Context-dependent reduction in corticomuscular coupling for balance control in chronic stroke survivors

Abstract

Balance control is an important indicator of mobility and independence in activities of daily living. How the functional coupling between the cortex and the muscle for balance control is affected following stroke remains to be known. We investigated the changes in coupling between the cortex and leg muscles during a challenging balance task over multiple frequency bands in chronic stroke survivors. Fourteen participants with stroke and ten healthy controls performed a challenging balance task. They stood on a computerized support surface that was either fixed (low difficulty condition) or sway-referenced with varying gain (medium and high difficulty conditions). We computed corticomuscular coherence between electrodes placed over the sensorimotor area (electroencephalography) and leg muscles (electromyography) and assessed balance performance using clinical and laboratory-based tests. We found significantly lower delta frequency band coherence in stroke participants when compared with healthy controls under medium difficulty condition, but not during low and high difficulty conditions. These differences were found for most of the distal but not for proximal leg muscle groups. No differences were found at other frequency bands. Participants with stroke showed poor balance clinical scores when compared with healthy controls, but no differences were found for laboratory-based tests. The observation of effects at distal but not at proximal muscle groups suggests differences in the (re)organization of the descending connections across two muscle groups for balance control. We argue that the observed group difference in delta band coherence indicates balance context-dependent alteration in mechanisms for the detection of somatosensory modulation resulting from sway-referencing of the support surface for balance maintenance following stroke.

Introduction

Poor ability to maintain an upright stance is a significant problem following stroke. It is known to restrict mobility, reduce independence in activities of daily living, and increase the risk of falling during functional tasks (Alenazi et al. 2018; Li et al. 2019; Wong et al. 2016). The control of balance and its recovery in people with stroke may depend on changes in the brain regions and the underlying descending pathways (Braun et al. 2007; Fujimoto et al. 2014; Grefkes and Ward 2014; Mihara et al. 2012b; Mima et al. 2001). Standing balance relies on the ability to maintain body’s center of mass within the base of support (Goel et al. 2019; Li et al. 2019) by integrating visual, vestibular, and somatosensory inputs (Goel et al. 2019; Mohapatra et al. 2014). This processing is usually driven by spinal and subcortical regions (Goel et al. 2019; Sherrington et al. 2017). When the standing balance is challenged, conscious cortical processing is triggered to control lower limb muscles for balance control (Fujimoto et al. 2014; Hülsdünker et al. 2015; Mihara et al. 2012a; Ozdemir et al. 2018; Slobounov et al. 2006). Following stroke, the adaptive and maladaptive changes in the cortex and its altered connections with spinal cord/muscles might influence the planning and execution of motor commands for balance compensation. There is a positive correlation between the integrity of underlying corticospinal pathways and motor function in stroke (Carlowitz-Ghori et al. 2014; Lindenberg et al. 2010; Yoo et al. 2019). How stroke affects the communication between the brain and the muscle involved in the maintenance of upright stance is not completely understood.

Cortico-muscular coherence quantifies the degree of oscillatory coupling (i.e., in the frequency domain) between the cortex and a muscle. In healthy adults, significant levels of coherence has been observed during challenging standing tasks that require a dynamic control of posture such as during foot stamping and stepping, changes in stance conditions, platform translation (Masakado et al. 2008; Peterson and Ferris 2019; Stokkermans et al. 2022). In contrast, corticomuscular coherence was found to be absent during quiet standing in young adults (Masakado et al. 2008). These findings are consistent with previous literature suggesting the involvement of cortical processing when the balance is challenged. Evidence suggests that corticomuscular coherence in young adults is observed in theta, beta, and gamma frequency bands for the leg muscles during challenging balance tasks (Jacobs et al. 2015; Masakado et al. 2008; Peterson and Ferris 2019; Stokkermans et al. 2022), but is not observed in the delta frequency band (Ozdemir et al. 2018). A delta band corticomuscular coherence has been observed in healthy older adults, along with coherence in theta, beta, and gamma frequency bands immediately following an unexpected translation of the support surface (Ozdemir et al. 2018). The occurrence of significant coherence in a broader frequency spectrum in older adults during a challenging balance task might reflect a shift from spinal to supraspinal control of the leg muscle activity due to age-related neuromuscular deterioration (Baudry 2016; Papegaaij et al. 2014).

In subacute and chronic stroke, corticomuscular coherence measures may indirectly represent cortical reorganization (Carlowitz-Ghori et al. 2014), and higher coherence may suggest better functional recovery (Grefkes and Ward 2014; Hara 2015; Mima et al. 2001; Olafson et al. 2021; Swayne et al. 2008; Wist et al. 2016). Several stroke studies have focused on understanding changes in corticomuscular coherence on the affected side during upper extremity tasks (Braun et al. 2007; Carlowitz-Ghori et al. 2014; Fang et al. 2009; Graziadio et al. 2012; Mima et al. 2001). A recent stroke study investigated corticomuscular coherence during a unilateral static ankle dorsiflexion task and found significantly lower coherence in beta and gamma frequency bands in stroke patients as compared to healthy individuals (Xu et al. 2023). The beta band coherence is associated with the ability to perform submaximal isometric contractions while the gamma band coherence might represent stronger muscle force production (Brown and Marsden 1998; Conway et al. 1995; Halliday et al. 1995; Kilner et al. 2000). Further, studies focusing on upper extremity tasks found significantly lower coherence for distal muscles and unchanged coherence for proximal muscles on the affected side when compared to the non-affected side in chronic stroke patients and to the dominant side in healthy controls (Mima et al. 2001; Zhou et al. 2021). These findings were explained in terms of distinct organization of pyramidal pathways to distal and proximal muscles (Grosse et al. 2002), differences in recovery of the stroke-affected descending connections to the muscle groups (Mima et al. 2001) and/or compensatory involvement of proximal muscles following stroke to accomplish the task goal (Zhou et al. 2021). The alterations in corticomuscular coherence of the affected leg muscles during a standing balance task following a stroke remain unexplored.

The purpose of this study was to investigate corticomuscular coherence for affected leg muscles during a standing balance task with varying difficulty (i.e., challenges) over multiple frequency bands in chronic stroke survivors when compared with healthy adults in the same age range (healthy controls). We instructed participants to maintain balance under different balancing contexts such as fixed support surface and sway-referenced support surface. The gain of the sway-referenced support surface was altered leading to distortions in somatosensory inputs, thus affecting balance performance (Goel et al. 2019). The distorted somatosensory inputs acted as a challenge to the upright stance, thus requiring online adjustments of postural movements without which the stability is compromised resulting in a fall. We simultaneously performed electroencephalography (EEG) and electromyography (EMG). We expected that, when compared with healthy controls, participants with stroke would demonstrate smaller coherence for the leg muscles over a broader frequency range. We also expect that this group difference would be greater during performance of the balance task under challenging balance contexts (i.e., sway-referenced support surface). Because a stroke lesion may differentially influence corticomuscular coherence for proximal and distal leg muscles (Mima et al. 2001), we separately investigated coherence for proximal and distal leg muscle groups.

Methods

Participants

Fourteen participants with stroke (age: 61.93 ± 8.97 years, mean ± SD; range: 37–70) and ten healthy adults (age: 55.3 ± 8.65 years, mean ± SD; range: 35–65) provided informed written consent to participate in this study. Demographics are presented in Table 1. The inclusion criteria for participants with stroke included middle cerebral artery (MCA) stroke for the first time, at least 12 months post stroke, ability to stand for 5 min independently without assistance, between 18 and 90 years of age, have MoCA score ≥ 26 (Bränström et al. 2021; Trzepacz et al. 2015), and absence of other neurological/ musculoskeletal impairments. Healthy adults (controls) were recruited if they had no history/symptoms of neurological/neuromuscular disorders affecting lower limbs with a MoCA score of 26 or greater. Healthy adults were right foot dominant based on self-reporting of the leg with which they preferred to take a step. This study was approved by the Institutional Review Board at the University of Houston.

Table 1.

Clinical information of patients

Stroke Patient	Age of testing	Sex	Lesion Site	Stroke Chronicity at testing (mths)	BBS Score (Total = 56)	TUG Score (sec)
1	61	M	Right MCA	12	44	60b
2	68	M	Right MCA	≥ 12a	44	20
3	67	F	Right MCA	56	46	25
4	68	M	Left MCA	> 12a	46	24
5	59	F	Left MCA	17	53	15
6	51	M	Right MCA	100	34	46.9
7	66	M	Right MCA	16	49	15.4
8	37	M	Right MCA	37	41	17.4
9	70	M	Right MCA	33	44	13.1
10	58	F	Right MCA	180	45	14.9
11	65	M	Left MCA	15	32	18.3
12	60	M	Right MCA	72	39	26.06
13	68	M	Right MCA	24	40	9.55
14	69	M	Right MCA	102	33	24.31

^aexact data is not available;

^bthe test was performed with ankle foot orthosis; F, female; M, male; MCA, middle cerebral artery; BBS: Berg Balance Scale; TUG: Timed Up and Go

Instrumentation

Computerized dynamic posturography (CDP)

The dynamic balance stability was assessed using a commercially available CDP force platform (Neurocom Balance Master, Natus Medical Incorporated, Pleasanton, CA). It is extensively used both in clinical (Cohen and Kimball 2008) and research settings (Wood et al. 2015) for monitoring sensory and motor performance aspects of the balance control system. The support surface is equipped with a motorized dual force plate system (45.72 cm × 45.72 cm), in which ground reaction forces (GRF) from under the subject’s feet are collected by normal and shear force transducers embedded within the force plate (support surface). The system can be used to adjust the orientation of the force plate with respect to the gravitational vertical by rotating it in the sagittal plane about an axis through the subject’s ankle joint in some proportion (a pre-selected gain between −2 to + 2) to the postural sway of the subject. A negative sway gain means the movement of the plate will be in the opposite direction of the subject’s COP, and a positive sway gain means the movement of the plate will be in the same direction as the subject’s COP. The transducer data was collected at 100 Hz and processed by pre-installed software on a Windows-based desktop connected to the Neurocom Balance Master (Research module, Neurocom software version 8.0, Natus Medical Incorporated, Pleasanton, CA). The Neurocom system also generated an analog timing signal, which was used to synchronize the electroencephalography (EEG) and the electromyography (EMG) system with the GRF data.

Clinical assessment of balance and mobility

Berg Balance Scale (BBS): The scale consists of a series of 14 predetermined tasks to assess static and dynamic balance. Each task is scored, ranging from 0 to 4; “0” indicates the lowest functional level, and “4” indicates the highest functional level. The score can take a value between 0 and 56 with higher scores suggestive of better balance (Berg et al. 1992).

Timed Up and Go (TUG) Test: The TUG test measures dynamic balance and functional mobility. The subjects were instructed to rise from an armchair, walk 3 m from the marker, turn around, walk back to the chair, and sit down. The time to complete this test was measured in seconds. More time taken is associated with poor dynamic balance/mobility (Podsiadlo et al. 1991). Three trials were performed, and a mean score (seconds) was used in subsequent analysis.

Laboratory based assessment of reactive balance control

We designed a dynamic balance paradigm, using a continuous balance task, to study the functional coupling between the cortex and leg muscles during dynamic balance control. The balance control system must continuously process information from multisensory systems to assess balance stability and orientation of the body and initiate appropriate behavioral responses in case of a disturbance to prevent a fall. During the continuous balance task, participants stood on the balance master with eyes closed while EEG, EMG, and GRF data were collected (Fig. 1A). The balance context was changed by varying the responsiveness of the support surface in proportion to the estimated center of mass. The sway-referencing of the support surface reduces the contribution of lower limb somatosensory receptors to the control of dynamic balance (Rasman et al. 2018) (Fig. 1B), and thereby manipulates the reliance on somatosensory inputs for the control of ongoing stance and modifies the feedback relationship within the balance control loop. The effects of such manipulation on dynamic stability are related to the action of the participant. We altered the relationships between postural sway and somatosensory inputs by varying the gain of the support surface in the sagittal plane in different proportions to the estimated instantaneous COM sway angle. We used a range of gains (−1.0, −0.4, 0, 0.4, 0.6, 1.0, 2.0) in the balance task of three different levels of balance difficulty: low, medium, and high. The lowest gain value of 0 (a quiet stance, fixed support surface with no sway referencing) may not be challenging to subjects and therefore classified as low difficulty. Higher gain values of 0.4, 0.6, and 1.0 were classified as medium difficulty as subjects were expected to exhibit greater balance instability. However, the gain of 2.0, or negative gains of −0.4 and − 1.0 were classified as that of high difficulty due to expectations that participants would have a further increase in postural instability, as done in our previous work (Goel et al. 2019; Ozdemir et al. 2016, 2018).

Fig. 1.

Experimental setup and the continuous balance task. (A) Experimental setup for the posture task. (B) Continuous balance task. (C) Representative EEG (Cz) and EMG traces for all five muscles for a single trial from one subject

The continuous balance task lasted for 180 s, with three 20 s testing conditions at a gain of 0 and one 20 s condition for each of the other four gain values presented in the following order for each participant: 0, 0.6, −0.4, 0, 1.0, −1.0. During the task, participants were instructed to maintain their balance. The order of these conditions was chosen so that we first had a low difficulty condition with a stable support surface (gain value of 0), and then the difficulty was gradually increased to medium level (gain values of 0.6 or 1.0), and then increased further to high level (gain values of −0.4 or −1.0). The general pattern of low, medium, and high difficulty was repeated three times. These varying balance conditions allowed us to gradually increase the difficulty of the balance control task while simultaneously monitoring the EEG, EMG, and GRF responses. Thus, this task was developed with the specific intent of challenging the balance control system in such a way that both its behavioral (GRF measures) and neuromuscular (EEG and EMG measures) underpinnings could be observed during the stable (low difficulty), cautious (medium difficulty), and threatened (high difficulty) stages of balance control. Participants were not informed when the gain changed. To ensure safety, participants always wore a safety harness. In addition, they also wore a physical therapy belt with a spotter standing nearby to prevent any falls. Participants were allowed to practice at gain values of 0, 0.4, 2.0. Data from these gain values were not used for analysis.

Electroencephalography (EEG)

Whole-scalp electroencephalography (EEG) was recorded using 64 active channel EEG electrodes (Brain Products GmbH, Germany; 1000 Hz). The electrodes were placed as per the International 10–20 system for EEG electrode placement. For instance, to locate Cz, four fiducial points namely nasion, inion, left preauricular, and right preauricular were identified. Cz was identified at the midpoint of the nasion-inion arc length and the left-right preauricular arc length. Four of the electrodes were used to record electrooculography (EOG) signals. EEG was recorded at rest and during the balance task. We constrained movement of the EEG cables by placing an elastic mesh on the top of the EEG cap, which aided in minimizing motion artifacts during the balance task on the Neurocom (Goel et al. 2019; Luu et al. 2017). We utilized a modified international 10–20 EEG system where we moved the GND and REF from the default locations (AFz and FCz) to the left and right earlobes, respectively (Luu et al. 2016). We made these modifications because the default channel locations for GND and REF are very close to the fronto-central cortex, a region of interest for this study. The empty spots were filled by moving T7 and T8 electrodes to AFz and FCz locations, respectively.

Electromyography (EMG)

We recorded activity of thigh and lower leg muscles bilaterally using differential surface electrodes from tibialis anterior (TA), gastrocnemius medialis (GM) and soleus (SOL) as distal leg muscle group, and rectus femoris (RF) and biceps femoris (BF) as proximal leg muscle group (1111.11 Hz, gain 1000; 20–450 Hz Delsys Trigno EMG System, Boston, MA). These muscles were selected because of their involvement in the control of vertical posture while dealing with symmetrical perturbations induced in the sagittal plane (Mohapatra et al. 2014; Santos et al. 2010). The placement of electrodes for recording EMG activity was based on recommendations reported in the literature (Mohapatra et al. 2014).

Experimental protocol

Each subject participated in a single session. First, we obtained their clinical measures using BBS and TUG test. Following this step, we prepared them for EEG and EMG measurements. Prior to electrode placement, the skin was prepared by cleaning with isopropyl alcohol pads and by shaving excessive hair, if necessary. Each participant then performed the challenging balance task with varying sensory conditions while standing on the force platform (support surface). We obtained baseline measurements for 60 s during quiet standing with eyes closed before and after the balance task. The dynamic balance task consisted of maintaining an upright stance with eyes closed during a nine sequential 20 s duration balance task conditions.

Data analyses

Behavioral data

BBS (out of 56) and TUG (average of three trials in seconds) were collected before the balance task. The ground reaction force data collected from the force plate were combined to create a center of pressure (COP) time series in medio-lateral (ML) and anterior-posterior (AP) directions for the continuous balance task (Neurocom 2009). Next, we estimated the COM by low pass filtering the COP data (second order Butterworth; f_c = 0.86 Hz) (Goel et al. 2019, 2021) for the AP direction for the continuous balance task as the sway of interest was in the AP direction due to the mechanical configuration for rotation of the support surface. Several studies have confirmed the reliability of this method to estimate COM from COP (Breniere 1996; Caron et al. 1997; Lafond et al. 2004). For the continuous balance task, we computed root mean square (RMS) COP (Prieto et al. 1996) and path length (PL) (Donath et al. 2012) from COP (Forth et al. 2011; Ozdemir et al. 2013, 2018). The RMS and PL of COP represent spatial measures of postural performance. We also calculated COP velocity (RMSCOPv) (Geurts et al. 1993), which is also a reliable measure in the quantification of balance control.

The number of falls induced by rotation of the support surface was also recorded using qualitative observations during testing. A fall was noted when a participant lost his/her balance and required either self-induced stepping and/or support from the overhead harness to prevent them from falling to the floor (e.g., participants applied force through the rope and harness system as evident from the rope and harness going taut). The spotter was instructed to assist the participant only after they had fallen (i.e., to prevent swinging caused by the rope).

EEG pre-processing

The EEG pre-processing used in this study was similar to that described in (Goel et al. 2018). Briefly, raw EEG signals sampled at 1000 Hz were first downsampled to 250 Hz. EOG channels (4 channels) were utilized for adaptive filtering of the EEG signals (60 Channels) through the H infinity algorithm to remove eye artifacts, correct baseline drifts, and remove other shared sources of noise (Kilicarslan et al. 2016). A standardized early-stage EEG processing pipeline (PREP) with default parameters was used to remove artifactual EEG channels and apply a robust common referencing method to increase the signal to noise ratio (Bigdely-Shamlo et al. 2015). PREP is reported to be more robust in detecting noisy channels compared to previously known noisy channel detection method: pop_rejchan from EEGLAB (Bigdely-Shamlo et al. 2015; Delorme and Makeig 2004). The PREP pipeline also replaces artifactual channels with surrogate data that is interpolated from neighboring sensors. This minimizes a bias when performing common average referencing. The signals were then band pass filtered at 0.1–100 Hz using a 4th order Butterworth filter to remove slow drifting noise. Artifact Subspace Reconstruction (ASR) was applied next to detect and denoise any artifactual sections in the EEG data (Mullen et al. 2013). The ASR uses clean EEG data to calibrate noise covariance and uses a standard deviation threshold of 15 with a sliding window (500 ms) based principal component analysis (PCA) to detect noisy sections in the EEG data (Artoni et al. 2017; Chang et al. 2020; Mullen et al. 2013).

Next, to compensate for the rank deficiency in the data, surrogate channels during the PREP pipeline were removed before running the independent component analyses (ICA) since the ICA assumes the number of independent components to be the same as the number of channels provided as the data. Adaptive mixture ICA (AMICA) was used to compute the maximally independent components (ICs) from the data (Palmer et al. 2011). The AMICA is reported to be the best ICA method available to date, and it provides more dipolar and minimized mutual information ICs (Delorme et al. 2012; Leutheuser et al. 2013). Using the constructed boundary elemental model (BEM) explained in the next paragraph and the digitized channel locations, dipole fitting method: DIPFIT in EEGLAB (Delorme et al. 2011; Delorme and Makeig 2004) was used to calculate the equivalent current dipole sources that explain at least 85% (Sipp et al. 2013) of topographic variance obtained from ICA results. The digitized channel locations were first warped onto the constructed head model before calculating the dipole locations. We removed dipoles located outside the individual BEM model as well as those with artifactual components such as muscle-related power spectral density characteristics and motion artifact related high-frequency noise.

The boundary element model (BEM) was created after computing surface meshes and constructing the head model using a standard MRI template. The constructed BEM was then saved and used as one of the arguments in the DIPFIT process. Each IC scalp projection, its equivalent dipole’s location, and its power spectra were then visually inspected and ICs related to non-brain artifacts (e.g., sensor movement, muscle artifact, stimulation artifact) were removed.

Wavelet coherence analysis

Wavelet coherence (WC) was used to investigate the relation in time-frequency space between EEG and EMG signals. It is a method of measuring the cross-correlation between two signals with respect to time and frequency. We computed WC between Cz (EEG) electrode and EMG activity measured from TA, GM, SOL, BF, and RF using signals from the affected side in participants with stroke and from the dominant side in healthy participants. In absence of exact age- and gender-matching between the two groups, it was not appropriate to study the corresponding side in healthy participants. We selected Cz electrode location because it is known to represent the sensorimotor area of the cortex overlying the leg representation (Neuper and Pfurtscheller 2001; Ushiyama et al. 2012; Xu et al. 2023). We used the Morlet wavelet as the mother wavelet, with default parameters set as one octave for 10 scales as it provides good time and frequency localization, and a complex shape that allows for capturing both amplitude and phase information (Grinsted et al. 2004). The localized coherence significance was evaluated using Monte Carlo analysis with 1000 iterations as this provide a way to estimate statistical significance and assess the reliability of the wavelet results. EMG signals were detrended and downsampled to 250 Hz. The final two seconds of the EMG signal was corrupted across several trial conditions and thus this was eliminated from further analysis. WC was computed over the first 18 s of trial (4500 samples) with a frequency resolution of 0.0526 Hz. To understand the dynamic nature of coherence between EEG and EMG, WC was averaged over 2 s non-overlapping trial segment (9 bins). Values outside the cone of influence (COI) were considered highly non-reliable and thus excluded from the analysis (Grinsted et al. 2004). This also led to the exclusion of the first and last bins. Coherence was considered significant (p < 0.05) if it was greater than $\mathrm{Z}$, which is given using the following formula:

$$\mathrm{Z}={1-0.05}^{(1/\mathrm{L}-1)}$$

where $\mathrm{L}$ is the total number of non-overlapping segments, equals to the available number of trials (6 trials; practice trials not included) multiplied by seven (Witham et al. 2011), resulting in a Z value of 0.07. The average coherence values were then converted to Fisher z-statistics (normalized) using arc hyperbolic tangent transformation (Baker and Baker 2003; Mima et al. 1999) prior to performing any statistical analysis. We analyzed corticomuscular coherence in delta (0.1 to 4 Hz), theta (4 to 8 Hz), alpha (8–12 Hz), beta (12–30 Hz) and gamma (30–50 Hz) frequency bands. Although the EMG signals were filtered with a built-in bandpass filter, it is not unusual to experimentally observe significant coherence at lower frequencies (Chen et al. 2018; Grosse and Brown 2003; Van Asseldonk et al. 2014).

Statistical analyses

For behavioral, EEG-EMG coherence data, we performed normality (Shapiro-Wilk) and equal variance Levene’s test to check for violation of assumptions for parametric statistical approach. For BBS and TUG scores, these assumptions were found to be violated (both p values < 0.05) and thus we used a non-parametric approach, Mann-Whitney U test, to compare the stroke group with the healthy control group. The assumptions were validated for the COP, EEG, and EMG data. For RMS COP, RMS COP_v and COP PL, we used restricted maximum likelihood linear models with a between-subject factor as Group (Stroke, Healthy) and within-subject factors such as Task Conditions (Low, Medium, High). Statistical analysis was conducted on coherence values obtained over the 14-sec trial window for each balance task condition. For normalized EEG-EMG coherence measure for frequency bands and bins, we used separate restricted maximum likelihood linear models with between-subject factor as Group (Stroke, Healthy) and within-subject factors such as Task Conditions (Low, Medium, High), Frequency Bands (Delta, Theta, Alpha, Beta, Gamma), and Time Bins (Bin1, Bin2, Bin3, Bin4, Bin5, Bin6, Bin7) for each of the five muscles. For each statistical model, post-hoc analysis was performed for significant results using Bonferroni correction for multiple comparisons.

Previous work including ours (Craig et al. 2017; Goel et al. 2019) have observed large changes in behavior following the first exposure to low, medium and high balance task conditions. The first three practice conditions were not included in the analysis. Also, we found no difference in the COP and coherence values between second and third set of low, medium, and high conditions (separate paired t-test; all p values > 0.05). Therefore, we considered data averaged across gain values of similar level of difficulty for further analysis (e.g., data averaged across two low, two medium, and two high conditions). Prior to this step, we excluded trials where participants fell. There was no significant difference between two groups in the total number of trials eliminated (16% in the stroke group versus 10% in the healthy control group; χ² (df = 1) < 1.2; p > 0.05). One participant in each group had both high balance task condition trials excluded. All other participants in both groups had at least one trial per balance task condition included for statistical comparisons. All values are displayed in the form of mean ± standard error of measurement unless noted otherwise. The significance level α was set at 0.05 (IBM SPSS v29.0).

Results

All participants were able to complete clinical assessments and the balance task. Although stroke participants and healthy controls were not exactly matched for age, we found no age-related difference between the two groups (t₂₂ = 1.81; p = 0.084).

Normalized EEG – EMG coherence

Distal muscle group

For the TA muscle, participants with stroke showed a significant difference in the modulation of normalized EEG-EMG coherence across balance task conditions (significant Group × Conditions interaction: F_2,6790.5 = 9.623, p < 0.001) and frequency bands (significant Group × Bands interaction: F_4,6803.9 = 28.629, p < 0.001) when compared to healthy controls. Pair-wise posthoc comparisons showed that participants with stroke showed significantly smaller coherence than healthy controls during the medium difficulty balance condition in the delta frequency band (t₆₄ = 2.668, p = 0.0097, adjusted α = 0.0167; Figs. 2 and 3). All other pairwise comparisons were non-significant (p > adjusted α for multiple comparisons). We observed a significant modulation in coherence across frequency bands (main effect of Bands: F_4,6803.7 = 4729.3; p < 0.001) and bins (main effect of Bins: F_6,6803.9 = 13.722, p < 0.001). No other main and interaction effects were found (all p values > 0.1).

Fig. 2.

The medium difficulty balance task condition. A. Spectrograms of the EEG signal (Cz) and all five muscles (TA, GM, SOL, BF, RF) showing relative power magnitude across different frequency bands. B. Coherence spectra plots for Cz and each of the five muscles. Data have been averaged across all subjects within a group (Stroke participants and Healthy controls)

Fig. 3.

Smaller cortico-muscular coherence for the TA muscle in stroke. Cortico-muscular coherence (CMC) magnitude between Cz (EEG) and tibialis anterior (TA-EMG) for all five frequency bands and all three task conditions for stroke participants and healthy controls. (A) CMC averaged across the entire 20 s trial. (B) CMC for seven trial bins. The dotted line represents coherence significance threshold Z. Data are means (± SEM) of all subjects. An asterisk indicates significant differences between groups

For the GM muscle, participants with stroke when compared with healthy controls showed a significant difference in the modulation of normalized EEG-EMG coherence across balance task conditions (significant Group × Conditions interaction: F_2,2358.9 = 5.212, p = 0.006), frequency bands (significant Group × Bands interaction: F_4,2358 = 14.334, p < 0.001) and bins (significant Group × Bins interaction: F_6,2358 = 4.222, p < 0.001). Pair-wise posthoc comparisons showed that participants with stroke showed significantly smaller coherence than healthy controls during the medium difficulty condition in the delta frequency band (t₆₄ = 2.580, p = 0.012, adjusted α = 0.0167; Fig. 4). All other pair-wise comparisons were non-significant (p > adjusted α for multiple comparisons). We also found significant differences in coherence across frequency bands (main effect of Bands: F_4,2358 = 4673.6; p < 0.001) and bins (main effect of Bins: F_6,2358 = 15.958, p < 0.001), but not across balance task conditions (no main effect of Conditions: F_2,2358.9 = 0.843; p = 0.430).

Fig. 4.

Smaller cortico-muscular coherence for the GM muscle in stroke. Cortico-muscular coherence (CMC) magnitude between Cz (EEG) and gastrocnemius medialis (GM-EMG) for all five frequency bands and all three task conditions for stroke participants and healthy controls. (A) CMC averaged across the entire 20 s trial. (B) CMC for seven trial bins. The dotted line represents coherence significance threshold Z. Data are means (± SEM) of all subjects. An asterisk indicates significant differences between groups

For the SOL muscle, participants with stroke when compared with healthy controls showed a significant difference in the modulation of normalized EEG-EMG coherence across balance task conditions (significant Group × Conditions interaction: F_2,2367 = 6.21, p = 0.002) and frequency bands (significant Group × Bands interaction: F_4,2358.1 = 3.619, p = 0.006). However, no pair-wise posthoc comparison reached statistical significance (p > adjusted α for multiple comparisons; Fig. 5). Moreover, we found significant difference in coherence across frequency bands (main effect of Bands: F_4,2358.2 = 2494.9; p < 0.001) and bins (main effect of Bins: F_6,2358.2 = 7.759, p < 0.001), but not across balance task conditions (no main effect of Conditions: F_2,2367.1 = 0.604; p = 0.547). All other main and interaction effects were not significant (all p values > 0.4).

Fig. 5.

No group difference in cortico-muscular coherence for the SOL muscle. Cortico-muscular coherence (CMC) magnitude between Cz (EEG) and soleus (SOL-EMG) for all five frequency bands and all three task conditions for stroke participants and healthy controls. (A) CMC averaged across the entire 20 s trial. (B) CMC for seven trial bins. The dotted line represents coherence significance threshold Z. Data are means (± SEM) of all subjects

Proximal muscle group

For the BF muscle, participants with stroke when compared with healthy controls showed a significant difference in the modulation of normalized EEG-EMG coherence across balance task conditions (significant Group × Conditions interaction: F_2,6717.6 = 9.038, p < 0.001) and frequency bands (significant Group × Bands interaction: F_4,6714.7 =14.931, p < 0.001). However, all pair-wise comparisons failed to reach statistical significance (p > adjusted α for multiple comparisons; Fig. 6). We found modulation in coherence across balance task conditions (main effect of Conditions: F_2,6717.6 = 11.854; p < 0.001), across frequency bands (main effect of Bands: F_4,6714.7 = 4710.6; p < 0.001), and bins (main effect of Bins: F_6,6714.7 = 10.639, p < 0.001). No other interaction or main effects were significant (all p values > 0.9).

Fig. 6.

No group difference in cortico-muscular coherence for the BF muscle. Cortico-muscular coherence (CMC) magnitude between Cz (EEG) and biceps femoris (BF-EMG) for all five frequency bands and all three task conditions for stroke participants and healthy controls. (A) CMC averaged across the entire 20 s trial. (B) CMC for seven trial bins. The dotted line represents coherence significance threshold Z. Data are means (± SEM) of all subjects

For the RF muscle, participants with stroke failed to show a significant difference in the modulation of normalized EEG-EMG coherence across balance task conditions (no significant Group × Conditions interaction: F_2,6808.03 = 0.907, p = 0.404 or main effect of Group: F_1,66902408.6 = 0.122, p = 0.727; Fig. 7) when compared with healthy controls. We found significant modulation in coherence with changes across balance task conditions (main effect of Conditions: F_2,6808.03 = 4.602; p = 0.01), across frequency bands (main effect of Bands: F_4,6805.2 = 5105.6; p < 0.001), and bins (main effect of Bins: F_6,6805.2 = 18.1, p < 0.001). Participants with stroke when compared with healthy controls showed modulation in normalized EEG-EMG coherence across frequency bands (significant Group × Bands interaction: F_4,6805.2 =8.796, p < 0.001), but not across bins (no significant Group × Bins interaction: F_6,6805.2 =1.685, p = 0.120; no main effect of Group). None of the pair-wise posthoc comparisons reached statistical significance (p > adjusted α for multiple comparisons).

Fig. 7.

No group difference in cortico-muscular coherence for the RF muscle. Cortico-muscular coherence (CMC) magnitude between Cz (EEG) and rectus femoris (RF-EMG) for all five frequency bands and all three task conditions for stroke participants and healthy controls. (A) CMC averaged across the entire 20 s trial. (B) CMC for seven trial bins. The dotted line represents coherence significance threshold Z. Data are means (± SEM) of all subjects

Balance and mobility performance

Performance on clinical tests of balance and mobility

As expected, participants with stroke showed poor scores on clinical tests of balance and mobility when compared to healthy controls. Stroke participants (42.14 ± 6.1; mean ± SD) when compared to healthy controls (55.2 ± 0.92) showed lower scores on the Berg Balance Scale (Fig. 8A reports median and quartiles for non-parametric statistical comparison; Mann Whitney U-test; U = 0.5, Z = − 4.102; p < 0.001). Consistent with this finding, participants with stroke (23.57 ± 13.8 s) when compared with healthy controls (8.52 ± 1.13 s) took longer to complete the Timed Up and Go test (Fig. 8B reports median and quartiles for non-parametric statistical comparison; Mann Whitney U-test; U = 2.0, Z = − 3.983; p < 0.001).

Fig. 8.

Behavioral variables. A. Berg Balance Score (BBS) shown as median and quartiles. B. Timed Up and Go (TUG) shown as median and quartiles. C. RMS COP, D. RMS COPv, and E. COP Path length (PL) during the continuous balance task. COP data are means (± SEM) of all subjects. Asterisks indicate significant differences between groups

Performance on the continuous balance task

We found an increase in RMS COP with an increasing difficulty of the balance task for both groups (Fig. 8C; main effect of Conditions: F_2,42.1 = 13.377, p < 0.001). However, this increase in RMS COP was similar for both groups (no significant Group × Conditions interaction: F_2,42.1 = 1.048, p = 0.360; no main effect of Group: F_1,22.01 = 2.84, p = 0.106).

Similarly, we found an increase in RMS COP_v with an increasing difficulty of the balance task for both groups (Fig. 8D; main effect of Conditions: F_2,42.695 = 77.4188, p < 0.001). However, this increase in RMS COP_v was similar for both groups (no significant Group × Conditions interaction: F_2,42.695 = 0.462, p = 0.633; no main effect of Group: F_1,22.335 = 0.903, p = 0.352).

Although increasing the gain of support surface increased COP PL (main effect of Conditions: F_2,42.777 = 70.164, p < 0.001), we found that this increase in PL to be similar for both groups (Fig. 8E; No Group × Conditions interaction: F_2,42.777 = 0.401, p = 0.672 and no main effect of Group: F_1,22.365 = 2.094, p = 0.162).

Discussion

The novel finding from this study is that participants with stroke demonstrated smaller normalized corticomuscular coherence in the delta frequency band when compared to healthy controls during a balance task. Importantly, this difference was primarily observed during medium difficulty balance task condition but not during quiet standing and high difficulty task condition. Finally, these balance context-dependent findings were seen for most of the distal leg muscles but not for proximal leg muscles. Below we discuss potential mechanisms associated with these findings.

Smaller corticomuscular coherence in the delta frequency band in stroke.

Both participants with stroke and healthy older adults showed significant delta band coherence between Cz and lower extremity muscles (coherence values > Z) during the continuous balance task. The later finding is consistent with our earlier work in older adults (Ozdemir et al. 2018), which found a significant coherence in the delta frequency band following unexpected translation of the support surface. It is unlikely that the coherence in delta frequency range is due to the presence of baseline drift common to both EEG and EMG signals. First, the EEG preprocessing pipeline used in this study included the H-infinity filter that is known to robustly correct baseline drifts (Kilicarslan et al. 2016). Second, despite the presence of EMG power in lower frequency bands in all muscles (see Fig. 2), the difference in the EEG-EMG coherence in delta frequency band between stroke participants and healthy controls was found in distal muscles (except SOL) but not in proximal muscles. Therefore, we argue that the finding of significant delta band coherence and the muscle-specific group effect indicate physiological underpinnings.

Novembre and colleagues have noted coherence in the delta frequency range following a delivery of unexpected auditory or electrical stimulus to the opposite hand during performance of an isometric grip force production task (Novembre et al. 2018, 2019). The presence of coherence in the delta frequency range may be related to the detection of a salient stimulus during an ongoing voluntary task (Novembre et al. 2018, 2019; Valentini et al. 2015). Salient stimuli have functional implications as they trigger reactive mechanisms for movement control (Menon and Uddin 2010). In Novembre’s studies, the delta band coherence co-occurred with negative and positive EEG cortical potentials following an unexpected stimulus known to be associated with error monitoring and correction (Novembre et al. 2018, 2019). Similar positive-negative EEG potentials in the fronto-central brain region have been observed following an unexpected perturbation of the standing balance as well as during continuous balance tasks similar to that used in this study (Dietz et al. 1984, 1985; Goel et al. 2018, 2019, 2021; Maki and McIlroy 2007; Marlin et al. 2014; Mochizuki et al. 2010; Ozdemir et al. 2018; Varghese et al. 2014, 2019). The current analysis does not provide information on the relative timing of delta band coherence and EEG potentials. We speculate that the presence of delta band coherence during the balance task might represent detection of unexpected events in the form of support surface movement requiring a reactive control of lower limb muscles for balance maintenance. Future work will focus on understanding the significance of delta band coherence during challenging balance tasks.

We observed significant differences in the modulation in delta band coherence across balance task conditions between stroke participants and healthy controls. Mainly, participants with stroke showed significantly smaller magnitude of normalized corticomuscular coherence when performing the task under the medium difficulty condition. This finding suggests impaired coordination and connectivity between sensorimotor cortical region and muscles involved in balance control. And the impact of this impairment is likely to be observed during the challenging balance task where the gain of the support surface was altered resulting in continuous sensory manipulation of the balance control loop (Rasman et al. 2018). Compromised balance context-dependent detection of sensory inputs for balance modulation might have led to significantly smaller delta band coherence in participants with stroke in the medium difficulty condition. However, similar group differences were not observed for the high difficulty balance condition. In the high condition, the gain of sway-referenced support surface was significantly manipulated. One may expect higher corticomuscular coherence when performing a balance task under the high difficulty condition in the healthy control group, with stroke participants facing even greater difficulty resulting in a larger group difference. However, healthy individuals did not show higher coherence magnitude for the high difficulty balance task condition in comparison to the medium difficulty condition (see. Figures 3A and 4A). It is possible that the distortions in somatosensory inputs were too large to react to by individuals in both groups resulting in similar coherence.

Stroke participants showed smaller delta band coherence for distal leg muscles but not for proximal muscles when compared with healthy controls. There is a suggestion for different organization of pyramidal pathways to distal and proximal muscle groups accounting for differences between the two muscle groups (Grosse et al. 2002). During upper extremity tasks, studies have found normal coherence for proximal elbow flexors but smaller coherence for distal wrist muscles on the affected when compared to non-affected side in chronic stroke participants and to the dominant side in healthy controls (Mima et al. 2001; Zhou et al. 2021). It is likely that the descending connections to proximal and distal leg muscle groups might have different recovery curves following stroke (Zhou et al. 2021). That is, the stroke-affected proximal muscle group showed faster recovery to an extent that resulted in indifferent coherence from healthy controls in this study. Motor recovery following stroke has been associated with an increase in corticomuscular coherence (Krauth et al. 2019; Zheng et al. 2018). Alternatively, a lack of group effect might highlight compensatory involvement of the proximal muscle group due to repair and remodeling processes affecting the neural networks for balance maintenance following stroke (Jones 2017). However, the present findings are unable to parse out the distinction between recovery and compensatory mechanisms (Dayan and Cohen 2011; Takeuchi and Izumi 2012). The proximal balance strategy might have assisted stroke participants to encounter more challenging high difficulty balance condition (McCollum et al. 1996; Runge et al. 1999). Overall, these findings emphasize that the corticomuscular coherence measure may be used to track the functional recovery or the development of compensatory strategies to regain balance function following a stroke (Grefkes and Ward 2014; Hara 2015; Mima et al. 2001; Olafson et al. 2021; Swayne et al. 2008; von Carlowitz-Ghori et al. 2014; Wist et al. 2016).

Within the distal muscle group, the group effect found in TA and GM muscles, but not in the SOL muscle might be attributed to nonhomogeneous influence of stroke on corticospinal projections to muscles with different muscle composition. Non-invasive brain stimulation studies have suggested distinct corticospinal projections to lower leg muscles based on differences in the ability to facilitate TA versus SOL muscles using transcranial anodal and magnetic stimulation techniques (Benecke et al. 1988; Brouwer and Ashby 1990; Cowan et al. 1986). TA as well as GM muscles are predominantly composed of fast-twitch type-II fibers (Abruzzo et al. 2013; Fujiwara et al. 2010) and are suited for dynamic activities (Arunganesh et al. 2021; Fujiwara et al. 2010) similar to medium and high difficulty balance conditions in our study. In contrast, the SOL muscle is rich in slow-twitch type-I fibers (Abruzzo et al. 2013), ideal for static postural tasks (Fujiwara et al. 2010) such as the low difficulty balance condition in our study. These muscles demonstrate different discharge frequencies of motor units during muscle contractions (Ushiyama et al. 2012). There is conflicting evidence regarding whether stroke-induced muscle atrophy predominantly targets type-I (Edstrom 1969; von Walden et al. 2012) or type-II muscle fibers (Frontera et al. 1997). Our findings in the SOL muscle (composed of type-I fibers) reflect these reports where one-third of stroke participants showed coherence magnitudes similar to healthy controls while the remaining participants showed lower magnitudes (see Fig. 5A, the medium difficulty condition), resulting in an overall lack of group difference for this muscle.

Similar coherence magnitude between stroke and healthy controls at higher frequency bands

We found significant coherence in the theta frequency band, but it was not different between participants with stroke and healthy controls during the continuous balance task. The theta band coherence has been previously shown while standing and walking in healthy adults (Peterson and Ferris 2019). These authors found that the theta band coherence increased with a pull perturbation during standing and suggested that coherence in the theta frequency band might be associated with increased muscle activity during perturbation. In our study, both participants with stroke and healthy controls might have used similar levels of muscle activity during performance of the continuous balance task resulting in no difference in the theta coherence band.

Earlier work has shown that when subjects are instructed to keep their eyes closed versus open during a quiet standing task, the coherence in the alpha frequency band is significantly smaller (Vecchio et al. 2008). The alpha activity has been associated with the visuo-spatial attentional aspects of balance control (Edwards et al. 2021) and possibly reflect processing/integrating visual and somatosensory information (Vecchio et al. 2008). In our study, all subjects were required to close their eyes during the continuous balance task that included two trials of quiet standing (low difficulty condition). We found non-significant coherence in the alpha frequency band in both groups (i.e., coherence values < Z). Our findings extend previous work by showing the absence of alpha band coherence during both quiet standing and standing on a sway-referenced platform with varying gain with eyes closed.

Although we found significant coherence magnitude in the beta frequency band during the continuous balance task, it was not different between participants with stroke and healthy controls. The beta band coherence is usually observed during tasks involving sub-maximal or isometric muscle contraction (Gwin and Ferris 2012; Ushiyama et al. 2012) and has been previously reported during standing tasks (Jacobs et al. 2015; Stokkermans et al. 2022). A recent study that instructed subjects to perform isolated ankle dorsiflexion found lower beta band coherence for lower leg agonist (TA) as well as antagonist (GM) muscles in participants with stroke as compared to healthy controls (Xu et al. 2023). The beta band coherence is believed to represent sensorimotor integration resulting from the interaction between motor cortex and peripheral structures such as spinal cord, muscles, and afferent nerves (Baker 2007; Grosse et al. 2002). A lesion that affects the corticospinal connections as it happens during stroke would influence the beta band coherence (Mima et al. 1999, 2001). Guo and colleagues reported findings like our study where participants with chronic stroke showed beta band coherence magnitude similar to healthy controls during performance of an upper limb isometric force production task (Guo et al. 2020). It is possible that the beta band might not be sensitive to detect alterations in the corticospinal function in participants with chronic stroke, who have regained independence in performing daily activities such as standing and walking.

Similarly, we found no difference in the low gamma band coherence between participants with stroke and healthy controls during the continuous balance task. The coherence in gamma frequency band has been observed during dynamic muscle contractions involving lower limbs (Gwin and Ferris 2012) and while standing on a moveable support surface (Stokkermans et al. 2022). The gamma band coherence is related to muscle dynamics, i.e., scales with the force produced by a muscle (Brown and Marsden 1998; Mima et al. 1999). Although the continuous balance task that subjects performed involved both static (quiet standing or low difficulty) and dynamic movements (medium and high difficulty conditions), it failed to cause a modulation in the low gamma band coherence in both groups with the task difficulty. It is possible that the participants with stroke in our study were in the chronic stage of recovery, and had achieved a level of recovery such that they were able to produce dynamic muscle activities similar to healthy controls.

Poor balance stability in participants with stroke

Although participants with stroke when compared with healthy controls showed poor scores on balance and mobility on clinical tests such as BB scale and TUG, we failed to observe group difference across COP measures of balance performance as assessed on the computerized balance platform. This finding is consistent with previous studies that found poor correlation between BBS score and COP measures (Fujimoto et al. 2014; Mansfield et al. 2012) and might be due to larger inter-subject variability in the balance performance on the computerized task, possibly due to different strategies used by individuals for balance maintenance (see Fig. 8). All COP measures were sensitive to the balance task difficulty conditions; however, they were unable to differentiate different strategies used by stroke participants and healthy adults. The power spectral-density analysis that parses out a relative contribution of different sensory systems to maintain an upright stance may provide a better resolution to detect alterations in balance control due to aging and stroke.

Limitations

Considering the small sample size of this study, it is important to confirm these findings in a larger cohort. The built-in narrow band-pass filter of the EMG amplifier might have introduced artifacts, distortions, or spurious oscillations in the EMG signal in the lower frequency bands (De Cheveigné and Nelken 2019; Widmann et al. 2015), thus impacting the EEG-EMG coherence analysis. Future studies are required to replicate our findings of significant delta band coherence and muscle-group specific effect in the delta band EEG-EMG coherence. Our analysis for the corticomuscular coherence focused on the EEG sensor level activity which limits the spatial resolution to identify specific brain regions engaged during the balance task. The application of advanced EEG source estimation approach can help localize the brain areas engaged during the continuous balance task (Goel et al. 2018, 2019; Mouraux and Iannetti 2009).

Summary and conclusion

To our knowledge, this is the first study investigating cortico-muscular coherence in relation to balance control among participants with stroke versus healthy controls in five frequency bands across distal and proximal leg muscles. Our finding of smaller delta frequency band corticomuscular coherence during the challenging continuous balance task in participants with stroke when compared with healthy controls might suggest impairment in context-dependent sensorimotor processes for the detection of somatosensory modulation to control the leg muscles for balance maintenance following stroke. Moreover, the observation of abnormal coherence for distal but not for proximal leg muscles on the affected side might suggest differences in the (re)organization of the descending connections across two muscles groups for balance control. The findings from these studies may provide insights into the neural mechanisms underlying poor balance control that may facilitate the assessment of balance behavior during a course of an intervention for balance rehabilitation post-stroke.

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.