Brain regions associated with periodic leg movements during sleep in restless legs syndrome

The neural substrates related to periodic leg movements during sleep (PLMS) remain uncertain, and the specific brain regions involved in PLMS have not been evaluated. We investigated the brain regions associated with PLMS and their severity using the electroencephalographic (EEG) source localization method. Polysomnographic data, including electromyographic, electrocardiographic, and 19-channel EEG signals, of 15 patients with restless legs syndrome were analyzed. We first identified the source locations of delta-band (2–4 Hz) spectral power prior to the onset of PLMS using a standardized low-resolution brain electromagnetic tomography method. Next, correlation analysis was conducted between current densities and PLMS index. Delta power initially and most prominently increased before leg movement (LM) onset in the PLMS series. Sources of delta power at −4~−3 seconds were located in the right pericentral, bilateral dorsolateral prefrontal, and cingulate regions. PLMS index was correlated with current densities at the right inferior parietal, temporoparietal junction, and middle frontal regions. In conclusion, our results suggest that the brain regions activated before periodic LM onset or associated with their severity are the large-scale motor network and provide insight into the cortical contribution of PLMS pathomechanism.

LM and PLMS measures. LM and PLMS were manually scored according to the World Association of Sleep Medicine/International RLS Study Group criteria 27 . LM was defined as any anterior tibialis EMG event with the criteria that the onset has EEG activity ≥8 μV in amplitude above resting baseline, the offset is under 2 µV, and the duration is ≤0.5 second. PLMS were defined as a series of four or more consecutive candidate LMs lasting 0.5-10 seconds and separated by intervals of 10-90 seconds during sleep.
PLMSI was defined as the average number of LMs included in PLMS per hour. Inter-movement interval (IMI) indicated the duration between the onset of one LM and the onset of the following LM 19 . The number of IMIs that were 10-90 seconds long and in sequences of ≥3 was divided by the total number of intervals to yield the periodicity index (PI); this index can vary between 0 (absence of periodicity, with none of the intervals between 10 and 90 seconds long) and 1 (complete periodicity, with all intervals between 10 and 90 seconds long).
In the present study, pLM indicated an LM that belonged to a PLMS series, and iLM indicated an LM that did not belong to a PLMS series.

LM selection and EEG preprocessing.
LMs only during non-REM sleep stages were included in the analysis. EEG data were processed using a band-pass filter (1-80 Hz) and a notch filter (60 Hz) to reduce background noise. EEG, EMG, and ECG data were segmented for −20~20 seconds around LM onset. LMs that had another LM in the preceding interval (−20~0 seconds) were rejected to exclude fluctuations in the baseline due to the preceding LM. LMs having signal fluctuations in the preceding period were also excluded by visual inspection. EEG segments exceeding an absolute amplitude of 150 μV were excluded from the analysis. EEG segments with any significant artifacts were also rejected by visual inspection. Independent component analysis was performed to remove EEG signal fluctuations, such as eye movement and skin potential, from the original data.
Time-frequency analysis of EEG in relation to LM. To identify event-related changes in specific EEG spectral characteristics across LM conditions, the time-varying spectral power was estimated using short-time Fourier transform (STFT) with the following parameters: Hanning window and a 1-second time interval with a non-overlap. We did not overlap the time window for clarifying the instantaneous power changes. The magnitudes of the STFT coefficient were corrected with that of the baseline interval according to the following formula: Baseline-normalized power = (power − average power of baseline interval)/(standard deviation of power of baseline interval).
To differentiate EEG changes associated with PLMS from those associated with spontaneous EEG arousal not accompanied by any other event, spectral analysis of spontaneous EEG arousal was performed using the same method used for LMs.
Heart rate changes. The time interval between two successive enhanced R-peaks was used to calculate heart rate. To obtain more highlighted R-peaks, band-pass filtering was conducted. The signal was filtered by fourth-order Butterworth filters with cut-off frequencies between 15 and 20 Hz. The heart rate was computed from the R-R interval, and the baseline-normalized heart rate was obtained using the same formula as that used for baseline-normalized power. EEG source localization. From the scalp-recorded EEG potentials, a three-dimensional reconstructed source distribution for the delta-band activity beginning prior to PLMS onset was obtained using the sLORETA method implemented in Brainstorm version 3.2, an open source toolbox 31 . The solution space was restricted to the cortical gray matter in 15,002 voxels with a spatial resolution of 1 mm in the Talairach coordinates 32 . To group the voxels into functionally similar cortical regions, we employed a total of 148 regions of interest (ROIs) adopted from a standard magnetic resonance imaging (MRI) template obtained from Destrieux et al. This ROI method was implemented for comparing normalized power between pLM and iLM or for correlating current densities with PLMSI. The principal component analysis was performed to determine the dominant activity among all voxels at each ROI. At each ROI, current density was standardized compared to that in the baseline to test for statistical significance.

Statistical analysis.
Clinical data are presented as the medians and 25th-75th interquartile ranges. EEG spectral power or heart rate data at every time point near LM onset were compared with the average of those obtained during the baseline interval using paired samples t-tests. An uncorrected p value of less than 0.01 was considered statistically significant to study data from scalp-recorded EEG channels or ECG heart rate. For source localization, a t-distribution was adopted to show brain regions that were significantly activated compared with the activity observed at baseline. To compensate for multiple comparisons when investigating the EEG sources, the false discovery rate (FDR) was applied to the statistically significant regions of t-distribution versus baseline (FDR-adjusted p value < 0.005) 33 . A correlation analysis between clinical scores, including IRLS, ferritin levels, or PLMSI, and current density for EEG delta power, was performed using Pearson's correlation, and p values of less than 0.05 were considered statistically significant. The data analyses and statistical analyses were performed using MATLAB software version R2014b (Mathworks, Inc., Natick, MA, USA).

Results
Clinical characteristics of RLS patients. The median age of the 15 subjects was 52 [25-75% interquartile range; 40-58] years old, and 13 of them were female. Their epidemiological characteristics, sleep questionnaire scores, and PSG parameters are summarized in Table 1. The median PLMSI was 10.5 [6.5-35.1], and the median value of the mean IMI in PLMS was 36.7 [32.9-39.9]. The total number of pLMs identified in the PSG of 14 subjects was 1,871. Of these, 274 pLMs (14.6%) were excluded according to the condition having preceding another pLM within 20 seconds, and 47 ones were excluded due to artifacts. Therefore, a total of 1,550 pLMs obtained from 14 subjects and 251 iLMs obtained from another 14 subjects were subjected to subsequent analyses.
Changes in heart rate and EEG spectral power preceding LM onset. EEG and ECG signals changed before pLM onset, in line with previous studies. The heart rate began to increase significantly at −1 second before pLM onset. Averaged spectral power over 19 EEG channels increased significantly at around −2 seconds before pLM onset (p < 0.01 compared with the baseline interval) in all frequency bands, especially beta and delta bands ( Fig. 1). In particular, the delta-band power was the most initial and prominent signal to increase (Fig. 2). The delta-band power in the central area was the first signal to increase at −3~−2 seconds prior to pLM in PLMS and was followed by signals in the frontal and parietal areas. Beta-band power was the second prominent signal following delta band and began to increase in the frontal area at −2~−1 seconds (Fig. S1). iLM showed slightly different patterns of changes in EEG spectral power from those of the pLM. Delta power prior to iLM onset increased in the frontal area at −3~−2 seconds followed by in the central and parietal areas (Fig. S2).
EEG sources associated with PLMS. Because the delta band was the most initial and prominent signal before pLM onset at approximately −3 seconds, further analysis employed the delta band to assess the brain regions associated with PLMS. We identified the EEG current sources for the delta bands preceding PLMS using the sLORETA method. Several brain regions, including the bilateral pericentral, dorsomedial prefrontal, and posterior cingulate regions, had higher EEG current densities before pLM onset (Fig. S3A,B). In contrast, cortical activity before   www.nature.com/scientificreports www.nature.com/scientificreports/ iLM onset began to activate in the bilateral dorsal, ventrolateral, medial frontal, and anterior cingulate regions (Fig. S3C,D). Statistically significant current sources of delta power emerged first at -4 ~ -3 seconds before pLM onset and were mainly located in the right pericentral, bilateral dorsolateral prefrontal, and cingulate regions (Fig. 3A). The current sources then spread out into the bilateral frontal and temporal areas with time (Fig. 3B). Current sources of delta power before iLM onset were sparsely detected until −4~−3 seconds (Fig. 3C) and after −3~−2 seconds appeared at the areas encompassing the right pericentral and left parieto-occipital regions (Fig. 3D).

Correlation between PLMSI and cortical activity.
Considering that the significant EEG sources before pLM emerged initially at between −4~−3 or −3~−2 seconds, the cortical activity represented by the current density of delta bands during these time intervals was correlated with PLMSI. Brain regions that had significant correlations with PLMSI were the right inferior parietal, right middle frontal region, and superior temporal sulci at −3~−2 seconds (Fig. 4). No brain regions were correlated with IRLS or ferritin levels.

Discussion
The results of our study reveal that the brain regions associated with PLMS were the large-scale motor network, including pericentral, dorsolateral prefrontal, and cingulate regions, at −4~−3 seconds before each repetitive LM. We also showed that part of default mode network and motor control area, comprising the right inferior parietal, temporoparietal junction, and middle frontal regions, had significant correlations with PLMSI. These results provide new insights into the neural substrates in relation to PLMS in RLS.
We first confirmed that EEG activity preceded LM by a few seconds, in line with the literature. This finding was observed in wide spectral bands from the delta to the beta band, but the delta band was the initial signal that showed the most pronounced increase. Delta activity has been known to be the main spectrum preceding motor movement and is understood to be the signal most related to motor initiation and preparation [28][29][30] . For these reasons, we analyzed the delta bands for EEG source localization in the present study.
EEG topographic data of the delta band showed that the pericentral area was the first to change before pLM onset and that the frontal and parietal areas followed afterwards. In the next step, the EEG current source of the www.nature.com/scientificreports www.nature.com/scientificreports/ delta band was determined using the sLORETA method, which identified the specific brain regions that were activated before PLMS. As a result, we confirmed that the neural substrates preceding pLM were localized at the pericentral, dorsolateral prefrontal, and cingulate regions. The dorsolateral prefrontal cortex has various functions, including motor control 13 . The posterior cingulate cortex, as a central node in the default mode network, plays a role in executive motor control in connection with the frontoparietal control network 34 . A positron emission tomography study demonstrated that the bilateral pericentral and right posterior cingulate regions corresponded to the sites activated during motor imagery in locomotor-related tasks 13 . Increasing evidence in neuroimaging and neuroanatomical studies supports the idea that some subset of anterior cingulate motor area neurons are involved in motor control preparation and execution [35][36][37] . The white matter volume of the anterior cingulate cortex is smaller in patients with RLS [38][39][40] . A functional MRI (fMRI) study revealed that the left primary motor and somatosensory cortices with ventral anterior cingulum were activated during periodic leg movements in RLS patients 41 . Therefore, the activated brain regions associated with PLMS in RLS patients constitute the large-scale motor network, including the sensorimotor cortex with motor control regions, and mostly in agreement with the areas associated with voluntary movement.
It is important to consider the possibility that brain regions associated with PLMS are specific to PLMS or represent only an EEG arousal response. LMs in PLMS are sometimes accompanied by arousal, and in some previous reports in the literature, PLMS has been considered a kind of arousal response 2,8,42 . In this regard, it is necessary to confirm that the EEG source of this study is not simply caused by arousal; therefore, we evaluated spectral power associated with spontaneous arousal during sleep. The delta power associated with spontaneous arousal, which was not associated with any other events, was initially observed at the medial frontal area at −2 ~ 0 seconds and spanned over the left temporal area at 0~2 seconds (Fig. S4). The topographic data obtained for PLMS were initially different but overlapped with the sites activated during spontaneous arousal: the spectral power of the delta band in the central area initially increased at −3 ~ -2 seconds before PLMS, and subsequently, the medial frontal area was activated at −2 ~ -1 seconds (Fig. 2). This medial frontal activation in PLMS may indicate the late arousal response after motor network activation. Hence, the sources of PLMS found in this study were more associated with motor initiation rather than a related arousal response in agreement with studies by Ferri et al. 43 .
Neural substrates related to the severity of PLMS are proposed in the present study including inferior parietal, temporoparietal junction, and middle frontal regions. For example, subjects with more delta activity in the right inferior parietal region tended to have higher PLMSI (Fig. 4). Thus, these cortical regions are likely to be the neural substrates underlying the severity of PLMS. Increasing evidence in neuroimaging and neuroanatomical studies supports the idea that inferior parietal lobule as the default mode network is involved in motor control preparation and execution 13,44,45 . Interestingly, aforementioned fMRI study found that multiple areas, including the right inferior parietal lobule, were activated in periodic leg movements during wakefulness in RLS patients 41 .
The results of the present study suggest that the default mode network with prefrontal motor control pathway may play a role in PLMS. In addition, the regions related to the severity were mainly observed in the right hemisphere, possibly due to hemispheric functional lateralization. Previous studies have shown that the right hemisphere has a specific role in motor control and that there is a right hemisphere preference for the control of slow repetitive movement 46,47 . From our results, it can be suggested that the severity in PLMS might be more associated with the right hemisphere.
Structural MRI data in RLS patients need to be reassessed because cortical structural changes have been mostly investigated in RLS patients. Previous studies using diffusion images or voxel-based morphometry in RLS patients in part support our results of brain regions associated with PLMS. Diffusion tensor imaging of RLS patients revealed that reduced fractional anisotropy was observed in the subcortical areas close to the motor and the somatosensory cortices as well as parts of the limbic system 40 . Connor et at. found decreases in white matter volume in RLS patients in the corpus callosum, anterior cingulum, and precentral gyrus using postmortem imaging analysis 39 . In another study with 28 RLS patients, the same group of researchers found decreases of cortical thickness in bilateral postcentral gyrus and corpus callosum posterior midbody indicating somatosensory pathway 48 . Interestingly, contrary to previous studies, recent multimodal MRI study by Stefani et al. with 87 RLS patients identified increases of gray matter volume of primary sensory cortex bilaterally, the left premotor cortex, and the right parietal inferior lobe which would fit to the concept of adaptive cortical plasticity as a result of increased neuronal activation during PLMS 49 . Further studies are needed to investigate the implication and causal relationship of a cortical role in PLMS with RLS.
Our results have clinical implications in terms of therapeutic targets for neuromodulatory treatment. Various neuromodulatory therapies, including transcranial direct current stimulation and repetitive transcranial magnetic stimulation, have been applied in patients with RLS [50][51][52][53] . One of the most important points to be considered is the therapeutic target to which stimulation should be given. Most studies have given stimulation on the primary motor cortex or supplementary motor area. EEG sources found in the present study may provide insight for new therapeutic targets for neuromodulation in the treatment of RLS with PLMS.
The present study needs to be interpreted in light of several limitations, as follows. First, the number of RLS patients analyzed in this study was small because PSG equipped with 19-channel EEG cannot be routinely performed. Further research with a larger number of patients using extended EEG channels is needed to support our results and the precise role and association of cortical contribution of PLMS. Second, the EEG source localization was relatively sparse because of the limitations inherent to our study methods. The number of EEG electrodes was not high (19 channels), and the standard electrode location coordinates and standard MRI template were applied without using individual data. Third, the role of subcortical regions during PLMS could not be evaluated in the present study, which was based on the electrophysiological analysis of cortical signals. Stereo-EEG studies, if researchers have opportunities such as presurgical evaluations using intracranial electrodes in patients with epilepsy, would therefore be helpful. Fourth, in the present study, PLMS was evaluated only in RLS patients, and various conditions associated with PLMS may have different characteristics 54 . Thus, whether the brain regions Scientific RepoRtS | (2020) 10:1615 | https://doi.org/10.1038/s41598-020-58365-0 www.nature.com/scientificreports www.nature.com/scientificreports/ associated with PLMS in the present study can be generalized to other conditions showing PLMS remains unclear. EEG sources in PLMS need to be validated in other disorders 18 . Finally, pLMs with short intervals of less than 20 seconds were excluded from the analysis due to baseline interval setting, which may have affected the overall results.
Our study suggests that the large-scale motor network may be involved in initiating PLMS in RLS. These factors contribute to our understanding of the pathophysiology of PLMS. Further validation with other neuroimaging investigations is warranted to delineate this association better and to increase our understanding of the pathogenesis of PLMS.

Data availability
Polysomnographic signals and preprocessed data analyzed during the current study are not publicly available due to compliance to privacy. Summary statistics are available from the corresponding author on reasonable request.