Lateral cerebellothalamic tract activation underlies DBS therapy for Essential Tremor

Abstract

Background:

While deep brain stimulation (DBS) therapy can be effective at suppressing tremor in individuals with medication-refractory Essential Tremor, patient outcome variability remains a significant challenge across centers. Proximity of active electrodes to the cerebellothalamic tract (CTT) is likely important in suppressing tremor, but how tremor control and side effects relate to targeting parcellations within the CTT and other pathways in and around the ventral intermediate (VIM) nucleus of thalamus remain unclear.

Methods:

Using ultra-high field (7T) MRI, we developed high-dimensional, subject-specific pathway activation models for 23 directional DBS leads. Modeled pathway activations were compared with post-hoc analysis of clinician-optimized DBS settings, paresthesia thresholds, and dysarthria thresholds. Mixed-effect models were utilized to determine how the six parcellated regions of the CTT and how six other pathways in and around the VIM contributed to tremor suppression and induction of side effects.

Results:

The lateral portion of the CTT had the highest activation at clinical settings (p < 0.05) and a significant effect on tremor suppression (p < 0.001). Activation of the medial lemniscus and posterior-medial CTT was significantly associated with severity of paresthesias (p < 0.001). Activation of the anterior-medial CTT had a significant association with dysarthria (p < 0.05).

Conclusions:

This study provides a detailed understanding of the fiber pathways responsible for therapy and side effects of DBS for Essential Tremor, and suggests a model-based programming approach will enable more selective activation of lateral fibers within the CTT.

1. Introduction

Deep brain stimulation (DBS) therapy can be a highly effective neurosurgical intervention for medication-refractory Essential Tremor [1-4]. However, clinical effectiveness continues to vary within [5] and across clinical DBS centers [6], stemming from deviations in the precise surgical targeting of the electrode array to the cerebellar-receiving area of motor thalamus and/or the posterior subthalamic area [7,8] and challenges with identifying stimulation settings that fully reduce tremor without inducing side effects [9,10]. Directional programming of DBS leads that contain split-band electrodes around the lead body enable directing [11] and orienting [12] the activation of axonal pathways, which can extend the therapeutic window [13-15], but the utility of these approaches is limited by an incomplete understanding of the pathways involved in therapy and side effects of DBS for Essential Tremor.

The cerebellar-receiving area of motor thalamus, also known as the ventral intermediate nucleus (VIM) [1,4,7,16-18], as well as the posterior subthalamic area, which includes the zona incerta (ZI) and prelemniscal radiations from the cerebellum, are thought to be key pathways in the propagation of tremorous activity [19-21]. Recent connectomic and tractography studies have focused on the cerebellothalamic tract (CTT) as an effective target for therapy [8,22-26]. However, other studies suggest that CTT may also be responsible for stimulation-induced side effects [27,28].

Biophysical modeling of DBS has become an important tool to identify relationships between the volume of tissue activated [29,30] or the predicted pathways activated [31-33] and the resulting clinical efficacy and side effects of each stimulus setting. Recent studies have shown that pathway activation models, which directly simulate multi-compartment axonal membrane dynamics, are significantly more accurate compared to activating function and driving-function based methods [34]. Further, subject-specific models [31], which are individually constructed from ultra-high field (7T) magnetic resonance (MR) imaging data [35,36], enable one to more accurately account for individual variability in brain pathway morphologies. In this study, we developed subject-specific pathway activation models of DBS for Essential Tremor to identify the specific neural pathways associated with the clinical effect of reducing tremor and inducing side effects, including paresthesia and dysarthria.

2. Materials and methods

2.1. Subject pre- and post-implant imaging and DBS programming

Pre-operative high-field, 7-Tesla (7T) MR imaging was collected from thirteen patients clinically diagnosed with medication-refractory Essential Tremor by a movement disorders specialist. Imaging was performed at the University of Minnesota Center for Magnetic Resonance Research on a 7T research scanner (Magnetom 7T Siemens, Chicago, IL, USA) and followed our published protocols [37,38]. Briefly, imaging included 0.6 mm isotropic T1-weighted, 0.4 × 0.4 × 1.0 mm T2-weighted, and 1.25 or 1.5 mm isotropic diffusion-weighted (b = 1500, 54 directions) images (Fig. 1A). Abbott Infinity DBS systems (Abbott Laboratories, Chicago, IL, USA) with segmented, split-band DBS leads (model 6172 or 6173) were then implanted either unilaterally (three subjects) or bilaterally (ten subjects). Subjects returned approximately one month post-implant for initial programming of their DBS systems, where monopolar review thresholds for therapy and side effects were identified and initial optimized settings determined by a movement disorders specialist. At that time, a CT scan (0.4 × 0.4 × 0.6 mm) was collected and fused with the pre-operative MRI to determine the final location and orientation of the DBS lead within 3D Slicer v4.8.0 [39]. Subject demographics and final programmed DBS settings are shown in Table 1. For subjects with bilateral DBS leads, the clinician-optimized settings were determined for each lead individually such that only one DBS lead was active during the clinical assessments. The study was approved by the University of Minnesota Institutional Review Board and all subjects gave written, informed consent prior to participation and in accordance with the Declaration of Helsinki.

Fig. 1. Subject-specific pathway activation models of DBS for Essential Tremor.

(A) High resolution brain imaging was performed on each subject to define anatomy, tissue conductivity, and lead localization. (B) Predicting pathway activation involved coupling finite element models, pathway morphologies, and biophysical representations of individual axons.

Table 1.

Subject demographics and clinician-optimized DBS settings.

Subject	Age	Gender	Implant to programming time		Initial optimized setting
			Left lead	Right lead	Left lead	Right lead
1	60	M	36 days	29 days	2A (−), case (+), 1.2 mA, 180 Hz, 60 μs	2ABC (−), case (+), 1.2 mA; 180 Hz, 60 μs
2	73	M	38 days	24 days	2ABC (−), case (+), 1.75 mA, 130 Hz, 60 μs	2ABC (−), case (+), 1.5 mA, 130 Hz, 60 μs
3	61	F	36 days	40 days	2B (−), case (+), 3.25 mA, 130 Hz, 90 μs	2B (−), case (+), 3.25 mA, 130 Hz, 90 μs
4	65	F	31 days	70 days	1 (−), case (+), 4.2 mA, 190 Hz, 60 μs	3ABC (−), case (+), 2.0 mA, 130 Hz, 30 μs
5	57	M	29 days	31 days	3A (−), case (+), 2.5 mA, 130 Hz, 60 μs	3ABC (−), case (+), 2.5 mA, 130 Hz, 60 μs
6	67	M	32 days	–	2C (−), case (+), 2.2 mA, 180 Hz, 60 μs	–
7	52	F	52 days	–	2ABC (−), case (+), 1.0 mA, 130 Hz, 30 μs	–
8	75	F	31 days	28 days	3ABC (−), case (+), 1.6 mA, 130 Hz, 60 μs	2ABC (−), case (+), 1.0 mA, 130 Hz, 60 μs
9	57	M	52 days	23 days	3ABC (−), case (+), 2.75 mA, 130 Hz, 30 μs	3C (−), case (+), 2.0 mA, 130 Hz, 30 μs
10	70	M	28 days	28 days	2ABC (−), case (+), 1.5 mA, 130 Hz, 60 μs	3ABC (−), case (+), 2.5 mA, 130 Hz, 60 μs
11	66	F	–	31 days	–	2ABC (−), case (+), 0.5 mA, 130 Hz, 60 μs
12	65	M	30 days	29 days	3ABC (−), case (+), 1.9 mA, 130 Hz, 60 μs	3ABC (−), case (+), 0.5 mA, 130 Hz, 60 μs
13	74	M	80 days	38 days	2ABC (−), case (+), 3.5 mA, 180 Hz, 30 μs	2AC (−), case (+), 2.0 mA, 130 Hz, 30 μs
Totals:	65a	5F/8 M	37 daysa		10 bilateral/3 unilateral leads
	65b		31 daysb		Amplitude: 1.9 mAa, 2.0 mAb
					Frequency: 141 Hza, 130 Hzb
					Pulse Width: 53 μsa, 60 μsb

^*Note that DBS lead numbers refer to the electrode row and A, B, C characters refer to the directional split-band electrode within a row. For, example, 2ABC refers to a grouped configuration in which stimulation was delivered through all three electrodes within the second row of electrodes.

^aAverage.

^bMedian.

2.2. Biophysical tissue voltage models

Using the previously collected imaging data, subject-specific finite element models (FEMs), which calculated the distribution of anisotropic and inhomogeneous tissue voltages for each stimulation setting, were constructed in COMSOL Multiphysics v5.4a (COMSOL Inc., Burlington, MA, USA) (Fig. 1B). The surface of each subject's brain was segmented from the T1 image using the FMRIB's Software Library (FSL) [40] Brain Extraction Tool (FMRIB, Oxford, UK). Patient-specific brain volumes were further segmented into grey matter, white matter, and cerebral spinal fluid using FMRIB's Automated Segmentation Tool (FAST) on the T1 image. Voxels classified as grey matter or white matter were assigned subject-specific anisotropic conductivities based on diffusion tensors [41,42] and using conductivity and permittivity values calculated with the Cole-Cole dispersion functions at the median frequency (3049 Hz) of the stimulus waveform applied (grey matter: σ = 0.106 S/m, ϵ_r = 65,898; white matter: σ = 0.065 S/m, ϵ_r = 29,790) [43-46]. Voxels classified as cerebrospinal fluid were assigned isotropic values (σ = 2.00 S/m and ϵ_r = 109) and bulk tissue outside of the brain was considered lumped tissue (σ = 0.065 S/m, ϵ_r = 29,790) [47]. Each FEM included a digital reconstruction of the Abbott 6172 segmented DBS lead (n = 21, diameter: 1.27 mm, contact height: 1.5 mm, contact spacing: 0.5 mm) or Abbott 6173 segmented DBS lead (n = 2, diameter: 1.27 mm, contact height: 1.5 mm, contact spacing: 1.5 mm). DBS leads had eight electrode contacts stacked in a 1-3-3-1 configuration with the two middle rows containing a split-band of three electrodes. A 0.25 mm encapsulation layer was added around the lead to represent the tissue response following implantation and was modeled with isotropic properties (σ = 0.065 S/m, ϵ_r = 29,790) [48]. The FEM was meshed using Delaunay triangulations and variable tetrahedral elements with extra fine resolution within and around the lead/encapsulation and within a 20 mm sphere around the lead tip. Increasing the sphere radius from 20 mm to 30 mm resulted in <0.5% change in maximum tissue voltage across all simulated axons. Final meshes contained 167,000–198,000 tetrahedra. In each model, the base of the neck was assigned a Dirichlet boundary condition of zero volts to represent the return electrode, and the insulation of the lead was assigned a Neumann boundary condition of zero flux. Within the FEM, stimulation was modeled as a 1 mA/m² normal current density applied to the active electrode's surface. The scaling factor between the subject-specific normal current density and the simulated 1 mA/m² was used to scale the resulting tissue voltages in the pathway activation models.

2.3. Pathway activation models

2.3.1. Nuclei segmentation

MR images were used to segment nuclei and pathways thought to be involved in therapy and side effects of DBS for Essential Tremor (Fig. 1B). Segmentation of the thalamic nuclei at 7T [49] was completed by resampling the T1 MRI image to AC-PC space and warping the Mai Atlas of the Human Brain to the T1-MRI images [31,50,51] to identify the internal and external aspects of the VIM (VIMi and VIMe, respectively), ventral-caudal nucleus (VC), ventral oral anterior and posterior nucleus (VOA/P), reticular nucleus (Rt), and ZI. Warping fiducials were placed external and internal to the thalamus by leveraging high-field 7T imaging with improved contrast, which enabled us to discern borders of nuclei within thalamus including the ventral lateral nuclei to the medial dorsal nucleus and the VC nucleus to the VIM [49]. Motor, premotor, and somatosensory cortices were segmented with FreeSurfer [52], whereas manual segmentation from T1 and T2 images was used for the red nucleus, subthalamic nucleus, crus cerebri, internal globus pallidus, and decussation of the superior cerebellar peduncle. All nuclei and cortex segmentations were visually confirmed by multiple investigators to mitigate error.

2.3.2. Pathway tractography and segmentation

Segmentation of tracks was done using output from probabilistic tractography applied to each subject's diffusion-weighted images. Diffusion preprocessing steps included motion, susceptibility, and eddy current distortions correction using FSL's [40] eddy and topup algorithms: FDT [53] was used for eddy-current correction, BEDPOSTX [54] was used to estimate the diffusion parameters, and DTIFIT [55] was used to fit the diffusion model to each voxel. The diffusion data were first acquired in the anterior-posterior phase encoding direction and then in the posterior-anterior phase encoding direction. Probabilistic tractography was completed using the diffusion images and segmented nuclei to extract subject-specific trajectories for pathways within and around the VIM. For cortico-thalamo-cortical pathways, seed-points and way-points were placed in the premotor cortex (pMC) and the VOA/P, the motor cortex (M1) and the internal and external aspects of the VIM (VIMi and VIMe, respectively), and the somatosensory cortex (S1) and the VC. The CTT was subdivided into six pathways: seed-points were placed in the deep cerebellar nuclei contralateral to the DBS lead, waypoints were placed in the decussation of the superior cerebellar peduncle and the red nucleus/lateral edge of the red nucleus ipsilateral to the DBS lead, and endpoints were placed in the ipsilateral reticular nucleus, VIMi, and VIMe. Each of those tracts were separated further into anterior and posterior subdivisions at the level of the ventral VIM resulting in six pathways. The corticospinal tract within the internal capsule (IC) was obtained with seed-points in the ipsilateral crus cerebri and waypoints in the ipsilateral motor cortex. For the medial lemniscus (ML), the region lateral to the superior cerebellar peduncle decussation was used as a seed-point and a waypoint was placed in the VC. The thalamic/ansa fascicularis (TF), the lenticular fasciculus (LF), and the ZI were segmented manually from the MR images and included axonal pathways that either projected along the tracts (TF and LF) or in the case of ZI were oriented along trajectories approximately parallel to the CTT [56,57]. Since tractography is subject- and scan-specific, we did not use the same threshold to segment tracts across patients. The subject-specific framework involved first segmenting the obvious high-valued hyperintense regions in the diffusion data and then in cases with disconnected segments, we also included lower-valued hyperintense regions until the segmented regions formed a single connected pathway. Pathway thickness was largely consistent within and across subjects. There was overlap in some pathways, especially in the medial and lateral portions of the cerebellothalamic tract before entering the thalamus, which is consistent with known branching patterns of the cerebellothalamic tract entering the thalamus [58-60]. All resulting pathway segmentations were visually confirmed in detail by multiple investigators and further validated against a human brain atlas [51] and histology and neurotracing studies [58-60] to mitigate error.

2.3.3. Axon population

Each pathway was populated with a distribution of multi-compartment myelinated axons (2 μm diameter) [61]. Axons that intersected the lead were displaced to account for tissue deformation using the following formula: r + R*exp (−r), where r is the distance from the closest compartment to the center of the DBS lead and R is the sum of the radius of the DBS lead and thickness of the encapsulation layer [62].

2.3.4. Predictions of axon activation

The FEM voltage distribution was interpolated at each membrane compartment and scaled according to in vivo recordings of electrical artifacts adjacent to a DBS lead in non-human primates, which served to account for the capacitive effects of stimulating through an electrode-tissue interface. If stimulation was applied through multiple electrodes, linear superposition was used to calculate the resulting FEM output. The time-varying extracellular potential was applied to each compartment along each axon, which was used to calculate the transmembrane currents (NEURON v7.3) [11,63]. Action potential counters were placed at the distal end of each axon to determine the number of axons in each pathway whose action potential counters were reliably ‘activated’ by stimulation (Fig. 1B). These simulations were performed at multiple stimulation amplitudes at each monopolar review setting (single electrode or grouped electrode configuration in reference to the metallic can of the IPG). All electrode configurations tested previously in the clinic during the entire monopolar review process (613 settings total) were included in these NEURON simulations. A subset of that data included three important settings: the optimized therapeutic setting, paresthesia threshold, and dysarthria threshold for each patient (230 settings).

2.4. Statistical analysis

2.4.1. Significance of pathway activation

Percent activation for each pathway was calculated as the number of axons activated for a given stimulation setting divided by the total number of modeled axons in each pathway. Freidman's test for variance was run to determine if statistical differences in pathway activation existed at therapeutic and side effect thresholds (MATLAB version R2019b; The MathWorks, Inc., USA). Post hoc analysis was done to test specific comparisons between pathways. A two-sided sign test was performed to compare pathway activation to zero.

2.4.2. Associations between pathway activation and clinical outcomes

For further statistical analysis, significant pathway activation was deemed as greater than 20%, a threshold based on previous studies in which 10–20% activation was necessary to exhibit behavioral changes [64,65]. Pathways – with this level of significant activation at clinician-optimized settings, paresthesia thresholds, or dysarthria thresholds – were analyzed using the larger dataset which included 613 settings (multiple stimulation amplitudes per electrode configuration) taken from the entire monopolar review process for all subjects, rather than just at the aforementioned thresholds. Using this larger data set, ordinal logistic regression models were run in SAS v9.4 (SAS Institute Inc., Cary, NC, USA) with random subject effect in which the dependent variable was categorized as a 0, 1, or 2, based on tremor reduction or side effect severity as interpreted from the clinical visit records. No tremor reduction with DBS was categorized as a 0, partial tremor reduction as a 1, and substantial to full tremor reduction as a 2. No manifestation of a stimulus-induced side effect was categorized as a 0, transient paresthesia or slight dysarthria as a 1, and persistent paresthesia or moderate to severe dysarthria as a 2. For ordinal logistic regression, the outcome was an odds ratio estimate for each individual term.

2.5. Data availability

Pathway activation modeling scripts and data will be uploaded and available through our GitHub repository (https://github.umn.edu/NRTL/PathwayModels_ET-DBS).

3. Results

Neural pathway activation was investigated for 23 directional DBS lead implants in 13 subjects with Essential Tremor. Implant locations and orientations relative to thalamic parcellations and fiber pathways coursing near and through those nuclei are shown for cases with electrodes near the VIM thalamus (Subject 2-R) and the posterior subthalamic area (PSA) (Subject 4-R) (Fig. 2). In this cohort, despite variation in lead location, clinician-optimized electrodes were located near the CTT with contacts positioned in the ventral VIM thalamus, ventral pole of VIM thalamus, or the PSA (see Supplementary Fig. 1 for all DBS lead implant locations).

Fig. 2. Pathways modeled for each DBS lead implant.

Examples are shown for lead implants with electrodes primarily in (A) VIM and (B) PSA, with visualization in the sagittal (left) and coronal (right) perspectives. A subsample of axonal fibers are shown for the cerebellothalamic tract with axons terminating in VIMi (blue) and VIMe (yellow); medial lemniscus (green); and corticothalamocortical pathways: S1-VC (green), M1-VIM (yellow, blue), pMC-VOA/P (violet). The arrow represents the row with the electrode(s) used in the clinician-optimized DBS setting. Pathways not pictured include axons terminating in Rt, and the thalamic and lenticular fascicularis pathways.

3.1. Pathway activation at clinician-optimized settings

Fourteen pathways (six subdivisions of the CTT and eight other tracts) were tested for pathway activation at the clinician-optimized stimulation settings (Fig. 3). While most modeled pathways were not robustly activated at these settings, three pathways did show stronger activation across subjects and leads (CTT, ZI, and ML). Additionally, these three pathways exhibited variance in pathway activation amongst the 23 DBS leads studied, which enabled testing how the degree of pathway activation related to clinical outcomes for the CTT, ZI, and ML.

Fig. 3. Subject-specific pathway activation for clinician-optimized DBS settings.

Activation percentages are shown (A) together for all 23 leads, and (B) for each subject.

Friedman's Test for variance indicated that there were significant differences in activation across pathways (p < 0.001). The highest mean pathway activation was found in the CTT (32.45 ± 14.01%), which was significantly larger than activation in ML (19.92 ± 18.49%; Wilcoxon signed rank test, p < 0.05) and ZI (16.55 ± 17.62% Wilcoxon signed rank test, p < 0.005) (Fig. 4A). An ordinal logistic mixed effects regression model including these three pathways showed that CTT was the only pathway with a significant effect on tremor suppression (p < 0.001) (Fig. 4B). ZI and ML were included for comparison in this analysis given their higher pathway activation values at settings eliciting side effects. Within the CTT (Fig. 4C), Friedman's test for variance indicated significant differences in sub-pathway activation (p < 0.001). Fibers innervating posterior VIMe showed the highest mean activation (59.08 ± 30.89%), followed closely by fibers projecting into anterior VIMe (41.97 ± 30.29%) and posterior Reticular nucleus (41.11 ± 34.56%) (Fig. 4B). All three aforementioned pathways were found to have a similar positive association with tremor suppression (p < 0.001) (Fig. 4C).

Fig. 4. Pathway activation associations with DBS therapy.

(A) Activation across all pathways (*p < 0.05, **p < 0.005), with the white/black dot representing the median and the horizontal line representing the mean activation for each pathway. A black median dot represents significant pathway activation compared to zero (p < 0.001). Inset shows the mixed effects model results for the ZI, CTT, and ML pathways (*p < 0.0001). Odds ratio estimates with 95% confidence intervals not overlapping the dotted grey line were considered to have a significant effect on therapy outcome. (B) The CTT was subdivided into 6 pathways with fibers entering the posterior (P) and anterior (A) Rt, VIMe and VIMi regions of thalamus. Significantly different pathway activations are demarcated by *p < 0.001. (C) Mixed effects model results are shown for subsections of CTT that had greater than 20% activation at clinician-optimized stimulation settings (*p < 0.001).

3.2. Pathway activation at side effect thresholds

Activation of the fourteen pathways were also assessed at the stimulation thresholds for inducing persistent paresthesias and dysarthria. For settings eliciting persistent paresthesias, Friedman's test revealed significant differences in pathway activation (p < 0.001), with the ZI, CTT, and ML pathways exhibiting greater than 20% pathway activation (Fig. 5A). Persistent paresthesias were found to be primarily associated with activation of ML (p < 0.001) and CTT (p < 0.001), whereas ZI fiber activation had a slight negative association with paresthesia severity (p < 0.001) (Fig. 5B). CTT subdivisions revealed significant differences in activation, with the posterior CTT fibers projecting to the VIMi showing the largest positive association with paresthesia severity (p < 0.001) (Fig. 5C).

Fig. 5. Pathway activation at persistent paresthesia thresholds.

(A) Activation of pathways at stimulation thresholds for persistent paresthesia was high for ML (*p < 0.001) along with CTT and ZI. A black median dot represents significant pathway activation compared to zero (p < 0.001). (B) A mixed effects model for paresthesia outcome showed that ML and CTT were similarly associated with paresthesia presence and severity (*p < 0.001). (C) Mixed effects model results for paresthesia outcome including the 6 subsections of CTT fibers showed posterior VIMi as the CTT fibers most associated with paresthesias (*p < 0.05, **p < 0.001).

Pathway activation at the dysarthria thresholds again showed significant differences (Friedman's test, p < 0.001). The CTT exhibited the highest activation at stimulation settings reported to induce dysarthria (p < 0.001) as shown in Fig. 6A. Additionally, CTT proved to have a significant positive association with dysarthria presence in the mixed effects model (p < 0.001) (Fig. 6B). When the CTT was subdivided into 6 regions for further analysis, despite all regions having greater than 20% activation and a Friedman's test revealing significant differences in pathway activation (p < 0.001), only CTT fibers innervating the anterior VIMi had a significant association with dysarthria (p < 0.05) (Fig. 6C). Comparisons of pathway activations amongst clinician-optimized, parethesia-inducing, and dysarthria-inducing DBS settings are shown in Fig. 7 for a single subject and for all subjects in Supplementary Fig. 2.

Fig. 6. Pathway activation at thresholds for stimulation-induced dysarthria.

(A) Pathway activation at thresholds for stimulation-induced dysarthria were elevated for both ZI and pathways within the CTT. (B) Mixed effects model results for dysarthria showed CTT as the only pathway significantly associated with dysarthria. (C) A mixed effects model for dysarthria including directional subsections of CTT fibers showed the anterior VIMi fibers of the CTT as the only subsection significantly associated with dysarthria.

Fig. 7. Pathway activation across therapeutic and side effect-inducing DBS settings.

(A) Representative example (Subject 12-L) of modeled pathways, including M1 to VIMe (yellow), M1 to VIMi (blue), pMC to VOA/P (lavender), S1 to VC (green), ML (green), CTT to VIMe (yellow) and CTT to VIMi (blue). (B) Pathway activation using ring mode (contact 3ABC) as the cathode (0.5 mA) resulted in complete arrest of tremor without inducing side effects. Activated axons appear in their corresponding pathway color. Non-activated axons are transparent grey. (C) Persistent paresthesias were induced using contact 1 as the cathode (2.5 mA), and (D) dysarthria was generated using contact 4 as the cathode (4 mA). In both cases, full tremor control was also achieved.

4. Discussion

In this study, detailed subject-specific biophysical models of DBS were used to investigate the relationships between pathway activation and clinical effects of DBS on tremor, paresthesia, and dysarthria. Similar to previous studies, our results showed significant activation of the CTT aligning with tremor improvement; however, the models went further in specifically identifying the lateral deep cerebellar nuclei projecting axons to the reticular nucleus and to VIMe as the key elements activated at clinician-optimized stimulation settings.

4.1. Improvement in tremor associated with lateral CTT activation

Therapeutic efficacy of thalamic DBS for Essential Tremor has been attributed to activation of the VIM nucleus as well as the PSA [31,66]. The PSA has been shown to have clinical outcome equivalence to VIM DBS, though the PSA achieves therapy at lower stimulation intensities on average [10,67]. In our patients, the directional DBS leads were implanted such that the segmented contacts were positioned through the VIM and PSA, enabling stimulation settings to target both the CTT and other pathways in the PSA region, including the ZI, TF, and LF. We first investigated the hypothesis that activation of one or more of these pathways is responsible for the therapeutic effects of DBS. Our models showed that, at the clinician-optimized settings, the CTT coursing through the PSA area was activated significantly more than the ZI. There was also little or no activation of the LF and TF. These subject-specific pathway activation modeling results confirm previous studies based on 3T MRI and volume of tissue activated calculations that found positive association with the CTT region and tremor reduction [22-25,30,31,68-70].

Given the improved image contrast and higher signal to noise ratio of the subject-specific imaging at 7T, we were able to further segment the CTT into 6 sub-pathways based on entrance to VIM at the ventral pole. These included the anterior reticular, VIMe, and VIMi, as well as the posterior reticular, VIMe, and VIMi subpathways. Based on electrophysiology studies [71], the medial to lateral demarcations relate to the somatotopic organization of VIM in that projections related to the lower extremities are more lateral (VIMe) and those responsive to facial muscle perturbations are more medial (VIMi). Recent thalamotomy studies have shown that a lesion restricted to the most lateral/external portion of VIM is sufficient for arresting tremor with less adverse effects than the traditional, more medial target [58]. We found that clinician-optimized settings had greatest activation of the VIMe and posterior reticular fibers, suggesting the postero-lateral afferents are the portion of the CTT responsible for the primary therapeutic effect seen with DBS for Essential Tremor.

4.2. Paresthesia associated with medial lemniscus activation

Paresthesias are thought to originate from activation of cells within the VC region of thalamus given their responsiveness to somatosensory stimulation [72,73]. Consistent with this hypothesis, the mixed effects models showed the presence and severity of a persistent paresthesia was associated with activation of the ML pathway. Activation of the ML aligns with a previous study noting association of the afferent inputs to the VC with paresthesias [31]. The additional association of paresthesias with activation of posterior VIMi CTT fibers may stem from the proximity of these fibers to the ML or modulation of kinesthetic cells at the posterior border of the VIM [74] and potentially an over-interpretation of fibers from the medial lemniscus on the posterior VIMi CTT. Additionally, in our study, two-thirds (50/75) of the stimulation settings inducing persistent paresthesias reported paresthesias located in the face, which aligns with the posterior VIMi as having a somatotopy of face-responsive cells [71].

4.3. Dysarthria associated with activating CTT fibers projecting to the anterior VIMi

Dysarthria is another common stimulation-induced side effect of DBS for Essential Tremor, reportedly occurring in up to 75% of cases [75,76]. In our study, dysarthria occurred at the upper limit of the therapeutic window in over half of the clinician-optimized settings. We assessed the pathways activated at dysarthria onset and found a significant association with activation of CTT fibers, and more specifically, anterior VIMi afferents. This medial activation profile aligns with the face-related cellular somatotopy of internal VIM [71] as well as a study by Matsumoto and colleagues showing that stimulation-induced dysarthria occurred with more medial thalamic stimulation [77]. Other studies have also suggested that stimulation overlapping the CTT was a possible cause of dysarthric worsening of intelligibility [27,28]. Our findings are consistent with these past studies and further suggest that the anterior VIMi afferents of the CTT are specifically those responsible for stimulation-induced dysarthria.

4.4. Study limitations

The patient-specific modeling pipeline developed here has the potential to improve directing and steering of the electric field in and around DBS leads. However, these studies have an inherent degree of error varying from registration of MR image volumes, to registration of a post-operative CT scan to the MRI scans, to manual segmentation of brain volumes later used for tractography. These sources of error are mitigated to the best of our ability through a combination of improved spatial resolution of images by using 7T MRI with optimized sequences, manual correction of automated image registration, multi-investigator validation of subject-specific segmentations, and detailed validation of tractography solutions with brain atlases. While the tissue models incorporated inhomogeneous and anisotropic details of each subject's brain anatomy, blood vessels were not explicitly segmented, which may affect the degree of pathway activation [33]. Additionally, the pathways modeled in this study did not include axonal branching [78-80] or synaptic connections amongst pathways within the VIM [81] or ZI, which can affect the overall cellular modulatory effect within both nuclei. The models also assumed a 2 μm axon diameter for all pathways, which while consistent with previous studies [61,81], could affect data interpretation since narrower diameter axons or axons without myelination would have weaker activation patterns. The models also did not include all known pathways [82] within and adjacent to the VIM (e.g. inhibitory feedforward connections from the Rt to the VIM [83] and structural differences between thalamo-cortical and cortico-thalamic pathways [80,84]. Further, the ZI contains a heterogenous mixture of neuronal cell types and neuronal pathways that project into and through it [85]. In this study, we modeled the ZI pathway in an orientation consistent with cortical-ZIc tracings as shown in Refs. [56,57], and thus did not account for other axonal orientations that may have been stimulated differentially [12]. Another consideration is that the therapeutic effects were based on the clinician-assessed Fahn-Tolosa Marin (FTM) and the Essential Tremor Rating Assessment Scale (TETRAS) scores. As such, the final programmed settings and thresholds for therapy were used primarily in the analysis; however, within the larger sample used for mixed effects modeling, tremor and side effects outcomes were assigned a severity value of 0, 1, or 2 based on interpretation of the clinical records, which enabled more nuanced comparisons. An additional consideration is the time and effort necessary to produce these detailed, subject-specific models, which were not optimized for practical implementation in today's standard clinical workflow. The use of these models based on ultra-high brain MRI, however, provide unique opportunities to identify the key pathways for both surgical targeting and post-operative programming on a subject-specific basis [35].

5. Conclusions

This patient-specific pathway activation modeling study utilizing high-field MR imaging suggests that DBS targeting the lateral cerebellothalamic tract is most closely associated with tremor reduction, whereas paresthesias are associated with activation of the medial leminiscus and the posterior-medial border of the CTT and dysarthria is associated with activation of the anterior medial CTT. These pathway-based targets provide a more refined template for future neurosurgical targeting and post-operative programming of DBS leads. While placing DBS leads too posterior is well known to result in low-threshold paresthesias, the modeling results also suggest that positioning the DBS lead near the medial CTT may result in low-threshold dysarthria. Future studies may consider using directional programming (i.e. utilizing the split-band electrodes) on directional DBS leads to sculpt and steer electric fields to more precisely target the lateral portion of the CTT.

Abbreviations

CT

computed tomography

CTT

cerebellothalamic tract

DBS

deep brain stimulation

IC

internal capsule/corticospinal tract

IPG

implanted pulse generator

LF

lenticular fasciculus

M1

motor cortex

ML

medial lemniscus

MR

magnetic resonance

PSA

posterior subthalamic area

Rt

reticular nucleus of the thalamus

S1

somatosensory cortex

TF

thalamic fasciculus

VC

ventral caudal nucleus of thalamus

VIM

ventral intermediate nucleus of thalamus

VIMe

external VIM

VIMi

internal VIM

VOA/P

ventral oral ant/posterior nucleus of thalamus

ZI

zona incerta