Resting-state electroencephalographic biomarkers of Alzheimer's disease

Abstract

Objective: We evaluated the value of resting-state EEG source biomarkers to characterize mild cognitive impairment (MCI) subjects with an Alzheimer's disease (AD)-like cerebrospinal fluid (CSF) profile and to track neurodegeneration throughout the AD continuum. We further applied a resting-state functional MRI (fMRI)-driven model of source reconstruction and tested its advantage in terms of AD diagnostic accuracy.

Methods: Thirty-nine consecutive patients with AD dementia (ADD), 86 amnestic MCI, and 33 healthy subjects enter the EEG study. All ADD subjects, 37 out of 86 MCI patients and a distinct group of 53 healthy controls further entered the fMRI study. MCI subjects were divided according to the CSF phosphorylated tau/β amyloid-42 ratio (MCIpos: ≥ 0.13, MCIneg: < 0.13). Using Exact low-resolution brain electromagnetic tomography (eLORETA), EEG lobar current densities were estimated at fixed frequencies and analyzed. To combine the two imaging techniques, networks mostly affected by AD pathology were identified using Independent Component Analysis applied to fMRI data of ADD subjects. Current density EEG analysis within ICA-based networks at selected frequency bands was performed. Afterwards, graph analysis was applied to EEG and fMRI data at ICA-based network level.

Results: ADD patients showed a widespread slowing of spectral density. At a lobar level, MCIpos subjects showed a widespread higher theta density than MCIneg and healthy subjects; a lower beta2 density than healthy subjects was also found in parietal and occipital lobes. Evaluating EEG sources within the ICA-based networks, alpha2 band distinguished MCIpos from MCIneg, ADD and healthy subjects with good accuracy. Graph analysis on EEG data showed an alteration of connectome configuration at theta frequency in ADD and MCIpos patients and a progressive disruption of connectivity at alpha2 frequency throughout the AD continuum.

Conclusions: Theta frequency is the earliest and most sensitive EEG marker of AD pathology. Furthermore, EEG/fMRI integration highlighted the role of alpha2 band as potential neurodegeneration biomarker.

Keywords: AD biomarkers; Alzheimer’s disease; EEG; Graph analysis; MRI; eLORETA.

Fig. 1

RS-fMRI and EEG analysis pipeline. Preprocessing of RS-fMRI data (A) and RS-EEG data (B) were applied for patient groups and healthy controls. To delineate EEG patterns within the AD continuum (AIM 1), current source density analysis was computed on EEG data (B.1) identifying the most informative frequency bands. Subsequently, fMRI-driven analysis was performed on RS-EEG data (AIM 2). Firstly, independent component analysis was performed on fMRI data (A.1) in order to select the most informative functional networks. Subsequently, network current source analysis was applied on RS-EEG data within the selected networks and frequencies (B.2). Furthermore, RS-fMRI and EEG graph theoretical analyses were performed only within the fMRI networks (A.2) and frequency bands (only for EEG analysis) (B.2). AIM 3: EEG Diagnostic performance driven by fMRI was tested (B.3). Abbreviations: BOLD = blood-oxygen-level-dependent; RS-EEG = resting-state electroencephalogram; RS-fMRI = resting-state functional magnetic resonance imaging.

Fig. 2

Lobar (A) and network (B) current source density analysis on EEG data. A) Mean values of lobar current density at the different frequencies are reported for MCIpos (A), MCIneg (B), ADD (C) and controls (D). Significant comparisons reported in the boxes are referred to age-, sex- and education-adjusted ANOVA models of rank transformed values, followed by post-hoc pairwise comparisons (Bonferroni-corrected for multiple comparisons, p value < 0.05). Error bars are shown. Lengthened boxes mark the two selected frequencies for the subsequent EEG analysis. B) Percentage of voxels within each selected RS-fMRI network. Only comparisons showing significant differences in at least one network are reported per each freuency band (p value < 0.05). P value refers to age-, sex- and education-adjusted ANOVA models (FDR-corrected, p < 0.05) of rank transformed values. Abbreviations: ADD = Alzheimer’s disease dementia; DMN = default mode network; MCIneg = Mild cognitive impairment with pTau/Aß42 < 0.13; MCIpos = Mild cognitive impairment with pTau/Aß42 ≥ 0.13; PVN = primary visual network; RFP = right frontal-parietal network; VISASS = visual-associative network.

Fig. 3

Independent Component Analysis (ICA) showing decreased functional connectivity in Alzheimer’s disease dementia patients compared with controls. First line shows the selected fMRI ICA-based networks (highlighted in light blue). Second and third lines report three dimensional rendered brains illustrating altered patterns of functional connectivity in Alzheimer’s disease dementia (ADD) patients compared with controls (p < 0.05 FDR-corrected) within networks: a) default mode network (left posterior cingulate cortex, right midcingulate cortex, left angular cortex and middle temporal gyrus and precuneus bilaterally), b) primary visual network (lingual and calcarine cortex bilaterally, right occipital and temporal middle gyrus and left fusiform gyrus), c) right frontal-parietal network (right middle occipital gyrus and cuneus, so as in right inferior parietal gyrus and angular gyrus) and d) visual associative network (calcarine cortex bilaterally, right lingual gyrus and bilateral middle and inferior occipital gyri and bilateral fusiform gyri). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 4

RS-fMRI and EEG graph analysis properties within fMRI networks. Violin plots of clustering coefficient and path length of selected networks (the two most significant metrics) are shown for patient groups and healthy controls. Both graph metrics were calculated based on the Linear Lagged Connectivity for alpha2 and theta frequency bands from RS-EEG data (on the top of the figure) and based on Pearson’s Correlation Coefficient from RS-fMRI data (on the bottom of the figure). Boxplots are reported within violin plots. The horizontal lines in each box plot represents, from the bottom to the top, the 25th percentile, the median and 75th percentile. Whiskers represent the minimum and maximum values. All the dots outside the confidence interval are considered as outliers. Significant comparisons are reported (p values < 0.05). P values refer to age, sex and education-adjusted ANOVA models of rank transformed values, followed by post-hoc pairwise comparisons (Bonferroni-corrected for multiple comparisons). ●: ADD vs Controls; *: ADD vs MCIneg; ■: MCIpos vs Controls; ▲: MCIpos vs MCIneg. Abbreviations: ADD = Alzheimer’s disease dementia, DMN = default mode network, MCI = mild cognitive impairment (pos = pTau/Aß42 ≥ 0.13, neg = pTau/Aß42 < 0.13), PVN = primary visual network, RFP = right frontal-parietal network, VISASS = visual-associative network.