Essential tremor disrupts rhythmic brain networks during naturalistic movement

Abstract

Essential Tremor (ET) is a very common neurological disorder characterised by involuntary rhythmic movements attributable to pathological synchronization within corticothalamic circuits. Previous work has focused on tremor in isolation, overlooking broader disturbances to motor control during naturalistic movements such as reaching. We hypothesised that ET disrupts the sequential engagement of large-scale rhythmic brain networks, leading to both tremor and deficits in motor planning and execution. To test this, we performed whole-head neuroimaging during an upper-limb reaching task using high-density electroencephalography in ET patients and healthy controls, alongside optically pumped magnetoencephalography in a smaller cohort. Key motor regions-including the supplementary motor area, premotor cortex, posterior parietal cortex, and motor cerebellum-were synchronized to tremor rhythms. Patients exhibited a 15 % increase in low beta (14-21 Hz) desynchronization over the supplementary motor area during movement, which strongly correlated with tremor severity (R² = 0.85). A novel dimensionality reduction technique revealed four distinct networks accounting for 97 % of the variance in motor-related brain-wide oscillations, with ET altering their sequential engagement. Consistent with our hypothesis, the frontoparietal beta network- normally involved in motor planning-exhibited additional desynchronization during movement execution in ET patients. This altered engagement correlated with slower movement velocities, suggesting an adaptation towards feedback-driven motor control. These findings reveal fundamental disruptions in distributed motor control networks in ET and identify novel biomarkers as targets for next-generation brain stimulation therapies.

Keywords: Essential tremor; Motor control; Movement disorders; Oscillations; Reaching.

Figure 1. Structure of delayed reach-to-target task and kinematic analysis.

(A) Illustration of the sequence and timings of the reaching task. Each block began with 30 s of postural hold and eyes open rest. The reaching task consisted of rest (3 s), posture (3 s), onset of a directional cue (2.5 ±1 s), reach execution (max. 4 s), and sustained hold (1.5 s). The task followed a 2x2 design: high versus low uncertainty (HUC vs LUC), and large versus small targets (LRG vs SML). (B) An example set of trajectories from a single subject for the centre-out reaches, colour coded by the target. (C) The group averaged velocity profiles of the reaches, plot in normalized time from reach onset (0%) to hold onset (100%). (D-G) Boxplots summarising the between-subject statistics of task kinematics. Each point gives the deviation from the subject’s mean value. Data is shown for EEG experiment controls (red) and ET patients (blue), with colour matched bars indicating significant post-hoc t-tests. Black bars demark significant post-hoc t-test when comparing ET and controls. Brackets indicate tests that did not survive Bonferonni correction for multiple comparisons. Data shown for the EEG cohort only.

Figure 2. Source maps of OPM/EEG data coherence with the 4-12 Hz peripheral tremor signal estimated using a Dynamic Imaging of Coherent Sources (DICS) beamformer from postural hold data and movement related beta (14-30 Hz) power changes.

Subjects’ anatomy was flipped such that all slices were aligned contralateral to the dominant hand prior to performing statistics. (A and B) EEG and OPM image of the overlap of the top 90^th percentiles subject level coherence at source level. The slice location corresponds to the peak of the ET group average peak at the contralateral supplementary motor area (cSMA). (C and D) T-contrast of EEG and OPM DICS images comparing coherence in tremor band between control and ET. Maps were thresholded at the critical T-value (P < 0.05, uncorrected). Positive T-statistics indicate control > ET. (E and F) Same as (C and D) but for overlaps computed from the auxiliary DICS analysis, performed after regressing out the cSMA virtual channel. (G and H) Grand averaged EEG and OPM source power images across both controls and ET patients. Contrasts show unthresholded t-statistics using a baseline period at each stage of the reaching task.

Figure 3. Analysis of movement locked beta event related desynchronization (ERD) in the supplementary motor area (SMA) comparing between Essential Tremor (ET) patients and control recordings in EEG.

(A) The grand-averaged accelerometer trace indicates the sequence of the task. (B) Time courses of low beta power (14-21 Hz) of the combined left and right SMA, scaled as the percentage change relative to at rest (0%), shown across the postural hold, reach planning, execution, and the hold period (columns left to right). Data is shown as the group average ±SEM, for controls (red) or ET patients (blue). Bars indicate significant decrease between controls and ETs, determined by cluster permutation statistics (P < 0.05). Brackets indicating tests that did not survive Benjamini and Hochberg False Discovery Rate correction for multiple comparisons. (C) Same as (A), but for OPM data. (D) Scatter plots of log scaled tremor amplitude versus the percentage change in beta power. In the case of significant Pearson correlation, bold blue lines indicate the associated regression for ET patients only. The circled point indicates the outlier (identified according to Cook’s criteria) removed from regression.

Figure 4. Identification of large-scale, movement related brain circuits using time-frequency principal component analysis (tfPCA) applied to group averaged data (both ET and controls) derived from source-space projected EEG virtual electrodes.

(A) Time-frequency spectrograms were computed for each region of interest (ROI) in the network, incorporating the dorsal prefrontal cortex, primary and supplementary motor areas, posterior parietal cortex, and cerebellum VI. tfPCA was applied to the group averaged EEG data, with spectrograms concatenated for each motor epoch. Prior to tfPCA, data was log scaled, and Z-normalized per subject. Visualization of the coefficients of each PCA component 1-4 (red, blue, green, and purple, respectively) in both the (B) frequency and spatial domains. (C-F) Back projection of empirical data allows for visualisation of the time-frequency dynamics of each component (averaged in space), across the four different epochs of the reaching task (corresponding to the rows). (G) Explained variance of the data by the four components. (H) Reconstruction of the original time-frequency data using the four components.

Figure 5. Visualization of motor responsive, latent circuit dynamics and a comparison between controls and ET patients recorded with EEG.

Components were computed using tfPCA applied to the group averaged EEG data (Figure 4). These coefficients were then used to project data to trial-level latent dynamics that could be used to explore differences between controls and ET patients. The grand averaged accelerometer traces are shown at the top to indicate the task sequence. (A) The latent dynamics exhibited during the postural hold for each component (columns) are plot for ET (blue) and controls (red) separately. Bars and associated P-values indicate the outcome of cluster permutation statistics between the two experimental groups. Brackets indicate tests that did not survive Benjamini and Hochberg False Discovery Rate correction for multiple comparisons (16 tests in total). (B-D) Same as (A) but for cue presentation, reach execution, and the sustained hold. (E-H) Plots of the 2D latent trajectories (components 1 and 4, only) indicate highly correlated dynamics between controls and ETs with quantitative differences in the weighting of the components such as increased engagement of the prefrontal beta network in ET subjects (4^th component, apparent in plot G). An equivalent analysis was performed for OPM recordings (Supplementary Figure 7).

Figure 6. Analysis of modulation of latent circuit dynamics altered in Essential Tremor, with movement kinematics during the reaching task.

Latent states were projected to the single trial level, using coefficients computed from group averaged time frequency EEG data. Mean latent dynamics (reproduced from Figure 5) for controls (red) and ET patients (blue). Differences in average velocity/hold variability are shown for both controls and ET patients, whereas tremor parameters for ET patients only. Trials were split into 1^st (light colours) and 4^th quartiles (dark colours) and then subject averages compared at the group level using a cluster permutation test (indicated by thick lines above, and relevant P value for the test statistic). Brackets indicate tests that did not survive Benjamini and Hochberg False Discovery Rate correction for multiple comparisons (16 tests per parameter). (A-D) Plots of the latent dynamics of the 1^st component over the postural raise, cue onset, reaching, and hold, split by the tremor band power during each hold period (blue lines), the mean velocity during the postural raise/reach (green lines), and the hold variability (red lines). (E-H) Same as (A) but for the 4^th component. The 1^st and 4^th components are shown as these were most modulated by ET,the remaining components are presented in Supplementary Figure 8.

Figure 7. Hypothetical neuronal mechanisms underlying alterations to motor induced dynamics following pathological synchronization in Essential Tremor.

(A) During movement preparation (first column), neurons entrain to brain rhythms such as the sensorimotor beta rhythm (shown in red), but also tremor (blue). During movement (middle column), motor cortical neurons exhibit firing rate changes (assumed to be asynchronous; shown in green) that encode movement parameters. Simultaneously, ERD of the beta rhythm relinquishes units to participate in motor encoding, followed by a post movement beta rebound (final column). (B) In moderate ET, the tremor rhythm entrains a fraction of cortical neurons reducing the effective information encoding capacity of the available pool. To compensate, during movement, increased beta ERD frees up the neuronal “real estate” required to effectively encode movement.