.8, p = .073, ηp2 = .2. The right-hand movement end times showed a similar trend to the onsets with the RTs increasing in the −300 to 0 msec window; however, the significance disappeared upon correction (−600 to −300 msec: 535 ± 18 msec vs. −300 to 0 msec: 548 ± 20 msec), t(14) = 2.32, pB = .072, BF10 = 1.7. Thus, looking at bursts collectively from the correlational analysis and specifically in particular time windows, our results suggest that the beta state delays movement times, specifically affecting the movement onsets.
Overall, this analysis of beta bursts for the left SM hemisphere was consistent with Experiment 1 in showing that various features of the bursts, especially amplitude and the timing, relate to the slowing of the right-hand movement.
Potential Effects of Physical Fatigue
It is a possibility, given that the participants performed the primary movement with the right hand, that physical fatigue could contribute partly to the effect as the right hand moves twice. A consequence of fatigue would be that the behavioral slowing of the right hand would be more pronounced late in the experimental session. To probe this, we investigated in both experiments the movement times for the right and the left hand in the first half and the second half of the session. In Experiment 1, the difference between the button press RTs, that is, ΔRTRight-Left, was not significantly different between the halves, although the second half was a bit slower (first half = 13 ± 9 msec vs. second half = 22 ± 8 msec), t(10) = 0.95, p = .367, BF10 = 0.4. We tried to control for this in Experiment 2 by ensuring that the primary and secondary movements used different muscle groups. The same analysis on the main task data revealed that (a) ΔRTRight-Left for movement end times were not different between halves (first half = 28 ± 8 msec vs. second half = 24 ± 7 msec), t(14) = 0.73, p = .478, BF10 = 0.3, and (b) ΔRTRight-Left for movement onsets were actually smaller for the second half (first half = 41 ± 8 msec vs. second half = 32 ± 7 msec), t(14) = 2.54, p = .023, BF10 = 2.5. This says, if anything, that the participants speeded up as the session continued and suggests our results here are unlikely due to physical fatigue.
In two experiments, we demonstrated that an induced SM beta state slows an instructed movement. The high beta state was created in each participant by asking them to perform a right-hand movement (right wrist flexion in Experiment 1 and right button press in Experiment 2). Its functional effect was then tested by embedding a secondary right or a left movement during the beta rebound period of the primary movement. In both experiments, the initial right-hand movement led to stronger left versus right hemisphere SM beta rebound and a slowing of subsequent movement for the right hand versus the left. This is consistent with our initial hypothesis that a high beta state has retardive properties. Furthermore, following the initial right hand movement, the mu–beta desynchronization for the secondary movement was reduced for the left SM area (for a right-hand movement) than for the right SM area (for a left-hand movement). This suggests that a high beta state in say the left hemisphere potentially impacts the physiological signature of subsequent movement for that same hemisphere. Finally, by examining specific features of the beta state, that is, transient beta bursts, we showed that both burst time and burst height related to the degree of slowing within the same hemisphere: Specifically, we found that bursts stronger and closer to the cue to move have a larger effect on behavior. This study has several novel aspects and useful implications. First, whereas most of the earlier studies that examined how the beta state affected behavior took advantage of endogenous variations in beta, here we developed a behavioral paradigm in which people proactively used an instruction to put them into a state that shows the functional properties of slowing. Second, our results have implications for theories of beta rebound—suggesting that it, partly at least, reflects a functional “suppressive” state.
By giving participants a behavioral instruction to move, we were able to proactively create a high beta state, that is, PMBR, and by embedding a cue to move in that period, we showed that this state has a functional effect on slowing. This approach of generating a beta state through a specific behavioral instruction (and then testing the functional effect) is different from some earlier studies, which instead took advantage of endogenous variations in beta oscillations to test relations with behavior. For example, Shin et al. (2017) showed that beta oscillations occurring before a sensory cue affected perception, whereas Torrecillos et al. (2018) and Little et al. (2019) showed that beta oscillations occurring before movement affected RTs (also see Gilbertson et al., 2005). Like those studies, we found that features of beta bursts, in our case both the timing and amplitude of the burst before the (secondary) movement, were related to the amount of slowing, except here we specifically showed this effect was stronger when it had been inculcated by an earlier instruction. Other studies have manipulated beta in different ways. For example, Pogosyan et al. (2009) used noninvasive alternating current stimulation to entrain the motor system at a beta frequency and then showed a small effect on movement. Similarly, another transcranial alternating current stimulation study driving motor cortex at beta frequencies led to decreased peak force development during movement (Joundi, Jenkinson, Brittain, Aziz, & Brown, 2012). Also, a study in monkeys that provided real-time neurofeedback of beta showed that the animals could regulate their motor cortical beta to different levels and that this in turn delayed movement onsets when beta power was high before they were cued to move (Khanna & Carmena, 2017). Our approach differs from all these by presenting a straightforward behavioral route to achieving an experimentally controlled increased beta state in humans. Future studies might test if and how people can manipulate PMBR at will to create a state that is even more retardive of movement and perhaps sensation/perception. As PMBR is sensitive to the amount of force (Fry et al., 2016), the briskness of movement (Stancák & Pfurtscheller, 1996), and task duration (Pakenham et al., 2020), there are several ways that people could modulate it. Also, beta rebound can be generated even after imagined movement (Pfurtscheller & Neuper, 1997; Pfurtscheller, Neuper, Flotzinger, & Pregenzer, 1997), making it potentially relevant as a “brain switch” for motor prosthetics in paralyzed people (Pfurtscheller & Solis-Escalante, 2009). Based on our findings, it is possible that clinically relevant approaches could be developed in neuropsychiatric disorders using imagined movement to generate high beta state in a way that would then affect, for example, tics (Niccolai et al., 2016), attention-deficit/hyperactivity disorder (Bluschke, Broschwitz, Kohl, Roessner, & Beste, 2016), or even stroke (Quandt et al., 2019). Furthermore, our task design will benefit studies that have been looking at the more cognitive effects of beta, especially testing the effect of a state created by the previous trial on the current trial. For instance, HajiHosseini, Hutcherson, and Holroyd (2020) have shown that variations in beta seen after reward feedback predict performance on a subsequent stimulus recall task. Transcranial alternating current stimulation over the motor cortices in beta frequency has shown to affect performance in task-switching conditions whose effects seem to spill over onto subsequent trials (Heise, Monteiro, Leunissen, Mantini, & Swinnen, 2019). Such studies could possibly take advantage of our method by modulating beta behaviorally and then embedding the task of their choice during this state to hone in on its functional properties.
Another implication of our results is for theories of SM beta. An early idea of SM beta, owing to its prominence at rest, was that it is an “idling” rhythm (Pfurtscheller, Stancák, & Neuper, 1996). Later, it was suggested that, rather than reflecting a mere lack of movement, SM beta may be a signature of an active process that promotes the existing motor set whilst impairing neural processing of new movements—something referred to as “the status quo” (Engel & Fries, 2010) or an “active inhibition of the motor network” (reviewed in Kilavik et al., 2013). For PMBR at least, the idea of active inhibition had some support from the fact that, across participants, those with more PMBR had higher levels of motor cortical Gamma aminobutyric acid (Gaetz et al., 2011), and further, that the time period of PMBR corresponded with reduced corticospinal excitability assessed via single-pulse transcranial magnetic stimulation (Chen et al., 1998). Furthermore, recently it was shown that fast termination of movement leads to PMBR in regions other than the motor cortex, thought to reflect a wider state of active motor inhibition (Heinrichs-Graham, Kurz, Gehringer, & Wilson, 2017). The behavioral slowing seen in our study supports this active inhibition account. An alternate school of thought is that PMBR might be linked to sensory evaluation. An early study shows that PMBR is markedly lower when a movement is forcefully terminated in comparison to when it is passively terminated, implying that it is evaluating some form of sensory prediction error (Alegre, Alvarez-Gerriko, Valencia, Iriarte, & Artieda, 2008; also see Cassim et al., 2001). This is also backed up by work from Tan, Jenkinson, and Brown (2014), where they showed that PMBR is attenuated for trials where participants made visuomotor errors. More recently, Torrecillos, Alayrangues, Kilavik, and Malfait (2015) showed that PMBR decreased during both SM and goal-related perturbations, suggesting that it might be modulated by any form of change in the current motor plan. We note that these sensory accounts are, to some extent, related to the status quo idea, where high PMBR signifies maintaining the current nonerroneous motor plan but reduces when you make an error leading to SM adaptation. Finally, motor beta oscillations are also known to be modulated by attention. For example, beta oscillations increase before an informative cue (Saleh, Reimer, Penn, Ojakangas, & Hatsopoulos, 2010). In our case, however, the secondary movement was unpredictable, and so the beta increase might not be a reflection of increased attention as that would predict a gain in performance for the secondary movement. Taken together, especially considering that PMBR relates to gamma aminobutyric acid levels (Gaetz et al., 2011) and reduced corticospinal excitability (Chen et al., 1998), we suppose our results are most compatible with this status quo or active inhibition idea. Still, we warrant that the functional role of PMBR is not monolithic and might reflect several processes, for example, active inhibition as well as sensory input (Cassim et al., 2001), perhaps reflected in different frequency bands.
Several aspects of our study design/parameters were critical to achieve the behavioral results that we observed in both experiments. The secondary movement had to occur infrequently (20%) to prevent the participants from preparing for it. Indeed, during initial piloting, we observed that the behavioral slowing diminished when the percentage of secondary movement was increased (say to 50%). By keeping the probability of the secondary movement low and not having interference from preparatory mechanisms, we hoped that the post-primary movement high beta state would interact with the preparation and/or execution of the secondary movement. This was confirmed experimentally by the EEG data showing lower mu–beta desynchronization for the secondary movement cue when there was high beta rebound before it (Figures 2F and 4F). Another important feature was that the primary movement had to be brisk and ballistic, as we needed to induce PMBR reliably on every trial. It is known that fast and brisk movements can give rise to a strong beta rebound (Stancák & Pfurtscheller, 1996). This helped in reducing the variability in both the timing and power of the PMBR within a participant (cf. Espenhahn, de Berker, van Wijk, Rossiter, & Ward, 2017); it also aided in detecting beta bursts more reliably at a single trial level. Finally, although both experiments showed the behavioral effect, with richer kinematics for the secondary movement in Experiment 2, we were able to hone in on the parameters most affected by the beta state. We saw that the high beta state mainly delayed the movement times (onset and end) and did not have much effect on the peak velocity. Previous studies that have looked at the influence of cortical/subcortical beta on movement have shown effects either on movement times or movement velocity (Little et al., 2019; Torrecillos et al., 2018; Khanna & Carmena, 2017; Pogosyan et al., 2009). It is important to dissociate these effects to understand the functional role of SM beta oscillations.
Our results have some limitations. First, we only used the right-hand movement to create the high beta state in the left SM cortex. The experimental design could have been more balanced by also using the left hand for the primary movement. It is always preferable to counterbalance hand in any experiment using one hand at a time. The main reason we did not use both right and left hand for the primary movement was that the number of trials for doing the trial-by-trial analysis between beta bursts and behavior for each hemisphere would have been very low. This is because the secondary movement trials only occurred 20%, of the time to prevent participants from generally preparing for them. If we had used both left and right hands for the primary movement and used both left and right hemispheres for analysis, we would only have had 30 trials per hemisphere per hand, which would have been too few (especially considering some reduction of trials because of artifact rejection). Second, there is the concern that slowing for a secondary right arm movement after a primary right arm movement, compared with a secondary left arm movement after a primary right arm movement might have related to fatigue. However, our Experiment 2 was designed to obviate this concern as the primary and secondary movements recruited mostly different muscle groups. A final limitation is that our results pertain entirely to the PMBR form of beta and not to other kinds of SM beta or beta in other parts of the brain. Thus, we cannot be sure if the conclusions are only narrowly relevant to this functional state or beta more broadly. Furthermore, a caveat to our methods. For the beta-burst analyses, we restricted the burst extraction to a particular spatial filter (SM) and spectral domain (beta band). Yet, a potential concern is that some domain reduction approaches (spatial or choosing a particular frequency) can lead to differences in estimating burst parameters (Zich, Quinn, Mardell, Ward, & Bestmann, 2020). However, we believe this does not affect our results as domain reduction has maximal effects on the estimation of burst length, interval length between bursts and burst onset, but in our case, the burst parameters that related to behavior were burst time (the time at the peak of a burst) and burst height (amplitude at the peak). These are less likely to be distorted by domain reduction approaches.
In conclusion, we show that a high beta state created via a movement instruction slows down new movements during this period. This suggests that PMBR corresponds, at least to some extent, to a functional–suppressive state. Our approach also provides a behavioral framework for future investigations of the functional role of beta oscillations and also for practical/clinical attempts to help participants voluntarily inculcate beta states with potentially retardive properties.
We thank Henri Skinner and Emma Cary for helping in data recording. This work was supported by the National Institutes of Health (DA026452) and the James S. McDonnell Foundation (220020375).
Reprint requests should be sent to Vignesh Muralidharan, Department of Psychology, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093, or via e-mail: firstname.lastname@example.org.
Vignesh Muralidharan: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Writing—Original draft; Writing—Review & editing. Adam R. Aron: Conceptualization; Formal analysis; Investigation; Methodology; Supervision; Writing—Original draft; Writing—Review & editing.
Adam R. Aron, National Institutes of Health (https://dx.doi.org/10.13039/100000002), grant number: DA026452. Adam R. Aron, James S. McDonnell Foundation, grant number: 220020375.
Diversity in Citation Practices
A retrospective analysis of the citations in every article published in this journal from 2010 to 2020 has revealed a persistent pattern of gender imbalance: Although the proportions of authorship teams (categorized by estimated gender identification of first author/last author) publishing in the Journal of Cognitive Neuroscience (JoCN) during this period were M(an)/M = .408, W(oman)/M = .335, M/W = .108, and W/W = .149, the comparable proportions for the articles that these authorship teams cited were M/M = .579, W/M = .243, M/W = .102, and W/W = .076 (Fulvio et al., JoCN, 33:1, pp. 3–7). Consequently, JoCN encourages all authors to consider gender balance explicitly when selecting which articles to cite and gives them the opportunity to report their article's gender citation balance. The authors of this article report its proportions of citations by gender category to be as follows: M/M = .615, W/M = .173, M/W = .096, and W/W = .115.