Beyond reduced alpha power: a weaker temporal link between alpha and other brain rhythms in Alzheimer's disease

Background: Reduced alpha power is among the most consistently reported resting electroencephalography (EEG) features of Alzheimer's disease (AD). However, alpha activity also interacts with other rhythms over time, a dynamic aspect that power measures do not capture. Whether AD weakens this temporal interaction, and whether such a change carries information beyond reduced alpha power, remains unclear. Methods: We analyzed public eyes-closed resting EEG from 29 healthy controls (HC), 36 patients with AD and 23 with frontotemporal dementia (FTD). After separating periodic from aperiodic activity, we tracked the aperiodic exponent and periodic theta, alpha, beta and gamma activity in consecutive 1-s windows. A vector autoregressive model predicted these five measures from the preceding window. Alpha's predictive contribution (Gα) was the increase in prediction error for theta, beta and gamma when the model was refitted without alpha as a predictor, with training and testing on separate time blocks of each recording. AD-HC differences were adjusted for age and sex, then for alpha expression (periodic alpha power) and signal quality (dynamic range and spatial reliability), with alpha self-persistence (alpha's carry-over between windows) as a benchmark. We also tested whether expression and Gα each added information about AD beyond the other. An exploratory Jansen-Rit-family circuit model examined whether expression and Gα can change partly independently. Results: Gα was lower in AD than in HC (adjusted difference in 10^3Gα, -3.19; 95% confidence interval -5.25 to -1.13; p = 0.0035), with a consistent direction across sensitivity analyses. Adjustment for alpha expression and signal quality retained 75.8% of this difference but only 4.4% of the self-persistence difference. Expression and Gα each added information beyond the other (Holm-adjusted p = 0.012 for both). FTD contrasts were inconclusive. In the circuit model, reducing local recurrent excitation lowered alpha expression without an estimated decrease in Gα, whereas reducing downstream input lowered Gα with a smaller change in expression. Conclusions: At the group level, the resting alpha abnormality in AD comprises two partly nonredundant dimensions: reduced expression and a weaker temporal link between alpha and subsequent multi-rhythm activity. Quantifying this temporal link may complement conventional spectral markers of AD.