Muscle-selective disinhibition of corticomotor representations using a motor imagery-based brain-computer interface

Accepted Manuscript Muscle-selective disinhibition of corticomotor representations using a motor imagerybased brain-computer interface Mitsuaki Takemi, Maeda Tsuyoshi, Yoshihisa Masakado, Hartwig Roman Siebner, Junichi Ushiba PII:

S1053-8119(18)30768-7

DOI:

10.1016/j.neuroimage.2018.08.070

Reference:

YNIMG 15231

To appear in:

NeuroImage

Received Date: 24 January 2018 Revised Date:

14 August 2018

Accepted Date: 28 August 2018

Please cite this article as: Takemi, M., Tsuyoshi, M., Masakado, Y., Siebner, H.R., Ushiba, J., Muscleselective disinhibition of corticomotor representations using a motor imagery-based brain-computer interface, NeuroImage (2018), doi: 10.1016/j.neuroimage.2018.08.070. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

1

ACCEPTED MANUSCRIPT 1 2

Muscle-selective disinhibition of corticomotor representations using a motor imagery-based brain-computer interface

3

Mitsuaki Takemi1, 2, Tsuyoshi Maeda1, Yoshihisa Masakado3, Hartwig Roman Siebner2, 4,

5

Junichi Ushiba5, 6*

RI PT

4

6 7

1

8

Technology, Keio University, Kanagawa, Japan

9

2

School of Fundamental Science and Technology, Graduate School of Science and

10

Hvidovre, Hvidovre, Denmark

11

3

12

Japan

13

4

14

Copenhagen, Denmark

15

5

16

University, Kanagawa, Japan

17

6

18

Japan

M AN U

Department of Rehabilitation Medicine, Tokai University School of Medicine, Kanagawa,

Department of Neurology, Copenhagen University Hospital Bispebjerg,

TE D

Department of Biosciences and Informatics, Faculty of Science and Technology, Keio

Keio Research Institute for Pure and Applied Sciences (KiPAS), Keio University, Kanagawa,

EP

19

SC

Danish Research Centre for Magnetic Resonance, Copenhagen University Hospital

*Corresponding author: Junichi Ushiba, Ph.D.

21

3-14-1 Hiyoshi, Kohoku-ku, Yokohama, Kanagawa, Japan

22

Tel/Fax: +81-45-563-1141; E-mail: [email protected]

AC C

20

2

ACCEPTED MANUSCRIPT

Abstract

2

Bridging between brain activity and machine control, brain-computer interface (BCI)

3

can be employed to activate distributed neural circuits implicated in a specific aspect

4

of motor control. Using a motor imagery-based BCI paradigm, we previously found a

5

disinhibition within the primary motor cortex contralateral to the imagined

6

movement, as evidenced by event-related desynchronization (ERD) of oscillatory

7

cortical activity. Yet it is unclear whether this BCI approach does selectively

8

facilitate corticomotor representations targeted by the imagery. To address this

9

question, we used brain state-dependent transcranial magnetic stimulation while

10

participants performed kinesthetic motor imagery of wrist movements with their right

11

hand and received online visual feedback of the ERD. Single and paired-pulse

12

magnetic stimulation were given to the left primary motor cortex at a low or high

13

level of ERD to assess intracortical excitability. While intracortical facilitation

14

showed no modulation by ERD, short-latency intracortical inhibition was reduced the

15

higher the ERD. Intracortical disinhibition was only found in the agonist muscle

16

targeted by motor imagery at high ERD level, but not in the antagonist muscle.

17

Single pulse motor-evoked potential was also increased the higher the ERD. However,

18

at high ERD level, this facilitatory effect on overall corticospinal excitability was not

19

selective to the agonist muscle. Analogous results were found in two independent

20

experiments, in which participants either performed kinesthetic motor imagery of

21

wrist extension or flexion. Our results showed that motor imagery-based BCI can

22

selectively disinhibit the corticomotor output to the agonist muscle, enabling

23

effector-specific training in patients with motor paralysis.

24

Keywords: electroencephalogram (EEG), event-related desynchronization (ERD),

25

short-latency intracortical inhibition (SICI), sensorimotor rhythm (SMR), state-dependent,

26

transcranial magnetic stimulation (TMS)

AC C

EP

TE D

M AN U

SC

RI PT

1

3

ACCEPTED MANUSCRIPT

1. Introduction

2

Brain-computer interface (BCI) links brain activity with machine control. BCI-based

3

paradigms are widely used to engage neural pathways associated with specific motor

4

tasks, especially in patients who are unable to generate overt movement due to severe

5

paralysis. BCI can help them to regulate motor-related brain activity (Mukaino et al.,

6

2014; Ono et al., 2014; Pichiorri et al., 2015; Soekadar et al., 2015). During BCI

7

training, patients are instructed to imagine hand movements. They concurrently receive

8

online feedback of their motor-related brain activity via visual or somatosensory

9

stimuli. Repeated use of such BCI is believed to strengthen residual neural networks

10

involved in motor control and facilitate motor recovery through error-based learning

11

and several neuroplasticity mechanisms, such as Hebbian-like plasticity and

12

use-dependent plasticity (Belardinelli et al., 2017; Kraus et al., 2016a; Pichiorri et al.,

13

2011; Sitaram et al., 2012; Soekadar et al., 2014; Vukelić and Gharabaghi, 2015;

14

Young et al., 2014).

15

Most BCI paradigms using motor imagery for rehabilitation purposes in post-stroke

16

hemiparesis exploit sensorimotor event-related desynchronization (ERD) as online

17

readouts of motor cortical activity (Mukaino et al., 2014; Ono et al., 2014; Pichiorri et

18

al.,

19

electroencephalography (EEG) and refers to the amplitude decrease of sensorimotor

20

oscillations in the mu (7–13 Hz) and beta (14–26 Hz) frequency bands. The use of ERD

21

as online EEG marker of cortical activity can be explained by the fact that the

22

physiological characteristics of ERD have been studied intensively over the last

23

decades. Sensorimotor mu ERD evoked by the imagery of voluntary hand movements

24

is known to show similar spatial profiles to the ERD associated with the preparation of

25

voluntary hand movements (Pfurtscheller and Neuper, 1997). Yuan et al. (2010) found

26

that

27

blood-oxygen-level-dependent signals in functional magnetic resonance imaging

28

during motor imagery. Furthermore, sensorimotor mu ERD during voluntary inhibition

29

of motor memory retrieval was significantly larger in dystonia patients whose motor

TE D

Young

et

al.,

2014).

ERD

can

be

readily

recorded

with

AC C

EP

2015;

M AN U

SC

RI PT

1

both

mu

and

beta

ERDs

co-localize

and

co-vary

with

the

4

ACCEPTED MANUSCRIPT

system has shown to be hyper-excitable relative to healthy subjects (Hummel et al.,

31

2002). A recent study demonstrated high test-retest reliability of the magnitude of beta

32

ERD induced by wrist movements across multiple sessions (Espenhahn et al., 2016).

33

Together, these studies qualify ERD during motor tasks as a reliable biomarker

34

representing motor cortical activities.

35

In line with these findings, we have shown that the magnitude of mu/beta ERD during

36

the use of a motor imagery-based BCI reflects the excitability of primary motor cortex

37

(M1) (Takemi et al., 2013) and spinal motoneurons (Takemi et al., 2015). Hence, the

38

use of motor imagery-based BCI can increase the excitability of corticospinal output

39

engaged for imagined movements. Robust increases in the corticospinal excitability

40

(CSE) have also been reported in relation to beta ERD magnitudes after BCI

41

interventions (Kraus et al., 2016a). Moreover, we recently demonstrated that the

42

association between sensorimotor mu ERD during a BCI control and the excitability of

43

ipsilateral M1 varies depending on whether the users imagine distal hand or proximal

44

arm movements (Hasegawa et al., 2017). Yet it remains to be clarified whether the

45

increase in induced M1 excitability in relation to ERD magnitude is confined to the

46

muscle representation engaged by the imagery task. Like an increase in CSE is specific

47

to the muscles involved in motor imagery (Fadiga et al., 1999; Hashimoto and

48

Rothwell, 1999; Ikai et al., 1996; Levin et al., 2004; Tremblay et al., 2001), the

49

BCI-induced excitability increase should be confined to the agonist muscle without

50

affecting the excitability of antagonist muscle. However, since somatotopy in ERD

51

patterns during arm/hand movements are highly overlapping (Miller et al., 2007), the

52

users of motor imagery-based BCI may elevate ERD by engaging imagery of incorrect

53

or unspecific movements. If this were the case, the BCI-induced excitability increase

54

could be expressed in the agonist and antagonist muscle representations.

55

To test these hypotheses, we used BCI-inspired brain state-dependent brain stimulation,

56

in which transcranial magnetic stimulation (TMS) is given depending on the current

57

state of sensorimotor EEG oscillations. In the experiment, participants were asked to

58

perform kinesthetic motor imagery of wrist movements (i.e., flexion or extension) and

AC C

EP

TE D

M AN U

SC

RI PT

30

5

ACCEPTED MANUSCRIPT

they received online visual feedback of the amount of ERD during the motor imagery.

60

Single-pulse or paired-pulse TMS was given over the M1 contralateral to the imagined

61

movement and triggered by the extent of ERD. Single-pulses were applied to assess

62

CSE, and paired-pulses were applied to probe GABA A-mediated short-latency

63

intracortical inhibition (SICI) (Ziemann et al., 1996) and NMDA-mediated intracortical

64

facilitation (ICF) (Ziemann et al., 1998). Motor-evoked potentials (MEPs) were

65

recorded with surface electromyography (EMG) from wrist flexor and extensor

66

muscles. This procedure enabled us to test whether motor imagery-based BCI

67

selectively facilitated corticomotor excitability in the imagined muscle movement (i.e.,

68

agonist) relative to the non-imagined movement (i.e., antagonist) and the association

69

between the amount of BCI-induced ERD increase and alterations in cortical

70

excitability.

AC C

EP

TE D

M AN U

SC

RI PT

59

6

ACCEPTED MANUSCRIPT

2. Materials and Methods

72

2.1 Participants

73

24 volunteers were recruited. 10 participated in experiment 1 (aged 21.8 ± 1.5 years; six men,

74

four women). The results of experiment 1 prompted a follow-up experiment (experiment 2).

75

We screened additional 14 volunteers of whom 10 volunteers participated in experiment 2

76

(aged 22.4 ± 0.8 years; eight men, two women). The screening procedure is described in detail

77

in the subsequent section. All participants were right-handed and had normal vision

78

(according to self-reports), with no history of brain disorders or other health problems. All

79

participants were naïve to the purpose of the experiment and the procedures. The purpose and

80

experimental procedure were explained to the participants, before written informed consent

81

was obtained. The study was approved by the local ethics committee of Keio University

82

(IRB approved number: 23-16, 24-23) and performed in accordance with the

83

Declaration of Helsinki.

84

2.2 TMS and EMG

85

Surface EMG was recorded from the right extensor carpi radialis (ECR) and flexor carpi

86

radialis (FCR) muscles using two pairs of bipolar Ag/AgCl electrodes ( φ = 10 mm). The

87

cathode electrodes were placed over the muscle belly, while the anode electrodes were placed

88

20 mm distal from the cathode electrodes. Impedance for all channels was maintained below

89

20 kΩ throughout the experiment. EMG signals were band-pass filtered (5–2000 Hz with 2nd

90

order Butterworth) with a 50-Hz notch to avoid power-line noise contamination, digitized at 5

91

kHz using a biosignal amplifier (Neuropack MEB-9200; Nihon Koden, Tokyo, Japan), and

92

monitored throughout the experiment. The EMG data of each trial were stored on a computer

93

from 50 ms before the TMS pulse to 150 ms after the pulse.

AC C

EP

TE D

M AN U

SC

RI PT

71

94

TMS was delivered using two interconnected single-pulse magnetic stimulators

95

(Magstim BiStim 2 ; Magstim, Whiteland, UK) producing a monophasic current

96

waveform in a 70-mm figure-of-eight coil. In experiment 1, we identified the optimal

97

coil position at which a single-pulse TMS evoked a MEP response in the ECR muscle

98

with the lowest stimulus intensity, referred to as the motor hotspot. At this position, the

99

coil was oriented approximately 45° to the sagittal plane. The resting motor threshold

7

ACCEPTED MANUSCRIPT

(RMT) was defined as the lowest stimulator output eliciting an MEP in the relaxed

101

ECR of >50 µV peak-to-peak in 5 out of 10 trials (Rossini et al., 1994). The motor

102

hotspot was marked with ink over the scalp to ensure an exact repositioning of the coil

103

throughout the experiment. In experiment 2, the motor hotspot and RMT for the FCR

104

muscle were first identified, using the same procedure as for experiment 1. We

105

subsequently tested the RMT of the ECR muscle by applying TMS to the FCR hotspot

106

in 14 volunteers in order to confirm that comparable MEP amplitudes can be elicited in

107

the FCR and ECR muscles. The difference in RMTs of the ECR and FCR muscles was

108

≤1% of the maximum stimulator output in 10 of the 14 participants. Participants with

109

matched RMTs were included in experiment 2.

SC

RI PT

100

Single-pulse TMS was set at an intensity of 120% of the RMT. In paired-pulse TMS

111

paradigm, a subthreshold conditioning stimulus was set at 80% of the RMT, and was

112

delivered through the same magnetic coil at 2, 3, 10 or 15 ms prior to the suprathreshold test

113

stimulus adjusted to 120% of the RMT (Vaalto et al., 2011; Takemi et al., 2013). The

114

stimulus intensity remained constant throughout the experiment for each participant.

115

2.3 EEG measurement and analyses

116

EEG was recorded with five Ag/AgCl electrodes ( φ = 10 mm) placed at C3 and 30 mm

117

anterior, posterior, medial and lateral to C3 to cover the sensorimotor hand area. Ground and

118

reference electrodes were placed at the forehead and left earlobe, respectively. Impedance for

119

all channels was maintained below 5 kΩ throughout the experiment. EEG signals were

120

band-pass filtered (0.1–100 Hz with 4th order Butterworth) with a 50 Hz notch, and digitized

121

at 512 Hz using a biosignal amplifier (g.USBamp; g.tec medical engineering, Graz, Austria).

122

The EEG data was transmitted to the workspace of MATLAB 2011a (The Mathworks, Natick,

123

MA) for online analysis and offline storage.

AC C

EP

TE D

M AN U

110

124

The EEG signal from C3 was then re-referenced by the four neighboring

125

electrodes. This derivation, referred to as a Laplacian, is a known spatial filter that

126

emphasizes sensorimotor oscillations originating immediately below the electrodes

127

(McFarland et al., 1997). The Laplacian EEG data were segmented into successive

128

512-point (1000 ms) windows with 480-point overlapping. Fast Fourier transformation

8

ACCEPTED MANUSCRIPT 129

with a Hanning window was applied in each segment to estimate the power spectrum

130

density. ERD was calculated at each segment with a time resolution of 62.5 ms and a

131

frequency resolution of 1 Hz, according to the following calculation:

ERD( f ,t) =

R( f ) − A( f ,t) R( f )

RI PT

132

A denotes the power spectrum density of the EEG, at time, t, and frequency, f. R is the

134

mean power spectrum of the baseline period, defined as the 3-s resting interval between

135

4–1 s prior to the onset of motor tasks. A large positive value indicates a large decrease in

136

the EEG power spectrum compared with a baseline period. Note that ERD value at time t

137

was delayed by 500 ms on average due to FFT over 1000 ms window with symmetrical

138

Hanning function.

139

2.4 Experimental protocol

140

The experiment consisted of three conditions in the following order: screening, resting, and

141

BCI condition (Fig. 1a). All conditions were performed on the same day with the same

142

EEG and EMG electrode settings. The participants sat in a comfortable armchair and

143

placed their hand, palm side down, on the armrest. A 20-inch computer monitor was

144

placed 60 cm in front of the participant’s eyes. In the screening condition (Fig. 1a, top), the

145

subjects repeatedly performed sustained right wrist extension (experiment 1) or flexion

146

(experiment 2). Each trial started with the presentation of the word ‘Rest’ in the center of the

147

monitor. Six seconds later, the word ‘Ready’ was presented for 1 s. Thereafter a vertical line

148

was displayed in the center of the monitor, which prompted participants to perform 5-s of

149

sustained wrist extension or wrist flexion with their right hand. A single trial lasted 12 s and

150

was repeated 30 times.

AC C

EP

TE D

M AN U

SC

133

151

ERD is typically observed over the sensorimotor cortex contralateral to the

152

imagery hand, but the most reactive frequency band displaying ERD slightly varies

153

between participants (Pfurtscheller et al., 2006; Takemi et al., 2013, 2015). Thus the

154

3-Hz width frequency band displaying the largest ERD was determined within mu and

155

beta frequencies (7–26 Hz) in each participant by using his/her EEG signals recorded

156

in the screening condition that may reflect cortical activity during voluntary wrist

9

ACCEPTED MANUSCRIPT

movement. The frequency band displaying the largest ERD was calculated immediately

158

after the screening condition and used for online ERD calculation in the BCI condition.

159

The resting condition (Fig. 1a, middle), which served as a control for the BCI condition,

160

was always conducted before the BCI condition. Similar to the screening condition,

161

each trial started with the presentation of the word ‘Rest’ for 6 s, before the word ‘Ready’ was

162

subsequently displayed for 1 s. The monitor then displayed the vertical line and a horizontal

163

bar that moved randomly for 5 s while participants remained at rest. In 80% of trials, one

164

single or paired-pulse TMS (with inter-stimulus intervals (ISIs) at 2, 3, 10, and 15 ms) was

165

randomly applied to the motor hotspot while the horizontal bar was moving to avoid

166

habituation with repeated stimulation. The experimenter continuously monitored EMG

167

activities of the target muscles to ensure that the participants maintained complete relaxation.

168

Trials contaminated by more than ±25 µV of background EMG activities were discarded

169

online. A total of 50 MEPs without any background EMG activities (10 MEPs for each pulse

170

configuration) were collected.

M AN U

SC

RI PT

157

The BCI condition was conducted using the same time scheme as the screening session

172

(Fig. 1a, bottom). Instead of performing the actual movement, participants remained their

173

right hand (palm side down) on the armrest and performed kinesthetic motor imagery of the

174

sustained wrist extension in experiment 1 and of the sustained wrist flexion in experiment 2

175

exerting maximum mental effort. Besides, we did not explicitly give instruction regarding

176

amplitude and force of movements. The length of the horizontal bar displayed on the monitor

177

was updated every 62.5 ms according to the ERD magnitude during the motor imagery

178

period. Thus the participants received online visual feedback of the ERD (Fig. 1b). The

179

center of the axis represented no ERD, without any amplitude changes in the EEG from the

180

baseline period. The position of the bar shifted to the right when the ERD value was positive;

181

the bar moved to the left when ERD value was negative. Before the BCI condition,

182

participants were informed that successful motor imagery (i.e., positive ERD) would shift the

183

bar to the right edge of the monitor, while unsuccessful motor imagery (i.e., negative ERD)

184

would shift the bar to the left edge.

AC C

EP

TE D

171

185

The BCI condition consisted of two blocks. In one block, single-pulse TMS or

186

paired-pulse TMS at short ISI (2 or 3 ms) or long ISI (10 or 15 ms) was given to the motor

10

ACCEPTED MANUSCRIPT

hotspot at the first instance when ERD exceeded 5% during the motor imagery period. In the

188

other block, TMS was triggered at the first instance that ERD exceeded 15% during the motor

189

imagery period. To test SICI, we chose the short ISI, which showed stronger inhibition (2 or 3

190

ms) in the resting condition. Similarly, for ICF we chose the long ISI, which showed stronger

191

facilitation (10 or 15 ms) in the resting condition. The order of the blocks and single and

192

paired-pulse TMS within a single block were pseudo-randomized and counter-balanced

193

between participants to minimize the accumulative effect of TMS in conjunction with ERD

194

(Kraus et al., 2016b). Trials contaminated by more than ±25 µV of background EMG activity

195

were discarded online and 30 MEPs without any background EMG activities were collected

196

in each block (10 MEPs for each pulse configuration). The BCI condition was paused every

197

15 min to give the participants a short break and avoid mental fatigue. In the breaks, we asked

198

participants whether they wanted to drink, readjust their seat position, and put eye drops in

199

order to avoid swallowing, neck muscle contraction and eye blinks, which induce large

200

artifacts in EEG, as much as possible. The TMS coil was then positioned to the motor hotspot

201

and the remaining trials were resumed. We confirmed that the RMT of the agonist muscle did

202

not change before and after any suspension.

203

2.5 MEP analysis

204

Peak-to-peak MEP amplitudes were measured offline. We analyzed the data in the resting and

205

BCI conditions separately. For the resting condition data, MEP amplitudes were averaged for

206

each TMS protocol (i.e., single-pulse and paired-pulses with short ISI (2 or 3 ms) and long ISI

207

(10 or 15 ms)) and muscle (i.e., FCR and ECR). To ensure that paired-pulse TMS with

208

different ISIs induced intracortical inhibition and facilitation, the mean MEP amplitude in the

209

resting condition was analyzed using a two-way ANOVA, with the TMS protocol and muscle

210

as within-subject factors. Here, SICI was evaluated using the mean MEP amplitude provoked

211

with a short ISI that showed smaller MEP. Likewise, ICF was evaluated using the mean MEP

212

amplitude with a long ISI showing larger MEP.

AC C

EP

TE D

M AN U

SC

RI PT

187

213

Peak-to-peak MEP amplitudes in the BCI condition were averaged for each TMS

214

protocol, muscle and ERD magnitude (5%ERD and 15%ERD). To quantify SICI and ICF the

215

mean MEP amplitude by paired-pulse TMS was expressed as a percentage of the mean MEP

11

ACCEPTED MANUSCRIPT

amplitude by single-pulse TMS. Mean single-pulse MEP amplitude in the BCI condition was

217

subtracted from the mean amplitude obtained in the resting condition. Also, mean SICI and

218

ICF strength in the BCI condition was subtracted from the mean strength obtained in the

219

resting condition. These subtracted values were then statistically analyzed using ANOVA and

220

bootstrap test. To test whether motor imagery-based BCI modulated different aspect of

221

intra-M1 excitability, an ANOVA was separately applied to the single-pulse MEP, SICI and

222

ICF. ERD magnitude and muscle were set as within-subject factors, while the imagery task

223

(wrist flexion or extension) was set as the between-subject factor. The normal distribution of

224

the data was confirmed using the Kolmogorov-Smirnov test. Post-hoc pairwise

225

comparisons were performed using the Sidak procedure to correct for multiple

226

comparisons. A bootstrap test was performed to examine whether a distribution of TMS

227

measures obtained during imagery of wrist movement was significantly different from

228

those in the resting condition. For each TMS measure, data were randomly sampled

229

with replacement and the bootstrapped mean was calculated 5000 times. A histogram

230

of the bootstrapped means was then used to test significance with a Bonferroni-Holm

231

correction. The type I error was set to 0.05. IBM SPSS Statistics, version 22 for

232

Windows, was used for ANOVA and post-hoc pairwise comparisons. Bootstrap test was

233

performed using MATLAB 2014a.

AC C

EP

TE D

M AN U

SC

RI PT

216

12

ACCEPTED MANUSCRIPT

3. Results

235

3.1 Electroencephalogram

236

In the screening session, all participants showed ERD over the sensorimotor cortex

237

contralateral to the hand executed wrist movement. The median and interquartile range of

238

experiment 1 were 12.6 and 9.2–13.9% (during 5-s of sustained right wrist extension),

239

respectively, and 14.9 and 11.0–19.2% in experiment 2 (during 5-s of sustained right wrist

240

flexion), respectively. The median and interquartile range of the frequency band showing

241

strongest ERD were 13 and 8–15 Hz in experiment 1 and 14 and 9–15 Hz in experiment 2,

242

respectively. The ERD characteristics in the screening session were not different between

243

experiments 1 and 2, based on Mann-Whitney test (ERD magnitude: p = 0.38; the frequency

244

band showing strongest ERD: p = 0.59).

M AN U

SC

RI PT

234

Since ERD is a relative measure and depends on the baseline power, we also compared

246

the Laplacian EEG power for the baseline period (4–1 s prior to the onset of motor

247

imagery in the BCI condition and 1–0 s before TMS was triggered in the resting condition)

248

in the trials where TMS was applied among conditions. The baseline EEG power at the 3-Hz

249

frequency band displaying the largest ERD, defined in the screening session, was averaged

250

over the trials, time, and frequencies. Friedman’s test revealed that the grand average of the

251

baseline EEG power was not different between conditions (rest, 5%ERD, 15%ERD) in both

252

experiment 1 ( χ 2 (2) = 2.6, p = 0.27) and 2 ( χ 2 (2) = 0.8, p = 0.67). Considering that the

253

baseline power was calculated with EEG signals more than 2 s later from the offset of

254

motor imagery, the baseline power was not contaminated by post-movement oscillatory

255

rebound.

256

3.2 MEP, SICI, and ICF in the resting condition

257

Figure 3 summarizes the group results. Data of a single subject are shown in figure 2 (left

258

column). Paired-pulse TMS at short ISI induced inhibition of the MEP (i.e., SICI), while

259

paired-pulse TMS at long ISI facilitated the MEP amplitude relative to single-pulse

260

stimulation (i.e., ICF). For the data acquired in experiment 1, ANOVA demonstrated an

261

interaction between the ISI and muscle (F(2,18) = 3.82, p = 0.042). For the ECR muscle,

262

post-hoc pairwise comparisons revealed smaller MEP amplitudes at short ISI (i.e., SICI

AC C

EP

TE D

245

13

ACCEPTED MANUSCRIPT

condition) relative to single-pulse TMS and paired-pulse TMS at long ISI (p < 0.001).

264

The FCR muscle showed only a trend difference towards smaller MEP amplitudes in

265

the SICI condition relative to single-pulse TMS (p = 0.061) and the ICF protocol (p =

266

0.057). In addition, MEP amplitudes of the ECR muscle tended to be larger than MEP

267

amplitudes of the FCR muscle for single-pulse TMS (p = 0.059) and the ICF protocol (p

268

= 0.069).

RI PT

263

The strength of SICI is known to be weaker when the test MEP amplitude is smaller

270

(Sanger et al., 2001). Given the trend showing differences in the MEP amplitudes of the ECR

271

and FCR muscles in experiment 1, this may complicate the comparison of SICI in the two

272

muscles with respect to the ERD magnitude. This prompted us to perform experiment 2, in

273

which we adjusted the mean test MEP amplitudes of the two muscles at rest. We only

274

enrolled those 10 subjects who showed no relevant difference in the RMTs between

275

ECR and FCR muscles (≤1% of the maximum stimulator output). In experiment 2, an

276

ANOVA using mean MEP amplitude as dependent variable demonstrated no interaction

277

between the TMS protocol and muscle (F(2,18) = 0.91, p = 0.42), suggesting matched MEP

278

amplitudes in the two muscles. A main effect of TMS protocol was confirmed (F(2,18) = 43.8,

279

p < 0.001). A post-hoc test revealed that the MEP provoked by the SICI protocol was

280

smaller than the MEP provoked by the single-pulse TMS and ICF protocol (p < 0.001).

281

In addition, the MEP of ICF was larger than the single-pulse MEP (p < 0.01).

282

3.3 MEP, SICI, and ICF in the BCI condition

283

Figure 4 summarizes the group results, and data from a single participant are shown in figure

284

2 (middle and right column). Motor imagery-based BCI increased the unconditioned MEP

285

amplitude evoked by single-pulse TMS. The increase in MEP depended on the magnitude of

286

ERD and was not limited to the agonist muscle engaged by motor imagery task (Fig. 4a).

287

Accordingly, ANOVA revealed a main effect of ERD (F(1,18) = 5.09, p = 0.037) and an

288

interaction between muscle and imagery task (F(1,18) = 9.76, p = 0.006). No other interactions

289

were found (p > 0.25). Post-hoc analysis revealed that single-pulse MEPs had larger mean

290

amplitudes at 15%ERD than 5%ERD (p = 0.037). During imagery of wrist flexion,

291

single-pulse MEPs of the FCR muscle were larger than the MEPs of the ECR muscle (p =

AC C

EP

TE D

M AN U

SC

269

14

ACCEPTED MANUSCRIPT

0.016). In addition, bootstrap tests revealed an increase in single-pulse MEP during motor

293

imagery relative to the resting condition regardless of the ERD magnitude, which mostly

294

occurred in the agonist muscle. During imagery of the wrist flexion, significant

295

differences were found in the FCR muscle at 5%ERD and 15%ERD (p < 0.05). During

296

imagery of wrist extension, significant differences were found in the ECR muscle at

297

5%ERD, and in the ECR and FCR muscles at 15%ERD (p < 0.05).

RI PT

292

In contrast to the unconditioned MEPs, the effects of motor imagery-based BCI on SICI

299

were specific to the muscle engaged by the imagery of wrist movement (Fig. 4b). This was

300

reflected by a three-way interaction among ERD, muscle, and imagery task (F(1,18) = 22.3, p

301

< 0.001), showing that both the amount of ERD and the imagery task influenced the relative

302

strength of SICI in a muscle-specific fashion. Post-hoc pair-wise comparisons provided

303

consistent results for the two imagery tasks. Attenuation of SICI scaled with the magnitude of

304

ERD in the agonist muscle that was targeted by the motor imagery task (p < 0.001). Moreover,

305

there was a significant difference of SICI between the agonist and antagonist muscle at

306

15%ERD (p = 0.001, by imagery of wrist extension; p = 0.002, by imagery of wrist flexion).

307

A bootstrap test for SICI strength during motor imagery relative to the resting

308

condition revealed that, regardless of the imagery task, SICI was only significantly reduced

309

in the agonist muscle at 15%ERD (p < 0.05).

TE D

M AN U

SC

298

In contrast to SICI, motor imagery-based BCI had no effect on the magnitude of ICF

311

(Fig. 4c). The respective ANOVA yielded no interaction between any factors (F(1,18) = 0.46−

312

1.65, p = 0.22−0.51). Further, there was no main effect for ERD (F(1,18) = 0.44, p = 0.51),

313

muscle (F(1,18) = 0.16, p = 0.70), and imagery task (F(1,18) = 0.47, p = 0.50).

314

3.4 Relationship between task compliance and SICI modulation in the BCI condition

315

Cortical excitability can be influenced not only by ERD magnitude but also by the degree of

316

task compliance (de Lange et al., 2008). Thus, we calculated the percentage of task success,

317

as the ratio of TMS applied trials to total number of motor imagery trials for each participant.

318

A paired t-test demonstrated a difference in the task success rate between the two conditions

319

in experiment 1 (p < 0.001: 54 ± 6% and 30 ± 4% at 5%ERD and 15%ERD, respectively) and

320

experiment 2 (p < 0.001: 57 ± 7% and 30 ± 6% at 5%ERD and 15%ERD, respectively).

AC C

EP

310

15

ACCEPTED MANUSCRIPT

However, there was no correlation between SICI strength in the agonist muscle, which was

322

reduced with a higher ERD, and the task success rates in those conditions (Experiment 1: r(18)

323

= -0.22, p = 0.34; Experiment 2: r(18) = -0.26, p = 0.27). This result indicated that the

324

reduction of SICI by the use of motor imagery-based BCI was not associated with task

325

difficulty.

RI PT

321

We also concerned that the duration of kinesthetic motor imagery (i.e., time since

327

participants started to perform motor imagery until TMS pulse was applied) affected the

328

amount of SICI reduction. Thus, using a paired t-test, we compared the duration of

329

kinesthetic motor imagery before TMS applied in the 5%ERD and 15%ERD conditions. The

330

results were as follows: 1.6 ± 0.1 s at 5%ERD and 3.0 ± 0.3 s at 15%ERD in

331

experiment 1 (p < 0.001); and 1.6 ± 0.2 s at 5%ERD and 3.1 ± 0.2 s at 15%ERD in

332

experiment 2 (p < 0.001). However, there was no correlation between the SICI strength

333

in the agonist muscle and duration of kinesthetic motor imagery in the BCI condition

334

(Experiment 1: r(18) = 0.31, p = 0.18; Experiment 2: r(18) = 0.34, p = 0.14). The present

335

results are compatible with the notion that the attenuation of SICI strength in the agonist

336

muscle was rather associated with ERD magnitude than the duration of kinesthetic

337

motor imagery. Yet a non-significant correlation is not sufficient to prove the lack of

338

causal relationship between the variables.

AC C

EP

TE D

M AN U

SC

326

16

ACCEPTED MANUSCRIPT

4. Discussion

340

In the present study, we tested whether the increase of M1 excitability by motor

341

imagery-based BCI is confined to the muscle representation engaged by the imagery

342

task. The findings can be summarized as follows. First, motor imagery-based BCI

343

increased the CSE, which was measured by single-pulse MEP. This increase was

344

associated with the ERD magnitude but not confined to the agonist muscle at high ERD

345

level. Second, although ICF showed no modulation by ERD, SICI was reduced when

346

the BCI required a high ERD level. This intracortical disinhibition was only found in

347

the agonist muscle targeted by motor imagery without affecting the excitability of the

348

antagonist muscle. Third, two experiments in which participants either performed

349

kinesthetic motor imagery of wrist extension or flexion yielded analogous results,

350

providing internal replication.

351

4.1 Changes in corticomotor excitability, evaluated with EEG-triggered TMS

352

At both levels of ERD, motor imagery-based BCI increased corticomotor excitability

353

as probed by single-pulse TMS. The facilitatory effect of our ERD-based BCI approach

354

was most prominently expressed in the muscle targeted by motor imagery, but also was

355

present in the antagonist muscle at the high ERD level. In agreement with the present

356

results, the M1 excitability profiles of an agonist muscle have been shown to increase

357

during a period of a voluntary movement (Hoshiyama et al., 1997; Reynolds and Ashby,

358

1999). We postulated that the BCI-induced increase in the corticomotor excitability

359

was mediated by analogous representations as the corresponding motor execution. Our

360

results also demonstrated larger mean MEP amplitudes at the 15%ERD relative to the

361

5%ERD. These results are compatible with the notion that the CSE of hand/arm

362

muscles may inversely scale with the synchronization of sensorimotor oscillations

363

(Hummel et al., 2002).

364

Contrary to M1 excitability profiles just before voluntary wrist movement (Hoshiyama

365

et al., 1997; Reynolds and Ashby, 1999), the BCI-induced increase in the CSE was not

366

confined to the agonist muscle of imagined movement, especially at the high ERD level.

AC C

EP

TE D

M AN U

SC

RI PT

339

17

ACCEPTED MANUSCRIPT

Although this may be surprising, a previous study evaluating CSE during motor

368

imagery actually demonstrated facilitation of the antagonist muscle MEP (Kasai et al.,

369

1997). Others revealed the inhibition of the antagonist MEP in 18 of the 28 tested

370

muscles during motor imagery (Ikai et al., 1996), but the majority of these changes

371

were not statistically significant. These observations led us to suggest that motor

372

imagery-based BCI have dynamic effects on the M1 excitability, which are similar but

373

not identical to those seen during motor execution. Note that the fixed order of the

374

conditions (resting and BCI) might have introduced order effects, as a TMS study

375

found single pulse TMS repeated every 2.5-5.5 s induced cumulative increases in the

376

CSE (Pellicciari et al., 2016). However, we expected such order effects to be stronger

377

if the BCI condition would be followed by measurements at rest. Therefore, we decided

378

against counter-balancing the two conditions.

379

We also found modulation of SICI by motor imagery-based BCI. SICI was

380

significantly reduced only at the high level of ERD. Moreover, contrary to the results

381

of the single-pulse MEP, BCI-associated disinhibition was confined to the agonist

382

muscle representation regardless of the amount of ERD. SICI was significantly

383

attenuated only at 15%ERD compared to the resting condition, unlike single-pulse

384

MEP, which was facilitated at 5%ERD. These discrepancies between MEP and SICI

385

results suggested a distinct role of MEP increase and SICI decrease in a motor

386

imagery-based BCI task. As a local intracortical phenomenon, SICI would be well

387

placed to modulate the relationship between adjacent intracortical representations via

388

changes in horizontal connections (Reis et al., 2008). Conversely, the increase in MEP

389

is a direct phenotype of corticospinal outputs, which are affected by inputs from

390

non-primary motor areas as well as spinal excitability. Thus, BCI-induced SICI

391

modulation may be not to directly drive an increase in the CSE, but rather serve to

392

focus it appropriately to the task. Furthermore, a previous study investigating the

393

precise timing of SICI changes in a simple reaction time protocol revealed that the

394

reduction of SICI occurred later than the increase of single-pulse MEP (Nikolova

395

et al., 2006). Augmentation of the MEP amplitude was initiated approximately 120

AC C

EP

TE D

M AN U

SC

RI PT

367

18

ACCEPTED MANUSCRIPT

ms before the reaction onset, though the inhibition was abolished less than 60 ms

397

prior to the onset. This relationship of early-MEP and late-SICI modulation in a

398

reaction time task was similar to our findings (Fig. 5). In our motor imagery-based

399

BCI task, an increase in the MEP was initiated with a weaker ERD than the

400

reduction of SICI. Considering that the ERD gradually increases toward the

401

movement onset (Deiber et al., 2012), voluntary movements and a BCI control may

402

share a specific aspect of motor control. A motor imagery-based BCI would thus

403

allow to primarily modulate corticomotor activity close to the changes

404

accompanying the actual movements. Indeed, corroborating physiological findings

405

have demonstrated that BCI may bridge the gap between motor imagery and actual

406

movements (Bauer et al., 2015).

407

Our results demonstrated that a motor imagery-based BCI exploiting ERD induces

408

deactivation of inhibitory M1 neurons innervating corticomotor outputs to the agonist

409

muscle. However, we still cannot determine the extent to which the observed SICI

410

changes were related merely to motor imagery. In this study, TMS pulse was given at a

411

certain ERD magnitude, of which the participant was aware due to the visual feedback.

412

Thus, the increased M1 excitability might be related to the imagery task and/or the

413

awareness of successful trials. Yet no correlation between the amount of SICI and task

414

success rates across participants suggested that motor imagery rather than

415

neurofeedback context caused modulation of motor cortex excitability. This

416

interpretation is supported by a recent study in which motor imagery with and without

417

neurofeedback induced significant modulation of ERD magnitude (Darvishi et al.,

418

2017). Our results also proved the possibility that neither task difficulty nor duration of

419

motor imagery modulated M1 excitability. Given that the motor imagery of hand

420

movements at different force levels did not alter both MEP amplitudes (Park and Li,

421

2011) and ERD magnitudes (Murphy et al., 2016; Zaepffel et al., 2013), the

422

neuromodulatory effects by the use of the present BCI approach may stem from brain

423

activities related to motor imagery rather than to task compliance. However, future

424

studies that compare the effects of motor imagery with and without online visual

AC C

EP

TE D

M AN U

SC

RI PT

396

19

ACCEPTED MANUSCRIPT

feedback of ongoing ERD are needed to clarify whether ERD-based BCI immediately

426

boosts corticomotor excitability relative to motor imagery alone.

427

4.2 How do sensorimotor EEG oscillations reflect TMS responses?

428

The present results demonstrated a negative relationship between the amount of ERD

429

and GABA A-mediated SICI in the agonist muscle of motor imagery task, but not in the

430

antagonist muscle. However, considering that EEG signal reflects a spatial sum of

431

neuronal activity, one may wonder how the disinhibition of the agonist muscle

432

representation by kinesthetic motor imagery was selectively reflected in the amount of

433

ERD. A key factor generating sensorimotor oscillations is a negative feedback loop

434

composed by excitatory thalamo-cortical relay nucleus and inhibitory thalamic

435

reticular nucleus (Steriade and Llinás, 1988). This negative feedback loop synchronizes

436

the firing of multiple cortical neurons. This generates EEG oscillations, with a voltage

437

that consists of a spatial sum of postsynaptic potentials. The amplitude is particularly

438

large when the loop is stable, such as at rest. However, the down-regulation of

439

inhibitory neural activity leads to the loop divergent. This results in desynchronization

440

of neural activities and decrease in the sum of postsynaptic potentials (Pfurtscheller

441

and Lopes da Silva, 1999; Suffczynski et al., 2001). Thus, suppression of the

442

sensorimotor oscillations could be regarded as a neural marker reflecting deactivation

443

of inhibitory neurons. The present results showed that motor imagery focally

444

attenuated the activity of inhibitory circuits in M1 in a dynamic manner. Given the

445

results, we assumed that the loop involving the cortical neurons projecting to the

446

agonist muscle was selectively divergent during motor imagery. This assumption is

447

supported by TMS studies in humans, which indicated that kinesthetic motor imagery

448

has dynamic and focal effects on the excitability of the agonist M1 area (Fadiga et al.,

449

1999; Hashimoto and Rothwell, 1999; Ikai et al., 1996; Levin et al., 2004; Tremblay et

450

al., 2001). Altogether, it suggests that partial desynchronization of the neuronal

451

populations innervating agonist muscle by the selective disinhibition might lead a

452

decrease in the sum of postsynaptic potentials, resulting in the attenuation of

453

sensorimotor EEG oscillations, even when the remaining neurons were in the loop and

AC C

EP

TE D

M AN U

SC

RI PT

425

20

ACCEPTED MANUSCRIPT synchronized.

455

4.3 Practical implication for the clinical use of BCI training

456

Several mechanisms have been hypothesized for neuroplasticity that facilitates motor

457

recovery, which might occur through the continued use of motor imagery-based BCI.

458

For example, somatosensory feedback triggered by ERD of ipsilesional M1 is believed

459

to induce Hebbian-like plasticity by which co-activation of the M1 and periphery

460

fosters the ipsilesional sensorimotor loop (Belardinelli et al., 2017; Kraus et al., 2016a;

461

Soekadar et al., 2014). Real-time visual feedback of the ERD strength allows the users

462

to voluntarily and repeatedly regulate the ipsilesional M1 excitability based on their

463

intention, which could promote use-dependent plasticity (Young et al., 2014). Note that

464

in either case a premise underlying neuroplasticity through BCI training is that ERD

465

magnitude during motor imagery reflects focal increase of the M1 excitability to the

466

targeted muscle of imagined movements.

467

In the current study, single-pulse TMS of M1 revealed that motor imagery-based BCI

468

can dynamically modulate CSE of the agonist muscle depending on ERD magnitude in

469

the healthy participants who were naïve to motor imagery task. This result suggests

470

people can perform motor imagery of targeted muscle contraction at a first instance

471

being the user of the BCI. However, since the BCI-induced increase in the CSE was not

472

confined to the agonist muscle at high ERD level, muscle-specific plasticity of

473

ERD-based BCI may be caused by other mechanisms. Our study points to a

474

muscle-selective effect on SICI as possible candidate mechanism. Motor imagery

475

caused a muscle-specific intracortical disinhibition of the agonist muscle but not the

476

antagonist muscle at a high level of ERD. Weakening the excitability of intracortical

477

inhibitory circuits concurrently with motor training is known to enhance synaptic

478

plasticity and foster the learning process (Ziemann and Siebner, 2008). The use of BCI

479

may acutely increase CSE not focally to the target muscles, but repeated use of such

480

BCI will aid to induce plasticity mostly in the M1 of the target muscle representation

481

by the selective disinhibition. Moreover, the somatosensory feedback (e.g., electrical

AC C

EP

TE D

M AN U

SC

RI PT

454

21

ACCEPTED MANUSCRIPT

stimulation to the targeted muscle of motor imagery and robotic orthosis that executes

483

imagined movement) contingent upon the occurrence of ERD allows sensorimotor

484

co-activation to be restricted to the targeted corticomuscular region (Kraus et al.,

485

2016a). This should help selectively reinforce the sensorimotor representations in a

486

BCI task (Ono et al., 2014; Soekadar et al., 2014). Adaptive modification of the ERD

487

threshold at which somatosensory feedback is given is known to foster an increase in

488

ERD throughout the intervention (Naros et al., 2016). This could further down-regulate

489

SICI in the brain, resulting in enhancement of motor improvement. However, to our

490

knowledge, no one has examined effects of the BCI training on cortical representations

491

of functionally reciprocal muscles. Yet, based on a previous study, we inferred that

492

cortical plasticity induced by motor imagery-based BCI is effector-specific and reflects

493

the muscular pattern required to overtly perform the imagined action (Pichiorri et al.,

494

2011).

495

4.4 Limitations

496

We found that the down-regulation of SICI in relation to ERD magnitude was only

497

prominent in the agonist muscle without affecting the SICI of the antagonist muscle.

498

While SICI reflects intracortical GABA A inhibition, other types of GABA-mediated

499

inhibition, such as long-interval intracortical inhibition (LICI) and late cortical

500

disinhibition (LCD), might also be concurrently modulated by the use of motor

501

imagery-based BCI (Chong and Stinear, 2017). LICI and LCD are thought to reflect the

502

degree of postsynaptic GABAB inhibition, and presynaptic GABAB disinhibition,

503

respectively, which should be both relevant to the facilitation of cortical plasticity

504

following motor training (Di Pino et al., 2014). Therefore, future studies are warranted

505

to test how GABAB related intracortical inhibition scales with ERD magnitude.

506

In the present study, the frequency band of ERD testing relationship with MEP, SICI,

507

and ICF has a bias towards mu frequency band. This is because ERD magnitude

508

accompanying actual wrist movements (evaluated in the screening session) was larger

509

in the mu frequency band than the beta band for most participants. It still remains to be

AC C

EP

TE D

M AN U

SC

RI PT

482

22

ACCEPTED MANUSCRIPT 510

determined which frequency band should be used for ERD-based BCI in post-stroke

511

hemiplegia

512

Ramos-Murguialday et al., 2013), beta ERD (Belardinelli et al., 2017), and both mu

513

and beta ERD (Mukaino et al., 2014; Pichiorri et al., 2015) have all shown significant

514

improvements of motor function in post-stroke patients. To tackle this question, future

515

studies need to systematically compare the effects on motor recovery that can be

516

achieved with motor imagery-based BCIs informed by mu- and beta-ERD. In our

517

previous study, we found that ERD magnitude in the frequency band showing largest

518

ERD during voluntary movements, but neither in the fixed mu (7−13 Hz) nor beta

519

(14−26 Hz) band, reflects spinal excitability (Takemi et al., 2015). This finding

520

suggests that individual adjustments of the target frequency band for BCI may be most

521

suited to promote functional recovery.

BCIs

operated

by mu

ERD

(Ono

et

al.,

2014;

AC C

EP

TE D

M AN U

SC

RI PT

rehabilitation.

23

ACCEPTED MANUSCRIPT

5. Conclusions

523

In the present study, we tested whether the increase of M1 excitability by motor

524

imagery-based BCI was confined to the muscle representation engaged by motor

525

imagery task. We found motor imagery-based BCI has distinct effects on CSE and

526

intracortical inhibition depending on ERD magnitudes. The increase of CSE, measured

527

by single-pulse MEP, was associated with the ERD magnitude, but not confined to the

528

agonist muscle when the BCI task required a high level of ERD. SICI was significantly

529

reduced only at the high level of ERD, and this intracortical disinhibition was only

530

found in the agonist muscle without affecting the SICI of the antagonist muscle

531

regardless of the amount of ERD. These electrophysiological results were confirmed in

532

two independent experiments, in which participants either performed kinesthetic motor

533

imagery of wrist extension or flexion. This meant that motor imagery-based BCI

534

selectively deactivates inhibitory interneurons in the M1 representation innervating agonist

535

muscle. This finding is important as it implies that cortical plasticity occurring through

536

repeated use of the BCI may be rendered effector-specific.

537

Acknowledgements

538

We thank Sayoko Ishii for her technical support. This work was supported by a KAKENHI

539

“Non-linear Neuro-Oscillology” (15H05880) from Japan Society for the Promotion of

540

Science (JSPS) to JU and a Novo Nordisk Foundation Interdisciplinary Synergy Program

541

“BaSiCs” (NNF14OC0011413) to HRS. MT was supported by a Grant-in-Aid for JSPS

542

Fellows (16J02485).

AC C

EP

TE D

M AN U

SC

RI PT

522

24

ACCEPTED MANUSCRIPT 543

References

544

Bauer, R., Fels, M., Vukelić, M., Ziemann, U., Gharabaghi, A., 2015. Bridging the gap

545

between motor imagery and motor execution with a brain-robot interface.

546

Neuroimage 108, 319–327. Belardinelli, P., Laer, L., Ortiz, E., Braun, C., Gharabaghi, A., 2017. Plasticity of

RI PT

547 548

premotor cortico-muscular coherence in severely impaired stroke patients with

549

hand paralysis. NeuroImage: Clinical 14, 726–733.

551 552

Chong, B.W.X., Stinear, C.M., 2017. Modulation of motor cortex inhibition during motor imagery. J Neurophysiol 117, 1776–1784.

SC

550

Darvishi, S., Gharabaghi, A., Boulay, C.B., Ridding, M.C., Abbott, D., Baumert, M., 2017. Proprioceptive Feedback Facilitates Motor Imagery-Related Operant

554

Learning of Sensorimotor β-Band Modulation. Front Neurosci 11, 60.

555

doi:10.3389/fnins.2017.00060

556

M AN U

553

Deiber, M.P., Sallard, E., Ludwig, C., Ghezzi, C., Barral, J., Ibañez, V., 2012. EEG alpha activity reflects motor preparation rather than the mode of action selection.

558

Frontiers in Integrative Neuroscience 6, 59. doi:10.3389/fnint.2012.00059

559

TE D

557

Di Pino, G., Pellegrino, G., Assenza, G., Capone, F., Ferreri, F., Formica, D., Ranieri, F., Tombini, M., Ziemann, U., Rothwell, J.C., Di Lazzaro, V., 2014. Modulation of

561

brain plasticity in stroke: a novel model for neurorehabilitation. Nat Rev Neurol 10,

562

597–608.

Espenhahn, S., de Berker, A.O., van Wijk, B.C.M., Rossiter, H.E., Ward, N.S., 2016.

AC C

563

EP

560

564

Movement-related beta oscillations show high intra-individual reliability.

565

Neuroimage 147, 175–185.

566

Fadiga, L., Buccino, G., Craighero, L., Fogassi, L., Gallese, V., Pavesi, G., 1999.

567

Corticospinal excitability is specifically modulated by motor imagery: a magnetic

568

stimulation study. Neuropsychologia 37, 147–158.

569

Hasegawa, K., Kasuga, S., Takasaki, K., Mizuno, K., Liu, M., Ushiba, J., 2017.

570

Ipsilateral EEG mu rhythm reflects the excitability of uncrossed pathways

571

projecting to shoulder muscles. J Neuroeng Rehabil 14, 85.

572

doi:10.1186/s12984-017-0294-2

25

ACCEPTED MANUSCRIPT 573 574 575

Hashimoto, R., Rothwell, J.C., 1999. Dynamic changes in corticospinal excitability during motor imagery. Exp Brain Res 125, 75–81. Hoshiyama, M., Kakigi, R., Koyama, S., Takeshima, Y., Watanabe, S., Shimojo, M., 1997. Temporal changes of pyramidal tract activities after decision of movement: a

577

study using transcranial magnetic stimulation of the motor cortex in humans.

578

Electroencephalogr Clin Neurophysiol 105, 255–261.

579

RI PT

576

Hummel, F.C., Andres, F., Altenmüller, E., Dichgans, J., Gerloff, C., 2002. Inhibitory control of acquired motor programmes in the human brain. Brain 125, 404–420.

581

Ikai, T., Findley, T.W., Izumi, S., Hanayama, K., Kim, H., Daum, M.C., Andrews, J.F.,

SC

580

Diamond, B.J., 1996. Reciprocal inhibition in the forearm during voluntary

583

contraction and thinking about movement. Electromyogr Clin Neurophysiol 36,

584

295–304.

M AN U

582

585

Kasai, T., Kawai, S., Kawanishi, M., Yahagi, S., 1997. Evidence for facilitation of

586

motor evoked potentials (MEPs) induced by motor imagery. Brain Res 744,

587

147–150.

Kraus, D., Naros, G., Bauer, R., Leão, M.T., Ziemann, U., Gharabaghi, A., 2016a.

TE D

588 589

Brain-robot interface driven plasticity: Distributed modulation of corticospinal

590

excitability. Neuroimage 125, 522–532.

Kraus, D., Naros, G., Bauer, R., Khademi, F., Leão, M.T., Ziemann, U., Gharabaghi, A.,

592

2016b. Brain State-Dependent Transcranial Magnetic Closed-Loop Stimulation

593

Controlled by Sensorimotor Desynchronization Induces Robust Increase of

594

Corticospinal Excitability. Brain Stimul 9, 415–424.

AC C

EP

591

595

de Lange, F.P., Roelofs, K., Toni, I., 2008. Motor imagery: a window into the

596

mechanisms and alterations of the motor system. Cortex 44, 494–506.

597

Levin, O., Steyvers, M., Wenderoth, N., Li, Y., Swinnen, S.P., 2004. Dynamical

598

changes in corticospinal excitability during imagery of unimanual and bimanual

599

wrist movements in humans: a transcranial magnetic stimulation study. Neurosci

600

Lett 359, 185–189.

601 602

McFarland, D.J., McCane, L.M., David, S.V., Wolpaw, J.R., 1997. Spatial filter selection for EEG-based communication. Electroencephalogr Clin Neurophysiol

26

ACCEPTED MANUSCRIPT 603

103, 386–394.

604

Miller, K.J., Leuthardt, E.C., Schalk, G., Rao, R.P.N., Anderson, N.R., Moran, D.W.,

605

Miller, J.W., Ojemann, J.G., 2007. Spectral changes in cortical surface potentials

606

during motor movement. J Neurosci 27, 2424–2432. Mukaino, M., Ono, T., Shindo, K., Fujiwara, T., Ota, T., Kimura, A., Liu, M., Ushiba,

608

J., 2014. Efficacy of brain-computer interface-driven neuromuscular electrical

609

stimulation for chronic paresis after stroke. J Rehabil Med 46, 378–382.

610

Murphy, B.A., Miller, J.P., Gunalan, K., Ajiboye, A.B., 2016. Contributions of

RI PT

607

Subsurface Cortical Modulations to Discrimination of Executed and Imagined

612

Grasp Forces through Stereoelectroencephalography. PLoS ONE 11, e0150359.

613

doi:10.1371/journal.pone.0150359

M AN U

614

SC

611

Naros, G., Naros, I., Grimm, F., Ziemann, U., Gharabaghi, A., 2016. Reinforcement

615

learning of self-regulated sensorimotor β-oscillations improves motor performance.

616

Neuroimage 134, 142–152.

617

Nikolova, M., Pondev, N., Christova, L., Wolf, W., Kossev, A.R., 2006. Motor cortex excitability changes preceding voluntary muscle activity in simple reaction time

619

task. Eur J Appl Physiol 98, 212–219.

620

TE D

618

Ono, T., Shindo, K., Kawashima, K., Ota, N., Ito, M., Ota, T., Mukaino, M., Fujiwara, T., Kimura, A., Liu, M., Ushiba, J., 2014. Brain-computer interface with

622

somatosensory feedback improves functional recovery from severe hemiplegia due

623

to chronic stroke. Front Neuroeng 7, 19. doi:10.3389/fneng.2014.00019

625 626

AC C

624

EP

621

Park, W.H., Li, S., 2011. No graded responses of finger muscles to TMS during motor imagery of isometric finger forces. Neurosci Lett 494, 255–259. Pellicciari, M.C., Miniussi, C., Ferrari, C., Koch, G., Bortoletto, M., 2016. Ongoing

627

cumulative effects of single TMS pulses on corticospinal excitability: An intra- and

628

inter-block investigation. Clin Neurophysiol 127, 621–628.

629

Pfurtscheller, G., Brunner, C., Schlögl, A., Lopes da Silva, F.H., 2006. Mu rhythm

630

(de)synchronization and EEG single-trial classification of different motor imagery

631

tasks. Neuroimage 31, 153–159.

632

Pfurtscheller, G., Lopes da Silva, F.H., 1999. Event-related EEG/MEG synchronization

27

ACCEPTED MANUSCRIPT 633 634 635 636

and desynchronization: basic principles. Clin Neurophysiol 110, 1842–1857. Pfurtscheller, G., Neuper, C., 1997. Motor imagery activates primary sensorimotor area in humans. Neurosci Lett 239, 65–68. Pichiorri, F., De Vico Fallani, F., Cincotti, F., Babiloni, F., Molinari, M., Kleih, S.C., Neuper, C., Kübler, A., Mattia, D., 2011. Sensorimotor rhythm-based

638

brain-computer interface training: the impact on motor cortical responsiveness. J

639

Neural Eng 8, 025020. doi:10.1088/1741-2560/8/2/025020

640

RI PT

637

Pichiorri, F., Morone, G., Petti, M., Toppi, J., Pisotta, I., Molinari, M., Paolucci, S., Inghilleri, M., Astolfi, L., Cincotti, F., Mattia, D., 2015. Brain-computer interface

642

boosts motor imagery practice during stroke recovery. Ann Neurol 77, 851–865. Ramos-Murguialday, A., Broetz, D., Rea, M., Läer, L., Yilmaz, O., Brasil, F.L.,

M AN U

643

SC

641

644

Liberati, G., Curado, M.R., Garcia-Cossio, E., Vyziotis, A., Cho, W., Agostini, M.,

645

Soares, E., Soekadar, S., Caria, A., Cohen, L.G., Birbaumer, N., 2013.

646

Brain-machine interface in chronic stroke rehabilitation: a controlled study. Ann

647

Neurol 74, 100–108.

Reis, J., Swayne, O.B., Vandermeeren, Y., Camus, M., Dimyan, M.A., Harris-Love, M.,

TE D

648

Perez, M.A., Ragert, P., Rothwell, J.C., Cohen, L.G., 2008. Contribution of

650

transcranial magnetic stimulation to the understanding of cortical mechanisms

651

involved in motor control. J Physiol (Lond) 586, 325–351.

653

Reynolds, C., Ashby, P., 1999. Inhibition in the human motor cortex is reduced just before a voluntary contraction. Neurology 53, 730–735.

AC C

652

EP

649

654

Rossini, P.M., Barker, A.T., Berardelli, A., Caramia, M.D., Caruso, G., Cracco, R.Q.,

655

Dimitrijević, M.R., Hallett, M., Katayama, Y., Lücking, C.H., 1994. Non-invasive

656

electrical and magnetic stimulation of the brain, spinal cord and roots: basic

657

principles and procedures for routine clinical application. Report of an IFCN

658

committee. Electroencephalogr Clin Neurophysiol 91, 79–92.

659 660 661 662

Sanger, T.D., Garg, R.R., Chen, R., 2001. Interactions between two different inhibitory systems in the human motor cortex. J Physiol (Lond) 530, 307–317. Sitaram, R., Veit, R., Stevens, B., Caria, A., Gerloff, C., Birbaumer, N., Hummel, F., 2012. Acquired control of ventral premotor cortex activity by feedback training: an

28

ACCEPTED MANUSCRIPT 663

exploratory real-time FMRI and TMS study. Neurorehabil Neural Repair 26,

664

256–265.

667 668 669 670 671

interfaces in neurorehabilitation of stroke. Neurobiol Dis 83, 172–179. Soekadar, S.R., Witkowski, M., Birbaumer, N., Cohen, L.G., 2015. Enhancing Hebbian

RI PT

666

Soekadar, S.R., Birbaumer, N., Slutzky, M.W., Cohen, L.G., 2014. Brain-machine

Learning to Control Brain Oscillatory Activity. Cereb Cortex 25, 2409–2415. Steriade, M., Llinás, R.R., 1988. The functional states of the thalamus and the associated neuronal interplay. Physiol Rev 68, 649–742.

Suffczynski, P., Kalitzin, S., Pfurtscheller, G., Lopes da Silva, F.H., 2001.

SC

665

Computational model of thalamo-cortical networks: dynamical control of alpha

673

rhythms in relation to focal attention. Int J Psychophysiol 43, 25–40.

674

M AN U

672

Takemi, M., Masakado, Y., Liu, M., Ushiba, J., 2015. Sensorimotor event-related

675

desynchronization represents the excitability of human spinal motoneurons.

676

Neuroscience 297, 58–67.

677

Takemi, M., Masakado, Y., Liu, M., Ushiba, J., 2013. Event-related desynchronization reflects downregulation of intracortical inhibition in human primary motor cortex. J

679

Neurophysiol 110, 1158–1166.

TE D

678

Tremblay, F., Tremblay, L.E., Colcer, D.E., 2001. Modulation of corticospinal

681

excitability during imagined knee movements. J Rehabil Med 33, 230–234.

682

EP

680

Vaalto, S., Säisänen, L., Könönen, M., Julkunen, P., Hukkanen, T., Määttä, S., Karhu, J., 2011. Corticospinal output and cortical excitation-inhibition balance in distal

684

hand muscle representations in nonprimary motor area. Hum Brain Mapp 32,

685

1692–1703.

AC C

683

686

Vukelić, M., Gharabaghi, A., 2015. Self-regulation of circumscribed brain activity

687

modulates spatially selective and frequency specific connectivity of distributed

688

resting state networks. Front Behav Neurosci 9, 181.

689

doi:10.3389/fnbeh.2015.00181

690

Young, B.M., Nigogosyan, Z., Walton, L.M., Song, J., Nair, V.A., Grogan, S.W., Tyler,

691

M.E., Edwards, D.F., Caldera, K., Sattin, J.A., Williams, J.C., Prabhakaran, V.,

692

2014. Changes in functional brain organization and behavioral correlations after

29

ACCEPTED MANUSCRIPT 693

rehabilitative therapy using a brain-computer interface. Front Neuroeng 7, 26.

694

doi:10.3389/fneng.2014.00026

695

Yuan, H., Liu, T., Szarkowski, R., Rios, C., Ashe, J., He, B., 2010. Negative covariation between task-related responses in alpha/beta-band activity and BOLD

697

in human sensorimotor cortex: an EEG and fMRI study of motor imagery and

698

movements. Neuroimage 49, 2596–2606.

699

RI PT

696

Zaepffel, M., Trachel, R., Kilavik, B.E., Brochier, T., 2013. Modulations of EEG beta power during planning and execution of grasping movements. PLoS ONE 8,

701

e60060. doi:10.1371/journal.pone.0060060

703

Ziemann, U., Chen, R., Cohen, L.G., Hallett, M., 1998. Dextromethorphan decreases the excitability of the human motor cortex. Neurology 51, 1320–1324.

M AN U

702

SC

700

704

Ziemann, U., Lönnecker, S., Steinhoff, B.J., Paulus, W., 1996. Effects of antiepileptic

705

drugs on motor cortex excitability in humans: a transcranial magnetic stimulation

706

study. Ann Neurol 40, 367–378.

TE D

709

homeostatic metaplasticity. Brain Stimul 1, 60–66.

EP

708

Ziemann, U., Siebner, H.R., 2008. Modifying motor learning through gating and

AC C

707

30

ACCEPTED MANUSCRIPT

Figure captions

711

Figure 1. (a) Experimental paradigm of the screening, resting, and brain-computer interface

712

(BCI) conditions. (b) Experimental setup for the BCI condition. Five electroencephalogram

713

(EEG) electrodes were placed around the right-hand motor area. The event-related

714

desynchronization (ERD) was calculated from EEG signals in online. Transcranial magnetic

715

stimulation (TMS) was applied when the ERD exceeded a predefined threshold during wrist

716

motor imagery. The motor-evoked potential (MEP) was recorded from both flexor carpi

717

radialis (FCR) and extensor carpi radialis (ECR) muscles. Participants received visual

718

feedback of the ERD magnitude, forming the BCI system.

719

Figure 2. MEP waveforms induced by single-pulse and paired-pulse TMS in the resting

720

and the BCI conditions. MEP data shown here were obtained from a single participant

721

of Experiment 1 so the participant performed imagery of wrist extension in the BCI

722

condition. Thin grey lines represent ten MEP traces overlaid per condition, while the

723

thick black line is average of the ten MEP traces. The filled triangles and vertical lines

724

below indicate timing of the test stimulus. The open triangles and vertical lines below

725

indicate the timing of the conditioning stimulus. Herein, the mean single-pulse MEP

726

amplitude was increased in the agonist muscle during motor imagery (ECR Rest: 0.85

727

mV, 5%ERD: 1.21 mV, 15%ERD: 1.30 mV; FCR Rest: 0.51 mV, 5%ERD: 0.59 mV,

728

15%ERD: 0.64 mV). Short-latency intracortical inhibition (SICI), which quantified by

729

the ratio between the mean conditioned MEP and the mean unconditioned MEP, was

730

attenuated as the ERD increased in the agonist muscle (ECR Rest: 30.7 %, 5%ERD:

731

40.3 %, 15%ERD: 78.0 %; FCR Rest: 67.0 %; 5%ERD: 63.5 %, 15%ERD: 58.4 %).

732

Intracortical facilitation (ICF) was not different among conditions in both muscles

733

(ECR Rest: 144.6 %, 5%ERD: 139.3 %, 15%ERD: 132.0 %; FCR Rest: 128.0 %,

734

5%ERD: 117.8 %, 15%ERD: 129.9 %).

735

Figure 3. Mean MEP amplitudes provoked by single and paired-pulse TMS in the resting

736

condition of experiment 1 (a) and experiment 2 (b). The SICI results consisted of paired-pulse

737

TMS at the short inter-stimulus interval (ISI; 2 or 3 ms) that showed smaller MEP amplitude.

AC C

EP

TE D

M AN U

SC

RI PT

710

31

ACCEPTED MANUSCRIPT

Likewise, the ICF results consisted of paired-pulse TMS at the long ISI (10 or 15 ms), which

739

showed larger MEPs. Paired-pulse results demonstrated inhibition by the short ISI protocol

740

and facilitation by the long ISI protocol. Error bars indicate the standard error of the mean. **

741

p < 0.01; *** p < 0.001.

742

Figure 4. Association between the ERD magnitude and TMS measures (the single-pulse MEP

743

amplitude, SICI and ICF) in the BCI condition of experiment 1 (a) and experiment 2 (b).

744

Increase in single-pulse MEP and reduction of SICI was both associated with the amount of

745

ERD. However, the increase in the MEP was not confined to the agonist muscle at the

746

high ERD level. Conversely, the disinhibition was selectively observed in the agonist

747

muscle. ICF showed no modulation by motor imagery. Data shows the mean values for all

748

participants. Error bars indicate standard error of the mean. ** p < 0.01; *** p < 0.001; # p <

749

0.05, different from a corresponding resting condition.

750

Figure 5. Changes in the excitability of agonist M1 representation by actual voluntary

751

movement and motor imagery-based BCI task. (a) According to Nikolova et al. (2006), the

752

MEP increased from approximately 120 ms prior to the onset of muscle activity; then

753

a reduction of SICI occurred. It was speculated that the relatively late reduction in

754

SICI allowed increase in the excitability to be better focused, enabling the

755

generation of appropriate motor commands. (b) In the motor imagery-based BCI

756

task, an increase of MEP was initiated with a weaker ERD than a reduction of SICI.

757

This relationship was similar to that seen with voluntary movement. A motor

758

imagery-based BCI may enable to induce corticomotor activity close to the

759

changes accompanying actual movements, especially when a sufficient ERD was

760

induced.

AC C

EP

TE D

M AN U

SC

RI PT

738

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT