Cortex-muscle coherence (CMC) reflects coupling between magnetoencephalography (MEG) and surface electromyography (sEMG), being strongest during isometric contraction but absent, for unknown reasons, in some individuals. We used a novel nonmagnetic high-density sEMG (HD-sEMG) electrode grid (36 mm × 12 mm; 60 electrodes separated by 3 mm) to study effects of sEMG recording site, electrode derivation, and rectification on the strength of CMC. Monopolar sEMG from right thenar and 306-channel whole-scalp MEG were recorded from 14 subjects during 4-min isometric thumb abduction. CMC was computed for 60 monopolar, 55 bipolar, and 32 Laplacian HD-sEMG derivations, and two derivations were computed to mimic “macroscopic” monopolar and bipolar sEMG (electrode diameter 9 mm; interelectrode distance 21 mm). With unrectified sEMG, 12 subjects showed statistically significant CMC in 91–95% of the HD-sEMG channels, with maximum coherence at ∼25 Hz. CMC was about a fifth stronger for monopolar than bipolar and Laplacian derivations. Monopolar derivations resulted in most uniform CMC distributions across the thenar and in tightest cortical source clusters in the left rolandic hand area. CMC was 19–27% stronger for HD-sEMG than for “macroscopic” monopolar or bipolar derivations. EMG rectification reduced the CMC peak by a quarter, resulted in a more uniformly distributed CMC across the thenar, and provided more tightly clustered cortical sources than unrectifed sEMGs. Moreover, it revealed CMC at ∼12 Hz. We conclude that HD-sEMG, especially with monopolar derivation, can facilitate detection of CMC and that individual muscle anatomy cannot explain the high interindividual CMC variability.
- corticomuscular coherence
- multichannel EMG
- sensorimotor cortex
- spatial filter
during sustained isometric contraction, surface electromyogram (sEMG), representing the algebraic sum of motor unit action potentials (MUAPs), is coherent with magnetoencephalographic (MEG) and electroencephalographic (EEG) signals recorded from the primary motor cortex (M1) (Brown et al. 1998; Conway et al. 1995; Gross et al. 2000; Halliday et al. 1998; Salenius et al. 1997). This coupling, referred to as cortex-muscle coherence (CMC), typically occurs at 15–40 Hz, depending on the exerted force and the muscle (Hari and Salenius 1999), and it likely reflects oscillatory discharge of the M1-pyramidal neurons projecting to the spinal α-motoneurons and thus modulating the population-level motor unit discharges (Baker et al. 1997; Conway et al. 1995; Salenius et al. 1997).
CMC was first demonstrated with MEG recordings that measure cortical activity with millisecond temporal resolution. MEG picks up magnetic fields that are generated mainly by postsynaptic currents in the apical dendrites of cortical pyramidal neurons (Hämäläinen et al. 1993; Hari et al. 2015; Hari and Salmelin 2012). As the skull does not distort the MEG signals, pointlike cortical sources can be located with spatial resolution of a few millimeters (Hämäläinen et al. 1993). Accordingly, MEG-based CMC has been used in functional mapping of, e.g., the hand area of M1 (Bourguignon et al. 2013; Mäkelä et al. 2001). Unfortunately, CMC is not detected in all individuals, missing in ∼15% (Pohja et al. 2005).
Recent evidence indicates that CMC can be detected more robustly with high-density surface electromyography (HD-sEMG) than with the classical bipolar sEMG recording (van de Steeg et al. 2014). Indeed, by sampling the spatial distribution of sEMG with a grid of small, closely spaced electrodes, it is possible to identify the anatomical features of the muscle, to reduce the filtering effect associated with the electrode size (Farina and Merletti 2001; Helal and Bouissou 1992), and to apply signal transformations (either linear or nonlinear) to the set of simultaneously recorded sEMG signals. The combination of these features allows a better estimation of the level of muscle activation (Merletti et al. 2010).
The level of sEMG-sEMG coherence between two synergistic muscles, which stems from the common oscillatory synaptic inputs to α-motoneurons, also varies according to the sEMG recording site over the hand muscles (Keenan et al. 2011, 2012). Therefore, it is expected that the sEMG recording site would influence the level of CMC and thus the reliability of cortical source estimation, important for functional mapping of the rolandic cortex.
Here we aimed to evaluate the effect of 1) sEMG recording site (i.e., spatial location of the electrode on the skin over the muscle) and 2) sEMG derivation (monopolar, bipolar, and Laplacian) on the strength of CMC and on the cortical source estimates. The results obtained with the HD-sEMG derivations were further compared with results obtained with computationally produced “macroscopic” bipolar and monopolar sEMG derivations, mimicking conventional bipolar or monopolar sEMG recordings, in terms of electrode size (diameter 9 mm), interelectrode distance (IED, 21 mm), and electrode placement.
It has been suggested that it is possible to obtain more accurate timings for motor units with rectified than with unrectifed sEMG (Halliday and Farmer 2010), meaning that rectification could also affect the detection of corticospinal coupling. In fact, rectification of sEMG prior to coherence analysis has in some cases hindered CMC detection (McClelland et al. 2012), although this is not always the case (Yao et al. 2007). In this study we thus also compared our main CMC results obtained with unrectified sEMG with those obtained with unrectified sEMG.
MATERIALS AND METHODS
We studied 14 healthy subjects (mean age 28.1 yr, range 23–44 yr; 9 men, 5 women) without any history of neuropsychiatric disease or movement disorder. According to the Edinburgh handedness inventory (Oldfield 1971), all subjects were right-handed (mean score 93.2, range 73–100 on a scale from −100 to 100). The study had prior approval by the ethics committee of Aalto University. The subjects gave informed consent before participation. Subjects were compensated monetarily for lost working hours and travel expenses.
Figure 1A illustrates the experimental setup. MEG was recorded concomitantly with HD-sEMG from thenar muscle while the subject maintained a steady isometric thumb abduction.
Before MEG recordings, maximum isometric thumb abduction force was measured with a custom-made device comprising a rigid load cell (model 1042, dynamic range up to 29.4 N; Vishay Precision Group, Malvern, PA). The subjects performed two or three maximum contractions, each lasting for 3–4 s, with 2-min rest periods in-between. The trial with the highest force value was used as the maximum voluntary contraction (MVC) force (mean force over 1 s of stable contraction).
During MEG recordings, the subjects were sitting with their left hand on their thigh and their right hand on a table in front of them, inserted in the device to measure the isometric force. They were asked to maintain a steady isometric force and to fixate a blue line (representing their force level for the past 1 s) on a gray background in the middle of a translucent screen 180 cm in front of them. The target force level (6% of their MVC) was visible as a white continuous line, and the boundaries of the target force level (5% and 7% of their MVC) were presented with continuous black lines. When needed, the subject's vision was optically corrected with nonmagnetic goggles. The recording lasted for 4 min. The subjects were instructed to avoid eye movements and excessive blinking and to focus on maintaining the isometric contraction as steady as possible.
The measurements were carried out at the MEG Core of Aalto NeuroImaging, Aalto University. MEG signals were recorded in a magnetically shielded room (Imedco, Hägendorf, Switzerland) with a 306-channel whole-scalp neuromagnetometer (Elekta Neuromag; Elekta, Helsinki, Finland). The recording passband was 0.1–400 Hz, and the signals were sampled at 2,000 Hz. The subject's head position inside the MEG helmet was continuously monitored by five head-tracking coils located on the scalp; the locations of the coils with respect to anatomical fiducials were determined with an electromagnetic tracker (Fastrak; Polhemus, Colchester, VT).
Monopolar sEMG was measured from the right thenar (targeting mainly the abductor pollicis brevis muscle) with a custom-made nonmagnetic electrode grid of 60 silver electrodes (contact area 0.8 mm2 in each; LISiN, Politecnico di Torino) arranged in five columns (from medial to lateral: 10, 12, 13, 13, and 12 electrodes in each) aligned in the longitudinal direction of the muscle. IED was 3 mm in both longitudinal and transverse directions. Monopolar sEMG signals were recorded with respect to a reference electrode (Neuroline 720; Ambu, Ballerup, Denmark) placed on the lateral edge of the distal third of the right radial bone and then band-pass filtered (10–400 Hz) and sampled at 2,000 Hz (Elekta Neuromag; Elekta). The impedance (at 30 Hz; electrode impedance meter EZM5; Grass, West Warwick, RI) between the reference electrode and each HD-sEMG electrode was below 100 kΩ, and the sEMG signal quality was always checked visually during real-time monitoring prior to data acquisition.
The same custom-made device with which we measured the MVC force was used during the steady 4-min isometric thumb abduction, but the force was measured with a more sensitive load cell (model 1004; Vishay Precision Group) with a dynamic range up to 5.9 N. The force signal was low-pass filtered at 400 Hz and sampled at 2,000 Hz, time-locked to MEG and sEMG signals.
Three-dimensional T1 magnetic resonance images (MRIs) were acquired with a whole-body General Electric Signa VR 3.0T MRI scanner (Signa VH/I; General Electric, Milwaukee, WI) or a 3-T MAGNETOM Skyra whole-body scanner (Siemens Healthcare, Erlangen, Germany) at AMI Centre, Aalto NeuroImaging, Aalto University.
Continuous MEG data were first preprocessed off-line with the temporal signal-space separation (tSSS) method implemented in MaxFilter 2.2.10 (Elekta Neuromag; Elekta) to suppress external interferences and to correct for head movements (Taulu and Simola 2006). A zero-phase filter was used to band-pass filter MEG signals between 1 and 195 Hz and monopolar sEMG signals between 10 and 400 Hz. Monopolar sEMG signals from electrodes contaminated by artifacts (mean 1.7, range 0–5 of 60 electrodes) were discarded.
Three HD-sEMG (monopolar, bipolar, Laplacian) and two computationally produced “macroscopic” sEMG derivations (bipolar and monopolar) were tested (see Fig. 1, D and E). HD-sEMG monopolar derivation consisted of the 60 monopolar sEMG signals detected with respect to the reference electrode on the right radial bone. Longitudinal bipolar sEMG signals (55 in total) were computed from the monopolar derivations as the differences between adjacent monopolar sEMG channels in the longitudinal direction of the muscle, i.e., for each of the five electrode columns separately. Laplacian sEMG signals (32 in total) were derived from the monopolar sEMG signals by combining five crosswise electrodes, with the central electrode weighted −4 and the four surrounding electrodes weighted +1.
Macroscopic bipolar and monopolar sEMG signals were based on large “virtual” electrodes computationally produced by averaging a subgroup of the small-sized electrodes of the HD-sEMG grid (Staudenmann et al. 2006), and thus mimicking well the conventional sEMG recordings using large surface electrodes (see Fig. 1D). The macroscopic bipolar signal was derived as differential signal between two “virtual” electrodes, each consisting of the average of 12 monopolar sEMG channels in “circular” derivation (see Fig. 1D). The resulting electrode diameter was 9 mm, and the distance between the centers of the two electrodes was 21 mm. Similarly, the macroscopic monopolar signal was derived by averaging 12 monopolar sEMG channels in the middle portion of the HD-sEMG grid.
Cortex-muscle coherence analysis.
Continuous data from the recording were split into 1,000-ms epochs with 800-ms epoch overlap (Bortel and Sovka 2007), leading to frequency resolution of 1 Hz. Epochs with MEG signals exceeding 6 pT (magnetometers) or 1.4 pT/cm (gradiometers) were rejected to avoid contamination of the data by eye movements, muscle activity, and artifacts in the MEG sensors. Power and cross-spectra, as well as cross-correlograms, were calculated between MEG and unrectified RMS-normalized sEMG signals with the multitaper approach [5 orthogonal Slepian tapers, yielding a spectral smoothing of ±2 Hz (Thomson 1982)]. Coherence spectra were then computed according to the formulation of Halliday et al. (1995). The analysis was repeated for each sEMG channel in HD-sEMG monopolar, bipolar, Laplacian, and macroscopic bipolar and monopolar derivations separately. To clarify the effects of rectification of sEMG on CMC, the same analysis was repeated using rectified sEMG, and the results were compared with those obtained using unrectified sEMG signals.
Location of innervation zone.
The location of the main innervation zone (i.e., the end-plate area where the individual neuromuscular junctions are concentrated) was determined individually for each of the five sEMG electrode columns based on visual inspection of the bipolar sEMG signals (Barbero et al. 2011; Piitulainen et al. 2009). The innervation zone corresponded to the bipolar sEMG channel showing a low amplitude, associated with a reversal in signal polarity in the adjacent bipolar signals (either the proximal or distal one). Because of the short 3-mm IED, the innervation zone of different motor units occupied an area covered by more than one bipolar sEMG channel. For this reason, innervation zone area was defined for each column of the electrode grid, as the bipolar sEMG channel directly on top of the main innervation zone and one channel proximal and one channel distal to it (i.e., including three consecutive channels).
To study the potential effect of innervation zone on the level of CMC, the mean CMC level in the sEMG signals over the innervation zone area was compared with that of sEMG signals outside the innervation zone (i.e., extrajunctional channels) for each HD-sEMG derivation separately.
Location of coherent cortical sources.
For each sEMG signal in the HD-sEMG and macroscopic derivations, the cross-correlograms with all MEG signals were band-pass filtered at 6–45 Hz and the timing of its most prominent peak was determined among a fixed selection of 40 gradiometers that comprised all the most responsive sensors over the rolandic area of the left hemisphere. Then the source was estimated by fitting an equivalent current dipole (ECD) at the determined time to the spatial distribution of the filtered cross-correlograms, using the same fixed selection of 40 gradiometers and the corresponding 20 magnetometers. ECDs were estimated within a spherical head model fitted to individual MRIs (10 subjects) or to the standard Montreal Neurological Institute (MNI) brain (4 subjects) so as to match the centroparietal brain region. ECDs were then visualized on the coregistered individual MRIs.
Spread of coherent cortical sources.
For each subject, the center of mass of the fitted ECDs was first calculated as the mean of all ECD coordinates across all sEMG channels, for each HD-sEMG derivation separately. Then, to characterize the spread of the ECDs, their coordinates relative to their center of mass were subjected to a principal component analysis to construct an ellipsoid centered on their center of mass, with the lengths of the axes along principal components equal to one standard deviation as derived from singular values (Fischer et al. 2005). The volume and major axis of this ellipsoid provided a measure of the degree of the ECD spread around the center of mass.
sEMG amplitude and spatial cortex-muscle coherence maps over the muscle.
For each subject and HD-sEMG derivation (monopolar, bipolar, and Laplacian), maps for sEMG amplitude and CMC over the thenar were calculated. sEMG amplitude maps were computed as the RMS of each sEMG signal from the start of the steady phase of the isometric contraction to the end of the recording (i.e., for the whole 4-min contraction). CMC maps were computed as the maximum coherence value in each sEMG signal across the 7–35 Hz frequency range and across the fixed selection of 40 gradiometers over the rolandic area of the left hemisphere. The individual maps were normalized by their maximum value and averaged across all subjects to obtain group-level maps.
To estimate the homogeneity of sEMG amplitude and CMC over the muscle (i.e., recording sites of the sEMG), coefficients of variations were calculated across all sEMG channels separately for each subject and HD-sEMG derivation (monopolar, bipolar, and Laplacian).
Spatial correlation between sEMG amplitude and cortex-muscle coherence maps.
To estimate the possible association between sEMG signal amplitude and the level of CMC, Pearson's correlation coefficient was calculated between the sEMG amplitude and CMC maps separately for each subject and HD-sEMG derivation. We then used a one-sample t-test to assess whether correlation coefficients differed from zero for each EMG derivation separately.
Statistical significance of cortex-muscle coherence.
The statistical significance of the coherence level was assessed with surrogate data separately for each sEMG signal (Faes et al. 2004). First, 1,000 surrogate coherence spectra were computed between MEG signals and Fourier transform surrogate sEMG signals. The Fourier transform surrogate restrains the power spectrum to remain the same as in the original signal but replaces the phase of the Fourier coefficients by random numbers in the range [−π; π]. Then, the peak coherence value across the preselected 40 gradiometers in the 7–35 Hz frequency range was extracted for each surrogate coherence spectrum to compute the cumulative density function of the peak coherence value occurring because of stochastic matching between sEMG and MEG signals. The coherence thresholds at P < 0.05 corrected for multiple comparisons across MEG channels were then evaluated as the 95th percentile of the corresponding cumulative density function.
Comparisons between sEMG derivations.
A nonparametric Friedman test was used to compare the peak and mean CMC values between the five sEMG derivations. The same test was used to compare the spread of the ECD cluster (volume and length of the ellipsoid) in the left rolandic cortex, the coefficient of variation in the sEMG amplitude and level of CMC, and the mean CMC value on and outside the innervation zone area between the three HD-sEMG derivations. In the case of statistically significant differences, Wilcoxon signed-rank tests were performed between all combinations of the sEMG derivations and the respective P values were corrected according to Bonferroni. Wilcoxon signed-rank test was used also to compare the mean CMC value on and outside the innervation zone area for each HD-sEMG derivation separately. The threshold for statistical significance was set at P < 0.05 for all comparisons. Friedman and Wilcoxon tests were performed with IBM SPSS Statistics (version 22, SPSS, Chicago, IL).
Effect of rectification of sEMGs on cortex-muscle coherence.
In the statistical evaluation of the effects of sEMG rectification, we used a Wilcoxon signed-rank test (implemented in SPSS) to compare the peak and mean CMC values, the coefficients of variation of CMC level across the thenar, and the spread of the ECD clusters (volume and length of the ellipsoid) in the left rolandic cortex between coherence analyses using unrectified vs. rectified sEMG. The tests were run separately for each HD-sEMG derivation, and the threshold for statistical significance was set at P < 0.05.
Figure 1F illustrates the force, MEG, and sEMG signals of a representative subject (S1) during an isometric right thumb abduction task. All 14 subjects were able to maintain the isometric force within the task limits throughout the 4-min measurement, and they did not report perceptible muscle fatigue during the task.
In two subjects (1 male, 1 female), the CMC level was low and statistically significant in <20% of the HD-sEMG channels. These two subjects were excluded from further analysis, as their data would have added noise to our analyses of spatial variability of CMC over the thenar. The remaining 12 subjects displayed significant CMC in more than half of the HD-sEMG channels (mean ± SD 95 ± 12% in monopolar, 94 ± 7% in bipolar, and 92 ± 11% in Laplacian) and showed significant CMC for macroscopic monopolar signal, whereas CMC did not reach significance in 2 subjects for macroscopic bipolar signal. Figure 2 shows the CMC spectra for the most and least coherent sEMG channels for each sEMG derivation. Table 1 summarizes the peak frequency of CMC that always peaked at ∼25 Hz for all five sEMG derivations.
Table 2 summarizes the CMC results. Peak CMC differed significantly between the sEMG derivations [χ2(4) = 23.5, P < 0.001]. The macroscopic bipolar signal showed lower peak CMC than the HD-sEMG monopolar (Z = −2.9, P = 0.04) and bipolar (Z = −2.8, P = 0.05) derivations. The macroscopic monopolar signal also showed lower peak CMC than HD-sEMG monopolar (Z = −3.1, P = 0.02) derivation but did not differ from the other derivations.
Significant differences were detected in mean CMC (across all sEMG channels) between the HD-sEMG derivations [χ2(4) = 18.3, P < 0.001]; monopolar derivation showed higher mean CMC (0.056 ± 0.052) than bipolar (0.039 ± 0.033, Z = −2.7, P = 0.018) and Laplacian (0.040 ± 0.033, Z = 3.1, P = 0.006) derivations. Interindividual CMC variability (i.e., coefficient of variation across subjects' mean CMC values) was 93% for monopolar, 84% for bipolar, and 86% for Laplacian derivation.
Effect of sEMG Recording Site on Cortex-Muscle Coherence
Figure 3 shows normalized sEMG amplitude and CMC maps superimposed with location of the main innervation zones. As expected, high sEMG amplitude was observed on the main innervation zone in the monopolar and Laplacian derivations and low amplitude in the bipolar derivation. As is evident also from the maps, the coefficient of variations of the sEMG amplitude [χ2(2) = 12.7, P < 0.05] and the CMC [χ2(2) = 13.2, P < 0.01] differed significantly between the sEMG derivations. Table 3 summarizes coefficient of variation results for CMC (i.e., intraindividual variation of CMC across the sEMG recording sites). More uniform distributions of sEMG amplitude and CMC were observed for the spatially least selective monopolar derivation (mean ± SD for coefficient of variations 0.28 ± 0.11 of EMGRMS) than for bipolar (0.43 ± 0.10, Z = −2.9, P = 0.012 for EMGRMS and Z = −2.7, P = 0.018 for CMC) and Laplacian (0.49 ± 0.18, Z = −3.2, P = 0.003 for EMGRMS and Z = −3.1, P = 0.006 for CMC) derivations, whereas the bipolar and Laplacian derivations did not differ significantly (Z = −1.2, P = 0.239 for EMGRMS and Z = −0.9, P = 0.388 for CMC).
Spatial Correlation Between sEMG Amplitude and Cortex-Muscle Coherence Maps
Individual subjects showed high correlation coefficients (up to r = 0.88) between sEMG amplitude and CMC maps. The coefficients were predominantly positive for monopolar (75%) and bipolar (80%) derivations but even for Laplacian (50%) derivation. However, the group mean (n = 12) of the correlation coefficients did not differ significantly from zero (monopolar P = 0.21; bipolar P = 0.06; Laplacian P = 0.68).
Effect of Innervation Zone on Cortex-Muscle Coherence
Figure 3B shows the mean ± SD location of the subjects' main innervation zones superimposed on group-level CMC maps. The spatial relationship between sEMG electrodes and the innervation zone did not affect the level of corticospinal coupling, as the mean CMC value did not differ between the innervation zone and extrajunctional areas in any of the HD-sEMG derivations (Z from −1.1 to −1.6 and all P > 0.1).
Location of Coherent Cortical Sources
Figure 4 shows the location of coherent sources and magnetic field patterns for different sEMG derivations. Two subjects are shown, one subject (S14) with statistically significant CMC in 98% of the sEMG channels and one subject (S12) whose data were excluded from further analyses because of very weak CMC, reaching the significance level only in 0–16% of the sEMG channels. In all subjects, the main cluster was located in the left-hand area of the rolandic cortex, either in the precentral or postcentral gyrus or spreading to both.
Spread of Coherent Cortical Sources
The spread of the source clusters (volume and length of the ellipsoid, including all sEMG channels) differed significantly between the sEMG derivations [volume: χ2(2) = 18, P < 0.001; length: χ2(2) = 12, P = 0.04]. The coherent sources derived from the monopolar sEMG signals were clustered within a smaller region (mean ± SD: volume 0.8 ± 1.7 cm3; length 0.8 ± 0.7 cm) compared with bipolar (volume: 4.0 ± 5.3 cm3, Z = −3.1, P = 0.006; length: 1.3 ± 0.7 cm, Z = −2.2, P = 0.048) and Laplacian (volume: 4.9 ± 6.5 cm3, Z = −3.1, P = 0.006; length: 1.4 ± 0.9 cm, Z = −3.0, P = 0.009) sEMG signals.
Effect of Rectification of sEMGs on Cortex-Muscle Coherence
Table 1 presents mean frequency of CMC for unrectifed and rectified sEMG derivations. Unrectified sEMG signals yielded significant CMC at ∼25 Hz. However, with rectified sEMG nine subjects showed significant CMC also at ∼12 Hz and up to four subjects had their dominant peak at ∼12 Hz.
Table 2 summarizes the CMC results for unrectified and rectified sEMG derivations. For monopolar and Laplacian derivations, the peak CMC was about a quarter stronger with unrectified than rectified sEMG (mean increase 25.6%, Z = −2.3, P = 0.023 for monopolar and 25.4%, Z = −2.0, P = 0.041 for Laplacian). No significant effect of rectification was observed for the bipolar or macroscopic derivations or for the mean CMC (across all sEMG channels).
For all derivations, the coefficient of variation of CMC across the thenar was significantly lower for rectified sEMG (see Table 3; Z = −2.1, P = 0.034 for monopolar, Z = −3.0, P = 0.003 for bipolar, and Z = −2.9, P = 0.004 for Laplacian).
Rectification strikingly reduced cortical source cluster volume (by 93%, Z = −2.3, P = 0.023 for monopolar; by 92%, Z = −3.1, P = 0.002 for bipolar; by 85%, Z = 3.0, P = 0.003 for Laplacian) and ellipsoid length (by 63%, Z = −2.8, P = 0.005 for monopolar; by 59%, −3.1, P = 0.002 for bipolar; by 54%, Z = 2.7, P = 0.006 for Laplacian). After rectification, the volumes were 0.05 ± 0.12 cm3, 0.30 ± 0.78 cm3, and 0.75 ± 1.54 cm3 and the cluster ellipsoid lengths were 0.3 ± 0.2 cm, 0.5 ± 0.6 cm, and 0.7 ± 0.6 cm for monopolar, bipolar, and Laplacian derivations, respectively.
We showed that the level of corticospinal coupling, measured with CMC, is affected by the recording site of sEMG over the thenar and by the derivation used to detect the sEMG signals. The monopolar HD-sEMG derivation yielded on average ∼20% stronger CMC, more uniform CMC distribution over the thenar, and tighter cortical source clusters than bipolar and Laplacian HD-sEMG derivations. In subjects showing robust CMC, the coherent cortical sources were tightly clustered in the right rolandic hand area, whereas more scattered source clusters were observed in subjects demonstrating weak CMC. The HD-sEMGs provided on average ∼20–30% stronger peak CMC than the respective derivations mimicking bipolar macroscopic or monopolar sEMG recordings. Finally, sEMG rectification reduced the peak CMC by a quarter, but it provided more uniform distribution of CMC values across the thenar, which might explain the tighter source clusters.
Spatial Variability of CMC Across Muscle Surface
Both bipolar and Laplacian HD-sEMG derivations provide spatially more selective signals than monopolar HD-sEMG (Disselhorst-Klug et al. 1997; Farina et al. 2003). The broad sensitivity pattern of the monopolar derivations largely explains the uniform distribution of the CMC level across the muscle surface.
The local muscle anatomy (e.g., orientation, size, and depth of muscle fiber) affects the amplitude distribution of sEMG, especially in spatially sensitive bipolar sEMG recordings (Mesin et al. 2009a; Farina et al. 2002), and it may in part explain the greater spatial variability of CMC level. Action potentials start to propagate from innervation zones that are typically located in the middle portion of the muscle and prominently visible in bipolar HD-sEMG (Masuda and Sadoyama 1986, 1988). The strongest monopolar sEMG signals are typically measured above the innervation zone, whereas bipolar sEMG signals are reduced in amplitude at the same area (Kleine et al. 2000; Roy et al. 1986). The locations of the innervation zones vary between individuals (Masuda et al. 1985) and muscles (Barbero et al. 2012; Rainoldi et al. 2004), and they can vary also with respect to the sEMG electrodes in a force level-dependent manner due to the relative motion between the muscle and the skin (Piitulainen et al. 2009).
Somewhat surprisingly, the local sEMG amplitude variations related to the locations of the innervation zones (especially the amplitude drops in bipolar derivation) were not systematically reflected in the strength of CMC. This result may reflect our short (3 mm) IED, which is likely less than (or similar to) the spread of the innervation zones. Therefore, the effect of the innervation zone can be distributed over a set of HD-sEMG signals and the drop in bipolar amplitude can be less than expected.
Signal-to-noise ratio of sEMG was sufficient for CMC analysis throughout the muscle, including the innervation zone. However, the amplitudes of both sEMG and CMC showed clear spatial variability. According to simulations, an increase in signal amplitude may increase the level of coherence (i.e., positive correlation), but only when the signal-to-noise ratio is low (Muthukumaraswamy and Singh 2011). However, the signal-to-noise ratio of sEMG is relatively high, including the coherent ∼25-Hz band. The spatial correlation coefficient between sEMG power and CMC was positive in 50–80% of our subjects depending on the HD-sEMG derivation, but it did not differ significantly from zero at group level. We thus conclude that the level of CMC was not affected by sEMG signal amplitude.
In addition to sEMG amplitude, the shapes of the MUAPs (and thus sEMG power spectra) vary with the sEMG recording site, and the change occurs more rapidly in space with spatially more selective sEMG recordings (Kleine et al. 2007). The variations in the spatial dimension of the sEMG signal may in part explain the overall variation in the level of CMC across the muscle surface, as well as the finding that the level of CMC computed with monopolar sEMG recordings showed a more uniform distribution across the muscle surface.
Finally, the individual muscle anatomy (i.e., sEMG recording site) cannot solely explain the high CMC variability across the subjects. The interindividual CMC variability was 84–93%, and, in comparison, intraindividual-CMC variability (i.e., across sEMG recording sites) was on average only 19–35% for the HD-sEMG derivations.
Contribution of End-of-Fiber Effects to Monopolar sEMG
Two main components can be identified in monopolar sEMG. The first reflects propagating MUAPs along the muscle fibers and the second the nonpropagating end-of-fiber effects arising when the MUAPs meet the muscle-tendon junction (Stegeman et al. 1997). Importantly, the currents and potentials associated with these two phenomena differ. The MUAPs are associated with quadrupolar intracellular currents within the muscle fibers, with one depolarizing current dipole in the leading edge of the MUAP and one repolarizing current dipole at the trailing end of the MUAP. In contrast, the end-of-fiber effects arise when the MUAP reaches the tendon at the muscle-tendon junction, and then for a short time (for the duration of the MUAP) only the trailing-end dipole of the MUAP prevails.
The electric potentials due to nonpropagating end-of-fiber effects (due to the intracellular current dipoles) spread further in space than those of the propagating MUAPs (intracellular current quadrupoles) because electric potential decays with distance r from the source approximately as 1/r−2 for dipolar and 1/r−3 for quadrupolar sources. The resulting monophasic potentials are positive over the entire muscle, and thus simultaneously arriving MUAPs effectively summate. Instead, the presence of positive and negative phases in the potential distributions associated to the propagating MUAPs easily cancel each other because of asynchronous firings of different motor units and variation in conduction velocities in muscle fibers of different sizes. Consequently, the monopolar sEMG signals are less affected by amplitude cancellation (Staudenmann et al. 2010; Tucker and Turker 2005), and they thus provide a better estimation of muscle force compared with bipolar and Laplacian derivations (Staudenmann et al. 2006).
When more than single motor units are recruited, the MUAPs of different sizes and firing frequencies are superimposed and start to form interference patterns where single MUAPs are difficult to discern and potentials of different phases cancel each other, thereby affecting the amplitude of the sEMG (Day and Hulliger 2001; Keenan et al. 2005). It may thus become more difficult to detect from the sEMG the common oscillatory synaptic input to α-motoneurons (Farina et al. 2013), which is suggested to be the coherent signal with the cortical activity (Baker et al. 1997; Conway et al. 1995; Salenius et al. 1997). The effective summation of the end-of-fiber effects could carry the coherent information (i.e., timing of the oscillatory inputs) and may thus in part explain the ∼20% stronger level of CMC when using monopolar rather than bipolar HD-sEMG signals. The degree of amplitude cancellation increases with muscle contraction intensity, i.e., with the number and discharge rate of active motor units (Farina et al. 2008; Keenan et al. 2005). Although the intensity of the muscle contraction was low (∼6% of MVC), the effects of amplitude cancellation on the present results cannot be ruled out.
Common-mode signals are effectively suppressed in bipolar and Laplacian derivations so that these recordings are thus rather selective to the muscle just under the electrodes, whereas the monopolar sEMG is more prone to cross talk from the other nearby muscles. Because thenar contains several muscles that cannot be selectively activated, cross talk did in our study more likely affect monopolar than bipolar sEMG (Dimitrov et al. 2003; Mesin et al. 2009b).
Finally, our results imply that other factors affecting the quality of sEMG amplitude estimation, such as the effect of anatomical properties of the muscle, influence the CMC to a lesser extent than do the end-of-fiber effects and amplitude cancellation (see Spatial Variability of CMC Across Muscle Surface).
HD-sEMG vs. “Macroscopic” sEMG Recordings
van de Steeg et al. (2014), using an 8 × 8 HD-sEMG electrode grid, found stronger CMC for monopolar and Laplacian sEMG than for bipolar sEMG. They also computed CMC after applying principal component analysis to remove the most common components (across all channels of the electrode grid) from the monopolar sEMG. The obtained CMC levels were similar to those with “raw” monopolar and Laplacian signals (no statistics performed) but, again, significantly higher than with the bipolar sEMG derivation. These results are not directly comparable with ours, because the monopolar sEMG was rereferenced to one sEMG channel of the electrode grid on the same muscle, while we used a remote reference electrode on the radial bone. However, our findings are in line with their observations that the use of HD-sEMG may enhance detection of statistically significant CMC.
We further compared the HD-sEMG derivations with sEMG derivations mimicking conventional macroscopic bipolar and monopolar sEMG recordings and observed ∼20–30% stronger CMC than with the respective HD-sEMG approaches. The benefit of HD-sEMG approaches is their ability to “scan” nearly the whole muscle surface with the HD-sEMG electrode grid to identify the optimal sEMG recording site for CMC. We expected the spatially more selective sEMG derivations (bipolar and Laplacian) to perform better in identifying the optimal sEMG recording site, but the peak CMC (see Table 2) did not differ between the HD-sEMG derivations. However, the bipolar and Laplacian derivations showed lower mean CMC and higher spatial variability in the level of CMC across the muscle, decreasing the probability to detect significant CMC.
Our results indicate that a more “global” measure of muscle activity can enhance the detection of the CMC, possibly because CMC likely reflects the modulation of α-motoneuron drive to a population of motor units rather than to single motor units (Williams and Baker 2009). In our case, monopolar sEMG recording appeared optimal for CMC detection. Whenever HD-sEMG grids are not available, our results suggest the use of a single monopolar sEMG recording with remote reference on an inactive site (e.g., on bone) to measure CMC, taking in account that the monopolar recordings may be more prone to interference (e.g., from power lines) in noisy environments.
Estimates of Coherent Cortical Sources
In accordance with previous studies, the main cluster of the coherent cortical sources was located in the hand area of the left rolandic cortex (Conway et al. 1995; Salenius et al. 1997). In subjects showing significant CMC for most sEMG channels (>90% of all channels), the sources were clustered within a small volume in the cortex (<0.02 cm3 for monopolar sEMG CMC), whereas more scattered source estimates were observed in subjects whose CMC barely reached the significance threshold. This negative correlation between the estimated source volume and CMC level suggests that the spread of the sources merely reflected the signal-to-noise ratio of cross-correlograms (which covaries with the level of coherence) from which the sources were estimated. Accordingly, the better spatial selectivity of the sEMG (obtained with bipolar derivations) was associated with lower CMC level and larger spread of cortical sources. Nevertheless, it was possible to define the hand area in the left M1 for all 12 subjects (statistically significant CMC for >58% of the sEMG channels), relying on the location of the main source cluster.
Prominent Cortex-Muscle Coherence at ∼12 Hz with Rectified sEMGs
In 9 of 12 participants, rectification of sEMG signals revealed corticospinal coupling at ∼12 Hz that was not visible in CMC computed with unrectified sEMGs. With rectified sEMG, CMC typically peaks between ∼15 and 30 Hz during weak isometric contractions (Brown et al. 1998; Conway et al. 1995; Gross et al. 2000; Halliday et al. 1998; Salenius et al. 1997), but peaks at ∼6–15 Hz have been detected in some individuals with MEG (Salenius et al. 1997) or epicortical recordings over M1 (Raethjen et al. 2002) as well as in epicortical recordings over the primary somatosensory cortex and the supplementary motor area (Ohara et al. 2000). It is apparent that the presence of the 12-Hz CMC only with rectified, and not with unrectified, EMG will require more rigid examination in the future, using for example variable contraction levels that result in different levels of motor unit recruitment and firing frequencies, and consequently to different levels of cancellation between the motor unit action potentials in the sEMG. Indeed, recent evidence suggests that the oscillatory cortical input components may be more strongly represented in rectified or unrectified sEMG signals depending on the degree of amplitude cancellation of sEMGs (Farina et al. 2013).
Rectification of sEMG has in some studies hindered CMC detection (McClelland et al. 2012), and van de Steeg et al. (2014) found a tendency toward reduced peak and mean CMC after rectification of sEMG. Our results support these observations, as rectification of the sEMG signals reduced the peak of the 20-Hz CMC (in our case, by a quarter) but, surprisingly, provided more uniform distribution of CMC values across the thenar and more tightly clustered cortical sources. It is likely that the more condensed source clusters simply reflect the smaller spatial variation of the CMC across the thenar. As rectification of sEMG may improve detection of motor unit timing (Halliday and Farmer 2010), the common oscillatory activity of the motor units could also be better presented, therefore enhancing detection of the CMC.
Implications for Functional Mapping of M1
To date, functional magnetic resonance imaging (fMRI) has been the main tool to pinpoint the rolandic cortex during noninvasive presurgical evaluation (Bartsch et al. 2006; De Tiège et al. 2009). Unfortunately, interpretation of fMRI maps is challenging in patients with altered neurovascular coupling due to various brain disorders (Bartsch et al. 2006; D'Esposito et al. 2003; Korvenoja et al. 2006; Krings et al. 2001). In such cases, MEG may represent an alternative to fMRI, as it provides direct information about neuronal activity. In some patients with space-occupying lesions, MEG may be superior to fMRI in identifying the central sulcus (Korvenoja et al. 2006; Mäkelä et al. 2006).
CMC-based functional mapping of the cortex can be used to locate, e.g., the primary motor hand area (Bourguignon et al. 2013; Mäkelä et al. 2001). With conventional macroscopic single-channel sEMG recording, the CMC is not detectable in all subjects (nonsignificant in 2 of 12 individuals in study of Pohja et al. 2005), but studies in larger populations are lacking. We observed that monopolar HD-sEMG enhances the detection of significant CMC, and it also yields multiple source estimates (one for each sEMG signal) to strengthen the validity of the source identification. However, the drawback of HD-sEMG is the time-consuming preparation, measurement, analysis, and interpretation, as well as the need for special apparatus and some special user expertise. Finally, if single-channel sEMG recordings are to be used in CMC-based functional mapping, it would be beneficial to use monopolar sEMG and to record activity from as many synergistic muscles as possible to obtain a large number of estimates for the coherent cortical sources.
We have shown that the level of coupling between the rolandic cortex and muscle activity depends on the sEMG derivation used but is not systematically affected by the site of the sEMG recording over the muscle. The individual muscle anatomy therefore does not explain the large interindividual variability of CMC levels. Nevertheless, HD-sEMG recordings can facilitate detection of CMC. Monopolar sEMG resulted in the strongest and most uniform CMC across the thenar and can thus be recommended for CMC recordings. The effects of EMG rectification on CMC need further exploration.
This study was supported by the Academy of Finland (Grants 131483 and 263800 to R. Hari and Grant 13266133 to H. Piitulainen) and by the SalWe Research Program for Mind and Body (Tekes—the Finnish Funding Agency for Technology and Innovation Grant 1104/10) and the European Research Council (Advanced Grant 232946 to R. Hari).
No conflicts of interest, financial or otherwise, are declared by the author(s).
Author contributions: H.P., A.B., V.J., and R.H. conception and design of research; H.P. and A.B. performed experiments; H.P., A.B., and M.B. analyzed data; H.P., A.B., M.B., and R.H. interpreted results of experiments; H.P., A.B., and R.H. prepared figures; H.P. and A.B. drafted manuscript; H.P., A.B., M.B., V.J., and R.H. edited and revised manuscript; H.P., A.B., M.B., V.J., and R.H. approved final version of manuscript.
We thank Helge Kainulainen for technical support.
- Copyright © 2015 the American Physiological Society