Differences in Multimodal EEG and Clinical Correlations Between Alzheimer’s Disease and Frontotemporal Dementia

Background. Alzheimer’s disease (AD) and frontotemporal dementia (FTD) are the two main types of dementia. We aim to investigate the difference between AD and FTD by use of multimodal EEG analyses. Additionally, the difference in correlations between EEG and clinical data was also investigated. Methods. Thirty-one patients diagnosed with AD and 15 patients with FTD were recruited (2008.1-2020.2), along with 24 healthy controls. Clinical data were reviewed. EEG microstate analysis, spectral analyses, and connectivity analysis were performed.


Background
Alzheimer's disease (AD) is the most common form of dementia, accounting for 60 to 80 percent of cases [1]. The most essential and often earliest clinical manifestation of AD is selective memory impairment.
Accumulation of abnormally folded amyloid beta (Aβ) and tau proteins in amyloid plaques and neural tangles are causally related to neurodegenerative processes [2]. Aβ is thought to be the trigger of the disease process. Tau is a prerequisite for diagnosis of AD, but also can act independently without amyloid plaques to cause neurodegeneration in other disease, like frontotemporal dementia (FTD) [1].
FTD accounts for approximately 10% of all dementias [3], characterized by prominent changes in social behavior and personality or aphasia accompanied by pathological changes in the frontal and/or temporal lobes. TAR DNA-binding protein 43, tau, and fused-in-sarcoma protein were the three major disease proteins in the neuropathology of FTD [4]. It can be di cult to distinguish clinically between AD and FTD, as AD may manifest with behavioral disturbances, and memory disturbances may manifest in FTD. Cerebrospinal uid (CSF) biomarkers Aβ, total tau (t-tau), and phosphorylated tau (p-tau) have good accuracy in predicting AD [5,6]. Low Aβ 42 levels, high concentrations of t-tau and p-ta, and the ratio of tau/Aβ 42 help to discriminate AD from healthy controls and other dementias [6,7]. Moreover, electroencephalography (EEG) is increasingly considered to be a potential biomarker for dementia differentiation recently.
EEG is a relatively cost-effective, non-invasive technique. It provides in vivo data on electrical activity with high temporal resolution. Several characteristics of the EEG have been put forward as biomarkers in AD and might be useful in the early recognition of neural signatures of dementias and differential diagnosis [8]. EEG microstates are de ned as quasi-stable brief patterns of coordinated electrical activity on the cortical surface, which was rst described by Lehman D et al. [9,10]. The topographies remained transiently stable for 60-150 ms before rapidly transitioning into a new state. These microstates have been shown to in uence cognition and perception [11,12]. Moreover, changes in consciousness state have been related to microstates changes [13]. Britz et al. reported that different cognitive functions were associated with speci c microstates [14]. Previous studies have investigated microstates changes in cognitive disorders [10,[15][16][17][18][19]. More researches are required to get more convincing conclusions.
Studies on the difference in microstate and the correlation between CSF biomarkers and microstates between AD and FTD are limited. The current study was set to investigate EEG microstate' characteristics in AD and FTD, along with EEG spectral and connectivity analysis, and the correlations between EEG and clinical data. The differences in EEG and clinical data were then analyzed to test the utility of EEG as a biomarker for clinical evaluations and differential diagnosis.

Patients
The study population consisted of patients with cognition impairment in Peking Union Medical College Hospital between June 2015 and October 2019. Patients were diagnosed based on information obtained from an extensive clinical history, physical examinations, and excluded mood disorders and schizophrenia. Clinical assessment scales included the Mini Mental State Examination (MMSE) [20], the Montreal Cognitive Assessment (MoCA) [21], and activities of daily living (ADL) score. Patients who had complications of other neurological or psychiatric disorders, and severe systemic diseases that may in uence the cerebral nervous system, were also excluded. For FTD diagnosis, the Neary and Snowden et al. or the Mckhann et al. criteria were employed [22,23]. The National Institute on Aging-Alzheimer's Association (NIA-AA) criteria were used to diagnose patients with dementia due to AD [24]. Dementia diagnoses were performed independently by two experienced clinicians.

Biomarkers assessments
Page 4/27 CSF t-tau, p-tau and Aβ 42 were measured using an enzyme-linked immunosorbent assay (Fujirebio, Ghent, Belgium). Samples were handled by experienced senior laboratory technicians blinded to patients' information.

EEG examination and data preprocessing
EEG monitoring was performed using a 19-channel video-EEG monitoring system (NIHON KOHDEN, EEG-1200C). Recording electrodes were placed according to the international 10-20 system with a sampling frequency of 500 Hz. The degree of visual EEG abnormality was scored as follow: (1) 0 = normal; (2) 1 = mild abnormal: mild asymmetry background activities (< 50%), or irregular alpha rhythm, or excess beta activities with amplitude more than 50 µV, or excess theta activities mainly over the frontal region, or mild excess delta activities; (3) 2 = moderate abnormal: 7-8 Hz alpha rhythm, or no certain occipital alpha rhythm, asymmetry (> 50%), or moderate excess delta activities, or sporadic epileptiform discharge; (4) 3 = severe abnormal: low-voltage or electric silence, or periodic waves, or delta or theta activities are the dominated activities in the background, or rhythmic epileptiform discharge.
EEG data without excessive noise or artifacts from subjects were preprocessed with EEGLAB (R13_6_5b) in MATLAB R2017a. An independent component analysis was used for further artifact removal. Data were band-pass ltered into the range of 0.1-40 Hz and were recomputed against the average reference. Resting state EEG data were split into non-overlapping epochs of two seconds. Patients with less than 25 epochs were excluded. It resulted in 27 patients with AD, 21 with FTD, and 24 age-matched healthy controls (HC) for further analyses. Frequency spectral and EEG coherence analyses were performed in the following frequency bands: delta (1-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (12-30 Hz).

Microstate analysis
The microstate analysis was conducted using the EEGLAB plugin Microstate 0.3 in MATLAB R2017a.
EEG data were further band-pass ltered into the 1-20 Hz range for microstate analysis. The overall variances across all electrodes were quanti ed by measuring the global eld power (GFP). EEG topographies tend to be stable during periods of high GFP [9]. The scalp maps at the momentary peaks of the GFP were extracted and clustered using a modi ed k-means spatial cluster analysis [18]. Previous studies revealed that the optimal number of microstate classes belonged to 2-6 classes (mean 3.7 classes), according to the agglomerative clustering procedure [25][26][27]. The current study used a crossvalidation criterion and the Krzanovski-Lai criterion by the Cartool software [28] to determine the optimal number of microstate classes, testing the entire range of 1 to 12 classes.
The cluster analysis resulted in mean microstate topographies for each class. Group model maps were created based on individual model maps. The resulting class-labeled group microstate maps were then t back to the templates to assign model maps to each participant.
Microstate topographies of each microstate class were compared between groups using a nonparametric randomization test (TANOVA, topographical analysis of variance), as implemented in the Ragu software [29]. Microstate duration (ms), frequency of occurrence of each microstate (/s), the percentage of total analysis time covered by each microstate (%), and transition probabilities were calculated.

Frequency spectral analysis
Frequency spectral analysis was performed using a fast Fourier Transform (FFT, 1000-point) algorithm.
The absolute power spectral density (PSD, dB, 10log 10 (V 2 /Hz)) for each channel based on the periodogram was calculated. Relative PSD (rPSD) was computed by normalizing the total power in the whole frequency range. The absolute and relative PSDs were averaged across channels within groups to obtain a measure for global comparisons between groups in each frequency band.

Statistics analyses
Relatively symmetrical data distribution of microstate, rPSD and absolute PSD was shown in the boxplots (Supplementary 1). Although there were outliers, it intuitively conformed to the normal assumption. Multivariate analysis of variance (MANOVA) was therefore performed to assess group differences of microstate variables. When overall signi cant effects were found, univariate ANOVAs followed by post hoc analyses with Bonferroni correction were performed. A Spearman correlation test was used for the correlation analysis. Continuous non-normal data were examined using a Kruskal-Wallis test for group comparisons. Chi square test was used for group comparison of categorical data. The level of signi cance was set at 0.05. Statistics analyses were performed using IBM SPSS Statistics v22.

Results
Clinical and demographics data between dementia and control groups were presented in Table 1. HC, AD and FTD participants had no difference in age. FTD group had higher proportion of male, and AD group had more females than males. The difference in gender between three groups was not signi cant. FTD group was signi cantly less impaired in terms of MMSE than the AD group (P = 0.015). Additionally, the two dementia groups did not differ signi cantly in terms of dementia duration, ADL, and MoCA. The percentage of patients taking cholinesterase inhibitors (AChEIs) did not differ between the two dementia groups.
CSF biomarker tests were performed in 8 patients with AD and 8 with FTD. As expected, AD group had lower Aβ 42 and higher levels of t-tau and p-tau than FTD group. The differences were not signi cant. The ratio of p-tau to Aβ 42 was shown to be signi cantly higher in AD, compared to FTD (P = 0.028). 44

EEG Microstates
The median optimal number of microstate classes in AD and control groups was four, while the median in FTD was ve. The overall median optimal number in the entire dataset was four. The number of microstate classes was therefore set to four for further analyses, which was also commonly used in most researches, labeled as A, B, C, and D [26,31]. The mean global explained variance (standard deviation, SD) of four microstates in each group was 79.8% (3.3%) for controls, 74.1% (2.3%) for AD, 77.0% (4.4%) for FTD.
Across all microstate classes, the mean microstate duration was 68.1 ms in controls, 75.1 ms in AD patients, and 76.6 ms in FTD patients. The mean duration in dementia groups tended to be increased compared controls, without statistical signi cance ( Table 2). The mean number of unique microstate occurrence per second was reduced in AD, compared to HC (P = 0.0021). Group microstate maps were illustrated in Fig. 1. TANOVAs for each microstate class showed that the AD maps were different from controls maps for class B, FTD maps were different from AD and control maps for class A. There were no signi cant group differences in model map topography for class C and D.
Microstate analysis results were presented in Table 2 and Fig. 2. There're no signi cant differences between AD and FTD groups. Compared to controls, microstate A duration in FTD and microstate B durations in AD were increased. Microstate C occurrence was reduced in both dementia groups compared to controls, with no signi cant difference between AD and FTD groups. Coverage of microstate C was reduced in FTD, compared to controls. The transition to microstate C from A was reduced in AD group, compared to control group (0.081 vs 0.097, P = 0.031). Other microstate transition probabilities were shown no group difference (Supplementary 2).

Relation between microstate and clinical data
We found that the degree of visual EEG abnormality was negatively correlated to MMSE score (r = -0.363, P = 0.025) in AD, and positively correlated to the ratio of t-tau to Aβ 42 (r = 0.756, P = 0.030) and the ratio of p-tau to Aβ 42 (r = 0.756, P = 0.030).
In the AD group, microstate B coverage was negatively correlated to the concentration of CSF Aβ 42 (r = -0.833, P = 0.010), and was positively correlated to the ADL score (r = 0.691, P = 0.006). Additionally, CSF Aβ 42 concentration was negatively related to transition probability from A to B (r = -0.714, P = 0.047), and p-tau concentration was negatively related to transition probability from A to C (r = -0.738, P = 0.037). MMSE score was negatively related to microstate C duration (r = -0.357, P = 0.028), and positively related to microstate A occurrence (r = 0.360, P = 0.026) and contribution A (r = 0.363, P = 0.025).
In the FTD group, the CSF Aβ 42 level was positively related to microstate D occurrence (r = 0.786, P = 0.021) and transition probability from D to A (r = 0.714, P = 0.047). There was a negative correlation between the mean occurrence and CSF t-tau concentration (r = -0.714, P = 0.047). ADL was negatively related to transition probability from D to B (r = -0.886, P = 0.019).
The microstate variables were not signi cantly correlated to MoCA scores, ratios of tau to Aβ 42 , the number of APOE ε4 copies in both groups. The spearman correlations with a relatively high signi cance level (P < 0.040) were illustrated in Fig. 3. The correlations with p-value > 0.040 required a larger sample to be con rmed.

EEG microstate in early and late onset AD
Thirty-nine patients with AD were divided into two subgroups: early onset AD (EOAD, age < 65, n = 30) and late onset AD (LOAD, age > = 65, n = 9). Demographics data and clinical assessment scales were presented in Table 3. There're no signi cant differences between two subgroups in termS of gender, MMSE, MoCA, ADL, and visual EEG score.

Frequency spectral analysis
The across-channel grand average of global EEG PSD in each group was illustrated in Fig. 4A&C. Means of the absolute PSD in control group were higher compared to dementia groups in alpha and beta bands with signi cance (  Fig. 4A). As shown in the Fig. 4C and Table 4, the global relative PSD in dementia groups was signi cantly reduced in alpha and beta bands, and increased in delta bands, compared to controls. The topographies calculated from the global absolute and relative PSDs over frequency bands were illustrated in Fig. 4B&D. The topographies revealed that PSD changes were presented in the whole scalp regions.
There was no signi cant difference between the two dementia groups for both absolute and relative PSDs. The rPSD of three separated scalp regions (anterior: Fp1, Fp2, F3, F4, C3, C4, Fz, Cz; posterior: P3, P4, O1, O2, Pz; temporal: F7, F8, T3, T4, T5, T6) was calculated and compared among groups. The rPSD in each region in dementia groups was markedly reduced in alpha and beta bands and increased in theta band, compared to controls. Moreover, the rPSD level in the temporal region in AD groups was signi cantly higher than rPSD level in FTD groups in delta band (P = 0.034) (Supplement 3).

Connectivity analysis
Mean coherence data in all groups were presented in In the FTD group, alpha and beta coherences for the electrode pair Fp2-F8 were signi cantly decreased (Alpha: P Fp2−F8 = 0.015; Beta, P Fp2−F8 = 0.018), compared to controls. There was a reduction of beta coherence for the Fp2-F4 pair (P Fp2−F4 = 0.043). There're no signi cant coherence differences between AD and FTD groups.

Discussion
The current study investigated microstate's changes, power spectral density, and EEG connectivity in Alzheimer's disease and frontotemporal dementia. The correlation between EEG microstate and clinical severity and CSF biomarkers in the two dementia groups was also analyzed.
Cognitive scores and CSF biomarkers were different between FTD and AD as expected. Previous study reported that FTD was associated with greater impairments in ADLs than AD [32]. We found that the two groups had similar ADL scores, but signi cant higher MMSE score in FTD than AD were revealed. It indicated that FTD needs a higher MMSE score to get the same ADL with AD. The present study revealed lower Aβ 42 levels and higher tau levels in AD than FTD. The ratio of p-tau to Aβ 42 was signi cantly increased in AD compared to FTD. These results are in line with previous studies [5,33].
The visual EEG severity was negatively correlated to the MMSE score. Previous studies reported the positive correlation between visual EEG scores and clinical severity [34,35], which was also con rmed in the present study. We didn't nd the correlation between CSF biomarkers and cognition scales, consistent with earlier study [36]. However, visual EEG scores were positively related to the ratios of t-tau to Aβ 42 and p-tau to Aβ 42 in AD group.
EEG microstate topographies in AD and FTD signi cantly deviate from controls. Microstate B map was different between AD and control, while class A map differed between FTD and control. Previous studies revealed very different results. Two studies revealed no topography differences between AD and controls [19,37], but topographies of classes B and C in semantic dementia, a variant of FTD, were different from maps in control [37]. Schumacher et al. reported that all ve classes (A-E) maps were different between AD and control groups [10]. Another study reported that AD had different topographies of classes A and D compared to control group [15].
Microstate variables changes were also different in FTD and AD. We found increased durations of class B in AD and class A in FTD. Microstate C occurrence was decreased in both dementia groups. Some studies revealed that microstate durations were decreased in patients with dementia or cognitive impairment [18,19,38,39]. However, more recent studies reported increased durations [10,15,16], and reduced occurrences [10,15] in AD. Consistent with the atter researches, increased durations were also demonstrated in our study. The increased duration and reduced occurrence re ect the loss of microstate dynamics, which may be related to the EEG slowing [10].
Moreover, after divided into EOAD and LOAD, marked increased microstate duration and decreased occurrence for class A-C were observed in EOAD, as well as mean duration and occurrence. Earlier studies reported that visual EEG abnormalities were more severe in EOAD [34,40]. Since visual EEG results showed no difference between the two subgroups, our results indicate that loss of microstate dynamics may be more sensitive than visual EEG slowing.
Microstate B was signi cantly different in AD for topography and duration, and correlated to CSF Aβ 42 and ADL score, with a high spearman's rank coe cient. These class B alterations were not presented in FTD. Previous studies revealed that microstate B was correlated with the bilateral occipital cortex [14]. In our study, EEG connectivity analyses demonstrated that fronto-occipital far coherence in alpha and beta bands were reduced in AD compared to controls, which was not detected in FTD group. AD patients have more atrophy in the occipital gyrus and precuneus than FTD patients [41]. The reduced fronto-occipital far coherence and prefer occipital atrophy may partially explained the difference of microstate B alteration between AD and FTD. For microstate A change in FTD, class A was correlated with superior and middle temporal lobe [14], consistent with the frontotemporal pathologic abnormalities in FTD.
Microstate correlation analysis demonstrated that microstate C duration was negatively correlated to MMSE score, while a positive correlation was revealed in microstate A occurrence and MMSE score in AD. We also found that CSF biomarkers were related to microstate. Aβ 42 level was related to microstate B coverage positively in AD and to microstate D occurrence negatively in FTD. CSF Aβ 42 and tau have high diagnostic accuracy. The correlation between biomarkers and EEG microstate and visual scores indicated that EEG could be a potential diagnostic method for dementia. Since EEG is a noninvasive and convenient examination, the diagnostic value of EEG for dementia is worthy of further work.
The spectral analysis demonstrated that FTD and AD had lower rPSD in alpha and beta bands, higher rPSD in delta bands, indicating the general EEG was slowing. It further suggests that loss of microstate dynamics may be attributed to EEG slowing. A diffuse slowing with reduction of power in faster rhythm and increased power in slow rhythm have been observed in AD [42,43] and FTD [44]. But the power in delta band was lower in FTD than AD [44,45]. The rPSD in slow rhythm tended to be lower in FTD than AD in the current study, with no signi cance. But a signi cant temporal delta power increase was detected in AD compared to FTD, indicating functional temporal impairment may be more severe in AD.
Our study didn't evaluate memory function but revealed that AD had lower MMSE than FTD.
Finally, we found that AD had more reduced coherence, compared to FTD. Fronto-temporal coherence in alpha and beta bands was reduced in the left hemisphere in FTD, but bilateral sides in AD. Moreover, AD had reduced fronto-central, fronto-occipital and temporal-central connectivity in alpha, beta and delta bands, compared to controls. The previous study demonstrated that major coherence reductions were in the alpha band, while delta coherence results were con icting [8]. Caso et al. revealed decreased fast rhythm values in central/temporal regions in AD, compared to FTD, by use of sLORETA [45]. The ndings in our study were consistent with the earlier studies. The frontal lobe and corpus callosum were vulnerably damaged [46], which may partially explained reduced connectivity to frontal and central areas in AD.

Limitations
The present study has some limitations. First, part of patients with dementia were taking AChEIs which can in uence EEG data [47]. There was no difference in the number of patients taking AChEIs between FTD and AD. However, group comparisons between dementia and control groups may be required further work to con rm without AChEI in uence. In addition, the sample size of patients who had CSF biomarkers was small. Therefore, the correlation analysis results with low spearman's rank coe cient and signi cance level were not strong enough. A larger sample will draw more convincing conclusions.

Conclusions
The current study demonstrated that EEG data of AD and FTD were different by use of microstate analysis, spectral analysis, and coherence analysis. More EEG changes and more areas involved were detected in AD, compared to FTD.

Consent for publication
Written informed consent was obtained from all the participants for the publication of this article.

Competing Interest
The authors declare that they have no competing interests. Availability of data and materials The data used during the current study are available from the corresponding authors on reasonable request.
Authors' contributions NL analyzed and interpreted the data, wrote the original manuscript. JG analyzed the data and diagnosed the patients, revised the manuscript. QL and CLY conceived the study, revised the manuscript. CHM acquired the data, diagnosed the patients. HYS performed EEG recordings, analyzed visual EEG data. All authors read and approved the nal manuscript.   Frequency spectral analysis. Across-channels grand average of absolute power spectral density (PSD, dB) and relative PSD (%µV2Hz-1) over with frequency for each group are illustrated. The topographies were calculated from PSD in each frequency band. A general slowing EEG was presented in both FTD and AD groups. FTD: frontotemporal dementia; AD: Alzheimer's disease; HC: healthy controls.