Adaptive Deep Brain Stimulation Synchronized to Gait Phase Improves Walking and Reduces Falls in Parkinson's Disease
核心洞察
A novel adaptive deep brain stimulation (搜索) (aDBS) system that modulates stimulation amplitude in real time during the leg swing phase of walking was developed and tested in five Parkinson's patients.
The gait-phase-synchronized aDBS improved step-length symmetry by 3.5% and reduced step-length variability by up to 39%, with a significant reduction in falls (odds ratio 4.35, P=0.047) compared to continuous DBS.
Personalized neural biomarkers derived from cortical and pallidal circuits outperformed canonical frequency bands, with optimal control signals identified from the pallidum in four hemispheres and motor cortex in five of nine hemispheres.
A research team at the University of California, San Francisco has developed and clinically tested a first-in-class adaptive deep brain stimulation (搜索) (aDBS) system that synchronizes stimulation amplitude modulation to specific phases of the gait cycle in patients with Parkinson's disease (搜索) (PD). The system, which operates on a subsecond timescale, represents a fundamental departure from existing aDBS approaches that have largely operated on slower timescales of seconds to minutes and targeted canonical beta activity without consideration of movement phase.
The findings, published in Nature Medicine, demonstrate that gait-phase-synchronized aDBS can improve walking symmetry, reduce step variability, and significantly decrease falls compared to conventional continuous DBS (cDBS).
Study Design and Personalized Biomarker Discovery
Five patients with PD and gait impairments underwent implantation of a bidirectional neural stimulator (Summit RC + S, Medtronic) with quadripolar depth electrodes targeting the internal segment of the globus pallidus (GPi) and subdural electrocorticography paddles placed over motor cortical areas. Four patients received bilateral implants, while one received unilateral implantation due to predominantly unilateral symptoms.
The research team recorded local field potentials (LFPs) from pallidal and cortical sites during overground walking at self-selected speeds, analyzing spectral power across canonical frequency bands—theta (4–7 Hz), alpha (7–12 Hz), low beta (13–20 Hz), high beta (20–30 Hz), and low gamma (30–50 Hz). Substantial gait-phase-dependent modulation was observed across both ipsilateral and contralateral swing phases.
A critical innovation was the use of a data-driven grid search across all recording locations and frequency band combinations between 2.5 Hz and 50 Hz to identify patient-specific biomarkers of contralateral leg swing. "Compared to canonical bands, these individualized frequency bands revealed subfrequency ranges that better distinguished contralateral swing phase and, in most cases, achieved higher decoding accuracy," the authors reported.
The origin of optimal biomarkers varied considerably: the pallidum provided the best control signal in four hemispheres, the premotor cortex in three hemispheres, and M1 and S1 in one hemisphere each. Notably, M1 exhibited the highest proportion of hemispheres with at least one discriminative canonical frequency band (eight of nine), followed by the pallidum (five of nine).
Real-Time Gait-Phase Detection and Stimulation Control
Patient-specific biomarkers were embedded into each patient's neural stimulator using an on-device linear discriminant analysis classifier. Two critical parameters—ramping rate and amplifier blanking—were titrated individually. Amplifier blanking was set to 1 ms, sufficient to minimize ramping-related artifacts.
In real-time testing, the stimulators accurately detected subsecond gait-phase transitions, switching stimulation amplitude between 0.5× and 1.0× the clinically optimized setting during contralateral leg swing. Across patients, stimulation amplitude was significantly increased during contralateral leg swing relative to chance, with an average increase of 18.6%.
Biomarker stability was assessed by comparing spectral power across clinical DBS optimization and aDBS testing phases. Effect-size analysis revealed mostly negligible-to-moderate differences: among 19 comparisons, 2 were negligible, 5 were small, 7 were moderate, and 5 were large. "Overall, patient biomarker spectral power varied modestly across timepoints, with a high degree of overlap between distributions," the researchers noted.
Acute Gait Improvements with aDBS
Group-level analyses indicated improved gait symmetry under aDBS compared to cDBS. Wilcoxon rank-sum tests identified significant differences in step-length asymmetry (n=383 versus n=423, W=109,982, P<0.001) and step-time asymmetry (n=390 versus n=476, W=100,426, P<0.05). Median step-length asymmetry showed a 3.5% decrease with a 3.1% reduction in interquartile range.
Bilateral reductions in median step-length variability reached 39.1% (left) and 31.6% (right), while step-time variability reductions were 16.0% (left) and 26.6% (right), though these changes did not reach statistical significance (W=18–22, P=0.056–0.310).
Individual responses varied. Three patients demonstrated significant improvements in step-length symmetry, shifting medians toward 0% and reducing variability by an average of 2.9% (W=958–2,067, P<0.0001). Patient 1, who experienced chronic instability of effective cDBS settings and severe dyskinesia requiring 520 days of clinical optimization, showed no significant improvement.
Double-Blind Crossover Trial: Reduced Falls
Three patients (patients 2, 3, and 4) completed a double-blind crossover experiment randomizing three conditions: cDBS, ramp-up aDBS (RU-aDBS; 0.5× to 1.0× clinically optimized amplitude during contralateral leg swing), and ramp-down aDBS (RD-aDBS; 1.0× to 0.5× during contralateral leg swing). Each condition was maintained for 8–10 days.
RU-aDBS was associated with a significant reduction in reported fall frequency compared to cDBS (odds ratio=4.35, P=0.047, 95% CI=1.07–20.22), indicating a higher likelihood of patients reporting lower fall-frequency categories. RD-aDBS had no significant effect on falls. No group-level changes were observed for freezing episodes.
At-home gait metrics captured via wearable triaxial accelerometers revealed that both aDBS conditions improved stride length and step symmetry compared to cDBS. RD-aDBS showed a larger effect on stride length (4.7%) than RU-aDBS (0.9%), while RU-aDBS showed greater reduction in step asymmetry (2.37%) than RD-aDBS (0.36%).
Blinded clinical assessments at the end of each testing phase showed that both aDBS strategies generally maintained or improved global motor control relative to cDBS, as measured by MDS-UPDRS-III scores.
Parallel Findings: Activity-Dependent STN Dynamics
In a complementary study also published in Nature Medicine, an independent research group characterized how subthalamic nucleus (搜索) (STN) dynamics encode locomotor activities across therapeutic conditions in 35 patients with advanced PD. This study found that STN dynamics encode key locomotor activities—sitting, standing, walking, and obstacle avoidance—through specific spectral modulations, with gamma power increasing and high-beta power decreasing during transitions to more demanding activities.
Critically, the researchers demonstrated that L-DOPA and STN DBS exert opposing effects on activity-dependent STN dynamics: L-DOPA amplified activity-dependent modulations in high-beta and gamma power, while STN DBS reduced power levels in these bands. These therapy-specific shifts prevented a single neural decoder from generalizing across therapeutic conditions.
To address this, the team developed a modular decoding framework integrating therapy-specific decoders with a classifier that dynamically switches between them based on the ongoing therapeutic state. This framework was then implemented in a feasibility clinical trial (AdapGAIT, NCT06791902) in four patients, demonstrating that activity-dependent DBS could alleviate locomotor deficits—including freezing of gait, leg dystonia, and reduced walking endurance—while preserving efficacy for cardinal motor symptoms.
Clinical Significance and Future Directions
The gait-phase-synchronized aDBS approach addresses a major clinical challenge: improving advanced gait disturbances with traditional DBS. The authors propose that continuous stimulation, while beneficial for certain cardinal symptoms, may interfere with physiological timing signals needed for dynamic balance and limb coordination—a concept they term the "information lesion" hypothesis.
"By targeting stimulation to the precise temporal windows when pathological activity is most disruptive, this approach overcomes the key limitations of cDBS, extends adaptive neuromodulation beyond conventional beta-state-based paradigms and reveals mechanistic insights into the circuit-level control of locomotion," the researchers concluded.
Study limitations include the small sample size reflecting the invasive nature of device implantation, the use of single-frequency band detection (future implementations may incorporate multiple bands), and heterogeneous clinical responses. Patient 1, who exhibited severe dyskinesia, showed no improvements, highlighting the need for longer adaptation periods in future studies.
The integration of home-based gait monitoring via wearable sensors further established real-world relevance, capturing benefits not consistently detectable during brief in-clinic testing. These findings position movement-state-driven aDBS as a viable and potentially transformative therapy for gait disorders in Parkinson's disease (搜索).