Article Images (0)
×![]()
featured_play_listSave article to folder
X
Create new folder
Full Text
PDF
Table of Contents
Abstract
1. Introduction
2. Materials And Methods
2.1. Subject Characteristics
2.2. MEG Recordings
2.3. MEG Data Analysis
2.3.1. Preprocessing
2.3.2. Time-Delay Embedded HMM (TDE-HMM) Algorithm
2.3.3. State Parameters Analysis And Spectral Analysis
3. Results
3.1. Impact Of L-Dopa On Motor Cortical β Oscillations In The Views Of HMM And Spectrum
3.2. Impact Of STN DBS On Motor Cortical β Oscillations In The Views Of HMM And Spectrum
4. Discussion
5. Conclusion
Acknowledgment
Abstract
Background: Parkinson’s disease (PD) is a movement disorder caused by dopaminergic neurodegeneration. Both Levodopa (L-dopa) and Subthalamic Deep Brain Stimulation (STN-DBS) effectively alleviate symptoms, yet their cerebral effects remain under-explored. Understanding these effects is essential for optimizing treatment strategies and assessing disease severity. Magnetoencephalogram (MEG) data provide a continuous time series signal that reflects the dynamic changes in brain activity. The hidden Markov model (HMM) can capture and model the temporal features and underlying states of the MEG signal to extract potential brain states and monitor dynamic changes. In this study, we employed HMM to investigate the cortical mechanism underlying the treatment of PD patients using MEG recordings. Methods: 21 PD patients treated with medication underwent MEG recording in both L-dopa medoff and medon conditions. Additionally, 11 PD patients receiving STN-DBS treatment underwent MEG recording in both dbsoff and dbson conditions. The MEG data were segmented into four states by Time-delay embedded Hidden Markov Model (TDE-HMM) algorithm. The state parameters including Fractional Occupancy (FO), Interval Times (IT), and Life Time (LT) for each state and power spectrum of β band were analyzed to study the effects of L-dopa and STN-DBS treatment respectively. Results: L-dopa significantly increased the motor state of HMM and power in the motor area of both high β (21–35 Hz) and low β (13–20 Hz); the motor state of high β in medoff were correlated with the Unified Parkinson’s Disease Rating Scale III (UPDRS III). Conversely, DBS significantly diminishes the motor state of HMM and power in motor area of high β oscillations. The score changes of tremor and limb rigidity after DBS treatment were significantly correlated with the changes of motor state of high β. Conclusions: This study demonstrates that L-dopa and STN-DBS exert differing effects on β oscillations in the motor cortex of PD patients, primarily in high β band. Understanding these distinct neurophysiological impacts can provide valuable insights for refining therapeutic approaches in motor control for PD patients.
The effect of L-dopa and DBS on cortical oscillations in Parkinson’s disease analyzed by hidden Markov model algorithm Kunzhou Wei a,b,1, Hang Ping a,b,1, Xiaochen Tang d, Dianyou Li c, Shikun Zhan c, Bomin Sun c, Xiangyan Kong a,b,* , Chunyan Cao c,* a School of Electrical Engineering and Computer Science, Ningbo University, Ningbo 315211, China b The Institute for Future Wireless Research (iFWR), Ningbo University, Ningbo 315211, China c Department of Neurosurgery, Affiliated Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China d Shanghai Mental Health Center, Shanghai, China A R T I C L E I N F O Keywords: PD L-dopa DBS MEG TDE-HMM Motor cortex A B S T R A C T Background: Parkinson’s disease (PD) is a movement disorder caused by dopaminergic neurodegeneration. Both Levodopa (L-dopa) and Subthalamic Deep Brain Stimulation (STN-DBS) effectively alleviate symptoms, yet their cerebral effects remain under-explored. Understanding these effects is essential for optimizing treatment strategies and assessing disease severity. Magnetoencephalogram (MEG) data provide a continuous time series signal that reflects the dynamic changes in brain activity. The hidden Markov model (HMM) can capture and model the temporal features and underlying states of the MEG signal to extract potential brain states and monitor dynamic changes. In this study, we employed HMM to investigate the cortical mechanism underlying the treatment of PD patients using MEG recordings. Methods: 21 PD patients treated with medication underwent MEG recording in both L-dopa medoff and medon conditions. Additionally, 11 PD patients receiving STN-DBS treatment underwent MEG recording in both dbsoff and dbson conditions. The MEG data were segmented into four states by Time-delay embedded Hidden Markov Model (TDE-HMM) algorithm. The state parameters including Fractional Occupancy (FO), Interval Times (IT), and Life Time (LT) for each state and power spectrum of β band were analyzed to study the effects of L-dopa and STN-DBS treatment respectively. Results: L-dopa significantly increased the motor state of HMM and power in the motor area of both high β (21–35 Hz) and low β (13–20 Hz); the motor state of high β in medoff were correlated with the Unified Parkinson’s Disease Rating Scale III (UPDRS III). Conversely, DBS significantly diminishes the motor state of HMM and power in motor area of high β oscillations. The score changes of tremor and limb rigidity after DBS treatment were significantly correlated with the changes of motor state of high β. Conclusions: This study demonstrates that L-dopa and STN-DBS exert differing effects on β oscillations in the motor cortex of PD patients, primarily in high β band. Understanding these distinct neurophysiological impacts can provide valuable insights for refining therapeutic approaches in motor control for PD patients.
1. Introduction
Parkinson’s disease (PD) is a progressive brain degenerative condition with some typical motor symptoms such as slowness, tremor, rigidity, gait disturbances, and postural instability. A crucial element for optimal information processing in the Basal Ganglia (BG)is the dopaminergic innervation of the striatum, stemming from the neurons in the substantia nigra. PD is marked by the degeneration of these dopaminergic neurons in the substantia nigra, which leads to the hallmark movement disorders (Bartels and Leenders, 2009; Chan et al., 2010; McAuley, 2003; Rinne, 1993). While treatments like Levodopa (L-dopa) and Subthalamic Deep Brain Stimulation (STN-DBS) can notably alleviate these symptoms. During DBS surgeries, Local Field Potential (LFP) recordings reveal overly synchronized β oscillations in STN (Rosa et al., 2011; Rossi et al., 2008; Weinberger et al., 2006). Notably, the amplitude (power magnitude) of β oscillations in STN * Corresponding authors. E-mail addresses: kongxiangyan@nbu.edu.cn (X. Kong), ccy40646@rjh.com.cn (C. Cao). 1 These authors are joint first author. Contents lists available at ScienceDirect NeuroImage journal homepage: www.elsevier.com/locate/ynimg https://doi.org/10.1016/j.neuroimage.2024.120992 Received 22 May 2024; Received in revised form 13 November 2024; Accepted 30 December 2024 NeuroImage 305 (2025) 120992 Available online 30 December 2024 1053-8119/© 2025 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC license ( http://creativecommons.org/licenses/bync/4.0/ ). directly correlates with the degree of the patient’s symptoms of dyskinesia in PD (Piña-Fuentes et al., 2020; Weinberger et al., 2006). Both L-dopa and DBS have demonstrated their capacity to counteract this pathological activity within the STN (Giannicola et al., 2010; Ramirez Pasos et al., 2019; Weinberger et al., 2006). While there’s a general agreement about the role of subcortical β synchronization in PD, the scientific community hasn’t reached a consensus concerning the significance of β oscillations in sensorimotor cortex of PD patients. Given these uncertainties, it becomes imperative to delve deeper into understanding the impacts of L-dopa and deep brain stimulation therapy on the sensorimotor cortex in PD patients. Magnetoencephalography (MEG) offers remarkable temporal and spatial resolution, which is an instrumental tool that non-invasively captures the faint magnetic field signals generated by the brain. This includes signals from the cerebral cortex (Cohen, 1968, 1972) as well as subcortical regions (Boon et al., 2017). Due to its capabilities, MEG is invaluable for scrutinizing oscillatory activity related to PD (Baillet, 2017; van Wijk et al., 2016). Recent studies have leveraged MEG in their exploration of PD. A longitudinal examination of PD patients revealed a surge in power for “slower” frequencies (specifically, the theta and alpha band). Conversely, there was a decline in power for “faster” frequencies, namely the β and gamma bands (Stoffers et al., 2008), (Cao et al., 2015). Some research indicates that upon administering L-dopa, β power in the cerebral cortex of PD patients experiences an uptick (Heinrichs-Graham et al., 2014; Melgari, 2014). Other studies associate Parkinson’s-related dysfunctions with heightened cerebral cortex β levels (Hall et al., 2014; Pollok et al., 2012). However, research by Stoffers et al. indicated that in advanced stages of PD, β oscillation power drops remain relatively unaffected by L-dopa administration (George et al., 2013; Stoffers et al., 2008). Researchers have traditionally regarded the β band as a unified functional entity. Nevertheless, recent evidence indicates that a more precise characterization would involve recognizing two distinct functional sub-bands within the β frequency range: low β (13–20 Hz) and high β (21–35 Hz) (van Wijk et al., 2016). Both L-dopa and STN-DBS have been shown to modulate low β signals within the STN specifically (van Wijk et al., 2016). Additionally, the phase-amplitude coupling within the STN is most pronounced in the low β range (van Wijk et al., 2016). Intriguingly, coherence between the supplementary motor area and STN is evident in the high β frequency range (21–30 Hz) but is absent in the low β range (13–21 Hz). This coherence aligns with the fiber densities of the `hyper-direct pathway’ connecting these structures (Oswal et al., 2021). Considering these distinctions, it is necessary to investigate the high β band and the low β band in somatomotor cortex individually to provide insights for brain-computer interface applications in motor control. The Hidden Markov Model (HMM) method serves as a powerful tool for analyzing sequence data, particularly in identifying the active state of brain networks across various time points. For example, HMM was used to segment EMG signals of PD patients into active and inactive regions, utilizing discrete wavelet transform for feature extraction to achieve accurate classification (Bengacemi et al., 2021). Similarly, Safi et al. employed HMM to differentiate between healthy individuals and PD patients using Balance Control Data, attaining accuracy levels of up to 98% (Safi et al., 2024). Roth et al. (2021) demonstrated the efficacy of an HMM-based stride segmentation method in PD patients, showing superior performance compared to traditional dynamic time warping (DTW) methods. Furthermore, Severson et al. (2020) developed a personalized input-output HMM to model disease progression, integrating individual drug effects and disease states in a time series framework. By leveraging HMM to brain activity data, researchers can effectively capture the dynamic nature of β oscillations, thus unveiling insights into the underlying mechanisms driving these oscillations (Heideman et al., 2020; Zhang et al., 2021). In this study, we utilized a HMM algorithm to segment distinct activation states across different brain regions and examine the power spectrum of β oscillations in different cortical regions. We hypothesize that the motor area activation identified through this method is correlated with the clinical status of the patients, and that L-dopa and STN-DBS have differing effects on cortical β oscillations in individuals with PD.
2. Materials and methods
2.1. Subject characteristics
We separately enrolled 21 PD patients under medication treatment before DBS operation and 11 PD patients receiving STN-DBS treatment (Table 1) from the Functional Neurosurgery Department of Ruijin Hospital. The data pertaining to the 21 PD patients undergoing pharmacological treatment were the same data used in our previous article (Cao et al., 2020). All participants met the clinical diagnostic criteria for possible PD as established by the UK Parkinson’s Disease Association Brain Bank (UK-PDSBB) (Gibb, 1988). Outside of PD, none of the patients exhibited any notable medical, neurological, or psychiatric conditions. All participants gave their written consent for involvement in the study. The research was sanctioned by the Ethics Committee of Ruijin Hospital, affiliated with Shanghai Jiao Tong University School of Medicine, adhering to the guidelines of the Declaration of Helsinki (Cao et al., 2020).
2.2. MEG recordings
MEG recordings were conducted using a 306-channel, whole-head Vector View MEG system (Elekta Oy, Helsinki, Finland) within a magnetically shielded room (Euroshield, Eura, Finland). To ascertain head position within the MEG system, the locations of bilateral preauricular and nasion fiducial points were digitized. As outlined in our previous study (Hall et al., 2014), PD patients on medication underwent MEG recordings after discontinuing parkinsonism medication for a minimum of 12 h (referred to as the `medoff’ state) and again 1 to 2 h after L-dopa administration (termed the `medon’ state). After discontinuing parkinsonism medication for a minimum of 12 h, PD patients undergoing DBS treatment were assessed with their impulse generator (IPG) activated (dbson state) and then 10 min following IPG deactivation (dbsoff state) (van Wijk et al., 2016). Throughout the recording, patients laid supine with their eyes open for 10 min. Raw MEG data were subjected to a band-pass filter ranging from 0.03 to 330 Hz and digitized at a 1000 Hz sampling rate. Magnetic and movement artifacts were meticulously eliminated using the temporal extension of the Signal Space Separation method (tSSS), facilitated by the MaxFilter software (Neuromag 3.4, Elekta Oy, Helsinki, Finland). Specifically, for PD patients under medication treatment, the tSSS was set with a 10-second raw data buffer and a subspace correlation limit of 0.98. For those undergoing DBS treatment, the tSSS parameters were adjusted to an 8-second raw data buffer with a subspace correlation limit of 0.8, in order to remove the artifacts generated by stimulation and the cables.
2.3. MEG data analysis
2.3.1. Preprocessing
The raw MEG data were processed using SPM12 (Litvak et al., 2011). The initial steps include downsampling the signal to 250 Hz and eliminating the 50 Hz power line noise. Time-independent component analysis (tICA) utilized the FastICA algorithm on the sensor data. During this phase, artifacts associated with breathing, heartbeat, and muscular movements were visually identified and discarded. The MEG data were co-registered with the individual structural MRI terms. Following this, the preprocessed MEG data were projected onto a standard 8 mm mesh source space using the linear constrained minimum variance (LCMV) beamformer (Van Veen and Buckley, 1988; Westner et al., 2022). Subsequently, the brain’s magnetic data were parceled into 39 distinct regions in accordance with a predefined brain template. This segmentation was weighted, identifying 39 cortical regions derived from resting-state ICA decompositions of fMRI data. This data originates from the first 200 subjects of the Human Connectome Project and has been employed in multiple MEG studies (Colclough et al., 2017, 2016; Quinn et al., 2018). Given the significant link between the β band and the motor zone function in Parkinson’s disease, the focus was on analyzing the β frequency band data. As a result, the data is subjected to band-pass filtering, creating two datasets by filtering in the frequency ranges of 13–20 Hz and 21–35 Hz, specifically targeting the high β band (21–35 Hz) and the low β band (13–20 Hz).
2.3.2. Time-delay embedded HMM (TDE-HMM) algorithm
The underlying premise of the HMM is that a time series can be represented by a finite set of hidden state sequences (Vidaurre et al., 2017). At any given time, only one state is active, and the likelihood of being in a particular state at time point t is influenced by the state at time point t-1 (Vidaurre et al., 2018b). HMM is tailored to uncover concealed brain states and the potential transitions between them. Despite the inherent complexity of neural activity, numerous studies have demonstrated that the brain exhibits relative stability during specific periods (Kohl et al., 2024; Kotz et al., 2023; Sharma et al., 2021). These stable states are linked to cognitive, motor, or pathological processes. Utilizing HMM to construct discrete states effectively captures this stability and identifies critical transitions between different brain states. For our analyses, two datasets are used: one is medon/medoff and the other is dbson/dbsoff. We employed the HMM-MAR toolbox to infer these states (Vidaurre et al., 2018a) (available at https://github.com/OHBA-analysi s/HMM-MAR). To address potential overfitting issues within the HMM, we adopted TDE-HMM. This model characterizes the state observation through a multivariate autocovariance matrix (Vidaurre et al., 2018b). The source-reconstructed time courses were integrated with time lags ranging from − 7 to 7. This led to a data matrix for each subject, sized at (N parcels × N lags) × N timepoints, where N represents the 39 parcels. Given that we had previously down-sampled the data to 250 Hz, a lag (L) of 15 equated to 30 ms lags in both positive and negative directions. This created an extended data matrix for each subject sized at (L lags × N nodes) × S time samples. Using principal component analysis, the first dimension of this matrix was condensed from 15 × N to 4 × N. Because the first four components already account for most of the variance, reducing to four dimensions is likely a balance between retaining sufficient information and optimizing computational efficiency and model complexity. Similarly, reducing dimensions would also decrease the risk of overfitting. To ensure consistency and robustness in our findings, the entire HMM training process was iterated 10 times. We selected the iteration that exhibited the most optimal performance, as indicated by the lowest free energy (see Fig. 1).
2.3.3. State parameters analysis and spectral analysis
Following the HMM analysis, we derived the time series of the posterior probability, signifying the likelihood of a specific state occurring at a given moment. We then determined the Viterbi path (Bishop and Nasrabadi, 2006), which outlines the most probable sequence of states, with each point in time exclusively assigned to one state. From this, we computed the Fractional Occupancy (FO), Interval Times (IT), and Life Time (LT) for each state. These metrics provide a dynamic representation of the brain state as captured by the HMM (Baker et al., 2014). Specifically, FO measures the proportion of time a particular state occupies relative to the entire duration. IT gauges the average time lapse between occurrences of the same state. Meanwhile, LT represents the average duration a state persists during a single occurrence (Brookes et al., 2018). These state parameters were determined individually for each patient. States from the HMM were determined based on two frequency bands: high β (21–35 Hz) and low β (13–20 Hz). We obtained the power spectra for each participant in two frequency bands, along with brain spatial energy maps, through power spectral calculation and spatial mapping. The spatial energy map of the brain is obtained by calculating the energy of each region in different states according to the time state sequence. and subsequently averaged them across all participants to discern the alignment of HMM states with intrinsic brain networks. Following this, we conducted group comparisons for each frequency band and specific regions of interest within the brain. The state parameter data of each group exhibited a normal distribution. To compare state parameters between PD patients in the medon and medoff states, as well as between the dbsoff and dbson states, paired T-tests were performed. Pearson correlation analysis was utilized to assess the linear relationships between state parameters within the same group of PD patients across different states. To account for multiple comparisons and control the false discovery rate (FDR), the FDR correction method was applied. An FDR threshold of 0.05 was selected to balance the detection of meaningful findings while minimizing the risk of false positives. All statistical analyses were conducted using GraphPad Prism 9.5.1 software, with significance levels set at p < 0.05.
3. Results
Utilizing the TDE-HMM algorithm, we identified four distinct Hidden Markov states from the data within the high β band (21–35 Hz) and the low β band (13–20 Hz) by comparing the medon and medoff conditions. The states greater than 4 does not change the most common states of the topological configurations. Conversely, using fewer than 4 states can cause some states to merge, which become visible at higher numbers of states, but results in a terrain map that lacks clarity and focus. we selected 4 as the number of states to identify the appropriate motor area states for the PD patients.
3.1. Impact of L-dopa on motor cortical β oscillations in the views of HMM and spectrum
In the results derived from the high β band, the four identified states exhibited distinct regional activations: State 1 was predominantly associated with frontal lobe activation; State 2 highlighted the occipital and temporal lobe; State 3 was characterized by motor activation with a more pronounced activation on the right side compared to the left; and State 4 was primarily linked to the activation of the left occipital lobe, as depicted in Fig. 2(A). A paired T-test, applied to the state parameter data of the `medon’ and `medoff’ groups, revealed significant differences particularly in States 3 and 4. For the motor lobe state (State 3), the Fractional Occupancy (FO) in the `medon’ group is notably higher than that in the `medoff’ group (p < 0.001), while the Interval Times (IT) is significantly reduced (p < 0.05). Conversely, for the occipital lobe state (State 4), the Life Time (LT) in the `medon’ group is markedly shorter than that in the `medoff’ group (p < 0.05), as depicted in Fig. 2(B). Among the four states derived from the low β band: State 1 primarily involves activation of the frontal and temporal lobes; State 2 is predominantly driven by the frontal lobe; State 3 is mainly activated in the right motor area; and State 4 is primarily linked to the occipital lobe, as depicted in Fig. 3(A). Specifically, in the motor lobe state (State 3), the Fractional Occupancy (FO) in the `medon’ group was notably elevated compared to the `medoff’ group, with a p-value <0.05 (illustrated in Fig. 3(B)). In state 3 of the high β and low β bands the power in the medon surpassed that of the medoff (see Figs. 4 and 5). This suggests that the intake of L-dopa amplifies the β power within the motor cortex of PD patients. Such an increase is evident in the elevated Fractional Occupancy (FO) and diminished Interval Times (IT) observed in the state parameters. In the high β band (21–35 Hz), for the occipital lobe state (state4), the power in the medoff exceeded that of the medon (Fig. 6). This observation indicates that L-dopa has inhibition on the power of the high-β band in the occipital cortex. Interestingly, this effect is contrary to what is observed in the motor cortex, suggesting a distinct influence of Ldopa on different brain regions. This stands in contrast to the motor cortex state (state3), where nearly every brain region demonstrated a pattern of medon power surpassing medoff power. The state parameters are designed to reflect the collective power across all brain regions. Consequently, within the occipital lobe state (state4), the Life Time (LT) for the medoff group is more extended than that for the medon group. L-dopa administration resulted in a marked improvement in motor symptoms. Within the high β band’s state 3 (motor state), a notable correlation was observed between the UPDRS III score and interval times under the medoff condition (r2 = 0.2154, p < 0.05) (Fig. 7). Longer state interval times indicate less β oscillation in motor zone, which is associated to more serious motor symptoms.
3.2. Impact of STN DBS on motor cortical β oscillations in the views of HMM and spectrum
Within the four states of the high-β oscillation, State 2 predominantly displayed activation in the motor cortex (Fig. 8(A)). A paired T-test conducted on the state parameter data between the `dbson’ (with electrical stimulation) and `dbsoff’ (without electrical stimulation) groups revealed significant differences in State 2 (Fig. 8(B)). For State 2 (motor state), the `dbsoff’ group demonstrated a significantly higher state proportion (p < 0.05), a reduced state interval (p < 0.05), and a prolonged state lifetime (p < 0.05) compared to the `dbson’ group. In addition, we found the average power spectrum of the whole brain region of motor state (state 2) in dbsoff was higher than that of dbson group (p < 0.001), (See Fig. 9). In the high β band’s state 2 (motor state), a notable positive correlation emerged between the UPDRS III score T + R (a composite score of tremor and limb rigidity) for dbsoff-dbson and the interval times parameter alterations of dbsoff-dbson (r2 = 0.3137, p < 0.05), as depicted in Fig. 10.
4. Discussion
The HMM approach is a powerful tool for analyzing sequential data, particularly noted for its ability to model temporal dynamics and uncover latent states that drive observed data patterns. In the context of PD, HMM has been effectively utilized to analyze time series data and identify hidden states that characterize disease-related phenomena. In our study, we applied the Time-Delay Embedded TDE-HMM to analyze MEG data from PD patients. This probabilistic model aims to uncover hidden brain states and the likely sequences of transitions between them. Our analysis revealed significant differences in the effects of L-dopa on brain activity. Specifically, we observed a substantially higher active state in the motor cortex for both high-β and low-β oscillations following L-dopa administration, indicating that L-dopa treatment increases the prevalence of motor states and shortens intervals between consecutive motor activations (Cao et al., 2020; Heinrichs-Graham et al., 2014). In additional, we found a notably shorter high-β active state in the occipital lobe for the ``medon’’ group, suggesting that PD patients experience increased occipital lobe activity in the absence of L-dopa treatment. Power spectrum analysis further supports these findings, showing that the ``medon’’ group exhibited increased power in the motor area across both high- and low-β bands during motor activation. This aligns with previous research (Cao et al., 2020), which found that L-dopa enhances β oscillation power in the motor cortex. Notably, during state 4, when the occipital lobe was active in the high-β band, power spectrum analysis revealed greater high-β power across multiple areas, including the primary motor cortex, in the ``medoff’’ state. This suggests that L-dopa may shorten occipital activation in this state, facilitating transitions between distinct brain states and potentially driving functional shifts. In the low-β oscillation during motor activation, L-dopa was associated with increased motor cortex activity but reduced activity in the visual and auditory cortices, indicating a possible redistribution of regional activity within the same state. This shift may reflect L-dopa’s role in modulating functional connectivity across cortical regions, adapting brain state dynamics to facilitate motor functions. Integrating clinical evaluation parameters from the Unified Parkinson’s Disease Rating Scale (UPDRS), we found a strong correlation in the `medoff’ group for state 3, suggesting that increased high β power following L-dopa administration reflects enhanced motor state presence, reduced state interval, and prolonged state life. Given the variance in severity between the left and right brain in Parkinson’s patients (Verreyt et al., 2011; Wu et al., 2009), we categorized the 21 subjects with medication treatment based on predominant left-sided and right-sided severity using their UPDRS III scores for separate analysis.. In total, we included 15 subjects with pronounced right-sided severity and 9 subjects with significant left-sided severity. As a result, we observed that the left motor region is active during the motor state. In our study, TDE-HMM analysis the high β band revealed motor state in the dbsoff condition exhibited a significantly higher proportion of states, shorter intervals between states, and longer state lifetimes compared to the dbson condition. These findings stand in remarkable contrast to those observed with levodopa in the high β band, indicating that the proportion of motor state is reduced when DBS is activated. Additionally, within the high β band, the sensor state in which the motor area deactivated exhibited shorter duration in the dbson condition comparing to the dbsoff condition. This suggests that DBS significantly diminishes power oscillations in the motor region within the high β band. In the high β band with the treatment of DBS, the change in the Interval Times of the motor state is proportional to the improvement in tremor and rigidity symptoms, indicating the greater symptom relief is associated with longer interval time of β burst, and reduced β oscillation in the motor cortex. Consistent with previous findings (Pauls et al., 2022), DBS treatment normalizes the characteristics of beta oscillatory activity in patients with Parkinson’s disease, meaning that beta bursts are shorter in duration and lower in amplitude. Both Levodopa and STN-DBS have been shown to alleviate PD symptoms significantly (Kim et al., 2015; Liu et al., 2019, 2014; Muthuraman et al., 2018; Xu et al., 2016; Yin et al., 2021). However, their mechanisms of action on cortical β oscillation differ. In PD, dopaminergic neuron degeneration in the SNc alters the brain’s neural signal equilibrium. The STN, under less inhibitory influence from the GPe, increases excitatory transmission to the Gpi and SNr. This hyperactivity induces thalamic suppression and motor cortex hypoactivity, manifesting as rigidity and bradykinesia. L-dopa treatment can mitigate these effects by compensating dopamine levels, thus enhancing motor cortex excitability (See Fig. 11(A)). Our prior research indicated that L-dopa increases β oscillation power (18–30 Hz) in the motor cortex, correlating with reduced motor symptoms like akinesia and rigidity on the contralateral side (Cao et al., 2015). STN-DBS counteracts aberrant circuitry by suppressing β activation within the STN. This inhibition results in the damping of excitatory discharge in Gpi and SNr, thereby normalizing thalamocortical dynamics. Conversely, through the hyperdirect pathway linking the motor cortex and STN, STN-DBS also dampens β oscillations in the motor cortex (Fig. 11(B)). Main effect of DBS is the functional inactivation (information jam) of the neurons in the stimulated area. The hypothesis was that this is generated by depolarization block. Continuous high frequency depolarization will prevent the generation of future action potentials because of the inactivation of sodium channels (depolarization block). The lodging of the discharge of basal ganglia neurons to DBS high-frequency stimulation inhibits the abnormal activity of these structures. On the other hand, Electrical stimulation at the STN can lead to antidromic inactivation of the afferent neurons to primary motor cortex. As shown in our results, the high β range (21–35 Hz) is more significantly influenced by the STN-DBS treatment in the motor cortex, which is consistent to the study by Abbasi, Omid which found that DBS facilitates antidromic activation via the hyperdirect pathway, and leads to a reduction in β wave oscillatory activity within the bilateral motor cortex (Abbasi et al., 2018). β oscillations play a critical role in maintaining tonic contractions and are suppressed before and during voluntary movement, with evidence suggesting they are essential for motor process integration (Barone and Rossiter, 2021). In our research, we observed that Levodopa and STN-DBS influence motor cortical oscillations differently. It is well-known that STN-DBS can adversely affect speech and gait with high amplitude stimulation, but L-dopa has no similar side effect. Martin M. Reich et al. highlighted that DBS could influence the functioning of brain regions linked to gait, leading to the emergence of gait instability and related symptoms, with the stimulation’s intensity, frequency, and location influencing specific brain areas (Reich et al., 2016). One of the limitations of our study is that clinical evaluations were conducted shortly before MEG recording by neurologists, which may not fully account for changes in condition. In addition, our sample size, comprising 21 PD patients for L-dopa analysis and 11 for DBS, restricts the generalizability of our findings. Factors such as disease duration, individual treatment responses, and other clinical variables may introduce confounding effects. Furthermore, the HMM algorithm relies on discrete states, which oversimplifies the continuous nature of brain dynamics, potentially resulting in the loss of crucial temporal information and details. The algorithm’s significant dependence on large datasets during the derivation process indicates that larger datasets may yield more precise and nuanced results. In the future, recruiting additional cases and utilizing wearable devices to assess clinical scales during the recording process could help address these limitations.
5. Conclusion
Our research employed the TDE-HMM algorithm to analyze MEG recordings from PD patients, revealing distinct effects of L-dopa and STN-DBS on β oscillatory activity in the motor cortex. L-dopa increased β power, whereas STN-DBS diminishes β activity, particularly high β band. Although both treatments alleviate motor symptoms, they affect β oscillations of motor cortex differently. Our results are consistent with previous studies, which have shown that high β oscillations in the motor cortex are closely linked to Parkinsonism symptoms, such as rigidity and bradykinesia, during the resting state (Cassidy et al., 2002). Additionally, our previous studies demonstrated that low β coherence within the motor circuit correlates with motor output during go/nogo tasks (Cao et al., 2024), and other studies have associated low β oscillations with motor preparation and postural control. Future research examining the effects of L-dopa and STN-DBS on both high and low β oscillations during dynamic motor tasks, such as gait, could further illuminate their neurophysiological mechanisms and contribute to the development of brain-computer interface applications for motor control. Financial disclosure None reported. Data and code availability statement Data and code can be shared after contact with the author. CRediT authorship contribution statement Kunzhou Wei: Writing – original draft, Visualization, Validation, Methodology, Formal analysis. Hang Ping: Writing – original draft, Visualization, Validation, Methodology, Formal analysis. Xiaochen Tang: Methodology, Formal analysis, Conceptualization. Dianyou Li: Project administration, Investigation. Shikun Zhan: Project administration, Investigation. Bomin Sun: Project administration, Investigation. Xiangyan Kong: Writing – review & editing, Supervision, Project administration, Funding acquisition. Chunyan Cao: Writing – review & editing, Resources, Investigation, Funding acquisition, Data curation, Conceptualization. Declaration of competing interest The authors declare that they have no conflict of interest.
Acknowledgment
This article is supported by National Natural Science Foundation of China (Grant nos. 62071265 & 82071547), One health Interdisciplinary Research Project, Ningbo University and Shanghai Municipal Science and Technology Commission (21Y11905300 to B.S.). Data availability Data will be made available on request.
Article Images (0)
×![]()
Bioz is the world’s most comprehensive AI search engine for scientific experimentation. The Bioz search engine offers researchers billions of data-driven product, technique, and protocol recommendations. Bioz taps into the latest advances in Artificial Intelligence (AI) – including Natural Language Processing (NLP), Machine Learning (ML) and Generative AI to mine and structure hundreds of millions of pages of complex and unstructured scientific papers. By harnessing the power of AI, Bioz provides researchers with an unprecedented amount of summarized scientific experimentation knowledge right at their fingertips, ultimately helping to speed up drug discovery and the rate of success in finding cures for diseases. Bioz is used by over 15 Million researchers from over 17,000 different universities and companies in 196 countries.
