Interaction of motor behaviour, cortical oscillations and deep brain stimulation in Parkinson’s disease

Abstract

Recent progress in the study of Parkinson’s disease has highlighted the pivotal role of beta oscillations within the basal ganglia-thalamo-cortical network in modulating motor symptoms. Predominantly manifesting as transient bursts, these beta oscillations are central to the pathophysiology of Parkinson’s disease motor symptoms, especially bradykinesia. Our central hypothesis is that increased bursting duration in cortex, coupled with kinematics of movement, disrupts the typical flow of neural information, leading to observable changes in motor behaviour in Parkinson’s disease.

To explore this hypothesis, we employed an integrative approach, analysing the interplay between moment-to-moment brain dynamics and movement kinematics and the modulation of these relationships by therapeutic deep brain stimulation (DBS). Local field potentials were recorded from the hand motor (M1) and premotor cortical (PM) areas and internal globus pallidus (GPi) in 26 patients with Parkinson’s disease undergoing DBS implantation surgery. Participants executed rapid alternating hand movements in 30-s blocks, both with and without therapeutic pallidal stimulation.

Behaviourally, the analysis revealed bradykinesia, with hand movement cycle width increasing linearly over time during DBS-OFF blocks. Crucially, there was a moment-to-moment correlation between M1 low beta burst duration and movement cycle width, a relationship that dissipated with therapeutic DBS. Further analyses suggested that high gamma activity correlates with enhanced motor performance with DBS-ON. Regardless of the nature of coupling, DBS’s modulation of cortical bursting activity appeared to amplify the brain signals’ informational content regarding instantaneous movement changes.

Our findings underscore that DBS significantly reshapes the interaction between motor behaviour and neural signals in Parkinson’s disease, not only modulating specific bands but also expanding the system’s capability to process and relay information for motor control. These insights shed light on the possible network mechanisms underlying DBS therapeutic effects, suggesting a profound impact on both neural and motor domains.

Mirpour and Pouratian investigate the interplay between movement, brain oscillations and deep brain stimulation (DBS) in Parkinson’s disease. They show that beta oscillations interfere with the moment-to-moment control of movement, and that DBS can enhance communication within brain networks, mitigating these effects.

Introduction

Bradykinesia is one of the primary motor symptoms of Parkinson’s disease (PD). It is characterized by a reduction in movement speed and amplitude during repetitive motions.^1-3 Bradykinesia has been associated with beta power amplification in both cortical⁴ and subcortical⁵ areas, enhanced beta synchrony within and between⁶ subcortical structures and increased subthalamic nucleus (STN) beta band bursting activity.⁷

To date, analyses have focused on investigating how sustained increases in oscillatory power and coherence relate to symptoms of bradykinesia, usually with an ordinal measure extracted from the Unified Parkinson’s Disease Rating Scale (UPDRS).⁸ However, dynamic changes in beta power play a crucial role in moment-to-moment movement regulation.⁹ Although oscillations suggest non-stationary dynamics, most current electrophysiology studies treat these signals as sustained, stationary signals.

A study on the neurophysiology of PD has shown that the temporal variability of the STN beta amplitude envelope correlates inversely with PD symptomatology.¹⁰ Tinkhauser⁷ likewise demonstrated that the duration of STN beta bursts, or transient increases in beta power amplitude,¹¹ relate to clinical scores of PD severity, and that beta burst duration decreases with levodopa therapy. We have previously shown that M1 cortical beta bursts are longer in patients with PD than those observed in patients with essential tremor.¹² Subsequent studies have indicated that targeting beta bursts in the STN using adaptive (‘closed-loop’) deep brain stimulation (DBS) can specifically reduce beta burst duration¹³ and offer greater symptomatic relief than continuous DBS.¹⁴ Among these studies, only a few established a real-time correlation between neural dynamics, as quantified by burst activity, and instantaneous movement kinematics.^12,15-20 This objective is fundamental to the aims of our study.

There is growing evidence that non-linear dynamics, evident in bursting activity in cortical and subcortical brain areas, play a significant role in the development and progression of PD. In comparison to healthy individuals, PD patients exhibit altered non-linear characteristics in brain activity across various scales, from individual neurons to global brain signals.^21,22 An examination of STN field potentials reveals that burst duration signifies a rise in the non-linearity of the signal, as determined by linear surrogates when transitioning from the medicated to the unmedicated state. This change in non-linearity is associated with motor dysfunction.¹⁵ These suggest that dynamic power fluctuation and functional connectivity, like studying the beta burst dynamics and mutual information, could represent the most effective method of understanding the genesis of the transient pathogenesis of Parkinson’s symptoms.

Despite these efforts, the moment-to-moment relationship between movement features and dynamics of local oscillations and functional connectivity, specifically in the cortex, remains unclear in PD. It is also unclear how DBS affects this relationship and thereby achieves alleviation of symptoms. We hypothesize that exaggerated bursting in cortical areas and increased coupling of bursts with movement kinematics disrupt the healthy flow of information, causing the moment-by-moment changes in motor behaviour in people with PD. To address this, our approach was to investigate the relationships between motor behaviour, beta burst duration, high gamma oscillations and functional connectivity with both DBS-ON and DBS-OFF in PD patients. Based on our hypothesis, we predict coupling between bursting dynamics and movement kinematics during DBS-OFF and that DBS-ON would modulate movement kinematics and brain electrophysiology dynamics, increasing the information exchange in motor control and improving motor performance.

Materials and methods

Surgical procedure

The Institutional Review Board at the University of California, Los Angeles, approved the study, and all participants gave written informed consent in accordance with the Declaration of Helsinki. DBS implantations were performed based on clinical necessity alone. The choice to target the internal globus pallidus (GPi) was made based on clinical evidence showing similar motoric efficacy, conversations with the patient about possible benefits and side effects and advice from an interdisciplinary clinical team, which included a neurosurgeon and a movement-disorder neurologist. The study involved a cohort of 26 individuals (18 males, 8 females) undergoing DBS for PD. Leads were implanted in bilateral GPi, except for Patients 25 and 26, who had unilateral left and right implantations, respectively (Supplementary Table 1, patient demographic details). All subjects received a comprehensive pre-operative neurological assessment, which involved clinical grading with the Unified Parkinson’s Disease Rating Scale (UPDRS) motor part III, both OFF and ON medication. In this study, we only used the hand subscore of UPDRS (sum of finger tapping, hand open/close and pronation/supination). PD medications were discontinued at least 12 h before recording, following standard clinical protocol.

All subjects underwent clinical pre- and post-operative imaging. Pre-operatively, T1-weighted magnetization prepared rapid acquisition gradient echo (MPRAGE) imaging (slice thickness = 1 mm, repetition time = 2100 ms, echo time, 2.98 ms, flip angle = 15°, 3 T, Siemens Skyra) was carried out. An eight-contact subdural electrocorticography (ECoG) strip with 4 mm platinum-iridium electrodes and 1 cm spacing (AdTech Medical) was inserted subdurally in a posterior direction through a burr hole made for DBS lead insertion. After creating the surgical burr holes based on preoperative clinical stereotactic planning, patients were awakened from anaesthesia. After establishing wakefulness for at least 30 min and clinical baseline, DBS leads were implanted using microelectrode recordings and mapping with kinesthetic testing and subsequent macroelectrode stimulation testing. For implantation, a Leksell stereotactic headframe (Elekta Instruments) was secured to the skull, followed by a full head CT scan with a 0.6 mm slice thickness (Siemens Sensation 64). The DBS lead (Model 3387, 1.27 mm lead body diameter, contact length 1.5 mm, inter-contact distance 1.5 mm, Medtronic Inc.) was targeted to the motor (ventral posterolateral) GPi using image-guided targeting, positioned 2–4 mm anterior, 19–24 mm lateral and 4–6 mm inferior to the mid-commissural point, depending on individual anatomy. All trajectories were verified with intraoperative microelectrode recordings based on firing activity and awake macrostimulation testing. All experimental deep brain leads, ECoGs and hand movement tests (Fig. 1A) were simultaneously recorded during the awake state at least 30 min after discontinuation of propofol sedation.

Figure 1.

Experimental design. (A) An example of a recording session over time. Subjects performed multiple blocks of rest and hand movement. The thin trace is an example of the hand movement data. The thick horizontal bars denote the period of deep brain stimulation (DBS)-OFF and -ON, respectively. (B) The schematic location of an ECoG strip on the cortex surface. The motor cortex (the dot denoted with M1) was the first contact in front of the central sulcus (CS line), and the contact before that was the premotor cortex (the dot denoted with PM). (C) A schematic illustration of a hand movement cycle showing the width of a movement cycle. (D) An example of burst analysis. Burst periods are highlighted on the filtered narrowband beta band signal, the analytical envelope of the signal and the 75th percentile threshold (dashed line). Bottom: The burst periods are highlighted with thick vertical bars on the original broadband signal. (E) An example of the movement cycle width over time in DBS-ON and -OFF blocks.

Experimental protocol

Recordings were obtained at a 2400 Hz sampling rate, with a bandpass filter of 0.1 and 1000 Hz using BCI2000 v3 connected to amplifiers (g.Tec, g.USBamp 2.0) and a single scalp reference. Electrode locations were determined postoperatively using preoperative MRI and CT, intraoperative fluoroscopy and postoperative CT using a technique previously described.^23,24 The first contact unambiguously anterior to the central sulcus was considered to lie over the motor cortex. The premotor cortex (PM) was identified as the contact immediately anterior to the primary motor cortex (M1). A map of all contacts used for analysis was projected onto a 3D reconstruction of a standardized cortical surface (Fig. 1B).

For experimental purposes, we used the second most ventral DBS lead contact (Medtronic Contact 1) to deliver cathodic stimulation with either a scalp (n = 3) or shoulder (n = 6) anode. We used the flanking contacts (Contacts 0 and 2) to measure local GPi activity in a bipolar manner. We recorded up to six cycles of 30-s rest and movement periods for each subject. Only blocks in which subjects performed at least 15 s of uninterrupted movement were included in the study (see the ‘Results’ section for details). The first cycle was without GPi-stimulation (DBS-OFF state), while the second cycle was with GPi-DBS stimulation (DBS-ON state), with subsequent alternation. We used a glove with five sensors that detect finger bending and straightening (5DT data glove 5 Ultra) on the left hand (except for Patient LG01) to provide concurrent hand movement recordings at a slower effective sampling rate, which was oversampled offline at 2400 Hz by BCI2000 using stair step interpolation. In this study, we only analysed data from movement periods.

Data analysis

All analyses were performed offline using MATLAB custom-made scripts (Mathworks Inc., NA, USA). Three bipolar signal pairs were used for analysis: GPi, PM and M1 cortices. M1 and PM (contact pair immediately anterior to the central sulcus, spanning precentral gyrus) were referenced with the electrode immediately posterior to the central gyrus (Fig. 1B).

For analyses of hand movement, slow fluctuation of the hand movement data was removed using local linear regression fit. Then, local maxima of movement velocity (first derivative of movement trace) were automatically detected using the findpeaks MATLAB algorithm (Fig. 1C). The full-width at half-maximum of each opening and closing hand movement cycle was calculated as the ‘movement cycle width.’ Each movement cycle width and the time it happened were used for the calculation of moment-to-moment correlation coefficients.

Neurophysiological data preprocessing involved removing 60 Hz line noise and discarding data segments containing electrical or movement artefacts, as previously described.²⁵ Removed segments primarily exhibited abnormally high power spectra values, excessive harmonics and high rates of voltage change. Automatic artefact removal involved full-wave rectification, first derivative calculation and filtering using a 5-point median filter. An artefact was defined as a region where the first derivative exceeded 5 SD for the subject, calculated across their entire recording. Linear interpolation from data 2 ms before and after the artefact replaced removed segments. The narrow band analytical signal was calculated using the Hilbert transform and zero-phase finite impulse response (FIR) bandpass filters with cut-off frequencies at low beta (12–20 Hz), high beta (21–35 Hz)²³ and high gamma (65–110 Hz) using the MATLAB signal processing toolbox. Power spectral density was calculated using Thomson’s multitaper method in 1 s consecutive time windows with no overlap for frequencies of 4 to 110 Hz with ±2 Hz frequency bandwidth with three tapers.²⁶ To account for inter-subject variability in baseline power, each segmented spectrum was normalized to the total power of the signal during each condition (excluding the line-noise and its harmonics). Periods of beta bursting were defined as any oscillation amplitude period that exceeded the 75th percentile power for that subject within a specific band and persisted for at least 50 ms as illustrated in Fig. 1D.^7,13 A common threshold for a subject was calculated by averaging all the thresholds of DBS-ON and -OFF blocks of a subject.¹³

We estimated the magnitude squared coherence to quantify the degree of co-variability between pairs of cortico-cortical or cortico-subcortical signals. This was calculated using the multitaper method, with time window and frequency smoothing parameters identical to those used in the power spectral density analysis (with a ±2 Hz frequency bandwidth and three tapers). Spearman’s rho test of significance was used to calculate correlations with clinical measurements. When more than one statistical test was performed against a dataset, the significance value was adjusted for multiple comparisons using the Bonferroni correction.

To calculate moment-to-moment correlation coefficients, the calculated physiological measures (burst features, power and coherence) were smoothed with a 3-s Gaussian kernel. Spearman’s correlation coefficients and P-values were then calculated at the time of each movement peak between the cycle-width and physiological measures. These coefficients reflect the dynamic interaction between neural activity and motor behaviour within each block, capturing increases and decreases over time that might not affect the overall average of the neural markers or movement properties. Additionally, we used separate linear mixed-effects (LME) models to examine how changes in movement cycle width were related to physiological measures in low and high beta bands, while controlling for confounding factors of subjects and blocks.

We used a LME model to account for both fixed and random effects within our data. This approach is particularly advantageous for estimating between-subject and between-item variance and accounts for both fixed and random effects, making them suitable for research with hierarchical structures like our data set of blocks and subjects. In our study, the dependent variable was the change in movement cycle width. The independent variables were burst durations or time. In addition to the empirical evidence highlighting the mechanistic divergence between low and high beta burst signals,²⁷ we discerned a substantial qualitative distinction in the effects produced within the low (13–20 Hz) and high (21–35 Hz) beta sub-bands (see the ‘Results’ section). The duration of M1 beta bursts and the hand movement cycle width variables were incorporated into the model separately for either the low or high beta sub-band. We also included intercepts and slopes as random factors grouped by subjects to account for cross-subject variability in the linear regression between the dependent and independent variables. Thus, fixed effects showed the estimated mean values of the slopes in the tested within-subject relationship across all subjects, and random effects were adjusted for average differences and confounding effects of subjects or blocks.

To investigate the dynamic relationship between hand movement cycle width and physiological brain neural markers, we conducted a time lag analysis. This analysis allowed us to explore the temporal dependencies between the variables by examining correlations at various time lags and visualized the relationship between movement and neural markers around the time of each cycle of movement.

We calculated the correlation, as explained above, between movement cycle width and neural markers across a range of time lags, from −1.5 s to +1.5 s relative to the centre of each movement cycle for DBS-ON and DBS-OFF blocks. Negative lags represented neural markers leading the movement, while positive lags represented the movement leading the neural markers. Significant differences between the DBS-ON and DBS-OFF conditions were identified and marked with a star on the plots using a paired t-test at P < 0.05 at each time lag point.

A mutual information statistic was calculated to determine how well movement features appear in the neural representation. The mutual information was calculated between the discretization of smoothed continuous narrowband signals (3-s Gaussian kernel, sigma 1 s) of the local field potential (LFP) signal and change in hand movement using algorithms introduced by Peng et al.²⁸ In brief, first, the entropy of the joint distribution of two variables was calculated. The joint entropy is given by:

$$H(A,B)=-{\displaystyle \sum _a}{\displaystyle \sum _b}P(a,b)\mathrm{lo}{\mathrm{g}}_2(P(a,b))$$ (1)

Where A and B represent the LFP signals and change in hand movement, and P(a, b) is the joint probability of observing values a and b.

Then, the marginal entropy of each signal was calculated by:

$$H(X)=-{\displaystyle \sum _y}P(x)\mathrm{lo}{\mathrm{g}}_2(P(x))$$ (2)

Where x is either LFP or a change in hand movement signal.

Finally, the mutual information (I) between the two variables was calculated given:

$$I(X;Y)=H(X)+H(Y)-H(X,Y)$$ (3)

Raw data were generated at the University of California, Los Angeles. Derived data supporting the findings of this study are available from the corresponding author on request.

Results

Kinematics of movement

The average age of 26 participants was 63.7 years (SD 7.02). We evaluated hand movement kinematics by instructing study participants to open and close their hands repeatedly as fast as they could for several blocks, each block lasting about 30 s (Fig. 1A). We included 3–6 (mean = 4.7) blocks of DBS-OFF and 1–5 (mean = 2.7) DBS-ON blocks from each subject. The block average of movement cycle width across all subjects showed a mean of 0.66 s [standard error of the mean (SEM) = 0.064] for DBS-OFF blocks and 0.49 s (SEM = 0.062) for DBS-ON blocks.

To quantify the bradykinesia, we segmented the opening/closing cycles and computed the width of each cycle as a proxy for the instantaneous movement velocity. We evaluated the moment-to-moment fluctuations of hand movement using Spearman’s correlation coefficient between movement cycle width and time for each block of movement (see Fig. 1E for an example). Positive correlation coefficients indicate a slowing of movement. 64.52% of all blocks exhibited positive correlation coefficients, with DBS decreasing this percentage to 39.71% [27/68, odds ratio = 2.7, 95% confidence interval (CI) = 1.4 to 5.3, χ²  P = 0.002], indicating a significant reduction in bradykinesia with DBS. Moreover, the odds of having significantly positively correlated blocks were significantly higher during DBS-OFF, compared to DBS-ON (odds ratio = 3.5, 95% CI = 1.0 to 12.5, χ²  P = 0.045), confirming that bradykinesia was decreased with DBS.

We compared the distribution of the correlation coefficients between DBS-ON and DBS-OFF blocks. Correlation coefficients were significantly skewed towards positive values during DBS-OFF (Wilcoxon signed rank test, P < 0.014), suggesting a consistent slowing of movement over time within each block. This skewness was not present with DBS-ON (Wilcoxon signed rank test, P = 0.202), indicating that DBS disrupted the temporal evolution of the movement. Moreover, there was a significant difference between the median values of the correlation coefficients during DBS-ON and -OFF (Wilcoxon rank sum test, P < 0.011; Fig. 2A), and a receiver operating characteristic (ROC) analysis confirmed that the two conditions could be discriminated based on the correlation coefficients, albeit with some overlap [area under ROC (AUC) = 0.62, 95% CI = 0.52 to 0.70; Fig. 2A, inset]. In addition, DBS significantly decreased the average cycle widths across blocks (t-test, P < 0.0001), indicating improved movement velocity. To account for the variability of subjects and blocks and to control for the unequal number of blocks per subject, we used a LME model with cycle width as the dependent variable, time as the fixed independent variable, and subjects as a random factor with a nested structure of blocks within subjects for each DBS condition. After removing the effect of random variables, we detected a significant increase in cycle width from the beginning to the end of the blocks (fixed effects coefficient = 1.1, P = 0.027). However, DBS reversed and rendered this effect insignificant (fixed effects coefficient = −1.6, P = 0.057). These results confirmed that the movement cycle width increased significantly within blocks in the DBS-OFF condition and that subject or block variability did not confound this effect.

Figure 2.

The relationship between movement cycle width and time, behaviour and burst duration. (A) The histogram of the correlation coefficients between primary motor cortex (M1) low beta burst duration and time of deep-brain stimulation (DBS)-ON and -OFF blocks. (B and C) The relationship between cycle width and bradykinesia score is shown in scatter plots. Each point represents a block of DBS-OFF (B) or DBS-ON (C) data. (D) The histogram of Spearman’s correlation coefficients between M1 low beta burst duration and the movement cycle width of DBS-ON and -OFF blocks. The red and blue triangles point to the median of the distributions. Inset: The receiver operator characteristic (ROC) curve between two distributions. (E) The normalized power (dB/Hz) as a function of frequency (Hz) for the DBS-ON and DBS-OFF conditions. (F) Burst duration (ms) as a function of frequency (Hz) for the DBS-ON and DBS-OFF conditions. In both E and F, the shaded areas represent the standard error of the mean.

Finally, we examined the relationship between the slopes of cycle width over time and unilateral UPDRS-III bradykinesia hand sub-scores in each subject. We found a significant positive correlation between the mean slope of cycle width over time and the bradykinesia scores for the DBS-OFF blocks (Spearman’s rho = 0.50, P = 0.02; Fig. 2B). With DBS-ON, the correlation of cycle width slope over time and unilateral UPDRS-III bradykinesia scores was absent (Spearman’s rho = −0.03, P = 0.90; Fig. 2C).

The relationship between physiologic markers and movement dynamics

We examined the relationship between the duration of low beta bursts in the M1 and movement cycle width at both the population and single-block levels. The M1 low beta burst and movement cycle width correlation is disrupted with therapeutic DBS, as shown in Fig. 2D. At the block level, 18.28% (17/93) of the blocks in DBS-OFF and 14.93% (10/67) showed significant Spearman correlation coefficients (P < 0.05). The number of blocks exhibiting a positive correlation (Spearman correlation coefficient > 0) between movement cycle width and M1 low beta burst duration significantly decreased from 62.37% (58/93) to 43.28% (29/67) using the chi-squared test (P = 0.017, odds ratio 2.17 with 95% CI = 1.14 to 4.11). The number of blocks with significant positive correlation (rho > 0 and P < 0.05) decreased from 13.98% in DBS-OFF to 5.97% in DBS-ON blocks. Furthermore, the median of the correlation coefficients was significantly larger than zero in DBS-OFF blocks (median = 0.086, Wilcoxon signed-rank test, P = 0.015) but significantly decreased (Wilcoxon rank-sum test, P = 0.02) to values not significantly different from zero (median = −0.059, Wilcoxon signed-rank test, P = 0.269) in DBS-ON blocks. Additionally, ROC analysis confirmed a significant shift in the correlation coefficient distribution toward negative values (AUC = 0.61, 95% CI = 0.54 to 0.71; Fig. 2D, inset) with DBS. Using the same methodology, we couldn’t find any significant relationship between the duration of high beta bursts or gamma in the M1 or PM and movement cycle width at both the population and single-block levels (for detailed statistics, see Supplementary Table 2).

Despite the correlation between cortical beta bursts and behavioural measures, M1 low beta and gamma burst duration and power were not significantly different between the DBS-ON and -OFF states on average across blocks (Wilcoxon rank sum test, P > 0.60; also see Fig. 2E and F). In contrast, we identified significant disparities in the normalized average low beta power in GPi between DBS-ON and -OFF states (Wilcoxon rank sum, P < 0.0001), consistent with prior reports.^23,29

While a moment-to-moment correlation was evident between M1 low beta burst duration and movement cycle width, there was no significant linear trend observed in the M1 low beta burst duration across block durations. The median correlation coefficients between M1 low beta burst duration and time did not differ significantly from zero (Wilcoxon sign rank, P = 0.135), and DBS did not induce any significant change (Wilcoxon rank sum test, P = 0.178). To investigate whether other neural markers within the low beta frequency band exhibited similar relationships, we examined the correlation between movement cycle width and both low beta M1 power and M1-GPi coherence. None of these additional metrics replicated the pattern observed with M1 low beta burst duration (refer to Supplementary Fig. 1 for details).

To control the confounding effect of subjects and blocks, we again employed a LME model. The movement cycle width was defined as the dependent variable, burst duration as a fixed factor, and subjects as a random factor with hierarchical construction of blocks as subgroups for both DBS-ON and DBS-OFF conditions. After excluding the effect of random variables, the LME model confirmed a significant correlation between burst duration and movement cycle width (fixed-effects coefficient = 0.039, P = 0.0098). This correlation became non-significant (fixed-effects coefficient = −0.0047, P = 0.798) in DBS-ON blocks, confirming the kinematic and beta burst duration correlation (for a full report on LME, see the Supplementary material).

Time-lag analysis

We assessed the relationship between physiological markers and movement cycle width, incorporating time lags up to a second before and after the centre of each movement cycle (Fig. 3A and B). In each panel of Fig. 3, the gap between DBS-OFF and -ON traces (red and blue) shows the decorrelation effect between neural markers and movement over time. The bifurcation between DBS-ON and DBS-OFF in M1 low beta burst duration starts prior to movement and continues after the movement (Fig. 3A). The distinction between the DBS-ON and DBS-OFF conditions remained significant across 10 consecutive bins, approximately 540 ms, as indicated by the red stars in Fig. 3A. This observation is noteworthy as it illustrates that, although DBS settings were constant throughout the DBS-ON blocks, they dynamically altered the relationship between movement and low beta bursts, especially at the onset of movement. This finding supports the hypothesis that low beta bursts are dynamically associated with movement features. On the other hand, DBS increased the correlation between M1 high gamma (65–110 Hz) burst duration and movement cycle width, starting before movement and lasting briefly after movement (P < 0.05; Fig. 3B, red stars). While the significant difference between the DBS-ON and DBS-OFF conditions in the high gamma band persisted for only half as long as that in the low beta band (lasting approximately 250 ms over three consecutive bins; Fig. 3B), it nonetheless represented a qualitatively distinct effect. Unlike in the low beta band, DBS increased the correlation between burst duration and movement cycle width in the high gamma band. This trend, evident over a longer duration of 10 consecutive bins (about 1 s), demonstrated a consistent, although not always statistically significant, effect of DBS that was also present in the correlation between movement cycle width and M1 power and M1-PM coherence (Supplementary Fig. 2). These findings suggested that the dynamics of high gamma activity may be positively associated with improved motor performance following DBS intervention, as further discussed in the ‘Discussion’ section. Further supporting this notion is the relationship between the correlation coefficients of blocks. The correlation coefficients of movement cycle width over time exhibited a significant positive correlation with the coefficient of movement cycle width with M1 low beta burst duration (rho = 0.30, P < 0.001; Fig. 3C) in DBS-OFF blocks. However, this correlation was not apparent with DBS-ON (rho = 0.02, P = 0.89; Fig. 3D). In contrast, this relationship was not significant for high gamma in DBS-OFF blocks (rho = −0.08, P = 0.42; Fig. 3E) but became negatively significant with DBS-ON (rho = −0.35, P < 0.001; Fig. 3F). The observed pattern suggests that an enhanced coupling between behaviour and beta burst duration is associated with more pronounced bradykinesia. However, DBS eliminated this correlation. In the context of high gamma bursts, no correlation was found between coupling and bradykinesia in the absence of stimulation. Intriguingly, with stimulation, an increased coupling was associated with reduced bradykinesia.

Figure 3.

The relationship between physiologic measures and movement cycle width over time and severity of symptoms. The median of the correlation coefficients of deep-brain stimulation (DBS)-ON and DBS-OFF physiologic measures averaged across all patients for DBS-ON and -OFF blocks separately. (A) Primary motor cortex (M1) low beta burst duration; (B) M1, 65–110 Hz burst duration with the introduction of the time lag between burst duration and movement cycle width. Negative x-axis values represent the neural data leading movement. The stars represent the significant difference between DBS-ON and -OFF traces using the paired t-test at P < 0.05. Scatter plots of Cycle width × Time correlation coefficients as a measure of severity of symptom across correlation coefficients of Cycle width × M1 low beta burst duration of (C) DBS-OFF and (D) DBS-ON. The same graphs were plotted for the 65–110 Hz band in E and F. Each filled black circle represents a block, and the dashed line is the least-squared line of the population. Bottom right in A–F: The results of the Spearman correlation test.

Information transmission and deep-brain stimulation effect

We examined the mutual information between bursting activity and instantaneous changes in hand movement. Interestingly, in both low-beta and high-gamma, DBS significantly enhanced the mutual information between burst activity and momentary alterations in movement cycle width (Wilcox rank-sum test, low beta burst amplitude P = 0.002, ROC AUC = 0.64; high gamma P = 0.006, ROC AUC = 062; Fig. 4A and B). This occurred even though the correlation between the burst amplitude and movement cycle width were in opposite directions This finding suggests that, despite differences in the correlation between behaviour and oscillatory activity across various frequency bands, DBS mechanistically introduced modifications that amplified information transfer (see the ‘Discussion’ section).

Figure 4.

The mutual information between burst duration and hand movement. (A) The histogram of the mutual information between (A) primary motor cortex (M1) low beta burst duration, (B) the 65–110 Hz M1 burst duration and movement cycle width of deep-brain stimulation (DBS)-ON and OFF blocks. The arrowheads point to the median of the distributions. Inset: The receiver operator characteristic (ROC) curve between two distributions.

Discussion

This study investigates the temporal dynamics of oscillatory activity in the cortex during continuous alternating movements in PD patients. Collectively, these findings suggest M1 low beta band features, especially burst duration, correlate with the emergence of bradykinesia, while high gamma oscillations appear when DBS alleviates disease symptoms, highlighting the modulation of neural oscillatory dynamics by therapeutic interventions. A pivotal aspect of our finding was the observation that cortical beta burst dynamics are associated with moment-to-moment variations in movement cycle width, and this association cannot be accounted for by changes in low beta power or synchrony within pallido-cortical signals (Supplementary Fig. 1). Notably, these dynamic shifts transpired without a significant change in average M1 burst duration at the block level. In addition, while there was a correlation between the cycle width and time correlation coefficient and bradykinesia, no difference in beta burst duration was observed between DBS OFF and ON. Therefore, the change in bradykinesia in this scenario does not appear to be associated with the change in a long-term average of beta burst duration; rather, this relationship is moment-to-moment and dynamic. The results indicate that the identified relationship between behaviour and M1 low beta burst likely originates not from a broad linear trend but rather from a transient coupling between changes in behavioural and neural data.

We found cortical low beta dynamics represented by burst durations to be correlated with moment-to-moment alterations in movement cycle width. This observation is in line with past literature that has consistently highlighted the role of beta band oscillatory dynamics in PD networks.^{7,12,13,30-32} Such a finding, coupled with the absence of a tangible correlation between other oscillatory and synchrony metrics with behaviour, reinforces the idea that cortical beta bursts emerge due to the non-linear accumulation of network oscillatory activity during ongoing motor actions.^33,34 This dynamic correlation underscores the therapeutic effects of DBS, suggesting that while the overall length of bursts remains consistent, the timing and relationship of these bursts relative to motor actions are modulated. Our results also highlight the role of cortical bursting in sequence effect—repetitive movement amplitude and speed that stands as a major morbidity contributor in PD.³⁵ Interestingly, despite existing knowledge of beta power suppression in the cortico-basal network during movement, there is no evidence for a direct correlation between oscillatory power shifts and sequence effect.

Time-lag analysis revealed a significant correlation that begins moments before movement initiation and persists thereafter. This highlights the close temporal association between movement cycle width and bursting, consistent with the hypothesis that cortical bursting maintains a dynamic causal relationship with both movement planning and execution control.

The correlation between beta burst duration and movement cycle width dissolves with DBS. While dopaminergic medication partially addresses bradykinesia, it does not significantly improve movement amplitude decline, unlike both open and closed-loop STN DBS, which show marked improvements.^16,36,37 The distinct responses to dopaminergic medications and DBS imply different underlying mechanisms for bradykinesia and the sequence effect.³⁸ Regardless of the nature of the correlation between M1 beta burst duration and movement cycle width, DBS’s efficacy in mitigating the sequence effect and decoupling behaviour from the cortical burst accentuates the modulating role of cortical bursts in movement genesis. This strongly aligns with the hypothesis that cortical bursting plays a pivotal role in real-time movement planning and execution control,^12,33,34 offering a potential avenue for counteracting bradykinesia during movements.

Our study further delved into the dynamics within the high gamma range (65–110 Hz). We observed a notable trend wherein DBS heightened the correlation between M1 high beta bursts and movement cycle width. While this observation didn’t meet the significance strength of M1 low beta burst duration, a consistent trend during a second lag-analysis insinuated that high gamma activity might be tied to the motor performance enhancements observed with DBS application. This aligns with earlier research advocating the significance of high gamma oscillations in motor function.^39,40

Our results extend this theoretical discourse, introducing a potential mechanism by which DBS enhances motor performance in PD, attenuating the linear correlation of low beta bursts and altering the disease’s symptomatic progression during time course of action. In contrast, the relationship between high gamma activity and behaviour is dynamic and non-linear, crucial for movement execution and regulation, but disrupted in PD. DBS may correct this imbalance, though its effect on high gamma bursts is less pronounced than on beta bursts. This observation may be due to our focus on simple behavioural metrics and linear analysis. A more profound understanding might require non-repetitive, complex movements, advanced kinematic measures and more sophisticated analytical methods.

Intriguingly, DBS reduces the influence of low beta cortical bursts on behaviour, while enhancing the impact of high gamma bursts. This dual modulation likely improves behavioural symptoms by increasing information flow, as indicated by a significant rise in mutual information between hand movement fluctuations and both low beta and high gamma burst amplitudes during DBS-ON blocks. The higher correlation coefficients suggest a stronger linear relationship between neural biomarkers and cycle width. However, this does not necessarily imply a higher mutual information, which could be influenced by the complexity and non-linearity of the relationships under different conditions. Here is why we introduced the mutual information, a non-linear metric that quantifies the shared information between two data channels. Our combined findings from correlation and mutual information analyses suggested that DBS amplifies the informational content conveyed by the cortical signal regarding hand movement. This enhancement may stem from a reduction in pathological oscillations, such as low beta oscillations, or an elevation in the signal-to-noise ratio, as seen with high gamma bursts.

It is important to note that bursts should not be oversimplified as mere pathological indicators, even if their progressive elongation during movements aligns with clinical manifestations. A deeper insight suggests that bursts act as a dynamic momentary channel for information flow, playing a pivotal role in real-time planning, control and execution of motor functions. While the analysis of burst activity successfully elucidated a few aspects of non-linear dynamic complexity, further research is warranted to comprehensively understand the intricacies of the non-linear dynamics of the basal ganglia-thalamo-cortical networks using more expressive analytical tools. Another key limitation of this study was that the results were obtained in the medication OFF state, as this condition is necessary for our current DBS therapeutic surgeries. Therefore, it remains unclear whether the observed relationships hold utility in the presence of dopaminergic medication.

This work unveils the critical role of cortical low beta bursts in relation to movement velocity and clinical symptoms. Our findings show how DBS alters the interaction between motor behaviour and neural indicators, especially M1 beta burst duration, and reveals the link between high gamma oscillations and improved motor performance. Such revelations pave the way for the evolution of more precise and potent therapeutic strategies for PD, with the overarching aspiration of elevating the well-being of those grappling with the ailment.

Data availability

Derived data supporting the findings of this study are available from the corresponding author on request.

Funding

This work has been supported by the National Institute of Neurological Disease and Stroke of the National Institutes of Health (https://www.ninds.nih.gov/) under award number R01NS097782.

Competing interests

K.M. reports no competing interests. N.P. receives consulting fees from Abbott and Sensoria Therapeutics.

Supplementary material

Supplementary material is available at Brain online.