Fatigue in Parkinson’s Disease Is Due to Decreased Efficiency of the Frontal Network: Quantitative EEG Analysis
Article information
Abstract
Objective
Fatigue is a common, debilitating nonmotor symptom of Parkinson’s disease (PD), but its mechanism is poorly understood. We aimed to determine whether electroencephalography (EEG) could objectively measure fatigue and to explore the pathophysiology of fatigue in PD.
Methods
We studied 32 de novo PD patients who underwent EEG. We compared brain activity between 19 PD patients without fatigue and 13 PD patients with fatigue via EEG power spectra and graphs, including the global efficiency, characteristic path length, clustering coefficient, small-worldness, local efficiency, degree centrality, closeness centrality, and betweenness centrality.
Results
No significant differences in absolute or relative power were detected between PD patients without or with fatigue (all p > 0.02, Bonferroni-corrected). According to our network analysis, brain network efficiency differed by frequency band. Generally, the brain network in the frontal area for theta and delta bands showed greater efficiency, and in the temporal area, the alpha1 band was less efficient in PD patients without fatigue (p < 0.0001, p = 0.0011, and p = 0.0007, respectively, Bonferroni-corrected).
Conclusion
Our study suggests that PD patients with fatigue have less efficient networks in the frontal area than PD patients without fatigue. These findings may explain why fatigue is common in PD, a frontostriatal disorder. Increased efficiency in the temporal area in PD patients with fatigue is assumed to be compensatory. Brain network analysis using graph theory is more valuable than power spectrum analysis in revealing the brain mechanism related to fatigue.
INTRODUCTION
Fatigue is a common disabling nonmotor symptom in Parkinson’s disease (PD), affecting 37% to 55% of patients with de novo PD [1,2]. Fatigue can be associated with apathy, anxiety, daytime sleepiness, sleep disturbance, and poor quality of life [3]. Among them, fatigue is highly related to poor quality of life [3].
Fatigue is considered an independent nonmotor symptom and may be due to a disease-specific factor because the prevalence of fatigue is not different between de novo and medicated patients, and it has a higher prevalence than in the general population, with a range from 20% to 25% [4]. The mechanism of fatigue in PD needs to be better understood due to its complicated nature involving multifactorial problems such as physiological and psychological factors. One of the main reasons for the lack of understanding of the disease mechanism is the absence of objective measurements to evaluate fatigue. Previous studies have attempted to objectively measure fatigue using various brain magnetic resonance imaging techniques, single photon emission computed tomography, and positron emission tomography [5]. However, these imaging techniques are expensive, involve radiation hazards, and require long inspection times.
Electroencephalography (EEG) is a relatively inexpensive and accessible tool for quantitatively measuring brain function, including brain connection analysis. This study aimed to determine whether EEG could be used to objectively measure fatigue in PD and to explore the pathophysiology of fatigue in PD by EEG analysis. To achieve this goal, we quantitatively analyzed EEG findings between PD patients with and without fatigue.
MATERIALS & METHODS
Patients
We retrospectively reviewed the medical records of 32 de novo PD patients who underwent EEG between November 2015 and October 2021. PD was diagnosed according to the UK Brain Bank criteria. We collected the following information from patients: Hoehn and Yahr stage (H&Y stage), Unified Parkinson’s Disease Rating Scale (UPDRS) Part I–III, the Korean Version of the Mini-Mental State Examination (K-MMSE), the Korean Version of the Montreal Cognitive Assessment (K-MoCA), education level, disease duration, Beck Depression Inventory (BDI), Fatigue Severity Scale (FSS), and the Korean version of the Nonmotor Symptoms Scale (NMSS) [6]. Depression was noted if the BDI score was 20 or higher [1], and fatigue was based on an FSS score of four or more [1]. The exclusion criteria were cognitive dysfunction (K-MMSE score less than the 2.5th percentile for age and educational-appropriate norm), and a history of stroke, epilepsy, pertinent head injury, or significant debilitating medical conditions [7]. The Institutional Review Board of Hallym University Dongtan Sacred Heart Hospital approved this study (IRB No. 2021-07-016). The need for informed consent was waived because of the retrospective nature of the study. All methods were performed in accordance with the Declaration of Helsinki and followed relevant guidelines and regulations.
EEG recording
Participants were asked to be relaxed and awake during the recording. Eyes-closed, resting-state EEG data were recorded for at least five minutes (Comet plus; Grass, West Warwick, RI, USA). The International 10–20 system was used for electrode placement, and 19 channels with referential montage were used: FP1, FP2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, and O2. One additional electrode applied to the left mastoid was used as a ground, and another electrode applied to the right mastoid was used as a reference. The EEG data were sampled at 200 Hz, and the bandpass filter was set between 1.0 Hz and 70 Hz. The impedance of every electrode was optimized to below 10 kΩ. Resting-state EEG was performed for 10 minutes. We selected three minutes of eyes-closed and artifact-free data based on visual inspection for further analysis. The mean time interval between clinical evaluation and EEG recording was 1.0 ± 2.4 days.
Data preprocessing
Quantitative EEG (QEEG) processing and group analyses were conducted using iSyncBrain (iMediSync Inc., Seoul, Korea; https://isyncbrain.com), a cloud-based, artificial intelligence EEG analysis platform [8]. EEG preprocessing was performed to denoise all the data and minimize the effects of artifacts. During the first stage of EEG preprocessing, the signals were sampled at 250 Hz and filtered with a bandpass filter in the 1–45.5 Hz range. The EEG signals were then passed through a notch filter in preparation for downstream processing, including re-referencing (using common average reference), wrong epoch rejection (using artifact subspace reconstruction), and adaptive mixture independent component analysis (AMICA). AMICA helps provide a generic, unsupervised approach to identifying and modeling changes in EEG dynamics [9]. Finally, artifacts identified via electromyography, cardiac signals, body movement, and electrooculography were removed to yield the clean QEEG normative data. To enhance the performance of iSyncBrain, we added an extra denoising process that included using a bandpass filter from 1 Hz to 45.5 Hz, re-referencing using a common average reference, and determining artifacts by AMICA [10].
Data processing and analyses
All EEG preprocessing processes, sensor-level data, source-level data calculations, and extractions were performed using a cloud-based EEG analysis platform (iSyncBrain). The power spectral density of the EEG rhythms was computed using Fast Fourier transform (FFT) analysis with 0.25 Hz of frequency resolution using iSyncBrain. Then, the signal was decomposed into the following frequency bands: delta (1–4 Hz), theta (4–8 Hz), alpha 1 (8–10 Hz), alpha 2 (10–12 Hz), beta 1 (12–15 Hz), beta 2 (15–20 Hz), beta 3 (20–30 Hz), and gamma (30–45 Hz). The alpha bands and beta bands were divided for more granular frequency analysis [11-13].
In the power spectrum analysis, the absolute and relative power of the seven frequency bands were measured using a FFT. Absolute band power is a spectral band power based on FFT, here provided by iSyncBrain. The relative power is the ratio of the power in a given frequency band to the power in the total frequency range. The topographical grouping of the electrodes was defined as frontal (Fp1, Fp2, F7, F3, Fz, F4, and F8), central (C3, Cz, and C4), temporal (T3, T4, T5, and T6), parietal (P3, Pz, and P4), or occipital (O1 and O2).
Source reconstructions were performed with sLORETA [14] using the Colin 27 Head model [15] with 68 ROI segmentations based on the Desikan-Killiany atlas [16].
In the brain network analysis, we used a popular graph theory approach that provides quantitative undirected binary metrics for measuring brain connectivity networks [17,18]. A network is a mathematical representation of a real-world complex system and is defined by a collection of nodes (vertices) and links (edges) between pairs of nodes [19]. Each ROI from the sLORETA analysis was considered a node. For the network analysis, we divided the brain into six regions and performed a network analysis for each region. We adopted the most commonly used network measures classified into global and local measurements. Global measurements included the global efficiency (GE), characteristic path length (CPL), clustering coefficient (CCO), and small-worldness (SW). The local measurements included the local efficiency (LE), degree centrality (DC), closeness centrality (CCE), and betweenness centrality (BC) [19,20]. Detailed explanations of these measurements are provided elsewhere [19-22]. Briefly, GE is the efficiency of remote information transmission in a network [23]. It indicates the average effect of the relevant brain network and its overall information transmission capacity [22]. The CPL measures the connectivity level of nodes in the network; lower values indicate more efficient information transfer between two nodes and better connectivity of the network [22]. The CCO is the ratio of the number of current edges between adjacent nodes to all possible connected edges; a higher value indicates a more highly segregated network. SW is the ratio of the normalized CCO to the normalized CPL and shows networks that are significantly more clustered than random networks, i.e., the density and efficiency of information exchange in the networks. The LE is the average efficiency of information transfer between a node and its nearest neighbors [24]. A higher LE means that the information transmission of adjacent nodes is more efficient and that the network is less differentiated [22,25]. DC is the number of adjacent nodes; a high DC means a more central node. CCE is the closeness of a node in the network to all other nodes (i.e., a measure of the relative importance of a node) [26]. Higher values of a specific node mean that it is close to all other nodes. BC is the tendency of a single node to be more central than all other nodes and the quantification that acts as a bridge between two other connected nodes along its shortest path length [27]. A higher BC value indicates a node closer to the center of the network. The brain network nodes represent the brain region, and the edges indicate connections between neurons or brain regions [20].
Statistical analysis
The statistical analysis was designed to examine the clinical characteristics and EEG data measured from 19 channels in two groups of PD patients: those without fatigue and those with fatigue. We used independent t tests (or Mann–Whitney U tests) and Fisher’s exact tests to compare the two groups, as appropriate. This analysis included a normality test using the Kolmogorov–Smirnov test for each group (Supplementary Tables 1 and 2 in the online-only Data Supplement). We also used one-way analysis of covariance to compare the EEG data between the two groups after adjusting for the BDI score. We did not find significant differences between the two groups in terms of the BDI score, but it in a prior study, it was argued that the BDI score was strongly correlated with fatigue in PD patients [3], and it is also worth considering that the p value approached significance (p = 0.059) (Table 1). The EEG data, obtained by averaging the power spectrum within each five lobe regions, and by averaging network activity across six lobe regions, were compared between PD patients without fatigue and those with fatigue. Significance was determined at a p value <0.05. Significance levels were Bonferroni-corrected for multiple comparisons: p value <0.05/5 (0.01) for the power spectrum analysis and <0.05/6 (0.008333) for the network analysis. Demographic and clinical data were analyzed using IBM SPSS 25 statistics (IBM Corp., Armonk, NY, USA).
Clinical features of patients without fatigue and with fatigue
RESULTS
Demographic and clinical characteristics
The demographic and clinical features of the drug-naïve PD patients with and without fatigue are shown in Table 1. No significant differences were found between the PD patients with and without fatigue in terms of the age at evaluation, sex, disease duration, K-MMSE score, K-MoCA score, years of education, BDI score, H&Y stage, UPDRS I-III score, or total NMSS score at evaluation (all p > 0.05). PD patients with fatigue had significantly greater FSS scores than PD patients without fatigue (p < 0.001).
EEG data analysis
Power spectrum analysis
In terms of absolute and relative power, for all frequency bands, there were no significant differences between PD patients with and without fatigue (all p > 0.02) (Figures 1 and 2).
Power spectrum analysis of patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). Topographic maps (A) and box plot (B) of seven electroencephalography frequency bands for absolute power. The box plot shows the range, median, and interquartile range. There were no significant differences between Parkinson’s disease patients with and without fatigue (all p > 0.05, Bonferronicorrected).
Power spectrum analysis of patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). Topographic maps (A) and box plot (B) of seven electroencephalography frequency bands for relative power. The box plot shows the range, median, and interquartile range. There were no significant differences between Parkinson’s disease patients with and without fatigue (all p > 0.09, Bonferroni-corrected).
Network analysis
Global measurements
PD patients without fatigue showed significantly greater GE for the theta bands in the left frontal area (p < 0.0001), whereas lower GE was observed for the alpha1 band in the left temporal region (p = 0.0001) (Figure 3). The CPL for the delta band was significantly lower in the left temporal area in PD patients without fatigue (p = 0.0059). The CPL for the alpha1 band was significantly greater in the left temporal area in PD patients without fatigue (p = 0.0005) (Figure 4). There were no significant differences in CCO or SW for any of the frequency bands (all p > 0.04) (Figures 5 and 6).
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The global efficiency of seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 < ***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The characteristic path length of seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 < ***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The clustering coefficient for seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 < ***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The small-worldness of seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 < ***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Local measurements
There were no significant differences in the LE for any of the frequency bands between PD patients without fatigue and PD patients with fatigue (all p > 0.01) (Supplementary Figure 1 in the online-only Data Supplement). For the alpha1 band, DC in PD patients without fatigue was greater in the lower part of the left temporal area (p < 0.001) (Supplementary Figure 2 in the online-only Data Supplement). CCE in the theta band was greater in the right parieto-occipital area (p = 0.0063) in PD patients without fatigue. CCE in the alpha1 band was lower in the left temporal area in PD patients without fatigue (p = 0.0007) (Supplementary Figure 3 in the online-only Data Supplement). Additionally, in PD patients without fatigue, BC for the delta band was greater in the left frontal area than in PD patients with fatigue (p = 0.0011) (Supplementary Figure 4 in the online-only Data Supplement).
DISCUSSION
Our study demonstrated several differing features between PD patients with and without fatigue via QEEG analysis, especially when using graph theory. The main findings were as follows: 1) We did not find significantly different features in the power spectrum analysis between PD patients without fatigue and PD patients with fatigue. 2) In global measurements, the efficient network according to the frequency bands seemed to vary by brain region. In PD patients without fatigue, the theta band showed a better overall information transmission capacity in the frontal region (GE values), and the delta band showed a more efficient network, whereas the alpha1 band showed less effective connectivity in the temporal area (GE and CPL values). 3) Regarding local measurements, the results were similar to those of the global network analysis. In PD patients without fatigue, more centrality was observed in the frontal areas, and less centrality was observed in the left temporal areas. For the theta band, DC was greater in frontal regions (i.e., more central areas). CCE was greater for the theta band (i.e., relatively more important) in the parieto-occipital area and for the delta band in the temporal area in PD patients without fatigue. The alpha1 band was less prominent in the temporal areas (DC and CCE values).
There have been no prior power spectrum analysis studies in PD patients with fatigue. Although our results revealed no differences between PD patients without fatigue and those with fatigue, a meta-analysis on the influence of fatigue on brain activity revealed increases in overall EEG activity (particularly in the theta and alpha bands) in the fatigued state [28]. The delta and alpha1 absolute powers in PD patients without fatigue seemed to be lower in all brain regions (Figure 1). It is not known why the EEG spectrum differs between PD patients with and without fatigue. Because the decreased power is known to be associated with increased cortical activity or an alert state [28,29], the lower EEG power in PD patients without fatigue may be related to the more active cerebral function in the awake state. The higher EEG power (i.e., synchronization) in PD patients with fatigue indicates an idle state (i.e., a less active state).
Our network analysis using graph theory showed that the theta band in the frontal area was related to fatigue in PD patients because the network of the frequency band in the frontal area was less efficient in PD patients with fatigue than in PD patients without fatigue. The theta band in the frontal area is related to the recruitment of cognitive control during monitoring, the management of memory function, and error processing [30]. These cognitive functions are reportedly related to fatigue in healthy people and people with chronic fatigue syndrome [31,32] or PD [33]. Previous brain network studies support our results. Theta band functional coupling was weakened in PD patients with fatigue [20,34]. However, because there were also contradictory results in other studies, possibly due to different study designs [20], more research may be needed.
It seemed that the delta band was also associated with fatigue because of the lower connectivity in the temporal area and the lower centrality in the frontal and temporal areas in PD patients with fatigue. We think that the delta band findings should be interpreted with caution because of possible EEG artifact contamination and previous contradictory reports. Some studies have shown that increased EEG power is correlated with fatigue, but others have not [35-37].
It seemed that the network activity for the alpha band in the temporal regions was a compensatory mechanism for fatigue in our study. A previous brain source analysis study may support our results. They showed greater source-current activity in the frontal-temporal-parietal regions in chronic fatigue syndrome patients [38]. Since alpha bands are involved in attention and memory processes [39], there may be an association between the efficiency of these brain functions and the fatigue process. Generally, alpha- and beta-band functional coupling are weakened in individuals with fatigue [39,40], but our results were not significant after multiple comparison correction.
We do not know the exact reason why there was no difference between patients without fatigue and those with fatigue in the power spectrum analysis but was a difference in the network analysis. A previous study also revealed significant differences in network measures between PD patients and healthy controls but this difference was not detected via power spectral analysis. They assumed that changes in the power spectrum were probably not evident in the early stages of PD because they studied de novo, early-stage PD patients, similar to our study [41]. Generally, power spectrum analysis shows cerebral activity in localized brain areas, and network analysis shows long-range communication among several brain areas [42]. Traditionally, localized brain areas have been investigated for neurologic symptoms, but network analysis of connected regions has gained much interest because it better explains neurological symptoms [43,44]. Brain connectivity using graph theory describes the characteristics of brain architecture, such as efficient communication between two brain areas. Another possibility is that fatigue in PD patients is less related to local brain oscillations and is a problem in the brain network. Brain oscillations play an essential role in both local activity and long-range communication. Many oscillatory bands in cerebral networks range from 0.05 Hz to 500 Hz, and their physiological roles are poorly understood. The function of local oscillations seems to differ from that of long-range communication, even in the same frequency bands [45].
Our study has some limitations. First, the number of patients studied was relatively small. Nevertheless, we think that our EEG results, especially the network analysis (our examined parameters), showed consistent findings. Second, we did not perform detailed sleep evaluations to exclude patients with severe sleep problems. Sleep problems can be a confounding factor because they can affect fatigue. However, there was no significant difference in sleep scores on the NMSS between PD patients without and with fatigue (numbers 3, 5, and 6; p > 0.100 in the Supplementary Table 3 in the online-only Data Supplement). Third, it can be argued that the sample size was biased toward women, especially in the PD population. However, there was no significant difference in the sex ratio between the PD patients with and without fatigue. In addition, the female-to-male ratio of the incidence and prevalence of PD in Korea is higher (1.4:1 and 1.6:1, respectively) [33]. This finding is different from that of Western population studies showing that PD is more predominant in men but is similar to that of many Asian population studies [33].
In conclusion, our study showed that EEG variables can be biomarkers of fatigue in PD, and that PD patients with fatigue have less efficient brain networks in the frontal area than PD patients without fatigue. These findings help us understand the brain mechanisms associated with fatigue in PD. In addition, brain network analysis using graph theory is more valuable than power spectrum analysis in revealing the brain mechanisms related to fatigue.
Supplementary Materials
The online-only Data Supplement is available with this article at https://doi.org/10.14802/jmd.24038.
Normality test results of power spectrum analysis
Normality test results of network analysis
Domain 2 (sleep/fatigue) of Non-Motor Symptom Scale
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The local efficiency of seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 <***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The degree centrality of seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 < ***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The closeness centrality of seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 < ***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Comparison of the network analysis using graph theory between patients with Parkinson’s disease with fatigue (YF) and without fatigue (NF). The betweenness centrality of seven electroencephalography frequency bands is shown. The box plot shows the range, median, and interquartile range. Data other than 25% to 75% of the value are displayed as diamond-shaped dots (0.00167 < *p ≤ 0.00833, 0.000167 < **p ≤ 0.00167, 0.0000167 < ***p ≤ 0.000167, ****p ≤ 0.0000167, Bonferroni-corrected).
Notes
Conflicts of Interest
The authors have no financial conflicts of interest.
Funding Statement
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (NRF-2021R1F1 A1051738). This research was supported by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: RS-2023-00266948).
Author Contributions
Conceptualization: Min Seung Kim, Suk Yun Kang. Data curation: Min Seung Kim, Sanguk Park, Ukeob Park, Suk Yun Kang. Formal analysis: Min Seung Kim, Sanguk Park, Ukeob Park, Suk Yun Kang. Funding acquisition: Suk Yun Kang. Investigation: Min Seung Kim, Sanguk Park, Suk Yun Kang. Methodology: Min Seung Kim, Sanguk Park, Ukeob Park, Suk Yun Kang. Project administration: Suk Yun Kang. Resources: Min Seung Kim, Suk Yun Kang. Software: all authors. Supervision: Suk Yun Kang. Validation: Min Seung Kim, Sanguk Park, Ukeob Park, Suk Yun Kang. Visualization: Sanguk Park, Ukeob Park. Writing—original draft: Min Seung Kim, Sanguk Park, Suk Yun Kang. Writing—review & editing: Ukeob Park, Seung Wan Kang, Suk Yun Kang.
Acknowledgements
None