Altered Low-Frequency EEG Connectivity in Mild Cognitive Impairment as a Sign of Clinical Progression

Recruitment and subjects

The data from the included participants have also been used for other studies [31 –33]. Furthermore, the power calculation are presented elsewhere [7].

The study was a prospective cohort study, conducted at two memory clinics at the Regional Dementia Research Centre, Department of Neurology, Zealand University Hospital and the Copenhagen Memory Clinic, Department of Neurology, University Hospital Copenhagen, Rigshospitalet. Patients consecutively referred from June 2012 to December 2014 for cognitive evaluation and diagnosed with either MCI or mild AD and at least a baseline Mini-Mental State Examination (MMSE) score of ≥22 were eligible for inclusion. The selection of patients was defined on preexisting exclusion criteria: If they had no close relatives who wished to participate, if they were participating in other intervention studies or if they were suffering from other neurological, psychiatric, or other severe disease. In addition, patients receiving sedative medication were excluded due to a potential sedative effect and patients who had any past or current addictions to alcohol or medications were excluded.

The HC were all volunteers recruited for scientific research through public advertisements at the memory clinics, at local associations for elders and through an online recruitment site for trial subjects. Inclusion criteria were age between 50–90 years, MMSE score ≥26, ACE ≥85, normal neurological and clinical examination, normal or age-related atrophy measured on a computed tomography (CT) scan of the brain, no and normal routine blood tests. Exclusion criteria were an inability to participate (including impaired vision or hearing) or presence of memory complaints or other cognitive symptoms as well as signs of major neurological, psychiatric or other severe disease, which potentially could elicit cognitive impairments including any signs of major depression and/or a Geriatric Depression Scale score >7. Furthermore, they could not be pregnant, have undergone general anesthesia or received electroconvulsive therapy in the past 3 months, receive sedatives, or have any past or current addictions to alcohol or medications.

In total, we included 17 patients with AD, 27 patients with MCI, and 38 HC. The study was reported to and approved by the Danish Data Protection Agency and by the Regional Ethical Committee according to Danish legislation.

Diagnostic assessment

At the time of referral, all patients underwent a standardized diagnostic assessment including a full physical and neurological examination, routine blood analysis, brain CT or MRI scan as well as cognitive screening, i.e., MMSE, Addenbrooke’s Cognitive Examination (ACE) and Digit Symbol Substitution Test (DSST) including Clinical Dementia Rating (CDR). Furthermore, as part of the diagnostic assessment, patients and relatives underwent Neuropsychiatric Inventory (NPI), Major Depression Inventory (MDI), and Activities of Daily Living Inventory (ADCS-ADL). All the CT and MRI scans were examined by an experienced neuro-radiologist. The majority also had a lumbar puncture (except two patients with MCI and six HC) performed with subsequent determination of CSF AD biomarkers, i.e., amyloid-β₄₂, total tau, and phosphorylated tau, and routine parameter analysis. If it was considered diagnostically relevant, the patients also had a full neuropsychological evaluation undertaken by a clinical neuropsychologist, but these were individualized for each patient with varying overlap and therefore not included in the current study. Diagnoses were settled by consensus of an experienced multidisciplinary team based on all available examination results. Patients with MCI were diagnosed according to the Winblad consensus criteria [34] and patients with AD were diagnosed according to the NIA-AA criteria [35].

At inclusion, all HC underwent the standardized diagnostic assessment, which included ACE, MMSE, MDI, and DSST, and analysis of CSF was performed on almost all HC. At the baseline visit, all HC were referred for a standardized EEG. The EEG recordings were not used in the assessment.

Study design

All tests, including CDR were repeated at inclusion. Recruiting of patients happened within 6 months after diagnosis. Afterwards, follow-up visits were carried out on a yearly basis, with serial cognitive tests, i.e., MMSE and ACE and the NPI, MDI, ADCS-ADL, and CDR scales. Clinical progression of MCI to AD was determined based on whether the patient clinically progressed to fulfilled the NIA-AA criteria [35]. If they progressed to another diagnosis (for example, Lewy body dementia), they were excluded from the comparison between pMCI and sMCI.

During the study period, the primary investigator performing the tests was blinded for the results of the EEG, imaging, and CSF analysis, and thereby blinded for the potential presence of underlying AD pathology.

Electroencephalography recording

All EEG recordings were performed at the two participating centers who beforehand had agreed on a common approach to record EEGs. EEG recordings were performed using NicoletOne EEG Systems (Natus^®) with a sampling rate of either 500 or 1000 Hz. Nineteen electrodes were positioned according to the International 10–20 system. The IS 10–20 system describes the location of the scalp electrodes during an EEG recording and has been developed to ensure standardized reproducibility. Most EEGs were recorded with alternating eyes closed and eyes open periods for three minutes each. The participants were alerted if they became visibly drowsy, since drowsiness influences recording. The neurophysiology assistant recording the EEG made marks in the EEG when the participant closed and open their eyes. In the current study, we used the eyes closed segments for analysis. After the recording, the files were exported as raw EEGs without any filtering.

Collection and analysis of cerebrospinal fluid

CSF was obtained by lumbar puncture between the L3/L4 or L4/L5 intervertebral space and collected in polypropylene tubes. CSF analyses included routine parameters (white cell count, erythrocytes, total protein, glucose, albumin, IgG-index, and oligoclonal bands) and the core AD biomarkers, i.e., Aβ₄₂, T-tau, and P-tau. CSF Aβ₄₂, T-tau, and P-tau were quantified with sandwich ELISAs (INNOTEST amyloid-β₄₂, hTau, and Phospho-Tau [181P], respectively; Fujirebio Europe, Ghent, Belgium). AD biomarkers analyses from both clinics were all carried out at one central laboratory.

Preprocessing of EEG

The EEG data were imported to MATLAB (Mathworks, v2016a) using the EEGLAB toolbox [36]. Only segments with eyes closed were selected either using markers placed doing recording or from the first 10 minutes of recording if markers were not present. The electrodes were computationally located on the scalp using the dipfit toolbox [37] with the standard 10–20 electrode model. The excessive channels were removed and the data were bandpass filtered from 1–70 Hz using the pop_firws function in MATLAB with a filter order of 2 and the Kaiser window parameter beta was estimated using a maximum passband ripple of 0.001. Furthermore, the data were bandstop filtered from 45–55 Hz using the same settings as described previously. Afterwards, the data were down sampled to 200 Hz. Then, the data were divided into one second epochs and the EEGs were visually inspected and epochs with excessive noise or artifacts were removed. Channels with excessive noise, drift, or bad connection were interpolated using spherical interpolation. The EEG had to have≤three electrodes with excessive artifact, otherwise the EEG was excluded from the analysis. Afterwards, the EEGs were re-referenced to average and independent component analysis (ICA) was performed using the extended infomax algorithm [38] for each file and components that contained eye blinks, eye movement, or specific line noise artifacts were removed manually. Lastly, the EEGs were inspected visually again and epoch with excessive noise or artifacts were removed. The investigator who performed the preprocessing was blinded to the diagnosis. Due to excessive artifacts, we excluded the following number of EEGs: two from patients with AD, two from patients with MCI, and one from HC. When comparing pMCI, and sMCI, one EEG from MCI was excluded due to clinical progression to vascular dementia.

Coherence calculations

Coherence is the square of the cross-spectrum of the electrodes divided by the product of the power spectra of the individual electrodes. This way of calculating coherence is a measure of consistency of a phase relationship between two signals ranging from 0 to 1. The imaginary part of coherency (iCoh) was calculated by taking the imaginary part of the cross-spectrum of the electrodes divided by the square root of the product of the power spectra of the individual electrodes [28]. Since we were interested in the magnitude, we calculated the absolute value of the imaginary result. Coherence and iCoh were calculated for each epoch separately and averaged. This was done between each pair of electrodes for each of the following frequency bands: delta (1–3.99 Hz); theta (4–7.99 Hz); alpha (8–12.99 Hz) and beta (13–29.99 Hz). Total coherence and iCoh were calculated by averaging across all electrodes and were done separately for each frequency band.

Statistics

All statistics were performed in MATLAB (vR2016a). When comparing demographics, number of epochs and cognitive scores for AD, MCI, and HC, we performed one-way ANOVAs. Independent t-tests were used to compare baseline cognitive scores between pMCI and sMCI. For coherence and iCoh, we log-transformed the data since the data was non-normally distributed. To compare all three groups, we performed an ANCOVA with age, gender, education, and current medication (antidepressants, anti-dementia medication and pain killers) as covariates and if we found a significant difference after correcting for multiple comparisons for each frequency band separately, we performed t-tests between AD versus HC, MCI versus HC, and AD versus MCI. For pMCI versus sMCI, we used ANCOVA with the same covariate as mentioned above and divided the baseline EEG from the patients with MCI into two groups (pMCI and sMCI) based on 2nd year follow-up. T-values were used for visualization of significant differences. Due to the exploratory nature of the study, we did not perform correction for multiple comparisons between pMCI and sMCI. For comparing total coherence values, we used ANCOVA between three groups (AD, MCI, and HC) and t-tests for comparing pMCI and sMCI.

Correlations

We correlated the CSF markers (Aβ₄₂, T-tau, and P-tau), MMSE, and ACE to the total coherence for each frequency bands for all included participants using Spearman’s correlation. If it was not possible to measure the value for Aβ₄₂ or T-tau due to a value that was too high to measure with standard techniques the values were excluded from the analysis. A correlation was significant if the p-values was below 0.05.

Demographics, cognitive tests, and EEG length

See Table 1 for characterization of the patients included in the analysis and comparisons between groups and Table 2 for the performance on cognitive screening instruments for each visit with comparison between the scores. See Table 3 for the comparison between demographics, baseline cognitive scores, and CSF biomarkers for pMCI and sMCI. Full overview of the cognitive scores for each visit including after one year and after three years can be found in the Supplementary Material (Supplementary Table 17) including a flow diagram (Supplementary Figure 1). The number of 1-second epochs between AD (mean(SD) = 147.13 (19.57)), MCI (mean(SD) = 153.56 (44.58)), and HC (mean(SD) = 177.46 (62.05)) were not significantly different (p = 0.078, F = 2.643). The number of epochs between pMCI (mean(SD) = 156 (56.75)) and sMCI (mean(SD) = 149 (34.28)) were not significantly different either (p = 0.713, t = 0.372).

Coherence and iCoh in AD and MCI compared to healthy aging

For full overview of coherence and total coherence results see Figs. 1 and 5A. The majority of significant differences were found in MCI compared to both HC and AD in the theta band. In AD, the significant increases were less pronounced in the theta band than in MCI and involved connections between central, frontal and temporal electrodes (i.e., F3-F7, C4-Fp2, and T4-F4). In the alpha band, the significant changes were decreases between frontal electrodes especially involving Fp1 and Fp2 both between AD and HC, and MCI and HC. Furthermore, we found temporal-temporal and temporal-frontal increased coherence and decreased coherence frontal-frontal, and frontal-occipital coherence between MCI versus HC in the beta band. In addition, the significant changes between AD and HC were less pronounced in beta compared to MCI. In the delta band, we found very mixed results with both increased and decreased coherence. However, in total coherence we found that the alpha band was overall most decreased in patients with AD and MCI compared to HC but was not significantly different. But we found a significant increase for MCI compared to HC (p≤0.000) in the theta band. All statistical test results for the ANCOVA can be found in Supplementary Tables 1–8.

OPEN IN VIEWER

Fig.1

The difference in coherence between groups (AD versus HC, MCI versus HC, and AD versus MCI) for all frequency bands in eyes closed condition. The difference is illustrated if the post-hoc test was significant, and t-values were used to illustrate the differences. Red indicates increased coherence in the first group listed. Blue indicates decreased coherence in the first group listed. Thicker lines represent larger t-scores, t-scores range from –5.17 to 6.58. HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

For full overview of iCoh results, see Figs. 2 and 5B. The most prominent findings in iCoh were decreases found in the delta band, which involved the temporal, central, and frontal electrodes in particular. The decreases were more pronounced in MCI compared to AD. In the theta band, we found an increase in coherence in the posterior electrodes for both AD and MCI, which involved the parietal, temporal (T5, and T6), and occipital electrodes. For the alpha band, increased iCoh between frontal electrodes were present in both AD and MCI. For total iCoh, no significant differences between AD, MCI, and HC were found. All statistical results for the ANCOVA can be found in the Supplementary Material (Supplementary Tables 17–24).

OPEN IN VIEWER

Fig.2

The difference in imaginary part of coherency between groups (AD versus HC, MCI versus HC, and AD versus MCI) for all frequency bands in eyes closed condition. The difference is illustrated if the post-hoc test was significant, and t-values were used to illustrate the differences. Red indicates increased coherence in the first group listed. Blue indicates decreased coherence in the first group listed. Thicker lines represent larger t-scores, t-scores range from –6.78 to 5.91. HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

Coherence differences between pMCI and sMCI

When comparing pMCI and sMCI, the majority of significant increases were found in the delta band between right-temporal and frontal electrodes (T6-F7, T6-F3, T4-F7, and T4-F8) (Fig. 3). In the beta, and alpha bands, only a few sporadic changes were found. When examining iCoh, the most pronounced changes were found in the delta band with decreased iCoh for frontal-frontal, temporal-frontal, and parietal-frontal connections (Fig. 4). All statistical test results for the ANCOVA can be found in the Supplementary Material (see Supplementary Tables 9–16 for coherence and Supplementary Tables 25–32 for iCoh).

OPEN IN VIEWER

Fig.3

The difference in coherence between progressed mild cognitive impairment (pMCI) and stable mild cognitive impairment (sMCI) for all frequency bands. Thicker lines represent larger t-scores, t-scores range from –9.38 to 13.28. Red indicates increased coherence in pMCI compared to sMCI. Blue indicates decreased coherence in pMCI compared to sMCI. HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

OPEN IN VIEWER

Fig.4

The difference in imaginary part of coherency between progressed mild cognitive impairment (pMCI) and stable mild cognitive impairment (sMCI) for all frequency bands. Thicker lines represent larger t-scores, t-scores range from –21.59 to 10.39. Red indicates increased coherence in pMCI compared to sMCI. Blue indicates decreased coherence in pMCI compared to sMCI. HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

For total coherence, no significant differences were found but pMCI had an increased coherence in the delta band (p = 0.1678) and especially in the theta band (p = 0.0617) (Fig. 6A). However, we found that total delta iCoh was significant different between pMCI and sMCI (p = 0.0182, t-stat: –2.551) (Fig. 6B).

OPEN IN VIEWER

Fig.5

Figure showing the total A) coherence and B) iCoh in each frequency band for all groups (AD, MCI, and HC).

OPEN IN VIEWER

Fig.6

Figure showing A) the total coherence in each frequency band for all pMCI and sMCI and B) the total iCoh in each frequency band for all pMCI and sMCI at 2nd year follow-up.

Correlations

When we performed correlations between ACE and total coherence, we found a significant negative correlation with total theta coherence (p = 0.038; rho = –0.239) and a positive correlation for total beta coherence (p = 0.013; rho = 0.283). Between the MMSE and total beta coherence, we found a significant positive correlation (p = 0.027; rho = 0.252). No significant correlations were found between CSF biomarkers and total coherence. Scatterplots can be found in Supplementary Figure 2.

Conclusion

In the current study, we found that patients with MCI, who progress to AD within the following two years have an increased low-frequency coherence and decreased delta iCoh. The increased low-frequency coherence may be due to increased atrophy of the hippocampus and may even be due to degeneration of subcortical cholinergic structures as suggested in previous studies. Furthermore, we found that the total alpha power was almost the same between AD and MCI, which may be due to an increased alpha coherence in the patients with AD receiving cholinesterase inhibitors. In addition, we found that iCoh may be able to differentiate between dementia due to AD and HC. Lastly, we found that total theta and total beta coherence were correlated to the total score of the ACE cognitive test, which may indicate that the low-frequency coherence are involved in the progression of MCI to AD. Overall, these findings suggest that low frequency coherence and iCoh could potentially be used to differentiate between pMCI and sMCI. However, larger studies are needed to confirm these findings.