A Convergent Pathway for Stimulation‐Induced Dyskinesia Following Deep Brain Stimulation

Abstract

Background

Stimulation‐induced dyskinesias (SID) from deep brain stimulation (DBS) of the subthalamic nucleus (STN) and globus pallidus internus (GPi) are uncommon; however, they are increasingly recognized. Once considered transient and indicative of effective neuromodulation, SID are now seen as potential therapy‐limiting side effects, akin to internal capsule activation. The mechanism and anatomical basis for SID remain poorly understood.

Methods

We conducted a retrospective study of individuals with Parkinson's disease with STN or GPi DBS who experienced SID in the dopaminergic medication OFF state during the monopolar review 1‐month post‐implantation.

Results

We analyzed 137 monopolar stimulation observations (105 GPi, 32 STN). In the GPi cohort, discriminative fiber tract analysis showed a strong association between SID and the modulation of subthalamo‐pallidal fibers. This correlation was confirmed using leave‐one‐out and five‐fold cross‐validation. We further validated this model by predicting SID in independent STN and GPi cohorts, with the GPi‐based model accounting for significant variance in SID occurrence in both cohorts.

Conclusions

SID from STN or GPi DBS likely shares a common pathway via subthalamo‐pallidal connectivity. DBS modulation of these fibers correlates with SID, as confirmed by multiple cross‐validation methods. These findings suggest that the fibers are part of a more extensive and yet‐to‐be‐fully‐characterized dyskinesia network. © 2025 The Author(s). Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

1. Introduction

Deep brain stimulation (DBS) of the internal pallidum (GPi) or subthalamic nucleus (STN) is an effective neuromodulatory therapy for many of the refractory motor symptoms in Parkinson's disease (PD) inclusive of both medication‐related fluctuations and dyskinesia. ¹ , ² However, in a minority of cases, the treatment has resulted in stimulation induced dyskinesias (SID), even in the absence of treatment with dopaminergic replacement therapy. ³ , ⁴ , ⁵ While classically associated with STN DBS for the treatment of PD, SID has become increasingly recognized in GPi DBS. ⁶ , ⁷ The underlying mechanism and circuits responsible for SID remain incompletely understood, and it is unknown whether STN‐ and GPi‐induced dyskinesia may share the same pathophysiology.

Historically, levodopa‐induced dyskinesias have been attributed to changes in the striatal outflow pathway resulting from both neurodegeneration and the pulsatile nature of oral levodopa exposure. ⁸ , ⁹ The emergence of dyskinesias has been hypothesized to relate to internal communication loops within the STN and the GPi as well as the globus pallidus externus (GPe). ¹⁰ We posited that the unintended modulation of the internal loops between the STN and the GPi/GPe complex may explain SID in both the STN and GPi targets.

Further, studies in rodents have identified striato‐pallidal projection neurons as crucial contributors to the development of levodopa‐induced dyskinesia. ¹¹ , ¹² Electrophysiological studies point to the conclusion that dyskinesias are associated with changes within the basal ganglia and other brain regions. ¹³ Interestingly, activation of the dorsolateral striatum in parkinsonian mice has been demonstrated to induce dyskinesias. ¹⁴ , ¹⁵ It has, however, remained a mystery why optogenetic inactivation of the striatum has resulted in limited success in suppressing dyskinesia. ¹⁶ , ¹⁷

These findings collectively suggest that the network underlying PD dyskinesia is complex and involves multiple nodes. While the striatum may be a central component of this network, other regions and pathways likely play significant roles in dyskinesia manifestation. This study characterized the connectivity profile and neural network underlying SID in PD DBS using human data from STN and GPi brain targets.

2. Methods

2.1. GPi DBS Cohort

We conducted a retrospective study that included PD patients who underwent GPi DBS at the University of Florida. This study was approved by the University of Florida's institutional ethics committee (IRB201901807). Preoperative clinical characteristics, surgical procedure details, and postoperative programming parameters were collected and analyzed. PD patients who experienced SID during DBS monopolar review in an outpatient setting were included in this study. A monopolar review was conducted, and each contact on the DBS lead was mapped for benefit and persistent side effect profiles in ring mode. Directional stimulation was not evaluated in these subjects. SID was documented as the persistent stimulation‐induced side effect if the clinician programmer observed onset of dyskinesias after increasing the stimulation amplitude above a certain threshold and resolution of dyskinesias after decreasing the stimulation amplitude below that threshold. The programmers were not blinded to the patient's preoperative dyskinesia status during monopolar review. This information was collected approximately 1 month after DBS lead implantation in the dopaminergic medication OFF state. Incomplete or unclear programming documentation of the postoperative monopolar review led to exclusion from the dataset.

2.2. Image Processing

The Lead‐DBS software package was used for image processing utilizing previously published methods. ¹⁸ , ¹⁹ Briefly, the postoperative high‐resolution, non‐contrast, brain computed tomography (CT) scan was co‐registered to the preoperative T1‐weighted MPRAGE magnetic resonance imaging (MRI) brain using a two‐stage linear registration using Advanced Normalization Tools (ANTs). ²⁰ The pre‐ and postoperative images were spatially normalized into MNI_ICBM_2009b_NLIN_ASYM template space using the symmetric normalization (SyN) registration approach implemented in ANTs. The DBS leads were spatially localized using the PaCER method within the Lead‐DBS software. All lead locations were manually verified, and corrections were applied if necessary. ²¹ The DBS lead models used were specific to the device manufacturer for each patient.

The electric field (EF) and volume of tissue activated (VTA) were estimated using finite‐element modeling within the Lead‐DBS software. The EF was generated over a tetrahedral mesh head model defined as an isotropic volume with a symmetric conductivity of 0.14 S/m. ²² An EF threshold of 0.2 V/mm was used to determine the VTA boundary. The EFs from the right brain hemisphere were nonlinearly warped to the left hemisphere based on the MNI_ICBM_2009b_NLIN_ASYM template.

2.3. Discriminative Fiber Tract Analysis

Discriminative fiber tract analysis (also referred to as DBS fiber filtering) was conducted using the EFs generated following previously published methods. ¹⁹ This method utilized a reference connectome to create a fiber tract population. In this study, we selected the Basal Ganglia Pathway Atlas as the reference connectome. ²³ For each fiber tract in the connectome, we calculated the maximum electrical magnitude projected by each EF onto the tract. Each EF was also assigned a binary designation of whether or not it was associated with SID. Then for each tract, we computed a two‐sample T‐test between the EF magnitudes of those associated with SID and those not associated with SID. The resulting T‐score, or ‘fiber score,’ was used to measure the extent to which a fiber tract was modulated by SID and was then compared with non‐SID EFs. This process was iterated through every fiber tract in the connectome. Fibers were labeled dyskinesia fibers if the fiber score was greater than 0. The fiber tract population was further refined to only consider tracts that spatially intersected with at least 20% of the EFs and intersected with an EF magnitude greater than 0.2 V/mm. After the dyskinesia fibers were defined, we labeled each fiber by comparing the three‐dimensional coordinates of each tract with the coordinates specified in the Basal Ganglia Pathway Atlas.

2.4. Cross‐Validation

We employed two cross‐validation methods to validate our connectivity findings related to SID. First, we applied a leave‐one‐out cross‐validation approach to predict whether a fiber was associated with SID based on its fiber score. ²⁴ In this method, each fiber was individually analyzed by leaving one fiber out of the dataset each time. The model was trained on the remaining fibers and tested on the excluded fiber, and we repeated this process for every fiber in the dataset to ensure that each fiber was used once as a test case.

To further validate the robustness of our findings, we implemented a second method using a five‐fold cross‐validation scheme. Here, the dataset was divided into five equal parts. Four parts were used in each iteration to train the model, while the remaining part was used for testing. This process was repeated five times, each part serving as the test set once. The results from all five iterations were then averaged to provide a comprehensive validation of the model's performance.

2.5. Independent STN DBS Cohort

We aimed to compare the connectivity profiles of GPi SID and STN SID through an exploratory analysis of basal ganglia connectivity. We conducted a second retrospective study involving PD patients who underwent STN DBS at the University of Florida to achieve this goal. We applied the same inclusion and exclusion criteria for the GPi cohort to ensure consistency between the two groups. This approach facilitated a direct comparison of the connectivity patterns associated with GPi SID and STN SID.

2.6. Cross‐Validation Via an Independent Target

The STN cohort used for the study underwent the same image processing and discriminative fiber tract analysis as the GPi cohort, resulting in a second set of dyskinesia fibers and fiber scores derived from the STN DBS cohort. Although the underlying connectome is the same, the fibers generated using the STN cohort represent a completely different set of fibers than the GPi cohort. We then used the dyskinesia fibers from the GPi DBS cohort predict whether stimulation volumes in the STN DBS cohort were associated with SID. This approach facilitated an application of the insights gained from the GPi cohort to evaluate the STN cohort's connectivity patterns in relation to SID. It further demonstrated the generalizability of the fibers identified from the GPi cohort. By applying the GPi‐derived model to a separate cohort from another DBS target, we sought to validate that the identified fibers were robust and applicable across different basal ganglia nodes.

2.7. Independent GPi DBS Cohort

Once the model was validated, a final test on completely unseen data was performed. Three patients with PD who underwent GPi DBS (two bilateral, one unilateral) at the University of Florida Fixel Institute and who were not included in the original dataset were included in this analysis. The three patients underwent imaging analysis and DBS monopolar review in an identical fashion to the original cohort. The model was defined to predict the development of acute SID if there was any voxel overlap between the dyskinesia fibers and the VTAs generated from the monopolar review data.

3. Results

A total of 24 PD patients who experienced SID during the monopolar review were analyzed in this study. Of these, 17 patients had GPi DBS, with 10 receiving bilateral implants and 7 receiving unilateral implants. All of the implanted GPi DBS leads had 1.5 mm axial spacing between contacts. The SID appeared on one or both sides when each DBS lead was individually tested in a monopolar review. When testing one lead in bilateral cases, the second lead was inactivated. The remaining 7 patients in the cohort had STN DBS, with 2 receiving bilateral implants and 5 receiving unilateral implants. The same criteria for SID were applied to the STN cases as the GPi cases. All of the implanted STN DBS leads had 0.5 mm axial spacing between contacts. Combined, these two cohorts represented 137 (105 GPi, 32 STN) observations of monopolar stimulation with persistent stimulation‐induced side effects. Baseline demographics for the GPi DBS and STN DBS groups are summarized in Table 1 and Table 2, respectively.

TABLE 1.

Baseline characteristics of the globus pallidus internus deep brain stimulation cohort.

Characteristic
Age (years)	65.0 (8.8)
Male gender (%)	65
Bilateral DBS (%)	59
Pre‐DBS dyskinesia (%)	82
Pre‐DBS UPDRS‐III total OFF	34.7 (13.1)
Pre‐DBS UPDRS‐III total ON	21.2 (13.9)

Note: Values are reported as mean (standard deviation) unless otherwise specified.

Abbreviations: DBS, deep brain stimulation; UPDRS‐III, Unified Parkinson's Disease Rating Scale‐Part III.

TABLE 2.

Baseline characteristics of the subthalamic nucleus deep brain stimulation cohort.

Characteristic
Age (years)	70.1 (7.8)
Male gender (%)	85
Bilateral DBS (%)	29
Pre‐DBS dyskinesia (%)	29
Pre‐DBS UPDRS‐ III total OFF	40.9 (17.8)
Pre‐DBS UPDRS‐III total ON	26.1 (6.1)

Note: Values are reported as mean (standard deviation) unless otherwise specified.

Abbreviations: DBS, deep brain stimulation; UPDRS‐III, Unified Parkinson's Disease Rating Scale‐Part III.

The discriminative tractography analysis identified that modulation of subthalamo‐pallidal fibers, which travel from the STN to the GPi and the GPe, were associated with SID (Fig. 1b). Figure 1c,f presents the results of the GPi‐based cross‐validation. Using a leave‐one‐out cross‐validation paradigm, dyskinesia fibers exhibited fiber scores that were significantly distinct from non‐SID fibers (P = 0.00). This analysis revealed a highly significant difference between the two fiber populations. A five‐fold cross‐validation further supported the observation (P = 0.01). This consistency across different validation methods underscored the robustness and reliability of the results and confirmed that the identified dyskinesia fibers were distinctly different from non‐SID fibers.

FIG. 1.

Acute stimulation‐induced dyskinesias (SID) is associated with modulation of the subthalamo‐pallidal fibers. (A) The deep brain stimulation (DBS) leads of the initial globus pallidus internus (GPi) cohort are shown relative to the subthalamic nucleus (STN) (purple), GPi (green), and globus pallidus externus (GPe) (blue). Fiber filtering identified subthalamo‐pallidal fibers (red) that travel from the STN to the GPi and GPe (B). The association of these fibers was verified using a (C) five‐fold and (F) leave‐one‐out cross validation paradigm. The fibers associated with acute SID, developed from a GPi‐only cohort, were cross‐validated using an independent STN‐only cohort and an independent GPi‐only cohort. The DBS leads from these cohorts are illustrated for comparison (D and E, respectively). Fiber score validation is shown for the STN cohort (G), and a confusion matrix is provided for the GPi cohort (H). PP, Predicted positive; PN, predicted negative; TP, true positive; TN, true negative. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 1g presents the results of cross‐validation testing of the initial GPi‐based model on the independent STN‐based cohort. This testing involved applying the constructed model to a connectome derived from an entirely different DBS surgical target, the STN. The model successfully identified STN DBS fibers associated with SID based upon fibers derived from GPi DBS (P = 0.02). The strong correlation between the identified fiber profile and SID demonstrated the model's robustness and its potential applicability across different DBS targets.

3.1. Confirmation Testing on Unseen GPi Data

After validating the model, a final test was conducted using completely new data – a GPi cohort independent from the initial cohort. The new dataset included three patients not part of the original analysis: two with bilateral and one with unilateral GPi DBS. This provided 20 additional data points to thoroughly evaluate the dyskinesia model. The results of this additional validation are shown in Figure 1h. This model predicted the presence of SID during monopolar review with 85% accuracy, 66.7% sensitivity, and 92.3% specificity.

4. Discussion

In this study, we aimed at translating the connectivity profile of GPi DBS‐related SID from connectivity and voxelwise based markers into an anatomical location. We developed a tract model based on the Basal Ganglia Pathway Atlas, which used discriminative fiber tract analysis to elucidate the pathways underlying SID (Fig. 2). We identified subthalamo‐pallidal connectivity as an essential pathway that may be associated with the induction of acute SID in GPi DBS. The fibers uncovered connect the STN to both the GPi and the GPe. The dyskinesia‐inducing fibers were then rigorously evaluated using two cross‐validation methods – leave‐one‐out and five‐fold cross‐validation – and this was followed by an independent validation using the STN brain target, which is well known for inducing hyperkinesia. ²⁴ Testing on new patients provided further support for the involvement of this pathway. Figure 3 illustrates the spatial relationship between the subthalamo‐pallidal fibers and the basal ganglia anatomy in a typical GPi DBS case. During monopolar review, stimulation parameters linked to acute SID were transformed into a VTA, which intersects the subthalamo‐pallidal fibers at the antero‐latero‐dorsal aspect of the GPi.

FIG. 2.

Path of the subthalamo‐pallidal tract. The stimulation‐induced dyskinesia pathway may be modulated by targeting the dorsal aspect of the subthalamic nucleus (STN, red) and the antero‐latero‐dorsal aspect of the globus pallidus internus (GPi, orange). This figure shows hypothetical volumes of tissue activated (VTAs, pink) representing brain areas commonly stimulated by deep brain stimulation intersecting with the subthalamo‐pallidal tract. [Color figure can be viewed at wileyonlinelibrary.com]

FIG. 3.

Relationship between subthalamo‐pallidal fibers and an example globus pallidus internus (GPi) deep brain stimulation (DBS) case. The volume of tissue activated (VTA) (orange) represents the stimulation parameters linked to acute stimulation‐induced dyskinesia (SID) during monopolar review of a GPi DBS lead. The subthalamo‐pallidal fibers (red) traverse between the subthalamic nucleus (purple), GPi (green), and globus pallidus externus (blue), intersecting the VTA associated with SID. [Color figure can be viewed at wileyonlinelibrary.com]

Despite these findings, the underlying mechanisms of SID remain incompletely understood. A previous study identified that antero‐latero‐dorsal stimulation of the GPi was associated with SID. ⁷ This specific localization aligns with the trajectory of subthalamo‐pallidal fibers identified in the current study. Despite our small test cohort, the SID model employed could be valuable for connectomic surgical targeting and for planning for both STN and GPi DBS surgeries. Understanding this circuitry preoperatively may in the future help to predict who will be likely to experience a more robust reduction in anti‐parkinsonian medication following DBS surgery.

However, a small proportion of patients receiving DBS struggle to achieve optimal motor benefits due to postoperative dyskinesias. These can be as challenging to manage as cases manifesting internal capsule activation. This issue is particularly prominent in patients with severe or brittle dyskinesias, leading to a see–saw dilemma that limits the overall effectiveness of DBS therapy. ²⁵ , ²⁶ Using DBS leads with 1.5 mm axial spacing between contacts may amplify this phenomenon in the GPi. GPi SID often occurs with antero‐latero‐dorsal stimulation, and the larger contact span increases the chance of reaching the GPi–GPe border, potentially causing SID. In contrast, DBS leads with 0.5 mm contact spacing might have a narrower stimulation zone, avoiding SID. However, emerging evidence suggests that modulating the GPi–GPe border could be a ‘sweet spot’ for DBS in PD, and smaller electrode arrays might struggle to reach this region. ²⁷ Ultimately, these patients often require frequent programming visits, advanced programming configurations, and significant medication adjustments. In extreme cases, additional DBS surgery such as lead revision or rescue may be required. ⁴ , ²⁸ Some studies suggest that SID shortly after DBS implantation can predict the best site for chronic stimulation; however, whether the SID is transient or becomes chronic and problematic remains incompletely understood. ³ , ⁵ , ²⁵ , ²⁹ Employing a connectomic map could assist clinicians in more efficiently navigating these complex cases.

To further understand this connectomic map, Shen et al. identified a population of cells within the GPe in mice that demonstrated vigorous activity during levodopa‐induced dyskinesias. ³⁰ Interestingly, levodopa‐induced dyskinesias were suppressed when these GPe cells were inhibited by using optogenetics. In contrast, optogenetic activation of these cells led to the induction of dyskinesia, even in the absence of levodopa. The study further revealed that GPe and striatum coactivation resulted in more severe dyskinesias than GPe activation alone. These results are consistent with our data and suggest the possibility of a homologous GPe connectivity pathway for driving human dyskinesias.

It is well established that inhibition of the STN through lesions or high‐frequency electrical stimulation can reduce levodopa‐induced dyskinesias in humans. ³¹ , ³² However, Shen et al. observed that optogenetic activation of the STN did not worsen levodopa‐induced dyskinesias, nor did it generate dyskinesia in the absence of levodopa. ³⁰ These findings support the notion that while SID has been observed in humans with STN DBS, it is more likely due to the modulation of the subthalamo‐pallidal fibers rather than the intrinsic activity of the STN target itself. Future prospective connectomic studies could provide valuable insights for guiding symptom‐specific precision neuromodulation. SID can be a challenging obstacle, and in many cases it delays, modifies, and/or hinders DBS programming. By understanding the precise connectivity and mechanisms involved in SID, clinicians could tailor neuromodulation therapies to more effectively alleviate these hyperkinesias.

Our study had several limitations. First, the study was retrospective, and the interpretation of these results should be limited to the observation of acute SID in the dopaminergic medication OFF state during the monopolar review. It is unclear if the SID phenomenon will translate to chronic SID or to levodopa‐induced dyskinesias. Future studies are needed to characterize the relationship between acute SID and the patient experience in the home setting. Second, the imaging analyses were conducted using normalized Montreal Neurological Institute (MNI) space and a group‐averaged normative connectome. These conditions may not wholly represent the patient‐specific disease state, although it could be argued that several studies have demonstrated normative connectome‐based results. Third, it might seem that SID is more common in GPi DBS than in STN DBS based on the cohort sizes in this study. However, we caution that this observation is influenced by a selection bias in our cohort as there were more GPi DBS implantations than STN DBS for PD at our institution during the retrospective review period. Lastly, this study's discriminative fiber tract analysis employed a basal ganglia‐specific connectome and was not designed to investigate connectivity beyond local anatomy. As such, there were limitations to information stored in the connectome. For instance, it does not provide the directionality of connectivity (ie, whether it is from STN to GPe or vice versa). Additionally, the connectome does not define the extent to which the subthalamo‐pallidal fibers intersect the GPi on their way to the GPe. Although we can make inferences based on the adjacent neuroanatomy, future studies will be required to examine a whole brain network.

In conclusion, we identified that SID was associated with modulation of the subthalamo‐pallidal pathway. The modulation of these fibers by DBS revealed a strong correlation with SID, and multiple cross‐validation methods provided confirmation. The findings suggest that these fibers in both STN and GPi DBS are part of a more extensive and yet‐to‐be‐fully‐characterized dyskinesia network.

Author Roles

(1) Research Project: A. Design, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript Preparation: A. Writing of the First Draft, B. Review and Critique.

J.K.W.: 1A, 1B, 1C, 2A, 2B, 3A, 3B.

A.H.: 1A, 3B.

E.H.M.: 1A, 3B.

M.R.B.: 3B.

M.S.O.: 3B.

Financial Disclosures

E.H.M. and M.R.B. have no disclosures. J.K.W. was supported by National Institutes of Health (NIH) KL2TR001429. A.H. was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, 424778381 – TRR 295), Deutsches Zentrum für Luft‐ und Raumfahrt (DynaSti grant within the EU Joint Programme Neurodegenerative Disease Research, JPND), the NIH (R01MH130666, 1R01NS127892‐01, 2R01 MH113929 & UM1NS132358), and the New Venture Fund (FFOR Seed Grant). A.H. was supported by the Schilling Foundation, the German Research Foundation (Deutsche Forschungsgemeinschaft, 424778381 – TRR 295), Deutsches Zentrum für Luft‐ und Raumfahrt (DynaSti grant within the EU Joint Programme Neurodegenerative Disease Research, JPND), the National Institutes of Health (R01MH130666, 1R01NS127892‐01, 2R01 MH113929 & UM1NS132358) as well as the New Venture Fund (FFOR Seed Grant). A.H. reports lecture fees for Boston Scientific, is a consultant for Modulight.bio, was a consultant for FxNeuromodulation and Abbott in recent years and serves as a co‐inventor on a patent granted to Charité University Medicine Berlin that covers multisymptom DBS fiberfiltering and an automated DBS parameter suggestion algorithm unrelated to this work (patent #LU103178). M.S.O. was supported by NIH R01NR014852, R01NS096008, UH3NS119844, and U01NS119562, and is Principal Investigator (PI) of the NIH R25NS108939 Training Grant. M.S.O. serves as Medical Advisor for the Parkinson's Foundation, and has received research grants from NIH, Parkinson's Foundation, The Michael J. Fox Foundation, the Parkinson Alliance, Smallwood Foundation, the Bachmann‐Strauss Foundation, the Tourette Syndrome Association, and the UF Foundation. M.S.O. has received royalties for publications with Demos, Manson, Amazon, Smashwords, Books4Patients, Perseus, Robert Rose, Oxford, and Cambridge (movement disorders books). M.S.O. is an Associate Editor for New England Journal of Medicine Journal, Watch Neurology, and JAMA Neurology. M.S.O. has participated in CME and educational activities (past 12–24 months) on movement disorders sponsored by WebMD/Medscape, RMEI Medical Education, American Academy of Neurology, Movement Disorders Society, Mediflix, and Vanderbilt University. The institution and not M.S.O. receives grants from industry. M.S.O. has participated as a site PI and/or co‐investigator for several NIH, foundation, and industry‐sponsored trials over the years but has not received honoraria. His research projects at the University of Florida receive device and drug donations.

Data Availability Statement

The data supporting the findings of this study are available from the corresponding author upon reasonable request. The image processing tools can be found online at https://github.com/netstim/leaddbs. The Basal Ganglia Atlas can be found online at https://osf.io/mhd4z/.