E/I unbalance and aberrant oscillation dynamics predict preclinical Alzheimer’s disease

Abstract

AbstractAlzheimer’s disease (AD) is a major cause of dementia linked with severe decrement of cognitive functions. Animal models and presence of epileptiform activity (subclinical/clinical) suggest that neuronal hyper-excitability could play a role in the continuum of AD. The prevalence of amyloid pathology is thought to impact GABAergic terminals potentially disrupting the excitatory-inhibitory (E/I) synaptic balance and causing network dysfunction and cognitive impairment. However, there is no direct neurophysiological evidence in humans of such mechanism throughout the different stages of the disease, nor reliable markers to predict its earliest manifestations (i.e., subjective cognitive decline, SCD). Here, we use novel metrics for directly testing the hypothesis that alterations in the brain temporal coordination and non-linear dynamics should reflect underlying changes in the E/I balance. To that aim, we measured brain activity with magnetoencephalography (MEG) in a cross-sectional cohort of 343 participants (116 neurotypical controls (NC), 85 SCD, and 142 MCI). We quantified non-linear dynamics as power-law long-range temporal correlations (LRTCs) in brain oscillations, a well-known proxy for brain criticality. We demonstrate that attenuations in LRTCs and changes in their spectral profile reliably dissociate NC, SCD, and MCI individuals and that such disrupted brain dynamics are caused by exacerbated excitation shifting the dynamical regime toward a supercritical phase. Additionally, we show that aberrant non-linear dynamics predict conversion from MCI to AD using data from a longitudinal cohort. Machine learning modeling with k-nearest neighbors classification demonstrated the ability of LRTCs and E/I estimates to discriminate early-stage AD patients with better accuracy than classical structural features, highlighting their potential role as early biomarkers of AD.