Background
Alzheimer’s disease (AD) is a progressive neurodegenerative disease and a growing public health concern. At the cognitive level, AD is mainly characterized by memory impairment but it also affects other cognitive domains [
1]. Meanwhile, AD patients show microscopic alterations in their brain, such as amyloid depositions and cell loss, which eventually may lead to macroscopic EEG changes. Cognition results from an optimal combination of local information processing and interregional integration of this information [
2]. This communication can be macroscopically approximated by the measurement of functional connectivity using time series that reflect brain activity. A functional brain network can be constructed by taking all functional connections (i.e., edges of the network) between all regions (i.e., nodes of the network). In these networks, nodes that have a central position within the network and therefore are important to the network structure and integrity, are called hubs. Previous research has shown that the parietal brain region, including the precuneus and posterior cingulate gyrus, is an important hub region in the healthy brain [
3]. In AD, this parietal hub region seems to be particularly affected [
4]. Electroencephalography (EEG) measures electrical brain activity and is used to study functional connectivity and networks in AD. In a group of early-onset AD patients, we observed reduced hub status in the posterior- and occipital brain regions with EEG [
5].
Studies of functional connections have revealed AD-related changes, in which functional connectivity is generally lower in AD [
6,
7], specifically in the higher frequency bands [
8,
9]. On the other hand, network characteristics seem to be altered in AD (e.g., [
9‐
11]). It is however not known how the severity of the disease influences both functional connections and brain networks in AD.
In this EEG study, we studied the hub strength and location and evaluated functional connectivity as a function of disease severity. Furthermore, we subdivided the EEG electrodes into frontal, central and posterior regions. Our hypotheses are three-fold. Firstly, we hypothesize that functional connectivity is reduced in mild stages of the disease and decreases further with increasing disease severity. Secondly, we expect hub strength to decrease in the same areas as the functional connectivity disruptions. Lastly, we expect hub strength to decrease (most likely in regions with decreasing functional connectivity) and therefore, we expect that other regions will become relatively more hub-like (a shifted hub location).
Discussion
In this study on topological patterns of physiological brain activity, we found that a decrease in the functional connectivity in the posterior brain regions was associated with increasing disease severity in the lower alpha band. In addition, the locations of the hubs in the functional networks of AD patients were located towards anterior brain regions compared to the hubs of the control networks, with a significant shift to a more anterior location in the more severely affected patients. The relative node importance of the frontally and centrally located brain areas, as quantified with the BC of the MST, increased with disease severity in AD.
Lower functional connectivity in AD has previously been reported in studies using different modalities, but the pattern of this functional connectivity and the methodology varied considerably between studies. In studies with high temporal resolution time series (EEG and magnetoencephalography (MEG)), functional connectivity in AD was found to be decreased in the higher frequency bands (alpha and beta) [
9,
21‐
26] as well as the lower frequency bands (theta) [
22,
23] and involving mainly brain regions that are connected by long (corticocortical) fibers [
24,
27]. In addition, the location of the largest decrease in patients compared to controls varied over the studies: main differences between groups were found in the anterior and central regions [
18,
28,
29] as well as the posterior regions [
21] and in one study, both regional increases and decreases were found [
30]. Functional connectivity increases of slower oscillations (theta band) were also reported [
8,
31]. In studies with lower temporal but high spatial resolution (functional magnetic resonance imaging (fMRI) and positron emission tomography (PET)), a similar pattern has been found with regionally dependent increases as well as decreases in functional connectivity but with a tendency for a general decrease in AD (for a review see: [
32]). These results indicate that the interpretation of functional connectivity changes in AD is, at least to some extent, dependent on the method used for the analysis. In addition, it can be conceived that during the course of the disease the functional connectivity is fluctuating, with a possible initial increase [
33] and a later decrease. Therefore, differences in inclusion criteria of AD patients across studies could partly account for the differences in the results [
4]. We included patient groups of different disease severity and studied the gradual effect of the AD severity on functional connectivity. Our results indicate that AD severity correlates with a functional connectivity decrease in the posterior brain areas in the lower alpha band. These results give rise to the hypothesis that loss of functional networks might be more valuable than increasing amyloid burden, which is supposed to have plateau’d at the stage of dementia [
34].
The posterior brain areas are main hub regions, and are known to be involved in AD [
35]. In healthy subjects, the posterior brain areas, including the precuneus and the posterior cingulate gyrus, contain hubs with many functional connections to other brain areas [
36‐
38] and are important for intellectual performance [
39]. Also, hubs seem electrically more active, as shown in an EEG simulation study [
4,
40]. Meanwhile, these hubs are more likely to be abnormal in a brain disorder like AD [
41]. Previously, the amyloid depositions were found to have a predilection for high activity brain areas [
42]. In addition, glucose metabolism in AD showed reduced activity in the cortex of the posterior cingulate gyrus [
43,
44] and precuneus [
45,
46]. We reported a shifted hub region, from posterior in controls, to more central regions in AD patients. However, since EEG has a low spatial resolution, any assumptions about regional effects should be made with caution. The functional meaning of the relocation of hubs to more anterior regions (i.e., EEG sensors) might have it’s origin in the heterochronicity of the pathophysiological processes in AD. This means that the pathological pattern is different in patients early in the disease as compared to later stages of the disease. This causes that the gradually degrading posterior region with rising disease pathology in this region to eventually be incapable of effectively conducting electrical activity.
Patients were diagnosed with AD based on clinical criteria using a standardized diagnostic protocol and international criteria [
1,
15]. The control subjects presented at the clinic with subjective memory complaints and can therefore not strictly be considered healthy. However, this group is clinically relevant since they represent daily practice in the memory clinic. The choice for the functional connectivity measure influences the results. In this study, we used the PLI. This measure might be biased towards long distance connectivity, because all zero-lag (mostly short distance) connections are discarded. However, the main advantage of this approach is the reduction of bias due to volume conduction and activity from common sources [
16]. Thus, the PLI may be an underestimation of functional connectivity and therefore, in our study of a large cohort, would show an underestimation of the real disease-effect.
The number of epochs used for analyses may influence the PLI results. In our analyses, we used 4 epochs of 8.192 s (4096 samples). We found that 4 epochs give as reliable PLI values as 5,6,7 or 8 epochs (see Additional file
1).
When comparing networks, several choices have to be made to handle networks of different sizes (number of nodes) and connection strengths. These choices influence the results of the network analysis and are arbitrary [
47]. The MST has the advantage of giving a unique representation of a connectivity matrix since no arbitrary choices have to be made. It is the minimal connected sub-network consisting of the strongest connections without forming cycles. Therefore, the MST can be considered as a backbone of the network that likely includes most of the important connections in the network [
28,
48].
Different centrality indices result in different values for the same graph. We choose the betweenness centrality as a measure for centrality in brain networks. It has previously been proposed to be robust to measure centrality of nodes in networks [
49]. Another often-applied centrality measure is the node degree that indicates the number of connection of a node in the network. Although the number of shortest paths through a node and the number of connections of that node are likely to be related, node degree is not sensitive to so-called connector hubs [
50]. Connector hubs are thought to connect high degree hubs to each other and therefore have a relatively low degree but at the same time include many shortest paths (e.g., a high betweenness centrality).
Competing interests
Dr Scheltens has received grant support (for the institution) from GE Healthcare, Danone Research, Piramal and MERCK. In the past 2 years he has received consultancy/speaker fees (paid to the institution) from Lilly, GE Healthcare, Novartis, Forum, Sanofi, Nutricia.
Research programs of dr van der Flier have been funded by ZonMW, NWO, EU-FP7, Alzheimer Nederland, CardioVascular Onderzoek Nederland, stichting Dioraphte, Gieskes-Strijbis fonds, Boehringer Ingelheim, Piramal Neuroimaging, Roche BV, Janssen Stellar. All funding is paid to her institution.
The remaining authors have no conflicts of interest to report. The medical ethical committee of the VU University Medical Center approved the study.
Authors’ contributions
ME made substantial contributions to acquisition of data and analysis and drafted the manuscript. CS participated in the design of the study and revised the manuscript for important intellectual content and made substantial contributions to the interpretation of the data. WF participated in the design and coordination of the study and revised the manuscript for important intellectual content. PS participated in the design of the study and revised the manuscript for important intellectual content. HW participated in the design of the study and revised the manuscript for important intellectual content. ES participated in the design of the study and revised the manuscript for important intellectual content and made substantial contributions to the interpretation of the data. All authors read and approved the final manuscript.