www.elsevier.com/locate/sleep

Sleep Medicine 9 (2008) 177–183

Original article

Cerebral correlates of muscle tone fluctuations in restlesslegs syndrome: A pilot study with combined functional

magnetic resonance imaging and anterior tibialmuscle electromyography

Kai Spiegelhalder a, Bernd Feige a, Dominik Paul b, Dieter Riemann a,Ludger Tebartz van Elst a, Erich Seifritz c, Jurgen Hennig b, Magdolna Hornyak a,*

a Department of Psychiatry and Psychotherapy, University Hospital, Hauptstraße 5, 79104 Freiburg, Germanyb Department of Diagnostic Radiology, Section of Medical Physics, University Hospital, Hugstetter Straße 55, 79106 Freiburg, Germany

c University Hospital of Clinical Psychiatry, Bolligenstraße 111, 3000 Bern 60, Switzerland

Received 23 October 2006; received in revised form 19 March 2007; accepted 30 March 2007Available online 16 July 2007

Abstract

Background: The pathology of restless legs syndrome (RLS) is still not understood. To investigate the pathomechanism of the dis-order further we recorded a surface electromyogram (EMG) of the anterior tibial muscle during functional magnetic resonanceimaging (fMRI) in patients with idiopathic RLS.Methods: Seven subjects with moderate to severe RLS were investigated in the present pilot study. Patients were lying supine in thescanner for over 50 min and were instructed not to move voluntarily. Sensory leg discomfort (SLD) was evaluated on a 10-pointLikert scale. For brain image analysis, an algorithm for the calculation of tonic EMG values was developed.Results: We found a negative correlation of tonic EMG and SLD (p < 0.01). This finding provides evidence for the clinical expe-rience that RLS-related subjective leg discomfort increases during muscle relaxation at rest. In the fMRI analysis, the tonic EMGwas associated with activation in motor and somatosensory pathways and also in some regions that are not primarily related tomotor or somatosensory functions.Conclusions: By using a newly developed algorithm for the investigation of muscle tone-related changes in cerebral activity, we iden-tified structures that are potentially involved in RLS pathology. Our method, with some modification, may also be suitable for theinvestigation of phasic muscle activity that occurs during periodic leg movements.� 2007 Elsevier B.V. All rights reserved.

Keywords: fMRI; EMG; Restless legs syndrome; Periodic leg movements; Pathology; Neurophysiology

1389-9457/$ - see front matter � 2007 Elsevier B.V. All rights reserved.

doi:10.1016/j.sleep.2007.03.021

* Corresponding author. Address: University Medical Center, Depart-ment of Psychiatry and Psychotherapy, University Hospital, Hauptstraße5, 79104 Freiburg, Germany. Tel.: +49 761 270 6501; fax: +49 761 270 6619.

E-mail address: magdolna.hornyak@uniklinik-freiburg.de (M.Hornyak).

1. Introduction

Despite the extensive growth in knowledge over thelast few years, there is as yet no generally accepted con-cept of the pathology of RLS [1,2]. The only functionalmagnetic resonance imaging (fMRI) study on RLS todate was published in 1997 by Bucher et al. [3]. It showedthat periodic leg movement (PLM) is associated withbilateral activation of the red nuclei and the brain stem,

and that sensory leg discomfort (SLD) is related to bilat-eral activation of the cerebellum, with contralateral acti-vation of the thalamus during the wake state. Assimultaneous electrophysiological recordings were notfeasible at that time, the analysis was based on visually

K. Spiegelhalder et al. / Sleep Medicine 9 (2008) 177–183 179

et al. [5]. In the present study, we used the same method-ology with a slightly modified electrode cap in which oneelectrocardiogram (ECG) and one EEG lead werereplaced with bipolar leg EMG electrodes. The modifiedcap includes 28 EEG, 1 ECG, 1 electrooculogram(EOG) and 2 tibialis anterior EMG electrodes.

The EMG electrodes were placed on one leg accord-ing to guidelines for recording PLM activity [10]. Theleg in which patients experienced stronger RLS symp-toms was used for the EMG measurement. If symptomsof equal strength occurred in both legs, electrodes wereplaced on the right leg. As a result, left tibialis EMG wasused in three and right tibialis EMG in four patients.EMG signals were sampled at 500 Hz with a high-passfilter set at 55 Hz. Since the fMRI image acquisitioncaused strong artifacts, only the inter-scan EMG epochswere used. Since artifacts remained up to 404 ms afterthe scan time and began 250 ms before the next scan,sections of EMG data with a length of 1848 ms couldbe used. The first four inter-scan EMG epochs were dis-carded in order to obtain 240 epochs just preceding thecorresponding fMRI scan. Raw EMG data was exam-ined thoroughly by an experienced physician to identifyPLM and to reject epochs with voluntary movements.The following procedure has been developed for thecomputation of a tonic EMG regressor for fMRI analy-sis from the rectified EMG. First, we computed the med-ian of the rectified EMG for each epoch to obtain a‘‘typical’’ rectified EMG value (median of 1 epoch:‘‘Median1’’, see Fig. 1). This value is robust againstshort bursts of EMG activity but not against occasionallonger twitches. In a second step, we detected epoch val-ues strongly deviating from the preceding and subse-quent epochs by comparing with the median across 4epochs before to 4 epochs after the epoch itself (medianof 9 epochs: ‘‘Median9’’). If Median1 was >2 · Med-ian9, Median1 was replaced by Median9. The regressorwas preprocessed by applying the natural logarithm in

Fig. 1. One epoch of rectified raw EMG data. The horizontal linesshow the median of all EMG values for this epoch (Median1, seeSection 2).

order to enhance the distribution of the parameters.At this stage, the tonic EMG was used for correlationwith the SLD (see below). For fMRI analysis, slowshifts were removed from the tonic EMG regressor witha 1/128 Hz high-pass filter (the same filter was applied tothe fMRI voxel time curves) and the regressor was con-volved with a canonical hemodynamic response functionusing the ‘‘waver’’ program of the software packageAnalysis of Functional NeuroImages (AFNI, http://afni.nimh.nih.gov/, [11]).

2.5. Statistics

2.5.1. Analysis of fMRI dataImage analysis was performed using the AFNI soft-

ware. The fMRI data was preprocessed using intra-ses-sion slice alignment and interslice time correction.Spatial alignment with the anatomical scans was per-formed by manual optimization of translation and rota-tion parameters. Anatomical scans were transformed tothe stereotaxic coordinates of Talairach and Tournoux[12] by manually marking the points AFNI requires forthe transformation. Slow shifts in functional voxel timeseries were removed with a 1/128 Hz high-pass filter. Spa-tial smoothing was performed with a full width at halfmaximum of 5 mm. Finally, functional datasets weretransformed into Talairach space by using the transfor-mation of the co-registered anatomical images. Sincebasic motor and somatosensory pathways are stronglylateralized, the functional datasets of the three subjectswith EMG electrodes on the left leg were mirrored atthe sagittal interhemispheric plane (x = 0 plane), causingthe hemispheres to be swapped. The normalisation of thefunctional datasets to the Talairach atlas ensures a preciseorientation of the x = 0 plane in the interhemispheric fis-sure; that is, the hemispheres can be mirrored by simplyswapping the sign of the x-axis. Voxel-by-voxel z-trans-formation was performed for each subject.

Multiple linear regression analysis of the seven con-catenated functional time series data was performedusing the AFNI program ‘‘3dDeconvolve’’ for the ‘‘eyesopen/closed’’ regressor and the tonic EMG regressor.Additionally, the analysis was performed without thedescribed mirroring to account for any potentiallynon-lateralized brain areas correlated with the EMGregressors. In both analyses, an additional head move-ment regressor was used to ensure that correlationsbetween EMG regressors and fMRI signal were notdue to head movements. This is because fluctuations inthe EMG regressors could possibly be correlated withhead movements, which can in turn create artifactualgray level changes at structural boundaries. As headmovement regressor, a displacement length was devel-oped using the three translational alignment parametertime curves of intra-session slice alignment as follows:after shifting each time curve to non-negative values

-4 -3 -2 -1 0 1

02

46

810

Tonic EMG / ln(μV)

Sen

sory

leg

disc

omfo

rt (

SLD

) / p

oint

s

Fig. 2. Correlation between sensory leg discomfort and tonic EMGcalculated by a linear mixed-effects model (p < 0.01). Sensory legdiscomfort was assessed on a 10-point Likert scale (‘‘no leg discom-fort’’ to ‘‘extreme discomfort’’).

180 K. Spiegelhalder et al. / Sleep Medicine 9 (2008) 177–183

by subtracting the minimum, their geometrical meanwas computed for each scan, then filtered and z-trans-formed like the other regressors.

Like the EMG regressors, the ‘‘eyes open/closed’’regressor (1 for eyes open, 0 for eyes closed) was con-volved with a canonical hemodynamic response func-tion. Standard parameters used by AFNI are peakamplitude: 1.0; delay time: 2 s; rise time: 4 s; fall time:6 s; undershoot in relation to the peak: 0.2; and restoretime: 2 s. In order to model the temporal relationshipsprecisely, the parameters were adjusted differently forthe ‘‘eyes open/closed’’ and the EMG regressors. Thedelay time was set to 0.680 s for the ‘‘eyes open/closed’’regressor, because the interslice time correction shiftsthe slices to the average slice time-offset, and the ‘‘eyesopen’’ and ‘‘eyes closed’’ instructions were given at theend of the silent interval, that is, 1.320 s before the aver-age slice time-offset. Given that the middle of the EMGsections precede the fMRI scan by 2.493 s, delay timewas set to 0 s and rise time to 3.507 s for the EMG regres-sors. In the statistical analysis, only voxels that corre-sponded to brain areas across all patients and runswere used (as determined by individual masks calculatedby the AFNI program ‘‘3dAutomask’’). In order toobtain overall statistical significance of p < 0.01 for eachanalysis, statistical significance was set at p < 0.005 usingthe Bonferroni adjustment procedure because two mainanalyses were carried out. For the whole-brain volume,this was achieved by considering cluster sizes above250 mm3 at a voxel threshold of p < 0.001 (Monte-Carlosimulation using the AFNI program ‘‘AlphaSim’’).

The free software ‘‘blender’’ (http://www.blender3-d.org/) was used for three-dimensional visualization ofthe statistically significant clusters together with thethalamus as derived from the Montreal NeurologicalInstitute (MNI) Talairach data set and cortex and brainstem surfaces as segmented from one of the healthy sub-jects examined in Feige at al. [5].

2.5.2. Correlation between SLD and tonic EMG

In order to test the hypothesis of a correlation betweenSLD and tonic EMG using the three SLD assessmentsavailable for each subject, we calculated the mean tonicEMG value within the first and second half of each runand related it to the nearest SLD value in time. A linearmixed-effects model was used to take into account thatboth a fixed EMG effect and a random subject effect maycontribute to the value of sensory discomfort in the legs.

3. Results

3.1. Clinical RLS symptoms and their correlation with the

EMG signal

Patients reported the severity of SLD during theexperiment as follows: at the beginning of the

experiment: 1.7 ± 1.8 (range 0–5); at the end of the ‘‘eyesopen/closed’’ condition: 3.0 ± 2.1 (range 1–7); and atthe end of the experiment: 4.9 ± 3.8 (range 1–10) points.Ratings of the RLS-related leg discomfort correlatednegatively with the tonic component of EMG (n = 28;p < 0.01); that is, the more the muscle tone decreasedthe more intensive the RLS-related sensory discomfortwas (Fig. 2). Possibly due to the fact that the investiga-tion took place in the afternoon and late evening, onlyone patient manifested a small number of leg move-ments [10].

3.2. fMRI analysis

3.2.1. Eyes open/closed paradigm

In the ‘‘eyes open/closed’’ paradigm, the primaryvisual cortex and visual association areas were foundto be deactivated bilaterally when eyes were closed, thusshowing the accuracy of the assessments. A frontocen-tral activation was attributed to an eye movementartifact.

3.3. Tonic EMG regressor, mirrored datasets

Fig. 3 shows the significant partial correlationsbetween the fMRI signal and the tonic EMG regressor.These results stem from the analysis in which datasetsfor the three subjects with EMG electrodes on the leftleg were mirrored so that the laterality (ipsilateral orcontralateral) of the brain activation was consistent forall patients independent of the side of EMG recording.Some clusters mainly comprised white matter or intra-thecal areas and, therefore, probably constitute artifacts.

Fig. 3. Partial correlations between the tonic EMG regressor and the fMRI signal in the analysis with mirrored datasets. Red colour indicatespositive correlations; blue colour indicates negative correlations. (For interpretation of the references to colour in this figure legend, the reader isreferred to the web version of this article.)

K. Spiegelhalder et al. / Sleep Medicine 9 (2008) 177–183 181

Table 1 shows the significant clusters resulting from theanalysis after omission of these artifacts.

3.3.1. CerebrumTonic EMG activity correlated positively with a clus-

ter within regions of the contralateral medial precentraland postcentral gyri (cluster A, cf. Table 1), areasknown to be associated with leg movements and sensoryleg input. Furthermore, one cluster within the postcen-tral gyrus (cluster B) and bilateral areas in the precentralgyrus (clusters C, D, E) were found to be positively cor-related with the tonic EMG. In addition, we found thetonic EMG to be positively correlated with a cluster inthe ipsilateral posterior cingulate (cluster F), with a clus-ter encompassing areas of contralateral cuneus and ipsi-lateral lingual gyrus (cluster G) and with one cluster inthe contralateral precuneus (cluster H). Within the ipsi-lateral cingulate gyrus we found one cluster that wasnegatively correlated with tonic EMG (cluster I).

Table 1Significant clusters correlated with tonic EMG resulting from the analysis w

Label X Y Z Volume/mm3 Max. F

Cerebrum

A �11 �34 +61 828 24.30B +57 �19 +30 484 24.65C +58 +3 �16 422 19.68D �59 �3 +17 375 18.74E +38 �24 +58 281 22.68F +13 �49 +10 1016 20.26G +3 �78 +5 703 34.88H �12 �47 +58 250 17.39I +15 +32 +29 1297 26.03

Cerebellum

J +1 �60 �9 375 18.73K �29 �60 �27 766 19.15L 19 �66 �24 266 15.82

3.3.2. Cerebellum

Three clusters in the cerebellum were negatively cor-related with the tonic EMG regressor: one in the vermisencompassing areas of culmen and declive (cluster J) aswell as two clusters in the cerebellar hemispheres, one ofwhich lies in the contralateral lobus anterior (cluster K)and one in the ipsilateral lobus anterior (cluster L).

3.4. Tonic EMG regressor, non-mirrored datasets

3.4.1. Cerebrum

The analysis of non-mirrored datasets revealed only afew major differences to the analysis of mirrored data-sets. We found one additional cluster encompassingareas of the right precentral gyrus and right middle fron-tal gyrus (x = +31, y = 0, z = +55, maximumF = 20.22, 641 mm3) and one additional cluster in theleft cingulate gyrus (x = +12, y = �42, z = +28,maximum F = 23.42, 1047 mm3) that were positively

ith mirrored datasets

Corr. Areas mainly involved

+ Left postcentral gyrus, left precentral gyrus (contralateral)+ Right postcentral gyrus (ipsilateral)+ Right precentral gyrus (ipsilateral)+ Left precentral gyrus (contralateral)+ Right precentral gyrus (ipsilateral)+ Right posterior cingulate (ipsilateral)+ Left cuneus (contralateral) right lingual gyrus (ipsilateral)+ Left precuneus (contralateral)� Right cingulate gyrus (ipsilateral)

� Vermis of cerebellum� Left hemisphere of cerebellum (contralateral)� Right hemisphere of cerebellum (ipsilateral)

182 K. Spiegelhalder et al. / Sleep Medicine 9 (2008) 177–183

correlated with the tonic EMG regressor. The cluster inthe ipsilateral cingulate gyrus that was negatively corre-lated with tonic EMG in the analysis using mirroreddatasets was now separated into three discrete parts inthe analysis with non-mirrored datasets, one in the rightand two in the left hemisphere.

3.4.2. Cerebellum

In comparison with the analysis using mirrored data-sets, we found one additional cluster in the right cerebel-lar hemisphere (x = +37, y = �58, z = �28, maximumF = 25.59, 1172 mm3) that was negatively correlatedwith the tonic EMG.

4. Discussion

We report on the first study to apply simultaneousrecording of EMG activity and fMRI in patients withRLS. The primary aim of the pilot study was to developa means to determine an EMG regressor for fMRI anal-ysis and to investigate the relationship of tonic EMG,SLD and the fMRI signal.

We hypothesized that the investigation of EMGactivity is essential to the understanding of the pathol-ogy of RLS because symptoms deteriorate during inac-tivity, leg movement leads to partial or total relief ofcomplaints, and most RLS patients show periodic legmovements in wake and sleep, which is considered tobe the typical motor symptom of the disorder [7]. In thispilot study, we investigated muscle tone-related brainactivation and assumed a relationship between changesof tonic EMG activity and symptom severity. For quan-tification of the severity of sensory symptoms we used a10-point Likert scale corresponding to the 100 mmvisual analog scale used in the suggested immobilizationtest (SIT) [13]. SLD increased over the course of thisexperiment in the same way it would in the SIT. Wedetected a negative correlation between the tonic com-ponent of EMG activity and SLD; that is, a decreaseof tonic muscle activity was associated with an increaseof SLD. Our study is the first to produce evidence forone of the main characteristics of the disorder, namelythat RLS-related subjective leg-discomfort increasesduring muscle relaxation at rest. This correlation shouldbe confirmed in a larger patient sample.

We identified several brain areas associated withtonic EMG activity. As we did not investigate healthysubjects in this study, conclusions about the significanceof the results for the pathology of RLS have to be drawncarefully and need to be corroborated in further studies.We found that tonic EMG activity was positively corre-lated with activation in pyramidal (precentral gyrus)motor system as well as in the primary somatosensorycortex. We also detected an activation cluster locatedin the contralateral precuneus that is closely relatedto motor and somatosensory functions (positive

correlation with tonic EMG). A low level of tonicEMG activity was additionally associated with a highlevel of cerebellar activation. Apart from these, wefound activation clusters in the cingulate cortex whichmight be related to the cognitive-emotional evaluationof fluctuations in the symptom severity of RLS. Interest-ingly, an EMG-correlated activation occurred also inthe Brodmann area 18 (Cuneus), a region that is knownto process visual information.

Our study differs from the only other fMRI study inRLS [3] in that we were able to quantitatively record andcorrelate EMG activity during fMRI. Furthermore, inthe prior study [3] the fMRI acquisition was limited tosix slice positions covering the primary motor area,basal ganglia, cerebellum, thalamus, red nucleus, andbrain stem while we performed a whole-brain volumeacquisition. Our findings concerning SLD partly matchthose described by Bucher et al. [3]. Both studies foundcerebellar activation associated with a high level of SLDand, in our study, associated also with a low level oftonic EMG, that is, an ‘‘indirect’’ correlation with a highlevel of sensory discomfort. It is noteworthy that in con-trast to Bucher et al. [3] we did not find any activationpattern in the thalamus corresponding to SLD.

We used two paradigms in the present study. In thefirst paradigm, we ensured signal and data analysisquality by evaluating an ‘‘eyes open/closed’’ conditionthat has been used previously in a study by our group[5]. A head movement regressor has been included inthe analyses as we did not want to mistake varianceinduced by head movement for true activations. Thisseems to be essential for the analysis of EMG-correlatedfMRI data in patients with RLS as we found that thetonic EMG was correlated with head movements inthe present study (for details, contact the correspondingauthor).

The primary aim of our pilot study was to implementcombined EMG-fMRI registration and, in a first step,assess tonic EMG fluctuations and the related BOLDsignal changes. The investigation of phasic muscle activ-ity, like that of periodic leg movements and of the rela-tionship between tonic and phasic muscle activity, is thegoal of a subsequent study. The introduction of the con-cepts of tonic and phasic EMG activity in RLS appearedreasonable to us as the occurrence of the RLS symptomsis closely related to the decrease of muscle tone (e.g.,during rest or during sleep), and short muscle dischargesmanifesting as PLM are considered to be the motorsymptom of the disorder. These concepts are used anal-ogous to the analyses performed in patients with rapideye movement (REM) sleep behavior disorder (e.g.,[14]). In the present study, only one patient presentedleg movements, which was probably due to part of theexperiment taking place in the afternoon, a time atwhich RLS symptoms are usually only of slight or mod-erate severity. Furthermore, a higher awareness and

K. Spiegelhalder et al. / Sleep Medicine 9 (2008) 177–183 183

concentration of patients during the unusual situation inthe MRI scanner might have contributed to decreasedsymptom intensity, which is a well-known phenomenonin RLS [7]. The fMRI signal could have been affected byphasic EMG activity, for instance, by short bursts or bythe above-mentioned leg movements. This phasic EMGactivity probably had a relationship to the fMRI signaland might have slightly increased the error variance inthe regression analysis of the fMRI signal for the tonicEMG regressor. However, the results of this analysiscould only be due to phasic EMG activity in case of astrong correlation between phasic and tonic EMG activ-ity. We developed an operationalisation of phasic EMGactivity (difference between mean and median for eachepoch of rectified EMG) that we did not report in thepresent study because none of our participants hadstrong phasic EMG activity. Pearson’s correlationbetween tonic and phasic EMG regressors wasr = �0.068. From our point of view, this very weak cor-relation coefficient provides evidence for the assumptionthat the fMRI activation in the present study was notdue to fluctuations in phasic EMG activity. To obtainperiodic leg movements and higher symptom intensity,further investigations should be preferably performedin the late evening hours.

Acknowledgments

This study was partly supported by an unrestrictedgrant from the Swiss National Science Foundation anda grant from Boehringer Ingelheim Pharma GmbH &Co. KG, Biberach, Germany.

References

[1] Hening W. The clinical neurophysiology of the restlesslegs syndrome and periodic limb movements. Part I:

diagnosis, assessment, and characterization. Clin Neurophysiol2004;115:1965–74.

[2] Trenkwalder C, Paulus W. Why do restless legs occur atrest?-pathophysiology of neuronal structures in RLS. Neuro-physiology of RLS (part 2). Clin Neurophysiol 2004;115:1975–88.

[3] Bucher SF, Seelos KC, Oertel WH, Reiser M, Trenkwalder C.Cerebral generators involved in the pathogenesis of the restlesslegs syndrome. Ann Neurol 1997;41:639–45.

[4] Richardson MP, Grosse P, Allen PJ, Turner R, Brown P. BOLDcorrelates of EMG spectral density in cortical myoclonus:Description of method and case report. NeuroImage2006;32:558–65.

[5] Feige B, Scheffler K, Esposito F, Di Salle F, Hennig J, Seifritz E.Cortical and subcortical correlates of electroencephalographicalpha rhythm modulation. J Neurophysiol 2005;93:2864–72.

[6] Bartels A, Zeki S. The chronoarchitecture of the human brain–natural viewing conditions reveal a time-based anatomy of thebrain. NeuroImage 2004;22:419–33.

[7] Allen RP, Picchietti D, Hening WA, Trenkwalder C, Walters AS,Montplaisir J. Restless legs syndrome: diagnostic criteria, specialconsiderations, and epidemiology. A report from the restless legssyndrome diagnosis and epidemiology workshop at the NationalInstitutes of Health. Sleep Med 2003;4:101–19.

[8] The International Restless Legs Syndrome Study Group. Valida-tion of the international restless legs syndrome study group ratingscale for restless legs syndrome. Sleep Med 2003;4:121–32.

[9] Mugler JPr, Brookeman JR. Three-dimensional magnetization-prepared rapid gradient-echo imaging (3D MP RAGE). MagnReson Med 1990;15:152–7.

[10] American Sleep Disorders Association. Recording and scoring ofleg movements. ASDA report: Atlas and scoring rules. Sleep1993;16:748–59.

[11] Cox RW. AFNI: Software for analysis and visualization offunctional magnetic resonance neuroimages. Comput Biomed Res1996;29:162–73.

[12] Talairach J, Tournoux P. Co-planar stereotaxic atlas of the humanbrain: 3-dimensional proportional system – an approach to cerebralimaging. New York: Thieme Medical Publishers; 1988.

[13] Montplaisir J, Boucher S, Nicolas A, Lesperance P, Gosselin A,Rompre P, et al. Immobilization tests and periodic leg move-ments in sleep for the diagnosis of restless leg syndrome. MovDisord 1998;13:324–9.

[14] Consens FB, Chervin RD, Koeppe RA, Little R, Liu S, Junck L,et al. Validation of a polysomnographic score for REM sleepbehavior disorder. Sleep 2005;28:993–7.