Replication and validation of two novel magnetoencephalography functional connectivity measures in Alzheimer’s disease

Two novel magnetoencephalography (MEG) functional connectivity measures were recently shown to be superior to well-known measures of amplitude correlation and phase synchronization in detecting early stage neurophysiological abnormalities, particularly neuronal hyperexcitability, in a computational model of Alzheimer's disease (AD). The reliability of these measures in empirical data remains to be evaluated. We investigated whether the Phase Lag Time (PLT) and Joint Permutation Entropy (JPE) could identify consistent patterns of between- and within-group functional connectivity in the theta (4-8 Hz), alpha (8-13 Hz), and beta (13-30 Hz) frequency bands in eyes-closed resting-state MEG recordings from two independent cohorts of AD patients (<i>n</i> = 28/<i>n</i> = 29) and control subjects (<i>n</i> = 29/<i>n</i> = 27). We assessed the classification performance of the measures and examined their construct validity using a computational approach. Results were compared with those obtained using two previously validated functional connectivity measures, the leakage-corrected Amplitude Envelope Correlation (AEC-c) and Phase Lag Index (PLI). In both cohorts, whole-brain analysis identified significant group differences (AD patients versus control subjects) in functional connectivity estimated by PLT and JPE across the theta and beta bands (effect sizes: 0.15-0.40; <i>p</i> < .05). Regional analysis revealed that 43-77 regions (out of 80) showed significant group differences in these frequency bands, with significant cross-cohort correlations between regional difference scores of <i>r_p</i> = 0.37-0.55, <i>p</i> < .001. PLT and JPE connectivity matrices revealed strong within-group consistency across cohorts for both AD patients and control subjects in all frequency bands (<i>r_s</i> > 0.80, <i>p</i> < .001). Logistic regression models trained on whole-brain PLT or JPE values in the theta or beta band achieved area under the curve values between 0.72 and 0.78. Finally, the PLT and JPE were shown to be sensitive to changes in coupling strength in a whole-brain computational model. Across all analyses, the PLT and JPE performed as well as, or better than, the AEC-c and PLI. The PLT and JPE can reliably detect neurophysiological abnormalities related to AD from eyes-closed resting-state MEG recordings. Since accumulating evidence identifies neuronal hyperexcitability as a key pathological process in early AD and a potential therapeutic target, these novel functional connectivity measures could play a valuable role in early disease detection and the evaluation of treatments aimed at restoring the excitation-inhibition balance in AD.