Frontal Lobe Connectivity and Network Community Characteristics are Associated with the Outcome of Subthalamic Nucleus Deep Brain Stimulation in Patients with Parkinson’s Disease

Abstract

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is nowadays an evidence-based state of the art therapy option for motor and non-motor symptoms in patients with Parkinson's disease (PD). However, the exact anatomical regions of the cerebral network that are targeted by STN-DBS have not been precisely described and no definitive pre-intervention predictors of the clinical response exist. In this study, we test the hypothesis that the clinical effectiveness of STN-DBS depends on the connectivity profile of the targeted brain networks. Therefore, we used diffusion-weighted imaging (DWI) and probabilistic tractography to reconstruct the anatomical networks and the graph theoretical framework to quantify the connectivity profile. DWI was obtained pre-operatively from 15 PD patients who underwent DBS (mean age = 67.87 ± 7.88, 11 males, H&Y score = 3.5 ± 0.8) using a 3T MRI scanner (Philips Achieva). The pre-operative connectivity properties of a network encompassing frontal, prefrontal cortex and cingulate gyrus were directly linked to the postoperative clinical outcome. Eccentricity as a topological-characteristic of the network defining how cerebral regions are embedded in relation to distant sites correlated inversely with the applied voltage at the active electrode for optimal clinical response. We found that network topology and pre-operative connectivity patterns have direct influence on the clinical response to DBS and may serve as important and independent predictors of the postoperative clinical outcome.

Full text

Vol.:(0123456789)
1 3
Brain Topogr
DOI 10.1007/s10548-017-0597-4
ORIGINAL PAPER
Frontal Lobe Connectivity andNetwork Community
Characteristics areAssociated withtheOutcome ofSubthalamic
Nucleus Deep Brain Stimulation inPatients withParkinson’s
Disease
NabinKoirala1· VinzenzFleischer1· MartinGlaser2· KirstenE.Zeuner3·
GüntherDeuschl3· JensVolkmann4· MuthuramanMuthuraman1· SergiuGroppa1
Received: 2 May 2017 / Accepted: 26 September 2017
© Springer Science+Business Media, LLC 2017
distant sites correlated inversely with the applied voltage at
the active electrode for optimal clinical response. We found
that network topology and pre-operative connectivity pat-
terns have direct influence on the clinical response to DBS
and may serve as important and independent predictors of
the postoperative clinical outcome.
Keywords Parkinson’s disease· Deep brain stimulation·
Structural connectivity· Community structures· Network
analysis
Abbreviations
AAL Automated anatomical labeling
AUC Area under the curve
BCT Brain connectivity toolbox
COG Center of gravity
DBS Deep brain stimulation
DWI Diffusion-weighted imaging
FWHM Full width at half maximum
H & Y Hoehn and Yahr
MED OFF/ON Medication off/on
MPRAGE Magnetization-prepared rapid
gradient-echo
ROC Receiver operating characteristic
ROI Region of interest
SMA Supplementary motor area
STN Subthalamic nucleus
UPDRS Unified Parkinson’s disease rating scale
VTA Volume of tissue activation
Introduction
Parkinson’s disease is one of the most common neurode-
generative diseases with no permanent cure. Deep brain
Abstract Deep brain stimulation (DBS) of the subthalamic
nucleus (STN) is nowadays an evidence-based state of the
art therapy option for motor and non-motor symptoms in
patients with Parkinson’s disease (PD). However, the exact
anatomical regions of the cerebral network that are targeted
by STN–DBS have not been precisely described and no
definitive pre-intervention predictors of the clinical response
exist. In this study, we test the hypothesis that the clini-
cal effectiveness of STN–DBS depends on the connectivity
profile of the targeted brain networks. Therefore, we used
diffusion-weighted imaging (DWI) and probabilistic tractog-
raphy to reconstruct the anatomical networks and the graph
theoretical framework to quantify the connectivity profile.
DWI was obtained pre-operatively from 15 PD patients who
underwent DBS (mean age = 67.87 ± 7.88, 11 males, H&Y
score = 3.5 ± 0.8) using a 3T MRI scanner (Philips Achieva).
The pre-operative connectivity properties of a network
encompassing frontal, prefrontal cortex and cingulate gyrus
were directly linked to the postoperative clinical outcome.
Eccentricity as a topological-characteristic of the network
defining how cerebral regions are embedded in relation to
Muthuraman Muthuraman and Sergiu Groppa have contributed
equally.
* Sergiu Groppa
segroppa@uni-mainz.de
1 Department ofNeurology, Johannes Gutenberg University,
55131Mainz, Germany
2 Department ofNeurosurgery, Johannes Gutenberg University,
55131Mainz, Germany
3 Department ofNeurology, University ofKiel, 24105Kiel,
Germany
4 Department ofNeurology, University ofWürzburg,
97080Würzburg, Germany

Brain Topogr
1 3
stimulation of the subthalamic nucleus (STN–DBS) is cur-
rently a standard evidence-based treatment for PD patients
that substantially improves the motor and non-motor symp-
toms (Weaver etal. 2012; Odekerken etal. 2013; Klingel-
hoefer etal. 2014). Despite the clear clinical benefits, the
STN–DBS mechanisms remain unclear (Udupa and Chen
2015). Furthermore, the STN target and the activity within
it do not have an unequivocal justification for the clinical
response. Earlier studies presented evidence that the stimula-
tion might modulate the neuronal activity within the STN,
while recent studies showed that the DBS rather targets the
fibers entering, exiting or passing the stimulation region and
not just the STN itself (McIntyre and Hahn 2010; Nambu
and Chiken 2015). Moreover, the modulation of the patho-
logical oscillations in distinct brain networks might be a
critical feature of the DBS-induced clinical response (Stein
and Bar-Gad 2013; Brittain and Brown 2014). Studies on
primates and recent studies on humans attested the existence
of the so-called hyperdirect cortical STN projections which
might be of special importance for the effects of STN–DBS
(Brunenberg etal. 2012a; Haynes and Haber 2013).
Apart from the therapeutic benefit, DBS can cause severe
adverse effects. These effects might be long-term and com-
plex like the cognitive decline and neuro-psychiatric symp-
toms which occur with STN–DBS (Fukaya and Yamamoto
2015). Hence, the understanding of the local and systemic
interactions, together with an exact description of the tar-
geted cerebral network would markedly advance our thera-
peutic strategies and improve the efficacy and reliability and
reduce the side effects of actual DBS protocols by refin-
ing the patient selection and developing improved targeting
strategies.
We hypothesize that the preoperative connectivity pattern
of the interconnected regions is closely related to the clini-
cal effectiveness of STN–DBS. Therefore, we use diffusion-
weighted imaging (DWI) and probabilistic tractography to
reconstruct the anatomical networks and the graph theoreti-
cal framework to quantify their connectivity profile. Net-
work topology properties defined within graph theoretical
analyses have become important measurable characteristics
that explain healthy brain dynamics and help to understand
and model brain disorders (Bullmore and Sporns 2009). The
study of network characteristics in PD using graph theory
in structural MRI has revealed specific global and local
functional network topological changes characterized by a
decrease in clustering and path length between the nodes in
comparison to healthy controls (Olde Dubbelink etal. 2014).
Additionally, structural network analysis using cortical and
subcortical anatomical measurements has shown larger
characteristic path length and reduced global efficiency in
addition to a lower regional efficiency in frontal and parietal
regions for mild cognitive impaired PD patients compared
to healthy controls. This provides support for the role of
aberrant network topology in motor and cognitive impair-
ment in patients with early PD (Pereira etal. 2015). In this
study, we link brain network properties with the postopera-
tive clinical outcome to test the hypothesis that the clinical
response to STN–DBS depends on the connectivity pattern
of the interconnected brain regions and stimulation site.
Methods
Subjects andData Acquisitions
In this study, fifteen patients (4 females and 11 males) with
idiopathic PD without dementia and receiving DBS treat-
ment were selected with a mean age of 67.87 ± 7.88years.
For all patients, a high resolution T1-image of the brain
using MPRAGE sequence (TR = 7.7ms, TE = 3.6ms, flip
angle = 8°, 160 slices) was obtained with a 3T MR-Scan-
ner (Philips Achieva) using an 8-channel SENSE head
coil before the DBS surgery. DWI of the whole brain at
2mm isometric voxel resolution covering a field of view of
224 × 224mm was obtained. We recorded three acquisitions
of DWI sequences encompassing 32 gradient directions and
five b0 (no diffusion weighting) images for each acquisi-
tion (b value = 1000s/mm2, TE = 59 ms, TR = 11,855ms,
fat saturation “on”, 60 contiguous slices). The total acquisi-
tion time for the whole protocol was 35min which included
24min (3 × 8min) for DWI sequences. On the first postop-
erative day a further MRI acquisition was performed on a
1.5T scanner (Philips Achieva, Philips Medical Systems,
Best, The Netherlands) with a protocol consisting of a
T1-weighted structural image of the whole brain using a
standard MPRAGE sequence (TR = 10.7ms, TE = 1.96ms,
flip angle = 8°).
Stimulation Parameters
The surgical procedure is previously explained in detail
elsewhere (Groppa etal. 2014). All patients were implanted
with bilateral STN electrodes (model 3389 DBS, Medtronic)
and pulse generators (Activa® PC, biphase stimulation)
with a pulse setting of 60µs at 130Hz. DBS electrode volt-
age is the stimulation intensity at the active electrode as
expressed in Volts for the optimal clinical response. The
voltage was adjusted for each individual patient while the
remaining stimulation parameters (pulse width and stimu-
lation frequency) were left unchanged. The stimulation
adjustment was performed by clinicians who were blinded
to the hypothesis and goals of this study, and values from the
follow-up at 3months after the implantation were included
in the analysis. We choose this time point since no more
impedance changes occur at this post-operative period and
voltage values remain constant. The medical treatment was

Brain Topogr
1 3
individualized after DBS according to the clinical indication.
The study protocol used was approved by the local ethics
committee and all patients signed a written consent regard-
ing the procedure. The clinical details are shown in Table1.
Data Analysis andNetwork Reconstruction
The obtained images were preprocessed using inbuilt func-
tionality in FSL described in detail elsewhere (Jenkinson
and Smith 2001; Jenkinson etal. 2002; Johansen-Berg
etal. 2004; Behrens etal. 2007). In brief, susceptibility and
motion artefacts correction and diffusion tensor modelling
were performed using the diffusion toolbox (topup and FDT,
part of FSL). Crossing fibers distribution was estimated
using BEDPOSTX (implemented in FSL) and the probabil-
ity of major (f1) and secondary (f2) fiber directions were cal-
culated. All images were aligned and affine-transformed into
the stereotactic space MNI-152. A multi-fiber model was
fit to the diffusion data at each voxel, allowing for tracing
of fibers through regions of crossing or complexity. Here,
we drew 5000 streamline samples from each seed voxel to
form an estimate of the probability distribution of connec-
tions from each seed voxel. When these streamlines reach
a voxel in which more than one direction is estimated, they
follow the direction that is closest to the direction at which
the streamline arrives. Tracts generated are volumes wherein
values at each voxel represent the number of samples (or
streamlines) that passed through that voxel. Here each tract
from every seed mask in the atlas is repeatedly sampled and
only those tracts which passed through at least one other
seed mask were retained. The obtained streamlines were
then used to build the connectivity matrix. For the elimina-
tion of spurious connections, tractography in individual sub-
jects was thresholded to include only voxels through which
at least 10 percent of all streamline samples had passed.
A connectivity matrix was obtained using the seed masks
for 116 regions of interest (ROI) defined by the Automated
Anatomical Labeling atlas (Tzourio-Mazoyer etal. 2002)
for each subject (Fig.1). The links or the entries in the con-
nectivity matrix represent the ratio of number of samples
(or streamlines) that passes through ROI (j) to all generated
streamlines from ROI (i). This weighted connectivity index
between ROIs in the matrix was then analyzed using various
network measures (both global and local) obtained via Brain
Connectivity Toolbox (Rubinov and Sporns 2010) (https://
sites.google.com/site/bctnet/).
Community Analysis
In the reconstructed network we aimed to detect intercon-
nected regions forming functionally relevant entities such
as modules. Modules are groups of nodes that have more
connections within themselves than expected in a randomly
sampled group of nodes (Meunier etal. 2009). Module
detection performs a partitioning of the brain into entities
with higher within, than between module correlations (Gir-
van and Newman 2002; Newman 2006). The modules were
identified using the Louvain modularity algorithm as imple-
mented in BCT (Blondel etal. 2008) for each individual sub-
ject connectivity matrix. We performed 5000 iterations with
the Louvain algorithm and the assignment of each region to
a particular module was based on the maximum number of
times/iterations a region was assigned to a module (Ritchey
etal. 2014).
Random Network Formation andComparison
The network measures (see below) were computed in the
obtained modules. To depict neurobiologically meaningful
network properties we compared the values obtained from
each module with values from generated random networks.
The random network was formed by weight reshuffling tech-
nique (Opsahl etal. 2008) for each individual subject. The
procedure reshuffles the weights globally in the network but
maintains the topology of the observed network (also for
each module) and permits a distinct delineation to the devel-
opment of neurodegeneration that leads to a loss of distinct
topology patterns in patients.
Network Measures
In this study we assessed three network measures: charac-
teristic path length, eccentricity and global efficiency. For
a given network, these measures can characterize the effi-
ciency of information transfer at different levels and reveal
the importance of distinct regions within the module.
Characteristic path length is the average shortest dis-
tance between the regions in the network and is a measure
of global integration (Stam etal. 2009). Global efficiency
has been introduced as a network integration parameter to
Table 1 Clinical detail. Clinical parameters assessed before and after
the DBS surgery
Here SD is the standard deviation
Parkinson’s disease patients (n = 15) mean ± SD
Sex (male/female) 11/4
Age (years) 67.8 ± 7.8
Disease duration (years) 13.6 ± 6.5
Preoperative Hoehn–Yahr stage (Med OFF) 3.8 ± 0.8
Preoperative Hoehn–Yahr stage (Med ON) 2.7 ± 0.5
Preoperative UPDRS III (Med OFF) 34.5 ± 8.4
Preoperative UPDRS III (Med ON) 17.4 ± 9.0
Preoperative dose of levodopa or equivalent (mg/day) 827 ± 397.0
Postoperative dose of levodopa or equivalent (mg/day) 335 ± 202.6

Brain Topogr
1 3
describe information flow over the entire network and is
computed as the average inverse shortest path length. Even
though global efficiency and characteristic path length are
very similar measures, here we selected both as charac-
teristic path length could possibly be more influenced by
disconnected or remote nodes, while global efficiency is
more robust to such extremes (Latora and Marchiori 2001).
Eccentricity is defined as the maximal shortest path length
between any two regions in the network and is a measure of
relative nodal importance (Sporns 2003).
Predicting Anatomical Regions intheNetworks
To establish the role of network modules and distinct ana-
tomical regions within the module, we designed a receiver
operating characteristic (ROC) curve-based analysis. There-
fore, we tested how the connectivity profile of each region
is ranked in comparison to others by comparing the areas
under the curve (AUC). The ROC curves were obtained
using the eccentricity values of the analyzed module and
compared against values from the same module obtained
from the random network. Moreover, for the selection of
these involved modules, only those modules with an AUC
above the threshold of no discrimination (i.e. greater than
0.5), and having a significantly (one sided t-test, p < 0.05)
higher network measures value than in the generated random
networks were considered. This approach ascertained the
modules and regions within the module, for each network
measure, with the strongest association to the outcome vari-
able (i.e. improvement in the motor UPDRS score or DBS
voltage) and having preserved network topology that dif-
fers from random networks. Furthermore in this study, we
concatenated the side-specific and axial values of UPDRSIII
value to obtain a complete evaluation of the clinical out-
come. Since the significant areas were not side specific and
we aimed to develop universal measures that can be imple-
mented in the clinical practice without a priori bias of handi-
ness and laterality of the symptoms, we opted to perform
whole brain analysis.
Determination ofContact‑Specific Masks
Electrode positions and electrode trajectory were determined
using post-operation T1 images. The detailed procedure is
explained elsewhere (Witt etal. 2013). Briefly, the lead
was mathematically modelled by a straight line along the
Fig. 1 Overview of methods. a Preoperative DWI obtained from 15
Parkinson’s patients. b Probabilistic tractography was used to obtain
the connectivity matrix using 116 ROIs as depicted in c. d Illus-
tration of DBS electrodes in STN which was performed in all 15
patients. e The connectivity matrix with the probability from a seed
region to another region. f The random network connectivity matrix
obtained by weight reshuffling of the obtained matrix. g Subdivision
of the whole brain into various modules using the Louvain algorithm
(gamma = 2) represented by different colors. h Visualization of the
network comparison performed between the random network and the
obtained network indicating various network analyses performed

Brain Topogr
1 3
electrode trajectory. Based on the optimized lead position
the T1 intensity profile was extracted along the trajectory
and the electrode contact positions were then determined
by observing the intensity dip apparent in the extracted
intensity profile. Geometrically determined electrode con-
tact positions were used to create spatially Gaussian weight-
ing masks (Fig.2). Gaussian weights were determined by
specifying the following two standard deviations: (i) along
the lead to model contact dimensions, known from manu-
facturer’s annotations; (ii) from the two other orthogonal
directions to model the stimulation depth. We restricted our
analysis to one different mask extension: a Gaussian shape
with 2 standard deviations along the lead and 2 standard
deviations in depth (corresponds to an isometric mask with
4.7mm full width at half maximum (FWHM), correspond-
ing to a radius of ca. 2.35mm). The multivariate Gaussians
were centered at the exact contact positions. These volumes
were selected considering existing literature that attest that
neural elements up to a distance of 2mm from the active
contact might be reached by the studied DBS stimulation
settings (Ranck 1975). The target coordinates for STN were
defined relative to the midpoint of the anterior and posterior
commissure [mid AC PC] on the T1 MPRAGE images used
for stereotactic planning. To assess if the position of the
electrodes in each patient has an effect on voltage applied or
the post-operative UPDRS, we computed the geodesic dis-
tance of each calculated center of gravity (COG) of the gen-
erated electrode volume of tissue activation (VTA) masks
from the obtained STN position and stereotactic target and
hence correlated it with the DBS parameters (collective
UPDRSIII values and DBS voltage).
Statistical Analysis
The significant difference between the values obtained
for each module in the analyzed network and the random
network generated was assessed by performing a t-test
(p < 0.05). The correlations between the obtained network
measures and DBS clinical parameters were obtained using
Pearson’s correlation coefficient. To ascertain the signifi-
cance of the correlation obtained, further leave-one-out
analysis was performed. Finally, ROC curve analysis was
performed to compare the connectivity profile of the mod-
ules with strongest association to the clinical outcome. All
of the statistical analyses were performed using MATLAB
(ver. 2013, Mathworks, Inc.).
Results
Clinical Assessment
We included 15 patients with PD who were selected for
STN–DBS therapy. The UPDRSIII scores improved after
DBS–STN in both MED OFF [Before: 34.46 ± 8.35, After:
13.86 ± 5.99 (t = 7.75, p < 0.001)] and MED ON state
[Before: 17.4 ± 9.03, After: 11.54 ± 5.77 (t = 2.09, p < 0.05)].
The H & Y scale scores improved after DBS–STN in MED
OFF state from 3.8 ± 0.78 to 2.7 ± 0.48 (t = 3.76, p < 0.01),
and in MED ON state from 2.75 ± 0.54 to 1.9 ± 0.51
(t = 3.59, p < 0.01).
Community Detection
Community detection using the Louvain algorithm
depicted 10 modules encompassing the regions shown in
Fig.3. The ROC curve analysis indicated that 6 of these
modules had an AUC significantly above the threshold
of no-discrimination (Table2). Significant differences in
the analyzed network measures in comparison with the
generated random networks were attested in two mod-
ules: Left-frontal and Central, signifying cerebral regions
with preserved anatomical architecture. Left-frontal com-
prises primarily the prefrontal cortical regions of the left
Fig. 2 DBS electrode
VTA masks. The Gauss-
ian weighted masks overlaid
in MNI152-1mm standard
template for all patients. ‘L’ and
‘R’ represent the left and right
hemisphere respectively

Brain Topogr
1 3
hemisphere (7 regions of AAL atlas) while module Cen-
tral chiefly includes the frontal lobe and premotor corti-
cal regions comprising 13 regions of the AAL atlas (see
Table2 for detail).
Network Properties andPostoperative Outcome
In the depicted modules, the network parameters eccentric-
ity, characteristic path length and global efficiency showed
Fig. 3 Modules visualization.
The whole brain subdivided
into various modules using the
Louvain algorithm (gamma = 2)
represented by different colors.
Anatomical annotations are
presented in Table2
Table 2 Modules and ROIs.
Regions of Interest (ROIs) from
AAL atlas assigned to the 10
modules obtained using the
Louvain algorithm as visualized
in Fig.2
1. Left frontal 4. Central Putamen_R
Precentral_L Supp_Motor_Area_L Pallidum_R
Frontal_Sup_L Supp_Motor_Area_R Thalamus_R
Frontal_Mid_L Cingulum_Mid_L Heschl_L
Frontal_Inf_Oper_L Cingulum_Mid_R Temporal_Sup_L
Frontal_Inf_Tri_L Cingulum_Post_L Temporal_Mid_L
Frontal_Sup_Medial_L Cingulum_Post_R 8. Left parieto-temporal
Postcentral_L Calcarine_L Hippocampus_L
2. Right frontal Cuneus_L ParaHippocampal_L
Precentral_R Occipital_Sup_L Amygdala_L
Frontal_Sup_R Precuneus_L Lingual_L
Frontal_Mid_R Precuneus_R Occipital_Inf_L
Frontal_Inf_Oper_R Paracentral_Lobule_L Occipital_Inf_R
Frontal_Inf_Tri_R Paracentral_Lobule_R Fusiform_L
Frontal_Sup_Medial_R 5. Occipital Postcentral_R
3.Frontal-subcortical Calcarine_R Parietal_Sup_R
Frontal_Sup_Orb_L Cuneus_R Parietal_Inf_R
Frontal_Mid_Orb_L Occipital_Sup_R Angular_R
Frontal_Inf_Orb_L Occipital_Mid_R Heschl_R
Rolandic_Oper_L 6. Frontoorbital Temporal_Sup_R
Olfactory_L Frontal_Sup_Orb_R Temporal_Pole_Sup_L
Olfactory_R Frontal_Mid_Orb_R Temporal_Mid_R
Frontal_Mid_Orb_L Frontal_Inf_Orb_R Temporal_Pole_Mid_L
Frontal_Mid_Orb_R 7. Right parieto-temporal Temporal_Inf_L
Rectus_L Rolandic_Oper_R 9. Temporal
Rectus_R Insula_R Temporal_Inf_R
Insula_L Occipital_Mid_L Hippocampus_R
Cingulum_Ant_L Parietal_Sup_L ParaHippocampal_R
Cingulum_Ant_R Parietal_Inf_L Amygdala_R
Caudate_L SupraMarginal_L Lingual_R
Putamen_L SupraMarginal_R Fusiform_R
Pallidum_L Angular_L Temporal_Pole_Sup_R
Thalamus_L Caudate_R Temporal_Pole_Mid_R
10. Cerebellum
Vermis and Cerebellum regions in module

Brain Topogr
1 3
an association with both the postoperative stimulation
parameter (DBS voltage) for an optimal clinical response
and the motor score (UPDRSIII). The computation of net-
work metrics over the entire network did not predict the
clinical outcome after DBS. In the modules with significant
differences compared to the random network (Left-frontal
and Central), eccentricity showed a significant interdepend-
ency with the DBS voltage (Left-frontal: r = − 0.36, p < 0.05;
Central: r = − 0.39, p < 0.05). Eccentricity values from the
Left-frontal and Central modules also correlated signifi-
cantly with the postoperative motor outcome as defined from
UPDRSIII (module Left-frontal: r = 0.47; Central: r = 0.45,
both p < 0.05) affirming the impact on the clinical state.
To determine the role of distinct anatomical regions
within a module, we compared the AUC from the ROC
analysis of eccentricity values from different areas within the
module. In the module Left-frontal, we determined that the
connectivity pattern from the medial superior frontal gyrus,
primarily involved in executive functions, the dorsolateral
superior frontal gyrus, and the opercular part of the inferior
frontal gyrus could be used as classifiers to assess predictive
model accuracy (Fig.4).
In the module Central, the eccentricity in the frontal
cortex and regions of the limbic network represented ana-
tomical areas within the module with the best accuracy for
effect delineation. In this module, the eccentricity values
from the SMA, posterior cingulate gyrus, median cingu-
late and paracingulate gyrus were predictors for the opti-
mal modeling accuracy (Fig.5).
The correlation between the position of the electrode
VTA mask and voltage or postoperative UPDRSIII was
not significant (p > 0.1). Furthermore, the permutation test
(randomise tool in FSL) performed using the VTA masks
and voltage and postoperative UPDRSIII also did not yield
any significant (i.e. p < 0.05, corrected) clusters.
Discussion
The proposed brain connectivity analysis and commu-
nity detection methods revealed a significant association
between the level of modular topology and connectiv-
ity in the frontal cortex and postoperative outcome after
STN–DBS in PD patients. Patients with a more efficient
(less degraded) network topology in the frontal cortex
(with higher eccentricity in comparison with random
network) would need less voltage to achieve the optimal
motor response. MRI-derived connectivity measures might
serve as important apparative and examiner independent
predictors for the clinical DBS-outcome and might be used
to optimize patient selection.
Fig. 4 Networks comparison in module ‘Left frontal’. Figure in
upper left shows the brain mesh and the included brain regions
(spheres). Yellow lines indicate the weighted connectivity between
them. Upper right: The histogram shows the difference in the net-
work parameters mean with standard deviation from eccentricity (E),
characteristic path length (λ) and global efficiency (G) of PD in com-
parison to random networks. The asterisk represents the statistically
significant difference observed (p < 0.05). Lower left: ROC curves
obtained from the regions within the module. The sensitivity–speci-
ficity plot was obtained from the analysis between the network meas-
ures from PD and random network. The highlighted lines show the
regions in this module with the highest AUC. The plot on the lower
right shows the correlation between eccentricity and DBS voltage in
this module

Brain Topogr
1 3
Network Structure Detection inPatients forSTN–DBS
Brain functionality can be characterized by local interactions
and a global integration in modules for specific brain func-
tions (Park and Friston 2013). The modular and hierarchi-
cal organization to support the effectivity of brain functions
has already been established non-invasively in humans as
derived from structural and functional MRI network analyses
(Sporns and Zwi 2004; van den Heuvel and Sporns 2011).
It is obvious from the complex layered, modular and hierar-
chical cerebral organization that high frequency stimulation
does not only affect the STN itself, but also distant areas.
Furthermore, the complex interaction with the PD pathology
is not only focused on substantia nigra or STN regions but
encompasses an altered connectivity between interconnected
regions comprising prefrontal, frontal and limbic regions.
To improve actual and future therapeutic strategies an exact
understanding of the complex interplay of PD pathology
and local and global effects of actual therapy strategies i.e.
STN–DBS is needed (Canu etal. 2015). Hence, the pro-
posed community and modular structures analysis has the
highly promising possibility of the exact characterisation of
the interconnected networks that are reached by STN–DBS.
The revealed modules encompass prefrontal, frontal
(SMA) and limbic cortex (cingulate and paracingulate
cortex). The described network of prefrontal and fron-
tal regions is mainly involved in executive functions and
is impaired early in the disease course in PD patients
(Nagano-Saito etal. 2005; Amboni etal. 2008; Koshimori
etal. 2015). Direct connections of STN to these regions
have been described (Brunenberg etal. 2012b). Two recent
studies have shown the importance of these connections
for STN–DBS efficiency, in Parkinsonian rats (Li etal.
2012) showed that STN–DBS activates the layer V neu-
rons in the motor cortex, which contributes to the disrup-
tion of abnormal neural activities, while (Gradinaru etal.
2009) demonstrated a direct and therapeutically beneficial
activation of motor and premotor areas with STN stimula-
tion using optogenetics and solid-state optics. In humans,
only indirect intimations regarding the distributed effects
of DBS exist. Grey matter cortical thickness analysis on
patients with PD has shown widespread cortical thinning
in frontal and premotor regions in comparison to healthy
controls (Pereira etal. 2012). In our previous study (Muth-
uraman etal. 2017), we found that the cortical integrity
is affected mainly in frontal lobe (paracentral area and
superior frontal region) and is correlated with DBS param-
eters and clinical outcome. Furthermore, a recent study has
shown that the modulation of white matter tracts directed
to the superior frontal gyrus and the thalamus is associated
Fig. 5 Networks comparison in module ‘Central’. Figure in upper
left shows the brain mesh an the included brain regions (spheres).
Yellow lines indicate the weighted connectivity between them. Upper
right: The histogram shows the difference in the network parameters
mean with standard deviation from eccentricity (E), characteristic
path length (λ) and global efficiency (G) of PD in comparison to ran-
dom networks. The asterisk represents the statistically significant dif-
ference observed (p < 0.05). Lower left: ROC curves obtained from
the regions within the module. The sensitivity–specificity plot was
obtained from the analysis between the network measures from PD
and random network. The highlighted lines show the regions in this
module with the highest AUC. The plot on the lower right shows the
correlation between eccentricity and DBS voltage in this module

Brain Topogr
1 3
with favorable clinical outcomes and may contribute to the
therapeutic effects of STN–DBS (Vanegas-Arroyave etal.
2016). Our data demonstrates the role of the connectivity
pattern in the frontal and central regions for the STN–DBS
clinical response in patients with PD and presents a first
and direct translational proof of the description of the tar-
geted networks from animal models.
Anatomical Regions ofRelevance forSTN–DBS
Eccentricity is computed as the maximal shortest distance
between two nodes (in this case two brain regions of inter-
est). It is a measure which shows how reachable a brain
region is from others in a specific module (Pavlopoulos
etal. 2011). A module with higher eccentricity values
has an architecture that permits a more efficient informa-
tion transfer to other regions; in this way, regions exert
a stronger influence over other ones and interact more
efficiently. In this study, we demonstrate that eccentricity
values in modules encompassing prefrontal, frontal and
limbic cortex are negatively correlated with the DBS-
electrode voltage. The location of the electrode contact
in STN has been shown in previous studies to be directly
associated with the improvements in the cardinal symp-
toms (tremor, rigidity etc.); with the best clinical outcome
achieved by stimulation in the dorsolateral motor part of
the STN (Voges etal. 2002; Herzog etal. 2004). In our
previous study, we also showed that the precise targeting
of the lateral region of the STN is essential for achiev-
ing sufficient stimulation efficacy (Wodarg etal. 2012).
Although it is possible to identify the STN using MRI
in image-guided surgery, it is still not feasible with the
post-operative low resolution MRI or T1—MPRAGE or
DWI sequences used in this study. The placement of the
lead and choice of active electrode in the STN holds sig-
nificant importance for the postoperative outcome. These
could be tested with available probabilistic STN atlases
however this would also be an approximation to the entire
STN region and not only the focused sensori-motor part.
Therefore we opted to calculate the COG of the registered
VTA in order to confirm only minimal differences in the
location of the active electrodes. In our study, the location
of the active contacts for DBS stimulation for the patient
group was not linked to the collective UPDRSIII values or
the applied voltage for the optimal postoperative outcome.
This could be due to the minimal variation of the location
of the active contacts between the subjects. Even though
the computation of the VTA is the gold standard for the
computation of the targeted neural tissue in DBS studies
(McIntyre etal. 2004; Butson etal. 2007), with image
processing and registration errors, further interpolations
of absolute distances occur which is one limitation of the
method and hence this study.
Predictive Effects ofNetwork Degeneration
andComparison toRandom Architecture
The human brain exhibits small-world topology, a favora-
ble property for efficient information transfer (Stam 2004;
Achard and Bullmore 2007). In addition, this topology is
associated with improved cognitive and motor performance
(Micheloyannis etal. 2006; van den Heuvel etal. 2009).
Neurodegenerative disorders cause disturbances in the opti-
mal organization of brain function and increase randomness
in the network (van Straaten and Stam 2013). Functional net-
works in Alzheimer’s disease lose their normal small-world
structure and regress towards a more random architecture
(Stam etal. 2007). The loss of overall functional connec-
tivity and small-world properties with increased random-
ness of the network was also shown for individuals with
schizophrenia (Liu etal. 2008; Yu etal. 2011). A reduced
network efficiency has been shown in the functional network
in PD patients by the aid of functional MRI (Skidmore etal.
2011). In our study, we observed that in modules encom-
passing frontal and central cortical areas, the eccentricity
values in patients with PD were higher than that of the ran-
dom network indicating the existence of small-world and
hierarchical topology in those modules. The network reor-
ganization with neurodegeneration toward architecture with
less small-world properties and increased randomness moti-
vated us to depict modules with highly organized topology
in comparison to random networks. Hence, we observed that
the effectiveness of STN–DBS is only depicted in modules
where the network topology is preserved and hierarchical
characteristics exist.
Conclusion
The performed network analysis revealed that connectivity
properties of a network in the frontal and central regions are
closely linked to the postoperative clinical outcome follow-
ing STN–DBS. Eccentricity as a topological-characteristic
of the network defining how cerebral regions are embedded
in relation to distant sites shows a clear association between
structural architecture and clinical outcome to functional
DBS neuromodulation in PD. The implementation of con-
nectivity profile analysis into the clinical setting might be an
important tool to help define and improve the postoperative
outcome.
Acknowledgements This work was supported by the German
Research Foundation (DFG; CRC-TR-128).

Brain Topogr
1 3
References
Achard S, Bullmore E (2007) Efficiency and cost of economical brain
functional networks. PLoS Comput Biol 3(2):e17
Amboni M, Cozzolino A, Longo K, Picillo M, Barone P (2008)
Freezing of gait and executive functions in patients with Par-
kinson’s disease. Mov Disord 23(3):395–400
Behrens TE, Berg HJ, Jbabdi S, Rushworth MF, Woolrich MW
(2007) Probabilistic diffusion tractography with multiple fibre
orientations: what can we gain? Neuroimage 34(1):144–155
Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E (2008) Fast
unfolding of communities in large networks. J Stat Mech
2008(10):10008
Brittain JS, Brown P (2014) Oscillations and the basal ganglia: motor
control and beyond. Neuroimage 85:637–647
Brunenberg EJ, Moeskops P, Backes WH, Pollo C, Cammoun L,
Vilanova A etal (2012a) Structural and resting state functional
connectivity of the subthalamic nucleus: identification of motor
STN parts and the hyperdirect pathway. PloS ONE 7(6):e39061
Brunenberg EJL, Moeskops P, Backes WH, Pollo C, Cammoun L,
Vilanova A etal (2012b) Structural and resting state functional
connectivity of the subthalamic nucleus: identification of motor
STN parts and the hyperdirect pathway. PloS ONE 7(6):e39061
Bullmore E, Sporns O (2009) Complex brain networks: graph theo-
retical analysis of structural and functional systems. Nat Rev
Neurosci 10(3):186–198
Butson CR, Cooper SE, Henderson JM, McIntyre CC (2007) Patient-
specific analysis of the volume of tissue activated during deep
brain stimulation. Neuroimage 34(17113789):661–670
Canu E, Agosta F, Sarasso E, Volonte MA, Basaia S, Stojkovic T
etal (2015) Brain structural and functional connectivity in
Parkinson’s disease with freezing of gait. Hum Brain Mapp
36(12):5064–5078
Fukaya C, Yamamoto T (2015) Deep brain stimulation for parkin-
son’s disease: recent trends and future direction. Neurologia
Medico-Chirurgica 55(5):422–431
Girvan M, Newman MEJ (2002) Community structure in social and
biological networks. Proc Natl Acad Sci USA 99(12):7821–7826
Gradinaru V, Mogri M, Thompson KR, Henderson JM, Deisseroth K
(2009) Optical deconstruction of parkinsonian neural circuitry.
Science 324(5925):354–359
Groppa S, Herzog J, Falk D, Riedel C, Deuschl G, Volkmann J (2014)
Physiological and anatomical decomposition of subthalamic neu-
rostimulation effects in essential tremor. Brain 137(Pt 1):109–121
Haynes WI, Haber SN (2013) The organization of prefrontal-sub-
thalamic inputs in primates provides an anatomical substrate
for both functional specificity and integration: implications for
Basal Ganglia models and deep brain stimulation. J Neurosci
33(11):4804–4814
Herzog J, Fietzek U, Hamel W, Morsnowski A, Steigerwald F,
Schrader B etal (2004) Most effective stimulation site in sub-
thalamic deep brain stimulation for Parkinson’s disease. Mov
Disord 19(9):1050–1054
Jenkinson M, Smith S (2001) A global optimisation method for robust
affine registration of brain images. Med Image Anal 5(2):143–156
Jenkinson M, Bannister P, Brady M, Smith S (2002) Improved opti-
mization for the robust and accurate linear registration and
motion correction of brain images. Neuroimage 17(2):825–841
Johansen-Berg H, Behrens TE, Robson MD, Drobnjak I, Rushworth
MF, Brady JM etal (2004) Changes in connectivity profiles
define functionally distinct regions in human medial frontal
cortex. Proc Natl Acad Sci USA 101(36):13335–13340
Klingelhoefer L, Samuel M, Chaudhuri KR, Ashkan K (2014) An
update of the impact of deep brain stimulation on non motor
symptoms in Parkinson’s disease. J Parkinson’s Dis 4(2):289–300
Koshimori Y, Segura B, Christopher L, Lobaugh N, Duff-Canning
S, Mizrahi R etal (2015) Imaging changes associated with cog-
nitive abnormalities in Parkinson’s disease. Brain Struct Funct
220(4):2249–2261
Latora V, Marchiori M (2001) Efficient behavior of small-world net-
works. Phys Rev Lett 87(19):198701
Li Q, Ke Y, Chan DCW, Qian ZM, Yung KKL, Ko H etal (2012)
Therapeutic deep brain stimulation in Parkinsonian rats directly
influences motor cortex. Neuron 76(5):1030–1041
Liu Y, Liang M, Zhou Y, He Y, Hao Y, Song M (2008) etal. Disrupted
small-world networks in schizophrenia. Brain 131(Pt 4):945–961
McIntyre CC, Hahn PJ (2010) Network perspectives on the mecha-
nisms of deep brain stimulation. Neurobiol Dis 38(3):329–337
McIntyre CC, Mori S, Sherman DL, Thakor NV, Vitek JL (2004)
Electric field and stimulating influence generated by deep brain
stimulation of the subthalamic nucleus. Clinical Neurophysiol
115(3):589–595
Meunier D, Achard S, Morcom A, Bullmore E (2009) Age-related
changes in modular organization of human brain functional net-
works. Neuroimage 44(3):715–723
Micheloyannis S, Pachou E, Stam CJ, Vourkas M, Erimaki S, Tsirka
V (2006) Using graph theoretical analysis of multi channel
EEG to evaluate the neural efficiency hypothesis. Neurosci Lett
402(3):273–277
Muthuraman M, Deuschl G, Koirala N, Riedel C, Volkmann J, Groppa
S (2017) Effects of DBS in parkinsonian patients depend on the
structural integrity of frontal cortex. Scientific Rep 7:43571
Nagano-Saito A, Washimi Y, Arahata Y, Kachi T, Lerch J, Evans A
etal (2005) Cerebral atrophy and its relation to cognitive impair-
ment in Parkinson disease. Neurology 64(2):224–229
Nambu A, Chiken S (2015) Mechanism of DBS: inhibition, Excitation,
or Disruption? In: Itakura T (ed) Deep brain stimulation for neuro-
logical disorders: theoretical background and clinical application.
Springer International Publishing, Cham, pp13–20
Newman MEJ (2006) Modularity and community structure in net-
works. Proc Natl Acad Sci USA 103(23):8577–8582
Odekerken VJJ, van Laar T, Staal MJ, Mosch A, Hoffmann CFE, Nijs-
sen PCG etal (2013) Subthalamic nucleus versus globus pallidus
bilateral deep brain stimulation for advanced Parkinson’s disease
(NSTAPS study): a randomised controlled trial. Lancet Neurol
12(1):37–44
Olde Dubbelink KT, Hillebrand A, Stoffers D, Deijen JB, Twisk JW,
Stam CJ etal (2014) Disrupted brain network topology in Par-
kinson’s disease: a longitudinal magnetoencephalography study.
Brain 137(Pt 1):197–207
Opsahl T, Colizza V, Panzarasa P, Ramasco JJ (2008) Prominence and
control: the weighted rich-club effect. Phys Rev Lett 101:168702
Park H-J, Friston K (2013) Structural and functional brain networks:
from connections to cognition. Science 342(6158):1238411
Pavlopoulos GA, Secrier M, Moschopoulos CN, Soldatos TG, Kossida
S, Aerts J etal (2011) Using graph theory to analyze biological
networks. BioData Mining 4:10
Pereira JB, Ibarretxe-Bilbao N, Marti M-J, Compta Y, Junqué C,
Bargallo N etal (2012) Assessment of cortical degeneration
in patients with Parkinson’s disease by voxel-based morphom-
etry, cortical folding, and cortical thickness. Hum Brain Mapp
33(11):2521–2534
Pereira JB, Aarsland D, Ginestet CE, Lebedev AV, Wahlund LO, Sim-
mons A etal. (2015) Aberrant cerebral network topology and mild
cognitive impairment in early Parkinson’s disease. Hum Brain
Mapp 36(8):2980–2995
Ranck JB (1975) Which elements are excited in electrical stimula-
tion of mammalian central nervous system: a review. Brain Res
98(3):417–440
Ritchey M, Yonelinas AP, Ranganath C (2014) Functional connectiv-
ity relationships predict similarities in task activation and pattern

Brain Topogr
1 3
information during associative memory encoding. J Cogn Neu-
rosci 26(5):1085–1099
Rubinov M, Sporns O (2010) Complex network measures of brain con-
nectivity: uses and interpretations. Neuroimage 52(3):1059–1069
Skidmore F, Korenkevych D, Liu Y, He G, Bullmore E, Pardalos PM
(2011) Connectivity brain networks based on wavelet correlation
analysis in Parkinson fMRI data. Neurosci Lett 499(1):47–51
Sporns O (2003) Graph theory methods for the analysis of neural con-
nectivity patterns. In neuroscience databases. Springer, New York,
pp. 171–185
Sporns O, Zwi JD (2004) The small world of the cerebral cortex. Neu-
roinformatics 2(2):145–162
Stam CJ (2004) Functional connectivity patterns of human magnetoen-
cephalographic recordings: a ‘small-world’ network? Neurosci
Lett 355(1–2):25–28
Stam CJ, Jones BF, Nolte G, Breakspear M, Scheltens P (2007) Small-
world networks and functional connectivity in Alzheimer’s dis-
ease. Cereb Cortex 17(1):92–99
Stam CJ, de Haan W, Daffertshofer A, Jones BF, Manshanden I, van
Cappellen van Walsum AM etal (2009) Graph theoretical analysis
of magnetoencephalographic functional connectivity in Alzhei-
mer’s disease. Brain 132(1):213–224
Stein E, Bar-Gad I (2013) Beta oscillations in the cortico-basal ganglia
loop during parkinsonism. Exp Neurol 245:52–59
Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard
O, Delcroix N etal (2002) Automated anatomical labeling of acti-
vations in SPM using a macroscopic anatomical parcellation of
the MNI MRI single-subject brain. Neuroimage 15(1):273–289
Udupa K, Chen R (2015) The mechanisms of action of deep brain
stimulation and ideas for the future development. Prog Neurobiol
133:27–49
van Straaten ECW, Stam CJ (2013) Structure out of chaos: functional
brain network analysis with EEG, MEG, and functional MRI. Eur
Neuropsychopharmacol 23(1):7–18
van den Heuvel MP, Sporns O (2011) Rich-Club organization of the
human connectome. J Neurosci 31(44):15775–15786
van den Heuvel MP, Stam CJ, Kahn RS, Pol HE (2009) Efficiency of
functional brain networks and intellectual performance. J Neurosci
29(23):7619–7624
Vanegas-Arroyave N, Lauro PM, Huang L, Hallett M, Horovitz SG,
Zaghloul KA etal. (2016) Tractography patterns of subthalamic
nucleus deep brain stimulation. Brain 139(4), 1200–1210.
Voges J, Volkmann J, Allert N, Lehrke R, Koulousakis A, Freund HJ
etal. (2002) Bilateral high-frequency stimulation in the subtha-
lamic nucleus for the treatment of Parkinson disease: correlation
of therapeutic effect with anatomical electrode position. J Neuro-
surg 96(2):269–279
Weaver FM, Follett KA, Stern M, Luo P, Harris CL, Hur K etal (2012)
Randomized trial of deep brain stimulation for Parkinson disease
Thirty-six-month outcomes. Neurology 79(1):55–65
Witt K, Granert O, Daniels C, Volkmann J, Falk D, van Eimeren T
etal (2013) Relation of lead trajectory and electrode position to
neuropsychological outcomes of subthalamic neurostimulation
in Parkinson’s disease: results from a randomized trial. Brain
136(7):2109–2119
Wodarg F, Herzog J, Reese R, Falk D, Pinsker MO, Steigerwald F
etal (2012) Stimulation site within the MRI-defined STN predicts
postoperative motor outcome. Mov Disord 27(7):874–879
Yu Q, Sui J, Rachakonda S, He H, Gruner W, Pearlson G etal (2011)
Altered topological properties of functional network connectivity
in schizophrenia during resting state: a small-world brain network
study. PloS ONE 6(9):e25423