Functional Connectome in Parkinson’s Disease and Parkinsonism

Abstract

Purpose of review:

There has been an exponential growth in functional connectomics research in neurodegenerative disorders. This review summarizes the recent findings and limitations of the field in Parkinson’s disease (PD) and atypical parkinsonian syndromes.

Recent findings:

Increasingly more sophisticated methods ranging from seed-based to network and whole-brain dynamic functional connectivity have been used. Results regarding the disruption in the functional connectome vary considerably based on disease severity and phenotypes, and treatment status in PD. Non-motor symptoms of PD also link to the dysfunction in heterogeneous networks. Studies in atypical parkinsonian syndromes are relatively scarce.

Summary:

An important clinical goal of functional connectomics in neurodegenerative disorders is to establish the presence of pathology, track disease progression, predict outcomes, and monitor treatment response. The obstacles of reliability and reproducibility in the field need to be addressed to improve the potential of the functional connectome as a biomarker for these purposes in PD and atypical parkinsonian syndromes.

Introduction

Parkinson’s disease (PD) and atypical parkinsonian syndromes are neurodegenerative disorders characterized by distinct patterns of progressive neuronal loss throughout the brain. Increasing evidence suggests two main converging pathways common to neurodegenerative diseases: 1) cell-to-cell propagation and seeding of misfolded protein aggregates (alpha-synuclein, tau) and 2) synaptic toxicity caused by the accumulation of the soluble forms of these proteins [1]. Both mechanisms eventually lead to widespread “disconnection” between neurons. Thus, PD and atypical parkinsonian syndromes should be considered progressive neural network disorders that lead to autonomic, motor, emotional, and cognitive dysfunction.

In the last two decades, resting-state functional magnetic resonance imaging (rs-fMRI) has been widely used to examine the (re)organization of the functional neural networks (i.e., functional connectome) in neurodegenerative disorders. Rs-fMRI is based on the task-free, spontaneous low frequency (~ 0.01–0.1 Hz) fluctuations in the blood oxygenation level-dependent (BOLD) signal. These fluctuations display a spatial structure comparable to task-related activations [2]. The correlations between the BOLD signal fluctuations (i.e., functional connectivity (FC)) represent functional resting-state networks that are linked to underlying neuronal modulations [3]. The relative ease of rs-fMRI (e.g., shorter scan times, no task requirement) has popularized its use and allowed researchers to investigate the intrinsic connectivity patterns of neural networks underlying the symptoms, disease processes, and treatment effects in neurological disorders.

This review summarizes the recent findings and provides a critical evaluation of the functional connectome studies in PD and atypical parkinsonian syndromes. The aim is to discuss the roots of heterogeneity and inconsistency in the findings and outline future directions to improve the potential of functional connectivity as a meaningful and reliable biomarker for these disorders.

Searches on PubMed were conducted using the terms “functional connectome”, “functional connectivity AND network”, linked to each disease separately with “AND”. All searches were limited to the original research articles using rs-fMRI with humans written in English in the last five years (January 2015- July 2020). Studies focusing only on measures of regional spontaneous neural activity and publications on vascular or drug-induced parkinsonism were not included. Frequently used methods and terms are summarized in Table 1. Figure 1 displays the major network hubs.

Table 1.

Glossary

Resting-state FC	The temporal correlation between the spontaneous fluctuations of the BOLD signal extracted from brain voxels or regions.
Resting-state neural network	A set of brain regions that show coherent spontaneous FC. The composition of the networks varies across studies. Most commonly reported networks and their functions are: Default mode network (DMN): Spontaneous, self-referential cognition Frontoparietal network (FPN): Executive control, a.k.a. central executive network (CEN). Dorsal attention network (DAN): Visuospatial attention Task-positive network (TPN): Anti-correlates with the DMN, composed of FPN and DAN Salience network (SN): Detection of salient events, “switch” between the DMN and FPN Sensorimotor network (SMN): Motor control Others: Visual, auditory, language, basal ganglia, and cerebellar networks
Independent component analysis (ICA)	A method to decompose the BOLD signal into its spatial components (i.e., networks) that are assumed to be statistically independent from each other. Once aggregated group level spatial components are estimated, these components are back-reconstructed to obtain subject-specific spatial components and their time series. ICA is also an efficient way of separating noise from the signal [4].
Graph theory	The study of graphs that are structures used to model the pairwise connections (edges) between elements (nodes). The neural networks are thought of as graphs. The brain regions and the connectivity between them that constitute these networks represent the nodes and edges, respectively [5].
Degree	Number of links (i.e., BOLD signal correlations) of a node.
Node strength	Sum of weights of links (i.e., BOLD signal correlation strengths) of a node. It indicates how strongly one node is connected to the rest of the nodes in the network.
Clustering coefficient	A measure of the extent to which immediate neighbors of a node are also connected to one another.
Local efficiency	A measure of how relevant a node is for the communication among neighbors.
Modularity	A measure of the community structure of a network. Communities are defined as groups of densely interconnected nodes that are sparsely connected with the rest of the network. The modular organization of a network is related to its segregation-integration properties and determines the network dynamics and information flow.
Functional segregation	A network’s ability for specialized processing within clusters of nodes.
Functional integration	A network’s ability to bind information efficiently from distributed regions.
Path	The shortest distance (i.e., minimum number of connections) between a node and every other node in the network.
Global efficiency	Inverse of the average shortest path length between all pairs of nodes in the network. It is a measure of functional integration across the network.
Node betweenness centrality	A measure of how central a node is to the communication among other nodes in the network. It is computed as the fraction of all shortest paths in the network that contain a given node.
Hub	Node with high values of betweenness centrality
“Small-worldness” of a network	The clustering coefficient of a small-world network is greater than that of a random network, whereas its average shortest path length is comparable to that of a random network. This allows efficient communication within the “cliques” of nodes and between remote nodes in the network. The brain is considered to have a small-world organization.
Eigenvector centrality (EC) mapping	The EC attributes a value to each voxel in the brain such that a voxel receives a large value, if it is strongly correlated with many other nodes that are themselves central within the network. EC mapping identifies hubs in an unbiased manner [6].
Dynamic causal modeling	It is used to estimate causal coupling (i.e., effective connectivity) between nodes in a network and how that coupling is influenced by experimental manipulations. Realistic models of interacting nodes are created. Using a Bayesian model comparison procedure, the best model is then identified from the observed fMRI data [7].
Dynamic (i.e., time-varying) FC	Denotes changes of network FC patterns over time. The sliding window method is the most common. The FC of the networks is computed over a time window iteratively by sliding the window by a specific step every time. This computation yields connectivity time courses representing the temporal evolution of the functional connectome. Finally, k-means clustering is used to identify the dynamic brain “states.” The amount of time spent in a particular state (dwell time), number of visits in each state, and number of switches between states are reported [8].
Support vector machine (SVM)	A data-driven machine learning model used in classification and regression of the rs-fMRI data. Based on a set of “training” data, the SVM algorithm builds a model that assigns new data in a “test” set to one class or the other. In the case of regression, SVM finds a regression model based on the training set that can correctly predict novel data (e.g., predicting behavioral performance based on fMRI data).

Figure 1. Resting-state networks.

The hubs of the major resting-state networks are displayed on the Montreal Neurological Institute brain template using the CONN functional connectivity toolbox (https://web.conn-toolbox.org). DMN: 1: Medial prefrontal cortex, 2: Posterior cingulate cortex, 3: Lateral parietal cortex. SN: 1: Rostral prefrontal cortex, 2: Anterior cingulate cortex, 3: Anterior insula, 4: Supramarginal gyrus. FPN: 1: Lateral prefrontal and 2: Posterior parietal cortex. DAN: 1: Frontal eye fields, 2: Intraparietal sulcus. SMN: 1: Lateral and 2: Superior sensorimotor areas. VIS: 1: Primary visual cortex, 2: Dorsal and 3: Ventral visual association areas. LAN: 1: Inferior frontal and 2: Superior temporal cortex. CB: 1: Anterior and 2: Posterior cerebellar lobes. Basal ganglia: Cd-caudate, Th-thalamus, Put-putamen, Gp: globus pallidus, Na-nucleus accumbens.

Parkinson’s Disease

Disease Severity

The network topology differs across disease stages and motor severity in PD. A meta-analysis focusing on the corticobasal ganglia-thalamocortical networks demonstrated increased FC in the left pre- and postcentral gyri that are parts of the sensorimotor network (SMN) in PD patients compared to controls. Higher right precentral gyrus FC correlated with lower (i.e., better) Unified PD Rating Scale part III (UPDRS-III) motor exam scores [9]. Lower FC between the anterior putamen and midbrain/substantia nigra in PD patients at early Hoehn & Yahr disease stages (H&Y<2) was associated with higher UPDRS-III scores. Steeper decline in this FC also correlated with worsening motor function over one year [10]. The FC changes are not limited to the SMN and basal ganglia (BG) circuits. PD patients at milder stages (H&Y 2) showed increased FC within the default mode network (DMN). Those at more advanced disease stages showed reduced FC between occipito-temporal regions [11]. Similarly, decreased FC was found in nodes mainly belonging to the SMN, DMN, and occipito-temporal regions in PD patients. Reduced local FC properties especially of the SMN nodes correlated with more advanced H&Y stage and higher UPDRS-III scores [12]. However, correlation between increased FC in the SMN and higher UPDRS-III scores has also been reported [13].

Local network efficiency was found to correlate with worse, and global network efficiency with milder disease severity (total UPDRS scores) in PD patients [14]. In a dynamic FC study, PD patients showed reduced expression of and reduced mean dwell time in a brain state characterized by functional segregation of networks. Reduced mean dwell time in this state correlated with worse motor severity [15]. Other studies investigating the segregation (i.e., modularity) properties of the connectome did not replicate these findings. Decreased modularity in cognitive and motor networks has been demonstrated in PD patients, however, these changes did not correlate with motor severity [16, 17].

Interpretation:

The functional connectome exhibits global and local changes as the disease progresses. The SMN, BG, and DMN seem particularly susceptible to these changes. The FC alterations in the SMN and BG tend to correlate with motor severity, however, the direction of these correlations varies across studies.

Motor Subtypes and Symptoms

Several studies focused on the neural correlates of the axial symptoms (gait/postural impairment, freezing of gait (FOG)) with an emphasis on the pedunculopontine nucleus (PPN) involved in arousal, postural control, and locomotion; and attention and executive networks.

PD patients exhibiting the postural instability and gait disorder phenotype showed decreased FC in the SMN and fronto-parietal network (FPN), and increased PPN-motor cortex FC compared to controls [18]. PD patients with falls showed increased FC between bilateral parietal regions compared to those without falls. This FC correlated negatively with gait speed, and balance and attention scores in both groups [19]. PD patients with postural instability and rapid eye movement sleep behavior disorder (RBD) demonstrated decreased FC between the PPN and supplementary motor area, which was associated with poorer postural responses. The same PD cohort and those with only RBD also showed reduced FC between the PPN and anterior cingulate cortex, which correlated with daytime sleepiness suggesting that sleep and postural control are interdependent and are mediated by locomotor and arousal networks linked to the PPN [20•]. PD patients with FOG showed decreased FC in the parietal regions [21, 22], decreased global efficiency in the dorsal attention network (DAN) [23], altered FC between the PPN and occipito-temporal and cerebellar regions [24], and increased FC between the PPN and medial prefrontal regions compared to those without FOG and controls [25]. Deficits in dual-task gait performance were associated with altered FC between the striatal and temporo-parietal regions [26] and between the PPN and medial prefrontal regions [25]. FOG severity and fear of falling were associated with specific FC patterns in the fronto-striatal-limbic circuits in PD-FOG [27].

Interpretation:

These studies suggest that the axial symptoms in PD are multifaceted including motor, emotional, and cognitive components. A dysfunctional network composed of the PPN, BG, limbic regions, and visuospatial attention and executive networks underlies these symptoms, but the level and direction of the dysfunction vary.

Tremor-dominant and akinetic/rigid subtypes of PD also demonstrate different FC profiles. The cerebellar-thalamic circuit has been implicated in the tremor-dominant subtype. The ventral intermediate (Vim) nucleus of the thalamus - a hub in the cerebello-thalamo-cortical circuits involved in tremor generation - showed increased FC with the dentate nucleus, BG, and motor cortices in the PD-tremor group compared to controls and PD-akinetic/rigid group. Increased Vim-primary motor cortex and Vim-dentate nucleus FC correlated positively with tremor scores, but a relationship between the Vim nucleus FC and akinetic/rigid motor scores was not observed [28]. Similarly, the dentate nucleus showed increased FC with other parts of the cerebellum in the PD-tremor group compared to those without tremor and controls, which also correlated positively with tremor scores [29]. Using the global spontaneity of movement score in UPDRS-III as a surrogate for akinesia, a negative correlation between this score and the FC between the posterior medial frontal cortex and left inferior parietal lobule was found. This was interpreted as decoupling in the higher-order motor control and initiation networks in akinesia [30].

Finally, a recent study has examined the “motor reserve”, as opposed to motor deficit, in de novo PD patients. Using the demographic variables and dopamine transporter availability in the posterior putamen, the UPDRS-III scores were predicted for each subject. Motor reserve was defined as the difference between the actual and predicted UPDRS-III scores. The associated network was then identified using the motor reserve estimates in a FC analysis. This network consisted of the BG, insula, hippocampus, amygdala, cerebellum, and inferior frontal gyrus. Stronger FC of the motor reserve network was associated with less increase in the dopaminergic medication need over two years [31•].

Treatment Effects

Dopaminergic Treatment (DT)

The effects of DT especially on the FC of the cortico-striato-thalamic circuits and the (re)organization of the SMN have been reported in a previous review [32]. It has been commonly assumed that DT (“ON” state) “normalizes” the aberrant disease-related FC patterns observed in the off-medication state (“OFF” state). More recent studies have revealed a complex relationship between DT and functional connectome that goes beyond normalization.

A meta-analysis reported convergence on the bilateral inferior parietal lobule in PD compared to controls. This region is important in sensorimotor integration. The right-hemispheric FC measures associated with this region were increased in PD-OFF and decreased in PD-ON compared to controls, demonstrating a differential FC pattern depending on medication state [33]. PD-OFF patients compared to controls showed increased putamen-cerebellum FC, which correlated with better motor performance and interpreted as “compensatory”, and increased putamen-primary motor cortex FC, which correlated with worse motor performance. DT normalized the putamen-cerebellum FC [34]. Similarly, PD-OFF showed higher eigenvector centrality (EC) in the SMN compared to PD-ON. The lower EC particularly of the putamen in the ON state correlated with motor symptom improvement [35]. Yet, other studies revealed conflicting results. PD-ON compared to PD-OFF exhibited higher EC in the cerebellum and brainstem and stronger FC between these regions and other motor regions. Higher cerebellar EC correlated with lower UPDRS-III scores regardless of medication state [36]. Decreased posterior putamen and increased subthalamic nucleus (STN) FC with other BG and motor cortical regions were found in PD-OFF compared to controls. DT normalized this FC pattern with a large effect on the more affected side, which also correlated with improvement in bradykinesia [37].

Dopaminergic modulation of the striatal FC was also implicated in dyskinesia. The primary sensorimotor cortex-putamen FC increased after DT in PD patients without dyskinesia and decreased in those with dyskinesia. Dopaminergic modulation of this FC in the most affected hemisphere also predicted whether an individual patient would develop dyskinesia [38].

Interpretation:

DT response in the motor cortico-striato-thalamic and cerebellar circuits is not uniform. Relatively higher FC in the OFF state was interpreted as “compensatory” and decrease in aberrant FC in response to DT was interpreted as “normalization.” However, the direction of these FC changes and their correlation with motor symptomatology have not been consistent across studies.

Other studies examined the DT-related FC changes with regard to motor phenotypes. In tremor-dominant PD patients, using electromyography recordings during scanning and dynamic causal modeling (DCM), DT was found to reduce tremor onset-related activity in the globus pallidus external part and tremor amplitude-related activity in the Vim nucleus. The DCM analysis showed that dopamine directly increased the self-inhibition of the Vim nucleus. Moreover, the magnitude of this self-inhibition predicted the clinical tremor response to DT [39•]. Another study investigating the disease- and medication-related FC changes identified distinct functional connections between the frontal, motor, and posterior cortical regions that separated PD-ON and PD-OFF, and controls. These connections were associated with different motor subtypes: Bradykinesia/rigidity with putamen FC, tremor with cerebellum and supplementary motor area FC, the latter of which also predicted tremor improvement with DT [40].

The effects of DT have also been investigated at the whole-brain level using graph theory. Altered global and local efficiency in cognitive networks in PD-OFF and DT-related modulations of the local efficiency in the SMN and subcortical regions were observed. These findings were interpreted as reorganization and normalization of disrupted network topology, respectively [41]. The network topology was found to be more strongly integrated (i.e., hyperconnected) in PD-OFF compared to PD-ON. The level of this integration in particular subnetworks also correlated with lower motor severity, and higher premorbid intelligence and whole brain gray matter volume, hence, was interpreted as a compensatory reconfiguration [42].

Interpretation:

The heterogeneity of all these findings suggests that the FC patterns in OFF and ON states in nondrug-naïve PD patients cannot be explained solely on the basis of compensation and normalization, respectively. Instead, the reorganization of the functional connectome seems to be differentially affected by the disease and DT in a nonlinear fashion.

Deep Brain Stimulation (DBS)

In PD patients under STN-DBS stimulation compared to their presurgical state under DT, stimulation was associated with higher EC in the bilateral motor cortex and increased FC in the motor cortical-thalamo-cerebellar network. The clinical improvement in response to DT and STN-DBS was comparable [43]. In PD patients who were on dopaminergic medication during scanning, turning the DBS on compared to off increased the whole brain FC and “normalized” it toward the level of controls. The amount of stimulation of the motor STN correlated with the increased FC specifically in the motor network. Furthermore, the motor thalamo-cortical FC increased, whereas the striatal FC with the cerebellum, STN, and globus pallidus external part decreased [44]. In PD patients who were off dopaminergic medication, turning the DBS on compared to off increased the effective connectivity in the cortico-striatal, thalamo-cortical, and BG direct pathways, and led to the decoupling of the STN FC [45].

Interpretation:

STN-DBS exerts its clinical effect via motor cortico-striatal and thalamo-cortical pathways mainly by increasing their FC, and by altering the STN FC profile.

Non-Motor Symptoms

Cognitive Decline

The DMN is the most frequently studied cognitive network in neurodegenerative disorders. Characteristic alterations in the DMN have been demonstrated consistently in Alzheimer’s disease. The DMN is also an essential network in PD-related cognitive impairment (PD-CI) [46], but the alterations are heterogeneous [47].

Reduced FC in the DMN in patients with PD-CI compared to those with normal cognition (PD-NC) and controls has been demonstrated [46, 47–51]. Weakening (or loss) of the natural anti-correlation between the DMN and task positive network (TPN) [48, 52] and between the DMN and salience network (SN) [53] have also been reported in PD-CI. Better cognitive performance was found to relate to the decoupling between the DMN and SN in controls, but this relationship was lost in PD patients [54]. On the other hand, better visual and verbal memory performance in PD-NC correlated with the FC between specific DMN nodes [55].

Several studies have also reported changes in the functional organization of the TPN and SN in PD-CI patients. These include reduced FC in the DAN [48, 56, 57], in the FPN [46], and between the DAN and FPN [48, 58••]; and reduced degree in the SN [53]. Moreover, FC profiles mainly of the frontal nodes distinguished PD-CI patients from PD-NC [59]. PD patients also exhibited different FC patterns depending on the type of their cognitive deficit. Dysexecutive deficits correlated with reduced FC in the SMN, and posterior cortical deficits (visuospatial skills, memory) correlated with reduced FC in the central executive network (CEN) and increased FC in the temporal network [51].

The role of the substantia innominata (SI) - a cholinergic hub involved in cognitive processes - has also been investigated in PD-related cognitive decline. Increased FC between the SI and parietal areas and decreased FC between the SI and right frontal cortex was found in drug-naïve PD-CI compared to PD-NC. These FC changes correlated with executive and visuospatial cognitive deficits, and were more prominent in PD patients with earlier onset of CI [60]. Increased FC between the SI and fronto-parietal and cerebellar regions was shown in drug-naïve PD patients, most of whom had CI, compared to controls. This pattern was more pronounced in patients with lower than those with higher SI volumes despite comparable cognitive profiles [61].

In addition to its membership in motor networks associated with parkinsonian motor symptoms, the cerebellum has also emerged as a candidate hub in PD-CI. Reduced cerebellar FC has been demonstrated in PD patients with non-amnestic compared to those with amnestic deficits [50]. Reduced cerebellar FC with occipital and frontal regions was also found in PD patients with impaired cognition [62].

More recently, dynamic FC methods have been employed to examine how evolution of network FC may contribute to cognitive deficits. PD-CI group showed reduced mean dwell time in a brain state with weaker FC within and between networks and reduced FC between motor and cognitive control networks compared to controls [63]. In a similar study, PD patients showed increased mean dwell time in a brain state with stronger within-network FC (i.e., more segregated) and visited this state more frequently than controls. These results were mostly driven by the subgroup of PD patients with dementia [64]. On the other hand, in a graph theory-based static analysis, loss of network segregation and higher global network efficiency were associated with worse CI in PD patients [65].

Interpretation:

Altered FC within and between the cognitive networks including the DMN, FPN, DAN, and SN plays a role in PD-CI. The SI and cerebellum have emerged as new cognitive hubs. The dynamic network topology seems to behave differently in PD patients with various levels of cognitive decline, yet the findings do not converge on a pattern.

Mood Disorders

Depression has been a major focus of FC studies of mood disorders in PD. Increased FC within and between the limbic hubs [66–70], and decreased FC mainly in the fronto-temporal regions and between fronto-temporal-limbic regions [66, 67, 71, 72] have been reported in PD patients with depression compared to those without depression. Similar FC alterations were also found to be associated with anxiety in PD [69, 70]. The FC of the right middle frontal gyrus correlated negatively with depression severity in PD patients with depression [72]. Depression scores in PD patients also correlated with the altered FC between the parahippocampus and fronto-temporal regions [70]. Furthermore, altered FC in the DMN [71, 73], increased FC in the SN [72, 73] and FPN [73] have been found in PD patients with compared to those without depression. Reduced FC between predominantly limbic fronto-striatal regions has been reported in PD patients with apathy [69, 74]. Finally, a recent study identified the caudate nucleus as a hub related to impairment in neuropsychiatric and cognitive domains in PD. The altered FC between the caudate and specific dorsal fronto-parietal and posterior cortical regions was associated with unique and shared influences in both domains [75•].

Interpretation:

These results suggest that mood dysregulation in PD is associated with altered FC within and between the networks involved in higher-order cognitive regulation and emotion processing.

Impulsive-Compulsive Behavior (ICB)

PD patients with compared to those without ICB showed increased FC between the mesocorticolimbic regions. Dopamine agonist use did not affect the FC results suggesting a pre-existing functional network organization in the PD-ICB group [76]. In drug-naïve PD patients, increased FC in the SN and decreased FC in the right CEN and DMN were associated with the development of ICB while on dopaminergic medication at three-year follow-up. Increased FC between the SN and left CEN also correlated with the severity of impulsivity [77]. PD-ICB exhibited higher dwell time in a brain state that showed stronger within-network FC involving brain regions that overlap with the SN and mesocorticolimbic hubs compared to those without ICB and controls [78]. PD-ICB compared to those without ICB showed reduced FC in the SMN, and ICB duration correlated positively with the FC between the left FPN and visual network [79]. PD patients with high impulsivity showed increased FC between the right FPN and medial visual network compared to those with low impulsivity [80].

Interpretation:

Altered FC in cognitive networks, specifically, enhanced FC in the SN and reward-related mesocorticolimbic circuits seem to be associated with ICB in PD.

Hallucinations

Increased FC within the DMN and between the DMN and DAN nodes was observed in PD patients with hallucinations compared to those without hallucinations [81]. The hippocampus showed higher FC with the DMN and frontal regions, but lower FC with the visual cortices in PD patients with hallucinations, which also correlated with their visuospatial memory impairment [82].

Interpretation:

Enhanced DMN FC may play a role in PD-related hallucinations.

Fatigue

PD patients with distressing fatigue compared to those without fatigue exhibited reduced FC in the SMN and increased FC in the DMN [83], and reduced FC between the SN and FPN and multiple premotor and parieto-temporal regions [84].

Sleep Disturbances

PD patients with and without subjectively reported nighttime sleep disturbance showed decreased FC between the major DMN, CEN, and DAN hubs and widespread cortical areas compared with controls. This decrease was less prominent in the PD cohort with compared to that without sleep disturbance, perhaps implicating a compensatory mechanism [85]. Reduced FC in the frontal, temporal, insular, and limbic lobes, and cerebellum was observed in PD patients with compared to those without excessive daytime sleepiness [86]. On the other hand, a trend toward increased FC in the DMN regions including the ventromedial prefrontal and middle temporal cortices in PD patients with compared to those without excessive daytime sleepiness was also demonstrated [87].

Interpretation:

Altered FC in major networks is associated with sleep disturbances in PD, but the pattern and direction of these changes are variable.

Prodromal PD and Non-Manifesting LRRK2 Carriers:

In a prodromal PD cohort with hyposmia and RBD compared with controls, intra- and inter-hemispheric FC in the striato-thalamo-pallidal circuit was found to be decreased. These changes did not extend to the cortex [88]. Yet, another study showed decreased FC not only within the BG network, but also between the BG and fronto-temporal cortices in patients with RBD compared with controls [89].

In a study with asymptomatic LRRK2 mutation carriers, reduced left striatal-parietal FC and reduced right substantia nigra-occipital FC was demonstrated compared to non-carriers [90]. In a similar study, asymptomatic LRRK2 carriers demonstrated reduced FC in the DMN, SN, and DAN, but not in the SMN; as well as reduced level of functional integration in the SN and DAN [91•]. Both studies show that the functional connectome, especially involving non-motor regions, is altered in genetically at-risk groups.

Disease Progression Models

Recent studies combined FC measures with other disease-related biomarkers to create prediction models for disease progression. Longitudinal follow-up of nondemented PD patients compared to controls showed greater decline in the SMN FC, which was faster in those with higher baseline alpha-synuclein levels in the cerebrospinal fluid; and decline between the DAN-FPN FC, which correlated with worsening cognition [58••].

Another study tested the hypothesis that neurodegeneration in PD would spread from the initially affected (i.e., atrophied) subcortical structures (“disease reservoir”) to the connected cortex. Indeed, cortical clusters (occipital, sensorimotor, frontal) that showed greater structural connectivity and FC with the subcortical disease reservoir showed greater atrophy in PD patients compared to controls at one-year follow-up [92••].

In an attempt to link the propagation of synucleinopathy to the connectome dysfunction, a simulation model was created based on the alpha-synuclein and glucocerebrosidase gene expression profiles from the Allen Human Brain Atlas, and structural and functional connectome data from young healthy controls. Simulated neuronal loss across the healthy structural connectome correlated with the empirical atrophy obtained from the PD patients. Furthermore, the functional connectome biased the simulated propagation to brain areas with higher co-activity suggesting that both structural connectivity and FC shape the spread of the neurodegenerative process [93••].

Atypical Parkinsonian Syndromes

Multiple System Atrophy (MSA)

MSA is characterized by alpha-synuclein-containing cytoplasmic inclusions in the glial cells causing neurodegeneration in striatonigral and olivopontocerebellar structures. Probable MSA diagnosis requires autonomic failure and poorly levodopa-responsive parkinsonism (MSA-P) or a cerebellar syndrome (MSA-C) [94].

Cerebellum has been the main region of interest in fMRI studies in MSA. In patients with MSA compared to those with PD, the cerebellum showed reduced FC with the striatum and other networks including the FPN, SMN, and SN [95]. Patients with MSA-C showed reduced cerebellar FC with other cerebellar and cortical regions compared to controls [96, 97], and disease severity correlated negatively with cerebellar FC [96]. On the other hand, MSA patients showed increased ponto-cerebellar FC (driven by the MSA-C subgroup) and decreased DMN FC (driven by the MSA-P subgroup) compared to controls. The ponto-cerebellar FC correlated positively with smooth pursuit impairment in the MSA group [98]. The dentate nucleus showed reduced FC with the DMN in MSA compared to PD patients [99]. Both MSA-C and MSA-P patients showed decreased dentate nucleus FC with cortical and subcortical regions compared to controls, but this decrease was more diffuse in the MSA-C subgroup and included the DMN nodes [100]. MSA-P patients exhibited altered dentate nucleus FC with several brain regions compared to PD patients and controls. The effective connectivity from the motor cortex to the dentate nucleus correlated with milder motor severity in the MSA-P group [101].

Furthermore, parts of the cerebellum that were components of the cognitive networks also showed reduced FC in the MSA group with cognitive deficits compared to controls [102]. Similarly, MSA patients with cognitive impairment compared to those with normal cognition showed decreased degree centrality and FC of the right middle fontal gyrus [103].

Interpretation:

Abnormal (predominantly decreased) cerebellar FC has been demonstrated consistently in MSA patients. However, in most studies the cerebellum was used as the sole or main seed biasing the results.

Progressive Supranuclear Palsy (PSP) and Corticobasal Syndrome (CBS)

PSP and CBS are tauopathies characterized by tau inclusions in neurons and glia with astrocytic plaques and neurofibrillary tangles. Differences in the tau fragments, histological lesions, and anatomical distribution of tauopathy distinguish the pathological processes in these conditions and determine their clinical manifestations [104, 105]. The clinical features of probable PSP (i.e., PSP-Richardson Syndrome (RS)) are oculomotor dysfunction and postural instability. Parkinsonism (PSP-P), gait freezing, and frontal dysfunction can also be seen [106]. Probable CBS is characterized by asymmetrical motor symptoms of the limb and higher cortical sensorimotor deficits [107]. Despite these differences, both syndromes may show considerable clinical overlap.

Given its strong involvement in the neurodegenerative process in PSP, the midbrain tegmentum (MT) has been used as a seed in FC studies. Reduced FC in the dorsal MT and several cortical and subcortical regions were found to correlate with motor and cognitive deficits in PSP. Decreased thalamic FC correlated with bulbar and mentation subscores of the PSP Rating Scale. Decreased pallidum FC correlated with lower Mini Mental State Exam scores. Finally, decreased dorsal MT FC correlated with lower Frontal Assessment Battery scores [108]. In a longitudinal study with PSP patients compared to controls, the baseline reduced FC between the rostral MT, BG, and paracingulate nodes shifted to reduced FC between the rostral MT, BG, and posterior mid-cingulate nodes. Longitudinal atrophy followed a similar pathway suggesting that atrophy and decline in FC follow a midbrain-to-posterior cortical gradient in PSP. Furthermore, reduced FC in the prefrontal-paralimbic regions correlated with worsening disease severity [109]. In both PSP-RS and PSP-P phenotypes, FC of the MT network and SMN was increased, and that of the DMN was decreased compared to controls. These changes were more pronounced in the PSP-RS group. Reduced MT FC also correlated with vertical gaze impairment in both PSP groups [110].

CBS patients showed decreased EC and FC in temporo-parietal and insular regions, and increased EC and FC in medial frontal regions and caudate compared to controls [111]. Both PSP and CBS groups showed within- and between-network FC changes involving multiple networks compared to controls. These changes were more pronounced in the CBS group and included widespread cortical motor and cognitive control regions [112]. Thalamus FC with the prefrontal regions, BG, and cerebellum was also found to be decreased in both PSP and CBS patients compared to controls. Yet, the dentate nucleus FC with the BG and prefrontal regions was decreased in the PSP, but it was asymmetrically increased with frontal regions in the CBS patients compared to controls suggesting a role for the dentate nucleus FC in distinguishing the two conditions [113].

Interpretation:

FC alterations in a widespread network anchored in the MT are observed in PSP. The FC changes in CBS are more elusive, but seem to involve cortical motor and cognitive regions.

Finally, in pathologically confirmed PSP and CBS cases, correlations between pathological and in vivo imaging findings were examined. A neurodegeneration score was calculated for each region of interest based on pathological changes. Rostral MT for the PSP and superior frontal sulcus for the CBS group were chosen as seeds in the FC analysis. No significant correlation was found between the mean rostral MT FC and the neurodegeneration scores or tau burden in the PSP group. A significant positive correlation was found between the mean superior frontal sulcus FC and neurodegeneration scores in the CBS group. The pathological measures showed more meaningful correlations with in vivo regional atrophy measures suggesting that seed-based FC is a less reliable predictor of tauopathy than structural imaging measures [114].

Conclusions

The past decade has witnessed an exponential growth in functional connectomics research on brain disorders. More recent methodological developments have aimed to advance our understanding from correlative to causal and predictive neural interactions with the ultimate goal of developing reliable imaging biomarkers to establish the presence of pathology, track progression, and monitor treatment response. Despite these developments, the field of functional connectomics continues to be burdened by reliability and reproducibility problems that lead to heterogeneous results [115]. The literature included in this review is a testament to this heterogeneity in PD and parkinsonism, which makes it difficult to draw clear-cut conclusions about the functional connectome changes associated with these disorders.

There are multiple sources of heterogeneity across studies that can be categorized under clinical and analytical factors. Clinical cohort characteristics vary considerably within and between studies regarding disease phenotypes, disease duration and severity, treatment duration, and medication state during scanning. The complex interactions between these characteristics, as well as between motor and non-motor symptoms are not always taken into account. Analytically, the sample size (PD cohorts:11–134 subjects, median 24; parkinsonism cohorts: 11–62 subjects, median 22), number of rs-fMRI data points (i.e., scan duration: 4.5–32 min, median 8 min), preprocessing methods of the rs-fMRI data (e.g., dealing with physiological artifacts, global signal, and head motion), definition of networks (e.g., atlas-based, data-driven), and methods of computing the connectome metrics (e.g., seed-based, ICA, graph) also vary substantially between studies. Relatively small sample sizes raise concern for the likelihood of inflating the observed positive effects [116]. Seed-based FC studies, though usually hypothesis-driven, tend to introduce selection bias. On the other hand, multivariate approaches such as graph theory provide a network-level assessment, but the neurobiological interpretation of functional graph metrics is not always straightforward. This is partly because FC is about the statistical interdependencies of the BOLD signal rather than the “real” brain connections. Besides the tangible sources of heterogeneity, the functional connectome of a single individual is inherently variable over the timescales of days and months [117], and exhibits topological features that share commonalities with, but are also distinct from the group connectome [118].

These challenges and the urgent need to overcome them are increasingly well-recognized by researchers, and roadmaps to more reliable and reproducible functional connectomics research have been proposed [116, 119]. Yet, all of these factors continue to introduce unique challenges to the use of the functional connectome as a biomarker for classification, prediction, and longitudinal monitoring purposes in neurodegenerative disorders in general. Importantly, acceptable classification accuracy (e.g., successfully distinguishing patients from healthy people or one disease population from the other based on the functional connectome) with high sensitivity and specificity requires well-characterized and large samples. The validity of predictions (e.g., predicting treatment response or behavioral outcome based on the functional connectome) also depends on large samples with several hundred observations and on the independence between training and test sets [120]. The Human Connectome Project provides fMRI data from 1200 healthy participants, but this scale has not been reached for PD and parkinsonism fMRI datasets.

Moving forward, it is imperative to address the sources of heterogeneity and implement steps to improve reliability and reproducibility of functional connectomics in general and its application to PD and parkinsonism in particular. Big data initiatives (e.g., Parkinson’s Progression Markers Initiative, EU Joint Programme - Neurodegenerative Disease Research) are valuable for pooling the data and increasing the statistical power of studies. Techniques to mitigate the between-scanner and between-site differences in these initiatives should also be developed (e.g., [121]). Finally, combining carefully chosen functional connectome features with other disease-related biological measures (e.g., atrophy, proteinopathy) may provide more accurate and comprehensive models for diagnosing the disease and monitoring progression.

Abbreviations

BG

Basal ganglia

BOLD

Blood oxygenation level-dependent

CBS

Corticobasal syndrome

CEN

Central executive network

DAN

Dorsal attention network

DCM

Dynamic causal modeling

DMN

Default mode network

DT

Dopaminergic treatment

EC

Eigenvector centrality

FC

Functional connectivity

fMRI

Functional magnetic resonance imaging

FOG

Freezing of gait

FPN

Fronto-parietal network

H & Y

Hoehn and Yahr

ICA

Independent component analysis

MSA

Multiple system atrophy, C: Cerebellar, P: Parkinsonian

MT

Midbrain tegmentum

PD

Parkinson’s disease

PD-CI

PD with cognitive impairment

PD-NC

PD with normal cognition

PPN

Pedunculopontine nucleus

PSP

Progressive supranuclear palsy

RBD

Rapid eye movement sleep behavior disorder

Rs-fMRI

Resting-state fMRI

SMN

Sensorimotor network

SN

Salience network

STN-DBS

Subthalamic nucleus deep brain stimulation

SVM

Support vector machine

TPN

Task positive network

UPDRS

Unified Parkinson’s Disease Rating Scale