© 2020 IOP Publishing Ltd

1. Introduction

Neurological disorders due to strokes and spinal cord injuries (SCI) are the leading cause of long-term disabilities [1], which limit these individuals’ daily activities and impose financial burdens on the health care system [1]. Among the different motor and sensory functions for individuals with paralysis, regaining arm and hand function is considered the first priority to help improve the quality of life [2].

Neuromuscular electrical stimulation is widely used to generate muscle contractions for assistive or therapeutic purposes. Electrical stimulation can evoke muscle contractions to help restore purposeful

functional movement, which is termed functional electrical stimulation (FES) [3, 4]. In conventional electrical stimulation, electrodes are placed at the mus-cle belly near the motor points. The distal branches of the motor axons are directly activated, and mus-cle activation patterns are non-physiological. First, superficial muscles are activated primarily [5], which can constrain movement kinematics due to a limited access of deep muscles. Second, motor units (MUs) are recruited either in a physiologically reversed order with large and fast-fatigable MUs recruited first [6] or in a random order [7]. In addition, the activation of MUs is highly synchronized [8], and the elicited twitch forces are time-locked to the stimulus pulses. This

Y Zheng and X Hu

Printed in the UK

AB5E09

JNEIEZ

© 2020 IOP Publishing Ltd

2020

00

J. Neural Eng.

JNE

10.1088/1741-2552/ab5e09

00

1

13

Journal of Neural Engineering

IOP

16

September

2019

28

November

2019

2

December

2019

Muscle activation pattern elicited through transcutaneous stimulation near the cervical spinal cord

Yang Zheng and Xiaogang Hu1,2

Joint Department of Biomedical Engineering, University of North Carolina at Chapel Hill and North Carolina State University, Chapel Hill, NC 27599, United States of America

1 University of North Carolina at Chapel Hill, 116 Manning Drive, 10206B Mary Ellen Jones Bldg, Chapel Hill, NC 27599-7575, United States of America

2 Author to whom any correspondence should be addressed.

E-mail: xiaogang@unc.edu

Keywords: transcutaneous electrical stimulation, spinal cord, hand function, arm function, muscle activation

Supplementary material for this article is available online

AbstractObjective. Neuromuscular electrical stimulation can help activate muscles of individuals with neurological disorders. However, conventional electrical stimulation targets distal branches of motor axons, and activates muscles non-physiologically. For example, stimulation at the muscle belly activates muscles in a highly synchronized manner. Accordingly, we investigated whether the muscle activation pattern was more asynchronized through transcutaneous stimulation near the cervical spinal cord (tsCSC). Approach. A stimulation array was placed on the posterior side near the cervical spinal cord, to target the arm and hand muscles. Stimulation trains of 10 Hz and 30 Hz were delivered. Electromyogram signals were recorded to quantify the muscle activation patterns. Arm and finger joint kinematics were also recorded using a motion capture system. Main results. After an initial synchronized activation prior to 35 ms after stimulation onset, we observed substantial asynchronized muscle activities with a long latency (>35 ms). The asynchronized activation is also more evident in distal muscles compared with the proximal muscles. In addition, the decreased synchronization level of muscle activities correlated with a reduced fluctuation of joint movement. The highly asynchronized muscle activities indicated an activation of the sensory axons and/or dorsal roots as well as a possible involvement of some spinal-supraspinal circuitry. Significance. Our tsCSC approach can improve the muscle activation pattern during electrical stimulation with a possible involvement of the spinal and supraspinal pathways, which can facilitate applications on rehabilitation/assistance of individuals with impaired motor function.

PAPER

RECEIVED

16 September 2019

REVISED

28 November 2019

ACCEPTED FOR PUBLICATION

2 December 2019

PUBLISHED

JNL:JNE PIPS: AB5E09 TYPE: PAP TS: NEWGEN DATE:5/2/20 EDITOR: IOP SPELLING: US

J. Neural Eng. 00 (2020) 000000 UNCORRECTED PROOF

2

Y Zheng and X Hu

muscle activation pattern is different from voluntary muscle contractions, in which asynchronous MU dis-charges produce steady forces at a range of different firing rates.

Recently, transcutaneous electrical stimulation at the proximal segment of the median/ulnar nerve bundles has been used to elicit hand grasp [9, 10] and hand opening [11] patterns. Since the nerve bundle encloses the motor axons innervating different muscle fibers, deep muscles and fatigue-resistant muscle fib-ers can be activated by changing the stimulation loca-tion and intensity, thereby leading to more dexterous finger movement [10] and reduced muscle fatigue compared with conventional FES [12]. In previous studies of proximal nerve stimulation, clustered sub-threshold current pulses with a carrier frequency of approximately 10 kilohertz have been demonstrated to disperse the activation of MUs [13], and are able to reduce muscle fatigue compared with the conventional single suprathreshold current pulses [14, 15]. How-ever, compared with voluntary muscle contractions, the synchronization level of MU activation is still substanti ally high, which is manifested as well-defined compound muscle action potentials (CMAPs) time-locked to the stimulation onsets.

Recently, noninvasive cervical spinal stimulation has been applied to improve voluntary control of the hand in individuals with muscle weakness [16–18], by modulating the excitability of the spinal cord circuity. The transcutaneous spinal cord stimulation can acti-vate the sensory axons in the dorsal roots even though simultaneous activation of motor axons is also pos-sible [19]. The activation of sensory axons can elicit muscle contractions through the spinal and supraspi-nal pathways with a physiological recruitment order of the MUs, which can help reduce muscle fatigue during electrical stimulation [8]. However, previous studies [16–20] mainly aimed to modulate the excitability of the spinal cord, in order to promote functional recov-ery rather than directly eliciting upper limb motions as a FES technique for rehabilitation/assistance.

Accordingly, in our current study, we sought to investigate the muscle activation patterns with upper limb motions elicited through the transcutaneous stimulation near the cervical spinal cord (tsCSC). Specifically, current was delivered through a stimu-lation electrode array placed near the cervical spinal cord. The elicited upper limb kinematics were meas-ured, and monopolar electromyogram (EMG) was recorded to capture the muscle activation pattern. After an initial synchronized muscle activation, our results showed highly asynchronized muscle activi-ties with a long latency (35 ms after stimulation onset), which indicated an activation of sensory axons and/or dorsal roots as well as a possible involvement of spinal-supraspinal circuitries. The asynchronized activation was also more evident in distal muscles compared with the proximal muscles. This asynchronized muscle acti-vation indicated that the tsCSC can potentially lead to a sustainable force output. The outcomes could help improve the application of electrical stimulation for rehabilitation/assistance of individuals with impaired motor function.

2. Methods

2.1. Experiment2.1.1. SubjectsSix male subjects (21–34 years of age) without any known neurological disorders were recruited in the study. All subjects gave written informed consent with protocols approved by the Institutional Review Board of the University of North Carolina at Chapel Hill.

2.1.2. Data recordingJoint kinematic data were acquired using an 8-camera OptiTrack System (Natural Point, Inc, Corvallis, OR) to track the 3D position of reflective markers with a sampling rate of 120 Hz (figure 1(a)). Sixteen markers with a 6.4 mm diameter were attached on the dorsal side of the hand to measure the angles of the metacarpophalangeal (MCP) flexion/extension of the thumb, index, middle, ring, and pinky, and the wrist

Figure 1. Monopolar EMG signals were recorded at the corresponding muscles according to the motions elicited (a). The flexion/extension angles of the elbow, the wrist, and the MCP joints from the thumb, index, middle, ring, and pinky were recorded with a motion capture system (a). Monophasic current pulses were delivered with the anode stimulation electrodes arranged in a 3 × 3 array near the cervical spinal cord (b), and three cathode electrodes placed above the clavicle.

J. Neural Eng. 00 (2020) 000000

3

Y Zheng and X Hu

flexion/extension. Five 9.5 mm markers were attached on the posterior side of forearm as the reference. The other four 9.5 mm markers were attached to the forearm and the upper arm with two markers each to measure the elbow flexion/extension.

Gel-based electrodes with a dimension of 23mm × 23mm were placed at the biceps, triceps, extensor digitorum communis (EDC), extensor carpi ulnaris (ECU), flexor carpi radialis (FCR), and flexor digitorum superficialis (FDS) (figure 1(a)). Monopo-lar EMG signals were recorded with the reference strap placed at the elbow and the ground strap at the wrist via the Sessantaquattro system (OT Bioelettronica, Torino, Italy). The EMG signals were amplified with a digital band pass filter of 10–500 Hz, and a sampling rate of 2000 Hz.

2.1.3. Stimulation paradigmMonophasic current pulses of 0.6 ms were generated with a multi-channel stimulator (Multichannel Systems, Reutlingen, Germany). The rows of a switch matrix (Agilent Technologies, Santa Clara, CA) were connected to the anode and cathode of a stimulator channel. The columns of the switch matrix were connected to 12 stimulation electrodes, of which three were used as cathode and placed above the clavicle, approximately 6 cm to the midsagittal plane. The other nine electrodes were used as anode and arranged as a 3 × 3 array at the posterior side near the cervical spinal cord, thereby targeting the arm and hand (figure 1(b)). The cathode electrodes were placed at the anterior side largely because stronger muscle contractions can be elicited compared with the configuration with cathode electrodes at the posterior side using the same current intensity. The stimulator and the switch matrix were controlled using a custom MATLAB user interface, and stimulation trains can be delivered to any cathode and anode pair.

During the experiment, the subjects were seated in an armchair. The forearm was supported in the pronated position with a soft foam pad on the desk with the MCP, the wrist and the elbow joints flexed naturally before stimulation onset. All the cathode and anode pairs that can elicit independent or mul-tiple joint movement were first identified through a searching procedure across all cathode and anode combinations. Although anode stimulation was used to target the dorsal root and the sensory neurons can discharge at high frequencies (100–200 Hz), a high stimulation frequency (such as 100 Hz) can mask the detailed muscle activations from each stimulation. In order to have enough time to observe the response from each stimulation input, a lower stimulation fre-quency was necessary. For each cathode and anode pair, two stimulation frequencies, i.e. 30 Hz and 10 Hz, were tested sequentially in two trials. In this setting, we would have 33.3 ms and 100 ms interval to analyze the muscle response. In each trial, fifteen 3 s stimula-tion trains were delivered with a 2 s resting interval to

avoid possible muscle fatigue. Within each 3 s stimula-tion train, the current amplitude was modulated with a trapezoid, i.e. the current amplitude of the pulses first increased from 1 mA to the designated current for the plateau period (2.4 s long) and then decreased to 1 mA. The stimulation intensity was adjusted to induce moderate or strong muscle contractions with matched joint range of motion between the two stimulation frequencies, such that the difference of joint range of motion between trials was within ±10%. The result-ing stimulation current amplitude was 6.13 ± 1.90 mA (mean ± standard deviation) and 7.10 ± 2.07 mA for the 30 Hz and 10 Hz trials, respectively. The subjects were instructed to keep relaxed during each 3 s stimulation train and put their limbs back to the initial position when the stimulation trial was over. Several trials were performed before data acquisition to ensure that the subjects were familiarized to the stimulation.

2.2. Data processing2.2.1. EMG activityFigure 2(a) illustrates exemplar EMG recordings of the EDC muscle from a representative trial with finger and wrist extension. The EMG signals that were 5 ms prior to and 33 ms after the onset of individual stimulation pulses were extracted from the trials with 30 Hz stimulation. The EMG signals that were 12 ms prior to and 98 ms after the onset of individual stimulation pulses were extracted from the trials with 10 Hz stimulation. The EMG segments were then averaged to obtain the average EMG of individual channels (figure 3). Stimulation artifacts were identified at time zero, and the EMG segments and the average EMG after stimulation artifacts were extracted for further analysis. The baseline activity within the 2 s intervals between stimulation trains were also segmented into 38 ms or 110 ms segments with no overlap according to the stimulation frequency for individual trials, and the average baseline acitivity was also obtained (figure 3). The baseline activity was extracted as a comparison in order to assess any residual muscle activation after stimulation ended.

Two measurements were calculated from the EMG signals to quantify the synchronization level of mus-cle activations. First, the maximum cross-correlation coefficients (termed correlation coefficient in the subsequent text) out of a range (maximum range of each 30 or 10 Hz condition) of lagged correlation coefficients were calculated to quantify the similarity of CMAPs. This maximum correlation coefficient can tease out the potential influence of temporal jitters in the long-latency responses. A high correlation indi-cates similar shapes between two CMAPs either with or without a delay, and therefore highly synchronized muscle activations. In contrast, asynchronized activa-tion can lead to more random shapes, and therefore low correlations. For the 30 Hz trials, the correlation coefficients were calculated for all possible pairs of the EMG segments (light blue in figure 3) and then

J. Neural Eng. 00 (2020) 000000

4

Y Zheng and X Hu

averaged across all pairs for indiviudal channels. For the 10 Hz trials, the muscle activation showed different patterns with the time division point at approximately 35 ms after the stimulation onset (figure 3). Before the time division point, the EMG signals were synchro-nized, and had well-defined CMAPs time-locked to the stimulation onset. On the contrary, the EMG sig-nals were highly asynchronized after the time division point. Therefore, the EMG segments and their average from the trials with the 10 Hz stimulation frequency were further separated into two segments (termed pre-35 EMG and post-35 EMG, respectively). The correla-tion coefficients were then calculated for all pairs of the pre-35 EMG segments and for all pairs of the post-35 EMG segment, respectively. Then, the average correla-tion coefficients across all pairs were obtained for the pre-35 EMG and post-35 EMG, respectively.

The second measurement (termed amplitude ratio) was the ratio between the average amplitude (root mean square) of all individual EMG segments (light blue in figure 3) and the amplitude of the aver-age EMG (dark blue in figure 3). If muscle activations were asynchronous, more cancellations were expected

when the average EMG was obtained, resulting in a larger amplitude ratio. Otherwise, a small amplitude ratio can be obtained. For the 10 Hz trials, the ampl-itude ratio was also calculated for the pre-35 and post-35 EMG, respectively.

The correlation coefficient and amplitude ratio measurements were also calculated using the baseline activity for comparison, in order to demonstrate that the post-35 signals were indeed muscle activities rather than pure background noise as in the baseline activity. The average amplitude (root mean square) across all segments was also calculated, and were compared with the average baseline activities. The correlation coef-ficient, the amplitude ratio, and the amplitude were all averaged across channels (muscles) for individual trials.

In order to investigate the amplitude change of the three types of EMG recordings (pre-35 EMG and post-35 EMG from 10 Hz trials, and EMG from 30 Hz trials) over successive stimulations within the 3 s stimula-tion trains, the EMG amplitude was first calculated for individual stimulations, and then averaged across all muscles for individual trials. Then, the amplitude was

Figure 2. The exemplar EMG signal of the EDC muscle (a) and the joint angles (b) from a representative trial with a stimulation frequency of 30 Hz. Red and black dashed lines mark the onset and the end of the stimulation trains, respectively.

J. Neural Eng. 00 (2020) 000000

5

Y Zheng and X Hu

normalized by the amplitude value when the stimula-tion current first reached the maximum, to eliminate the influence of trials with large EMG amplitudes on the trials with small EMG amplitudes. In the end, the EMG amplitude was averaged across all trials for indi-vidual stimulations and individual subjects.

2.2.2. Joint kinematicsThe joint angle definitions of the MCP, wrist, and elbow joints are shown in figure 1(a). The joint angles were calculated as either the angle between two vectors (MCP and elbow) or the angle between two planes (e.g. wrist) determined by the corresponding markers. Figure 2(b) illustrates the joint angles of the MCP, wrist, and elbow from a representative trial with 30 Hz stimulation. A fused muscle contraction was evident at 30 Hz stimulation. When the stimulation was 10 Hz, obvious oscillations can be observed from the joint angles. Figure 4(a) illustrates the original trajectory of the elbow joint from a representative trial. To investigate whether asynchronized EMG activities can reduce the fluctuation of joint angles, the fluctuation amplitude in the 10 Hz trials was also quantified. Specifically, the joint angles were first band-pass filtered (4th order Butterworth filter, zero-

phase digital filtering, pass band: 8–12 Hz) to remove the absolute joint angle and high frequency noise (figure 4(b)). Then, the 2 s segments from 1 s to 3 s after the stimulation onset were extracted from the filtered trajectories. The standard deviations of individual segments were estimated and were then averaged across segments to represent the fluctuation amplitude for individual joint angles. The kinematic fluctuation of the 30 Hz trials was not analyzed, because the contraction with the 30 Hz stimulation frequency was fused, and the resolution of the motion capture system was not high enough to capture the possible small kinematic fluctuations.

3. Results

Figure 5 shows the average correlation coefficient, amplitude ratio, and amplitude of EMG and baseline activities across all subjects. The three measurements of all types of recordings (EMG from 30 Hz trials, pre-35 and post-35 EMG from 10 Hz trial, and baseline activities from 10 Hz and 30 Hz trials) were first averaged across all trials for individual subjects, and the three measurements of the baseline activities from 10 Hz and 30 Hz trials were further averaged

Figure 3. The EMG and baseline activity of different muscles from six different representative trials with the stimulation frequency of 30 Hz (left column) and 10 Hz (right column). The stimulation onset is at 0 ms. The illustrated data of the same muscle (each row) came from the same subject and the same stimulation electrode pair. The segment of baseline activities was the same with that of the EMG segments.

J. Neural Eng. 00 (2020) 000000

6

Y Zheng and X Hu

before statistical analysis. The correlation coefficients between CMAPs from the 30 Hz trials (figure 5(a)) showed a large variation ranging from 0.55 to 0.91. In contrast, the correlation coefficients from the 10 Hz trials had a relatively small variation, and the correlation coefficient of the pre-35 EMG was larger and the post-35 EMG was smaller, respectively, compared with the correlation coefficient from the 30 Hz trials. Moreover, the average correlation coefficient of the post-35 EMG was close to that of the baseline activity (0.53 versus 0.48). Repeated measures analysis of variance (ANOVA) showed that there were significant differences (F(3,15) = 82.88, p < 0.0001) of the correlation coefficient among the four types of recordings (baseline activity, pre-35 EMG

and post-35 EMG from 10 Hz trials, and EMG from 30 Hz trials). Further post-hoc test with the Holm–Bonferroni correction showed that the correlation coefficient differed significantly between any two types of recordings (p < 0.05). The amplitude ratio showed similar results to the correlation coefficient, which indicated that the pre-35 EMG had the highest synchronization level, and the post-35 EMG had the lowest synchronization level. Repeated measures ANOVA showed that the amplitude ratio was significantly different between four types of recordings (F(3,15) = 56.2, p < 0.0001). Further post-hoc test showed that there were significant differences between any two types of recordings (p < 0.05). As for the amplitude of recordings, repeated measures

Figure 4. The original (a) and band-pass filtered (b) trajectory of the elbow joint angle from a representative trial with 10 Hz stimulation frequency. Only data between 40 s to 66 s are shown here for better illustration.

Figure 5. The average correlation coefficient, amplitude ratio, and amplitude of EMG and baseline activities across all subjects with the stimulation frequency of 30 Hz and 10 Hz, respectively. The circles represent the measurement values of individual trials from all subjects. The error bars represent the standard deviation. *p < 0.05.

J. Neural Eng. 00 (2020) 000000

7

Y Zheng and X Hu

ANOVA showed that there were significant differences between the four types of recordings (F(3,15) = 29.54, p < 0.0001). The post-hoc test demonstrated that the amplitude differed significantly between any two types of recordings (p < 0.05).

As shown in figure 3, the activation of the EDC muscle was more asynchronized compared with the biceps and triceps muscles. In order to investigate whether the synchronization level differs between dif-ferent types of muscles, such as between the forearm muscles and the upper arm muscles, and between the extensor muscles and the flexor muscles, the correla-tion coefficient and the amplitude ratio of EMG activi-ties were compared between the forearm and upper arm muscles (figures 6(a) and (b)) and between the extensor and flexor muscles (figures 6(c) and (d)). These two variables were first averaged across all fore-arm muscles and all upper arm muscles, separately, and then averaged across all trials for individual sub-jects. With the 30 Hz stimulation, the correlation coef-ficient of the CMAPs in the forearm muscles was sig-nificantly smaller than that of the upper arm muscles (t(5) = −2.312, p = 0.0344). With the 10 Hz stimu-lation, the correlation coefficients of both the pre-35 EMG (t(5) = −4.1551, p = 0.0044) and post-35 EMG

(t(5) = −2.0378, p = 0.0486) activities of the fore-arm muscles were also significantly smaller than that of the upper arm muscles, and the amplitude ratio of the pre-35 EMG activities of the forearm muscles was significantly larger than that of the upper arm muscles (t(5) = 2.7160, p = 0.0210). The amplitude ratio of the post-35 EMG activities was larger compared with the upper arm muscles. However, the differences were only near significant (t(5) = 1.6945, p = 0.0755). The correlation coefficient and the amplitude ratio were first averaged across all extensor muscles and all flexor muscles, separately, and then averaged across all trials for individual subjects. The results showed that the synchronization level of the post-35 EMG activities from the 10 Hz trials of the extensor muscles was significantly lower (correlation coef-ficient: t(5) = −4.222, p = 0.0042; amplitude ratio: t(5) = 2.7929, p = 0.0192) than that of the flexor muscles. However, there was no significant difference in the pre-35 EMG activities from the 10 Hz trials or the EMG activities from the 30 Hz trials (p > 0.05).

With the same firing rate, asynchronous activa-tions of muscle fibers can induce more fused con-tractions, leading to reduced fluctuations of joint kinematics. We also investigated whether the synchro-

Figure 6. The comparison of the correlation coefficient (a) and the amplitude ratio (b) of EMG activities between forearm muscles and upper arm muscles, and the comparison of the correlation coefficient (c) and the amplitude ratio (d) of EMG activities between extensor and flexor muscles. The error bars represent the standard deviation. *p < 0.05. **p < 0.01.

J. Neural Eng. 00 (2020) 000000

8

Y Zheng and X Hu

nization level of muscle activations can influence the fluctuation amplitude of joint motions, quantified as the amplitude of the band-pass filtered joint angles. In this study, only the elbow joint angle fluctuation was analyzed because, first, elbow flexion/extension was the most common movement across subjects; second, joint angle fluctuations of fingers and wrist could arise from the oscillation of the elbow joint. The scatter plots in figure 7 illustrate the relation between the cor-relation coefficient of CMAP or the amplitude ratio of upper arm muscles and the fluctuation amplitude of the elbow joint using the 10 Hz trials. The linear regres-sion results showed a significant positive correlation between the joint angle fluctuation and the correla-tion coefficient for both the pre-35 EMG (figure 7(a), p < 0.05) and post-35 EMG (figure 7(b), p < 0.01). There was also a significant negative correlation between the joint angle fluctuation and the amplitude ratio for the post-35 EMG (figure 7(d), p < 0.01). However, the regression showed no significant correla-tion between the joint angle fluctuation and the ampl-itude ratio of the pre-35 EMG (figure 7(c), p > 0.05).

To evaluate the time dependent changes of the elicited muscle activities, we also quantified the ampl-itude of the three types of EMG recordings over suc-cessive stimulations within the 3 s stimulation trains for individual subjects (figure 8). After the stimulation intensity reached the maximum, the EMG amplitude decreased at the second stimulation in the 30 Hz trials for all subjects, and then increased at the third stimu-lation in five subjects (top panel). Paired t-test with Holm–Bonferroni correction showed that the EMG

amplitude of the first stimulation was significantly larger than that of the second stimulation (p < 0.01), and there was no significant difference either between the first and third stimulations, or between the second and third stimulations (p > 0.05). After the first sev-eral stimulations, the EMG amplitude showed differ-ent patterns for different subjects. In three subjects, the amplitude increased slowly until the stimulation cur-rent began to decrease. In the other three subjects, the amplitude decreased progressively over time. When the stimulation frequency was 10 Hz, there was no con-sistent trend within the first three stimulations for the pre-35 EMG (middle panel). As for the post-35 EMG (bottom panel), the amplitude increased at the second stimulation in five subjects. However, the increase was not significant (p > 0.05).

4. Discussion

In this study, we investigated whether muscle activation can be elicited using the tsCSC technique. The muscle activation patterns were captured with monopolar EMG recordings, and the synchronization level of muscle activities was quantified. The results showed that the tsCSC technique can activate the upper limb muscles through reflex pathways. Specifically, after the initial synchronized activation, highly asynchronized muscle activities were elicited 35 ms after the stimulation, which demonstrated the activation of sensory axons and/or dorsal roots. The short-latency response with a high synchronization level may contain the H-reflex component due to the

Figure 7. Regression analysis between the correlation coefficient (a) and (b), or the amplitude ratio (c) and (d) of upper arm muscles and the elbow joint fluctuation for the pre-35 and post-35 EMG, respectively.

J. Neural Eng. 00 (2020) 000000

9

Y Zheng and X Hu

monosynaptic connection between the afferent fibers and the motoneurons, while the long-latency response indicated the potential involvement of spinal-supraspinal circuitries, such as spinal interneurons and even motor cortex. Our findings indicated that the tsCSC method can potentially lead to a physiological pattern of muscle activation and thereby a sustainable force output. The outcomes can promote the application of the electrical stimulation technique for individuals with muscle weakness after central nervous system (CNS) injuries.

Consistent with the previous study [19], the well-defined CMAPs in the pre-35 EMG could be the mon-osynaptic H-reflex response and/or a possible M-wave, which appeared at approximately 5–10 ms after the stimulation onset. When the stimulation frequency was 30 Hz, the correlation coefficients of EMG activi-ties had a large range and can be as small as 0.3 in some trials. The asynchronized muscle activities demon-strated the involvement of afferent fibers and dorsal roots.

Highly asynchronized muscle activities occurred approximately 35 ms after stimulation onsets till the next stimulation pulse, and the amplitudes were sig-nificantly higher compared with the background noise, which demonstrated highly asynchronized and sustained muscle activation after a single stimulation pulse. Compared with the pre-35 EMG activities that generally showed well-defined CMAPs, the post-35 EMG had a significantly lower synchronization level. The asynchronized muscle activities with a much longer latency (>35 ms) could arise from the involve-ment of spinal-supraspinal circuitries, par ticularly interneuronal circuitry [21], brainstem and even

motor cortex because of the following reasons. First, the highly asynchronous activities should not come from direct activation of motor axons through stimu-lation, and there was likely an involvement of sensory pathways. Second, the long-latency EMG activities indicated that the involved pathways are potentially polysynaptic rather than monosynaptic H-reflexes. In previous studies, it has been demonstrated that the sudden stretch of a voluntarily contracting muscle in humans can elicit reflex activities with two comp-onents [22–27]. The early one (M1 response) occurs at a short latency following the monosynaptic path-way of the tendon jerk. The second one (M2 response) has a long latency, which has been demonstrated to be evoked, like the M1 response, by the intramuscular receptors. The long latency is partially due to a longer reflex pathway carrying sensory information from Ia afferents to supraspinal structures, like the motor cor-tex [28, 29]. Furthermore, the time required for infor-mation transmission from motor cortex to spinal cord is approximately 32.1 ms for upper arm muscles (such as the biceps) [25]. The time for the action potential to propagate from the motoneurons in the spinal cord to the end plates at the muscles is approximately 10 ms [19]. Assuming that the tsCSC in our study also acti-vated the Ia afferents and involved the cortical-spinal reflex pathway (M2 response), the time delay of the long-latency activities should be the summation of the two, which is approximately 42.1 ms. This is consistent with our results in that the onset of the long-latency EMG activities occurs approximately from 35 ms to 50 ms after the stimulation onset.

When stimulating at the proximal segment of the median/ulnar nerve bundles, the Ia afferent fibers can

Figure 8. The normalized amplitude of the three types of EMG activities over successive stimulations within 3 s stimulation trains for individual subjects. The mean and standard deviation of the first three stimulations are shown. Within each stimulation train, the stimulation current first increased, plateaued, and then decreased shown in the red trace. **p < 0.01.

J. Neural Eng. 00 (2020) 000000

10

Y Zheng and X Hu

also be activated. However, the long-latency reflex activity was not observed as shown in figure S1 (stacks.iop.org/JNE/00/0000/mmedia) during peripheral nerve stimulation. Earlier work has shown that the long-latency responses can be substantially influenced by voluntary effort [24, 27], which can modulate the excitability of the pathway. It is possible that the tsCSC may affect the excitability of the CNS due to its stimu-lation position near the spinal cord [30, 31], thereby activating the long-loop circuitry.

The possibility that the asynchronous activ-ity was due to voluntary effort can be excluded from two aspects. First, long-latency asynchronous activity was not observed through the peripheral nerve bun-dle stimulation at 10 Hz stimulation frequency even though a similar large joint motion can be observed. Two subjects undertook the experiment with stimu-lation at the proximal segment of the median/ulnar nerve bundles. They were also instructed to relax entirely during the stimulation and move their joints back to the initial position at the end of the trial. EMG was recorded from the FCR and FDS muscles as shown in figure S1. There was no long-latency EMG activity after approximately 35 ms. Second, voluntary reaction to changes in muscle length or other stimuli is typically larger than 100 ms [32]. In this study, the long-latency EMG activity occurred approximately 35 ms after the stimulation onset, which we believe was too fast for typical voluntary reactions. The baseline activity (red traces) between stimulation trains in figure 3 also showed no obvious EMG activity.

The asynchronized muscle activation pattern has multiple benefits in muscle force generation. First, it can help reduce muscle fatigue, because the muscle activation pattern mimics the voluntary contraction in which muscle fiber activations are highly asynchro-nized and fused contraction force can be obtained with a relatively low firing rate of MUs. In previous studies with the stimulation electrodes placed on the muscle belly or at the proximal segment of nerve bundles, it has been demonstrated that asynchronized muscle activation pattern can help mitigate muscle fatigue [14, 15, 33]. In addition, muscle activation through the spinal and supraspinal pathways indicated that MUs can be recruited in a physiological order with fatigue-resistant MUs being recruited first, which can also help to reduce muscle fatigue [8, 34]. Second, the active involvement of the CNS circuitry can potentially lead to both short-term and long-term neuroplasticity after CNS damage [35], which demonstrated the potential to enhance functional recovery after CNS injuries. Overall, the potential of our stimulation method in reducing muscle fatigue and promoting neuroplasti-city allows for more prolonged use of FES and provide greater long-term benefits to the users.

Beside the spinal cord stimulation, the peripheral nerve stimulation can also elicit muscle activities by activating afferent fibers [11, 35, 36]. They both have advantages and disadvantages. The peripheral nerve

stimulation can have better selectivity of targeted mus-cles, which can activate individual joints more inde-pendently such as single finger movements. Although peripheral nerve stimulation can activate afferent pathways, the elicited H-reflex is still time-locked well with the stimulation onset, and the firings of different recruited motoneurons are highly synchronized. In addition, different electrodes need to be placed care-fully to target different nerve bundles, and the place-ment may still be distorted due to joint motions, which could limit clinical practice. In contrast, the tsCSC activates the dorsal roots composed of a large number of nerve fibers associated with different muscles. The selectivity of targeted muscle activation may be chal-lenging. Given the observed partially asynchronized activity, the tsCSC method could lead to more asyn-chronized firings of motoneurons, which resembles physiological activation of neurons during voluntary effort. Lastly, the electrode grid is placed on the trunk, near the spinal cord. Different muscles can be activated by tuning the stimulation parameters, and the skin deformation of the trunk is small compared with that in arm segments, which can help to stabilize the elec-trode positions.

Our results showed that joint kinematics can reflect the difference of muscle activation patterns. Specifi-cally, there was a positive correlation between the syn-chronization level of EMG activities and the fluctua-tion amplitude of joint kinematics, especially for the post-35 EMG. However, the relation was not significant for the pre-35 EMG, possibly because the synchroniza-tion level of EMG activities was too high (correlation coefficient was mostly >0.8, figure 7(a)) to result in distinguishable joint kinematic fluctuations. The fluc-tuation amplitude of joint kinematics or contraction forces between the tsCSC and conventional FES can be compared in a further study, in order to further confirm whether the tsCSC can obtain fewer variable contrac-tions with lower stimulation frequency.

To improve voluntary control of the hand in tetra-plegic individuals using noninvasive cervical spinal stimulation, the electrode configuration is incon-sistent across studies [16–18]. Some studies utilized monophasic current pulses [19, 20]. Milosevic et al placed the cathode and the anode over the posterior and anterior sides of the neck, respectively [19], while Einhorn et al placed the anode over the posterior neck and the cathode at the clavicle [20]. Other studies uti-lized biphasic current pulses with one electrode placed over the posterior neck and the other one over the iliac crests [16, 17]. Milosevic et al explored the reflex activ-ity and demonstrated the ability to activate the dorsal roots [19]. However, the study also showed the lack of complete abolishment of the second response in the double-pulse stimulation test, and suggested that the evoked potentials may partially be a direct activa-tion of motor axons [19]. In Einhorn et al’s study, the attenuation of the second response in the double-pulse stimulation test was inconsistent across subjects [20].

J. Neural Eng. 00 (2020) 000000

11

Y Zheng and X Hu

In our current study, the cathode was placed above the clavicle and the anode was near the spinal cord mainly because stronger muscle contractions can be elicited compared with the reversed electrode polarity. The cathode is generally considered to induce depo-larization of neurons in conventional neuromuscular stimulation with electrodes placed at muscle bellies or nerve bundles. The situation could be different for the tsCSC, considering the spatial relation such as distance and orientation between the electrodes and the neu-rons [37] and the low-conductive tissues like the ver-tebra that enclose the spinal cord, dorsal/ventral roots, and dorsal root ganglions. The negatively charged ani-ons can accumulate near the spinal cord with a certain distance away from the anode, which has positively-charged cations. This effect has been shown in tran-scranial stimulation. For example, anode instead of cathode stimulation was used to elicit motor evoked potentials during transcranial electrical stimulation [38]. After all, our results demonstrated the involve-ment of the spinal and supraspinal pathway, which could help elicit upper limb motions in a more physi-ological way. However, a direct activation of the motor axons could still be possible. Since the activation loca-tion of the motor and sensory neurons were both close to the spinal cord, the M-wave and H-reflex response latency was too small to distinguish. The well-defined CMAP prior to 35 ms after stimulation onsets could be a combination of both M-wave and H-reflex activities.

A post-activation depression has been observed in H-reflex with a recovery time up to 200 ms for upper extremity muscles at rest [39], which was believed to be caused by the depletion of synaptic neural trans-mitters [40]. Our results showed that the EMG activ-ity in the 30 Hz trials first decreased significantly at the second stimulation, potentially due to post-activation depression. The EMG activity recovered from the third stimulation and then showed either slowly decreasing or increasing trend over subsequent stimulations. On the contrary, both the pre-35 and post-35 EMG in the 10 Hz trials showed no significant difference between the first three stimulations, and had a larger amplitude or at least comparable amplitude compared with the first stimulation over subsequent stimulations. This effect can arise from the longer recovery time between stimulations compared with the 30 Hz trials. In addi-tion, since the tsCSC could possibly activate some spinal-supraspinal circuitries, the activity with long latency might facilitate the motoneuron pool excitabil-ity, which could result in a decreased depression and a shorter recovery time. Other factors could also contrib-ute to the sustainable H-reflex activity over successive stimulations. The high-frequency stimulation can also trigger voltage dependent persistent inward cur rent, which can amplify the H-reflex amplitude [41, 42].

Our results showed that the synchronization level of extrinsic finger muscle activations from the fore-arm was lower compared with the upper arm elbow muscles. Typically, muscles that are necessary for fine

movement contain more spindles (therefore, a higher number of sensory axons) than muscles that are used for posture or coarse movement [43]. Therefore, the possibility of activating sensory axons is higher for the forearm muscles, thereby leading to more asynchro-nized muscle activation patterns.

One limitation of this study was that there was no optimization procedure for the stimulation param-eters (electrode position, stimulation intensity, fre-quency, et al) in order to maximize the muscle acti-vation elicited through the spinal and supraspinal pathways and minimize the co-activation of antago-nists. Our results showed that the tsCSC may elicit co-activation of antagonists, since the dorsal root contains the afferent fibers from all the arm and hand muscles. However, the preliminary results showed that the co-activation can be minimized or even eliminated (figure S2) by changing the stimulation parameters. Even though the co-activation of antagonists can be minimized, it remains unclear the extent to which the multi-muscle activation patterns induced by tsCSC may resemble a physiological activation of muscles. The dependence of the muscle activation pattern on the stimulation parameters was not explored either. The dorsal and ventral roots originate from the dor-sal and ventral horns of the spinal cord respectively and unite to form a spinal nerve within the vertebrate. Therefore, the distance between the stimulation posi-tion and the midsagittal plane can potentially deter-mine which part is mostly activated. In addition, the stimulation intensity could also affect the components that were mostly activated.

The other limitation of this study was that no par-ticipants with neurological disorders were involved. As a preliminary exploration, the current study mainly showed that the tsCSC can activate afferent pathways and induce H-reflex and long-latency reflex activities to elicit asynchronous EMG activation pat-tern. With impairment in the CNS, the ascending and descending pathways may be altered, which can result in abnormal muscle activation patterns. It is likely that the applicability of the tsCSC on different neurological conditions will be different. For exam-ple, the hyperexcitable reflex pathways and disinhi-bition of supraspinal pathways (e.g. reticulospinal) in stroke survivors partly contribute to the observed muscle spasticity [44]. As demonstrated in the cur-rent study, the tsCSC can elicit muscle activities though reflex pathways and longer latency supraspi-nal pathways, but the activation patterns induced by tsCSC in stroke survivors should be verified in a future study. It would also be interesting to test the tsCSC on individuals with a complete or incomplete SCI. The involvement of supraspinal circuit could be beneficial for incomplete SCI in terms of promoting plasticity. For complete cervical SCI, however, if the highly asyn-chronized long-latency activity is absent, it could pro-vide some support for our hypothesized supraspinal involvement.

J. Neural Eng. 00 (2020) 000000

12

Y Zheng and X Hu

5. Conclusions

The study demonstrated the feasibility of activating arm and hand muscles in a partially asynchronous way possibly through spinal and supraspinal pathways. Highly asynchronized muscle activities were observed with a long latency after the stimulation onset in tsCSC and the synchronization level was demonstrated to correlate with the fluctuation amplitude of the joint trajectories. This muscle activation pattern indicated that the sensory axons and spinal-supraspinal circuitries were involved in the tsCSC. Although the possibility of direct activation of motor axons was not excluded, our stimulation method still shows promise in generating upper limb movement by activating the CNS circuits. The motions elicited through the spinal and supraspinal pathways can be more fatigue-resistant and can also potentially facilitate neuroplasticity. These advantages can potentially alleviate some of the issues like muscle fatigue in electrical stimulation and can enhance the functional recovery of individuals with neurological injuries.

ORCID iDs

Xiaogang Hu https://orcid.org/0000-0002-8565-5940

References

[1] Benjamin E J et al 2017 Heart disease and stroke statistics-2017 update: a report from the American Heart Association Circulation 135 e146–603

[2] Anderson K D 2004 Targeting recovery: priorities of the spinal cord-injured population J. Neurotrauma 21 1371–83

[3] Howlett O A, Lannin N A, Ada L and McKinstry C 2015 Functional electrical stimulation improves activity after stroke: a systematic review with meta-analysis Arch. Phys. Med. Rehabil. 96 934–43

[4] Makowski N S, Knutson J S, Chae J and Crago P E 2014 Functional electrical stimulation to augment poststroke reach and hand opening in the presence of voluntary effort: a pilot study Neurorehabil. Neural Repair 28 241–9

[5] Gobbo M, Maffiuletti N A, Orizio C and Minetto M A 2014 Muscle motor point identification is essential for optimizing neuromuscular electrical stimulation use J. Neuroeng. Rehabil. 11 17

[6] Kubiak R J Jr, Whitman K M and Johnston R M 1987 Changes in quadriceps femoris muscle strength using isometric exercise versus electrical stimulation J. Orthopaed. Sports Phys. Ther. 8 537–41

[7] Gregory C M and Bickel C S 2005 Recruitment patterns in human skeletal muscle during electrical stimulation Phys. Ther. 85 358–64

[8] Barss T S, Ainsley E N, Claveria-Gonzalez F C, Luu M J, Miller D J, Wiest M J and Collins D F 2018 Utilizing physiological principles of motor unit recruitment to reduce fatigability of electrically-evoked contractions: a narrative review Arch. Phys. Med. Rehabil. 99 779–91

[9] Shin H, Zheng Y and Hu X 2018 Variation of finger activation patterns post-stroke through non-invasive nerve stimulation Frontiers Neurol. 9 1101

[10]Shin H, Watkins Z and Hu X 2017 Exploration of hand grasp patterns elicitable through non-invasive proximal nerve stimulation Sci. Rep. 7 16595

[11]Zheng Y and Hu X 2019 Elicited finger and wrist extension through transcutaneous radial nerve stimulation IEEE Trans. Neural Syst. Rehabil. Eng. 27 1875–82

[12]Shin H, Chen R and Hu X 2018 Delayed fatigue in finger flexion forces through transcutaneous nerve stimulation J. Neural Eng. 15 066005

[13]Zheng Y and Hu X 2018 Improved muscle activation using proximal nerve stimulation with subthreshold current pulses at kilohertz-frequency J. Neural Eng. 15 046001

[14]Zheng Y, Shin H and Hu X 2018 Muscle fatigue post-stroke elicited from kilohertz-frequency subthreshold nerve stimulation Frontiers Neurol. 9 1061

[15]Zheng Y and Hu X 2018 Reduced muscle fatigue using kilohertz-frequency subthreshold stimulation of the proximal nerve J. Neural Eng. 15 066010

[16]Inanici F, Samejima S, Gad P, Edgerton V R, Hofstetter C P and Moritz C T 2018 Transcutaneous electrical spinal stimulation promotes long-term recovery of upper extremity function in chronic tetraplegia IEEE Trans. Neural Syst. Rehabil. Eng.

[17]Gad P, Lee S, Terrafranca N, Zhong H, Turner A, Gerasimenko Y and Edgerton V R 2018 Non-invasive activation of cervical spinal networks after severe paralysis J. Neurotrauma 35 2145–58

[18]Freyvert Y, Yong N A, Morikawa E, Zdunowski S, Sarino M E, Gerasimenko Y, Edgerton V R and Lu D C 2018 Engaging cervical spinal circuitry with non-invasive spinal stimulation and buspirone to restore hand function in chronic motor complete patients Sci. Rep. 8 15546

[19]Milosevic M, Masugi Y, Sasaki A, Sayenko D G and Nakazawa K 2019 On the reflex mechanisms of cervical transcutaneous spinal cord stimulation in human subjects J. Neurophysiol.

[20]Einhorn J, Li A, Hazan R and Knikou M 2013 Cervicothoracic multisegmental transpinal evoked potentials in humans PLoS One 8 e76940

[21]McLean D L, Fan J, Higashijima S, Hale M E and Fetcho J R 2007 A topographic map of recruitment in spinal cord Nature 446 71–5

[22]Ghez C and Shinoda Y 1978 Spinal mechanisms of the functional stretch reflex Exp. Brain Res. 32 55–68

[23]Colebatch J, Gandevia S, McCloskey D and Potter E 1979 Subject instruction and long latency reflex responses to muscle stretch J. Physiol. 292 527–34

[24]Hammond P 1956 The influence of prior instruction to the subject on an apparently involuntary neuro-muscular response J. Physiol. 132 17

[25]Chan C, Jones G M, Kearney R and Watt D 1979 The ‘late’electromyographic response to limb displacement in man. I. Evidence for supraspinal contribution Electroencephalogr. Clin. Neurophysiol. 46 173–81

[26]Shemmell J, Krutky M A and Perreault E J 2010 Stretch sensitive reflexes as an adaptive mechanism for maintaining limb stability Clin. Neurophysiol. 121 1680–9

[27]Rothwell J, Traub M and Marsden C 1980 Influence of voluntary intent on the human long-latency stretch reflex Nature 286 496

[28]Phillips C G 1969 The Ferrier Lecture, 1968-Motor apparatus of the baboon’s hand Proc. R. Soc. B 173 141–74

[29]Marsden C, Merton P and Morton H 1976 Stretch reflex and servo action in a variety of human muscles J. Physiol. 259 531–60

[30]de Lima-Pardini A C et al 2018 Effects of spinal cord stimulation on postural control in Parkinson’s disease patients with freezing of gait Elife 7 e37727

[31]Bentley L, Duarte R, Furlong P, Ashford R L and Raphael J H 2016 Brain activity modifications following spinal cord stimulation for chronic neuropathic pain: a systematic review Eur. J. Pain 20 499–511

[32]Forgaard C J, Franks I M, Maslovat D, Chin L and Chua R 2015 Voluntary reaction time and long-latency reflex modulation J. Neurophysiol. 114 3386–99

J. Neural Eng. 00 (2020) 000000

13

Y Zheng and X Hu

[33]Malešević N M, Popovic L Z, Schwirtlich L and Popovic D B 2010 Distributed low-frequency functional electrical stimulation delays muscle fatigue compared to conventional stimulation Muscle Nerve 42 556–62

[34]Bergquist A J, Wiest M J, Okuma Y and Collins D F 2014 H-reflexes reduce fatigue of evoked contractions after spinal cord injury Muscle Nerve 50 224–34

[35]Bergquist A, Clair J, Lagerquist O, Mang C, Okuma Y and Collins D 2011 Neuromuscular electrical stimulation: implications of the electrically evoked sensory volley Eur. J. Appl. Physiol. 111 2409

[36]Klakowicz P M, Baldwin E R and Collins D F 2006 Contribution of M-waves and H-reflexes to contractions evoked by tetanic nerve stimulation in humans J. Neurophysiol. 96 1293–302

[37]Merrill D R, Bikson M and Jefferys J G 2005 Electrical stimulation of excitable tissue: design of efficacious and safe protocols J. Neurosci. Methods 141 171–98

[38]Burke D, Hicks R G and Stephen J 1990 Corticospinal volleys evoked by anodal and cathodal stimulation of the human motor cortex J. Physiol. 425 283–99

[39]Rossi-Durand C, Jones K E, Adams S and Bawa P 1999 Comparison of the depression of H-reflexes following previous activation in upper and lower limb muscles in human subjects Exp. Brain Res. 126 117–27

[40]Hultborn H, Illert M, Nielsen J, Paul A, Ballegaard M and Wiese H 1996 On the mechanism of the post-activation depression of the H-reflex in human subjects Exp. Brain Res. 108 450–62

[41]Powers R K, Nardelli P and Cope T C 2012 Frequency-dependent amplification of stretch-evoked excitatory input in spinal motoneurons J. Neurophysiol. 108 753–9

[42]Heckman C, Gorassini M A and Bennett D J 2005 Persistent inward currents in motoneuron dendrites: implications for motor output Muscle Nerve 31 135–56

[43]Pittman J 2019 An Integrated Approach to Neuroscience (Cambridge Scholars Publishing) pp 121–2

[44]Nielsen J B, Crone C and Hultborn H 2007 The spinal pathophysiology of spasticity–from a basic science point of view Acta Physiol. 189 171–80

J. Neural Eng. 00 (2020) 000000