Motor unit distribution and recruitment in spastic and non-spastic bilateral biceps brachii muscles of chronic stroke survivors

Abstract

Objective-

This study aims to characterize the motor units distribution and recruitment pattern in the spastic and non-spastic bilateral biceps brachii muscles (BBMs) of chronic stroke survivors.

Approach-

High-density surface electromyography (HD-sEMG) signals were collected from both spastic and non-spastic BBMs of fourteen chronic stroke subjects during isometric elbow flexion at 10%, 30%, 50% and 100% maximal voluntary contractions (MVCs). By combining HD-sEMG decomposition and bioelectrical source imaging, motor unit innervation zones (MUIZs) of the decomposed motor units were first localized in the 3D space of spastic and non-spastic BBMs. The motor unit depth defined as the distance between the localized MUIZ and its normal projection on the skin surface was then normalized to the arm radius of each subject and averaged at given contraction level. The averaged MU depth at different contraction levels on a specific arm side (intra-side) and the bilateral depths under a specific contraction level (inter-side) were compared.

Main Results-

The average depth of decomposed MUs increased with the contraction force and significant differences observed between 10% vs 50% (p < 0.0001), 10% vs 100% (p < 0.0001) and 30% vs 100% MVC (p = 0.0017) on the non-spastic side, indicating that larger MUs with higher recruitment threshold locate in deeper muscle regions. In contrast, no force-related difference in MU depth was observed on the spastic side, suggesting a disruption of orderly recruitment of MUs with increase of force level, or the MU denervation and the subsequent collateral reinnervation secondary to upper motor neuron lesions. Inter-side comparison demonstrated significant MU depth difference at 10% (p=0.0048) and 100% force effort (p=0.0026).

Significance-

This study represents the first effort to non-invasively characterize the MU distribution inside spastic and non-spastic bilateral BBM of chronic stroke patients by combining HD-sEMG recording, EMG signal decomposition and bioelectrical source imaging. The findings of this study advances our understanding regarding the neurophysiology of human muscles and the neuromuscular alterations following stroke. It may also offer important MU depth information for botulinum toxin (BoNT) injection in clinical post-stroke spasticity management.

1. Introduction

Information pertaining to the motor unit distribution inside of human muscles is critically important in understanding the neurophysiology of muscles and tracking the progress and treating of motoneuron diseases and neuromuscular disorders. Up to now, histochemistry, intramuscular and surface electromyography (EMG) are the primary techniques that have been employed to study the motor unit distribution inside of muscles.

The limitation of histochemical staining lies in that it can only be applied to cadaveric human muscles or animal muscles and controversy exists between findings [1–3]. Intramuscular EMG (iEMG) enables the accurate estimation of motor unit (MU) size and localization, owing to its high spatial selectivity[4–5]. However, iEMG is invasive and spatially constrained to a very small number of MUs that are in close proximity of intramuscular needle tip [6], which hinders us from studying the distribution of a large number of MUs.

Recently, with advances in surface EMG (sEMG) signal acquisition and processing techniques, investigators attempted to apply this technique to non-invasively study the MU distribution in vivo. Although sEMG offers a more comprehensive sampling of the entire MU pool compared to the iEMG, the signal itself is a complicated function of the placement of sEMG electrodes and the depth and size of underlying motor units. Therefore, it is difficult, if not impossible, to characterize the MU distribution without taking into consideration of these factors. An indirect or elusive approach of comparing the simulated and experimental sEMG signal during either transcutaneous electrical stimulation or voluntary contractions was adopted to study the MU depth distribution with respect to MU size [7–9]. Their findings suggest that simulations matched the experimental results only when larger MUs were preferably located deeper, conflicting to the iEMG results. Clearly, despite the tremendous effort made, MU distribution inside of human muscle remains largely inconclusive and an unmet need exists.

By taking advantage of the capability of the newly developed three-dimensional innervation zone imaging (3DIZI) technique in characterizing the motor unit innervation zone (MUIZ) location in the 3D space of target muscle [10–15], this study aims to study the MU distribution inside the spastic and non-spastic bilateral biceps brachii muscles (BBMs) of chronic stroke survivors based on the well documented and commonly accepted “size principle” [9, 16–19], specifying that small MUs are recruited first and larger MUs get involved with the increase of contraction force. BBM was chosen because it is functionally crucial in upper limb and is one of the most often affected muscles in post-stroke patients [20]. Characterizing motor unit distribution and recruitment pattern inside of spastic and non-spastic human muscle can not only dramatically enhance our understanding regarding the neurophysiology of healthy muscles but also allow for assessing and documenting disturbed recruitment pattern in spastic muscle. Furthermore, the finding of the current study will offer important depth information for guiding precise botulinum toxin (BoNT) injection in clinical spasticity management. The high rate of synaptic activity (i.e. high firing rate) of the small motor units can accelerate the uptake of BoNT and further potentiate its paralytic effect in clinical post-stroke spasticity management, as reported previously [21].

2. Methods

2.1. Participants and consent

Fourteen hemiparetic stroke survivors (five females, mean age 53.1 ± 11.4 years) were recruited from TIRR Outpatient Spasticity Management Clinic at TIRR Memorial Herman Hospital. The inclusion criteria included: diagnosed with a hemorrhagic or ischemic stroke with at least six months duration, with an elbow flexors Modified Ashworth Scale (MAS) between 1+ to 3 and capable of following instructions and giving informed consent. The exclusion criteria included: occurrence for multiple strokes, presence or history of other central/peripheral nervous system disorders, direct muscle injuries or fixed contracture and/or bone deformity at the elbow and old age (> 75 years) to avoid confounding factors that may affect the neuromuscular function and IZ distribution. The average duration of the chronic stroke is 61.6 ± 44.1 months (six ischemic and eight hemorrhagic). The experiment was approved by the University of Houston and University of Texas Health Science Center-Houston Institutional Review Board.

2.2. Experiment protocol

Subjects were seated with the target arm secured firmly in two adjustable metal plates, as presented by figure1. The skin overlaying the BBM was prepared with Nuprep Skin Prep Gel (Bio-Medical Instruments, MI, USA) and electrode prep pads (PDI Health, NY, USA) to lower contact impedance. Two high-density sEMG (HD-sEMG, 8 by 8 sensors, with a diameter of 4.5 mm and an inter-electrode distance of 8.5 mm) electrode arrays (TMSi, Enschede, The Netherlands) were placed over the muscle belly of BBM with the midline of the combined electrode grids longitudinally aligned with the midline of biceps, defined from the biciptal groove to the biceps tendon insertion. The reference electrode was placed on the lateral epicondyle of the humerus and a ground electrode was attached to the wrist of the contralateral arm using a fully soaked Velcro wrist band (TMSi, Enschede, the Netherlands). Three maximal voluntary contractions (MVCs) were performed and the highest force value observed was used as the 100% MVC force. Then three 10-second isometric contractions were performed at 10%, 30%, 50% and 100% MVC. The contraction can be compromised to 5 seconds considering the difficulty of maintaining steady elbow flexions at higher force levels with the affected arm. Visual feedback of the contractile force was provided via a screen monitor and subjects were required to maintain the force as steady as possible. A 1–2 min rest interval was given between trials to avoid muscle fatigue. HD-sEMG signals were recorded via a Refa-136 amplifier (TMSi, Enschede, the Netherlands) at a sampling rate of 2,048 Hz with a 24-bit resolution.

Figure 1.

Experimental setup.

2.3. Signal processing

HD-sEMG signals were examined offline and the corrupted channels were excluded from further analysis. Then the raw HD-sEMG signals were band-pass filtered (10–500 Hz) and notch filtered (60 Hz, stopband 59.5–60.5 Hz) using second order, zero-phase Butterworth filters. The K-means-clustering-modified convolution kernel compensation (KmCKC) method [22] developed by our group was adopted to decompose the HD-sEMG signals into constituent pulse trains. Briefly, time instant n₀ with significant muscle activity was selected from the global activity index and V_n0 reflecting the activity of MUs that fire at the time instant of n₀, was calculated as specified in a previous study [23]. A time instant with strongest amplitude in V_n0 but different from n₀ was selected as n₁ and V_n1 was calculated accordingly. Ten time instants with the highest amplitude were selected from the element-wise product of V_n0 and V_n1 and classified into two groups using a K-means clustering method. The group with the larger number of firing instants was used to construct the initial innervation pulse train by using a linear minimum mean square error method. Finally, a modified multi-step iterative convolution kernel compensation (CKC) method was employed to update the initial pulse train until desirable decomposition accuracy was achieved. The monopolar MUAP was then constructed using a spike-triggered averaging method for each channel [18] and the largest negative peak among all channels is considered as the amplitude of the decomposed MU. The bipolar MUAP mapping was obtained by subtracting the consecutive monopolar MUAP with its proximal neighbor in the muscle fiber direction. Given the fact that MUAP generates from the neuromuscular junction (indicated as IZ) and propagates in two opposite directions towards the fiber endings [24], the surface location of MUIZ can be visually identified from the bipolar MUAP mapping by observing either the reversal of signal polarity or minimum amplitude [11, 20]. The 128-channel sEMG measurement corresponding to the time instant that the electrode closest to the identified MUIZ reached its negative peak were extracted to characterize the MUIZ location in the 3D space of BBM by using the 3DIZI approach.

2.4. Modeling

A template upper arm model, consisting of the BBM, triceps, compact and cancellous bone, fat and skin, was constructed from a general MR image data set by following the same procedure as our previous work [25]. Then this template model was scaled based on the averaged ratio between the tissue thickness of template model and that of the participants acquired with a portable B-mode ultrasound scanner (M Turbo, SonoSite, Bothell, WA) at the muscle belly. The adjusted model then was meshed into finite element models with 234,103 tetrahedral elements and 40,865 nodes, using Abaqus 6.12 (SIMULIA, Providence, RI). Isotropic conductivity values of 4.55×10⁻⁴, 0.0379, 0.02 and 0.075 S/m were respectively assigned to the skin, fat tissue, compact and cancellous bones and anisotropic conductivity values of 0.2455 and 1.23 S/m (anisotropy ratio of 5) were assigned to muscles in transverse and muscle fiber direction [26]. Totally, 34,619 current dipoles were evenly distributed in the 3D space of BBM with a spatial resolution of 2 mm × 2 mm × 2 mm.

2.5. Localization of MU IZ using 3DIZI

The 3DIZI algorithm has been described in previous studies [10–13, 15]. In brief, the relationship between the dipole strength j and the sEMG measurement ∅ can be described by a linear equation,

$$\varnothing =G\cdot J+n$$ (1)

where G and n denotes the corresponding lead field matrix and measurement noise, respectively. The weighted minimum norm solution [11, 13, 15] to the above ill-posed problem of 3DIZI can be expressed as:

$$\widehat{J}=R{\left(W^{T}W\right)}^{-1}{G}^T{(G(R{\left(W^{T}W\right)}^{-1})G^T+\lambda I)}^{-1}\varnothing$$ (2)

where R is source covariance matrix constructed based on the surface MUIZ location; W is a weighted matrix which compensates for the undesired depth dependence; λ is the regularization parameter determined by the L-curve method [27]. For more details about the construction of the aforementioned matrices, please refer to [11]. The strength of reconstructed current dipoles was normalized to their maximal absolute value and only those with normalized strength higher than 0.3 were supposed to be active. The geometric center of active dipoles was considered as the MUIZ location of corresponding MU. Then the MU depth was defined as the distance between this geometric center to its normal projection on the skin surface. As the arm dimensions vary between subjects, the calculated MU depth was respectively normalized to the arm radius calculated from the measured arm circumference for each subject.

2.6. Determination of MU depth distribution with respect to MU size

Based on the commonly accepted size principle [9, 16–19], the averaged MU depth at four contraction levels were calculated in this study to explore the MU depth distribution with respect to MU size. Besides, previous finding indicated that depth of MU recruited in healthy BBM during an contraction level up to 10% MVC can be empirically estimated as the 20% of full width half maximum (FWHM), defined as the muscle-fiber-transverse-direction distance over the skin surface where the absolute value of MUAP negative peak is higher than 50% of the maximal absolute value [28]. For the purpose of comparison and validation, the FWHM of each decomposed MU was calculated one inter-electrode distance away from the identified MUIZ towards the muscle belly and normalized to arm circumference of each subject. Cubic spline extrapolation and the assumption that sEMG amplitude is left-right symmetrical about the IZ column were employed to estimate the FWHM when the electrode grids are not wide enough for the determination of FWHM.

2.7. Statistical analysis

The result of Jarque-Bera test showed the data were not normally distributed. Therefore, the non-parametric Kruskal-Wallis test was adopted to test the existence/absence of any statistically significant differences of MU depth and FWHM between four contraction levels on each side of BBM. Statistical significance was accepted at p<0.05. Mann-Whitney U test was used to perform the post-hoc comparisons, as well as to determine the significance of MU depth and FWHM difference between spastic and non-spastic sides at a given contraction level. Bonferroni correction was used for multiple comparisons, resulting in p < 0.0083 and p<0.0017 as significant and highly significant threshold, respectively. Data are reported as mean±SD within the text and in the figures.

3. Results

Fourteen non-spastic and ten spastic BBMs were investigated, with four subjects incapable of voluntary elbow flexion on the spastic side. Initially, 537 and 822 MUs were respectively decomposed from the spastic and non-spastic side BBM at four contraction levels. By excluding MUs with pulse-to-noise-ratio lower than 30 dB [29], the coefficient of variance of the inter spike interval greater than 0.3 [30–31], one-sided MUAP propagation and symmetrical distribution of MUAPs without propagation [14], 411 and 764 MUs were yielded from the spastic and non-spastic BBM, respectively. The total MU number yielded at 10%, 30%, 50% and 100% MVC for the spastic side was 97, 127, 105 and 82, respectively, and 169, 210, 197 and188 respectively for the non-spastic side, corresponding to an average number of 10±3, 12±5, 10±4 and 9±3 at 10%, 30%, 50% and 100% MVC for the spastic side, respectively, and 11±3, 14±4, 14±4 and 12±3 respectively for the non-spastic side.

A representative example of superficial and deep MUs in the non-spastic and spastic BBMs, along with the corresponding high-density monopolar MUAP, bipolar MUAP, FWHM and 3DIZI imaging results, were visualized in figure 2 and 3, respectively. In the non-spastic BBM, the normalized FWHM of superficial MU recruited at 10% MVC is 17.33% (5.72 cm) corresponding to a normalized depth of 30.26% (1.59 cm); and that for a deep MU recruited at 50% MVC is 36.39% (12.01 cm) corresponding to a depth of 49.86% (2.62 cm). For the spastic side, the normalized FWHM of superficial MU recruited at 10% MVC is 12.34% (3.95 cm) corresponding to a depth of 25.12% (1.28 cm) and that for a deep MU recruited at 30% MVC is 34.58% (11.06 cm) corresponding to a depth of 48.67% (2.48 cm).

Figure 2.

Representative results of superficial and deep MUs in the non-spastic BBM of subject #1. (a1) and (a2) show the monopolar MUAP maps of two MUs, in which the red MUAP represents the IZ channel and the red rectangular represents the row that was used to determine the FWHM; (b1) and (b2) present the bipolar MUAP maps of two MUs, in which the red line indicates the IZ location; (c1) and (c2) show the calculation of FWHM; (d1) and (d2) present the 3DIZI results of two MUs.

Figure 3.

Representative results of superficial and deep MUs in the spastic BBM of subject #1. (a1) and (a2) show the monopolar MUAP maps of two MUs, in which the red MUAP represents the IZ channel and the red rectangular represents the row that was used to determine the FWHM; (b1) and (b2) present the bipolar MUAP maps of two MUs, in which the red line indicates the IZ location; (c1) and (c2) show the calculation of FWHM; (d1) and (d2) present the 3DIZI results of two MUs.

The MU depth distribution was presented in figure 4. Intra-side comparison shows no significant depth difference (p=0.0829) for the spastic side, yet significant depth difference exists across the four contraction levels on the non-spastic side (p < 0.0001). MU depth increases with MVC levels and the depth at 100% MVC is significantly deeper than that at 10% (p < 0.0001) and 30% (p = 0.0017) MVC levels, and that at 10% MVC is also significantly different from that of 50% (p < 0.0001) MVC. As for the inter-side comparison, the spastic side MUs recruited at low contraction levels of 10% (p=0.0048) and 30% (p=0.3577) MVC were averagely deeper than those of non-spastic side and more superficial at high contraction levels of 50% (p=0.3028) and 100% (p=0.0026) MVC. The average MUAP amplitude defined as largest negative peak amplitude among all channels is 87.17±58.18, 152.85±84.93, 324.45±144.81 and 491.52±469.79 uV at 10%, 30%, 50% and 100% MVC respectively for the non-spastic side and 109.98±80.07, 144.41±92.56, 172.42±124.88 and 205.04±118.08 uV respectively for the spastic side.

Figure 4.

Normalized MU depths at four contraction levels. *, p < 0.005; **, p < 0.001.

The results of FWHM, as a reflection of MU depth [28], were shown in figure 5. Intra-side comparison reveals no significant FWHM difference for the spastic side (p=0.1004). The non-spastic side presents a significant, force-dependent FWHM difference (p < 0.0001), in which the FWHM increases with contraction level and the FWHM at 100% MVC is significantly different from those at 10% (p < 0.0001), 30% (p < 0.0001) and 50% (p=0.0015) MVC levels. For the inter-side comparison, the spastic side average FWHMs at 10% (p=0.1734), 30% (p=0.6496) and 50% (p=0.5962) MVC were larger than those of non-spastic side, yet statistical significance was only observed at 100% MVC level (p= 0.0018). For the purpose of comparison and validation, the relationship between the normalized MU depth and FWHM was evaluated on 68 non-spastic MUs with a FWHM smaller than 7 inter-electrode distance (with more accurate FWHM result) at contraction level of 10% MVC. A slope of 1.1537 and R² of 0.5768 were acquired by a linear regression analysis, as presented in figure 6.

Figure 5.

Normalized FWHMs at four contraction levels. *, p < 0.005; **, p < 0.001.

Figure 6.

Linear regression between the normalized MU depth and FWHM for 68 motor units recruited at 10% MVC with FWHM smaller than 7 inter-electrode distance (p<0.01).

4. Discussion

This study presents the first effort to non-invasively characterize the MU distribution inside spastic and non-spastic bilateral BBMs of chronic stroke patients by combining the cutting-edge techniques of HD-sEMG recording, EMG signal decomposition and bioelectrical source imaging.

Due to the technical limitation for noninvasive characterization of MU depth, determining the distribution of different size of MUs remains challenging; and conflicting findings exist between, even within, histochemical, iEMG and sEMG studies. By combining the autopsy with the histochemical techniques, previous studies reported that the fast-twitch type II fibers, usually innervated by large MUs, are predominant at the surface of the muscle, while slow-twitch type I fibers, usually innervated by small MUs, are deep [1–3]. While some other findings suggested that the fiber types randomly distribute inside of a muscle and the non-randomness can be as an indication of a pathological denervation and reinnervation process [32–34]. The research based on iEMG claimed that large MUs are superficial and small ones are deep[4]. Nevertheless, results acquired by sEMG are conflicting with those of iEMG [7–9]. Our findings indicate that the average MU depth increases with contraction force in the non-spastic side, suggesting that large MUs locate deeper according to the “size principle”. Therefore, our result is in consistent with previous sEMG findings.

Across four contraction levels, average MU depths at higher force levels (50% and 100% MVC) is significantly different from depths at lower force levels (10% and 30% MVC) on the non-spastic side. When the force increased from 50% to 100% MVC, the average MU depth increased, though not statistically significant. Therefore, the max recruitment threshold at which the last MU is recruited is higher than the reported 50% MVC by Jensen et al. [35], but still falls in the expected range of 40% and 80% MVC [36]. Adding one extra evaluations between 50% and 100% MVC, such as 70% MVC, would offer a relatively accurate estimation regarding the max recruitment threshold in BBM. On the contrary, the average MU depth on the spastic side was not force-related across four contraction levels. The very subtle depth change from 30% to 50% MVC on this side indicates a max recruitment threshold consistent with Jensen et al’s observation of 40% MVC in post-stroke patients [35]. Across four contraction levels, the average MU depth on the spastic side changed from 37.05 % to 40.60% of arm radius, which provides invaluable depth reference for BoNT injection in clinical spasticity management.

For the inter-side comparison, the spastic side average MU depth is greater than that of non-spastic side at low contraction levels of 10% and 30% MVC; the opposite result was acquired at high contraction levels of 50% and 100% MVC. This phenomenon may be due to the MU recruitment pattern disruption [6, 37] in which large MUs are prematurely recruited to compensate the compromised MU firing rate after upper motoneuron lesions [37–39]. An alternative explanation for this phenomenon is the large motor units which locate in the deeper region of muscle are denervated following stroke and then the muscle fibers comprising these motor units are reinnervated by the superficial small MUs [40–41], leading to a larger average MU depth at lower contraction levels and a relatively smaller average MU depth at high contraction levels. In addition, both of spastic and non-spastic average MUAP amplitude increased with the contraction force, suggesting the recruitment of larger size MUs at higher contraction level and evidencing the “size principle”. We acknowledge the average stroke duration of 61.6± 44.1 months in this study presents a wide time span. Nevertheless, previous studies indicated that the motor unit denervation may start to occur as early as 4 to 30 hours after stroke [42–43] and the stroke with a duration over six months was commonly considered as chronic [6, 41]. Therefore, we considered all our participants as chronic stork survivors and did not differentiate the stroke duration in this study. We agree that factors, such as the stroke duration, rehabilitation/recovery progress, etc could also play an important role in neuromuscular property alterations, a rigorous and quantitative comparison between patients with different stroke duration remains as open research question, and is worthy of further investigation in the future.

FWHM was also employed as one metric for MU depth evaluation. The spastic side and inter-side comparison results are similar to corresponding depth comparisons; whereas a significant FWHM difference was observed between 50% and 100% MVC on the non-spastic side. This discrepancy may attribute to two reasons. First, FWHM is not only related to the MU depth, but also depends on MU size. Therefore, the linear relationship between the normalized MU depth and FWHM and the empirical MU depth estimation approach reported by Roeleveld et al. are only applicable to low contraction level of 10% MVC to ensure small MUs are mainly recruited [28]. Second, due to the small number of electrodes in muscle fiber transverse direction in this study, the assumption of symmetry and data extrapolation may lead to erroneous estimation of FWHM. However, the high consistency between the linear regression results of the current study and Roeleveld et al’s (slope:1.1537 vs 1.2556, R^2: 0.5768 vs 0.7569) demonstrates the accuracy of MU depth determined by 3DIZI approach.

5. Conclusion

This study presents the first attempt to non-invasively and quantitatively characterize the MU distribution inside the spastic and non-spastic bilateral BBMs of chronic stroke patients by combining the cutting-edge techniques of HD-sEMG electrode array, EMG signal decomposition and bioelectrical source imaging. Our findings not only provide direct evidence that larger MUs with higher recruitment threshold locate in deeper muscle regions in non-spastic BBM, but also demonstrates that upper motor neuron lesions can alter MU distribution via MU recruitment pattern disruption and motor unit denervation and the subsequent collateral reinnervation. The findings of this study significantly advance our understanding regarding the neurophysiology of human muscles and the neuromuscular alterations following stroke, as well as offer important depth information for BoNT injection in clinical spasticity management.