FORSCHUNG
Parkinson’s disease speech production network
as determined by graph-theoretical
network analysis
Jana Schill1
, Kristina Simonyan2,3
, Simon Lang1, Christian Mathys4,5,6,
Christiane Thiel5,7, and Karsten Witt1,5
1Department of Neurology, School of Medicine and Health Sciences, University of Oldenburg, Oldenburg, Deutschland
2Department of Otolaryngology, Head and Neck Surgery, Harvard Medical School, Boston, MA, USA
3Department of Otolaryngology, Head and Neck Surgery, Massachusetts Eye and Ear, Boston, MA, USA
4Institute of Radiology and Neuroradiology, Evangelisches Krankenhaus, University of Oldenburg, Oldenburg, Deutschland
5Research Center Neurosensory Science, University of Oldenburg, Oldenburg, Deutschland
6Department of Diagnostic and Interventional Radiology, University of Düsseldorf, Düsseldorf, Deutschland
7Abteilung für Psychologie, School of Medicine and Health Sciences, University of Oldenburg, Oldenburg, Deutschland
Schlüsselwörter: Parkinson’s disease, Speech production network, Functional magnetic resonance
Bildgebung, Netzwerkanalyse, Emotional distraction
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
/
T
/
e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D
l
F
/
/
/
/
/
7
2
7
1
2
2
1
1
8
3
7
8
N
e
N
_
A
_
0
0
3
1
0
P
D
T
.
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
ABSTRAKT
Parkinson’s disease (PD) can affect speech as well as emotion processing. We employ whole-
brain graph-theoretical network analysis to determine how the speech-processing network (SPN)
changes in PD, and assess its susceptibility to emotional distraction. Functional magnetic
resonance images of 14 Patienten (gealtert 59.6 ± 10.1 Jahre, 5 weiblich) Und 23 healthy controls
(gealtert 64.1 ± 6.5 Jahre, 12 weiblich) were obtained during a picture-naming task. Pictures were
supraliminally primed by face pictures showing either a neutral or an emotional expression.
PD network metrics were significantly decreased (mean nodal degree, P < 0.0001; mean nodal
strength, p < 0.0001; global network efficiency, p < 0.002; mean clustering coefficient, p <
0.0001), indicating an impairment of network integration and segregation. There was an absence
of connector hubs in PD. Controls exhibited key network hubs located in the associative cortices,
of which most were insusceptible to emotional distraction. The PD SPN had more key network
hubs, which were more disorganized and shifted into auditory, sensory, and motor cortices after
emotional distraction. The whole-brain SPN in PD undergoes changes that result in (a) decreased
network integration and segregation, (b) a modularization of information flow within the network,
and (c) the inclusion of primary and secondary cortical areas after emotional distraction.
AUTHOR SUMMARY
Parkinson’s disease (PD) features a variety of motor and nonmotor symptoms, which relate to
changes in functional brain connectivity. This study investigated how the whole-brain speech
production network (SPN) is affected by PD. The paradigm included an emotional distraction
component to assess the SPN’s susceptibility to an emotional influence. The SPN was found to be
less integrated and less segregated in PD than in healthy controls. Emotional distraction had a
greater effect on the SPN in PD than in healthy controls. Key nodes for information flow (network
pillars) were mainly located in associative cortices. They were robust against emotional
distraction in healthy controls, but shifted to primary and secondary cortical areas in PD.
a n o p e n a c c e s s
j o u r n a l
Citation: Schill, J., Simonyan, K., Lang,
S., Mathys, C., Thiel, C., & Witt, K.
(2023). Parkinson’s disease speech
production network as determined by
graph-theoretical network analysis.
Network Neuroscience, 7(2), 712–730.
https://doi.org/10.1162/netn_a_00310
DOI:
https://doi.org/10.1162/netn_a_00310
Supporting Information:
https://doi.org/10.1162/netn_a_00310
Received: 6 July 2022
Accepted: 13 February 2023
Competing Interests: The authors have
declared that no competing interests
exist.
Corresponding Author:
Jana Schill
schill.jana@gmail.com
Handling Editor:
Richard Betzel
Copyright: © 2023
Massachusetts Institute of Technology
Published under a Creative Commons
Attribution 4.0 International
(CC BY 4.0) license
The MIT Press
Parkinson’s disease speech production network
INTRODUCTION
Parkinson’s disease (PD) is the second most common neurodegenerative disease (Tysnes &
Storstein, 2017). Due to a predominant loss of dopaminergic cells in the substantia nigra
(SN) and other brain regions, a cascade of alterations in brain structure and function takes
place, leading to a variety of symptoms (Dickson, 2018; Obeso et al., 2000). While the causal
relation between SN cell loss and the emergence of motor symptoms has been well researched
(e.g., Foffani & Obeso, 2018; Frosini, Cosottini, Volterrani, & Ceravolo, 2017), the pathogen-
esis of PD nonmotor symptoms is more complex and elusive.
One class of symptoms of PD are speech difficulties, describing impairments of voice, flu-
ency, and articulation (Ho, Iansek, Marigliani, Bradshaw, & Gates, 1998). The neurological
basis of speech impairment in PD is not fully understood, as speech and language control
is affected by both, motor symptoms as well as nonmotor symptoms. Speech production is
a highly complex task involving perception, reasoning, emotional and cognitive skills, motor
programming, as well as a proper motor output. In PD the precision control of muscle inner-
vation necessary for a proper speech output is affected, reflecting impaired motor perfor-
mance. Language and speech control is furthermore affected by nonmotor symptoms related
to disturbances in emotion regulation, mood changes and cognitive difficulties as well as
drawling and salivation (Critchley, 1981; Jankovic, 2008). The targeted examination of
whole-brain functional brain networks and their changes has proven useful in studying and
understanding natural language formation (Fuertinger, Horwitz, & Simonyan, 2015) as well
as its changes in other movement disorders (Fuertinger & Simonyan, 2017). As PD has been
linked to changes within network structures due to different pathophysiological mechanisms, it
is likely the speech production network (SPN) is altered in PD. However, it has not been inves-
tigated in detail whether this is indeed the case.
In recent years, more and more disorders have been attributed to changes in brain net-
work function (Ko, Spetsieris, & Eidelberg, 2017; Palop & Mucke, 2016; Pini et al., 2020;
Schirinzi, Sciamanna, Mercuri, & Pisani, 2018). Instead of investigating the role of a specific
set of brain regions for a task as measured by their individual activation patterns, network
analysis looks at the interplay of these regions (Bullmore & Sporns, 2009). Given the multi-
factorial origin of PD symptoms driven by changes in multiple neurotransmitter systems and a
heterogeneous spread of pathology in the course of the disease, network analysis seems to be
specifically fruitful for revealing changes in PD brain network control. Graph-theoretical network
analysis is one category of network analyses that considers the different regions of the brain as
nodes of a graph and defines the edges of the graph by measures of association between
these brain regions (Bullmore & Sporns, 2009). Thereby, mathematical graph computations
can be used to characterize the functional connectivity of the brain. Network integration (the
network’s ability to propagate information efficiently throughout the network) and network
segregation (the network’s ability for specialized processing within densely interconnected
groups of nodes, i.e., clusters) provide insight into the workings of a network and therefore,
in the case of brain network analyses, shed light on how the brain processes information
(Rubinov & Sporns, 2010). Another approach for investigating network function is a hub
analysis (van den Heuvel & Sporns, 2013). Here, nodes that contribute most to the network
are identified as hubs and then analyzed in terms of their spatial distribution throughout the
network and their communication patterns with other nodes. By assigning highly intercon-
nected groups of nodes into different network modules, the modularity of the network can
be assessed. Hubs that connect different modules (connector hubs) facilitate network integra-
tion, whereas hubs that are dominantly connected to nodes of their own module facilitate
network segregation.
713
Graph-theoretical network analysis:
Model-free, data driven network
analysis approach that describes the
brain in form of a graph.
Functional connectivity:
Statistical association between
regions’ activation time series.
Hub analysis:
Analysis of the communication
patterns within the network, focusing
on the most connected nodes (hubs).
Connector hub:
A hub that is predominantly
connected to nodes outside of its
own module.
Network Neuroscience
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
/
t
e
d
u
n
e
n
a
r
t
i
c
e
-
p
d
l
f
/
/
/
/
/
7
2
7
1
2
2
1
1
8
3
7
8
n
e
n
_
a
_
0
0
3
1
0
p
d
t
.
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Parkinson’s disease speech production network
Network analyses have revealed changes in integration and segregation associated with PD. In
an MRI resting-state study with drug-naïve PD patients, Luo et al. (2015) found that while network
integration was unchanged, patients showed reduced functional segregation as compared to
healthy controls, which they interpret as the PD brain network being more random and less opti-
mally organized. Reduced functional segregation was also reported by Kim et al. (2017) in a study
on MRI brain network dynamics in PD patients. A longitudinal MEG study in PD patients revealed
that while network integration was not different from controls in de novo patients, it decreased
with disease progression (Olde Dubbelink et al., 2014). Functional segregation, however, was
reduced right from the beginning. Taken together, these studies indicate that resting-state network
integration and segregation are affected differently by PD. However, to our knowledge there are
no studies investigating whole-brain functional connectivity during speech and emotion process-
ing. Furthermore, the analysis of resting-state network hubs revealed that PD affects hub function
and organization (Koshimori et al., 2016). Here, we extend this research to task-based networks.
In clinical practice, cognitive and emotional challenges are often accompanied by a wors-
ening of motor symptoms in PD, best seen in an increase of tremor amplitude in a cognitive or
an emotional task (Raethjen et al., 2008). This emotional–motor interaction demonstrates the
influence of a secondary task on motor performance, which has been documented also in dual-
task situations during gait (Raffegeau et al., 2019). In the present study we challenge the SPN by
a distractor task to assess how speech production is affected by an emotional challenge. Emo-
tion processing is a nonmotor domain that is affected by PD, and an impairment in recognizing
and discriminating emotions based on facial cues has been described (Aiello et al., 2014;
Castner et al., 2007; Iyer, Au, Angwin, Copland, & Dissanayaka, 2019; Wagenbreth et al.,
2019). Therefore, a detailed assessment of the emotional prime is important to evaluate the
conscious recognition of the emotional distractor.
In this study, we investigated the whole-brain functional networks of PD patients and elderly
healthy controls solving a picture-naming task primed with affective face stimuli. The facial
primes were intended to redirect processing resources toward the emotional distraction, without
necessarily inducing an emotion in participants. We analyzed the brain network topology in
terms of its segregation and integration and investigated communication patterns within the net-
work by a hub analysis. Furthermore, we illustrated which regions carry the communicative load
of the network by looking at network pillars, that is, exceptionally well connected hubs (Schill
et al., 2023). We assessed how robust they were against emotional challenge and neurodegen-
eration as present in PD. We answered two main research questions: First, we determined how
the SPN is affected by PD. Second, we assessed how emotional distraction influenced the SPN
and how this influence was affected by PD.
Our hypotheses were mainly based on resting-state network studies. We assumed that resting-
state findings would translate into the SPN: We expected network integration in the SPN of PD
patients to be similar to that of controls, while network segregation should be decreased. Further-
more, a hub analysis was performed to assess differences in the distribution of key nodes of infor-
mation processing between modules (connector hubs) and within a given module (provincial
hubs). Lastly, we explored how the network changed after presentation of an emotional distrac-
tor. We expected the PD patients’ network to be more susceptible to emotional distraction,
given the emotional–motor interference clinically evident in the everyday life of patients.
METHODS
Participants
Nineteen patients with PD and 25 age-matched healthy volunteers were recruited in this study.
The study only included right-handed native German speakers. Participants did not suffer from
Provincial hub:
A hub that is predominantly
connected to nodes within its own
module.
Network Neuroscience
714
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
t
/
e
d
u
n
e
n
a
r
t
i
c
e
-
p
d
l
f
/
/
/
/
/
7
2
7
1
2
2
1
1
8
3
7
8
n
e
n
_
a
_
0
0
3
1
0
p
d
.
t
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Parkinson’s disease speech production network
neurological and/or psychiatric diseases (other than PD in patients) or speech disorders and
did not take psychoactive drugs (medical or recreational). Diagnosis of PD was established
by a neurologist according to the medical history and the MDS-Clinical Diagnose Criteria
for PD (Postuma et al., 2015). Patients were on their usual medication scheme during the study
(levodopa equivalent dose: 539.2 mg ± 319.9 mg). The short form of the Edinburgh Handed-
ness Inventory (Veale, 2014) was used to verify handedness (a score of >60 classified the par-
ticipant as right-handed). To exclude dementia the Montreal Cognitive Assessment (MoCA;
Nasreddine et al., 2005) with a cutoff score of >23 was applied (scores range from 0 Zu 30,
with higher scores indicating better cognitive performance). Beck Depression Inventory II
(Beck, Steer, & Braun, 1996) with a cutoff score of <20 was used to establish the absence
of depressive symptoms (scores range from 0 63, with higher scores indicating more depres-
sive burden).
Seven participants had be excluded analysis for following reasons: motion
artifacts (1), other imaging artifacts misunderstood task structural brain abnormality (1),
depressive disorder and abnormal mean connectivity strength (2), determined by an outlier
analysis using Tukey’s fences (Tukey, 1977). Therefore, 23 healthy volunteers (64.1 ± 6.5 years
of age, 12 females>
