Structural Connectivity of the Basal Ganglia from Patient-Individual Tractography Is Key for Understanding the Effects of Deep Brain Stimulation in Parkinson’s Disease

Abstract

Introduction

In Parkinson’s disease (PD) patients, modulation of the fibre tracts of the cortico-basal ganglia-thalamo-cortical loop is the presumed mechanism of action of deep brain stimulation (DBS) of the subthalamic nucleus (STN). Therefore, we explored patient-individual cortical structural connectivity of the volume of tissue activated (VTA), as well as DBS-induced modulation of fibre tracts connecting the STN with cortical and subcortical nodes, and their correlation with therapeutic effects.

Methods

A retrospective cohort of n = 69 PD patients treated with bilateral DBS of the STN was analysed. Clinical response was assessed from the DBS-induced change in the UPDRS-III motor scores (total and symptom-specific sub-scores) under regular medication after a median follow-up of 9.0 (range 2.6–20.2) months. Tractography based on patient-individual diffusion-weighted MRI was employed in two ways. Whole-brain tractography was used to identify the cortical connections of fibres passing the VTAs, and reconstruction of specific white matter pathways of the motor loop connecting the STN with the basal ganglia and cortex was used to identify the proportion of fibres within these pathways which was modulated by STN-DBS. This proportion of pathway modulation was used in a correlative analysis with clinical outcomes.

Results

Fibres traversing the VTAs were primarily connected to the supplementary motor area (SMA) and to a lesser degree to the premotor cortex. Within the pathways connecting the STN with the cortical and subcortical nodes, on average 30–40% (range 10–80%) of the fibres were modulated by STN-DBS. This proportion correlated significantly with the percentage change in UPDRS motor score for fibres connecting the STN with the SMA (ρ = 0.28), pre-SMA (ρ = 0.26), ventral and dorsal premotor cortices (ρ = 0.26 and ρ = 0.29, respectively), and the globus pallidus externus (ρ = 0.26) and internus (ρ = 0.29). Also, good clinical responses for both tremor and rigidity were associated with a significantly (p < 0.05) higher proportion of modulated fibres for the same cortico- and sub-cortico-STN connections.

Conclusion

Patient-individual tractography reveals that, in PD, most of the cortical fibres traversing the VTA are connected to the SMA. In addition, clinical efficacy is related to the proportion of DBS-affected fibres connecting the STN with nodes of both the hyperdirect (cortex-STN) and the indirect pathways (STN-basal ganglia). As such, patient-specific tractography, in particular in the basal ganglia, could be used in a clinical context as a tool to guide therapy.

Introduction

Everyday voluntary movements are the result of a well-orchestrated and balanced activity of a large number of skeletal muscles, which are initiated and executed almost effortlessly. It is assumed that the desired sequences of muscle activations are generated by temporally and spatially organised neuronal activity within the cortico-basal ganglia-thalamo-cortical motor loop [1] (see Fig. 1). According to a provisional model of movement and action control [2–4], an intended motor action first leads to fast, excitatory input from the cortex to the subthalamic nucleus (STN) via the hyperdirect pathway (HDP), resulting in broad inhibition of unnecessary or interfering movements by activation of the globus pallidus internus/substantia nigra pars reticularis (GPi/SNr), which in turn inhibits the thalamus and cortex. An appropriate motor program is then selected by activation of the striatum (STR), subsequent inhibition of the GPi/SNr via the direct pathway and release/disinhibition of the desired activity in the thalamus and cortex. This program is antagonised and/or terminated via the indirect pathway, which connects the STR to the GPi/SNr via the globus pallidus externus (GPe) and the STN.

Fig. 1.

Schematic of the functional neuroanatomy of the cortico-basal ganglia-thalamo-cortical motor loop. Some connections are omitted for the sake of simplicity.

In Parkinson’s disease (PD), loss of dopaminergic neurons in the substantia nigra pars compacta (SNc) leads to changes in the excitability of neurons in the putamen (PUT), a part of the STR, which results in pronounced motor symptoms including akinesia/bradykinesia, rigidity, and tremor. Loss of dopamine in the STR reduces the excitability of the direct pathway and increases the excitability of the indirect pathway, rendering the STN and GPi/SNr hyperactive and inducing pathological low-beta-band oscillatory synchrony of the basal ganglia and motor cortex, which is associated with the motor symptoms [5].

Deep brain stimulation (DBS) of the STN is an established therapy for patients with advanced PD who do not respond to drug treatment. Despite its well-documented efficacy, the exact mechanism of STN-DBS is only partially understood. The classical view is that high-frequency DBS suppresses local pathological STN activity and, as such, represents a reversible type of lesion [6]. However, DBS also induces remote effects such as ortho- and antidromic generation and propagation of action potentials, as well as affection of local synaptic transmission of pathways afferent to the site of stimulation, and may, as such, interfere with the pathological neural activity at various sites [7–9]. These network-level effects seem to play an important role in the mechanism of action of DBS, and increasingly more work is now focussing on identifying the fibre tracts that connect the above-mentioned cortical and subcortical structures and how these can be targeted by DBS [10, 11]. This connectomic approach is also based on the realisation that the STN has a three-part structure in which parts connected to the motor, associative, and limbic loops are in close proximity [12–14]. In addition, the supplementary motor area (SMA) seems to have a strong connection with the motor part of the STN via the HDP, which could represent a main target for DBS [15].

In several previously published reports, normative connectomes were used where average fibre tracts are constructed in standard brain templates and analysed in combination with patient-specific volumes of tissue activated (VTAs) by DBS [16–24]. Although this approach has led to valuable insights, it remains unclear to what extent the normative connectomes can represent the actual structural connectivity of individuals, mostly elderly patients. In the present study, we, therefore, performed state-of-the-art patient-specific tractography from preoperative diffusion-weighted MR imaging (DWI) in patients that received STN-DBS for PD. We first applied a whole-brain tractography reconstruction approach where we determined the specific cortical areas whose connections are modulated by STN-DBS by isolating the streamlines passing through the VTAs. To investigate correlations between clinical outcome and modulation of specific pathways, we used a second tracking approach where, based on the results of the first approach, we reconstructed the relevant pathways of the cortico-basal ganglia-thalamo-cortical loop for each patient and then computed the fraction of fibres within these tracts that were modulated by the VTAs. We hypothesised that these measures represent the magnitude of the modulatory effect in the different tracts of the motor loop and allow the identification of the fibre tracts whose modulation correlates with motor outcomes as defined in the Unified Parkinson’s Disease Rating Scale part III (UPDRS-III) scoring system.

Patients and Methods

Patients and Clinical Response Assessment

This retrospective study included 69 patients (23 females) with a median age of 63.7 (range 46.8–78.1) years with PD (Hoehn-Yahr stages 1–4; average disease duration 10.4, range 2.6–27.2 years, Table 1) who were treated with bilateral DBS of the STN at the University Hospital Cologne from 2016 to 2021 and received longitudinal neurological evaluation including UPDRS scores by the treating neurologists. DBS surgery was performed under the guidance of MR imaging fused with a stereotactic CT scan, as well as intraoperative electrophysiological and/or neurological monitoring. Clinical response was assessed from the DBS-induced change in the total UPDRS-III motor scores under regular medication (Med On) after an average interval of 9.0 (range 2.6–20.2) months, at which point the stimulation settings were taken for VTA calculation. In addition, pre- and postoperative sum scores for individual symptoms (tremor, rigidity, and bradykinesia) were extracted from the UPDRS-III score sheets. Pre- and postoperative medications were transformed into the levodopa-equivalent dose (LEDD) [25]. The study was approved by the Local Ethics Committee (No. 21-117).

Table 1.

Patient characteristics

Age, years	63.7 (46.8–78.1)	​
Sex (female/male)	26/43	​
Hoehn-Yahr stage (1/2/3/4)	5/28/25/11	​
Duration of disease, years	10.4 (2.6–27.2)	​
VTA, µL	54.0 (2.2–305.4)	​
Interval, months	9.0 (2.6 – 20.2)

​	Pre-DBS	Post-DBS
Levodopa-equivalent daily dose	1,096.6 (150–2,159)	541.5 (0–1,431)***
UPDRS-III (Med ON)	21.1 (4–43)	16.8 (2–47)**
Tremor	3.4 (0–20)	1.5 (0–7)**
Rigidity	6.0 (0–17)	3.7 (0–14)***
Bradykinesia	11.8 (0–28)	8.3 (0–19)***

Numbers are average (range) or counts.

***p < 0.001.

**p < 0.01.

Imaging Procedures

Preoperatively, patients underwent an MRI scanning battery (Philips 3T Ingenia Elition, Eindhoven, The Netherlands), which included structural T1-weighted (165 slices; TR = 9.7 ms; TE = 4.8 ms; flip angle = 8°; voxel size = 0.5 × 0.5 × 1 mm³) and T2-weighted (80 slices; TR = 3,000 ms; TE = 80 ms; flip angle = 90°; voxel size = 0.5 × 0.5 × 2 mm³), as well as DWI. The diffusion-weighted data were acquired using an echo-planar imaging sequence (60 slices without slice gap; TR = 8,141 ms; TE = 100 ms; 40 gradient directions with b-value 1,000 s/mm²; and a voxel size = 2 × 2 × 2 mm³). Additional echo-planar imaging images were acquired without diffusion weighting (b-values = 0 s/mm²) and opposing phase-encoding directions for correction of susceptibility artefacts. One day postoperatively, patients underwent a CT scan to control the position of the DBS electrodes.

Pre-processing of the DWI data included noise reduction [26] and correction of eddy current artefacts using TORTOISE’s (https://tortoise.nibib.nih.gov) [27, 28] diffprep routine. Susceptibility artefacts were corrected via drbuddi [29], using the T2-weighted image as a reference. Afterwards, field inhomogeneities were corrected using the routine dwibiascorrect of MRtrix3 (https://www.mrtrix.org) [30] with the ants algorithm [31], and finally the images were resampled to 1.3 × 1.3 × 1.3 mm³ resolution according to the MRtrix recommendations.

MRtrix3 was also used to model the diffusion signal based on constrained spherical deconvolution (CSD) [32]. First, patient-specific tissue response functions were calculated using dwi2response, and study-specific tissue responses were obtained by averaging the patient-specific ones, using responsemean. These latter response functions were then used to calculate the patient-specific fibre orientation distributions (FOD) using sst3_csd_beta1 from MRtrix3Tissue (https://3tissue.github.io), a fork of MRtrix3 which has been shown to be better suited for DWI data acquired with a single b-value [33]. Finally, the FOD data were normalised using mtnormalise.

Further image processing included the registration of T2-weighted images to the MNI 2009c non-linear asymmetric standard brain template [34] through a combination of linear and non-linear transformations, calculated using ANTs [35]. In addition, postoperative CTs were registered to the structural MRI data using linear transformations (ANTs) and VTAs were generated using the SimBio [36] and FieldTrip [37] suites, using the default settings available through LEAD-DBS (https://www.lead-dbs.org, v2.5.3) [38].

Tractography

We investigated two different fibre tracking approaches, both using MRtrix3 tckgen with the ifod2 algorithm. First, we reconstructed a whole-brain tractogram using anatomically constrained tractography [39], where 20 million tracts were generated by seeding from the white matter-grey matter interface generated using the T1-weighted image. The VTAs were then superimposed on the whole-brain tractogram and used to filter streamlines running through them. The number of streamlines crossing the VTA and terminating in ipsilateral cortical regions, as defined in the fine-grained Glasser atlas [40], was identified for both hemispheres, averaged across hemispheres for each patient, and visualised in FreeSurfer (https://surfer.nmr.mgh.harvard.edu).

Since the streamline counts obtained in the first approach cannot be readily pooled across patients – see Discussion – they are not suited for a correlation analysis with the clinical outcomes. Therefore, in a second approach, we generated tractograms of relevant pathways in the cortico-basal ganglia-thalamo-cortical loop – (Fig. 1) – and identified the proportion of modulated fibres for each pathway. Cortical regions-of-interest (ROI), as identified in the first approach – see Results – included the primary motor (M1) area, the SMA, the pre-SMA, and the dorsal and ventral premotor areas (PMd and PMv, respectively), and were defined by the human motor area template [41]. Subcortical structures of interest included the putamen, the GPe, the GPi, the STN, and the thalamus as defined by the DISTAL atlas [42]. These structures were warped to the individual T2 images and then resliced to match the resolution of the diffusion-weighted images. Connections between pairs of nodes in this network were individually tracked for each patient, in which each voxel in the start ROI was seeded 2,197 (13³) times – an isotropic lattice with step size 0.1 mm, overlayed on the 1.3 × 1.3 × 1.3 mm³ voxels. Streamlines were excluded if they entered any node in the network apart from those involved in the tracked pathway. The proportion of modulated fibres was calculated for each hemisphere by taking the ratio of affected streamlines (those overlapping with the VTA) to the total amount of reconstructed streamlines. These values were then averaged from both hemispheres and used to identify connections associated with an improvement in the total UPDRS-III score and in the symptom scores.

Statistical Analysis

Clinical scores for the pre- and postoperative evaluations were compared via Wilcoxon’s signed rank test. Since the total UPDRS-III score has a range between 0 and 108, it was regarded as a quasi-continuous variable despite being composed of ordinal items. The percentage change in UPDRS-III motor score was calculated by ΔUPDRS = 100 × (UPDRS_preop − UPDRS_postop)/UPDRS_preop, resulting in positive values in patients with an improvement. This measure was then correlated to the proportion of fibres modulated by STN-DBS using Spearman correlation. The same approach was applied to LEDD, where ΔLEDD = 100 × (LEDD_preop − LEDD_postop)/LEDD_preop.

In addition, pre- and postoperative scores for the symptoms tremor, rigidity, and bradykinesia were computed by adding the respective single items from the UPDRS scale (tremor: items 20 and 21, 7 sub-items; rigidity: item 22, 5 sub-items; bradykinesia: items 23–26, 8 sub-items). Since these scores stem from a reduced number of ordinal items in the UPDRS-III scale, a different definition of clinical response was employed. A linear regression between the baseline and follow-up values was used to model the expected response, from which the postoperative effect of STN-DBS for each patient could be determined based on the preoperative values. Patients who fared better than the mean response as indicated by the regression line were categorised as “good responders” and those who did not were categorised as “bad responders,” as demonstrated in Figure 2. The proportion of fibres modulated by the VTA in the different pathways and the ΔLEDD were then compared between these two groups with regard to the specific symptoms using the Mann-Whitney U-test. False discovery rate correction was conducted using Benjamini-Hochberg’s method, with an alpha value of 0.95.

Fig. 2.

Scatter plot of pre- vs. postoperative rigidity score. The black dashed line is the result of linear regression of the displayed points, showing the expected STN-DBS effect. Patients whose postoperative score is below the prediction line are considered good responders (blue circles), while patients whose score is above the predicted one are considered bad responders (red circles).

Results

The clinical course of the patients after STN-DBS is summarised in Table 1. All clinical metrics were significantly reduced postoperatively. Of note, the average LEDD decreased from 1,096.6 mg preoperatively to 541.5 postoperatively (p < 0.001). Average VTA size was 54.0 µL, ranging from 2.2 to 305.4 µL.

Examples of whole-brain VTA-filtered tractograms are shown in Figure 3, as well as the resulting cortical projections maps. The majority of the filtered streamlines were seeded in the SMA, followed by the pre-SMA, PMd, and PMv. Some streamlines seeded in the M1 and primary somatosensory cortex (S1) regions also traversed the VTAs, but to a much lesser extent.

Fig. 3.

Examples of patient-individual fibre density map of connections between the clinical VTA’s and the cortex (left) and the mean fibre density of the cortical connections of the VTAs filtered from whole-brain tractography in the total cohort in template space (right). S1, primary somatosensory cortex; M1, primary motor cortex; SMA, supplementary motor area; pre-SMA, pre-supplementary motor area; PMd, dorsal premotor cortex; PMv, ventral premotor cortex.

Typical examples of the tractography within the structures of the cortico-basal ganglia-thalamo-cortical loop are depicted in Figure 4. The number of reconstructed streamlines from the STN to the GPe and GPi were orders of magnitude higher than those connecting with the cortex, shown in Figure 5. With regard to the HDP, the premotor regions had the most anatomical connections with the STN, followed by the primary motor and sensory cortices, and finally the pre-SMA and SMA. Only fibre tracts involving the STN were affected by the VTAs. The proportion of modulated fibres is shown in Figure 6, where the average proportion amounted to 36.0 ± 16.7%, without any significant differences between the different pathways.

Fig. 4.

Pathway reconstructions in a representative patient. On the left, the involved ROIs are shown. On the other panels, reconstructed streamlines in each pathway (blue: unaffected by DBS, red: modulated by DBS) are shown. ROIs: cortex – green (involved in HDP); putamen – orange; GPe – pink; GPi – blue; STN – cyan; thalamus – purple; VTA – yellow.

Fig. 5.

Streamline counts of structural connections between the STN and cortical regions and basal ganglia nodes.

Fig. 6.

Percentage of fibres modulated by STN-DBS for each pathway.

Regarding the clinical effects assessed by the total UPDRS-III score, a positive significant correlation between the degree of modulation and symptom improvement was found for almost all pathways involving the STN, except those for M1 and S1, as summarised in Table 2. The strongest correlations were seen for the fibre connections between the STN and GPi, the STN and PMd, as well as STN and SMA, whose scatterplots can be seen in Figure 7.

Table 2.

Correlation analysis between proportion of fibres modulated by DBS and percentage change in UPDRS-III score

Fibre tract	Spearman correlation coefficient	p value
STN – SMA	0.28	0.019*
STN – pre-SMA	0.26	0.031*
STN – PMv	0.26	0.034*
STN – PMd	0.29	0.016*
STN – M1	0.21	0.086
STN – S1	0.21	0.088
STN – GPe	0.26	0.031*
STN – GPi	0.29	0.015*

STN, subthalamic nucleus; SMA, supplementary motor area; pre-SMA, pre-supplementary motor area; PMd/PMv, dorsal/ventral premotor area; M1, primary motor cortex; S1, primary somatosensory cortex; GPe/GPi, external/internal globus pallidus.

*p < 0.05; p values which survived Benjamini-Hochberg correction.

Fig. 7.

Scatterplots of the proportion of fibres modulated by STN-DBS and the percentage improvement in UPDRS-III score for the most significantly correlated pathways.

The distributions of the proportion of targeted fibres in the groups of good vs. bad responders as defined from the linear models for the UPDRS-III sub-scores are shown in Table 3. For rigidity, the good responders had a significantly higher proportion of fibres targeted for all tracts involving the STN (about 40% vs. 30%), and for tremor, this pattern was also seen except for connections between STN-M1 and STN-S1. For bradykinesia, no significant differences in the proportion of modulated fibres between the groups were observed. Of note, no significant correlation between ΔLEDD and ΔUPDRS was found, and the ΔLEDD in good responders did not significantly differ from that of the bad responders with respect to any of the sub-scores (see online supplementary Table S1, available at https://doi.org/10.1159/000546716).

Table 3.

Degree of pathway modulation by DBS in good and bad responders with respect to symptom scores

​	Good response	Bad response	p value
Tremor
STN – SMA	39.8±15.9	31.1±15.0	0.021*
STN – pre-SMA	42.9±18.1	31.0±16.3	0.006*
STN – PMv	40.4±16.0	31.8±16.8	0.025*
STN – PMd	41.3±16.5	31.6±16.8	0.011*
STN – M1	34.4±14.9	30.2±19.1	0.146
STN – S1	34.8±15.3	31.4±18.9	0.124
STN – GPe	39.6±16.0	30.1±14.7	0.010*
STN – GPi	42.8±17.8	30.5±15.1	0.005*
Rigidity
STN – SMA	40.6±17.3	29.3±11.0	0.006*
STN – pre-SMA	42.7±19.7	30.4±13.0	0.009*
STN – PMv	41.7±17.7	29.3±12.2	0.004*
STN – PMd	42.4±18.1	29.3±12.2	0.002*
STN – M1	36.7±17.4	26.4±14.2	0.012*
STN – S1	37.2±17.1	27.7±15.2	0.012*
STN – GPe	40.1±17.4	28.6±10.9	0.005*
STN – GPi	43.5±18.8	28.7±11.3	0.001*
Bradykinesia
STN – SMA	36.0±14.4	36.0±17.9	0.768
STN – pre-SMA	37.4±15.6	38.1±21.1	0.796
STN – PMv	36.3±14.7	37.1±19.2	0.966
STN – PMd	36.8±14.9	37.3±19.7	0.861
STN – M1	31.7±14.2	33.6±19.3	0.899
STN – S1	32.2±14.7	34.6±19.3	0.687
STN – GPe	34.4±13.2	36.6±19.0	0.995
STN – GPi	35.8±15.4	39.4±20.1	0.535

*p values which survived Benjamini-Hochberg correction.

Discussion

Main Findings

Preoperative patient-specific tractography was used to identify the fibre tracts associated with optimal DBS of the bilateral STN in a cohort of PD patients. The fibres traversing the VTA resulting from stimulation parameters at follow-up were predominantly connected to the ipsilateral SMA, as shown by the whole-brain tractography reconstructions. Tractography of the cortico-basal ganglia-thalamo-cortical loop revealed that, on average, between 30 and 40% of the fibres connecting the STN with the cortical and subcortical regions of the motor loop were modulated by DBS. DBS-induced improvements of the UPDRS-III total score and sub-scores for tremor and rigidity correlated significantly with the VTA-modulated fraction of fibre connections between the STN and the pre-SMA, SMA, ventral/dorsal premotor cortex, as well as GPe and GPi.

Normative vs. Patient-Specific Connectivity Analyses

Many of the recent studies investigating the structural networks targeted by STN-DBS have applied normative connectomes in combination with patient-individual estimation of the VTAs DBS [16–24]. This procedure does not require the time-consuming acquisition of preoperative DWI data and can, therefore, be performed more easily in larger patient groups. However, it depends on the assumption that the average connectome of a cohort of healthy individuals or even PD patients from other remote centres is representative of an individual patient under investigation, at least in terms of general patterns of connectivity between the structures involved. As shown in Table 4, most of these analyses compared symptom scores 6–24 months after DBS to the preoperative situation, both under current medication (Med On) or after withdrawal of medication (Med Off). Quite consistently, these studies indicate that the improvement in overall UPDRS-III motor scores is associated with the targeting of fibres projecting to the (ipsilateral) SMA and M1, while targeting projections to other frontal and prefrontal structures were less often found to be predictive.

Table 4.

Overview of studies using normative connectomes

Study	N	Med.	DBS test condition	Follow-up, months	Outcome	SFG	preMC	PFC	Pre-SMA	SMA	M1	Tha
Horn et al. [16] (2017)a	51	Off	DBS On vs. Off	12–24	UPDRS-III	+	​	​	​	+	​	​
	44	Off	DBS On vs. pre-op	6–12	UPDRS-III	+				+
Wang et al. [17] (2021)	33	Off	DBS On vs. pre-op	3–12	UPDRS-III	​	​	+	+	​	​	​
Chen et al. [18] (2022)	98	Off	DBS On vs. pre-op	1	UPDRS-III	​	​	​	​	+	+	​
Strelow et al. [19] (2022)	47	On	DBS On vs. pre-op	6	Freezing of gait	​	+	+	​	+	+	​
Hacker et al. [22] (2023)	14	On	DBS On vs. pre-op	24	UPDRS-III	​	​	​	​	+	+	​
Gadot et al. [21] (2023)	40	Off	DBS On vs. pre-op	6	UPDRS-III	+	​	​	​	+	​	+
Fan [20] (2023)	76	On	DBS On vs. pre-op	12	Freezing of gait	​	​	​	+	+	​	​
Hollunder et al. [23] (2024)a	94	On	DBS On vs. Off	12	UPDRS-III	​	​	​	​	+	​	​
Rajamani et al. [24] (2024)a	237	Off	DBS On vs. Off	12–24	Tremor	​	​	​			+	​
			DBS On vs. pre-op	6–12	Bradykinesia					+
					Rigidity				+

SFG, superior frontal gyrus; preMC, premotor cortex; PFC, prefrontal cortex; Pre-SMA, pre-supplementary motor area; SMA, supplementary motor area; M1, primary motor cortex; Tha, thalamus.

^aPartially overlapping data sets.

In contrast, patient-specific connectomes were usually studied in smaller cohorts of 15–25 patients with shorter follow-up times and more variable outcome measures [17, 43–51] (Table 5). As with normative connectomes, most of the studies found that modulating the connection between the STN and SMA and M1 was associated with clinical improvements. As such, the general view that DBS targeting the fibres between the STN and SMA or M1 is of clinical benefit seems to be supported by both types of analyses. Furthermore, in a direct comparison between patient-specific connectomes, group connectomes from PD patients, and normative connectomes of healthy subjects, Wang et al. [17] observed a high degree of agreement between brain connectivity profiles of the clinical STN-VTAs. However, the pattern of cortical STN connections that were correlated with a favourable outcome differed substantially between the three types of connectomes, probably due to a greater variance in the patient-specific connectomes. The present study found a significant correlation between clinical benefit and SMA-STN pathway modulation, further confirming its role in the STN-DBS therapy. However, no significant correlation was found between clinical outcome and M1-STN pathway modulation, contradicting some of the literature, possibly due to variability in connectome approaches.

Table 5.

Overview of studies using patient-specific connectomes

Study	N	Med.	DBS test condition	Follow-up, months	Outcome	SFG	preMC	PFC	Pre-SMA	SMA	M1	Tha	GPi	GPe
Koirala et al. [43] (2016)	15	On/Off	DBS On/Med Off vs. pre-op Med On	n.d.	UPDRS-III					+	+
Vanegas-Arroyave et al. [44] (2016)	22	Off	DBS On	1	Clinical effect	+						+
Akram et al. [45] (2017)	20	Off	DBS On vs. Off	12	Tremor						+
					Bradykinesia					+
					Rigidity			+		+
Krishna et al. [47] (2019)	24	Off	DBS On vs. pre-op	1	Tremor		+				+
					Bradykinesia			+			+
					Rigidity			+
Avecillas-Chasin et al. [46] (2019)	13	On	DBS On vs. pre-op	25	UPDRS-III					+	+
					Bradykinesia					+
Vassal et al. [48] (2020)	9	Off	DBS On vs. pre-op	6	UPDRS-III		+			+	+
Wang et al. [17] (2021)	33	Off	DBS On vs. pre-op	3–12	UPDRS-III			+	+
Kahkola et al. [50] (2022)	22	On	DBS On vs. pre-op	12	UPDRS-III unilat.				+
Gonzalez-Escamilla et al. [49] (2022)	15	Off	DBS On vs. pre-op	3	UPDRS-III					+	+
Segura-Amil et al. [51] (2023)	20	Off	DBS On	4–6	Clinical effect					+	+
Present study	69	On	DBS On vs. pre-op	3–20	UPDRS-III		+		+	+
					Tremor		+		+	+			+	+
					Rigidity		+		+	+	+	+	+	+

n.d., not determinable from the manuscript; SFG, superior frontal gyrus; preMC, premotor cortex; PFC, prefrontal cortex; Pre-SMA, pre-supplementary motor area; SMA, supplementary motor area; M1, primary motor cortex; Tha, thalamus.

We here applied a high-angular resolution diffusion-weighted imaging method that has major advantages compared to the more standard diffusion tensor imaging often used in clinical practice. By using 40 diffusion directions, it allows the determination of the fibre orientations with reasonable resolution and to account for crossing fibres. The associated tractography method is based on a model applying so-called CSD [32] that delivers a FOD function in every voxel, which can accommodate multiple fibre populations. Coupled with probabilistic tractography, this pipeline is much more sensitive to complex fibre arrangements and, therefore, suited for this kind of analysis. Still, it can be applied in the clinical setting where it takes about 10 min to acquire the MR images.

It is important to note, however, that studies which employ patient-specific connectomes rely on the number of streamlines as a measure of connectivity between the VTA and specific target regions. The underlying assumption in such cases is that tractography-generated streamlines correlate with the number of axons which connect two nodes of interest, e.g., SMA and VTA. However, it is a well-known limitation of tractography that this is not the case: MR acquisition parameters, image quality, processing pipeline, pathway length, and underlying microstructure will all heavily influence the number of reconstructed streamlines, therefore confounding the analysis [52, 53]. A concrete example of such biases is shown in Figure 5, where shorter pathways (STN-GPe/GPi) resulted in many more streamlines than longer pathways (STN-Cortex). Part of the elegance of normative connectome analyses is that these avoid such confounding factors by making use of a predefined set of streamlines for the whole cohort, from which statistics are then drawn. In this work we stepped away from using raw streamline counts at the patient level, opting instead for tracking specific pathways of interest and investigating the degree of modulation in each individual hemisphere. In a way, this approach is similar to the application of a normative connectome, in that statistics are drawn from a set of common streamlines: whole tract and modulated tract share the same streamline set, with the difference being that we generate these common streamline sets for each patient. While the aforementioned confounds might not be fully resolved with the proposed approach, we believe this method is more reflective of the DBS effects than simply counting the number of streamlines emanating from the VTA.

Efficacy of DBS Targeting STN-Cortical and STN-Subcortical Connections

The high level of connectivity consistently observed between the VTAs and the SMA and premotor cortex suggests that the clinically useful VTAs in PD preferentially target fibre tracts of the HDP. As indicated by several studies in non-human primates [54] and humans [14, 55, 56-57], the HDP connects the primary motor cortex (M1), SMA, and premotor cortex with the motor part of the STN and, thus, constitutes a significant part of the motor loop in the tripartite model of cortico-basal ganglia-thalamo-cortical circuits [1, 58]. Of note, both experimental [54] and clinical [59] studies have identified separable clusters of the motor STN connected to the M1 and the SMA, each of which follows a somatotopic organisation. DBS targeting of fibres connected to different clusters of the motor STN could, therefore, have differential effects on the motor symptoms of PD. The results from the connectivity studies shown in Tables 4 and 5 suggest that tremor is attenuated by modulating fibre connections between the STN with the M1, while bradykinesia and rigidity are improved by stimulation of the SMA-STN or pre-SMA-STN fibre tracts [24, 45]. In the present study, we found that the clinical response of tremor and rigidity was associated with significantly higher fibre targeting not only between STN and pre-SMA/SMA, but also between STN and premotor cortex and STN and sensorimotor cortex, suggesting that modulation of a network rather than a single node contributes to the clinical efficacy of STN-DBS. Comparable observations have been made with MR guided high-focused ultrasound subthalamotomy, where distinct clusters corresponding to symptom improvements were identified within the motor STN [60]. The tremor-effective cluster corresponded to the part of the motor STN connected with the sensorimotor cortex; the bradykinesia-effective cluster fitted best with the SMA-connected cluster while the rigidity-effective cluster was located in between [60].

Apart from the HDP, the VTAs from STN-DBS usually also overlap with fibre tracts connecting the STN with the subcortical structures and, therefore, potentially modulate neural activity in the indirect pathway. Due to the dense arrangement and the complex fibre architecture, tractography of the basal ganglia nuclei and thalamus is more difficult and less reliable and has, therefore, been predominantly performed in publicly available high-quality diffusion MR data sets [42, 56, 61–63]. Hollunder et al. [23], by application of a normative connectome in conjunction with the Basal Ganglia Atlas [56], found that UPDRS-III improvement was associated with DBS targeting the fibres connecting the GPe with the premotor STN territory. In the present study, we also observed that STN-DBS targeted fibres connecting the STN with the GPe and GPi to a substantial amount and that this contributed to the clinical efficacy of DBS.

In the present study, we calculated the proportion of fibre connections between two structures passing through the VTA as a measure of modulation strength. Although this approach may not seem obvious at first glance, it has been shown to be useful in both clinical and theoretical studies. In a recent article by Segura-Amil et al. [51], the term “activation” was applied to streamlines passing through a given VTA, and it was shown that an activation of 50% of the HDP was required to achieve a clinical effect in PD patients. Kähkölä et al. [50] used patient-specific tractography to identify the cluster of the STN connected with the pre-SMA and found that only when this cluster was stimulated, i.e., more than 50% of the cluster covered by the VTA, did patients respond well to treatment in terms of unilateral motor improvement. Interestingly, in one of the most detailed large-scale simulation models of the cortico-basal ganglia network, a similar measure, i.e., DBS-induced percent fibre activation, was found to be a useful variable for predicting pathological thalamic activity [64], where the authors found that the optimal stimulation activated 88% of the HDP fibres, 56% of the STN-GPi fibres, and 46% of the STN-GPe fibres.

Relation between STN Structural Connectivity and Neurophysiological Effects of DBS

According to a recent comprehensive update on the neurophysiological mechanisms, DBS has both local and remote electrophysiological effects [9]. In the STN, DBS seems to mainly induce suppression of neural activity by activation of local GABA-ergic synapses, while the network effects are probably the result of antidromic propagation of action potentials along the long-range afferents. This view is supported by evidence from EEG, ECoG, MEG, and local field potential recordings showing that STN-DBS stimulation not only suppresses pathologic synchronised beta activity in the basal ganglia, but also modulates cortical beta activity [65, 66]. In addition, STN-DBS reduces cortical pathologic beta-gamma cross-frequency coupling and high beta-band coherence between cortex and STN [15]. Of note, at least the latter effect seems to be confined to the part connected to the SMA [7] and, as such, points to the HDP as an important structure supporting the modulatory effect of STN-DBS. This view is further strengthened by a combined local field potential-MEG analysis where the high beta-band SMA activity was found to selectively drive STN activity by means of propagation along the HDP which depended on the fibre density of the HDP [15].

Recently, a new neurophysiological biomarker has been proposed that studies phase-amplitude coupling of the beta-band with high-frequency oscillations in the STN, rather than looking at beta-power itself. Beta-high-frequency oscillation phase-amplitude coupling was specifically present in the motor STN, and the respective VTAs were structurally mainly connected to the SMA in contrast to the other contacts that influenced much wider frontal and parietal regions [67]. Together, these results confirm the hypothesis that modulation of the activity of the HDP and especially the of the fibres connecting the SMA with the motor STN plays an important role for the clinical effect of STN-DBS.

In principle, activity in the HDP could also be modulated by direct stimulation of the motor cortex, which has, indeed, been evaluated in a limited number of patients not suitable for STN-DBS [68]. However, as argued by Cioni et al. [68], motor cortex stimulation is performed below the threshold for movements and presumably interferes with small inhibitory axons in the cortex and orthodromic or antidromic activation of fibres connecting the motor cortex to the basal ganglia, rather than acting on the pyramidal cells. It may, as such, decrease cortical excitability or disrupt oscillatory rhythms and abnormal patterns of activity and seems to have some positive long-term effects, mainly on axial symptoms.

Limitations

The data of the present study originate from patients who were examined at different time intervals after implantation of the DBS electrodes. After STN-DBS surgery, lesion effects can take some months to wane and patients often experiment in multiple stimulation settings until an optimal programme is found. Our cohort thus potentially included patients with clinically sub-optimal settings. On the other hand, DBS effects may also wane over time and even with perfectly implanted electrodes, effects might not be seen at long follow-up times. It is, however, worth noting that including sub-optimal stimulation parameters can be of statistical benefit as it reduces selection bias and helps reveal the true underlying effect. Although we aimed to identify the main individual structural connections within the basal ganglia, the present tractography method is probably still too crude to resolve the dense arrangement and complex architecture of curved and collateralised fibre tracts within the Forel field. These fibres connect, among others, the GPi and STN to the thalamic subnuclei, run in close proximity to the zona incerta and have also been shown to be effective targets in DBS for PD [69–72].

Our cohort was also evaluated in the “on medication” state. Clinical outcomes evaluated here are thus potentially an effect of both medication and STN-DBS, and these cannot be truly separated. However, as indicated in a separate analysis, clinical response was not significantly correlated with the reduction of LEDD, supporting the view that the reduced need for medication represents an additional effect of DBS. Additionally, the UPDRS-III scores, which reflect only motor symptoms, are a single domain of affliction of PD patients. Patients with PD are often also afflicted by non-motor symptoms, which were not evaluated here. A more comprehensive, multi-centre analysis is already planned, where all these effects will be taken into consideration to better characterise the relation between STN-DBS and overall clinical outcome.

Conclusion

Patient-individual tractography reveals that, in PD, most of the cortical fibres that run through the VTA connect to the SMA. In addition, clinical efficacy is related to the proportion of DBS-modulated fibres connecting the STN with nodes of both the HDP and the indirect pathway. As such, patient-specific tractography, in particular in the basal ganglia, could be used in a clinical context as a tool to guide surgical therapy.

Acknowledgement

We would like to thank E. Güngör for her support with data collection.

Statement of Ethics

This study protocol was reviewed and approved by the Ethics Committee of the Medical Faculty of the University of Cologne, Approval No. 21-1117. All patients gave their informed written consent for the treatment with deep brain stimulation. Due to the retrospective character of the study, the Ethics Committee granted an exemption from requiring written informed consent for the analysis of clinical and imaging data.

Conflict of Interest Statement

Andres M. Lozano is a consultant for Medtronic, Boston Scientific, Abbott, Insightec, and Functional Neuromodulation. Veerle Visser-Vandewalle and Andres M. Lozano were both members of the journal’s Editorial Board at the time of submission. All other authors have no conflicts of interest to declare.

Funding Sources

Funds have been provided by the European Joint Programme Neurodegenerative Disease Research (JPND) 2020 call “Novel imaging and brain stimulation methods and technologies related to neurodegenerative diseases” for the Neuripides project “Neurofeedback for Self-Stimulation of the Brain as Therapy for Parkinson Disease.” The Neuripides project has received funding from the following funding organisations under the aegis of JPND: The Netherlands Organisation for Health Research and Development (ZonMw), the Netherlands; Federal Ministry of Education and Research (BMBF), Germany; Ministry of Education, Youth and Sports (MEYS), Czech Republic; French National Research Agency (ANR), France; Canadian Institutes of Health Research (CIHR), Canada; and Scientific and Technological Research Council of Turkey (TUBITAK), Turkey.

Author Contributions

R.L. – conceptualisation, methodology, formal analysis, writing of original draft, and manuscript review and editing. M.K. – conceptualisation, methodology, writing of original draft, and manuscript review and editing. G.A.B., J.N.P.-S., M.B., H.D., and J.W. – resources, data curation, and manuscript review and editing. J.M. and R.J. – methodology and manuscript review and editing. V.V.-V. – supervision, conceptualisation, funding acquisition, and manuscript review and editing. P.A. and D.E.J.L. – supervision, conceptualisation, and manuscript review and editing. H.B., M.L., B.S., A.L., A.B., B.J., and T.C. – conceptualisation and manuscript review and editing.

Funding Statement

Funds have been provided by the European Joint Programme Neurodegenerative Disease Research (JPND) 2020 call “Novel imaging and brain stimulation methods and technologies related to neurodegenerative diseases” for the Neuripides project “Neurofeedback for Self-Stimulation of the Brain as Therapy for Parkinson Disease.” The Neuripides project has received funding from the following funding organisations under the aegis of JPND: The Netherlands Organisation for Health Research and Development (ZonMw), the Netherlands; Federal Ministry of Education and Research (BMBF), Germany; Ministry of Education, Youth and Sports (MEYS), Czech Republic; French National Research Agency (ANR), France; Canadian Institutes of Health Research (CIHR), Canada; and Scientific and Technological Research Council of Turkey (TUBITAK), Turkey.

Data Availability Statement

The individual clinical and imaging data supporting the results of this study are not publicly available as they contain information that could compromise patient privacy but aggregated and/or anonymised data are available upon reasonable request to the author R.L.