A systematic review of MEG‐based studies in Parkinson's disease: The motor system and beyond

Abstract Parkinson's disease (PD) is accompanied by functional changes throughout the brain, including changes in the electromagnetic activity recorded with magnetoencephalography (MEG). An integrated overview of these changes, its relationship with clinical symptoms, and the influence of treatment is currently missing. Therefore, we systematically reviewed the MEG studies that have examined oscillatory activity and functional connectivity in the PD‐affected brain. The available articles could be separated into motor network‐focused and whole‐brain focused studies. Motor network studies revealed PD‐related changes in beta band (13–30 Hz) neurophysiological activity within and between several of its components, although it remains elusive to what extent these changes underlie clinical motor symptoms. In whole‐brain studies PD‐related oscillatory slowing and decrease in functional connectivity correlated with cognitive decline and less strongly with other markers of disease progression. Both approaches offer a different perspective on PD‐specific disease mechanisms and could therefore complement each other. Combining the merits of both approaches will improve the setup and interpretation of future studies, which is essential for a better understanding of the disease process itself and the pathophysiological mechanisms underlying specific PD symptoms, as well as for the potential to use MEG in clinical care.


| INTRODUCTION
Parkinson's disease (PD) is the second most common neurodegenerative disease after Alzheimer's disease, with a global disease burden of more than five million people (Olanow, Stern, & Sethi, 2009). The neuropathological hallmark of PD is the deposition of Lewy bodies, of which alpha synuclein is the main constituent. Nigrostriatal dopaminergic neurons are notoriously affected, and loss of these neurons leads to prominent motor features that can be treated symptomatically using levodopa suppletion and deep brain stimulation (DBS). In early disease stages, the alpha synuclein depositions mainly affect the brainstem and the surviving neurons of the nigrostriatal dopamine system, and extend to widespread cortical brain regions in later disease stages (Braak et al., 2003). PD is therefore increasingly recognized as a whole-brain disease with functional disturbances at both subcortical and cortical levels, and is characterized clinically by both motor and nonmotor symptoms.
The past two decades have seen rapid developments in functional imaging techniques aimed at the detection, characterization and localisation of brain activity. These techniques have yielded important insights into the neuronal mechanisms that may underlie PD and its broad range of clinical symptoms. One such technique is magnetoencephalography (MEG), which noninvasively records the weak magnetic fields that are induced by electrical activity in the cerebral cortex (Cohen, 1968(Cohen, , 1972 and subcortical structures (Boon, Hillebrand, Dubbelink, Stam, & Berendse, 2017;Jha et al., 2017). MEG's high temporal resolution can be used to study neuronal activity as well as functional interactions between distinct brain regions in great detail (Baillet, 2017).
Using MEG, PD-related neurophysiological characteristics have been studied both within the motor system and for the brain as a whole.
MEG analyses aimed at motor networks are spatially restricted to the motor cortex and are usually performed in source-space. They can be combined with neurophysiological signals of different origin, such as muscle activity recorded using electromyography (EMG;Timmermann et al., 2003;Volkmann et al., 1996) or local field potentials (LFPs) from the subthalamic nucleus (STN) recorded during DBS (Hirschmann et al., 2011;Litvak et al., 2011)). The study of whole-brain networks using MEG generally involves resting state recordings. Roughly three different approaches have been used in the analysis of whole-brain networks: the analysis of oscillatory brain dynamics using measures of band-limited power or peak frequency, investigation of functional (or directed/effective (Friston, 2011)) connectivity (FC) between brain areas, and assessment of the topological organization of brain networks.
MEG studies increasingly use source reconstruction techniques, such as beamforming, to project the extracranially recorded (sensorlevel) signals to source-space. In sensor-level analysis, several factors that may lead to erroneous estimates of functional connectivity should be considered. Multiple sensors pick up the signal from a single source because of volume conduction (the transmission of electromagnetic fields from a primary current source through biological tissue) and field spread (multiple sensors picking up activity of a common source). In addition, the same sensor picks up signals of multiple sources due to signal mixing. Moreover, the neuronal generators are generally not located directly underneath the sensor with the maximum power (particularly for axial gradiometers). The source-level approach can resolve some of these ambiguities and enables interpretation of the functional results in an anatomical context (Baillet, Mosher, & Leahy, 2001;Brookes et al., 2007;Hillebrand, Barnes, Bosboom, Berendse, & Stam, 2012;Hillebrand, Singh, Holliday, Furlong, & Barnes, 2005;Schoffelen & Gross, 2009). So far, review articles tend to treat motor network-focused studies (Burciu & Vaillancourt, 2018;Magrinelli et al., 2016) and wholebrain studies  separately. Although some efforts have been made to relate findings from motor networks to nonmotor symptoms (Oswal, Brown, & Litvak, 2013b), it is unknown to what extent findings from motor networks and whole-brain networks can be compared and if so, which similarities and discrepancies are present. A full understanding of the neurophysiological changes associated with PD is a stepping-stone toward the development of biomarkers and novel therapies that are urgently needed. Therefore, we set out to systematically review the MEG literature on PD not only to provide an overview of the neurophysiological characteristics of PD, their relationship with clinical symptoms, the effect of disease progression, and the influence of treatment on these characteristics, but also to explore how the results of motor network studies and wholebrain approaches can be integrated.

| METHODS
We performed this systematic review of the MEG literature in PD in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Liberati et al., 2009). We carried out web-based searches using medical databases: PubMed, Embase, Web of Science, Emcare, Academic Search Premier, and ScienceDirect. We used combinations of the key-words MEG and PD. The full search strategies can be found in Supplement A. References up to October 15, 2018 (date of latest search) were used for further study. Two researchers (LIB and VJG) independently screened all articles on title and abstract using the following inclusion criteria: original research article, published in English or Dutch, including a separate cohort of a minimum of five PD patients, and quantification of at least one MEG-parameter.
Although the underlying sources of MEG and EEG are the same, these techniques measure different components of the generated electromagnetic fields (resulting in different sensitivity profiles (Goldenholz et al., 2009)). In addition, MEG is more suitable for source-space analysis than EEG (Baillet, 2017), as it typically uses a higher number of sensors and is less affected by the details of the volume conductor. Even though neurophysiological information obtained using both techniques might be complimentary, a direct comparison would be challenging. We have therefore chosen to limit this review to MEG studies in PD (see Geraedts et al. (2018) for a recent review of quantitative electroencephalography (EEG) studies in PD (Geraedts et al., 2018)). Studies in which data analysis was confined to evoked fields were excluded, but studies aimed at induced/event-related MEG activity were included. Induced/event-related activity differs from evoked fields by not being phase-locked to a certain stimulus (David, Kilner, & Friston, 2006). Cohen's kappa for inter-rater agreement was calculated during this selection process. In case of disagreement, relevant sections were reread until agreement was reached.
Next, both reviewers evaluated the full-text of all included articles using the Joanna Briggs Institute (JBI) checklist for case series, extended with an item addressing clear reporting of MEG data acquisition and analysis (see Supporting Information). Articles had to score a minimum of five points (indicating a sufficient quality study) to be included in this review, of which at least one point was scored on the first three items, at least two points on item 4-8, and one point on item 11. In this descriptive review, we chose to include a much-cited article (Timmermann et al., 2003) that did not fulfill the latter (item 11) more stringent criteria on conducting and reporting the MEG research. Nonetheless, the importance of clear reporting of MEG data acquisition and analysis procedures is obvious (Gross et al., 2013). We subdivided the included articles into two main groups according to the brain network the analysis was focused on: motor network-focused, in which we treated the tremor network as a sub-category, and whole-brain network focused. Since a series of articles on the neurophysiological basis of neuronal entrainment in PD (Te Woerd, Oostenveld, Bloem, De Lange, & Praamstra, 2015;Te Woerd, Oostenveld, De Lange, & Praamstra, 2014;Te Woerd, Oostenveld, De Lange, & Praamstra, 2017;Te Woerd, Oostenveld, de Lange, & Praamstra, 2018), as well as four other articles (Anninos, Adamopoulos, Kotini, & Tsagas, 2016;Boesveldt, Stam, Knol, Verbunt, & Berendse, 2009;Gomez et al., 2011;Suntrup et al., 2013) tended to stand alone from the rest of this review, these will not be discussed in the results section, but the main findings are provided in Table 1.     (Continues) Neurophysiological measures relevant for this review; explanation of the measures can be found in Table 3.
Note included for title and abstract screening, leading to 79 articles meeting the pre-specified in-and exclusion-criteria (Kappa = 0.832). These articles were selected for full-text analysis, risk of bias assessment was performed, and data extraction took place. Three articles were excluded based on the JBI checklist (see Supporting Information) and 26 articles were excluded based on the inclusion/exclusion criteria.

| Motor network-focused research
A summary of the data extraction and risk of bias assessment of the motor network-focused articles can be found in Table 1 and a schematic overview to place the main findings in an anatomical context are provided in Figure 2. Unless stated otherwise, motor network-focused studies in this review have been performed in source-space.

| Early disease stages
Larger sensori-motor cortical (S1/M1) beta band power has been reported both in early-stage PD patients on dopamine replacement therapy (DRT) and in medication naïve patients as compared to controls, recorded during the resting state (Pollok et al., 2012). In this study, during isometric contraction of the contralateral forearm, beta band power was suppressed in controls, but not in PD patients. Only during isometric contraction, contralateral beta band power correlated with Unified Parkinson's Disease Rating Scale (UPDRS)-III scores in PD patients (Pollok et al., 2012). Hall and coworkers found larger resting-state beta band power in the motor cortex contralateral to the most affected hemibody in DRT-naïve patients. The benzodiazepine zolpidem, known for its modulating effects on PD motor symptoms, normalized the ratio in resting-state beta band power between the "affected" and "nonaffected" motor cortex and this correlated positively with improvement in UPDRS-III scores (Hall et al., 2014).
Cortico-muscular coherence (CMC) has been studied by correlating M1 activity with EMG signals recorded in the forearm. CMC was not different between PD patients and controls during steady-state contraction of the forearm (Pollok et al., 2012).

| Later disease stages
Studies in later-stage PD patients found that beta band power in cortical motor regions was lower during the resting state compared to con- Flowchart for inclusion of studies Luoma et al., 2018). However, no correlation with motor improvement has been observed. In addition, during a motor task, as well as during eyes-closed, no differences between ON and OFF stimulation were found.
Even in the absence of stimulation, MEG data are contaminated by high-amplitude, low frequency artifacts mainly originating from the influence of cardiovascular pulsations and breathing on the percutaneous extension wire (before implantation of a stimulator; Litvak et al., 2010), and the stimulator itself . Upon stimulation, electromagnetic artifacts generated by the stimulator, such as jump artifacts and ringing artifacts, obscure neuronal activity (see  for a detailed description of DBS-artifacts). However, MEG recordings are still technically feasible as DBS artifacts can be When studying induced MEG activity, prior to movement onset, in healthy individuals a desynchronization in cortical motor oscillations (beta band) occurs, that disappears during the actual execution of the movement: Event-related desynchronization (ERD). This is followed by a postmovement beta band rebound: Event-related synchroniza- (in color) Overview of main findings in motor network-focused research. A schematic representation of a coronal view of the brain, combined with the forearm muscle extensor digitorum communis. All displayed findings involve undirected functional connectivity, depicted using lines with double arrow heads. A: motor cortex; B: subthalamic nucleus; C: forearm muscle; D: temporal cortex. Red and blue represent higher respectively lower values found in PD patients compared with controls; Black lines represent no significant difference between PD patients and controls, or no comparison with a control group. References: (Hall et al., 2014;Heinrichs-Graham, Kurz, et al., 2014;Hirschmann et al., 2011;Litvak et al., 2011;Litvak et al., 2012;Oswal, Beudel, et al., 2016;Pollok et al., 2012;Pollok et al., 2013;Salenius et al., 2002;van Wijk et al., 2016;Vardy et al., 2011). (b)Overview of main findings in tremor network-focused research. A schematic representation of a coronal view of the brain, combined with the forearm muscle extensor digitorum communis. All displayed findings involve coherence at tremor frequency and its (sub)harmonics. A: sensorimotor and premotor cortex; B: cingulate motor area; C: thalamus; D: subthalamic Nucleus; E: cerebellum; F: forearm muscle. Not depicted in this figure: Posterior parietal cortex. References: Pollok et al., 2009;Timmermann et al., 2003;Volkmann et al., 1996) Brown, 2002 and this correlated with higher akinesia and rigidity subscores . This difference normalized after DRT in one study (Salenius et al., 2002), but not in another . It was speculated by Hirschmann and colleagues that this differential response to DRT was caused by the fact that tremor-dominant PD patients were not excluded from the study from Salenius and colleagues Salenius et al., 2002), as CMC increases might be a characteristic of tremor alleviation (Park et al., 2009). Transcranial alternating current stimulation (tACS) of the motor cortex at beta frequency (20 Hz), but not at 10 Hz, further attenuated both the beta band CMC during isometric contraction and reduced performance (amplitude variability) on a finger tapping task in PD patients, but not in controls (Krause et al., 2013). In a sensor-space study on the effect of DBS on motor CMC, results varied and the correlation with improvement in motor function inconsistent (Airaksinen et al., 2015).
By combining LFP recordings with MEG recordings in STN-DBS patients, a frequency-dependent coherence has been demonstrated between signals from the STN and the ipsilateral S1/M1 cortex in the beta and gamma band during the resting state (Hirschmann et al., 2011;Litvak et al., 2011;Litvak et al., 2012;Oswal, Beudel, et al., 2016;van Wijk et al., 2016). Beta coherence was most dominant in the high beta band (van Wijk et al., 2016), which was mainly located in the mesial premotor regions (Hirschmann et al., 2011;Litvak et al., 2011;Oswal, Beudel, et al., 2016). Resting-state M1-STN beta band coherence was inversely correlated  or not correlated with bradykinesia/rigidity UPDRS-III scores (DRT ON and OFF; Litvak et al., 2011. DRT increased beta band coherence between the STN and a small region in the prefrontal cortex in one study (Litvak et al., 2011), but in other studies DRT suppressed  or did not modulate (van Wijk et al., 2016) beta band coherence between the motor cortex and the STN. In one study, stimulation of the STN suppressed resting-state high-beta band coupling of the STN with mesial cortical motor regions, yet the degree of suppression did not correlate with motor improvement .

| Tremor network-focused research
Tremor most likely involves neuronal mechanisms different from those underlying bradykinesia and rigidity, as the latter symptoms worsen at the same rate as gait and balance impairments, whereas tremor does not (Louis et al., 1999). MEG studies aimed at revealing PD-related tremor networks have identified a number of brain regions with oscillatory activity that is coherent with forearm EMG signals at tremor frequency. First, a motor network contralateral to the 3-6 Hz Parkinson resting tremor has been identified involving the diencephalic level (likely corresponding to the thalamus), the lateral premotor cortex, S1 and M1 (Volkmann et al., 1996). Thereafter, cortico-cortical coherence analysis with contralateral M1 as a seed region (i.e., in which signals from the selected brain region are used to calculate correlations with the rest of the brain) revealed harmonic involvement (at single and double frequency) of the ipsilateral cerebellum, contralateral cingulate motor area (CMA) and contralateral posterior parietal cortex (PPC; Pollok et al., 2009;Timmermann et al., 2003). Over the years, several interesting additional observations have been made: (a) using MEG in combination with LFP recordings in DBS-patients, a muscular-STN-M1 coupling was found during tremor . (b) when controls were asked to imitate a tremor, an oscillatory network could be identified that is comparable to the PD-tremor network observed in the dopamine-OFF state (Pollok et al., 2009). (c) beta band power in cortical motor regions was lower during simultaneous measurement of an intermittent tremor Makela, Hari, Karhu, Salmelin, & Teravainen, 1993).

| Whole-brain focused research
A summary of the data extraction and risk of bias assessment of the whole-brain focused articles can be found in Table 2 and a schematic overview of the main findings is provided in Figure 3. Unless stated otherwise, the whole-brain focused studies have been performed in sensor-space.

| Spectral power
The mean frequency of cortical oscillations in PD patients decreases over the course of the disease. In a study involving PD patients at the earliest (drug-naïve) disease stage, oscillatory slowing was already present, most pronounced over the posterior brain regions (Stoffers et al., 2007). When more advanced PD patients were studied, oscilla-  (Continues)

| Functional connectivity
In sensor-space studies, local FC can be estimated by averaging FC values for all possible pairs of sensors within a given region of interest (ROI), whereas between-ROI FC can be estimated by averaging FC for all possible pairs of sensors between ROIs. In a sensor-space study, recently diagnosed (drug-naïve) PD patients showed an overall higher local and between-ROI alpha1 FC compared to controls (measured using synchronization likelihood (SL; (Stam & Van Dijk, 2002; an FC measure that captures both linear and nonlinear interactions). When moderately-advanced PD patients were compared with controls, higher local functional connectivity (SL in one study and the phase lag index (PLI; less sensitive to volume conduction) in another) was found in PD patients, involving the theta, alpha1, alpha2, and beta band (Cao et al., 2018;. Motor symptom severity and disease duration were positively associated with higher local and between-ROI SL-values . Furthermore, in one study DRT further increased between-ROI beta band FC, as well as local FC in the range of 4-30 Hz in association with clinical motor improvement (especially over centroparietal brain regions; . These findings are in contradiction with the findings of Cao and colleagues, who found the higher alpha PLI in PD patients to normalize upon DRT administration, in correlation with UPDRS-III improvement (Cao et al., 2018). This discrepancy could perhaps be explained by a differential response to DRT observed by Stoffers and coworkers: in the majority of patients, already elevated levels of resting-state local FC (4-30 Hz) further increased, but in patients with a strong improvement in motor function local beta band FC decreased . It was speculated that the differential response to DRT points at differences in the susceptibility to the development of response fluctuations and/or dyskinesias.
Longitudinal follow-up of PD patients using the PLI in sourcespace (the average PLI from a ROI with all other ROIs) revealed a A strong decrease in overall task-related cortical activation was found in all PD patients, most prominent in dysfagic patients.
In nondysfagic patients a compensatory activation toward lateral motor, premotor and parietal cortices seems to take placed upon swallowing, whereas the supplementary motor area was markedly reduced in activity.
Wiesman et al.  (Lachaux et al., 1999); comparable to PLI but sensitive to volume conduction/field spread) and directed Phase Transfer Entropy (dPTE; (Lobier, Siebenhuhner, Palva, & Palva, 2014)), a measure of directed connectivity. The PLV study demonstrated that during a working memory task, PD patients had significantly lower alpha band (9-16 Hz) PLV within the left-hemispheric fronto-temporal circuitry compared to controls, which correlated negatively with verbal working memory performance (Wiesman et al., 2016). The dPTE has been used to reveal lower beta band directed connectivity from posterior cortical brain regions toward frontal and subcortical brain regions in PD versus controls. In this study, lower directed connectivity from posterior cortical regions with the rest of the brain correlated with poor global cognitive performance in PD patients (Boon et al., 2017).
Comparison of a cohort of PDD patients with nondemented PD patients using two different processing pipelines led to conflicting outcomes that could at least partly be explained by differences in methodology Ponsen et al., 2012): in the first study, analysis was based on (ten) clusters of extracranial sensors and SL was used as FC measure. Compared to PD, PDD was characterized by lower fronto-temporal SL in lower frequency bands (delta, theta and alpha1), and higher left-sided parieto-occipital SL in the higher frequency bands (alpha2 and beta; ). In the second (source-level) analysis, FC was calculated using PLI. In the PDD group, PLI between pairs of regions was generally lower for the delta, alpha and beta band, and higher in the theta band. In the gamma band, differences went both ways (Ponsen et al., 2012).

| Topological organization
Olde  characterized the topological organization of PD brain networks in source-space using graph analysis techniques. In early-stage PD patients, lower local integration with preserved global efficiency of the whole-brain network has been observed in the delta band. A longitudinal analysis demonstrated a tendency toward a more random brain topology, in which both local integration (multiple frequency bands) and global efficiency

Oscillatory behaviour
Band power Average spectral power in a particular frequency band.
Mean frequency Average frequency of the spectrum within a given frequency range.

Complexity
Lempel-Ziv complexity Related to the number of distinct patterns and the rate of their occurrence along a given sequence. A high value indicates a high variation of the binary signal (Lempel & Ziv, 1976).

Functional connectivity Coherence
The degree of similarity of frequency components of two time series. Field spread and volume conduction, as well as power, influence the estimate. High values indicate strong functional connectivity (White & Boashash, 1990).
Phase lag index Instantaneous phases of two time series are compared at each time point and the asymmetry of the distribution of the phase differences between these time series is quantified. A high value indicates that there is a consistent nonzero (modulus π) phase relation between the two time series, indicative of functional coupling (Stam, Nolte, & Daffertshofer, 2007). Relatively insensitive to the effects of field spread and volume conduction.
Phase locking value Reflects the consistency of the phase covariance between two signals in a frequency range over time (phase-locking). Field spread/volume conduction affect the estimate (Lachaux, Rodriguez, Martinerie, & Varela, 1999).

Synchronization likelihood
The strength of synchronization between two time series based on state-space embedding. High values indicate strong functional connectivity, but field spread/volume conduction affects the estimate (Stam & Van Dijk, 2002).

Directed functional connectivity
Directed phase transfer entropy Based on the Wiener-Granger Causality principle, namely that a source signal has a causal influence on a target signal if knowing the past of both signals improves the ability to predict the target's future compared with knowing only the target's past: dPTE was implemented as a ratio between "incoming" and "outgoing" information flow .

Granger causality
Quantifies whether the past of one time series contains information that helps to predict the future of another signal. Does not capture nonlinear effects and requires construction of a model of the data (Granger, 1969).
Partial directed coherence Based on the notion of Granger causality. Frequency-domain approach to describe the (direction of) relationships between time series. Decomposes the relationships into "feedforward" and "feedback" aspects (Baccala & Sameshima, 2001).
(alpha2 band) were affected. Worsening global cognition was associated with more random topology in the theta band, and motor dysfunction was associated with lower alpha2 global efficiency. In contrast to the more conventional application of graph analysis techniques, minimum spanning tree (MST) analysis is free of threshold and normalization biases. MST analysis revealed a progressive decentralization of the network configuration, starting in the early-stage, untreated patients, which correlated with deteriorating motor function and cognitive performance (Olde Dubbelink, .
FIGURE 3 Legend on next page.

| DISCUSSION
In this review of the MEG literature on PD, we provide an overview of the neurophysiological characteristics of PD, their relationship with clinical motor and nonmotor symptoms, the effect of disease progression, and the influence of treatment on these characteristics. The design of the studies included in this review is very diverse, regarding both the MEG-recordings itself (e.g., task-based vs. resting-state, eyes-closed vs. eyes-open, MEG signals alone or in relation to other measures, such as LFPs from the STN) and data analysis (e.g., sourcespace vs. sensor-space, different FC measures). Despite these challenging differences in data analytical approaches, we were able to extract several robust findings.
Motor-network focused studies have uncovered a tremor network involving the motor cortex. In addition, these studies support the notion that, in contrast with the pathophysiology of bradykinesia and rigidity, not only basal-ganglia-cortical motor circuits, but also cerebello-thalamo-cortical circuits are important for PD-related tremor (for further reading see (Helmich, 2018)). Another robust finding is the presence of functional loops between the STN and the temporal lobe (alpha band) and the STN and the sensorimotor cortex (beta and gamma band), although the clinical relevance and the effect of DRT on these loops remain to be established. Furthermore, as illustrated in Figures 2 and 3, the neurophysiological characteristics of the PD brain may vary over the course of the disease. For motor networkfocused studies this could be exemplified by increased cortical motor beta band power early in the disease and decreased cortical motor beta band power later in the disease. Whole-brain studies showed a gradual slowing of the power spectrum and an initial increase in functional connectivity, which decreased over time in relation to disease progression, especially cognitive decline. Posterior cortical dysfunction seems to play a crucial role here (Boon et al., 2017;Stoffers et al., 2007). Treatments such as DRT and rivastigmine generally normalized disrupted neurophysiological characteristics in both research fields, although many discrepancies exist, for example the increase in cortical motor beta power upon DRT (Heinrichs-Graham, Kurz, et al., 2014), versus the decrease observed upon DBS (Abbasi et al., 2018;Luoma et al., 2018), or the differential effect of DRT on whole-brain functional connectivity (Cao et al., 2018;. Potential explanations for these discrepancies include methodological differences and differences in the underlying neurophysiological characteristics between PD patients (Figures 2 and 3).
When comparing the MEG findings discussed in this review with the EEG studies recently reviewed by Geraedts and colleagues (Geraedts et al., 2018), there is a prominent agreement on the link between spectral slowing and cognitive decline. Lower peak frequency and higher delta/theta power were the best predictors for future conversion to PDD in longitudinal EEG studies (Caviness et al., 2015;Klassen et al., 2011;Latreille et al., 2016) and in an MEG study a lower beta band power was the best predictor (Olde Dubbelink, . The effect of DRT on whole-brain power was inconclusive for both EEG (e.g., (Mostile et al., 2015) and MEG studies  The results section of this review reflects the clear distinction between motor network-focused MEG research and whole-brain MEG research. Although this distinction often leaves little room for direct comparisons, both fields do share common grounds and we will further explore these in the next two sections.

| Motor network-focused research from a whole-brain point of view
Beta band hypersynchrony within the STN and the basal ganglia-thalamo-cortical, cortico-cortical and cerebro-muscular loops is a wellestablished electrophysiological phenomenon in PD, not only in the MEG field (Brown, 2003;Hammond, Bergman, & Brown, 2007;Kühn, FIGURE 3 (in color) Overview of main findings in whole brain network-focused research: Band power. Schematic representation of observed statistical differences in relative band power between groups. Both sensor-space and source-space analyses are included in the figure. In case of sensor-space analysis, the brain region underlying the relevant sensor was colored. In case of source-space analysis results for each ROI are displayed as a color-coded map on a parcellated template brain viewed from, in clockwise order, the left, right, and top. An area is colored red when the mean power early PD > controls, late PD > early PD, and PDD > PD and blue when the difference was in the opposite direction. The three color codes of magnitudes (from light to dark) illustrate the effect size of the observed difference. Areas that did not show statistically significant differences are represented in white/gray. In the study by (Ponsen et al., 2012) the alpha1 and alpha2 band were combined. PD, Parkinson's disease without dementia; PDD, Parkinson's disease related dementia; L or R, cortical area on the left (L) or right (R) side of the head; C, central; F, frontal; O, occipital; P, parietal; T, temporal. Figure adapted from (Bosboom et al., 2006;Olde Dubbelink et al., 2013a;Ponsen et al., 2012;Stoffers et al., 2007). (b) (in color) Overview of main findings in whole brain network-focused research: Functional connectivity. Schematic representation of observed statistical differences. In case of a sensor-space analysis differences are depicted for local (colored regions) and interregional (arrows) functional connectivity (FC; synchronization likelihood and phase lag index) between groups. In case of a source-space analysis differences in FC from one ROI to the rest of the brain (using phase lag index) are displayed as a color-coded map on a parcellated template brain viewed from, in clockwise order, the left, right, and top. An area is colored red when the FC of early PD > controls, moderately advanced PD > controls, and PDD > PD and blue when the difference was in the opposite direction. Areas that did not show statistically significant differences are represented in white/gray. In the study by Ponsen et al. (2012)  Kupsch, Schneider, & Brown, 2006;Salenius et al., 2002;Stoffers, Bosboom, Deijen, et al., 2008). It has been suggested that the changes in beta band power/connectivity in PD might be a causal mechanism underlying the motor symptoms bradykinesia and rigidity, also considering the indirect evidence that treatment (either DRT or highfrequency DBS) alleviates symptoms and at the same time causes a normalization of local band power and interregional coupling of beta activity (Hammond et al., 2007;Heinrichs-Graham, Kurz, et al., 2014;Levy et al., 2002;Silberstein et al., 2005). However, there is no clear evidence that beta band synchronization directly accounts for the motor deficits in PD. Neurophysiological changes in motor network studies did not correlate with UPDRS-III scores when recorded during the resting state (Abbasi et al., 2018;Litvak et al., 2011;Pollok et al., 2012;Pollok et al., 2013;Vardy et al., 2011). Furthermore, several unexpected negative correlations were observed when late-stage PD patients were recorded during isometric contraction or a motor task of the forearm in the DRT-OFF state Pollok et al., 2013). It has therefore been speculated that excessive beta band power and/or connectivity may not represent a pathological disinhibition with an anti-kinetic effect, but could rather be interpreted as a compensatory mechanism that becomes redundant when DRT is administered Pollok et al., 2013). Hyperconnectivity has also been demonstrated in wholebrain (both source-space and sensor-space) studies in the early stages of PD, most pronounced in the alpha1 band (Olde Dubbelink et al., 2013b;Stoffers, Bosboom, Deijen, et al., 2008). The interpretation of hyperconnectivity in early disease stages is not trivial and the discussion on this matter takes place in a broader context than that of PD only (de Haan, Mott, van Straaten, Scheltens, & Stam, 2012;Hillary & Grafman, 2017). Both pathological disinhibition and compensatory mechanisms may lead to higher FC values, but only a compensatory mechanism would be a purposeful reaction to a pathological process.
However, it is unlikely that the latter mechanism is the sole explanation, since the majority of the studies in the present review did not show a positive correlation between higher FC and better clinical performance (Litvak et al., 2011;Pollok et al., 2012;Pollok et al., 2013;Stoffers, Bosboom, Deijen, et al., 2008;Vardy et al., 2011).
The functional subdivision between low and high-beta frequencies might be of value in unraveling the relationship between interregional coupling of beta activity and clinical functioning. Whereas dopaminergic treatment mainly affected low-beta spectral power in the STN, STN-cortical coherence was strongest in the high-beta band frequencies and was not modulated by levodopa (Litvak et al., 2011;van Wijk et al., 2016). Perhaps more complex functional interactions, such as cross-frequency coupling (see also, ), could play a role in the pathophysiology of PD motor symptoms.
Cross-frequency coupling was previously found within the STN (van Wijk et al., 2016) and within the motor cortex ((de Hemptinne et al., 2013), but see also (Cole et al., 2017)) but not between these two structures.
Alternatively, negative correlations such as between M1-STN beta band synchrony and UPDRS-III scores could merely reflect normal physiology, in which case one would expect healthy individuals to show stronger M1-STN coherence than PD patients . Obviously, it is not possible to perform invasive recordings of brain activity in controls to confirm this, but a case study in an obsessive-compulsive disorder patient, treated with STN-DBS, confirmed the presence of a high STN-motor cortical connectivity in the beta band (Wojtecki et al., 2017). Furthermore, advances in source reconstruction techniques, such as beamforming, increasingly allow the study of subcortical regions by means of MEG (Boon et al., 2017;Hillebrand, Nissen, et al., 2016;Jha et al., 2017). At this point, however, additional methodological and experimental studies are necessary to evaluate the ability of beamformer techniques to reliably distinguish between individual subcortical brain regions.
Another important consideration is that the local neurophysiological processes observed in the motor network take place in a brain that is both structurally (Braak et al., 2003) and Along the same line, the higher beta band functional connectivity between cortical motor regions (Heinrichs-Graham, Kurz, et al., 2014) should be considered against the background of global increases in beta band cortico-cortical FC that have been observed both using EEG and MEG in moderately advanced PD patients, and which correlated with both bradykinesia sub scores and disease duration (Silberstein et al., 2005;Stoffers, Bosboom, Deijen, et al., 2008). In contrast, in early disease stages larger beta band power has been observed in cortical motor regions in both PD patients and animal models of PD (Brazhnik et al., 2012;Degos, Deniau, Chavez, & Maurice, 2008;Hall et al., 2014;Javor-Duray et al., 2015;Pollok et al., 2012), yet this has not been mirrored by the results of whole-brain studies (Olde Dubbelink et al., 2013a;Stoffers et al., 2007).
Variability in ongoing brain activity contributes to the way the brain responds to certain sensory stimuli and therefore might indirectly influence differences in event-related/induced motor responses between controls and PD patients (Sadaghiani, Hesselmann, Friston, & Kleinschmidt, 2010). Furthermore, whole-brain band power changes are known to confound estimates of coherence between two neurophysiological signals and can thereby influence findings in motor network MEG studies (Schoffelen & Gross, 2009). In studies that estimated motor CMC, beta band power in cortical motor regions (and possibly also global beta band power) also differed between PD patients and controls and could therefore have impacted the CMC findings Salenius et al., 2002). In addition, the occipital dominant alpha band rhythm, mainly present when the eyes are closed, may dilute differences observed in the motor network studies (Luoma et al., 2018).

The interpretation of cortico-subcortical interactions in DBS
patients is hampered by the fact that these patients are generally in an advanced stage of disease and therefore have often received high doses of DRT for several years. Chronic DRT is known to influence cortical oscillations via neuronal plasticity (Degos et al., 2008). Furthermore, a longitudinal evaluation of the effect of STN-DBS on beta band oscillations within the STN, coherence with cortical regions, and cortical oscillations along the disease course has not been performed yet. Therefore, when studying cortico-subcortical coherence, the effects of the underlying disease, chronic use of medication and DBS itself on whole-brain cortical oscillations should be taken into account.

| Whole-brain research: Toward a more focused approach
In whole-brain MEG studies in PD, global oscillatory slowing, widespread changes in the strength of functional connectivity within and between brain areas, and a disruption of functional brain network organization have been observed. The consistent relationship between these findings and cognitive decline, motor dysfunction and disease duration support the notion that these whole-brain neurophysiological changes may represent a general marker of the disease processes underlying PD (Bosboom et al., 2006;Olde Dubbelink et al., 2013a; Olde Dubbelink, Stoffers et al., 2007), a conclusion that is further supported by the results of EEG studies (Fonseca, Tedrus, Letro, & Bossoni, 2009;He et al., 2017;Morita, Kamei, Serizawa, & Mizutani, 2009). However, the mechanisms that lead to these widespread neurophysiological changes remain unknown, as well as the way in which these neurophysiological changes induce the clinical symptoms of PD, particularly the nonmotor symptoms.
Observations that highlight the importance of cortico-subcortical interactions in PD include the influence of STN-DBS on whole-brain oscillations (Airaksinen et al., 2012;Cao et al., 2015;Cao et al., 2017), brain studies could build on these observations by including estimation of cortico-subcortical interactions using source reconstruction techniques, and correlate findings to both motor and nonmotor symptoms.
The neurophysiological changes observed in whole-brain restingstate studies correlated with both motor and nonmotor symptoms of PD (Bosboom et al., 2006;Olde Dubbelink et al., 2013a;Olde Dubbelink et al., 2013b;Stoffers, Bosboom, Deijen, et al., 2008), hence the interpretation of these changes might be more ambiguous than the observations in task-related conditions. On the other hand, whole-brain resting-state neurophysiological changes might be a more accurate marker of the underlying disease process. A reliable (noninvasive) in vivo marker of the disease process can be used to predict the disease course in individual patients and to monitor the effects of modulatory techniques such as DBS or future disease-modifying drugs.
The approach of focusing on average FC from a ROI with all other regions in a whole-brain analysis might be too diffuse to pick up changes restricted to certain sub systems. When trying to bridge the gap between the underlying disease and specific PD-related symptoms-referred to as pathophysiology in this context-a more focused approach would be preferable. A seed-based analysis could be used to confirm hypotheses that have arisen based on whole-brain research. In addition, particular symptoms such as cognitive dysfunction in specific domains may be correlated to changes in (dynamic) connectivity between specific subnetworks (Kucyi, Hove, Esterman, Hutchison, & Valera, 2016;Park, Friston, Pae, Park, & Razi, 2017). A more focused approach can provide important additional information on the pathophysiology of specific disease-related symptoms, which may prove useful for the development of symptomatic treatments, for example, targeting key brain regions or subnetworks using TMS or DBS. These exciting therapeutic possibilities are already being tested in PD patients (Freund et al., 2009;Manenti et al., 2016).

| Clinical utility of MEG in PD
Of the robust findings we have presented in this review, up to now only MEG-derived spectral markers (markers of spectral slowing) as predictors for conversion to PDD have potential for routine clinical use (Olde Dubbelink, . As these in-vivo biomarkers of disease progression can also be derived from cheaper and more widely available EEG recordings (Geraedts et al., 2018), the need to include MEG in standard clinical care is currently low. However, with MEG, patients would benefit from a more comfortable and faster recording technique. In addition, when the higher spatial resolution of MEG over EEG is exploited, application of MEG in routine clinical care could become more rational (see (Hillebrand, Gaetz, Furlong, Gouw, & Stam, 2018) for further reading on the clinical application of MEG).
Future studies are required to establish whether measures of functional connectivity or brain network structure, which could be determined more reliably using MEG, can surpass spectral slowing as an in-vivo biomarker of cognitive decline or disease progression in a broader sense.
The optimization of stimulation settings after DBS-placement could also benefit from MEG-recordings, both for nonmotor and motor effects. Potentially, beta band power in the sensorimotor cortex could serve as a biomarker for optimal motor effects, although the link between cortical beta oscillations and motor function is not clear yet (Abbasi et al., 2018;Luoma et al., 2018). Alternatively, a more dispersed cortical fingerprint could serve as a biomarker for optimal clinical (both motor and nonmotor) effects.

| Conclusion
Macro-scale neurophysiological changes in the PD brain have classically been studied from two different perspectives. Some research groups have studied PD-related changes in the brain as a whole, while others have explored relationships between more localized brain activity and motor symptoms, thereby focusing on pathophysiological mechanisms. However, the two research fields are certainly not mutually exclusive and the knowledge gained from both approaches may even be complementary: motor network function is influenced by whole-brain changes in neuronal activity related to the ongoing disease processes, whereas whole-brain analysis may not fully capture local pathophysiological mechanisms underlying specific symptoms.
Up to now, results of MEG studies have been very diverse and the application of MEG in standard clinical care is limited. Future studies that combine the merits of both approaches could increase reproducibility and interpretation of results, which will undoubtedly lead to valuable insights into the neuronal mechanisms underlying PD as well as into the pathophysiology of the broad range of clinical symptoms that characterize this disease.

ACKNOWLEDGMENTS
The authors thank J.