Oscillatory activity in dystonia and its association with clinical severity — a reproducible framework for spectral peak definition and classification
DOI: Pending DOI
Synopsis
Oscillatory biomarkers in dystonia remain poorly defined, partly due to inconsistent criteria for identifying spectral peaks. We developed a framework to define and classify peak-like spectral components in local field potentials and applied it to awake, pre-activation recordings from 8 dystonia patients. High beta and low beta activity were most prevalent, but no feature was universally present and substantial inter-patient variability was observed. Peak presence was generally consistent across channels, while power varied. Peak power, computed at patient-specific frequencies or surrogate band centers when no peak was detected, was not significantly associated with clinical severity. These findings highlight heterogeneity in dystonia-related activity and the need for refined biomarkers.
Background
Dystonia is a neurological movement disorder characterized by involuntary movements and abnormal postures1. Although deep brain stimulation (DBS) is an established therapy, its effects are often delayed and variable across patients2. This highlights the critical need to determine if electrophysiologic signals from target nuclei could serve as reliable biomarkers to track symptom fluctuations and predict clinical improvement. While Parkinson’s disease has established biomarkers like beta oscillations, no equivalent biomarkers have been defined for dystonia3–5.
Prior studies have reported oscillatory activity in low-frequency and gamma bands in dystonia, but findings vary across cohorts and recording conditions6–8. A key limitation is the lack of clear and reproducible criteria for defining spectral peaks, making it difficult to determine which oscillatory features are consistently present across patients with diverse clinical phenotypes. In addition, most existing studies reported perioperative recordings, often performed under anesthesia and may not reflect chronic activity9. Therefore, characterizing local field potential (LFP) activity prior to stimulation activation is a critical first step for adaptive DBS development.
This study aimed to (1) formalize a reproducible framework for defining and classifying spectral peaks in LFP recordings, (2) characterize the prevalence and spatial distribution of these features in awake, pre-activation recordings from dystonia patients, and (3) evaluate whether LFP power relates to dystonia severity and other clinical characteristics.
Methods
We studied 8 consecutive isolated dystonia patients who underwent DBS implantation with Medtronic Percept capable of recording LFP activity (7 globus pallidus internus, 1 ventral intermediate nucleus of thalamus). LFPs were recorded in indefinite streaming mode during the initial programming visit prior to device activation while sitting at rest. Three bipolar montages per hemisphere (contacts 0-3, 1-3, 0-2) yielded up to 6 channels per patient (45 total channel observations; one patient had a unilateral implant) (Fig. A).
Signals were bandpass filtered (1-100 Hz), and the electrocardiogram artifact was removed using template subtraction10. Power spectral density was calculated using Welch's method11. The aperiodic component was modeled with Fitting Oscillations & One Over f Algorithm (FOOOF)12 over 4-90 Hz and subtracted from the log10 spectrum, resulting in a flattened residual spectrum on which all peak quantification was performed (Fig. B, C1). Peak-like spectral components were identified and classified using a custom framework, with detailed definitions and detection procedures described in the Results section.
Clinical severity was assessed using the Global Dystonia Rating Scale (GDRS)13 scored from video recording by a movement disorders neurologist. Dystonia was classified by etiology (genetic or idiopathic), body distribution (segmental or generalized), age of onset (<40 or >40 years)1. Associations between peak power and severity were evaluated using Pearson correlation. Group-level comparisons across age, distribution, and etiology were assessed using t-tests.
Results
Peak definition and detection framework. Peak frequencies were identified using a custom Gaussian-based framework. Candidate spectral features on the flattened spectrum were used to guide Gaussian modeling, and retained peaks after multiple thresholding steps were manually confirmed (Fig. C2). A peak was therefore defined as a discrete Gaussian-like spectral component in the flattened spectrum and was then assigned to a frequency band depending on identified peak frequency (theta (4-7Hz), alpha (8-12Hz), low beta (13-20Hz), high beta(20-35Hz), while gamma peaks (35-90Hz) were further categorized into narrowband and wideband patterns based on spectral spread (Fig. A). Broadband increase above 1/f baseline and low-frequency power at filter cutoff (1-3Hz) were defined as non-peak features reflecting distributed spectral elevation rather than discrete components (Fig. B).
Peak power was the primary feature used for analysis and was computed as the average power within ±1 Hz of the peak center frequency (except +/- 5Hz for wideband gamma) (Fig. C3). To ensure comparability across patients, when no peak was identified within a given band, a surrogate frequency was assigned at the band center. Peak power was calculated from both the original spectrum and the spectrum with aperiodic component removed, with absolute power used for primary analysis. For channel comparison, power values were normalized to 0-1 across six channels within each frequency band by dividing by the maximum power among 6 channels for each patient.
Application to the cohort. Dystonia etiology was genetic in 3 patients (2 TOR1A, 1 SGCE), idiopathic in 4, and 1 patient has pending genetic testing. Body distribution was segmental (n=4) and generalized (n=4). Age of dystonia onset was < 40 years (n=4) or >40 years (n=4), range 12-67. Total GDRS ranged from 11 to 37. The patient with DBS implant in the thalamus was excluded from the group-level clinical comparison.
Across patients, we observed peaks in all frequency bands, as well as low-frequency and broadband elevations (Fig. D). High beta peaks were the most prevalent feature (n=7), followed by low beta (n=4) and alpha (n=3). Theta, wideband, and narrowband gamma were less common (n=2, 2, 1). Broadband increase was observed in 4 of 8 patients, including two patients who showed only broadband elevation without discrete peaks. Low-frequency elevation was observed only in the thalamus patient.
Spatial distribution. Peak power followed a consistent spacing-dependent trend with the largest peaks in 0-3 channel, then 1-3 channel and the smallest peaks in 0-2 channel. This spatial gradient was more pronounced for absolute power than for aperiodic-removed measures. Normalized peak power was generally higher in left-hemisphere channels than in right-hemisphere channels, although this lateralization was not observed after aperiodic removal. Despite these group-level patterns, variability remained substantial, with large standard deviations and widely dispersed individual observations. Within a patient, the same peaks were typically observed across channels, but their amplitudes varied. In addition, peak characteristics could differ between hemispheres within the same patient. Detailed lead location reconstruction is pending.
Clinical correlations. Group-level comparisons across age (<40 vs >=40), body distribution (segmental vs generalized), and etiology (genetic vs idiopathic) did not reveal significant differences in power measures across frequency bands (Fig. E). Peak power or average band power was not significantly associated with dystonia severity scores in any frequency band. This was consistent for both absolute power and aperiodic-removed power.
Discussion
Peak definition and detection framework. Peak frequencies were identified using a custom Gaussian-based framework. Candidate spectral features on the flattened spectrum were used to guide Gaussian modeling, and retained peaks after multiple thresholding steps were manually confirmed (Fig. C2). A peak was therefore defined as a discrete Gaussian-like spectral component in the flattened spectrum and was then assigned to a frequency band depending on identified peak frequency (theta (4-7Hz), alpha (8-12Hz), low beta (13-20Hz), high beta(20-35Hz), while gamma peaks (35-90Hz) were further categorized into narrowband and wideband patterns based on spectral spread (Fig. A). Broadband increase above 1/f baseline and low-frequency power at filter cutoff (1-3Hz) were defined as non-peak features reflecting distributed spectral elevation rather than discrete components (Fig. B).
Peak power was the primary feature used for analysis and was computed as the average power within ±1 Hz of the peak center frequency (except +/- 5Hz for wideband gamma) (Fig. C3). To ensure comparability across patients, when no peak was identified within a given band, a surrogate frequency was assigned at the band center. Peak power was calculated from both the original spectrum and the spectrum with aperiodic component removed, with absolute power used for primary analysis. For channel comparison, power values were normalized to 0-1 across six channels within each frequency band by dividing by the maximum power among 6 channels for each patient.
Application to the cohort. Dystonia etiology was genetic in 3 patients (2 TOR1A, 1 SGCE), idiopathic in 4, and 1 patient has pending genetic testing. Body distribution was segmental (n=4) and generalized (n=4). Age of dystonia onset was < 40 years (n=4) or >40 years (n=4), range 12-67. Total GDRS ranged from 11 to 37. The patient with DBS implant in the thalamus was excluded from the group-level clinical comparison.
Across patients, we observed peaks in all frequency bands, as well as low-frequency and broadband elevations (Fig. D). High beta peaks were the most prevalent feature (n=7), followed by low beta (n=4) and alpha (n=3). Theta, wideband, and narrowband gamma were less common (n=2, 2, 1). Broadband increase was observed in 4 of 8 patients, including two patients who showed only broadband elevation without discrete peaks. Low-frequency elevation was observed only in the thalamus patient.
Spatial distribution. Peak power followed a consistent spacing-dependent trend with the largest peaks in 0-3 channel, then 1-3 channel and the smallest peaks in 0-2 channel. This spatial gradient was more pronounced for absolute power than for aperiodic-removed measures. Normalized peak power was generally higher in left-hemisphere channels than in right-hemisphere channels, although this lateralization was not observed after aperiodic removal. Despite these group-level patterns, variability remained substantial, with large standard deviations and widely dispersed individual observations. Within a patient, the same peaks were typically observed across channels, but their amplitudes varied. In addition, peak characteristics could differ between hemispheres within the same patient. Detailed lead location reconstruction is pending.
Clinical correlations. Group-level comparisons across age (<40 vs >=40), body distribution (segmental vs generalized), and etiology (genetic vs idiopathic) did not reveal significant differences in power measures across frequency bands (Fig. E). Peak power or average band power was not significantly associated with dystonia severity scores in any frequency band. This was consistent for both absolute power and aperiodic-removed power.
Disclosures
The authors have no relevant financial disclosures.
References
1. Albanese, A. et al. Definition and Classification of Dystonia. Movement Disorders 40, 1248–1259 (2025).
2. Vidailhet, M., Jutras, M.-F., Grabli, D. & Roze, E. Deep brain stimulation for dystonia. Journal of Neurology, Neurosurgery & Psychiatry 84, 1029–1042 (2013).
3. Piña-Fuentes, D. et al. Adaptive DBS in a Parkinson’s Patient with Chronically Implanted DBS: A Proof of Principle. Mov Disord 32, 1253–1254 (2017).
4. Little, S. et al. Adaptive deep brain stimulation in advanced Parkinson disease. Annals of Neurology 74, 449–457 (2013).
5. Stanslaski, S. et al. Sensing data and methodology from the Adaptive DBS Algorithm for Personalized Therapy in Parkinson’s Disease (ADAPT-PD) clinical trial. npj Parkinsons Dis. 10, 174 (2024).
6. Scheller, U. et al. Pallidal low‐frequency activity in dystonia after cessation of long‐term deep brain stimulation. Movement Disorders 34, 1734–1739 (2019).
7. Zhang, R. Balance between pallidal neural oscillations correlated with dystonic activity and severity - ScienceDirect. https://www.sciencedirect.com/science/article/pii/S0969996123001936 (2023).
8. Miocinovic, S. et al. Cortical gamma oscillations in isolated dystonia. Parkinsonism & Related Disorders 49, 104–105 (2018).
9. Lozano, A. M. et al. Deep brain stimulation: current challenges and future directions. Nat Rev Neurol 15, 148–160 (2019).
10. Woody, C. D. Characterization of an adaptive filter for the analysis of variable latency neuroelectric signals. Med. & biol. Engng. 5, 539–554 (1967).
11. Welch, P. The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms. IEEE Transactions on Audio and Electroacoustics 15, 70–73 (1967).
12. Donoghue, T. et al. Parameterizing neural power spectra into periodic and aperiodic components. Nat Neurosci 23, 1655–1665 (2020).
13. Comella, C. L. et al. Rating scales for dystonia: A multicenter assessment. Movement Disorders 18, 303–312 (2003).