Elevated synchrony in Parkinson disease detected with electroencephalography

Abstract

Objective: Parkinson disease (PD) can be difficult to diagnose and treat. Development of a biomarker for PD would reduce these challenges by providing an objective measure of disease. Emerging theories suggest PD is characterized by excessive synchronization in the beta frequency band (∼20Hz) throughout basal ganglia-thalamocortical loops. Recently we showed with invasive electrocorticography that one robust measure of this synchronization is the coupling of beta phase to broadband gamma amplitude (ie, phase-amplitude coupling [PAC]). Other recent work suggests that high-frequency activity is detectable at the scalp using electroencephalography (EEG). Motivated by these findings, we tested whether beta-gamma PAC over sensorimotor cortex, recorded noninvasively with EEG, differs between PD patients off and on medications, and healthy control subjects.

Methods: Resting EEG was compared from 15 PD patients and 16 healthy control subjects. PD patients were tested on and off medications on different days, in a counterbalanced order. For each data set we calculated PAC and compared results across groups.

Results: PAC was elevated in the patients off medications compared to on medications (p = 0.008) and for patients off medications compared to controls (p = 0.009).

Interpretation: Elevated PAC is detectable using scalp EEG in PD patients off medications compared to on medications, and compared to healthy controls. This suggests that EEG PAC may provide a noninvasive biomarker of the parkinsonian state. This biomarker could be used as a control signal for closed-loop control of deep brain stimulation devices, for adjustment of dopaminergic treatment, and also has the potential to aid in diagnosis.

Figure 1

PAC calculation. The raw signal (A) for one electrode is first filtered (B) at both a lower frequency (for instance beta, left side), and a higher frequency (such as broadband gamma, right side). The lower frequency activity is thought to be important for long distance interactions throughout brain networks, ^, whereas broadband gamma activity is more local and is thought to be a surrogate for neural firing^{, ,} . The phase of the lower frequency activity and the amplitude of the higher frequency activity are then extracted using a Hilbert transform (C). Next the high frequency amplitude is binned according to low frequency phase (D). This provides a way of quantifying the degree to which the broadband gamma amplitude (‘spiking’) is modulated by ongoing low frequency oscillation. This relationship between neuronal spiking and oscillations is probably a fundamental mechanism of neural processing which is abnormal in PD^-. PAC provides a mechanism to study this relationship without unit recordings.

Figure 2

A. Group comodulograms showing median modulation index across all subjects in each group. Prior to taking the median across subjects, comodulograms were averaged over EEG channels that correspond most closely to M1 (C3 and C4). Results are shown for each subject's raw modulation index (MI) values using a CSD reference, but results are similar if MI values are first converted to z-scores or if an average reference is used. B. Boxplots showing MI values averaged over beta (13-30 Hz) phase and broadband gamma (50-150 Hz) amplitude. There is a significant difference between PD patients On and Off medications (two-tailed, paired, Wilcoxon sign rank test, p=0.008) and between controls and PD patients Off medications (two-tailed, unpaired, Wilcoxon rank sum test p = 0.009.) Red asterisks represent significant comparisons (p < 0.01). C. Individual patients’ MI values On and Off medications averaged over beta (13-30 Hz) phase and broadband gamma (50-150 Hz) amplitude. Each circle represents a patient.

Figure 3

Beta and Broadband gamma power for each group. A. Boxplots of average log beta power from 13-30 Hz. There is no significant difference between groups. B. Average log beta power for patients On versus Off medications. C. Boxplots of average log broadband gamma power from 50-150 Hz (excluding harmonics at 60 and 120 Hz). There is no significant difference between groups. D. Average log broadband gamma power for patients On versus Off medications.

Figure 4

Comparison of EEG and EMG Data. Because EMG picked up by EEG electrodes could generate non-neural PAC, we repeated the analysis we performed for sensorimotor EEG, but for the EEG electrodes closest to temporalis muscles (F7 and F8), and for bipolar EMG recorded from the right FDI muscle. A. Group comodulograms identical to Figure 2A except with the Phase Frequency extending to 150 Hz to allow visualization of harmonics at higher phase frequencies. B. Same as Figure 4A except showing the median across subjects of the EEG electrodes closest to temporalis muscles (F7 and F8), calculated with the CSD reference and raw MI values. C. Boxplots of PAC averaged over beta (13-30 Hz) phase and broadband gamma (50-150 Hz) amplitude for the EEG electrodes closest to temporalis muscles. There are no significant differences between groups (p>0.59). D. Individual patients’ MI values for the EEG electrodes closest to temporalis muscles, On and Off medications averaged over beta (13-30 Hz) phase and broadband gamma (50-150 Hz) amplitude. Each circle represents a patient. E. Same as Figure 4A except showing the median bipolar EMG across subjects. Note the PAC extends to the lowest phase frequencies examined and is present at the higher frequency harmonics. F. Boxplots of PAC averaged over beta (13-30 Hz) phase and broadband gamma (50-150 Hz) amplitude for the bipolar EMG signal. There are no significant differences (p>0.09). F. Individual patients’ MI values for the bipolar EMG signal On and Off mediations averaged over beta (13-30 Hz) phase and broadband gamma (50-150 Hz) amplitude. Each circle represents a patient.

Figure 5

Scalp Topography of Comparisons. We calculated PAC for all channels (using a CSD reference) and plotted the scalp topography for both the comparison of patients Off medications compared to patients On medications (A) and patients Off medications compared to controls (B). For the within patient comparison (A), paired tests were performed, and for the between group comparison (B), unpaired tests were performed. The normalized test statistic for each comparison is plotted for each electrode with a significant difference (p < 0.05; Wilcoxon sign-rank for A, Wilcoxon rank-sum for B). Test statistics for all non-significant tests were set to 0. For each comparison the test statistic for each electrode was normalized to the percent of the maximum test statistic for all channels. This normalization was to avoid confusion that might result from the different ranges of the rank sum versus signed rank test statistics.