Abstract

We studied magnetic signals from the human brain recorded during a second before a self-paced finger movement. Sharp triangular peaks were observed in the averaged signals about 0.7 second before the finger movement. The amplitude of the peaks varied considerably from trial to trial, which indicated that the peaks were concurrent with much longer oscillatory processes. One can cluster trials into distinct groups with characteristic sequences of events. Prominent short trains of pulses in the beta frequency band were identified in the premovement period. This observation suggests that during preparation of the intended movement, cortical activity is well organized in time but differs from trial to trial. Magnetoencephalography can capture these processes with high temporal resolution and allows their study in fine detail.

1  Introduction

Signal averaging is the basic tool in the study of event-related phenomena in the brain. Averaging over many trials is routinely used to extract relatively small signals associated with the process under study from presumably irrelevant brain activity and ambient noise. This approach shows its efficiency in the EEG studies of evoked potentials (Picton et al., 2000). However, averaging implies some simplifying assumptions concerning the brain activity. The signal of interest is assumed to be a sequence of event-locked waves with invariable latencies and shapes, while the noise can be approximated by a random process that is uncorrelated between trials and not time-locked to the event. These assumptions may be appropriate for some types of EEG studies, though the advent of magnetic encephalography (MEG) casts doubt on their validity, at least in some important cases. Better spatial and directional selectivity of magnetic sensors and their immunity to the interference from many cortical electrical sources make recorded signals clearer. This leads to the conclusion that “the traditional model underlying trial averaging that discards all ongoing background activity as noise is no longer tenable in the light of recent research. A large number of studies have demonstrated that the phase and amplitude of ongoing brain activity at the time of stimulus presentation significantly affects behaviour” (Gross et al., 2013).

There is a growing body of evidence that spontaneous activity effectively shapes the brain response to external stimulus. This was shown in studies of spiking activity in the visual cortex of animals (Arieli, Sterkin, Grinvald, & Aertsen, 1996; Fiser, Chiu, & Weliky, 2004). Human EEG demonstrates event-related modifications of the ongoing activity, which are interpreted as responses to the stimulus (Jansen, Agarwal, Hegde, & Boutros, 2003; Makeig et al., 2002). Fluctuations in the human brain activity affect event-related fMRI (functional magnetic resonance imaging) responses (Fox, Snyder, Zacks, & Raichle, 2006). However, the relationship between spontaneous and evoked brain activity is still debated. For example, one can ask whether the signal recorded in a particular experiment is due to the reorganization of the ongoing activity or some additional activity triggered by the stimulus. To resolve this type of question, analysis of the prestimulus period is essential. Numerous studies on this topic have nearly exclusively focused on the cortical oscillations, with special emphasis on the phase synchronization and partial phase resetting (Jansen et al., 2003; Pfurtscheller & Lopes da Silva, 1999; Jackson, Spinks, Freeman, Wolpert, & Lemon, 2002).

Cortical activity is not restricted to rhythmic phenomena, and even signals that look like oscillations are not identical repetitions of the same process. EEG from scalp electrodes and subdural or depth probes often shows relatively brief events, such as spikes or sharp waves, and many rhythms are present for only a short duration. Therefore, it becomes a matter of definition whether to speak of an oscillation or an event (Bullock, McClune, & Enright, 2003). We often observe signals of short duration in our MEG measurements on healthy subjects. In this study, we analyzed cortical events that precede self-paced finger movement. Sharp signals were seen on averaged traces well before the muscular activity started, and they should be attributed to the spontaneous processes in the brain related to the preparation of the movement. Such signals are highly relevant for predicting muscular activity that follows the cortical event. They can be used to control brain-computer interfaces (BCI), a powerful new technology for people severely disabled by neuromuscular disorders (Mellinger et al., 2007; Schalk & Leuthardt, 2011; Shih, Krusienski, & Wolpaw, 2012).

2  Materials and Methods

Eight right-handed volunteers took part in the study—four males and four females, 22–31 years old. None of the participants had known neurological or psychiatric disorders. The subjects were instructed to make quick index finger extensions at their own will, keeping the finger in the up position for some time and then moving it back into the original position. This series of events took about 3 s, and the duration of each phase was chosen as convenient by the subject in each trial. Natural scatter in the duration was about 15%. Each subject made more than 140 movements in a single session. Magnetic brain responses were recorded using a helmet-shaped whole-head MEG system (Elekta Neuromag, 306 channels) at the Moscow University of Psychology and Education.

During the measurement, the participants were sitting in a magnetically shielded room. The movement was monitored by a 3D accelerometer fixed to the tip of the finger. The instant of maximum acceleration of the finger could be clearly distinguished in each trial and was used as a reference time point for averaging. The peak time instants were identified manually using a special program visualizing accelerometer traces. Eight subsequent traces for one subject are shown in Figure 1. The MEG signals were recorded with a sampling rate of 1000 Hz and native hardware filters with bandpass 0.1 Hz to 330 Hz. To avoid distortion of the wave shapes, no additional filtering was applied to the data. For all sensors, 2000 ms epochs centered at the maximum acceleration point for each trial were extracted and subsequently analyzed using custom-made Matlab scripts. We focused on the signals in planar gradiometers and did not consider magnetometers in this work. We intend to study most localized events in the brain, which are clearly seen in only a few neighbor sensors. We did not analyze here characteristic signals seen in any single sensor (there are plenty) since we are aware of the possibility of recording spurious transient signals due to the operation of the complex magnetometer with different levels of signal processing. The experimental data analyzed here are the same as presented in Vvedensky (2014), though we focus on the events preceding finger movement for about 1 s, whereas previously we concentrated on movement-evoked activity.

Figure 1:

Magnetic field gradient measured on the pairs of neighboring sensors for four subjects performing voluntary movements of the right-hand index finger. Traces for different subjects are drawn in different colors. The signals in one planar gradiometer at each indicated location are presented. Frequency band is 0.1–330 Hz. Positions of the sensors in different colors for each subject are shown on a realistic sketch of the helmet-shaped sensor array. The cross is over the vertex, nose pointing to the left. Brisk movement occured in the middle of each 2 s plot. The traces are averages of 147 trials for subject 1, 175 for subject 2, 240 for subject 3, and 141 for subject 4. Yellow vertical stripes show time windows preceding the movement for about 0.7 second, where characteristic sharp peaks congruent in the neighboring sensors were observed. Accelerometer traces for eight subsequent trials (subject 4) are also shown to display consistency of the time reference point in each trial.

Figure 1:

Magnetic field gradient measured on the pairs of neighboring sensors for four subjects performing voluntary movements of the right-hand index finger. Traces for different subjects are drawn in different colors. The signals in one planar gradiometer at each indicated location are presented. Frequency band is 0.1–330 Hz. Positions of the sensors in different colors for each subject are shown on a realistic sketch of the helmet-shaped sensor array. The cross is over the vertex, nose pointing to the left. Brisk movement occured in the middle of each 2 s plot. The traces are averages of 147 trials for subject 1, 175 for subject 2, 240 for subject 3, and 141 for subject 4. Yellow vertical stripes show time windows preceding the movement for about 0.7 second, where characteristic sharp peaks congruent in the neighboring sensors were observed. Accelerometer traces for eight subsequent trials (subject 4) are also shown to display consistency of the time reference point in each trial.

3  Results

3.1  Congruent Signals in Pairs of Neighboring Sensors

We often see brief transient signals in our averaged records. Their specific shape with sharp edges readily pops out from the background during routine visual inspection of the data. Examples for four subjects are shown in Figure 1. The presence of these sharp events in the averaged records implies a robust time lock between the neural process generating this signal and the instant of the fast phase of the finger movement, which happened 0.7 s later. The reasoning behind choosing the instant of maximum acceleration as the reference time point is that the brain processes should coherently converge on the instant of the target action rather than on the beginning of the muscular activity. Myoelectric discharges in the muscles start 30 to100 ms before the actual voluntary limb movement (Waldmann, Schauer, Woldag, & Hummelsheim, 2010).

We focused on events that looked similar in just two neighboring planar gradiometer sensors and could not be clearly seen in the others. Basic knowledge of MEG (Hämäläinen, Hari, Ilmoniemi, Knuutila, & Lounasmaa, 1993) tells us that in these cases, the active patch of the cortex that generates these wave shapes is likely to be located between the pair of sensors and close to the head surface. These sources had different locations for individual subjects, and they surrounded the sensory-motor cortex, where high-amplitude responses to the finger movements were observed for these subjects (Vvedensky, 2014).

The high intersubject variability of the source location is not surprising, considering that individual variability seems to be characteristic of the action initiation process. High intersubject variability of active source location was reported for verb generation task in a group of seven epileptic patients with implanted ECoG electrodes (Miller, Abel, Hebb, & Ojemann, 2011). In spite of different locations of the sources for our subjects (we also observed a similar pair of signals in the opposite, right hemisphere for subject 5), their behavior was quite similar. These characteristic signals were observed about 0.7 s before the brisk finger movement; what is more important, these signals show common patterns in single trials.

3.2  Variation of the Amplitude of the Precursor Peak in Different Trials

The good signal-to-noise ratio of the magnetic signals allowed analysis of individual trials. The wave shapes in the time windows highlighted in Figure 1 are multidimensional vectors. For each trial and subject, one can determine the projection of the actual time course onto the corresponding average vector. This gives in each trial the amplitude of the component, congruent with the average wave shape. This amplitude varies notably from trial to trial, as shown in Figure 2. The amplitude of the congruent component even changed sign, and these changes were in concert on the neighboring sensors, as shown by the trend lines in Figure 2.

Figure 2:

Amplitudes of peaks in different trials in pairs of sensors, shown in Figure 1. Sensor numbers are indicated on the horizontal and vertical axes. Each point represents a single trial. The trials are subdivided into three groups. Blue points show trials with peak amplitude close to the average, red points show trials with high peak amplitude, and green points show trials with inverse amplitude. Blue points make 50% of all trials, red and green points 25% each. Amplitudes of the peaks change in concert in the pairs of sensors, as indicated by the trend lines. The large dot shows the average amplitude.

Figure 2:

Amplitudes of peaks in different trials in pairs of sensors, shown in Figure 1. Sensor numbers are indicated on the horizontal and vertical axes. Each point represents a single trial. The trials are subdivided into three groups. Blue points show trials with peak amplitude close to the average, red points show trials with high peak amplitude, and green points show trials with inverse amplitude. Blue points make 50% of all trials, red and green points 25% each. Amplitudes of the peaks change in concert in the pairs of sensors, as indicated by the trend lines. The large dot shows the average amplitude.

We attribute the observed behavior of the signals to the process that takes place in the cortex between the indicated pairs of sensors for each subject. We call it the precursor peak. MEG spikes, widely distributed across the cortex, were reported for time intervals preceding human saccades (Ioannides, Fenwick, & Liu, 2005). They probably play the same role of preparation for impending saccade as our precursor for the finger movement.

The large scatter of the precursor amplitude in different trials is similar to the amplitude variations of the components evoked by the finger movement 0.7 s later in the same group of subjects (Vvedensky, 2014, Figure 2). These cortical signals differ essentially. For an evoked event, one can expect amplitude dependence on the preevent ongoing brain activity, while the precursor itself is part of spontaneous brain activity. We believe that no external parameters should be taken into account when one looks for the cause of precursor amplitude instability. The mere observation of the scatter shown in Figure 2 suggests fragmentation of the set of trials into characteristic groups. We defined the central group to include 50% of all trials, nearly symmetric around the average value, and two surrounding groups with high and inverse amplitudes—25% of trials each. Such a splitting into homogeneous groups of single-trial patterns sharing similar characteristics (Laskaris, Fotopoulos, & Ioannides, 2004; Ioannides et al., 2002) is a useful tool that can contribute to better understanding of the brain mechanism of the initiation of action. Comprehensive analysis of different classification methods for ongoing EEG and MEG is given in Besserve et al. (2007). These authors stress that the “number of chosen features can influence classification performance drastically.” We are aware of that and tried different approaches to find proper clustering of the trials, which would emphasize in the best way what the human observer sees in the premovement period.

3.3  Regular Sequences of Cortical Events during Preparation of Movement

Simple arithmetic averages for each of the selected groups revealed different behaviors of the signals in these groups of trials (see Figure 3). In contrast to the central group (blue), which produced averages virtually identical to the averages for all trials (shown in Figure 1), the two other groups (red and green) display signals with regular structure extending well outside the time window, where the initial precursor peak was analyzed. This probably means that the brief precursor peak is precisely time-locked to a much longer regular process, often extending over the entire duration of the premovement period about 1 s long. A time lock between spikes and alpha oscillations was observed in the experiment where MEG spike-triggered averaging revealed oscillatory dependence and coupling between cortical areas in the human brain during quiet wakefulness with eyes closed (Ioannides, 2007). Each of our subjects generated individual sequences of the processes, though common features were evident. For a considerable number of trials, we saw enhanced activity in the alpha or beta frequency bands centered in time around 0.7 s before the movement. This is well in accord with the data from similar MEG experiment performed for brain-computer interface studies (Kauhanen, Nykopp, & Sams, 2006; Kauhanen, Nykopp, Lehtonen et al., 2006). Characteristic maxima in the time-frequency representations are clearly seen for their subjects about 0.7 s before brisk index finger extension (Kauhanen, Nykopp, & Sams, 2006, Figure 2). The authors did not discuss this part of their findings, though they stressed the existence of typical groups of trials, displaying different beta-band rebound after the finger movement. The peaks in the averaged time-frequency plots in their study cover the 10 Hz to 25 Hz range, and sometimes they are clearly split into alpha and beta bands. Details vary from subject to subject, though this activity is always present and is clearly different from the cortical activity of tetraplegic patients attempting index finger movements (they are unable to move their extremities).

Figure 3:

Partial averages for the groups of trials presented in different colors in Figure 2 for four subjects. Data from sensor 69 for subject 1, sensor 13 for subject 2, sensor 15 for subject subject 3, and sensor 60 for subject 4. The instant of the finger movement is in the middle of each plot. Blue curves show averages for central groups of trials (50%), which were nearly the same as the total averages. Red curves show averages for 25% of trials with high amplitude of the peak in the time window highlighted in yellow. Green curves correspond to 25% of trials with inverse amplitude of the peak in the highlighted time window. Separation into groups reveals individual coherent behavior of the signals in each group during preparation of the movement, though the signals evoked by the movement itself are the same in these groups.

Figure 3:

Partial averages for the groups of trials presented in different colors in Figure 2 for four subjects. Data from sensor 69 for subject 1, sensor 13 for subject 2, sensor 15 for subject subject 3, and sensor 60 for subject 4. The instant of the finger movement is in the middle of each plot. Blue curves show averages for central groups of trials (50%), which were nearly the same as the total averages. Red curves show averages for 25% of trials with high amplitude of the peak in the time window highlighted in yellow. Green curves correspond to 25% of trials with inverse amplitude of the peak in the highlighted time window. Separation into groups reveals individual coherent behavior of the signals in each group during preparation of the movement, though the signals evoked by the movement itself are the same in these groups.

We believe that in both experiments, we see the same preparatory process in the cortex, which accurately predicts the movement starting 0.7 s later. It also shapes the timetable of the cortical events, which eventually generate muscular activity. The spectral analysis methods used in Kauhanen, Nykopp, and Sams (2006) and Kauhanen, Nykopp, Lehtonen et al. (2006) describe general features of this mechanism, though fine details of temporal evolution of cortical activity preparing the action should be studied in the time domain. Spectral analysis reveals the existence of oscillations of certain frequency during a time window preceding the movement, while analysis in the time domain indicates accurate timing of the peaks with respect to the instant of intended action, which will start more than half a second later.

We see in Figure 3 that the separation into groups of trials on the basis of the precursor peak has little effect on the later signals, which are evoked by finger movement. This probably means that the organization of the cortical processes at the stage of preparation of action and during the action itself are quite different. For many repetitions of the same movement, a considerable fraction of magnetic signals tends to cancel each other out when averaged. These signals are not random noise (see Figure 4) but have a high degree of regularity. We conjecture that the brain somehow balances the frequency of the occurrence of different scenarios of cortical activity (Liu & Ioannides, 1996; Laskaris, Liu, & Ioannides, 2003) that lead to the same intended action.

Figure 4:

Magnetic signals in individual trials in sensor 14 for subject 2, 1 second before and 1 second after the self-paced finger movement. The trial number is indicated for each epoch. Spacing of the vertical gridlines is 50 ms. Regular trains of peaks, observed before the finger movement, are highlighted in different colors. Repetition rates correspond to alpha (green), beta (yellow), delta (blue), and gamma (pink) bands.

Figure 4:

Magnetic signals in individual trials in sensor 14 for subject 2, 1 second before and 1 second after the self-paced finger movement. The trial number is indicated for each epoch. Spacing of the vertical gridlines is 50 ms. Regular trains of peaks, observed before the finger movement, are highlighted in different colors. Repetition rates correspond to alpha (green), beta (yellow), delta (blue), and gamma (pink) bands.

3.4  Groups of Trials with Similar Behavior

We have described splitting the trials into three groups using the amplitude of the precursor peak as a classification parameter. More detailed analysis revealed a larger number of groups with characteristic behavior. All the trials could be separated into clusters that differed considerably, as is shown in Figure 4, where most representative trials from different groups are displayed. Basic features are evident.

Oscillations with beta-band repetition rate dominated the range 1000 to 500 ms before the movement. The number of peaks in a train varied from a single peak to more than seven. A beta train was often followed by a train in the alpha band, which could last up to the beginning of the movement. This is reflected in the top red plot for subject 2 in Figure 3. All high-amplitude oscillations were suppressed after the instant of the finger movement. Occasional trains of oscillations in the low-gamma and delta bands were also observed. It is worth noting that during considerable number of trials, no detectable magnetic signal was seen on the sensor we looked at (trial 130 in Figure 4). This behavior was not specific for just this sensor and subject; it was observed over parietal, temporal, and partly frontal lobes of the brain of all subjects, though the details differed quite a lot.

Identification of characteristic episodes in brain signals and the development of reliable episode detection methods are extremely important for monitoring changes in the brain state on the timescale of about 1 s. Several techniques are used for detection and analysis of interictal epileptiform spikes (Wilson & Emerson. 2002; Nowak, Santiuste, & Russi, 2009), though the algorithm accuracy is still less than that of experts. A somewhat different approach is needed for detecting short oscillatory episodes, and appropriate methods are developed (Whitten, Hughes, Dickson, Jeremy, & Caplan 2011; Lee et al., 2009). In our experiments, we are looking for a relationship between spikes and rhythmic episodes, and we rely on both visual inspection and computations

MEG signals in our experiments often displayed accurate wave shapes, and in some groups of trials they were remarkably similar, as shown in Figure 5. The displayed events are separated in time by about 1 minute, and yet the timing of the peaks with respect to the instant of impending movement is the same. These are nearly identical long beta trains, though they differ by a short interference in the middle. Precisely repeating motifs of spontaneous synaptic activity were detected in neocortical neurons of mice brain (Ikegaya et al., 2004). The synfire chains in their experiments repeated after a few minutes, maintaining millisecond accuracy. We believe that the beta trains recorded in our experiments are manifestations of similar temporally precise firing sequences of spikes in the human brain that prepares impending action.

Figure 5:

Beta trains during trials 54 and 73 included in the same group (trial 18 in Figure 4) which are quite similar and display accurate phase-lock relative to the finger movement that occurred at time instant 0.

Figure 5:

Beta trains during trials 54 and 73 included in the same group (trial 18 in Figure 4) which are quite similar and display accurate phase-lock relative to the finger movement that occurred at time instant 0.

The observed signals were far from noise like, but rather resembled the functioning of a control device permanently changing operating mode. Our data show that probably a system of devices is active during initiation and preparation of action, since characteristic sequences of events were observed simultaneously in multiple cortical sites. An example is shown in Figure 6. Similar beta trains were observed in several sensors, and it is tempting to try to find location of a single-cortical source that generated these recurring waveforms. However, the data from other sensors are incompatible with the “single-source” assumption, which means that several sources in different cortical sites were working in concert and contributed to the observed signals. We saw similar signals on the other side of the head as well. Remotely located sources oscillate nearly coherently in the beta band, though phase delays could change after a couple of peaks in the train. This resembles cross-talk between cortical sites and is in line with the suggestion that the phase is a key parameter for neural computations organized across multiple oscillatory cycles (Wilson, Varela, & Remondes, 2015).

Figure 6:

Simultaneous traces of magnetic signals in different sensors during a single trial for the subject 2. Time window from 1000 ms before to 1000 ms after the finger movement, grid – 50 ms. Characteristic beta trains are highlighted in yellow, and the alpha train in green. A single peak highlighted in green is clearly seen in sensor 8. In spite of the visual similarity, the beta trains in sensors 14, 68, and 8 should be attributed to sources having different locations in the cortex. This example illustrates coherent activity of several cortical sites in the beta band during preparation of action and their relationship with other types of activity.

Figure 6:

Simultaneous traces of magnetic signals in different sensors during a single trial for the subject 2. Time window from 1000 ms before to 1000 ms after the finger movement, grid – 50 ms. Characteristic beta trains are highlighted in yellow, and the alpha train in green. A single peak highlighted in green is clearly seen in sensor 8. In spite of the visual similarity, the beta trains in sensors 14, 68, and 8 should be attributed to sources having different locations in the cortex. This example illustrates coherent activity of several cortical sites in the beta band during preparation of action and their relationship with other types of activity.

Our observations are in excellent agreement with intracortical recordings of beta-synchronized neuronal activity in pre- and postcentral areas of the monkey brain during motor tasks (Murthy & Fetz, 1996; Brovelli et al., 2004). The number of cycles of beta oscillations per episode was variable in those experiments, with a mean of about four. Oscillations were recorded simultaneously at multiple sites in the sensorimotor cortex. Their results indicate that “episodes of 20- to 40-Hz oscillations occur often and become synchronized over a large cortical area during exploratory forelimb movements. However, they have no reliable relation to particular components of the movement and therefore seem unlikely to be involved directly in movement execution; instead, they may represent a neural correlate of attention during demanding sensorimotor behaviors” (Murthy & Fetz, 1996). The role of these beta oscillations in preparation of movement is still poorly understood. We believe that comfortable MEG measurements combined with proper setup of the experiment with cooperative subjects can shed light on the nature of the beta-synchronized large-scale cortical network subserving premovement maintenance behavior.

A beta train was sometimes followed by an event in another frequency band, as shown in Figures 3, 4, and 6. There is no reason to consider them as independent phenomena; rather, this is a special type of time-delayed cross-frequency coupling, which is considered to play a major functional role in neural computation (Canolty & Knight, 2010). Switching from one frequency band to another within a short time window was characteristic of the cortical events preceding the intended movement. This is in line with the suggestion that concerted alpha, beta, and gamma band oscillations are required for unified cognitive operations (Palva & Palva, 2007). Each frequency band in the series of alpha, beta, and low gamma nearly doubles the frequency in the previous band. During 1 s before the movement, we permanently observed these “doubling-halving” transitions, which are presumably due to the discontinuous change of the loop time—the conduction time for traversing a reciprocal connection between two populations of neurons (Klimesch, 2013). Magnetic signals from the brain preparing action provide abundant and reliable data for analysis that can tell us how this machinery is working.

4  Discussion

Neuroimaging studies of a wide range of cognitive tasks reveal simultaneous activation in multiple cortical regions, suggesting distributed networks of brain areas involved in cognition (Laskaris et al., 2004; Bressler & Menon, 2010). High variability of our signals, picked up from a restricted cortical area, indicates involvement of the monitored neural population in a larger system, which as a whole performs all the necessary actions leading to the self-paced finger movement. Actual events in the monitored area during a single trial depend on the state of the other components of the system. One might expect extraordinary complex relationships between the cortical areas involved, though the accumulated evidence suggests a rather small number of basic cortical networks that describes activity of the brain in the resting state. EEG (Mehrkanoon, Breakspear, & Boonstra, 2014; Hipp, Engel, & Siegel, 2011), MEG (Brookes et al., 2011) and fMRI (Yeo et al., 2011) data converge on the number 7, or slightly above, for the resting state of the brain when no task is given to the subject. Our case is somewhat different from the resting state, since action is assumed to be performed in the near future, and still we found nearly the same number of groups of trials. This may reflect a relationship between the events we observe and the processes in the basic cortical networks.

For every subject, we could define groups of behavior of cortical populations of neurons that lead to the same detectable overt action. The groups were clearly distinct, and each had its own fine-grained timing of events time-locked to the instant of the planned movement. There was nothing very special in the grouping approach. Classification of trials is quite common, and even a routine operation, in experiments with brain signals. One has to remove different kinds of artifacts from the records—ocular movements, heartbeats, muscular contractions, and unidentified interference. The selection of trials can be extended to the cortical processes, which manifest clearly different behaviors. Our observations show that preliminary classification of trials is an important tool; it can separate different spontaneous processes that are active during preparation of the voluntary movement. Classification of single trials is already commonly used in BCI research, and for MEG measurements in particular (Laskaris et al., 2004; Kauhanen, Nykopp, & Sams, 2006). Experiments on primates with BCI, using signals from intracranial probes (Carmena et al., 2003), led to the conclusion that several cortical areas should be monitored for reliable operation of the BCI-driven robotic arm. Accurate performance was possible because large populations of neurons from multiple cortical areas were sampled. The authors argue that motor programming and execution are represented in a highly distributed fashion across frontal and parietal areas and that each of these areas contains neurons that represent multiple motor parameters. They state that accurate real-time prediction of all motor parameters, as well as a high level of BCI control, can be obtained from multiunit signals. We conjecture that our observation of the groups of trials reflects different involvement of these separate cortical areas into the movement initiation process during individual trials.

Direct electrical stimulation of cortical areas of individuals with brain tumors during surgery revealed at least two distinct cortical regions that have to work in concert in order to execute movement that the patient is aware of (Desmurget et al., 2009). The authors report two main contrasting findings: (1) stimulation of the posterior parietal cortex caused human participants to intend to move and to report having moved, even in the absence of actual motor responses, and (2) stimulation of the premotor cortex triggered limb and mouth movements that were not consciously detected by the patients.

It is likely that many cortical sites operate in harmony in order to initiate and perform a coordinated planned action. Magnetic measurement is a subject-friendly technique to study ensembles of cortical neurons involved in activity scattered over the brain.

The analysis of electric or magnetic signals from the human brain is often restricted to frequency domain, which smears the results in the rather short time window preceding the movement. The actual time course of the magnetic signal in each trial reveals a finely grained structure of the cortical processes. In previous work (Vvedensky, 2014), we showed that the MEG spectra during the preevent period of time tend to be exponential in frequency, whereas the resting state displays power dependence (Miller, Sorensen, Ojemann, & den Nijs, 2009; Dehghani, Bédard, Cash, Halgren, & Destexhe, 2010). During movement preparation, brain behavior in the involved regions starts to be more regular and less noisy. When analyzing corresponding cortical processes, it is important to keep in mind that signals that look like noise may in fact be mixtures of regular processes sampled by the brain from its repository. At the neuronal level, this was shown quite a time ago: brain activity resembles a “vast sea of (high-dimensional) chaos (deterministic disorder) and random fluctuations, out of which structured events pop once in a while” (Tsodyks, Kenet, Grinvald, & Arieli, 1999).

During an experiment, transitions between brain states occur permanently from one trial to the next trial, which follows in a couple of seconds. This seems natural for a system that is alive. On the timescale of about half an hour (the duration of the experiment), all the possible states are visited more or less equally, and average signals are recorded in spite of considerable variability in single trials. An experiment with self-paced movements is a kind of mental state monitoring, where individual trials provide information about the ongoing changes in this state. MEG measurements are quite comfortable and are a good tool to study how the variations of the mental state are linked to the faculty of the human brain called volition or free choice. Volition matures late in individual development and probably recruits cortical sites specific for each human being. This means that the experiments should be set up and analyzed individually for each subject.

Precise timing is crucial to many aspects of human performance, especially the prediction of the future events. In our experiments, only the timing of the finger movement was “the free parameter” that was to some extent independently chosen in each single trial. A detailed study of the cortical processes during the brain’s preparation of action can shed light on the nature of decision making and conscious intention, a topic of particular interest (Haggard, 2005; Donner, Siegel, Fries, & Engel, 2009). Many different techniques are used in this area of research, though we believe that MEG with appropriate signal processing tools is most appropriate at the moment for the study of subtle and evasive mental phenomena related to volition and choice. In our experiments, we see magnetic signals with repetition rates characteristic of alpha, beta, and low gamma frequency bands, which are clearly seen in individual trials and mostly cancel each other in the averages for more than 100 trials. There is no reason to consider them as irrelevant noise. We see sharp transitions from beta band to alpha band and back, as well as transitions from beta band to gamma band and back. Beta band plays a central role in these processes. Although beta-band oscillations in motor cortex have been the subject of much experimental investigation, reviews (Baker, 2007; Kilavik, Zaepffel, Brovelli, MacKay, & Riehle, 2013) lead to the following conclusion: “After reviewing the large body of experimental literature, it becomes obvious that many processes might influence the beta-band activity, explaining why the functional significance of the different beta components is still poorly understood” (Kilavik et al., 2013). In our opinion, such uncertainty in the functional role is due to the generality of the basic process, reflected in the concatenated trains of pulses in different frequency bands. These are general neural computations that can be used for any purpose in the current environment.

The importance of using several frequency bands for neural computation was elegantly demonstrated in the study of the echolocation abilities of bats (Suga, 1990). Different processing tasks are parceled out among several anatomically distinct areas. Reciprocal connections between these areas form computational loops. Processing of the weak reflected signal needs suppression of the basic frequency of the emitted signal. Received echo signal is not precisely a second harmonic of the fundamental tone because of the Doppler shift. Suppression of the first harmonic confers an important advantage when the bat hunts in dense vegetation. Another important feature is the personalized auditory cortex of each bat. Bats live in colonies, and the personal signal should be recognized among hundreds of other signals. Similar personalization is needed for different pools of neurons in the human cortex, which have to perform their neural computations in the “swarm” of other pools busy with their own tasks. The concatenated trains of pulses in different frequency bands observed in our experiments are similar to the strings of symbols, resembling words, which can be identifiers of a particular cortical ensemble performing a specific neural computation.

5  Conclusion

Magnetic brain signals of short duration preceded self-paced finger movement for about 0.7 s. Sharp triangular peaks generated by the sources in the areas surrounding the sensory-motor cortex were precisely time-locked to the fast phase of the movement, which started much later. The peaks were also time-locked to the sequences of oscillations in alpha, beta, and low gamma frequency bands. These oscillations followed specific timing patterns during 1 s before the intended movement and highly resembled a process of computation. Parameters of these cortical processes were individual for each subject, though the main features of a general mechanism of the movement initiation could be extracted from the MEG measurements.

Acknowledgments

We thank T. A. Stroganova for general support, interesting discussions, and critically reading the manuscript. The MEG Center is supported by core funding from the Russian Ministry of Science and Education (RFMEF161914X0006). A.O.P. is supported by the Russian Science Foundation, Grant 14-28-00234 to Kurchatov Institute. V.L.V. is supported by Russian Fund for Basic Research grant 15-29-03814-ofi_m.

The authors declare no competing financial interests.

The study was approved by the local ethics committee of the Moscow University of Psychology and Education and was conducted following the ethical principles regarding human experimentation (Helsinki Declaration). Informed consent was obtained from all participants prior to recordings. The letter does not contain clinical studies or patient data.

References

Arieli
,
A.
,
Sterkin
,
A.
,
Grinvald
,
A.
, &
Aertsen
,
A.
(
1996
).
Dynamics of ongoing activity: Explanation of the large variability in evoked cortical responses
.
Science
,
273
(
5283
),
1868
1871
.
Baker
,
S. N.
(
2007
).
Oscillatory interactions between sensorimotor cortex and the periphery
.
Curr. Opin. Neurobiol.
,
17
(
6
),
649
655
.
Besserve
,
M.
,
Jerbi
,
K.
,
Laurent
,
F
,
Baillet
,
S.
,
Martinerie
,
J.
, &
Garnero
,
L.
(
2007
).
Classification methods for ongoing EEG and MEG signals
.
Biol. Res.
,
40
,
415
437
.
Bressler
,
S. L.
, &
Menon
,
V.
(
2010
).
Large-scale brain networks in cognition: Emerging methods and principles
.
Trends in Cognitive Sciences
,
14
,
277
290
.
Brookes
,
M. J.
,
Woolrich
,
M.
,
Luckhoo
,
H.
,
Price
,
D.
,
Hale
,
J. R.
,
Stephenson
,
M. C.
, …
Morris
,
P. G.
(
2011
).
Investigating the electrophysiological basis of resting state networks using magnetoencephalography
.
Proc. Natl. Acad. Sci. USA
,
108
(
40
),
16783
16788
.
Brovelli
,
A.
,
Ding
,
M.
,
Ledberg
,
A.
,
Chen
,
Y.
,
Nakamura
,
R.
, &
Bressler
,
S. L.
(
2004
).
Beta oscillations in a large-scale sensorimotor cortical network: Directional influences revealed by Granger causality
.
PNAS
,
101
(
26
),
9849
9854
.
Bullock
,
T. H.
,
McClune
,
M. C.
, &
Enright
,
J. T.
(
2003
).
Are the electroencephalograms mainly rhythmic? Assessment of periodicity in wide-band time series
.
Neuroscience
,
121
(
1
),
233
252
.
Canolty
,
R. T.
, &
Knight
,
R. T.
(
2010
).
The functional role of cross-frequency coupling
.
Trends Cogn. Sci.
,
14
(
11
),
506
515
.
Carmena
,
J. M.
,
Lebedev
,
M. A.
,
Crist
,
R. E.
,
O’Doherty
,
J. E.
,
Santucci
,
D. M.
,
Dimitrov
,
D. F.
, …
Nicolelis
,
M. A.
(
2003
).
Learning to control a brain–machine interface for reaching and grasping by primates
.
PLoS Biol.
,
1
(
2
),
e42
.
Dehghani
,
N.
,
Bédard
,
C.
,
Cash
,
S. S.
,
Halgren
,
E.
, &
Destexhe
,
A.
(
2010
).
Comparative power spectral analysis of simultaneous elecroencephalographic and magnetoencephalographic recordings in humans suggests non-resistive extracellular media
.
J. Comput. Neurosci.
,
29
(
3
),
405
421
.
Desmurget
,
M.
,
Reilly
,
K. T.
,
Richard
,
N.
,
Szathmari
,
A.
,
Mottolese
,
C.
, &
Sirigu
,
A.
(
2009
).
Movement intention after parietal cortex stimulation in humans
.
Science
,
324
,
811
813
.
Donner
,
T. H.
,
Siegel
,
M.
,
Fries
,
P.
, &
Engel
,
A. K.
(
2009
).
Buildup of choice-predictive activity in human motor cortex during perceptual decision making
.
Current Biology
,
19
(
18
),
1581
1585
.
Fiser
,
J.
,
Chiu
,
C. Y.
, &
Weliky
,
M.
(
2004
).
Small modulation of ongoing cortical dynamics by sensory input during natural vision
.
Nature
,
431
,
573
578
.
Fox
,
M. D.
,
Snyder
,
A. Z.
,
Zacks
,
J. M.
, &
Raichle
,
M. E.
(
2006
).
Coherent spontaneous activity accounts for trial-to-trial variability in human evoked brain responses
.
Nat. Neurosci.
,
9
(
1
),
23
25
.
Gross
,
J.
,
Baillet
,
S.
,
Barnes
,
G. R.
,
Henson
,
R. N.
,
Hillebrand
,
A.
,
Jensen
,
O.
, …
Schoffelen
,
J. M.
(
2013
).
Good practice for conducting and reporting MEG research
.
NeuroImage
,
65
,
349
363
.
Haggard
,
P.
(
2005
).
Conscious intention and motor cognition
.
Trends Cogn. Sci.
,
9
(
6
),
290
295
.
Hämäläinen
,
M.
,
Hari
,
R.
,
Ilmoniemi
,
R. J.
,
Knuutila
,
J.
, &
Lounasmaa
,
O. V.
(
1993
).
Magnetoencephalography—theory, instrumentation, and applications to noninvasive studies of the working human brain
.
Rev. Mod. Phys.
,
65
,
413
497
.
Hipp
,
J. F.
,
Engel
,
A. K.
, &
Siegel
,
M.
(
2011
).
Oscillatory synchronization in large-scale cortical networks predicts perception
.
Neuron
,
69
(
2
),
387
396
.
Ikegaya
,
Y.
,
Aaron
,
G.
,
Cossart
,
R.
,
Aronov
,
D.
,
Lampl
,
I.
,
Ferster
,
D.
, &
Yuste
,
R.
(
2004
).
Synfire chains and cortical songs: Temporal modules of cortical activity
.
Science
,
304
,
559
564
.
Ioannides
,
A. A.
(
2007
).
Dynamic functional connectivity
.
Current Opinion in Neurobiology
,
17
,
161
170
.
Ioannides
,
A. A.
,
Fenwick
,
P.B.C.
, &
Liu
,
L.
(
2005
).
Widely distributed magnetoencephalography spikes related to the planning and execution of human saccades
.
J. Neurosci.
,
25
(
35
),
7950
7967
.
Ioannides
,
A. A.
,
Kostopoulos
,
G. K.
,
Laskaris
,
N. A.
,
Liu
,
L.
,
Shibata
,
T.
,
Schellens
,
M.
, …
Khurshudyan
,
A.
(
2002
).
Timing and connectivity in the human somatosensory cortex from single trial mass electrical activity
.
Human Brain Mapping
,
15
,
231
246
.
Jackson
,
A.
,
Spinks
,
R. L.
,
Freeman
,
T.C.B.
,
Wolpert
,
D. M.
, &
Lemon
,
R. N.
(
2002
).
Rhythm generation in monkey motor cortex explored using pyramidal tract stimulation
.
Journal of Physiology
,
541
(
3
),
685
699
.
Jansen
,
B. H.
,
Agarwal
,
G.
,
Hegde
,
A.
, &
Boutros
,
N. N.
(
2003
).
Phase synchronization of the ongoing EEG and auditory EP generation
.
Clin. Neurophysiol.
,
114
,
79
85
.
Kauhanen
,
L.
,
Nykopp
,
T.
,
Lehtonen
,
J.
,
Jylänki
,
P.
,
Heikkonen
,
J.
,
Rantanen
,
P.
, …
Sams
,
M.
(
2006
).
EEG and MEG brain-computer interface for tetraplegic patients
.
IEEE Trans. Neural. Syst. Rehabil. Eng.
,
14
(
2
),
190
193
.
Kauhanen
,
L.
,
Nykopp
,
T.
, &
Sams
,
M.
(
2006
).
Classification of single MEG trials related to left and right index finger movements
.
Clin. Neurophysiol.
,
117
(
2
),
430
439
.
Kilavik
,
B. E.
,
Zaepffel
,
M.
,
Brovelli
,
A.
,
MacKay
,
W. A.
, &
Riehle
,
A.
(
2013
).
The ups and downs of beta oscillations in sensorimotor cortex
.
Exp. Neurol.
,
245
,
15
26
.
Klimesch
,
W.
(
2013
).
An algorithm for the EEG frequency architecture of consciousness and brain body coupling
.
Front. Hum. Neurosci.
,
7
,
766
.
Laskaris
,
N. A.
,
Fotopoulos
,
S.
, &
Ioannides
,
A. A.
(
2004
).
Mining information from event-related recordings
.
IEEE Signal Processing Magazine
,
21
(
3
),
66
77
.
Laskaris
,
N. A.
,
Liu
,
L. C.
, &
Ioannides
,
A. A.
(
2003
).
Single-trial variability in early visual neuromagnetic responses: An explorative study based on the regional activation contributing to the N70m peak
.
NeuroImage
,
20
,
765
783
.
Lee
,
P.-L.
,
Shang
,
L.-Z.
,
Wu
,
Y.-T.
,
Shu
,
C.-H.
,
Hsieh
,
J.-C.
,
Lin
,
Y.-Y.
, … 
Shyu
,
K.-K.
(
2009
).
Single-trial analysis of cortical oscillatory activities during voluntary movements using empirical mode decomposition (EMD)-based spatiotemporal approach
.
Annals of Biomedical Engineering
,
38
(
7
),
1683
1700
.
Liu
,
L.
, &
Ioannides
,
A. A.
(
1996
).
A correlation study of averaged and single trials MEG signals: The average describes multiple histories each in a different set of single trials
.
Brain Topography
,
8
,
385
396
.
Makeig
,
S.
,
Westerfield
,
M.
,
Jung
,
T. P.
,
Enghoff
,
S.
,
Townsend
,
J.
,
Courchesne
,
E.
, &
Sejnowski
,
T. J.
(
2002
).
Dynamic brain sources of visual evoked responses
.
Science
,
295
,
690
694
.
Mehrkanoon
,
S.
,
Breakspear
,
M.
, &
Boonstra
,
T. W.
(
2014
).
Low-dimensional dynamics of resting-state cortical activity
.
Brain Topography
,
27
(
3
),
338
352
.
Mellinger
,
J.
,
Schalk
,
G.
,
Braun
,
C.
,
Preissl
,
H.
,
Rosenstiel
,
W.
,
Birbaumer
,
N.
, &
Kübler
,
A.
(
2007
).
An MEG-based brain-computer interface (BCI)
.
NeuroImage
,
36
(
3
),
581
593
.
Miller
,
K. J.
,
Abel
,
T. J.
,
Hebb
,
A. O.
, &
Ojemann
,
J. G.
(
2011
).
Rapid online language mapping with electrocorticography
.
J. Neurosurg. Pediatr.
,
7
(
5
),
482
490
.
Miller
,
K. J.
,
Sorensen
,
L. B.
,
Ojemann
,
J. G.
, &
den Nijs
,
M.
(
2009
).
Power-law scaling in the brain surface electric potential
.
PLoS Comput. Biol.
,
5
,
e1000609
.
Murthy
,
V. N.
, &
Fetz
,
E. E.
(
1996
).
Oscillatory activity in sensorimotor cortex of awake monkeys: Synchronization of local field potentials and relation to behavior
.
Journal of Neurophysiology
,
76
(
6
),
3949
3967
.
Nowak
,
R.
,
Santiuste
,
M.
, &
Russi
,
A.
(
2009
).
Toward a definition of MEG spike: Parametric description of spikes recorded simultaneously by MEG and depth electrodes
.
Seizure
,
18
,
652
655
.
Palva
,
S.
, &
Palva
,
J. M.
(
2007
).
New vistas for alpha-frequency band oscillations
.
Trends Neurosci.
,
30
(
4
),
150
158
.
Pfurtscheller
,
G.
, &
Lopes da Silva
,
F. H.
(
1999
).
Event-related EEG/MEG synchronization and desynchronization: Basic principles
.
Clin. Neurophysiol.
,
110
(
11
),
1842
1857
.
Picton
,
T. W.
,
Bentin
,
S.
,
Berg
,
P.
,
Donchin
,
E.
,
Hillyard
,
S. A.
,
Johnson
,
R.
, …
Taylor
,
M. J.
(
2000
).
Guidelines for using human event-related potentials to study cognition: Recording standards and publication criteria
.
Psychophysiology
,
37
,
127
152
.
Schalk
,
G.
, &
Leuthardt
,
E. C.
(
2011
).
Brain-computer interfaces using electrocorticographic signals
.
IEEE Reviews in Biomedical Engineering
,
4
,
140
154
.
Shih
,
J. J.
,
Krusienski
,
D. J.
, &
Wolpaw
,
J. R.
(
2012
).
Brain-computer interfaces in medicine
.
Mayo Clin. Proc.
,
87
(
3
),
268
279
.
Suga
,
N.
(
1990
).
Biosonar and neural computation in bats
.
Sci. Am.
,
262
(
6
),
60
68
.
Tsodyks
,
M.
,
Kenet
,
T.
,
Grinvald
,
A.
, &
Arieli
,
A.
(
1999
).
Linking spontaneous activity of single cortical neurons and the underlying functional architecture
.
Science
,
286
,
1943
1946
.
Vvedensky
,
V. L.
(
2014
).
Individual trial-to-trial variability of different components of neuromagnetic signals associated with self-paced finger movements
.
Neurosci. Lett.
,
21
(
569
),
94
98
.
Waldmann
,
G.
,
Schauer
,
M.
,
Woldag
,
H.
, &
Hummelsheim
,
H.
(
2010
).
Choosing the optimal trigger point for analysis of movements after stroke based on magnetoencephalographic recordings
.
Stroke Research and Treatment
,
arti. 467673
.
Whitten
,
T. A.
,
Hughes
,
A. M.
,
Dickson
,
C. T.
,
Jeremy
,
B.
, &
Caplan
,
J. B.
(
2011
).
A better oscillation detection method robustly extracts EEG rhythms across brain state changes: The human alpha rhythm as a test case
.
NeuroImage
,
54
,
860
874
Wilson
,
M. A.
,
Varela
,
C.
, &
Remondes
,
M.
(
2015
).
Phase organization of network computations
.
Current Opinion in Neurobiology
,
31
,
250
253
.
Wilson
,
S. B.
, &
Emerson
,
R.
(
2002
).
Spike detection: A review and comparison of algorithms
.
Clin. Neurophysiol.
,
113
(
12
),
1873
1881
.
Yeo
,
B. T.
,
Krienen
,
F. M.
,
Sepulcre
,
J.
,
Sabuncu
,
M. R.
,
Lashkari
,
D.
,
Hollinshead
,
M.
, …
Buckner
,
R. L.
(
2011
).
The organization of the human cerebral cortex estimated by intrinsic functional connectivity
.
J. Neurophysiol.
,
106
(
3
),
1125
1165
.