Principal Investigator:
Dr. Ami Kumar,
Postdoctoral Research Scientist, Kuo Lab
Department of Neurology
Columbia University Irving Medical Center
New York, NY
SPECIFIC AIMS:
Essential tremor (ET) is a neurological disorder associated with cerebellar dysfunction, characterized by involuntary rhythmic movements (i.e., tremor). Approximately 7 million people in the United States suffer from ET, affecting their daily lives; however, the pharmacological therapies for ET are rather limited, leaving many disabled. To develop better therapies for ET, several clinical trials have been conducted in the past five years, but none of them were successful because of the heterogeneous responses of ET patients. We learned that some ET patients have dramatic responses while others do not respond at all, suggesting diverse brain circuit dysfunctions within the ET population. Therefore, we need a tool to determine the circuit abnormalities of each ET patient, enabling the development of targeted therapies tailored to individualized circuit dysfunctions.
Ample clinical and neuroimaging evidence have suggested that the cerebellar circuit is the key region for tremor generation in ET. To study the cerebellar circuit in ET, we recently developed a novel technique, cerebellar electroencephalography (EEG) (Science Translational Medicine, 2020, 2024, 2025). Using this technique, we identified that ET patients have four distinct types of cerebellar oscillatory activity patterns. Among 26 ET patients studied (Preliminary data), we discovered 46.2% of ET patients have Type 1 cerebellar oscillatory activity that matches with cerebellar physiology of a mouse model with cerebellar synaptic over-growth. Whereas 26.9% of ET patients have Type 2 cerebellar oscillatory activity that resembles the cerebellar physiology of a mouse model with increased inferior olive gap junction coupling. These findings suggest that cerebellar EEG has the potential to determine the underlying cerebellar circuit abnormalities in ET. Additionally, Type 3 (15.4%) and Type 4 (11.5%) may have less cerebellar involvement, suggesting that the tremor may come from other brain regions. However, we still do not know the percentage of each type in a larger ET population, their clinical correlates, or their responses to therapies, and this knowledge will serve as the basis to stratify ET patients based on their cerebellar circuit abnormalities. In addition to the four types of cerebellar EEG patterns based on oscillatory activity, cerebellar EEG also contains highly enriched and dynamic physiological information, that can be used to refine the subtype classification. Therefore, we will implement a machine learning approach to further explore ET circuit heterogeneity. Collectively, these insights will enable us to understand the individual circuit differences in the ET population and will help in the development of precise therapeutic targeting.
TEAM:
This proposal will provide training of a postdoctoral research scientist with expertise in EEG and data analysis (Ami Kumar, PhD), mentored by an ET physician-scientist (Sheng-Han Kuo, MD), EEG physician scientist (Ming-Kai Pan, MD, PhD) and machine learning expert (Lewis Tomalin, PhD) to test the hypothesis that ET subtypes based on their distinct cerebellar EEG patterns correspond to unique clinical characteristics and can be classified using machine learning.
Aim 1a: To determine the prevalence of each ET subtype and its association with clinical variables.
Rationale: We still do not know the percentage of each subtype in the ET population or if each ET subtype has different tremor or clinical characteristics, the essential first step in understanding the circuit heterogeneity of ET.
Approach: We will record cerebellar EEG in a large cohort of ET patients (n = 150) to determine the percentage of each subtype and assess the association with factors such as age, gender, age of tremor onset, tremor severity, cognition, and self-reported alcohol and ET medication responsiveness.
Aim 1b: To determine the association of each ET subtype with tremor characteristics and gait
patterns measured by wearable sensors.
Rationale: We still do not know if associated clinical features of ET, such as tremor characteristics and gait patterns, would differ in each ET subtype.
Approach: We will use wearable sensors to measure tremor variability, frequency, amplitude, and gait metrics in 150 ET patients.
Aim 2: To determine if a machine learning approach can further refine ET subtypes.
Rationale: Our ET subtyping is based on the cerebellar oscillatory activity pattern, but cerebellar EEG contains enriched physiological features that can be used to refine the subtyping.
Approach: We will use the EEG data obtained from 150 ET patients and employ an unsupervised machine learning clustering algorithm, such as the k-means clustering, for refining and validating the subtypes.
TIMELINE:
Aim 1a & Aim 1b: Recruitment, and data analysis Year 1- then first half of Year 2,
Aim 2: Year 3. Prepare for NIH K99 or R01 submission for longitudinal EEG biomarker validation and manuscript writing and submission in Year 3.
IMPACT: This study will advance our understanding of ET heterogeneity and pave the way for the development of tailored therapeutic strategies based on the cerebellar circuit dysfunction. The proposal will also provide cerebellar EEG/clinical training for Dr. Kumar to become an independent ET researcher.
Read the full proposal here.
Progress Reports
November 2025
Project Overview
Essential tremor (ET) affects nearly 7 million people in the U.S. and causes disabling rhythmic movements. Current treatments are often ineffective, as patients have heterogeneous responses, likely due to differences in underlying brain circuits. To address this challenge, we need a tool to probe the circuit heterogeneity of ET. We developed cerebellar electroencephalography (EEG), a novel method for measuring human cerebellar physiology, and identified four distinct oscillatory patterns that suggest ET subtypes. Two of these patterns correspond to well-characterized mouse models with known circuit alterations, supporting cerebellar EEG as a tool for probing circuit dysfunction. However, we still do not know the prevalence of each ET subtype and its associated clinical features. Therefore, we propose to study a larger group of 150 ET patients (Year 1: 60, Year 2: 60, and Year 3: 30) to determine the prevalence of each subtype and its relationship to tremor characteristics such as tremor frequency and amplitude, as well as associated clinical features, including cognitive impairment and gait abnormalities. Finally, we will apply a machine learning approach to sort out complex cerebellar EEG patterns for ET subtype classification, potentially refining and validating subtypes. By identifying distinct ET subtypes, this study will enhance our understanding of ET heterogeneity and serve as a foundation for developing personalized treatment strategies tailored to each patient’s unique brain circuit dysfunction.
Aim 1a: To determine the prevalence of each ET subtype and its association with clinical variables. Rationale: We still do not know the percentage of each subtype in the ET population or if each ET subtype has different tremor or clinical characteristics, the essential first step in understanding the circuit heterogeneity of ET. Approach: We will record cerebellar EEG in a large cohort of ET patients (n = 150) to determine the percentage of each subtype and assess the association with factors such as age, gender, age of tremor onset, tremor severity, cognition, and self-reported alcohol and ET medication responsiveness.
Aim 1b: To determine the association of each ET subtype with tremor characteristics and gait patterns measured by wearable sensors. Rationale: We still do not know if associated clinical features of ET, such as tremor characteristics and gait patterns, would differ in each ET subtype. Approach: We will use wearable sensors to measure tremor variability, frequency, amplitude, and gait metrics in 150 ET patients.
Aim 2: To determine if a machine learning approach can further refine ET subtypes. Rationale: Our ET subtyping is based on the cerebellar oscillatory activity pattern, but cerebellar EEG contains enriched physiological features that can be used to refine the subtyping. Approach: We will use the EEG data obtained from 150 ET patients and employ an unsupervised machine learning clustering algorithm, such as the k-means clustering, for refining and validating the subtypes.
Accomplished activities for the study Aims (June-November 2025)
1. Personnel hiring, training, and workflow development
• Successfully hired a research coordinator (Ian Smith) to support participant scheduling, data collection, preprocessing, quality control, and documentation.
• Conducted comprehensive training for the research coordinator in EEG setup, troubleshooting, data management protocols, artifact identification, patient interaction, and safety procedures.
• Established a standardized workflow for patient recruitment, screening, and consent.
• Completed full task setup, creating detailed documentation sheets (checklists and data-tracking systems). Each sheet is designed to be concise and clear, allowing for the capture of data thoroughly and accurately. Sheets are dated and labeled with the subject ID, ensuring that each document is easily identifiable. These resources guide every step of the clinic workflow, including equipment setup, patient preparation, task execution, and data logging, ensuring accurate and fully standardized procedures across all patient visits.
2. Regulatory Approvals
• Obtained IRB approval, including amendments to increase patient numbers, update consent language, and add the research coordinator to the study protocol (IRB-AAAU0162).
• Completed all required institutional and compliance documentation.
• Finalized study consent forms, data-sharing agreements, and updated clinical assessment measures to be incorporated into the study workflow.
• Created recruitment flyers for patient enrollment and obtained necessary approvals before distribution.
3. Recruitment
• Began screening patients based on the project’s inclusion and exclusion criteria. (The inclusion criteria of ET: 1) The diagnosis of ET by a trained movement disorders neurologist based on the 2018 MDS classification. 2) absence of clinically significant dystonia, Parkinsonism, and ataxia. Exclusion criteria: 1) a medical history of cerebellar injury, alcohol abuse, or chemotherapy exposures. 2) a family history of cerebellar ataxia. 3) the presence of head tremor, because head tremor may create artifacts for cerebellar EEG measurement.)
• Reached out to movement disorder ET specialists in the clinic for patient referrals and posted the study on recruitment platforms of Columbia University.
4. EEG recordings
• As part of Aim 1a, we have recorded 10 ET patients, with additional participants actively scheduled for the coming weeks.
• Each recorded session included clinical assessments using the Essential Tremor Rating Assessment Scale (TETRAS) to evaluate tremor severity and functional disability, and the Montreal Cognitive Assessment (MoCA) for detecting cognitive impairment in ET. We also collected patient demographics and relevant clinical data, including age, age of tremor onset, self-reported responsiveness to alcohol, and current ET medication use.
• Cerebellar EEG was recorded during resting-state and motor tasks.
Preliminary Results: Preliminary assessments of data quality, signal stability, and recording consistency using Brainstorm software confirm that the cerebellar EEG recordings are good and suitable for further analyses. All patient datasets have undergone full preprocessing, including band pass filtering, artifact detection, including removal of ocular, muscular, and electrical-noise components. The resulting clean signals will be used to extract spectral power for task-related activity across sessions, helping to identify subtypes in ET patients.
5. Body sway and gait measurements
• As part of Aim 1b, the recruited 10 ET patients underwent body sway and gait measurements using wearable sensors.
Preliminary Results: Among the first 10 recorded patients, 50% showed increased body sway during feet-apart trials with eyes open, 50% during feet-apart trials with eyes closed on a firm surface, and 60% during feet-together trials with eyes closed. For gait assessments, 40% exhibited increased stance width, and 20% demonstrated increased toe-out angle. These early findings indicate that measurable balance and gait abnormalities are already evident in a proportion of ET patients.
6. Next Steps (Next 6 Months)
• Continue recruitment and EEG recordings for Aim 1a, targeting an additional 50 ET patients to complete Year 1 enrollment.
• Perform spectral analysis on all EEG recordings.
• Initiate group-level statistical analyses and ET subtype classification.
• Once patterns are identified, conduct correlation analyses between EEG subtypes and body sway/gait measurements for Aim 1b.
7. Summary
The project is progressing as planned. Key milestones, including research coordinator hiring, IRB approval, and initial data collection have been successfully completed. The study is now fully operational, and the team is well positioned for accelerated recruitment and analysis in the upcoming phase.
Activities for the Study Aims (December 2025 -June 2026)
1. Recruitment and EEG recordings
- Patients are being continuously screened and recruited according to the project’s inclusion and exclusion criteria. To facilitate recruitment, we collaborated with IETF, which distributed an email to patients within a 50-mile radius of New York. This outreach generated substantial interest, and many patients subsequently contacted our study team.
- We have also engaged the General Neurology clinics and requested referrals of eligible patients who may be appropriate for study participation.
- Based on current recruitment efforts, we have enrolled and recorded data from 30 patients with ET, with additional participants scheduled for evaluation in June. For all enrolled participants, we have collected cerebellar electroencephalography (EEG) recordings during motor tasks, as well as body sway and gait measurements using wearable sensors, as outlined in Aims 1a and 1b.
2. Preliminary results for Aim 1a
2.1 Patient demographics
A total of 20 participants were analyzed. The mean age was 63.75 ± 14.64 years, the mean age at disease onset was 48.70 ± 19.00 years, and the mean disease duration was 15.05 ± 12.54 years. The cohort consisted of 60% males and 40% females. Clinical severity was assessed using the Essential Tremor Rating Assessment Scale (TETRAS), which yielded a 2 mean Activities of Daily Living (ADL) score of 17.10 ± 7.54, a mean performance score of 8.27 ± 3.32, and a mean total TETRAS score of 25.25 ± 9.99. Global cognitive function, was further evaluated using the Montreal Cognitive Assessment (MoCA) in this cohort, showing a mean score of 26.95 ± 2.35.
2.2 Cerebellar EEG preliminary results
Cerebellar EEG data were preprocessed and cleaned to remove ocular, muscular, and electrical artifacts, and power spectral density (PSD) was calculated from the artifact-free data. For each participant, PSD values within the tremor frequency range (4-12 Hz) were extracted and compared with the mean PSD of the non-tremor frequency range (13-20 Hz). A threshold-based algorithm was then applied to determine the presence or absence of a spectral peak within the tremor frequency range during both action and rest conditions, enabling objective classification of subjects based on their cerebellar oscillatory activity pattern. Among the 20 participants analyzed, the distribution across subtypes was as follows:
- Type 1 (cerebellar oscillatory activity during both action and rest): 60% of participants
- Type 2 (oscillatory activity during action only): 10% of participants
- Type 3 (oscillatory activity during rest only): 15% of participants
- Type 4 (minimal oscillatory activity during both action and rest): 15% of participants
These findings demonstrate that cerebellar EEG can identify distinct oscillatory patterns in ET patients, with Type 1 being the predominant subtype in this preliminary cohort.
2.3 Accelerometer-based tremor recordings
Tremor recordings were obtained using an accelerometer positioned on the index finger of the dominant tremor hand and analyzed to quantify peak tremor frequency (Hz) during rest and action conditions. Given the preliminary nature of these data and small subtype sample sizes (Type 1: n = 12; Type 2: n = 2; Type 3: n = 3; Type 4: n = 3), findings are reported as descriptive observations only. Mean action tremor frequency was 3.62 Hz for Type 1, 5.88 Hz for Type 2, 3.30 Hz for Type 3, and 1.53 Hz for Type 4, compared to resting frequencies of 7.24 Hz, 7.21 Hz, 9.01 Hz, and 6.49 Hz, respectively. We observed that, Type 2 appeared to show the highest tremor frequency during action while Type 4 showed the lowest, whereas tremor frequency was broadly similar across subtypes during rest. Type 3 showed the most notable difference between rest and action conditions (9.01 Hz vs. 3.30 Hz). These preliminary patterns require further investigation in the full cohort.
3. Preliminary results for Aim 1b
3.1 Postural sway
Postural sway was assessed in ET patients across three standardized balance conditions using APDM wearable sensors placed on both feet, both hands, the chest, and the lumbar region. Both RMS sway (m²/s²) and sway area (m²/s⁴) were quantified for each condition. Given the preliminary nature of these data and small subtype sample sizes (Type 1: n = 12; Type 2: n = 2; Type 3: n = 3; Type 4: n = 3), findings are reported as descriptive observations only.
- Condition 1 (eyes open, feet apart, firm surface): Sway was minimal across all subtypes. RMS sway ranged from 0.04 m²/s² (Type 3) to 0.10 m²/s² (Type 1), and sway area ranged from 0.01 m²/s⁴ (Type 3) to 0.04 m²/s⁴ (Type 1), with Type 1 showing the highest values across both measures.
- Condition 2 (eyes closed, feet apart, firm surface): Removal of visual input led to increased sway across all subtypes, indicating greater reliance on somatosensory and vestibular inputs for postural control. RMS sway ranged from 0.06 m²/s² (Type 3) to 0.16 m²/s² (Type 1), and sway area ranged from 0.03 m²/s⁴ (Type 4) to 0.14 m²/s⁴ (Type 3), with Type 3 showing the highest sway area despite Type 1 having the highest RMS sway.
- Condition 3 (feet together, eyes closed, firm surface): Sway was markedly elevated across all subtypes. RMS sway ranged from 0.11 m²/s² (Type 3) to 0.23 m²/s² (Type 1), and sway area ranged from 0.10 m²/s⁴ (Type 3) to 0.49 m²/s⁴ (Type 1), with Type 1 showing the highest values on both measures and Type 3 consistently the lowest. Narrowing the base of support combined with visual deprivation resulted in marked instability across all subtypes, consistent with impaired balance control under sensory-constrained conditions.
These preliminary findings from cerebellar EEG recordings and postural sway measurements demonstrate the successful implementation of the study approach and establish the analytical pipeline needed to support the proposed study aims.