Changes in cortical activity measured with EEG during a high-intensity cycling exercise

Abstract

This study investigated the effects of a high-intensity cycling exercise on changes in spectral and temporal aspects of electroencephalography (EEG) measured from 10 experienced cyclists. Cyclists performed a maximum aerobic power test on the first testing day followed by a time-to-exhaustion trial at 85% of their maximum power output on 2 subsequent days that were separated by ∼48 h. EEG was recorded using a 64-channel system at 500 Hz. Independent component (IC) analysis parsed the EEG scalp data into maximal ICs. An equivalent current dipole model was calculated for each IC, and results were clustered across subjects. A time-frequency analysis of the identified electrocortical clusters was performed to investigate the magnitude and timing of event-related spectral perturbations. Significant changes (P < 0.05) in electrocortical activity were found in frontal, supplementary motor and parietal areas of the cortex. Overall, there was a significant increase in EEG power as fatigue developed throughout the exercise. The strongest increase was found in the frontal area of the cortex. The timing of event-related desynchronization within the supplementary motor area corresponds with the onset of force production and the transition from flexion to extension in the pedaling cycle. The results indicate an involvement of the cerebral cortex during the pedaling task that most likely involves executive control function, as well as motor planning and execution.

coordinated motor skills require the integration of information from peripheral sensors, spinal locomotor networks as well as supraspinal commands (Grillner et al. 2008). However, only in the last decade have researchers begun to investigate the function of the human brain with respect to voluntary, mostly rhythmic, locomotor tasks in real time. Owing to the dynamic nature of human locomotion, a primary requirement to quantify cortical dynamics during locomotion is a time resolution that allows the recording of modulations in cortical activity within one cycle of a repeating movement sequence. Classical neuroimaging tools, such as functional magnetic resonance imaging or positron emission tomography are not well suited to resolve such rapid modulations of activity. Contrary, electroencephalography (EEG) is a mobile brain imaging technique with more-than-sufficient time resolution to capture cortical dynamics during human locomotion. While historically EEG has been considered too prone to noise and artifacts for dynamic recordings, recent advancements in hardware and software have opened the possibility for mobile brain imaging using EEG (Gramann et al. 2011; Makeig et al. 2009).

In the last decade, researchers have begun to study time-dependent cortical dynamics in locomotor and exercise-related situations, with the majority of studies focusing on upright human walking and slow jogging. It was proposed that the low-frequency component (0.1–2 Hz) of EEG recordings can be used to infer the overall time-dependent linear and angular kinematics of the ankle, knee and hip joints during walking (Presacco et al. 2011). Another study has shown that specific brain regions, such as the primary motor, the parietal, and frontal cortexes, exhibit distinct time-frequency patterns that are closely linked to the overall movement kinematics of the gait cycle (Gwin et al. 2011). Similarly, it was shown that task-dependent EEG can be extracted during slow to moderate walking speeds (De Sanctis et al. 2012), and that cortical modulations seen in EEG differ between active and passive walking conditions (Wagner et al. 2012). More recently, Wagner and colleagues (2014) showed that providing interactive feedback during walking enhances premotor and parietal areas in the brain that are thought to play an important role for motor planning and intention. These studies demonstrate that cortical dynamics respond to changes in task parameters, indicating that EEG can be used to study neural control during human locomotor tasks. The typical frequency bands that are investigated in movement-related tasks are theta- (θ: 4–8 Hz), alpha- (α: 8–12 Hz), beta- (β: 12–30 Hz) and low gamma- (γ: 30–50 Hz) frequency bands, as they are believed to reflect different aspects with respect to motor planning, execution and control (Gwin et al. 2011; Jurkiewicz et al. 2006; Pfurtscheller et al. 2000; Taniguchi et al. 2000; Waldert et al. 2008). This was supported by a recent study that found sustained suppression of cortical oscillations in the α- and β-band, while cortical activity in the low γ-band seemed to be modulated during the gait cycle phases (Seeber et al. 2014), suggesting that these responses characterize different elements of human walking. A follow-up study found that low and high γ-amplitudes are modulated throughout the gait cycle; however, their amplitude envelopes were negatively correlated throughout the movement (Seeber et al. 2015). The majority of the mentioned studies have been conducted during human walking at low intensities. The aim of this study was to apply similar analysis techniques to high effort locomotor tasks. Cycling avoids the high-impact forces that are typically observed during high-effort running and has often been used in studies investigating neuromuscular strategies due to the well-controlled movement and repeatability of the task (Enders et al. 2013). In fact, it has previously been used for EEG measurements during high-intensity cycling, and evidence was provided for increased communication between mid/anterior insular and motor cortex (Hilty et al. 2011). This increased communication between cortical areas was found at the end of the cycling exercise and returned to baseline after a rest period, suggesting that (muscle) fatigue alters interaction between cortical structures. Expanding upon these findings, the purpose of this study was to localize electrical brain activity using EEG during cycling at a constant workload and quantify changes in temporal brain activation throughout an exhaustive bout of exercise. Based on previous observations in cycling (Brümmer et al. 2011; Schneider et al. 2009), we hypothesized that electrocortical sources will be located in the motor areas, the frontal cortex and the parietal cortex. Furthermore, it was hypothesized that EEG activity within motor areas will demonstrate an oscillating pattern of increased and decreased activation throughout the pedaling cycle. We further hypothesized that overall EEG activity would increase as fatigue develops throughout the exercise duration.

MATERIALS AND METHODS

Subjects and ethics approval.

Ten healthy, experienced male cyclists (Table 1), with no history of lower limb injury within the past 2 yr and no known neurological or muscular deficits, volunteered for this study. All subjects provided written, informed consent prior to testing. All procedures complied with the standards of the Declaration of Helsinki and were reviewed and approved by the University of Calgary research ethics review board.

Table 1.

Subjects' characteristics including lactate thresholds and 85% of MAP that was used for the time-to-exhaustion trial

Subject ID No.	Age, yr	Weight, kg	MAP, W	MAP, W/kg	LT1, W	LT2, W	85% MAP, W
1	35	77.4	425	5.5	250	300	360
2	23	82.5	386	4.7	200	250	330
3	22	68.2	437	6.4	250	300	370
4	29	71.4	438	6.1	250	300	370
5	30	62.2	360	5.8	175	225	310
6	31	72.4	351	4.8	150	225	300
7	35	67.0	450	6.7	275	325	380
8	18	75.1	430	5.7	225	300	365
9	25	65.8	413	6.3	225	275	350
10	27	75.5	413	5.5	225	275	350
Mean	27.5	71.8	410.3	5.8	222.5	277.5	348.5
SD	5.6	6.1	33.8	0.7	38.1	34.3	26.9

MAP, maximum aerobic power; LT1 and LT2, lactate thresholds 1 and 2, respectively.

Experimental protocol.

Each subject performed testing on 3 different days. All experimental protocols were supervised by a Canadian Society of Exercise Physiology Certified Exercise Physiologist, and participants were cleared for exercise using a Physical Activity Readiness Questionnaire. Resting blood pressure and heart rate were measured prior to daily testing, and standardized ceiling levels, according to the Canadian Society of Exercise Physiology, were adhered to. All testing was completed using a Velotron Dynafit Pro cycle ergometer (RacerMate, Seattle, WA) that was calibrated for each testing session (Fig. 1A). All participants used their own pedals (clipless) and corresponding footwear that locked into the pedals to accommodate their usual bike racing equipment. The first testing day was used to establish the maximum aerobic power (MAP) capacity of each subject based on an incremental test. Testing on the second day consisted of a time-to-exhaustion (TTE) test at 85% of the individual's MAP and was used for familiarization with the protocol and equipment. The third day was a repeat of the second day and was used for the data analysis. The three testing sessions for each subject were separated by 48 ± 1 h. To increase repeatability of the results, caffeine intake was restricted to 2 h prior to testing, and the subject was asked to maintain a normal diet and refrain from any resistance training for the duration of the study.

Fig. 1.

A: schematic drawing of the experimental setup. Participants were equipped with a tight-fitting 64-channel electroencephalography (EEG) system and several electrodes to record lower limb electromyography (EMG). Cables were securely attached to the subject's back to avoid artifacts due to cable movement. B: the cadence of a representative subject for a complete time-to-exhaustion trial (gray line). The black line shows the cadence when applying a moving average with a window size of 50-pedal revolutions. The dashed line shows the cadence that the participant voluntarily chose to start the trial. The dotted line shows the corresponding cut-off cadence to define task failure. The hatched areas with the horizontal lines depict the time frame corresponding to the fresh and fatigue condition used for the analysis. The shaded gray area corresponds to the last 5% of the trial ultimately prior to task failure. C: the average cadence across all subjects for the fresh and fatigue phase as well as immediately before task failure. *Significant difference compared with the fresh condition (P < 0.01). †Significant difference compared with the fresh and fatigue condition (P < 0.01). rpm, Revolutions per minute.

MAP.

On the first day of testing, an incremental cycling exercise test was used to establish blood lactate thresholds LT1 and LT2 for each subject, according to the double breakaway model (Anderson and Rhodes 1989; Kindermann et al. 1979). Subjects were given 20 min of rest after the incremental test prior to starting the MAP protocol. The MAP test was started at a workload slightly below the LT1, and the workload increased every minute by 25 W. The test was terminated when the cadence dropped by more than 15 revolutions per minute (rpm) or below 70 rpm (Fig. 1B). Subjects were included in the study if they reached an MAP of at least 4.5 W/kg.

TTE trial.

Prior to the TTE trial, each subject performed a standardized warm-up protocol based on three 5-min stages corresponding to resistance values below, at, and above the measured LT1. After the warm-up, the TTE trial was performed at a constant workload of 85% of the MAP with the same seat and handlebar settings of the ergometer that were used to perform the MAP test. On both days, subjects were unaware of the workload and the elapsed time of the TTE trial to ensure consistency between days. The preferred cadence was freely chosen by the cyclists (typically between 90 and 100 rpm), and the test was terminated if the cadence dropped by 15 rpm or below 70 rpm. Verbal encouragement was provided throughout the test on both days to ensure maximum performance.

EEG data collection.

Subjects cycled at constant intensity using the ergometer described above. A magnetic switch was used to synchronize the data streams and identify the pedaling cycle. The switch sent a square-wave pulse that was recorded in synchrony with the EEG data to identify individual pedaling revolutions. EEG was recorded using an active electrode 64-channel (10–20 positioning) BrainVision actiCHamp system (Brain Products). Subjects wore a tight-fitting electrode cap corresponding to the measured head circumference using a chinstrap to reduce movement of the cap (Fig. 1A). For each subject, nasion, inion and preauricular points were used as anatomical landmarks to position the electrode cap. Prior to data collection, SuperVisc electrode gel (EasyCap) was used to ensure that the impedance was less than 20 kΩ for each channel. During acquisition, all EEG channels were referenced to the parietal midline (Pz) electrode, and a frontal midline (AFz) electrode was used as ground. EEG signals were recorded with a sampling frequency of 500 Hz per channel and stored on a computer for offline processing. All data processing was performed in Matlab (The Mathworks, Natick, MA) using custom-written software in combination with scripts based on EEGLab 13.2.2b, an open-source toolbox for the analysis of electrophysiological data sets (Delorme and Makeig 2004).

EEG data analysis.

Analysis of continuous EEG signals was carried out according to previously used methodologies (Gwin et al. 2011; Kline et al. 2014; Wagner et al. 2012, 2014). Briefly, EEG signals were band-pass filtered (1–50 Hz) to remove signal drifts as well as line noise and were re-referenced to a common average reference. Channels were removed if one of the following criteria were true: 1) channels with standard deviation (SD) larger than 1,000 μV; 2) channels whose kurtosis was more than 5 SD from the mean; and 3) channels that were uncorrelated with neighboring channels (r < 0.4) for more than 0.1% of the time. On average, 51.6 channels were used for further analysis (SD: 1.96, range 49–54). Independent component analysis (ICA) was used to decompose the signal into maximally independent components (IC) (Makeig et al. 1996). ICA was applied to the data set of individual subjects. EEG data corresponding to the entire TTE trial (approximately 7 min) were used for the ICA decomposition. ICs representing line noise, movement artifacts and eye activity were rejected prior to further analysis (Jung et al. 2000). To locate the neural sources in brain space, an equivalent current dipole model was computed for individual component scalp maps using the DIPFIT plugin within EEGLab (Oostenveld and Oostendorp 2002). Each IC was modeled as a single dipole. We used a boundary element head model based on the Montreal Neurological Institute template that utilizes the average of 152 MRI scans from healthy, young adult individuals. No specific spatial constraints were enforced, as spatial accuracy is limited without the individual anatomy of subjects. ICs were excluded from further analysis if the projection of the dipole to the scalp accounted for less than 80% of the scalp map variance. All remaining ICs of all individual subjects (∼100 in total) were clustered across all subjects using k-means clustering algorithms implemented in EEGLab. The algorithm used the dipole location in brain space and the power spectra of each individual IC for clustering purposes and was reduced to the first 10 principal dimensions (as in Gwin et al. 2011). Clusters and dipoles indicating eye movement artifacts were not considered for further analysis. Additionally, any components that were located near the lower back part of the head were not considered for further analysis as they typically represent electromyographic (EMG) activity from neck muscle that is characterized by an increase in spectral power above 30 Hz. Lastly, any clusters that were located in deeper brain structures near the cerebellum were not considered for analysis, as EEG is not able to localize neural activity in these areas reliably.

Spectral analysis of EEG data.

For all individual components in the identified electrocortical clusters, we calculated power spectra using Welch's method. The average power spectra for each cluster were compared for different stages of the exercise corresponding to the first, middle and last 20% of all pedal revolutions corresponding to the fresh, middle and fatigue condition during the TTE trial. An ANOVA with three conditions (fresh, middle, fatigue) was computed within the EEGLab toolbox to compare power spectra between conditions controlling for false discovery rate (Benjamini and Hochberg 1995) in case of multiple comparisons (P < 0.05).

Time-frequency analysis.

Specifically for electrocortical clusters in the motor areas, we calculated spectrograms for each pedal revolution as a function of crank angle. Spectrograms were calculated based on a set of sinusoidal wavelets (Delorme and Makeig 2004). The number of cycles increased slowly with increasing center frequency, which allows for a better frequency resolution at higher frequencies compared with a traditional wavelet approach using constant cycle length. The time-frequency spectrograms were time-locked to the start of the pedal revolution and then linearly time warped so that the onset and offset of each pedal revolution occurred at the same latencies (Castermans et al. 2014; Gwin et al. 2011; Gwin and Ferris 2012; Seeber et al. 2015). The time-warping procedure uses event latencies of each individual pedal revolution onset and offset and warps these latencies to the average onset and offset latency, thereby time normalizing each pedal revolution. To display electrocortical fluctuations within a pedal revolution, we expressed the time-frequency transformed data relative to the average spectrogram (averaged across the time domain of the pedal revolution) similar to previous studies (Gwin et al. 2011; Kline et al. 2014; Wagner et al. 2012, 2014). This was done based on the following equation:

$$Y_{\text{dB}}=10\cdot {\log }_{10}\frac{Y_t}{Y_0}$$

where Y_t is the power at time t and Y₀ is the average power over the pedaling cycle.

These fluctuations from baseline are typically referred to as event-related spectral perturbations (ERSP). We calculated ERSPs for each component within the motor area cluster. A grand average ERSP was obtained by averaging the individual ERSPs. Significant pedaling ERSPs were identified using bootstrapping methods available within EEGLab (200 iterations). Bootstrapping is based on surrogate data distributions by randomly resampling the observed data, recomputing the outcome variable and comparing the experimental data to the generated surrogate data distribution. Multiple comparisons were controlled using the false discovery rate with a significance level set at 0.05. All nonsignificant fluctuations from baseline in the ERSP plots were masked for visualization in this paper. To reveal the relative timing of brain activity, we calculated waveforms representing spectral power fluctuations in the α- (8–12 Hz), β- (12–30 Hz) and low γ-band (30–50 Hz). ERSPs and waveforms were visualized for the first 20% of the exercise (fresh) and the last 20% of the exercise (fatigue). These timeframes correspond to ∼120–135 pedal revolutions. Comparisons between the ERSP conditions was made within EEGLab using the bootstrapping procedure outlined above using the false discovery rate to correct for multiple comparisons.

EMG data collection and analysis.

While the analysis of muscle activity was not the primary focus of this paper, surface EMG data were collected to test whether potential cortical changes were associated with altered muscular activation patterns. Details of the data collection and analysis techniques are provided in a previously published paper that utilized the same techniques (Enders et al. 2015). Briefly, muscle activity was collected from muscles spanning the ankle, knee and hip joints with a sampling rate of 2,000 Hz. Raw EMG signals were separated into individual pedal revolutions according to a magnetic pedal switch. To obtain temporal waveforms, the EMG activity was wavelet transformed (von Tscharner 2000). The output of the wavelet transform is an intensity pattern indicating the EMG activity at each time point for specific center frequencies. Summing up the activity across all frequency bands results in the total intensity for each time point. These temporal waveforms were normalized to the pedaling cycle with a time resolution corresponding to 0.5° of the crank cycle. Grand average waveforms were obtained by averaging the trials of each subject corresponding to the first and last 20% of the TTE trial.

RESULTS

TTE.

On average, participants were able to produce 85% of their MAP for 7:04 min (range: 6:01 to 8:58 min). Participants started the trial at a cadence of ∼97 rpm. The cadence significantly dropped to 90 rpm during the last 20% of the TTE trial. As per our definition of task failure, a drop in cadence of 15 rpm or below an absolute number of 70 rpm, the average cadence for the last 5% of the pedal revolution decreased to 86 rpm (Fig. 1C).

Location of clusters in brain space.

Across all subjects, we identified four electrocortical clusters that contained components from more than one-half of our subjects. The corresponding Talairach coordinates (Lancaster et al. 2000) and Brodmann areas (BA) are illustrated in Table 2. Two clusters (left and right hemisphere) were identified in the parietal cortex with corresponding BA 39 and 7. One cluster was identified in the superior frontal gyrus of the frontal cortex corresponding to BA 8. Lastly, we identified one cluster in the supplementary motor area (SMA) of the brain in the precentral gyrus of the frontal cortex corresponding to BA 6. Additionally, two more clusters were identified corresponding to the right frontal and right premotor areas; however, only four and five subjects, respectively, contributed to these clusters. According to our inclusion criteria, if a cluster did not contain dipoles of at least more than one-half of the tested subjects, it was not included in any further analysis.

Table 2.

Talairach coordinates for identified clusters (geometric mean)

Cluster	Nearest Gray Matter Brodmann Area	Talairach Coordinates
Frontal cortex, superior frontal gyrus	BA 8	−21, 48, 52
Frontal cortex, precentral gyrus	BA 6	−23, −19, 70
Parietal cortex, superior parietal lobule	BA 7	38, −60, 60
Parietal cortex, inferior parietal lobule	BA 39	−48, −65, 46

Brodmann area (BA) was determined using the Talairach client (www.talairach.org; Lancaster et al. 2000).

Changes with exercise duration.

All clusters of cortical activity showed a significant increase in EEG power (P < 0.05). No differences were found between the fresh and middle portion of the TTE trial, and, therefore, the results are focused on the fresh and fatigued states. The left frontal cortex showed an increase in α-, β- and γ-band, while the clusters in the SMA and left parietal cortex only showed a significant increase in the α- and β-band, and the right parietal cortex only for the α-band (Fig. 2). The strongest increase in EEG power was observed in BA 8, followed by increases of smaller magnitude in the SMA (BA 6) and the parietal cortex.

Fig. 2.

Scalp map projections from the dipole clusters and power spectra for electrocortical sources corresponding to frontal cortex [Brodmann area (BA) 8; top left], supplementary motor area (SMA) (BA 6; bottom left), right parietal cortex (BA 7; top right), and left parietal cortex (BA 39; bottom right). The scale indicates the strength of the cluster average scalp projection to each electrode. The traces indicate the fresh (black) and fatigue (gray) conditions with shaded gray areas indicating the 95% confidence intervals. The MRI images show the dipole clusters in the sagittal and coronal planes. ICs, independent components.

Spectral fluctuations within motor area.

The ERSP pattern for BA 6 (SMA) is visualized with lighter gray, indicating increased brain activity, and darker gray, indicating decreased brain activity, with respect to the average spectrogram that was considered as baseline (Fig. 3). All nonsignificant modulations of spectral power were masked (medium gray) for better visualization. All frequency bands showed an oscillating pattern of increasing and decreasing activity throughout the pedaling cycle in both conditions. In the fresh condition, significant modulations in α-, β- and γ-band power are characterized by two local minima representing event-related desynchronization (ERD), which occur before the onset of right and left down stroke, respectively. Modulations of γ-band power remain similar in the fatigue condition, with the second negative peak occurring slightly later compared with the fresh condition. On the contrary, modulations of α- and β-band power show an altered pattern in the fatigue condition with a pronounced ERD occurring prior to the onset of the right down stroke, followed by a phase of event-related synchronization (ERS) prior to the left down stroke. Overall, there was a larger amplitude modulation in the α- and β-band in the fatigue condition.

Fig. 3.

A: event-related spectral perturbations (ERSP) plot of BA 6 showing modulations in EEG spectral power throughout the pedaling cycle (x-axis) as a function of frequency (y-axis, log scaled) with light gray and dark gray indicating increased and decreased activity with respect to the average spectrogram, respectively. Left plot shows data for the fresh condition (first 20%), and the middle plot for the fatigue condition (last 20%). The right plot shows the difference (fresh minus fatigue) between the two conditions with any nonsignificant differences set to zero. B: the plots below show the time-dependent brain activation across the pedaling cycle in the α-, β- and low γ-frequency bands. The gray and white areas indicate the down stroke (power production phase) of the right and left leg, respectively. The onset of right down stroke (RDS) and left down stroke (LDS) corresponds to the right and left leg passing the top dead center (TDC) and is indicated by the arrows below the x-axis.

Muscle activation patterns.

Grand average temporal waveforms of EMG activity were visualized for both conditions (Fig. 4). The waveforms are in general agreement with previous results investigating muscle activity during cycling (Baum and Li 2003; Enders et al. 2015; Ryan and Gregor 1992). Most muscles showed a trend toward increased muscle activity; however, across all subjects the differences were not statistically different. The overall temporal pattern of muscle recruitment, however, seemed very robust, despite the development of significant fatigue.

Fig. 4.

Grand average EMG waveforms in the fresh (black) and fatigue (gray) condition for six lower limb muscles. The x-axis shows the crank cycle with zero referring to the TDC. The y-axis shows the EMG activity normalized to the mean activity of each muscle across the pedaling cycle in the fresh condition.

DISCUSSION

Modulation of cortical activity during cycling.

This is the first study to quantify changes in the temporal pattern of electrocortical activity during high-intensity cycling in a TTE trial. This expands on previous work investigating intracortical communication during a cycling TTE trial where communication between mid/anterior insular and motor cortex significantly increased with the development of fatigue (Hilty et al. 2011). Similar to previous studies on gait, the results of our study highlight temporal patterns of human brain activity that are spatially localized to the motor area within the frontal lobe (BA 6) (Gwin et al. 2011; Wagner et al. 2012). Significant modulations from baseline were observed in EEG power for an electrocortical cluster that was localized in BA 6, representing the SMA. Significant ERD in EEG power were observed across the α-, β- and γ-frequency range for the SMA during the transition from the recovery phase to the power production phase for both legs, indicating activation and increased neuron excitability for these brain areas at these specific latencies (Neuper and Pfurtscheller 2001; Pfurtscheller and Lopes da Silva 1999). Thus neural activation within the SMA is greatest when muscles are recruited that are critical in transitioning the lower limb from a flexed to an extended position. This result is in agreement with previously described cortical activation patterns during cycling (Jain et al. 2013). Interestingly, a study quantifying cortical dynamics during walking supports the notion that modulation of activity in the BA 6 is related to movement with specific phases of repetitive locomotor activity (Wagner et al. 2012). They also found a trend for increased neural activation in the γ-frequency band in active compared with passive walking. Our results during active cycling also show modulation of γ-activity in the SMA, supporting the suggestion by Wagner and colleagues (2012) that it might be an important contributor to motor planning and sensorimotor processing.

The role of the motor areas in the cycling task used in this experimental protocol can be explored in more detail when interpreting the alternating pattern of desynchronization and synchronization of EEG power (Fig. 3). Desynchronization refers to clusters of neurons that are not firing in synchrony and reflect activity distributed across various brain areas. Neural desynchronization is often interpreted as active information processing (Jurkiewicz et al. 2006; Pfurtscheller and Berghold 1989). An increase in desynchronization can be indicative of an increase in task complexity with a higher demand for information processing (Pfurtscheller and Lopes da Silva 1999). The opposite phenomenon, an increase in EEG power at a given frequency, is typically observed due to a synchronized firing of neurons. Although it is generally believed that ERS in the α- and β-band reflects coherent activity and a deactivated communication between neural networks (Pfurtscheller and Lopes da Silva 1999), in the context of movement tasks it is often interpreted as movement-related sensory processing (Cassim et al. 2001). In summary, ERD reflects active information processing with activity distributed across neural networks, while ERS reflects a more focal activation pattern, indicating a neural cell assembly responding in the same way. The results in this study show different responses for the α- and β- compared with the γ-frequency band and will, therefore, be discussed separately.

In the α- and β-band, during the fresh condition, we see a small ERD that occurs during the phases of the right and left down stroke followed by ERS toward the end of the crank cycle that is stronger in the β- compared with the α-band. In both frequency bands, this pattern is changed during the fatigue condition and pronounced in amplitude. A clear phase of ERD during the down stroke of the right leg is followed by a phase of ERS in the second half of the cycle. The magnitude of ERD and ERS is increased in both α- and β-frequency band. Specifically, during the fresh condition, the phases of ERD, representing an active information processing behavior, correspond to the phases when the leg prepares for force production and transitions from a flexed to an extended position (see Fig. 3).

The γ-frequency band shows similar behavior to the α- and β-frequency bands during the fresh condition with oscillations of ERD and ERS throughout the pedaling cycle. Again, phases of ERD, interpreted as active information processing, correlate well with the timing of leg force production. While the temporal pattern seemed to be altered in the α- and β-band, the oscillations in the γ-band remained similar. In both conditions, the γ-band shows a pattern of two local minima, corresponding to the initiation of power production and preparation for the extension phase of the right and left leg.

The oscillating pattern of ERD and ERS could be interpreted as an alternating behavior in the motor areas, reflecting a distributed activity and active information processing in preparation for leg extension (ERD), followed by a focal activation behavior (ERS) during the main phase of force production. This pattern was most pronounced in the γ-frequency band. This observation was made in a cluster slightly localized in the left SMA. One may expect a mirror effect in the other hemisphere or a cluster that is located over the central midline. Our identified cluster contained dipoles located in both the left hemisphere as well as in the midline of BA 6. While we did find a cluster in the right part of the premotor cortex, this cluster did not contain ICs from more than one-half of our subjects.

Overall, we speculate that differences in EEG power modulation between the fresh and fatigued state are due to an increased effort and task difficulty with which participants need to cope. An alternative option would be a substantial change in neuromuscular strategies of the leg muscles to keep up with the cycling task. However, preliminary analysis of EMG data that were collected in synchrony with the EEG data does not support the hypothesis of altered lower limb neuromuscular strategies, as the overall EMG pattern remained similar (Fig. 4).

Spectral power increase in EEG with fatigue.

Our study observed a broadband increase in EEG power, even though for most identified clusters this was only true in the α- and β-band. This increase across a broader frequency band is in contrast to studies that have investigated different walking conditions and found changes limited to the α- and β-band (Wagner et al. 2012, 2014). It is unclear, however, if our fatigue protocol directly relates to comparisons of active and passive walking (Wagner et al. 2012) or different visual feedback conditions (Wagner et al. 2014). Typically, active movement is associated with desynchronization in the α- and β-band, and, therefore, it is not surprising to find such differences between active and passive movement. Instead our comparison was between two conditions that were both based on intense movements. Interestingly, a study that has investigated fatigue on a recumbent bicycle ergometer shows results that are very similar to this study. They showed that, as participants fatigued during the cycling exercise, there was a broadband increase in spectral power that was significant in the θ- (4.5–7.99 Hz), low α- (8–10.49 Hz), high α- (10.5–12.99 Hz), low β- (13–17.99 Hz) and high β- (18–30 Hz) frequency bands (Bailey et al. 2008). Another possibility is noise in the acquired EEG that was not removed by our data analysis. While this may partially contribute to our observation, the changes in modulation of ERD and ERS also suggest a response in cortical involvement during the exercise. Limitations with respect to noise and motion artifacts are further discussed at the end of this section.

Interaction of electrocortical clusters.

The four identified clusters of electrical activity in specific cortical domains (prefrontal, premotor and posterior parietal cortices) were consistently active in the majority of subjects in this study. Common activation between these areas indicate a sequence of cognitive processing that includes executive control, motor planning and execution as well as sensorimotor integration. Indeed, previous research suggests that activation in the prefrontal cortex area relates to the executive behavioral control function (Tanji and Hoshi 2008) and motor planning (Rizzolatti and Luppino 2001; Tanji 2001; Wise et al. 1997). Information about motor planning may be passed on to the SMA and premotor cortex as it is well documented that these areas have strong connections to the prefrontal cortex (Chouinard and Paus 2006; Rizzolatti and Luppino 2001), specifically the dorsolateral prefrontal cortex (Lu et al. 1994). Thus the output of the prefrontal cortex likely targets specific premotor areas. It has been proposed that the prefrontal-dependent motor areas (anterior parts of the premotor cortex and SMA) receive higher order cognitive information regarding motor plans and motivation (Rizzolatti and Luppino 2001), which supports the executive control function of the frontal areas. Interestingly, the largest response in neural activity accompanying fatigue is seen in the frontal areas. We speculate that this is consistent with an increase in executive control to keep up with the challenging task demand. However, this speculation needs careful evaluation in a future study.

In our analysis, no cluster was directly associated to activity in the primary motor cortex. Several technical as well as functional explanations may describe this observation. It has to be noted that EEG has a limited spatial resolution, limiting the exact location of activity within the motor areas. Furthermore, individual anatomy varies across participants, and this study did not utilize individual MRI head models, which would improve source localization. However, it is well known that strong interconnections between premotor and primary motor areas exist. It suggests that information that is processed and passed on to the premotor cortex is directed further to the primary motor cortex (Chouinard and Paus 2006; Künzle 1978) and the spinal cord (Dum and Strick 1991; Münchau et al. 2002). In fact, the majority of cortico-cortical connections of the primary motor cortex are with BA 6 (Leichnetz 1986; Stepniewska et al. 1993), implying that reciprocal information processing between premotor and primary areas plays a crucial role in motor execution. It is further known that both premotor and primary motor areas have strong connections to the spinal cord through corticospinal neurons that are activated during movement (Dum and Strick 1991; Quallo et al. 2012).

Lastly, the connectivity between parietal areas and motor areas (specifically the SMA and premotor cortex) has been demonstrated in many different scenarios. For a detailed review of this network, the reader is referred to more in-depth literature (Desmurget and Sirigu 2009). Studies link activity in the parietal cortex to motor awareness and movement intention prior to movement (Desmurget et al. 2009), coordinative aspects of movement, and functioning as an interface between sensorimotor and visually guided aspects of movements (Buneo and Andersen 2006; Taira et al. 1990). The fact that the posterior parietal cortex plays a role in sensory and motor function led to the proposal that it is heavily involved in sensory-motor interaction. Interestingly, the parietal cortex has typically been identified in studies focusing on real-time EEG recordings during repetitive locomotor tasks (Gwin et al. 2011; Wagner et al. 2014). Specifically, a premotor-parietal network has been identified during rhythmic walking, which is believed to represent motor planning and intention and has been shown to increase activity in the presence of task feedback (Wagner et al. 2014). In summary, connectivity between parietal and premotor areas seems to be crucial for motor planning as well as sensory motor integration. Increased EEG power was observed in this area, which might suggest the importance of motor planning and the integration of sensory input toward the end of an intense exercise.

Limitations: muscle drive vs. sensory afferent integration.

The spectral analysis performed in this study adds to the emerging understanding that the cortex may play a more important role during rhythmic movement activity than previously thought. It confirms similar findings of alternating latencies of ERD and ERS within a movement cycle (Gwin et al. 2011; Jain et al. 2013; Wagner et al. 2012, 2014). However, solely based on the spectral analysis, it remains unknown if the observed cortical activation patterns are directly involved in the transmission of motor commands to facilitate muscle activation. Another possibility, a hypothesis that was previously discussed (Gwin et al. 2011), is the integration of sensory afferent signals that are used to modulate efferent corticospinal transmission. Most likely, the observed activity reflects involvement in both descending efferent pathways, as well as sensory information processing. To investigate this proposal in more detail, a coherence analysis should be used to address the idea of an anatomical coupling between cortical and muscular signals during movement (Petersen et al. 2012).

Limitations: movement-related artifacts.

Movement-related and specifically gait-related artifacts have been a central issue to the debate about the interpretation of EEG data recorded during locomotor activities. Following the recommendation of previous reports, we excluded any components that were clearly indicative of EMG artifacts (neck, jaw, facial). Additionally, we recorded the data in an electrically shielded chamber to minimize electromagnetic interference from power lines and devices. Due to the nature of the experimental protocol, we had to address the issue of perspiration. Sweating typically results in very slow drifts of the EEG signal. We found these drifts to be well below 1 Hz, and thus we were able to remove these drifts using the applied filters. Additionally, sweating was relatively constant throughout the exercise, as subjects started to sweat during the warm-up period. Therefore, electrode impedance did not change significantly throughout the TTE trial. Participants were not allowed to stand up from the saddle to minimize head and upper body movement as much as possible. However, some movement still occurs during an intense cycling trial, and, while we have addressed common noise issues from an experimental and computational side, there is a remaining issue as artifact sources were shown to greatly vary across subjects and efforts (Kline et al. 2015), making it difficult to reject artifacts based on templates and models. Therefore, most likely, many reports of EEG during locomotion contain a mixture of true electrocortical and artifact signals, which should be kept in mind by the reader.

GRANTS

H. Enders is supported by Vanier Canada [National Sciences and Engineering Research Council (NSERC)], the Killam Foundation, as well as the CREATE program of the NSERC. J. Baltich is supported by Vanier Canada (Canadian Institute of Health Research). A. B. Protzner is supported by NSERC, Canadian Foundation for Innovation Leaders Opportunity Fund, and Alberta Enterprise and Advanced Education Research Capacity Program, Alberta Alignment Grants.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: H.E., F.C., C.M., A.B.P., and B.M.N. conception and design of research; H.E., C.M., and J.B. performed experiments; H.E. and F.C. analyzed data; H.E., F.C., C.M., A.B.P., and B.M.N. interpreted results of experiments; H.E. prepared figures; H.E. drafted manuscript; H.E., F.C., C.M., J.B., A.B.P., and B.M.N. edited and revised manuscript; H.E., F.C., C.M., J.B., A.B.P., and B.M.N. approved final version of manuscript.