Magnetoencephalography-detected Phase-amplitude Coupling in Parkinson Disease


 To characterize Parkinson disease, abnormal phase-amplitude coupling is assessed in the cortico-basal circuit using invasive recordings. Whether the same phenomenon might be found in areas other than the cortico-basal ganglia circuit is unknown. We hypothesized that using magnetoencephalography to assess phase-amplitude coupling in the whole brain can characterize Parkinson disease. We recorded resting-state magnetoencephalographic signals in patients with Parkinson disease and in healthy age- and sex-matched participants. We compared the whole-brain signals from the two groups, evaluating the power spectra of 3 frequency bands (alpha, 8–12 Hz; beta, 13–25 Hz; gamma, 50–100 Hz) and the coupling between the gamma amplitude and the alpha or beta phases. Compared with the healthy participants, the patients with Parkinson disease showed significant beta–gamma phase-amplitude coupling in the sensorimotor, occipital, and temporal cortices. In contrast, the two groups showed no significant difference in their resting-state powers. Further, in a resting state, the beta–gamma phase-amplitude coupling in the sensorimotor cortices correlated significantly with motor symptoms of Parkinson disease (P < 0.05); the beta-band power did not. We thus demonstrated that beta–gamma phase-amplitude coupling in the resting state characterizes Parkinson disease.


Introduction
Dysrhythmia contributes to the motor symptoms of Parkinson disease. In previous studies, patients with Parkinson disease in a resting state were observed to have abnormal Tanaka 4 synchronization in cortico-basal ganglia circuits, including the subthalamic nucleus, globus pallidus internus, and primary motor cortex 1,2 . Beta oscillations in the subthalamic nucleus have been correlated with motor symptoms such as bradykinesia and rigidity 3,4 , although recent evidence has been contradictory 5 . Excessive beta oscillations in the cortico-basal ganglia circuit have been shown to represent pathologic oscillations in Parkinson disease.
The beta oscillation phase measured during a resting state in patients with Parkinson disease has been significantly coupled with the gamma oscillation (>50 Hz) amplitude in the subthalamic nucleus 6,7 and motor cortex 8,9 , termed beta-gamma phaseamplitude coupling (PAC) 10,11 . Cortical beta-gamma PAC has been correlated with Parkinsonian motor symptoms and is attenuated by deep brain stimulation and levodopa 12 .
During movement, PAC attenuated earlier in patients with Parkinson disease than in study participants without movement disorders 9 . Excessive beta oscillations and beta-gamma PAC in the cortico-basal ganglia circuit are therefore potential cortical biomarkers of Parkinsonian motor symptoms 13 .
Previous evaluations of PAC mostly used intracranial electrodes in the cortico-basal ganglia circuit, such as electrodes for deep brain stimulation or subdural electrodes on the sensorimotor cortex to measure electrocorticographic signals. Whether the PAC in the gamma oscillation in patients with Parkinson disease differs from that in healthy study participants (HSPs) of similar ages has not been well examined. Although some studies using electroencephalography revealed exaggerated PAC in Parkinson disease, anatomic differences in PAC have not been compared because of the low spatial resolution of electroencephalography 14,15 . However, recent studies showed that magnetoencephalography (MEG) can evaluate cortical PAC with high reliability 16-18 . We therefore recorded MEG signals to evaluate PAC during a resting state both in patients Tanaka 5 with Parkinson disease and in age-matched HSPs. We hypothesized that beta-gamma PAC estimated from MEG signals characterizes patients with Parkinson disease and thus acts as a biomarker for Parkinsonian motor symptoms.

Results
Participants Between May 2015 and February 2018, 32 patients with Parkinson disease and 54 HSPs were recruited at Osaka University Hospital in Japan. Before data analysis, 9 patients and 17 HSPs were excluded: 2 patients and 5 HSPs because they fell asleep or moved during the resting-state recording; 2 patients and 7 HSPs because of metal artefact contamination; 1 patient and 5 HSPs because of equipment trouble; and 4 patients because of another movement disorder (essential tremor) or motor impairment (limb paresis caused by polio) and a change in diagnosis to multiple system atrophy. The subsequent analyses included the remaining 23 patients with Parkinson disease (11 men; mean age: 65.3 ± 7.9 years; Table 1) and 23 of the remaining 37 HSPs, age-and sexmatched to the patients (11 men; mean age: 62.8 ± 5.7 years). were significantly higher than the shuffled SI values ( Figure 1A; P < 0.05, 2-tailed Welch t-test, false discovery rate corrected). In contrast, the HSPs had significant beta-gamma PACs in a part of the motor cortices ( Figure 1B). The exaggerated beta-gamma PAC during the resting state was noninvasively observed in the patients with Parkinson disease.
The supplementary material details the areas in which significant beta-gamma PAC was observed. No significant alpha-gamma PAC during the resting state was observed in either the patients with Parkinson disease or the HSPs.   Figure 3A). However, the averaged Z scores of the beta-band power in the same area did not correlate with the MDS-UPDRS-III scores for akinesia (r = −0.02, P = 0.943; Figure 3B).    12 , and not to differ between the on and off cycles 14,24 . In our study, beta-band power in the sensorimotor cortex did not significantly explain Parkinsonian motor symptoms, consistent with findings in those previous studies.

Conclusions
Our results demonstrate that beta-gamma PAC measured during the resting state in the sensorimotor and visual cortices is a biomarker for Parkinson disease. As a non-invasive measure, PAC estimated from MEG signals could help to monitor a patient's symptoms and reveal the pathology behind Parkinson disease. Trained neurologists examined all patients with Parkinson disease and ensured that they met the Parkinson's UK Brain Bank criteria 28 . We excluded patients who had essential tremors and limb paralysis because those symptoms could change the PAC 9 . Trained clinicians used the MDS-UPDRS-III to assess motor impairments in the patients 29 . We did not require patients to stop their medications to participate. The total dose of dopaminergic antiparkinsonian drugs was converted to a levodopa equivalent daily dose. 30 HSPs had to meet these inclusion criteria: (1) Japanese ethnicity; (2)    All participants were instructed to remain awake in a resting state in the MEG scanner with eyes closed and without thinking about anything in particular. The instruction to keep eyes closed was designed to avoid artefacts from eye blinking. With participants in this state, continuous MEG data were acquired for more than 240 seconds.

Tanaka 15
Pre-processing of MEG signals We applied the continuously adjusted least-squares method, 31 performed with the MEG Laboratory software (Yokogawa Electric Corporation) using two reference magnetometers, to eliminate environmental (offline) noise. We used Brainstorm 32 with default parameters for MEG data pre-processing and MEG source imaging.
Of the original 160 channels, we excluded 10 in the temporal region for all participants because those channels were easily contaminated by muscle artefacts. MEG data resampling at 1000 Hz and high-pass filtering at 0.5 Hz were performed for baseline correction. In an independent component analysis, we used the Infomax algorithm implemented in Brainstorm to isolate ocular and cardiac artefacts. That algorithm calls the function runica.m from the EEGLAB toolbox 33 . Afterward, we visually inspected the MEG data to detect segments that remained contaminated by muscle artefacts or environmental noise. Contaminated segments were discarded from subsequent analyses.
To eliminate powerline contamination, we applied a band-stop filter at 60 Hz with a width of 1.5 Hz to the clean MEG data. Finally, to increase calculation speed, the MEG data were resampled at 500 Hz. Data with irremediable artefacts, usually caused by dental implants, were excluded from further analysis. We used permutation testing to determine the significance of correlations between MDS-UPDRS-III scores for akinesia and the PAC or power, evaluating whether more or fewer significant correlations were observed than would be expected by chance. The SI values and Z scores were randomly permuted for participants while the MDS-UPDRS-III scores were kept the same. We computed a surrogate correlation coefficient based on those shuffled data and repeated the procedure for 2000 permutations, yielding a null distribution of numbers of significant correlations. Significant correlation coefficients were then defined as those that differed significantly from the null distribution.

Estimation of cortical currents from MEG signals
The number of patients included in the study was calculated in an a priori power analysis in the G*Power program 41 . Our null hypothesis-that the correlation coefficients Tanaka 19 would differ from 0 in the negative direction-was based on previous studies. 12,27 The effect size (ρ), power to be achieved (β), and probability of an alpha error were set to 0.60, 0.80, and 0.05 respectively. We determined the sample size to be 30, with an expected loss of 30%.