Abstract
Essential tremor is the most prevalent movement disorder and is often refractory to medical treatment. Deep brain stimulation offers a therapeutic approach that can efficiently control tremor symptoms. Several deep brain stimulation targets (ventral intermediate nucleus, zona incerta, posterior subthalamic area) have been discussed for tremor treatment. Effective deep brain stimulation therapy for tremor critically involves optimal targeting to modulate the tremor network. This could potentially become more robust and precise by using state-of-the-art brain connectivity measurements. In the current study, we utilized two normative brain connectomes (structural and functional) to show the pattern of effective deep brain stimulation electrode connectivity in 36 essential tremor patients. Our structural and functional connectivity models were significantly predictive of post-operative tremor improvement in out-of-sample data (p < 0.001 for both structural and functional leave-one-out cross-validation). Additionally, we segregated the somatotopic brain network based on head and hand tremor scores. These resulted in segregations that mapped onto the well-known somatotopic maps of both motor cortex and cerebellum. Crucially, this shows that slightly distinct networks need to be modulated to ameliorate head vs. hand tremor and that those networks could be identified based on somatotopic zones in motor cortex and cerebellum.
Finally, we propose a multi-modal connectomic deep brain stimulation sweet spot that may serve as a reference to enhance clinical care, in the future. This spot resided in the posterior subthalamic area, encroaching on the inferior borders of ventral intermediate nucleus and sensory thalamus. Our results underscore the importance of integrating brain connectivity in optimizing deep brain stimulation targeting for essential tremor.
Introduction
Essential tremor (ET) is the most common movement disorder that is encountered in clinical practice (Deuschl, 2000a). A satisfactory pharmacotherapeutic treatment is difficult if impossible to attain in 25-55% of ET cases (Flora et al., 2010). Therefore, deep brain stimulation (DBS) has been accepted as an efficacious alternative to control medication-refractory tremor symptoms.
To date, multiple DBS targets have been proposed to effectively treat ET (Deuschl et al., 2011). Targeting the VIM nucleus was regarded as a historical gold-standard since the beginnings of modern-day DBS (Benabid et al., 1991). Increasingly, the ventrally adjacent white matter has been proposed to lead to superior effects (Hamel et al., 2007; Sandvik et al., 2012; Eisinger et al., 2018). This target has been referred to as the posterior subthalamic area (PSA). Thus, the optimal treatment coordinates are still a matter of debate.
Pathophysiological evidence has accumulated that a cerebello-thalamo-cortical tremor network plays a crucial role in mediating abnormal oscillatory tremor activity and its modulation is related to the therapeutic effects of DBS (Schnitzler et al., 2009, Raethjen and Deuschl, 2012). The cortical and subcortical nodes constituting the proposed network have been described with fMRI and MEG (Sharifi et al., 2014, Schnitzler et al., 2009). In light of such a network-based mechanism, strong connectivity between DBS electrodes and network tremor nodes should lead to effective treatment response. This approach has been followed in individual cases by Coenen and colleagues who proposed DTI-based targeting in tremor patients focusing on the connectivity between the cerebellum and the thalamus (Coenen et al., 2011a, 2011b, 2017). Recently, a different approach has been proposed to use whole brain connectivity patterns to predict clinical outcome after DBS. This was first demonstrated in Parkinson Disease across cohorts, and improvement scores could be predicted across DBS centers and surgeons (Horn et al., 2017b, 2017a). In case of ET, few studies addressed the relationship between DBS connectivity and clinical outcome and so far, none has actually used brain connectivity to predict the DBS effects in out-of-sample data (Pouratian et al., 2011; Gibson et al., 2016; Akram et al., 2018; Middlebrooks et al., 2018).
Here, we aimed at constructing a “therapeutic network” model for DBS in ET. Following the concept of (Horn et al. 2017b), we postulated that similarity to this connectivity fingerprint could linearly predict clinical outcome in ET patients. We traced DBS-electrode connectivity to other brain regions using high resolution normative connectomes (functional and structural) as surrogate neuroimaging models in a data-driven fashion. We validated the resulting optimal connectivity fingerprints by predicting individual tremor improvements in a leave-one-out design. In a further step, we used DBS connectivity to investigate somatotopic treatment effects. Specifically, we analyzed how tremor improvement of hand and head could be associated with segregated DBS connectivity maps. Finally, we condensed findings to define an optimal surgical target for ET, which is made publicly available in form of a probabilistic atlas dataset.
Materials and methods
Patients: demographic and clinical details
Thirty-six patients underwent DBS (72 DBS electrodes) for severe, medically intractable ET (13 female) were retrospectively included in the current study (mean age = 74.3 ± 11.9 years). Diagnosis of ET followed the consensus criteria proposed in 1998 (Deuschl et al. 1998). Patients with bilateral symmetric postural or kinetic tremor of the upper limb with the possibility of additional head tremor, were included as ET cases. Any isolated voice, chin, tongue or leg tremor patients were excluded. Additionally, patients with dystonic, neuropathic, orthostatic, physiological or psychological tremor were excluded. Patients had a mean disease duration of 24.33 ± 14.99 years before DBS surgery. All patients received bilateral DBS implants in Charité–Universitätsmedizin, Berlin for the period between 2001 and 2017 (see Table 1 for clinical and demographic information and supplementary table S1 for individual patient clinical characteristics). All implanted DBS electrodes were Medtronic 3387 (except for three patients in which two were implanted with Boston Scientific Vercice Directed and one with St Jude Medical). Preoperative MRI was used to define VIM/zona incerta DBS targets. Microelectrode recordings and test stimulation were utilized intraoperatively to guide DBS lead placement. Correct lead placement was confirmed by postoperative imaging using LEAD-DBS to localize DBS electrodes in standard MNI space (Fig. 1). Percent improvement in the Fahn-Tolosa-Marin (FTM) tremor score served as an index of clinical outcome (Fahn et al., 1988). FTM scores before (baseline) and at least 3 months after electrode implantation have been obtained from archival video material. All videos have been rated by three clinicians experienced in movement disorders. Each clinician (BA, DKu and DKo) rated separate part of the cohort (so no video was rated twice) and was blinded to the timing of the video (preoperative vs postoperative). Postoperative FTM scores indicate tremor severity during the chronic DBS ON condition. Upper limb subscores contralateral to DBS electrodes were summed up and used in the calculation of the main clinical outcome quantifying therapeutic effect. Upper limb subscore comprised the following items: rest tremor, postural tremor, action tremor, drawing of Archimedes spiral and repeated letter L writing (modified FTM score). For somatotopy related analyses, bilateral upper limb subscores and head scores were used. The head score consisted of the sum of head, face, tongue, speech and voice related subscores. All patients showed a reduction in FTM score of at least ∼ 27% with a mean decrease of 22.4 ± 9.9 points of the average total FTM score (from 33.3 ± 9.6 at baseline to 10.9 ± 5.5 with chronic DBS). The average postoperative time at which patients were assessed for postoperative FTM scoring was 12 ± 9.86 months.
The study was approved by the local ethics committee of the Charité University Medicine - Berlin.
DBS electrode localizations
Preoperative MRI as well as postoperative MRI or CT were obtained in all patients. DBS electrodes were localized using Lead-DBS software (Horn & Kühn 2017; www.lead-dbs.org) following the enhanced methodology described in (Horn & Li et al. 2018 NeuroImage). Briefly, preoperative and postoperative patients’ images were linearly co-registered using Advanced Normalization Tools (ANTs, Avants et al., 2009; http://stnava.github.io/ANTs/) and manually refined when necessary.
Pre- and postoperative images were then normalized into ICBM 2009b NLIN asymmetric space using the symmetric diffeomorphic image registration approach implemented in ANTs (Avants et al., 2009; http://stnava.github.io/ANTs/). Electrodes were then localized and volumes of tissue activated (VTA) modeled using Lead-DBS based on patient-specific stimulation parameters.
Functional and Structural Connectivity Estimation
Using VTAs as seed regions, functional and structural connectivity estimates were computed using pipelines implemented in Lead-DBS. Two normative connectomes were used: First, a structural connectome (Horn et al., 2014; Horn, 2015) which consists of high density normative fibertracts based on 20 subjects. Diffusion data were collected using single-shot spin-echo planar imaging sequence (TR = 10,000 ms, TE = 94 ms, 2 × 2 × 2 mm3, 69 slices). Global fiber-tracking was performed using Gibb’s tracking method (Reisert et al., 2011) (for more methodological details, see Horn and Blankenburg (2016)). Structural connectivity was estimated by extracting tracts passing through VTA seeds and calculating the fiber counts in a voxel-wise manner across the whole brain. Second, a functional connectome which was defined on 1000 healthy subjects resting-state fMRI scans (Yeo et al., 2011; https://dataverse.harvard.edu/dataverse/GSP) and is based on data of the Brain Genomics Superstruct Project. Data were collected with 3T Siemens (Erlangen, Germany) MRI and the resting state BOLD processed with signal regression and application of spatial smoothing kernel of 6mm at full-width at half maximum (Yeo et al., 2011). For the purpose of the current study, connectivity estimates were performed for each of the 72 VTAs (36 bilateral implants) after nonlinearly flipping right sided VTA to the left side using Lead-DBS.
Models of optimal connectivity
Following the concept described in (Horn et al., 2017b), clinical improvements in the contralateral upper limb were correlated with structural and functional connectivity from the VTA (while these were accumulated on the left side of the brain) to each brain voxel across electrodes. This process resulted in R-maps that carry Spearman’s rank-correlation coefficients for each voxel. The maps fulfill two concepts. First, they denote to which areas connectivity is associated with beneficial outcome. Second, their spatial distribution describes an optimal connectivity profile of DBS electrodes for ET (Horn et al., 2017b).
Thus, to make predictions, each VTA-derived structural or functional connectivity pattern was then tested for spatial similarity with this optimal connectivity model. Specifically, similarity between each VTA’s connectivity profile and the “optimal” connectivity profile (as defined by the R-map) was calculated using spatial correlation. The resulting similarity index estimates “how optimal” each connectivity profile was and was used to explain clinical improvement in a linear regression model. To cross-validate the model, we correlated aforementioned predicted and empirical individual upper limb tremor improvements in a leave-one-out design. Furthermore, we calculated discriminative fibertracts following the approach introduced recently by Baldermann and colleagues (Baldermann et al. 2019). Briefly, fibertracts connected to VTAs across cohort were isolated from the normative group connectome. In a mass-univariate analysis, for each fibertract, a two-sample t-test was performed between improvement scores of VTAs connected versus improvement of non-connected VTAs and fibers were labelled according to this t-score. The resulting positive t-score streamlines represent fibertracts that may discriminate between poor and good responders. Again, this analysis was performed across the left-sided accumulated VTAs using contralateral upper limb improvement subscores. This analysis was used to confirm the main analysis using a slightly different statistical concept.
Prospective Case Validation
We pre-operatively scanned one patient with diffusion weighted imaging (see supplementary methods for scan parameters) to investigate the validity of our model in predicting patient improvement using patient-specific tractography. The patient received an unilateral implant on the left (Abbott’s St. Jude Medical Infinity model) for treatment of refractory ET affecting the upper limbs. The VTA was modeled with the same pipeline as the main patients cohort. Patient-specific diffusion weighted imaging (dMRI) data was then used to calculate fiber streamlines seeding from the modeled left-VTA. The resulting connectivity profile was then fed into the structural predictive model created on the main cohort (using the normative connectome). Patient’s empirical right upper limb FTM score was calculated pre- and post-operatively following the same methodological description as in the main cohort.
Side-effects Related Connectivity Profile
Connectivity seeding from electrodes associated with DBS-related side effects were also calculated in a subgroup of patients in which information about side effects were available using the same functional connectome (Yeo et al. 2011). We then compared the resulting connectivity to a sample of control patients DBS-induced side-effects could be excluded. To do so, mass-univariate voxel-wise two-sample t-tests were calculated between connectivity strengths seeding from VTAs associated with gait ataxia or dysarthria and that of control patients. Connectivity difference images were then masked by significant p-values (<0.05, uncorrected) and presented as positive t-scores images.
Deriving Somatotopic Maps
In a further step, we segregated somatotopic maps informed by optimal functional connectivity models based on upper limb (hand) and head tremor improvements. Since head tremor is an axial feature modulated by both left and right VTAs, those were combined in this analysis. Hence, bilateral VTAs were used to estimate functional somatotopic maps (i.e. connectivity was estimated seeding from both VTAs). The resulting connectivity maps were correlated with either summed bilateral hand scores or head scores. The resulting R-maps were overlaid on the cerebellum and primary motor cortex to investigate somatotopy.
Defining an optimal DBS target
As a final step, we applied our optimal predictive structural and functional models to define an “optimal” DBS target. We masked our functional and structural R-maps to include only voxels in the cortical and cerebellar regions. This was done since otherwise the design would have been recursive (with subcortical information already present in the R-maps). The subcortical region with maximal connectivity to those R-maps was determined using Lead-DBS. The resulting connectivity maps were then overlapped to show where exactly they converged. This spot is characterized by optimal functional and structural brain connectivity for maximal therapeutic outcome.
Data availability
Patients dataset are not publicly available due to data privacy restrictions but can be made available from the corresponding author upon reasonable request. All code used in the present manuscript is available within Lead-DBS software (https://github.com/leaddbs/leaddbs).
Results
In total, 72 DBS electrodes were included in the analyses. Connectivity based R-maps highlighted positively predictive voxels in multiple regions (Fig. 2) such as paracentral gyrus (M1 and sensory cortex), visual cortices (V1 and V2), superior temporal gyrus, and superior and inferior cerebellar lobules. Additionally, functional connectivity to part of the premotor cortex and supplementary motor area was associated with beneficial DBS outcome. On the other hand, structural optimal connectivity outlined additional regions such as superior parietal lobule and precuneus. Except those, the beneficial functional and structural connectivity profiles were largely congruent.
Functional connectivity profiles could explain 16.4% of the variance in DBS outcome (R = 0.41, p < 0.001), while structural connectivity profile could explain 25% of the variance in DBS outcome (R = 0.50, p < 10-5). In a leave-one-out cross-validation, both structural (R = 0.40, p < 0.001) and functional connectivity (R = 0.36, p = 0.0017) remained significant predictors of individual clinical improvement. On average, predicted tremor improvements deviated from empirical improvements by 17.98 ± 10.73 % for structural and 18.09 ± 11.22 % for functional connectivity. As a proof of concept, similarity between VTA-seed connectivity in one modality and the R-map model of the other was also significantly predictive of tremor improvement (functional VTA-seed connectivity explained by structural model R = 0.41, p < 0.001; structural connectivity explained by functional model R = 0.33, p = 0.005). This may further illustrate similarities between optimal functional and structural connectivity maps. While our main analysis focused on improvements of hand-tremor scores, we repeated the main analysis for improvements of full tremor scores which led to near identical results (see Fig. S1).
Structural DBS connectivity showed voxel clusters intersecting with a DBS target commonly used in ET treatment (Papavassiliou et al., 2004) and with the cerebello-thalamo-cortical tract (Fig. 3). The cluster extended from the M1 cortex down to the thalamic-subthalamic region. Discriminative fibertract analysis delineated a well-defined tract connecting M1 and cerebellum (Fig. 5), passing through the motor thalamus. Crucially, based on our results, this tract represented the part of the cerebello-thalamo-cortical pathway that was associated with optimal improvement.
Beneficial structural connectivity (based on normative connectome) successfully predicted the magnitude of tremor improvement in a single prospective patient (empirical clinical improvement 61%, predicted clinical improvement 72%). This prediction was performed using patient-specific structural connectivity (Fig. S2).
Next, we aimed at defining functional connectivity maps that could explain therapeutic response in different body parts (hand vs. head tremor, Fig. 4). Of note, only 22 patients were included in the functional connectivity model of head tremor since the symptom was not present in the remaining 11. All patients responded well to head-tremor at baseline, thus a subanalysis comparing good vs. bad responders was not possible. The topology of M1 and cerebellar voxels predictive of hand and head tremor improvement followed the known homuncular organization of M1 and somatotopy of the cerebellum (Buckner et al., 2011). Furthermore, connectivity to these somatotopy-specific sub-regions of the cerebellum and M1 could explain improvement of hand (R = 0.44, p = 0.008), and head tremor (R = 0.59, p = 0.004), respectively.
Additionally, we investigated functional connectivity patterns that could differentiate patients with DBS-related side-effects (namely gait ataxia and dysarthria) from control patients. Our analysis revealed side-effect specific clusters. Interestingly, these cortical and cerebellar clusters overlapped minimally with voxels positively correlated with optimal DBS outcome. Of note, these results are not corrected for multiple comparisons and should be interpreted with caution.
Our final goal was to define a clinically relevant surgical target that maximizes beneficial connectivity within the thalamo-subthalamic area. In order to obtain such target, we seeded back from cortical voxels in our structural and functional R-maps (using their entries as a weighted connectivity seeds in Lead-DBS). Only cortical voxels were included to avoid confusion with already highlighted voxels in the sub-cortex. The resulting functional and structural connectivity patterns converged at the inferoposterior border of the VIM and extended inferiorly and posteriorly to overlap with the dorsal part of the zona incerta (Fig. 6).
Discussion
We demonstrated that optimal tremor reduction with DBS is significantly correlated with a specific pattern of functional and structural connectivity including sensorimotor areas and cerebellum. Importantly, the connectivity fingerprint of brain tissue activated by DBS can predict tremor improvement in out-of-sample data. Our models of optimal “therapeutic connectivity” largely overlap with brain regions that were linked to ET pathophysiology, before. More importantly, we demonstrated that tremor in distinct body parts is optimally ameliorated by modulating a specific network that includes somatotopic regions of both M1 and the cerebellum.
Finally, we defined an “optimal” DBS target that maximizes beneficial functional and structural connectivity.
The tremor network and pattern of beneficial DBS connectivity
The mechanism of tremor generation has been attributed to multiple central oscillators (Schnitzler et al., 2009) that are synchronized in a tremor specific frequency (Marsden et al., 2000; Hellwig et al., 2001) and distributed across nodes of the cerebello-thalamo-cortical pathway. It has been thought that the cerebellum drives tremorogenic oscillations (Deuschl, 2000b). However, several studies unveiled the involvement of cortical (sensorimotor, supplementary motor and premotor cortices) and subcortical (thalamus) nodes in tremor generation (McAuley, 2000; Pinto et al., 2003; Schnitzler et al., 2009; Helmich et al., 2013). Theoretically, interference with any of these cerebello-thalamo-cortical nodes should suppress tremor oscillation. The thalamic (VIM) nucleus which receives most of the cerebellar afferent fibers (Asanuma et al., 1983) has been of much interest in tremor research (Pedrosa et al., 2012; Basha et al., 2014; Fang et al., 2016; Milosevic et al., 2018). The VIM also projects to the aforementioned tremor-related motor areas (McFarland and Haber, 2002; Haber and Calzavara, 2009). This property gives it a central position in the cerebello-thalamo-cortical tremor pathway. Historically, it was considered an excellent target for lesioning surgery (thalamotomy) yielding a satisfactory outcome of tremor control (Deuschl et al., 2011). Later, DBS surgery started to replace thalamotomy in the majority of cases, given its reversible and adjustable stimulation (Tasker, 1998). Nonetheless, clear visualization of the VIM region with conventional MRI is difficult even in contemporary DBS surgery with modern imaging protocols (Yamada et al., 2010), this is why connectivity has already been used to target VIM-DBS surgeries (Anderson et al., 2011; Coenen et al., 2014).
This said, the optimal DBS target has to have tight functional and structural connectivity to the tremorogenic nodes in order to remotely modulate the nuisance tremor oscillations. Our results showed a connectivity pattern which agrees with this concept. Both structural and functional connectivity demonstrated areas in the pre- and postcentral gyri in addition to the superior and inferior cerebellar lobules. This is in line with the results of most studies that showed tremor related alterations of the sensorimotor and cerebellar areas (Colebatch, 1990; Jenkins et al., 1993; Wills et al., 1995; Czarnecki et al., 2011; Fang et al., 2013; Mueller et al., 2017). Additionally, target connectivity to the aforementioned areas was associated with tremor improvement in VIM-DBS and ablative (thalamotomy) surgeries (Klein et al., 2012; Gibson et al., 2016; Akram et al., 2018; Middlebrooks et al., 2018; Tuleasca et al., 2018a).
Other regions that could potentially play a role based on present findings are primary and associative visual cortices. The importance of brain visual areas in tremor pathogenesis has been recently investigated by using a visual task of increasing difficulty to illustrate the impact of visuospatial network in tremor augmentation (Archer et al., 2018). Furthermore, recent series of investigations suggested that structural and functional changes of the visual cortex could be a preoperative predictor of optimum tremor outcome after ablative radiosurgery (Tuleasca et al., 2017, 2018b, 2018c).
Somatotopic organization of beneficial DBS connectivity
Finely tuned DBS targeting with respect to the somatotopy of body regions has been considered in dystonia patients (Vayssiere et al., 2004). We leveraged the nature of anatomical somatotopic distributions in order to explain how DBS related connectivity profile could vary accordingly. Our results demonstrated two distinct connectivity profiles corresponding to hand and head in M1 and cerebellar regions. Crucially, these areas corresponded to formerly determined hand and tongue brain regions in the human M1 homunculus (Penfield and Boldrey, 1937) and cerebellum (Buckner et al., 2011). Furthermore, they predicted DBS tremor reduction in their respective body regions. Our finding supports the utility of hand and head tremor driven connectivity profiles in guiding DBS targeting, which could be an important future step for further refinement of DBS treatment of focal motor symptoms. Head tremor is the second most common body distribution of tremor symptoms encountered in ET patients that is highly disabling beside the predominant upper limbs tremor (Hoskovcová et al., 2013; Bhatia et al., 2018). Correspondingly, controlling head tremor has been an outcome issue in many patients undergoing DBS surgery (Obwegeser et al., 2000; Putzke, 2005). Our results may pave the way for personalized DBS targeting that is dependent on the tremor symptoms each patient may have. It is even conceivable to scan patients in the fMRI while they perform (imaginary) tasks involving hand and head to identify their specific somatotopic organization of M1 and the cerebellum. These regions could then be used in single patients to define the tremor target optimally corresponding to their symptomatology.
Connectivity-derived predictive models
The beneficial connectivity profiles that were estimated in the present work were built using a completely data-driven design. This means that these profile maps can be interpreted as an answer to where in the brain connectivity may explain most of the variance in clinical improvement. The concept of using connectivity patterns to predict functional capacity and clinical symptoms has been a central dogma in contemporary studies (Beaty et al., 2018; Cao et al., 2018). We relied on this concept in order to draw conclusions about the optimal connectivity fingerprint that will ensure the best outcome. Of note, connectivity associated with the emergence of side-effects involved inverse patterns of brain areas compared to beneficial DBS outcome. The cerebellar vermis was shown as a key region in ataxia related analysis, which is in accordance with previous results (Reich et al., 2016). Our models could significantly predict tremor improvement in out-of-sample data as well as in a single prospective patient using patient-specific dMRI data. Future work should focus on validating such connectivity fingerprints in a larger sample of prospective patients. Furthermore, the isolated discriminative tract emphasized the importance of targeting cerebello-thalamo-cortical pathways for determining DBS outcome (Coenen et al., 2014, Sammartino et al., 2016).
A connectomic DBS target for Essential Tremor
The exact DBS target for optimal therapeutic benefit in ET is not yet entirely clear. Four main surgical targets have been suggested for essential tremor treatment. Located within the thalamus, the VIM nucleus has been regarded as the mainstay therapeutic target (Benabid et al., 1991; Pahwa et al., 2006; Zhang et al., 2010; Baizabal-Carvallo et al., 2014), while the other three targets within the subthalamic area (the PSA, which encompasses caudal zona incerta and the radiation prelemniscalis, and subthalamic nucleus) were the focus of other studies (Herzog et al., 2004; Plaha et al., 2008; Fytagoridis and Blomstedt, 2010). VIM DBS has proven to be an effective tremor target since the beginnings of modern-day DBS (Benabid et al., 1991; Deuschl et al., 2011). On the other hand, there is a growing evidence that DBS to the directly adjacent PSA is similarly effective (Plaha et al., 2004, 2011; Fytagoridis et al., 2012; Barbe et al., 2018). Deciding which target is optimal for tremor suppression is a critical step in stereotactic surgery. The results of the present study showed that the discussed targets may in fact be the same – fibers that pass along the red nucleus toward the thalamus and in doing so traverse through the PSA and zona incerta. Structural and functional connectivity maps converged in a region that impinge the inferior-thalamic border and extend to the PSA. Moreover, the proposed DBS spot is located ventrolateral to the thalamus, in an area medial to the internal capsule and directly inferior to the VIM and sensory thalamic nuclei, encroaching on their inferior borders. This area has been described in the literature as the entry of the afferent cerebellar fibers to the thalamus (particularly, the VIM nucleus; Gallay et al., 2008). Our results further imply the importance of the cerebellothalamic tremor pathway and encourage tract-based targeting for ET treatment (Sammartino et al., 2016; Fenoy and Schiess, 2018). Intriguingly, the identified spot is in accordance with a recently described optimal location for focused ultrasound thalamotomy in essential tremor treatment (Boutet et al., 2018) and with a previously published sweet spot (Papavassiliou et al., 2004).
Limitations of the study
We used normative connectome data to estimate seed-based connectivity in individual patients. This concept has been introduced for studies in clinical domains such as stroke (Darby et al., 2018; Joutsa et al., 2018a, 2018b), DBS (Horn et al. 2017b, Fox et al., 2014) or TMS (Weigand et al., 2018) where patient-specific connectivity data is often lacking. Although these connectome atlases do not represent patient-specific connectivity, they in turn have the benefit of high signal-to-noise ratios. The functional connectome we used was defined on a high N (1000 subjects) and was acquired using specialized MR hardware (Yeo et al., 2011). In addition, the structural connectome was calculated using a modern approach that was best performer among 10 different tractography processing algorithms in an open competition (Fillard et al., 2011). Finally, this limitation should bias our results toward non-significance to predict out-of-sample data, but instead, the models proved highly robust in cross-validation.
Second, the retrospective design of our study poses a limitation. Needless to say, our exemplary attempt to validate the model on a single case scanned with patient-specific diffusion MRI should only be considered as an anecdotal evidence. Despite the good performance of our models in predicting individual outcome, a prospective multi-center study is needed to translate our results into clinical practice. Additionally, our side-effects connectivity analysis was based on small number of patients and did not involve a quantitative assessment of side-effects. As a consequence, results did not survive correction for multiple comparisons. Nevertheless, these results could be used to form hypotheses for further studies that may specifically address the connectivity fingerprints of VIM-DBS induced side-effects.
Third, inter-individual anatomical variability implies another challenge in predicting individual optimal DBS target using an optimal target from a group analysis. Nevertheless, our target was built on a connectome-based model which emphasizes the importance of targeting structural and functional connectivity between DBS electrode and regions delineated by the predictive models (specifically M1 and the cerebellum).
Lastly, our cohort assumed a single category of tremor syndromes, namely essential tremor. this could be of concern since other tremor syndromes equally benefit from DBS surgery (Kumar et al., 2003; Herzog et al., 2004; Foote et al., 2006; Mandat et al., 2010; Kilbane et al., 2015; Cury et al., 2017). For example, Parkinsonian tremor is successfully treated with subthalamic nucleus (Diamond et al., 2007) and VIM (Kumar et al., 2003) DBS. How connectivity patterns of effective DBS therapy could predict tremor reduction across different targets and tremor semiology remains to be established.
Conclusion
We identified patterns of connectivity that allow to predict individual clinical outcomes of DBS in ET patients. More specifically, we introduced somatotopic connectivity maps that bear the potential of steering DBS targeting and programming toward patient-specific profiles with respect to the body distribution of symptoms. Finally, we estimated an “optimal” DBS target and set it into relationship to known ET-DBS targets. Our target is based on the convergence of beneficial functional and structural connectivity patterns and is available as a probabilistic, deformable atlas that we made openly available within the software Lead-DBS. Our results add to the ongoing effort of connectivity-based DBS targeting and foster the advance of connectomic surgery.
Funding
B.A. is supported by a Doctoral Research Grant from the German Academic Exchange Service – DAAD. A.A.K is supported by DFG project grants KU 2261/13-1 and EXC 2049/1. The study was supported by DFG Emmy Noether grant 410169619 to A.H
Competing interests
A.H. reports one-time lecture fee by Medtronic unrelated to the present work. A.A.K. reports personal fees and non-financial support from Medtronic, personal fees from Boston Scientific, personal fees from Ipsen Pharma, grants and personal fees from St. Jude Medical outside the current work. B.A., D.Ku, D.Kr and S.E. have nothing to disclose.
Acknowledgements
We would like to thank Igor Ilinsky and Kristy-Kultas Ilinsky (University of Iowa) for their helpful comments on our results. We also thank Wolf-Julian Neumann (Charité – University Medicine Berlin) for his valuable advice and comments.
Footnotes
We added two major analyses: First, we analysed connectivity profiles associated with the emergence of side effects (dysarthria and ataxia). Crucially, largely distinct regions were involved in electrodes associated with side effects. In addition, we preoperatively scanned a novel patient with diffusion MRI to successfully predict his improvement using patient specific fibertracts. Albeit this analysis merely adds anecdotal evidence (and we present the results in supplementary material), it may line out that from a methodological standpoint, our pipeline could be used in clinical practice. Finally, we replicated our main findings using a different statistical method that works on the level of fibertracts directly (instead of voxels) and was recently introduced by our group (Baldermann et al., Biological Psychiatry). In doing so, we further confirm the validity of our results using a different method.