Master internship: research report
Excitation/Inhibition Imbalance in Preclinical and Clinical Alzheimer’s Disease
A Magneto-Encephalography Study
Master Brain and Cognitive Science – University of Amsterdam
Research Project 1
Danique Mulder
Supervisor: M.Sc Anne M. van Nifterick
Examiner: Dr. Alida A. Gouw
Abstract
Objective: The late dementia stage of Alzheimer’s Disease (AD) has been associated with reduced excitatory neuronal activity, whereas preclinical AD may be characterized by hyperactivity. One promising way to capture changes in neuronal activity is by looking at the ratio between excitatory and inhibitory synaptic currents (E/I ratio). In this study, we compared the E/I ratio between preclinical and clinical AD patient groups. Additionally, we aimed to provide a link between E/I ratio and continuous disease stage, by linking E/I ratio to cognitive and pathophysiological disease severity
Methods: Resting-state magnetoencephalography (MEG) recordings were acquired in healthy elderly, and preclinical, translational and demented AD patients. The E/I ratio was computed for the parieto-occipital regions and the hippocampi using three different measures: the Detrended Fluctuation Analysis (DFA), the Fitting Oscillations and One Over F (FOOOF) and the functional E/I (fE/I). All estimates were compared between groups, and correlated to the MMSE and phosphorylated tau (p-tau) concentrations.
Results: We found no changes in the E/I ratio in preclinical AD. However, decreased DFA exponents and increased FOOOF exponents, indicating increased inhibition, were observed in AD dementia. Those measures were also associated with the MMSE. Results from the fE/I were contradictory, with higher estimates in AD dementia, indicating less inhibition, as compared to preclinical AD.
Conclusion: Neuronal excitatory activity seems to be unaltered in the early stages of AD, but decreased in demented AD patients. Shifts in the E/I ratio were associated with cognitive disease severity. With this, we provide direct evidence for excitatory hypoactivity in AD dementia.
Introduction
Alzheimer’s Disease (AD) can be viewed as a continuum, where a long preclinical stage precedes the dementia stage. In this early stage, patients have no cognitive deficits, while their brain already shows changes associated with the disease, including positive biomarkers for amyloid-beta (Aβ) and synaptic dysfunction (1,2). Focusing treatment on these early changes may be beneficial to delay or prevent later neurodegeneration (1,3). Hence, it is essential to understand what processes underlie these changes. In the late dementia stage of AD, the brain is considered to be in a hypoactive state, characterized by decreased excitatory neuronal activity. Studies using electroencephalography (EEG) or magnetoencephalography (MEG) consistently demonstrated oscillatory slowing, characterized by lower peak frequencies and a shift in power from higher to lower frequencies (4,5). Studies using fluorodeoxyglucose positron emission tomography (FDG-PET) found decreased glucose metabolism (6–8). Interestingly, studies with animal models of early-stage AD reported the opposite. Hyperactivity was found for excitatory hippocampal neurons of a young APP23xPS45 double transgenic mice, before the development of Aβ plaques, but in the presence of increased soluble Aβ (9). Additionally, enhancing activity levels of excitatory neurons resulted in increased Aβ production (10) and long-term attenuation of neuronal excitatory activity caused a decrease of Aβ aggregation (11). These results indicate that there may be a vicious circle, where high levels of soluble Aβ cause hyperactivity, and where hyperactivity leads to increased Aβ production. Although evidence for excitatory hyperactivity during the early stage of AD is growing, it has mostly been studied with animal models. Remarkably, findings from computational modelling with neural masses are comparable: the neural masses that are most active and that are highly connected to other areas (so-called hub regions) appeared to be more vulnerable to degeneration. The hub regions exhibited oscillatory slowing and loss of spectral power (12), comparable to hallmarks observed in AD (4,5,13). This is in agreement with observations in AD, where highly active regions, such as the precuneus and posterior cingulate cortex, were found to be affected early in the disease progression (14,15). However, translation of these findings to human
patient groups remains challenging, as a direct non-invasive measure of neuronal hyper/hypoactivity is still lacking.
At present, different methods, relying on different modalities, have been applied to estimate neuronal activity non-invasively, but the findings diverge. MEG and functional Magnetic Resonance Imaging (fMRI) studies have provided evidence for resting-state hyperconnectivity between the precuneus and the inferior parietal lobules (16) and within the medial temporal lobe (17). Additionally, fMRI studies reported increased hippocampal and occipital blood-oxygen-level-dependent (BOLD) responses during memory encoding (18,19), possibly indicating hyperactivity of excitatory neurons. Conversely, there is also MEG evidence for the slowing of alpha rhythm (20) and decreased relative alpha power (21), and evidence from FDG-PET for decreased glucose metabolism (22), which would better fit with decreased levels of excitatory activity. Findings from these indirect methods could reflect either increased or decreased excitatory neuronal activity, but strong evidence for this association is lacking. For example, the strength of functional connectivity is modulated by neuronal oscillatory power, but this is not a one-to-one relationship (23). Similarly, although there are interactions between task-induced activity and resting-state activity, the two methods measure something different (24). Hence, in order to solve the ambiguity in the literature, a direct measure of neuronal hyper- and hypoactivity is essential.
Neuronal activity and the excitation/inhibition (E/I) ratio of local neuronal circuits are tightly linked. This ratio reflects the ratio between synaptic currents from inhibitory interneurons and excitatory pyramidal cells (25). In a healthy system, neuronal circuits show a delicate balance between synaptic excitation and inhibition (25), which is critical for the formation of oscillations (26). Small disruptions in the E/I balance (i.e., more excitatory relative to inhibitory synaptic currents, or vice versa) can disrupt oscillatory activity and, given the importance of oscillations in functional brain networks (23), cause dysfunction at the network level. E/I imbalance has been proposed as underlying pathophysiology of AD (25), where hyperactivity of excitatory neurons initially causes an increase in E/I, followed by a drop in E/I ratio due to hypoactivity. Thus, observing changes in the E/I ratio in AD patients may provide support for a state of hyperactivity in the early preclinical stage and hypoactivity during the late dementia stage of the disease.
Until recently, E/I ratio could only be measured invasively on a restricted spatial scale (27), but three algorithms were developed to extract the E/I balance non-invasively and on a whole-brain level, using electrophysiological M/EEG signals.
i) The detrended fluctuation analysis (DFA) measures critical-state dynamics of oscillatory activity by calculating dynamic long-range temporal autocorrelations (LRTC) on different time lags (28). The autocorrelation is highest when there is no time lag (i.e., the signal is correlated with itself at the same point in time), and it decays when increasing the time lag. The rate of the decay, which is reflected by the DFA exponent, determines the strength of the correlation: a faster decline, thus high DFA exponents, reflect the presence of LRTC. The DFA exponent was linked to the E/I ratio. When adjusting levels of inhibitory and excitatory activity in a neuronal network model, the DFA would change in a predictable way (29): the DFA exponent was highest for an E/I balanced network, and increasing levels of excitation or inhibition led to a decrease in the DFA exponent (Fig. 2a). Hence, DFA informs about E/I balance, but not about directionality (i.e., whether the network is dominated by excitation or inhibition) (Fig. 2a). Previously, a reduced DFA exponent was reported for both schizophrenia (30) and AD (31), indicating that this measure may also be able to capture changes in preclinical AD.
ii) The fitting oscillations and one-over-f algorithm (FOOOF) identifies oscillatory peaks and calculates the slope of the logarithmic power spectrum (32). A power spectrum, obtained by electrophysiological recordings, contains both periodic oscillations and an aperiodic one-over-f (1/f) component. The 1/f component shows an inverse linear relation between log power and log frequency, meaning that there is high power in low frequencies and low power in high frequencies (33). Gao et al. (25) showed that the exponent of the 1/f component was a good estimate of the E/I ratio. In rats, changes in E/I synapse density across hippocampal CA1 layers could be tracked by the 1/f exponent. Additionally, in macaques, a change in the spectral exponent, computed based on electrocorticography recordings, was reported after increasing inhibition through medication. In both experiments, lower exponents (i.e., flatter slopes) correspond to increased excitation, and higher exponents (i.e., steeper slopes) correspond to increased inhibition. Hence, the FOOOF does not provide information on a tipping point between domination by inhibition or excitation (Fig. 2b), but it does inform about relative levels of excitation and inhibition when comparing groups. In humans, a lower spectral exponent (i.e., higher E/I) has been linked to ageing (34), and the spectral slope appeared to be a better predictor of schizophrenia than the power of neuronal oscillations (35).
iii) The functional E/I (fE/I), developed specifically to measure the E/I ratio, combines alpha amplitude and (a derivative of) the DFA exponent (36). In a Critical Oscillation (CROS) in silico model, alpha amplitude correlated positively with the E/I ratio, where a higher amplitude was associated with more excitation (36). Additionally, the DFA exponent informs about the balance or imbalance of E/I. Therefore, a correlation between alpha amplitude and the DFA exponent would be able to inform about both the E/I tipping point as well as the extent of imbalance (Fig. 1). For the resting-state EEG signal of a healthy population, the fE/I estimate indicated a balanced E/I, and administration of GABAergic medication caused the E/I ratio to shift towards inhibition (36). Additionally, differences in fE/I estimates were reported in children with autism spectrum disorder (36), whose E/I ratio was also proposed to be imbalanced (37,38).
The present study used MEG data to compare the E/I ratio, estimated with three different measures (DFA, FOOOF and fE/I), between healthy controls and patients in three different stages of AD: Subjective Cognitive Decline (SCD; considered preclinical AD), Mild Cognitive Impairment due to probable AD (MCI; considered a translational stage) and AD dementia. Additionally, to treat AD as a continuous disease, E/I was linked to the severity of cognitive impairment and the pathophysiological disease severity, using the Mini Mental State Examination (MMSE) score and phosphorylated tau (p-tau) concentrations, respectively. Figures 2a-c display how each measure, in theory, reflects the E/I ratio. We hypothesized that, in line with animal studies on
Figure 1 fE/I. (A) How alpha amplitude and the DFA exponent behave under different E/I ratio. The DFA is highest for a balanced E/I and reduces when E/I is imbalanced. The alpha amplitude is positively correlated to E/I: it is lowest in an inhibition dominated regime and highest in an excitation dominated regime. (B) The correlation of alpha amplitude and nF(t) (i.e., the normalized fluctuation function; a derivative of the DFA exponent), and how they form the fE/I estimate. In an inhibition dominated network there is a positive correlation (left panel); for balanced E/I
neuronal activity, the E/I ratio would shift from a balanced E/I in healthy elderly to an excitation dominated network in preclinical AD, reflecting hyperactivity. Subsequently, E/I was expected to shift towards an inhibition dominated network from preclinical AD to AD dementia, reflecting hypoactivity. Therefore, patients in the MCI stage could exhibit either increased excitation or increased inhibition. This pattern was expected to be the same for group-level and continuous analyses. Figures 2d-f graphically display the specific hypotheses for each E/I measure. Note that the fE/I would be the only measure that was able to draw conclusions on the direction of the network imbalance, as its value is relative to a tipping point.
Methods
Participants. Existing data of 135 subjects from the Amsterdam Dementia Cohort of the Alzheimer Center Amsterdam at the VU University Medical Center (VUmc) (39) were analyzed. All patients underwent extensive dementia screening, consisting of history taking, neurological and neuropsychological examination (incl. MMSE), testing for amyloid and tau concentrations, and eyes-closed resting-state MEG. During a multidisciplinary consensus meeting, patients were diagnosed based on the most recent National Institute of Aging Alzheimer’s Association (NIA-AA criteria) (1). All patients who received the diagnosis SCD (n = 17), MCI (n = 19) or dementia due to Alzheimer’s Disease (AD dementia, n = 52) were included in the study.
Figure 2 The Hypothetical Relationship between E/I ratio and the DFA, FOOOF and fE/I. (A–C) Changes in the DFA, FOOOF and fE/I as a result of underlying changes in E/I ratio (25,29,36). (A) The maximum DFA exponent reflects a balanced E/I. The same values for DFA could correspond to both an inhibition dominated regime and an excitation dominated regime. (B) The spectral exponent (i.e. slope) is negatively correlated with E/I ratio, where a smaller exponent (i.e. flatter slope) is related to increased excitation. The slope contains no information on a tipping-point between an excitation dominated network and an inhibition dominated network. (C) The fE/I is positively correlated with E/I ratio, where a value of 1 reflects a balanced E/I, a value below 1 reflects an inhibition dominated network, and a value above one reflects an excitation dominated network. (D–F) Hypotheses on how the DFA, FOOOF and fE/I are expected to change during AD disease progression. The network is initially expected to shift from an E/I balance towards an excitation dominated network (from HC to SCD), followed by a shift towards an inhibition dominated network (from SCD to MCI and AD dementia). The DFA is hypothesized to be highest in HC and lower in all AD patients groups. Additionally, the DFA exponent is expected to be lower in AD dementia as compared to SCD and MCI. (D). The FOOOF the exponent is hypothesized to initially decrease (from HC to SCD), followed by an increase (from SCD to MCI and AD) (E). The fE/I is hypothesized to have a value of 1 in HC, a value above one in SCD and a value below 1 in MCI and AD.
Patients with subjective cognitive complaints but no objective cognitive impairment, as assessed by neuropsychological testing, and a positive biomarker reflecting Aβ deposition received the SCD diagnosis. In this study, and according to the most recent NIA-IAA research framework (1,2), we consider SCD patients to be in a preclinical stage of AD. MCI and AD patients also required a positive Aβ biomarker in order to be included in the study. The severity of cognitive impairment distinguished between the MCI and AD dementia groups: for MCI patients, cognitive impairment remained limited to one cognitive domain, and the decline did not significantly interfere with daily functioning. For AD dementia patients, cognitive disorders were present in at least two domains, and the impairment interfered with everyday activities (1,2). Subjects from the ADC cohort with subjective cognitive complaints (without objective cognitive impairment) and a negative Aβ biomarker were included as non-demented healthy elderly controls (HC; n = 47). Exclusion criteria were: a medical history of significant neurological (i.e., other than dementia) or psychiatric disorders and use of acetylcholine-esterase inhibitors, antipsychotics, anti-epileptics, lithium or neuropathic pain medication at the time of MEG-measurement. All subjects gave written informed consent for the use of their clinical data for research purposes.
Mini Mental State Examination (MMSE).
The cognitive disease severity was assessed with the MMSE (40). This is a short neuropsychological dementia screening test, assessing the severity of cognitive impairment in five different domains. The score reflects the severity of cognitive impairment, with 30 indicating no cognitive impairment and 0 indicating severe cognitive impairment.
Biomarkers analyses.
Aβ status was determined with either amyloid-PET (Florbetaben, Florbetapir, or PIB; n = 45) or in the cerebrospinal fluid (CSF; n = 124) obtained with a lumbar puncture. Thirty-four subjects underwent both amyloid-PET and a lumbar puncture. For these subjects, the amyloid-PET was leading for the Aβ biomarker status, as concentrations determined based on the CSF contain false negatives more often (41). For the 124 CSF samples, Aβ-42 and p-tau concentrations were determined. Up till December 2017, concentrations were determined with manual Innotest immunoassays. All other samples were automatically analyzed using Roche Elecsys®. Concentrations determined based on Innotest samples were converted to Elecsys concentrations (as described in (42)) to maintain comparability between scores. A CSF sample was considered Aβ positive for Aβ-42 concentrations lower than 1000 pg/ml. P-tau concentrations were used to assess pathophysiological disease severity. This concentration plateaus late in the disease and is positively correlated with brain atrophy (43) and the severity of cognitive impairment (44), which makes it suitable to track disease progression. The p-tau analyses were conducted for the subset of subjects who underwent a lumbar puncture (n = 124).
MEG acquisition and preprocessing.
All subjects underwent two five-minute eyes-closed resting-state MEG recordings, performed in a magnetically shielded room using a 306-channel whole-head system (Elekta Neuromag. Oy, Helsinki, Finland), with a sample frequency of 1250 Hz and online anti-aliasing (410 Hz) and high-pass (0.1 Hz) filters.
Subjects were instructed to stay awake and reduce eye movements during the recording. At the first signs of drowsiness, subjects were alerted by an auditory stimulus or instructed to open their eyes for a few seconds. Only the first five-minute recording was analyzed.
To determine the location of the head relative to the MEG sensors, signals from five head localization coils were continuously recorded. The positions of the head localization coils, and the outline of the participants’ scalp (∼300 points), were digitized using a 3D digitizer (Fastrak; Polhemus). The scalp surface was
co-registered with a head-size based MRI template, for all subjects. The sphere that best fitted the scalp surface served as a volume conductor model for the beamformer analysis. Channels containing excessive artefacts were identified by visual inspection and removed before applying the temporal extension of Signal Space Separation (tSSS) in MaxFilter software. The tSSS filter was used to remove artefacts caused by external sources (45). The data was offline high-pass filtered at 0.5 Hz and low-pass filtered at 100 Hz. An atlas-based centroid beamforming approach (as described in (46)) was applied to source reconstruct the MEG signal. The beamformer sequentially projected sensor level signals to the centroid voxel of the 78 cortical and 12 subcortical regions of interest (ROIs) from the automated anatomical labelling (AAL) atlas (47), by weighting the contribution from each MEG sensor to a voxel’s time series. Weights were computed based on a dipole source model at the voxel location. Lastly, data segments with artefacts (e.g., excessive muscle activity and eye blinks, not discarded by the tSSS and Maxfilter) and drowsiness were identified and removed from the recording. The final length of the time series differs per subject ([137s – 276s]). Spectral changes in AD are generally most pronounced for parieto-occipital regions (13) and hippocampi (4). Also, both parietal regions and the hippocampi are amongst the first to show amyloid deposition and neurodegeneration (14,15) and are therefore expected to show early differences in functionality as well. Therefore, analyses were performed for the parieto-occipital region (24 ROIs1) and the hippocampi (2 ROIs).
Detrended fluctuation analyses of long-range temporal correlations (DFA).
The DFA is a measure of LRTC, commonly used as an index of critical oscillation dynamics (28). Hardstone et al. (28) described the algorithm to compute the DFA in detail. In short: the signal is filtered in a frequency band of interest. As slowing of the alpha rhythm is commonly observed in AD (13), we decided to filter the signal in the extended alpha frequency band, ranging from 6 to 13 Hz. The cumulative sum of the time series was computed to create the signal profile. The signal was split into different length time windows that are equally spaced on a logarithmic scale (in this study ranging from 2 to 30 seconds). For each window length, the signal was split into time windows with 50% overlap. Each time window of each window length was detrended, by subtracting the linear trend (using a least-squares fit). The standard deviation (SD) of the detrended signal was computed, and the SDs for each window length was averaged and plotted on logarithmic scales. The slope of the plotted function is the DFA exponent. The network has positive LRTC for exponents between 0.5 and 1, whereas an exponent of 0.5 characterizes an uncorrelated signal. MATLAB scripts with an implementation of the DFA estimation were provided by K. Linkenkaer-Hansen (28).
Fitting Oscillations and One Over F (FOOOF).
The FOOOF algorithm separates the aperiodic 1/f component from the periodic oscillations, to make a precise estimation of the spectral exponent and the properties of the oscillatory peaks. The algorithm is explained in Haller et al. (32). In summary: The Fast Fourier Transform (FFT) was applied to the time series, and the result was log-log transformed, resulting in a log-spaced power spectrum. Secondly, Gaussians were fitted to each peak in the spectrum, after which they were subtracted from the power spectrum. Afterwards, a regression line was fitted to the remaining 1/f spectral component:
𝑦 = 10!""#$%∗ ( 1 𝑥$&'!($(%)
𝑥 denotes the frequency bin, and 𝑦 indicates the power in this frequency bin. The 1/f slope was fitted in the 3 – 40 Hz frequency range, with a maximum of two Gaussians (i.e., oscillatory peaks) per spectrum (Appendix
1Parietal: superior parietal gyrus, inferior parietal gyrus, supramarginal gyrus, angular gyrus, precuneus, posterior cingulate gyrus. Occipital: superior occipital lobe, middle occipital lobe, inferior occipital lobe, calcarine fissure and surrounding cortex, cuneus and lingual gyrus
A). The exponent was used as an estimate of the E/I ratio. Python scrips with an implementation of the FOOOF algorithm, as provided in Haller et al. (32) were used to compute the spectral exponent.
Estimation of functional excitation/inhibition ratio (fE/I).
This method combines alpha amplitude and (a derivative of) the DFA exponent to estimate the (functional) E/I ratio. See Bruining et al. (36) for a comprehensive explanation of the fE/I algorithm. In short: the signal was filtered in the alpha band. Again, we computed the fE/I over a 6 – 13 Hz frequency range to account for slowing of the alpha rhythm in AD patients (13). Secondly, the Hilbert transform was applied to obtain the amplitude envelope. The envelope was split into time windows of five seconds each (80% overlap) and converted to an amplitude normalized signal profile by dividing each window by the original amplitude. Thirdly, the signal was detrended, and the SD of the fluctuations was calculated, separately for each window. The result is the normalized fluctuation function (nF(t)), which was used as the derivative of the DFA exponent. Finally, the fE/I was calculated by correlating the nF(t) with the alpha amplitude and subtracting this correlation coefficient from one. If this value is equal to one, fE/I is balanced (i.e., no correlation between nF(t) and alpha amplitude). If the value is smaller than one, the system is inhibition dominated (i.e., a positive correlation between nF(t) and alpha amplitude). If the value exceeds one, the system is excitation dominated (i.e., a negative correlation between nF(t) and alpha amplitude) (Fig. 1, 2c). If the DFA exponent is too low, the algorithm will falsely give an fE/I estimate of 1 (the DFA and alpha amplitude do not correlate), even though the network is inhibition or excitation dominated (Fig 1a). To account for this, the fE/I was computed with a DFA cutoff. DFA cutoffs of 0.55 and 0.6 were compared to determine the optimal cutoff (Appendix B). The final analyses were conducted with a DFA cutoff of 0.55. MATLAB scripts with an implementation of the fE/I estimation were provided by K. Linkenkaer-Hansen (36).
Statistical Analysis.
Statistical analyses were performed in IBM SPSS Statistics 26.0. Normality was checked for by visual inspection of Q-Q plots and the Shapiro-Wilk test. The assumption of homoscedasticity was checked using Levene’s test for each ANOVA, and based on visual inspection for correlation analyses. Demographic group differences were analyzed with chi-square for gender and education, and with a one-way ANOVA for age. A one-way ANOVA was carried out for all group-level analyses, with clinical diagnosis serving as the independent variable and either the 1/f spectral exponent, DFA exponent or fE/I estimate serving as the dependent variable. Separate analyses were conducted with age as a covariate, to correct for group differences (see participant characteristics). If the assumption of homoscedasticity was violated, the F statistic was adjusted using Welch’s statistic. In the case of non-normality, the data was randomly resampled using the bootstrap method with 1000 replicates. In the presence of a main effect of group, post hoc analyses were conducted with either a Tukey HSD test (for homogeneous variances) or Games-Howell Multiple
Comparison test (for heterogeneous variances). These tests allow comparing all groups while correcting for the family-wise error rate (48). For testing the association between MMSE/p-tau and the spectral exponent, DFA exponent, and fE/I, Pearson Correlation analyses were conducted. If data were not normally distributed, a Spearman Correlation was used instead. Correlation analyses were conducted without the data for healthy controls (n = 97 for MMSE; n = 80 for p-tau), to explore the association within the AD continuum. Results with a p-value of lower than 0.05 or a confidence interval (CI) not including zero were considered statistically significant.
Results
Participant characteristics.
The main participant characteristics are listed in Table 1. All assumptions were met unless otherwise specified. There were no significant differences in gender distribution (χ2(3) = 2.404, p = .493) and distribution
education level (χ2(15) = 18.110, p = .257). However, there were age differences (F(3, 129) = 8.013, p < .001),
where the HC group was younger than the MCI (p = .001) and AD dementia group (p < .001). To account for this difference, all ANOVAs were repeated with age as a covariate. The MMSE score also differed between groups (F(3, 45.089) = 30.98, p < .001): the AD dementia group scores were significantly lower than those for all other groups (p < .001). For subjects whose Aβ-42 and p-tau concentrations were available: the Aβ-42 and p-tau concentrations also differed between groups, F(3, 119) = 94,58, p < .001 for Aβ-42 and F(3, 36.014) = 16.14, p < .001 for p-tau. The HC group had higher Aβ-42 concentrations than all other groups (p < .001). There were no significant differences in Aβ-42 concentrations between the AD groups. The HC group had lower p-tau concentrations than participants with MCI (p = .001) and AD dementia (p < .001). Similarly, the average p-tau concentrations were lower for the SCD group as compared to the MCI (p = .018) and AD (p < .001) groups. There were no group differences in the percentage of data that was cut out due to artefacts (F(3, 130) = 0.25, p = .859).
DFA.
The DFA exponents for the parieto-occipital region did not follow a normal distribution. The bootstrap method was applied to correct for this. There was a main effect of group on the DFA exponent, F(3, 131) = 8.543, p < .001. The AD dementia group had significantly lower DFA exponents than the HC (CI [-0.10, -0.03]) and SCD group (CI [-0.12, -0.03]). There were no differences between HC, SCD, and MCI or between AD dementia and MCI (Fig. 4a). The group effects persisted after adding age as a covariate, F(3, 128) = 8,15, p < .001, with no significant amount of variance explained by age, F(1, 128) = 0.239, p = .626. Spearman correlation was computed for the correlation analyses, as the assumption of normality was violated. The
Table 1. Participant Characteristics.
HC n SCD n MCI n AD n Sex (M/F) 29/18 47 9/8 17 10/9 19 24/28 52 Age (years) 56.7 (9.4) 47 60 (8.0) 17 65.6 (5.9) 19 64.0 (8.6) 52 Education 5.37 (1.4) 47 5.47 (1.0) 17 5.37 (1.1) 19 5.14 (1.0) 52 MMSE (1.73) 27.98 47 26.88 (3.3) 17 26.16 (2.1) 19 20.47 (5.3) 52 AB42 elecsys (pg/ml) (347.3) 1721.2 44 (462.3) 808.5 15 (299.2) 866.5 17 (205.3) 635.0 47 Tau elecsys (pg/ml) 16.9 (5.8) 44 17.3 (11.2) 15 33.3 (18.3) 17 35.9 (20.2) 47
Values are the average (SD). HC, healthy control; SCD, Subjective Cognitive Decline; MCI, mild cognitive impairment; AD, dementia due to Alzheimer’s Disease; M/F, male/female
analyses yielded a moderate positive correlation between the parieto-occipital DFA exponent and MMSE score, r = .414, p < .001. Additionally, a weak negative association between p-tau concentration and the DFA exponent was found, r = -.280, p = .012 (Fig. 4d,g). Similar patterns were observed for the occipital and parietal regions (Appendix C).
There were also significant group differences for the hippocampus, F (3,131) = 5.570, p = 0.001. The AD dementia group had a lower DFA exponent than the HC (p = .028) and SCD (p = .030) group. There were no significant differences between AD dementia and MCI or between HC, SCD, and MCI (Fig 5a). Adding age as a covariate did not change the results, as there was still a group effect, F(3,128) = 5.193, p < .002. There was no significant main effect of age (F(1, 128) = 0.239, p = .626. The DFA exponent for the hippocampus correlated positively with MMSE r = .397, p < 0.001. However, there was no association between the DFA exponent and p-tau in this cluster, r = -.168, p = 0.061 (Fig 5d,g).
FOOOF.
Figure 6 displays the average spectrum for each group. The average spectrums seemed to contain a bend around 20Hz. However, upon inspection of individual power spectrums, this bend was found to be a result of individual differences of the peak and width of the beta-band oscillatory peak. For the parieto-occipital cluster, there was a significant main effect, F(3, 131) = 5.991, p = .001, indicating group differences in the spectral exponent. Post hoc tests demonstrated that for the AD dementia group, the spectral exponent was significantly higher (i.e., steeper slope) as compared to the HC (p = .003) and the SCD groups (p = .020). There were no significant differences between MCI and AD or between HC, SCD, and MCI (Fig. 4b). The effects persisted after adding age as a covariate, F(3, 128) = 4.97, p = .028. Age did also explain a significant amount of variance in the exponent, F(1, 128) = 7.13, p < .001. The parieto-occipital spectral exponent was negatively correlated to the MMSE score, r = -.393, p < .001, indicating that lower MMSE scores were associated with steeper power spectral slopes. There was no significant relationship between the spectral exponent and the p-tau concentration, r = .080, p = .479 (Fig. 4e,i). Group differences in spectral slopes displayed comparable patterns for occipital and parietal regions (Appendix C).
Before adding age as a covariate, there was no main effect of group for the hippocampus, F(3,131) = 2.421, p = .069. However, significant group differences appeared after adding age as a covariate to the model, F(3,128) = 3.290, p = .023. The effects were driven by a lower spectral exponent in AD dementia as compared to HC (p = .007) and SCD (p = .031). The difference between MCI and AD dementia was not significant, nor were there any differences between HC, SCD, and MCI (Fig. 5b). There was no significant effect of age on the spectral exponent, F(1,128) = 3.898, p = .051. Results from the correlation analyses for the hippocampus were comparable to the parieto-occipital cluster: there was a significant negative correlation between
Table 2. Estimates for the DFA, FOOF, fE/I, and percentage of DFA exponents under 0.55 for each group. Values represent average (SD in parenthesis) unless otherwise specified
DFA FOOOF fEI % DFA < 0.55
Parieto-occipital
(median (IQR)) Hippocampus Parieto-occipital Hippocampus Parieto-occipital (median (IQR)) Hippocampus
Parieto-occipital Hippocampus HC 0.69 (0.07) 0.66 (0.06) 0.47 (0.08) 0.53 (0.09) 0.86 (0.23) 0.79 (0.13) 2.04 (4.24) 4.26 (14.1) SCD 0.70 (0.15) 0.70 (0.06) 0.45 (0.10) 0.52 (0.08) 0.82 (0.16) 0.78 (0.13) 5.39 (7.76) 8.82 (26.4) MCI 0.70 (0.09) 0.66 (0.06) 0.48 (0.09) 0.51 (0.09) 0.87 (0.15) 0.85 (0.17) 3.7 (5.87) 2.63 (11.47) AD 0.62 (0.12) 0.62 (0.06) 0.54 (0.10) 0.57 (0.14) 0.92 (0.12) 0.90 (0.16) 13.78 (16.15) 7.7 (18.21)
spectral slope and the MMSE (r = -.350, p < .001), but no association between p-tau and the spectral slope, r = .019, p = .832 (Fig. 5e,i).
fE/I.
The fE/I was computed with a DFA cutoff score of 0.6 and 0.55. The two different cutoff scores did not result in different fE/I estimates, but a lower cutoff of 0.55 allowed for the inclusion of more ROIs
(Appendix B). For both the parieto-occipital cluster and the hippocampus, significantly more ROIs had to be excluded from fE/I analysis for the AD dementia group as compared to the HC, SCD and MCI groups (Appendix B). The assumptions of normality of the data and homoscedasticity were both violated. The bootstrap method and Welch’s correction were applied to correct for this. For analyses with a DFA cutoff of 0.55, there was a main effect of group on the parieto-occipital fE/I estimate, F(3, 47.006) = 3.765, p = .017, where the AD dementia group had a higher fE/I estimate than the HC group, CI [0.01, 0.14] (Fig. 4c). There were no differences between any of the other groups. After adding age as a covariate the group effects remained,
Figure 4 DFA exponent, FOOOF exponent and fE/I estimates for the parieto-occipital cluster. The DFA exponent (A), FOOOF exponent (B) and fE/I estimate (C) for all groups computed for the parieto-occipital cluster. The horizontal lines represent the median (A,C) or the mean (B); the vertical lines represent the interquartile range (IQR) (A,C) or the SD (B). Each comparative test was conducted with age as a covariate (A-C). Effects significant at the 0.5 level are marked with an asterisk. (D-I) Scatter plot with all individual data points for the DFA (left panels), FOOOF (middle panels) and fE/I (right panels) together with the MMSE scores (D-G) and p-tau concentrations (G-I).
F(3, 128) = 3.236, p = .024. There was no significant main effect of age (F(1,128) = 0.064, p = .801. Finally, there was a significant negative correlation between the fE/I and the MMSE (r = -.204, p = .019), but there was no correlation between the fE/I and the p-tau concentration (r = .107, p = .236) (Fig. 4f,i). Group differences in fE/I estimates were comparable between occipital and parietal regions (Appendix C).
For the hippocampus there was also a main effect of group, F(3,130) = 5.327, p = 0.002. fE/I estimates were significantly higher for the AD dementia group as compared to the HC (p = .030) and the SCD group (p = .025) (Fig. 5c). The effects remained after adding age to the model, F(3,128) = 4.649, p < .004. Age explained no significant part of the variance, F(1,128) = 0.098, p = .755. Lastly, there was a significant negative correlation between the fE/I estimate and the MMSE (r = -.337, p < .001), but no significant association between p-tau and the fE/I, r = .142, p = .118 (Fig. 5f,i). See table 2 for the average DFA, FOOOF and fE/I estimates.
Discussion
This study aimed to demonstrate excitatory hyperactivity in early preclinical stages, and hypoactivity in clinical stages of AD using three methods to estimate E/I balance: the DFA, FOOOF, and fE/I. The current study does not provide evidence for early-stage hyperactivity. Results from the spectral exponent and LRTC
Figure 5 DFA exponent, FOOOF exponent and fE/I estimates for the hippocampus. The DFA exponent (A), FOOOF exponent (B) and fE/I estimate (C) for all groups computed for the hippocampus (left and right averaged). The horizontal lines represent the mean; the vertical lines represent the SD. Effects significant at the 0.5 level are marked with an asterisk. (D-I) Scatter plot with all individual data points for the DFA (left panels), FOOOF (middle panels) and fE/I (right panels) together with the MMSE scores (D-G) and p-tau concentrations (G-I).
for AD dementia were in line with the hypotheses: they provide evidence for increased inhibition in the AD dementia. However, the fE/I estimates were contradictory to the hypotheses, where AD dementia patients showed less inhibition as compared to healthy controls and preclinical AD.
Previously, animal studies (9–11) found increased spiking of hippocampal excitatory neurons of young AD mice models, and with the use of computational modelling, hyperactivity was demonstrated to be a plausible explanation for neurodegeneration (12). There are some human studies that have indicated increased excitatory activity levels, such as increased hippocampal task-related BOLD response in both familial (19) and non-familial AD (18) and hyperconnectivity within the medial temporal lobe (17). However, findings supporting hyperactivity are accompanied by studies supporting early-stage hypoactivity (20–22), and the applied measures may not be directly linked to levels of neuronal activity. In our study, with direct methods of neuronal activity, we did not find E/I ratio differences for the early preclinical and MCI of AD.
Perhaps the hyperactivity occurs ever earlier in the disease progression, even before the occurrence of subjective cognitive complaints. If so, brain networks in patients in preclinical and MCI stages of AD could already be shifting from hyperactive to hypoactive, hereby weakening the observed effects. Secondly, although the findings may reflect the truth, the lack of difference between healthy controls, preclinical AD and MCI could be due to too little power to pick up on the differences between healthy elderly and preclinical AD: the SCD and MCI groups were much smaller than the healthy elderly and AD dementia groups. Additionally, patients in these stages undergo many changes, and all subjects could be at different stages of progression. Finally, only a small portion of the preclinical AD patients will progress to AD dementia (49). Future studies should include more patients to elucidate if E/I ratio is genuinely unaltered, or if the absence of change in E/I were due to methodological limitations. Ideally, this study should be repeated with dominantly inherited genetic mutation carriers, to avoid dilution of the results by patients who will not develop AD dementia. Additionally, because familial AD patients are in an even earlier preclinical stage than the SCD group, this population could provide clarification on when during the preclinical AD stage hyperactivity occurs.
As opposed to early-stage hyperactivity, indications of a hypoactive state in AD dementia has been frequently reported: multiple studies find oscillatory slowing (4) (5) and decreased glucose metabolism (6,7). Again, there is no evidence that the measures used in these studies are directly associated with levels of neuronal activity. However, with direct measures of neuronal activity, we were able to replicate this finding. We provide new and strong evidence for excitatory hypoactivity in AD dementia. In both parieto-occipital regions and the hippocampus, AD dementia patients had steeper power spectral slopes, and an increased deviation from a critical state, both associated with imbalanced E/I (25,29). Deviation from a critical state could mean increased inhibition or excitation, but steeper power spectral have been specifically associated with increased inhibition. Taken together, these results provide evidence for increased inhibition in AD dementia as compared to earlier AD disease stages.
Figure 6. The average power spectrum for HC (A), SCD (B), MCI (C) and AD dementia (D) on log-log axes. The regression line represents the final fit of the 1/f component.
Unexpectedly, based on the fE/I estimates, networks of healthy elderly were found to be most inhibition dominated (i.e., lowest fE/I), whereas the networks of AD dementia patients, although still inhibition dominated, approximated a balanced E/I (i.e., highest fE/I). The fE/I algorithm was developed using a computer model and was validated with subjects who had unimpaired DFA exponents. In the computer model and the healthy validation sample, there was a clear correlation between the fE/I and the (derivative of the) DFA exponent. In our data, there was no clear association between the two: any DFA exponent could result in an fE/I estimate of one (i.e., balanced E/I). Validation with human data was performed on sensor level EEG data of a young sample. With sensor level data, volume conduction can deteriorate both the spatial and temporal resolution of a signal. Hence, the association between the DFA and fE/I may be driven by volume conduction, rather than the real neuronal activity (50). If so, the association between the DFA and fE/I could disappear after source reconstructing the signal.
Although this could explain why the fE/I estimates were low in all groups, it remains unclear why the fE/I estimate was higher in the AD dementia patients as compared to the other groups. The fE/I was demonstrated to be less reliable in the lower bound of DFA exponents (36). For very low DFA exponents, there is no correlation with alpha amplitude, and the fE/I gives a false estimate of one. For the AD dementia patients, the DFA exponent decreased towards the lower bound. This could have introduced a bias towards higher fE/I estimates, and thereby possibly explain higher fE/I estimates. Additional work needs to clarify how the fE/I behaves under low DFA exponents and if there are alternative ways to correct for low DFA exponents besides a cutoff. Additionally, future work should clarify the effect of volume conduction on the fE/I estimates. Until then, the fE/I results, at least for AD dementia patients with low DFA exponents, should be interpreted with caution.
Finally, when treating AD as a continuous disease (i.e., the correlation of the E/I measures with MMSE score and p-tau concentration), similar results were obtained. Those patients who had the steepest slopes and lowest DFA exponents tended to have more severe cognitive impairment. Again, the association was opposite for the fE/I. P-tau concentrations showed the same pattern, but only with the DFA in the parieto-occipital region: pathophysiological disease severity was negatively correlated to the DFA exponent
Some potential shortcomings need consideration. First, the E/I methods have been developed based data from healthy brains or computer models. Neurodegenerative diseases like AD are characterized by progressively severe brain atrophy. It has not been investigated how changes in the structure of the brain affect the outcome and reliability of the E/I methods. It is unlikely that structural differences explain the described effects: we found similar patterns for regions that often show brain atrophy, such as the hippocampus and parietal region, as well as regions that do not, such as the occipital lobe (51). Nevertheless, future studies should investigate how brain structure affects each measures’ output, to rule out that brain atrophy interferes with the methods’ reliability. A second limitation concerns the number of arbitrary choices that had to be made before applying each method. The decisions made during the analysis process may have influenced the outcome of that method (i.e., spectral exponent, DFA exponent, fE/I). To make an informed decision on the algorithm settings, we compared different algorithm settings for a small subset of the subjects. We observed that there were either no differences between settings or that the fit was best for chosen settings (e.g., DFA threshold in Appendix B). However, perhaps the small subset was not representative for the entire sample. To better clarify the effect of each (arbitrary) decision on the outcome, a systematic comparative study, where results are compared across different settings, is preferred. Finally, the E/I was only computed for parieto-occipital regions and the hippocampus. Possibly frontal and temporal regions exhibit different patterns. As all regions within the parieto-occipital region and the hippocampus showed similar patterns, it is plausible that changes are global rather than regional. Future studies could compare E/I on the whole-brain level, rather than a lobar level to confirm this.
In conclusion, the current study observed an unaltered E/I ratio in preclinical and MCI AD stages and was thus not able to provide evidence for the preclinical hyperactivity hypothesis. However, in line with the predictions, an increase in inhibition was found for demented AD patients. When treating AD as a continuous disorder, similar results were obtained. After the present study, it is still unknown if preclinical AD is characterized by neuronal hyperactivity, or if neuronal activity levels really remain stable until later in the disease progression. To gain a better understanding of the pathophysiology of AD, studying neuronal activity levels in preclinical AD remains an important research topic.
References
1. Jack CR, Bennett DA, Blennow K, Carrillo MC, Dunn B, Haeberlein SB, et al. NIA-AA Research
Framework: Toward a biological definition of Alzheimer’s disease. Alzheimer’s and Dementia. 2018. 2. Alzheimer’s Association. 2018 Alzheimer’s disease facts and figures. Alzheimer’s Dement. 2018; 3. Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, Fagan AM, et al. Toward defining the
preclinical stages of Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Aging-Alzheimer’s disease. Aging-Alzheimer’s Dement. 2011;
4. Engels MMA, van der Flier WM, Stam CJ, Hillebrand A, Scheltens P, van Straaten ECW. Alzheimer’s disease: The state of the art in resting-state magnetoencephalography. Clinical Neurophysiology. 2017.
5. Jeong J. EEG dynamics in patients with Alzheimer’s disease. Clinical Neurophysiology. 2004. 6. Edison P, Archer HA, Hinz R, Hammers A, Pavese N, Tai YF, et al. Amyloid, hypometabolism, and
cognition in Alzheimer disease: An [11C]PIB and [18F]FDG PET study. Neurology. 2007;
7. Kato T, Inui Y, Nakamura A, Ito K. Brain fluorodeoxyglucose (FDG) PET in dementia. Ageing Research Reviews. 2016.
8. Langbaum JBS, Chen K, Lee W, Reschke C, Bandy D, Fleisher AS, et al. Categorical and correlational analyses of baseline fluorodeoxyglucose positron emission tomography images from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Neuroimage. 2009;
9. Busche MA, Chen X, Henning HA, Reichwald J, Staufenbiel M, Sakmann B, et al. Critical role of soluble amyloid-β for early hippocampal hyperactivity in a mouse model of Alzheimer’s disease. Proc Natl Acad Sci U S A. 2012;
10. Bero AW, Yan P, Roh JH, Cirrito JR, Stewart FR, Raichle ME, et al. Neuronal activity regulates the regional vulnerability to amyloid-β 2 deposition. Nat Neurosci. 2011;
11. Yuan P, Grutzendler J. Attenuation of β-amyloid deposition and neurotoxicity by chemogenetic modulation of neural activity. J Neurosci. 2016;
12. de Haan W, Mott K, van Straaten ECW, Scheltens P, Stam CJ. Activity Dependent Degeneration Explains Hub Vulnerability in Alzheimer’s Disease. PLoS Comput Biol. 2012;
13. De Haan W, Stam CJ, Jones BF, Zuiderwijk IM, Van Dijk BW, Scheltens P. Resting-state oscillatory brain dynamics in alzheimer disease. J Clin Neurophysiol. 2008;
14. Palmqvist S, Schöll M, Strandberg O, Mattsson N, Stomrud E, Zetterberg H, et al. Earliest
accumulation of β-amyloid occurs within the default-mode network and concurrently affects brain connectivity. Nat Commun. 2017;
15. Buckner RL, Snyder AZ, Shannon BJ, LaRossa G, Sachs R, Fotenos AF, et al. Molecular, structural, and functional characterization of Alzheimer’s disease: Evidence for a relationship between default activity, amyloid, and memory. J Neurosci. 2005;
16. Nakamura A, Cuesta P, Kato T, Arahata Y, Iwata K, Yamagishi M, et al. Early functional network alterations in asymptomatic elders at risk for Alzheimer’s disease. Sci Rep. 2017;
17. Das SR, Pluta J, Mancuso L, Kliot D, Orozco S, Dickerson BC, et al. Increased functional connectivity within medial temporal lobe in mild cognitive impairment. Hippocampus. 2013;
18. Mormino EC, Brandel MG, Madison CM, Marks S, Baker SL, Jagust WJ. Aβ Deposition in aging is associated with increases in brain activation during successful memory encoding. Cereb Cortex. 2012;
19. Quiroz YT, Budson AE, Celone K, Ruiz A, Newmark R, Castrillõn G, et al. Hippocampal hyperactivation in presymptomatic familial Alzheimer’s disease. Ann Neurol. 2010;
20. Garcés P, Vicente R, Wibral M, Pineda-Pardo J ángel, López ME, Aurtenetxe S, et al. Brain-wide slowing of spontaneous alpha rhythms in mild cognitive impairment. Front Aging Neurosci. 2013; 21. López-Sanz D, Brunã R, Garcés P, Camara C, Serrano N, Rodríguez-Rojo IC, et al. Alpha band
disruption in the AD-continuum starts in the Subjective Cognitive Decline stage: A MEG study. Sci Rep. 2016;
22. Herholz K. Cerebral glucose metabolism in preclinical and prodromal Alzheimers disease. Expert Review of Neurotherapeutics. 2010.
23. Tewarie P, Hunt BAE, O’Neill GC, Byrne A, Aquino K, Bauer M, et al. Relationships between Neuronal Oscillatory Amplitude and Dynamic Functional Connectivity. Cereb Cortex. 2019;
24. Northoff G, Qin P, Nakao T. Rest-stimulus interaction in the brain: A review. Trends in Neurosciences. 2010.
25. Gao, R., Peterson, E. J., & Voytek B. Inferring synaptic excitation/inhibition balance from field potentials. Neuroimage. 2017;158:70–8.
26. Atallah B V., Scanziani M. Instantaneous Modulation of Gamma Oscillation Frequency by Balancing Excitation with Inhibition. Neuron. 2009;
27. Monier C, Fournier J, Frégnac Y. In vitro and in vivo measures of evoked excitatory and inhibitory conductance dynamics in sensory cortices. J Neurosci Methods. 2008;
28. Hardstone R, Poil SS, Schiavone G, Jansen R, Nikulin V V., Mansvelder HD, et al. Detrended fluctuation analysis: A scale-free view on neuronal oscillations. Front Physiol. 2012;
29. Poil SS, Hardstone R, Mansvelder HD, Linkenkaer-Hansen K. Critical-state dynamics of avalanches and oscillations jointly emerge from balanced excitation/inhibition in neuronal networks. J Neurosci. 2012;
30. Nikulin V V., Jönsson EG, Brismar T. Attenuation of long-range temporal correlations in the amplitude dynamics of alpha and beta neuronal oscillations in patients with schizophrenia. Neuroimage. 2012;
31. Montez T, Poil SS, Jones BF, Manshanden I, Verbunt JPA, Van Dijk BW, et al. Altered temporal correlations in parietal alpha and prefrontal theta oscillations in early-stage Alzheimer disease. Proc Natl Acad Sci U S A. 2009;
32. Haller M, Donoghue T, Peterson E, Varma P, Sebastian P, Gao R, et al. Parameterizing neural power spectra. bioRxiv. 2018;
33. Pritchard WS. The brain in fractal time: 1/f-like power spectrum scaling of the human electroencephalogram. Int J Neurosci. 1992;66(1–2):119–29.
34. Voytek B, Kramer MA, Case J, Lepage KQ, Tempesta ZR, Knight RT, et al. Age-related changes in 1/f neural electrophysiological noise. J Neurosci. 2015;
35. Peterson EJ, Rosen BQ, Campbell AM, Belger A, Voytek B. 1/F Neural Noise Is a Better Predictor of Schizophrenia Than Neural Oscillations. DoiOrg. 2017;
36. Bruining, K. Hardstone, R., Juarez-Martinez, E. L., Sprengers, J., Avramiea, A., Simpraga, S., Houtman, S. J., Poil, S., Dallares, E., Palva, S., Oranje, B., Matias Palva, J., Mansvelder, H. D., Linkenkaer-Hansen K. Measurement of excitation-inhibition ratio in autism spectrum disorder using critical brain dynamics. Sci Rep. 2020;10(1):1–15.
37. Bozzi Y, Provenzano G, Casarosa S. Neurobiological bases of autism–epilepsy comorbidity: a focus on excitation/inhibition imbalance. Eur J Neurosci. 2018;
38. Rubenstein JLR, Merzenich MM. Model of autism: Increased ratio of excitation/inhibition in key neural systems. Genes, Brain and Behavior. 2003.
39. Van Der Flier WM, Pijnenburg YAL, Prins N, Lemstra AW, Bouwman FH, Teunissen CE, et al. Optimizing patient care and research: The Amsterdam dementia cohort. J Alzheimer’s Dis. 2014; 40. Burns A, Brayne C, Folstein M. Mini-Mental State: A practical method for grading the cognitive state
of patients for the clinician. M. Folstein, S. Folstein and P. McHugh, Journal of Psychiatric Research (1975) 12, 189-198. International Journal of Geriatric Psychiatry. 1998.
41. Zwan M, van Harten A, Ossenkoppele R, Bouwman F, Teunissen C, Adriaanse S, et al. Concordance between cerebrospinal fluid biomarkers and [11C]PIB PET in a memory clinic cohort. J Alzheimers Dis. 2014;
42. Willemse EAJ, van Maurik IS, Tijms BM, Bouwman FH, Franke A, Hubeek I, et al. Diagnostic performance of Elecsys immunoassays for cerebrospinal fluid Alzheimer’s disease biomarkers in a
nonacademic, multicenter memory clinic cohort: The ABIDE project. Alzheimer’s Dement Diagnosis, Assess Dis Monit. 2018;
43. Hampel H, Bürger K, Pruessner JC, Zinkowski R, DeBernardis J, Kerkman D, et al. Correlation of cerebrospinal fluid levels of tau protein phosphorylated at threonine 231 with rates of hippocampal atrophy in Alzheimer disease. Arch Neurol. 2005;
44. Fagan AM, Xiong C, Jasielec MS, Bateman RJ, Goate AM, Benzinger TLS, et al. Longitudinal change in CSF biomarkers in autosomal-dominant Alzheimer’s disease. Sci Transl Med. 2014;
45. Taulu S, Hari R. Removal of magnetoencephalographic artifacts with temporal signal-space separation: demonstration with single-trial auditory-evoked responses. Hum Brain Mapp. 2009; 46. Hillebrand A, Tewarie P, Van Dellen E, Yu M, Carbo EWS, Douw L, et al. Direction of information flow
in large-scale resting-state networks is frequency-dependent. Proc Natl Acad Sci U S A. 2016; 47. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, et al. Automated
anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage. 2002;
48. Field A. Discovering statistics using IBM SPSS statistics. Statistics. 2013.
49. Slot RER, Sikkes SAM, Berkhof J, Brodaty H, Buckley R, Cavedo E, et al. Subjective cognitive decline and rates of incident Alzheimer’s disease and non–Alzheimer’s disease dementia. Alzheimer’s Dement. 2019;
50. Burle B, Spieser L, Roger C, Casini L, Hasbroucq T, Vidal F. Spatial and temporal resolutions of EEG: Is it really black and white? A scalp current density view. Int J Psychophysiol. 2015;
51. Frisoni GB, Fox NC, Jack CR, Scheltens P, Thompson PM. The clinical use of structural MRI in Alzheimer disease. Nature Reviews Neurology. 2010.
Appendix A: FOOOF frequency range
In a small subset of the data (N = 12; n = 3 in each group) the fit of aperiodic 1/f spectral exponents was compared over three different frequency ranges: 3 – 40 Hz, 3 – 50 Hz, and 3 – 70 Hz. The total spectral fit (i.e. both the aperiodic 1/f component and the periodic oscillatory peaks) explained the same amount of variance in all frequency ranges (R2 = 0.93 for the parieto-occipital cluster; R2 = 0.98 for the hippocampus).
However, based on visual inspection, the 3 – 40 Hz frequency range seemed to result in a better fit of the 1/f spectral exponent. Therefore, the FOOOF algorithm was applied to this range.
Appendix B: DFA cutoff
Low DFA exponents were demonstrated to biased fE/I estimates (36). We examined whether to maintain a DFA cutoff score of 0.55 or 0.60. The fE/I was slightly higher for a cutoff score of 0.55 as compared to a cutoff score of 0.60 (an increase of 0.007 and an increase of 0.002 for the parieto-occipital and hippocampus cluster, respectively; see S1)). The main difference between the two cutoffs lies in the number of ROIs to exclude. For a DFA cutoff of 0.6, on average, 5,39 ROIs (SD = 6.02) had to be excluded (out of 24 ROIs) in the parieto-occipital cluster. In the hippocampus cluster, on average, 0.43 (SD = 0.66) had to be excluded (out of 2 ROIs). For a DFA cutoff of 0.55, on average 1,73 (SD = 2.90) and 0.06 (SD = 0.17) ROIs had to be excluded, for the parieto-occipital and hippocampus cluster, respectively (Fig S2). Concluding, in the current study, a cutoff of 0.55 is best. There is only a minimal difference in fE/I estimate, but considerably more ROIs can be included in fE/I computation.
A linear regression analysis was conducted to investigate if the number of DFA exponents below 0.55 could predict the fE/I. A significant regression equation (F(1, 132) = 5.807, p = 0.017), with an R2 of 0.042,
demonstrated that the number of excluded ROIs is indeed weakly associated with the fE/I estimate.
Figure S1. The fE/I estimates for a DFA cut-off of 0.55 (left upper and lower panels) and a DFA cut-off of 0.6 (right upper and lower panels) for the parieto-occipital cluster (upper panels) and the hippocampus (lower panels).
Figure S2. All fE/I estimates and DFA exponents of parieto-occipital and hippocampal ROIs for each group. The grey data-points have a DFA exponent below 0.55 and were excluded from further analyses. The AD dementia group had more data-point with a DFA under 0.55. These point were excluded as they give an unreliable fE/I estimate.
Appendix C: E/I estimates for parietal and occipital ROIs
Figure S3. The group differences for the DFA (top panels), FOOOF (middle panels) and fE/I (bottom panels) follow a similar pattern for parietal (left panels) and occipital (right panels) regions. This justifies the clustering of the regions into one cluster.
