Abnormalities of Resting State Cortical EEG Rhythms in Subjects with Mild Cognitive Impairment Due to Alzheimer’s and Lewy Body Diseases

Subjects

In the present retrospective exploratory study, we used the rsEEG data of an international archive, formed by clinical, neuropsychological, and electrophysiological data in 30 Nold, 30 ADMCI, and 23 DLBMCI subjects matched for relevant demographic variables. These subjects were recruited outside a formal multicenter clinical trial by the following qualified clinical recording units of the informal European PDWAIVE Consortium: University of Rome “La Sapienza” (Italy), IRCCS Fatebenefratelli of Brescia (Italy); IRCCS SDN of Naples (Italy); IRCCS Oasi of Troina (Italy); University of Genova (Italy); Hospital San Raffaele of Cassino (Italy); IRCCS Hospital San Raffaele Pisana of Rome (Italy); and, University “G. d’Annunzio” of Chieti and Pescara (Italy). General Hospital of Linz (Austria); Dokuz Eylul University (Turkey); Istanbul Medipol University (Turkey); and University of Basel (Switzerland). The three groups of subjects were carefully matched for age, gender, and education. The ADMCI and DLBMCI groups were also carefully matched for the Mini-Mental State Examination (MMSE) score as a measurement of the global cognitive status [30].

Table 1 summarizes the relevant demographic and clinical (MMSE score) data of the Nold, ADMCI, and DLBMCI groups, together with the results of the statistical analyses computed to evaluate the presence or absence of statistically significant differences between the groups for the age (ANOVA), gender (Fisher-Freeman-Halton test), education (ANOVA), and MMSE score (Kruskal-Wallis test). As expected, a statistically significant difference was found among the Nold and the other two groups for the MMSE score (H = 34.7; p < 0.00001). Specifically, there was a higher MMSE score in the Nold than the ADMCI and DLBMCI groups (p < 0.0001). On the contrary, a statistically significant difference was found neither for the MMSE score between the ADMCI and the DLBMCI groups nor the age, gender, and education among the three groups (p > 0.05).

Local institutional Ethics Committees approved data sharing for scientific purpose. All experiments were performed with the informed and overt consent of each participant or caregiver, in line with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and the standards established by the local Institutional Review Board.

Diagnostic criteria

The inclusion criteria for the enrollment of the ADMCI patients were age between 55 and 90 years, complaints of memory deficits by the patient (and confirmed by a relative) or a relative, MMSE score ≥24, overall Clinical Dementia Rating (CDR) [31] score of 0.5, score on the logical memory test of 1.5 standard deviation (SD) lower than the age-adjusted mean, 15-item Geriatric Depression Scale (GDS) [32] score ≤5, modified Hachinski ischemia [33] score ≤4, and at least 5 years of education. As this retrospective study was based on data of several clinical units that did not follow a harmonized protocol, there was a jeopardized availability of the criteria of MCI status. In particular, the MCI status could be single or multidomain. The status of ADMCI was based on the positivity to one or more of the following biomarkers: Aβ_1-42/phospho-tau in the CSF, fluorodeoxyglucose PET (FDG-PET) mapping, and structural magnetic resonance imaging (MRI) [34]. Exclusion criteria were other significant neurological, systemic, or psychiatric illness, enrolment in a clinical trial with experimental drugs, major depression disorders described in the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), high dose of neuroleptics or chronic sedatives or hypnotics, antiparkinsonian medication and the use of narcotic analgesics. Of note, the use of cholinesterase inhibitors and memantine was allowed.

All ADMCI subjects underwent a battery of neuropsychological tests to evaluate the status of MCI. This battery included neuropsychological tests assessing the general cognitive performance in the domains of memory, language, executive function/attention, and visuoconstruction abilities. Specifically, the tests assessing memory included the delayed recall of Rey figures [35] and/or the delayed recall of a story [36]. The tests assessing language included the 1-min verbal fluency for letters, fruits, animals, or car trades [37] and/or the Token test [38]. The tests assessing executive function and attention included the Trail Making Test Part A and B [39]. Finally, the tests assessing visuoconstruction included the copy of Rey figures.

The diagnosis of the probable DLB was carried out in agreement with the consensus guidelines [3, 4]. Twenty-two out of 23 DLBMCI patients performed DaTSCAN to confirm the diagnosis of probable DLB. Concerning the detection of the core and suggestive features of DLB, the Neuropsychiatric inventory item-2 investigated the occurrence frequency and the severity of hallucinations [40]. Frontal Assessment Battery [41], and Clinician Assessment of Fluctuations [18, 19] were included to investigate, respectively, the severity of the frontal dysfunction and the presence and severity of the cognitive fluctuations. Unified Parkinson Disease Rating Scale-III assessed the presence and severity of the extrapyramidal signs [42]. The presence and/or absence of rapid eye movement sleep behavior disorder was determined according to minimal International Classification of Sleep Disorders criteria (1992) [43]. As this retrospective study was based on data of several clinical units that did not follow a harmonized protocol, the DLBMCI subjects underwent a different battery of clinical scales including the Neuropsychiatric Inventory [40], the scale for the assessment of Behavioral and Psychological Symptoms of Dementia, the MMSE, the Dementia Rating Scale-2 [44], the Epworth Sleepiness Scale for estimating subjective sleep disturbances, and the Alzheimer’s Disease Cooperative Study for the Activities of Daily Living. Furthermore, DLBMCI subjects underwent different battery of neuropsychological tests to evaluate the status of MCI [5]. This battery included neuropsychological tests assessing the general cognitive performance in the domains of memory, language, executive function/attention, and visuoconstruction abilities (some of them received the CERAD-plus battery).

The diagnosis of PD was based on a standard clinical assessment of tremor, rigidity, bradykinesia, and postural instability without major cognitive deficits for 12 months [45]. As measures of severity of the motor disability, the Hoehn and Yahr stage [46] and the Unified Parkinson Disease Rating Scale-III [42] for extrapyramidal symptoms, were used. The diagnosis of PDMCI was based on the Diagnostic Criteria for Mild Cognitive Impairment in Parkinson’s Disease [47]. The inclusion criteria comprised: 1) a diagnosis of PD as based on the UK PD Brain Bank Criteria [45]; 2) a gradual decline, in the context of an established PD, in the cognitive status reported by either the patient or informant, or observed by the clinicians; 3) cognitive deficits not sufficient to interfere significantly with functional independence in the activities of the daily life, although slight difficulties on complex functional tasks may be present. On the basis of clinical features and neuroradiological findings, the exclusion criteria for PDMCI included the following forms of parkinsonism: 1) dementia with Lewy bodies [3, 49], 2) drug-induced parkinsonism, 3) cerebrovascular parkinsonism, and 4) atypical parkinsonism with absent or minimal responses to antiparkinsonian drugs [48, 49].

All Nold subjects underwent a cognitive screening (including MMSE and GDS) as well as physical and neurological examinations to exclude any dementia or major cognitive deficit. No Nold subject referred subjective cognitive impairment. The subjects affected by chronic systemic illnesses (e.g., diabetes mellitus) were excluded, as were the subjects receiving chronic psychoactive drugs. The subjects with a history of previous or present neurological or psychiatric disease were also excluded. All Nold subjects had a GDS score lower than the threshold of 5 (no depression) or no depression after an interview with a physician or clinical psychologist.

The use of psychoactive drugs was allowed in both ADMCI and DLBMCI subjects. Specifically, 12 out of 30 (40%) ADMCI patients and 10 out of 23 (43.48%) DLBMCI patients used psychoactive drugs. No Nold subject used psychoactive drugs. In detail, 36.67% of the ADMCI patients and 39.13% of the DLBMCI patients used antianxiety and antidepressant drugs, while 3.33% of the ADMCI patients and 17.39% of the DLBMCI patients used psychotic drugs. When given, the psychoactive drugs were suspended for about 24 h before EEG recordings to harmonize the time of the last administration of the drug before those recordings across patients. A statistical analysis (Fisher test) was computed to evaluate the presence or absence of statistically significant differences between the ADMCI and DLBMCI groups as the consumption of the psychoactive drugs. Furthermore, the consumption of the psychoactive drugs was used as a covariate in the main rsEEG statistical analysis, to exclude that some differences in the consumption of psychoactive drugs may influence rsEEG statistical differences between the ADMCI and the DLBMCI group.

Table 2 reports the number and the percentages of the ADMCI and DLBMCI patients assuming the psychoactive drugs for mental disorders (e.g., sedative, anxiety, antidepressant, antipsychotic), before the EEG recordings. The psychoactive drugs were categorized in two classes: the first class included anxiety and antidepressant drugs, while the second class included antipsychotic drugs. Some patients assumed more than one psychoactive drug. Furthermore, the Table 2 reports the number and the percentages of the ADMCI and DLBMCI patients who took others drugs (e.g., AChEIs, and drugs for extrapyramidal symptoms and dementia). All types of drugs are reported in detail in the Table 3.

EEG recordings

EEG data were recorded while the subjects were sitting comfortably with eyes closed in a standard resting state condition (rsEEG). At least 5 min of rsEEG data were recorded (128 Hz or higher sampling rate, with a bandpass between 0.01 Hz and 100 Hz) from a minimum number of 19 exploring scalp electrodes positioned over the whole scalp according to the 10–20 System (i.e., Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, and O2). Linked earlobe reference electrode was preferred, but not mandatory to respect the methodological facilities and standard internal protocols of the clinical recording units. A ground electrode was typically located between the AFz and Fz electrodes, and electrodes impedances were kept below 5 Kohm. Horizontal and vertical electro-oculographic activities (0.3–70 Hz bandpass) were also recorded to monitor blinking and eye movements. The EEG recordings were performed, in all subjects, in the late morning to minimize drowsiness. Furthermore, an operator controlled on-line the subject and the EEG traces to keep constant the level of vigilance.

For all Nold subjects, the clinical diagnosis (i.e., intact cognitive status) and the EEG recording were performed on the same day. Instead, the mean time from the clinical diagnosis of MCI to the EEG recording was 1.2 months (±0.3 standard error mean, SE) in the ADMCI patients and 19.5 months (±4.2 SE) in the DLBMCI patients. Furthermore, the mean time from the clinical diagnosis of MCI to the last follow-up with the diagnosis of dementia or MCI was 22.9 months (±1.4 SE) in the ADMCI patients and 49.4 months (±7.2 SE) in the DLBMCI patients. Finally, 6 out of 30 (20%) ADMCI patients and 13 out of 23 (53.3%) DLBMCI patients converted to the dementia at the last follow-up. Statistical analyses were computed to evaluate the presence or absence of statistically significant differences between the MCI groups as the time from the clinical diagnosis of MCI to the EEG recording (T-test), the time from the clinical diagnosis to the last follow-up with the diagnosis of dementia or MCI (T-test), and the conversion from MCI to dementia (Fisher test). Statistically significant differences were found between the two MCI groups (i.e., ADMCI and DLBMCI) as the time from the clinical diagnosis of MCI to the EEG recording (p < 0.00005; longer time for the DLBMCI than ADMCI group), the time from the clinical diagnosis of the last follow-up with the diagnosis of dementia or MCI (p < 0.001; longer time for the DLBMCI than ADMCI group), and the percentage of conversion from MCI to dementia (i.e., DLB or AD; p < 0.01; higher converted subjects in the DLBMCI than ADMCI group). Therefore, the mentioned three variables were used as covariates in the main rsEEG statistical analyses, to minimize their effects on the statistical findings.

Preliminary analysis of the EEG data

The recorded rsEEG data were band-passed to avoid aliasing, down-sampled to 128 Hz (when recorded with higher sampling frequency), segmented in consecutive 2-s epochs, and analyzed off-line. We rejected the rsEEG epochs associated with operator’s markers indicating drowsiness, verbal warnings, eyes opening, arm/hand movements, or other events (e.g., sweat, sway, head movements, etc.) disturbing the EEG recordings. Furthermore, the rsEEG epochs with ocular (e.g., rapid eye opening despite the request to maintain the eyes closed), muscular, and other types of artifacts were preliminarily identified by an automatic computerized procedure. The rsEEG epochs with sporadic and well-shaped blinking artifacts were corrected from the EOG activity by an autoregressive method [50]. Two independent experimenters (blind to the diagnosis at the time of the rsEEG analysis) manually revised the rsEEG epochs accepted for further analysis. The rsEEG epochs with signs of a sleep intrusion (an ongoing increase of theta, K complex, spindles, etc.) were rejected. To harmonize the rsEEG data collected with different reference electrodes, all artifact-free rsEEG epochs were re-referenced to the common average for further analysis.

Spectral analysis of the rsEEG epochs

A standard digital FFT-based power spectrum analysis (Welch technique, Hanning windowing function, no phase shift) computed the power density of the 2-s rsEEG epochs with 0.5 Hz of frequency resolution. This standard FFT procedure was implemented in a software script developed under Matlab 6.5 (Mathworks Inc., Natick, MA).

According to a previous study of our group [29], the frequency bands of interest were individually identified based on the following frequency landmarks: the transition frequency (TF) and the individual alpha frequency peak (IAF). In the EEG power density spectrum, the TF marks the transition frequency between the theta and alpha bands, defined as the minimum of the rsEEG power density between 3 and 8 Hz (between the delta and the alpha power peak). IAF was defined as the maximum power density peak between 6 and 14 Hz. In precedence, these frequency landmarks were well described by Klimesch et al. [51–53].

The TF and IAF were computed for each subject involved in the study. Based on the TF and IAF, we estimated the frequency band range for each subject as follows: delta from TF -4 Hz to TF -2 Hz, theta from TF -2 Hz to TF, low-frequency alpha band (alpha 1 and alpha 2) from TF to IAF, and high-frequency alpha band (or alpha 3) from IAF to IAF + 2 Hz. The other bands were defined based on standard fixed frequency ranges: beta 1 from 14 to 20 Hz, beta 2 from 20 to 30 Hz, and gamma from 30 to 40 Hz. The alpha 1 and alpha 2 bands were computed for each subject as follows: alpha 1 from TF to the midpoint of the TF-IAF range and alpha 2 from this midpoint to IAF.

Cortical sources of rsEEG epochs as computed by eLORETA

We used the freeware tool called “exact LORETA” (eLORETA) for the linear estimation of the cortical source activity generating scalp-recorded rsEEG rhythms [27].

The present implementation of eLORETA uses a head volume conductor model composed of the scalp, skull, and brain. In the scalp compartment, exploring electrodes can be virtually positioned to give EEG data as an input to the source estimation [27]. The brain model is based on a realistic cerebral shape taken from a template typically used in the neuroimaging studies, namely that of the Montreal Neurological Institute (MNI152 template) [54]. The eLORETA solves the so-called EEG inverse problem estimating “neural” current density values at any cortical voxel of the mentioned head volume conductor model. The solutions are computed rsEEG frequency bin-by-frequency bin.

The input for this estimation is the EEG spectral power density computed at 19 scalp electrodes. The output is the electrical brain source space formed by 6,239 voxels with 5 mm resolution, restricted to cortical gray matter of the head volume conductor model [37]. An equivalent current dipole is located in each voxel. For each voxel, the eLORETA package provides the Talairach coordinates, the lobe, and the Brodmann area (BA).

In line with the general low spatial resolution of the present EEG methodological approach (i.e., 19 scalp electrodes), we performed a regional analysis of the eLORETA solutions. For this purpose, we collapsed the eLORETA solutions within frontal, central, temporal, parietal, occipital, and limbic macroregions (ROIs) considered separately. Table 4 reports the list of the BAs used for the ROIs considered in the present study. Of note, the main advantage of the regional analysis of eLORETA solutions was that we could disentangle the rsEEG source activity in contiguous cortical areas. For example, the rsEEG source activity in the occipital ROI was disentangled from that estimated in the parietal and temporal ROIs, etc. This was made possible by the fact that eLORETA solves the linear inverse problem by taking into account (at least in part) the effects of the head as a volume conductor. In contrast, the solutions of rsEEG power density computed at a parietal scalp electrode reflect the contribution of source activities not only of the underlying parietal cortex but also of surrounding occipital and temporal cortices.

For the present eLORETA cortical source estimation, a frequency resolution of 0.5 Hz was used, namely, the maximum frequency resolution allowed by the use of 2-s artifact free EEG epochs. The frequency bands of interest were delta, theta, alpha 1, alpha 2, alpha 3, beta 1, beta 2, and gamma, which were defined subject-by-subject as described above.

Statistical analysis of the eLORETA solutions

A statistical session was performed by the commercial tool STATISTICA 10 (StatSoft Inc., http://www.statsoft.com) to test the hypothesis that the rsEEG source activity as revealed by eLORETA solutions might differ between the ADMCI and DLBMCI groups using the Nold group as a control reference. To this aim, an ANOVA was computed using the regional normalized eLORETA solutions (normalized current density at all voxels of a given ROI) as a dependent variable (p < 0.05). The ANOVA factors were Group (Nold, ADMCI, and DLBMCI), Band (delta, theta, alpha 1, alpha 2, alpha 3, beta 1, beta 2, and gamma), and ROI (frontal, central, parietal, occipital, temporal, and limbic). The TF, IAF, the conversion from MCI to dementia (i.e., DLB or AD), the time from the clinical diagnosis of MCI to the EEG recording, the time from the clinical diagnosis to the last follow-up with the diagnosis of dementia or MCI, and the consumption of psychoactive drugs (antidepressant, antianxiety, and antipsychotic drugs) were used as covariates. In particular, we used two different covariates concerning the use of psychoactive drugs: the first covariate included the information about the use of antidepressant and antianxiety drugs, while the second covariate included the information about the use of antipsychotic drugs. The use of two different covariates for the psychoactive drugs was motivated by the equal distribution of the antianxiety and antidepressant drugs in the ADMCI and DLBMCI (e.g., 36.67% versus 39.13%) patients before the EEG recordings and a wider use of psychotic drugs in the DLBMCI group compared with the ADMCI group (17% versus 3%).

The degrees of freedom were corrected by the Greenhouse-Geisser procedure when appropriate. Duncan test was used for post-hoc comparisons (p < 0.05).

The planned post-hoc testing also evaluated the above prediction about the differences in the rsEEG source solutions between the ADMCI and DLBMCI groups using the Nold group as a control reference. Specifically, we predicted: 1) a statistical 3-way interaction effect including the factors Group, ROI, and Band (p < 0.05); 2) a post-hoc test indicating statistically significant differences of the regional normalized eLORETA solutions with the pattern Nold¬ =ADMCI¬ =DLBMCI (Duncan test, p < 0.05).

The above statistical analyses were controlled by the Grubbs test (p < 0.005) for the presence of outliers in the distribution of the eLORETA source solutions.

Accuracy of the rsEEG source activity in the discrimination among the Nold, ADMCI, and DLBMCI individuals

The rsEEG sources showing statistically significant differences among the three groups were used as discriminant (not diagnostic as the abnormalities in those sources were not necessarily disease-specific) variables for the following classification trials: 1) the Nold versus the ADMCI individuals; 2) the Nold versus the DLBMCI individuals; and 3) the ADMCI versus DLBMCI individuals. The correct blind classifications of these rsEEG source activities were performed by GraphPad Prism software (GraphPad Software, Inc., California, USA) for the production of the receiver operating characteristic (ROC) curves [55]. The following indexes measured the classification performance of the above binary classification: 1) Sensitivity. It measures the rate of the positives who were correctly classified as positives (i.e., “true positive rate” in the signal detection theory); 2) Specificity. It measures the rate of the negatives (control) who were correctly classified as negatives (i.e., “true negative rate” in the signal detection theory); 3) Accuracy. It is the mean between the sensitivity and specificity (the amount of subjects in the groups was the same); and 4) Area under the ROC curve (AUROC). The AUROC was another standard index of the global classification accuracy.

Statistical analysis of the EEG cortical sources

Table 5 reports the mean values of the TF and IAF for the three groups (i.e., Nold, ADMCI, and DLBMCI), together with results of the statistical comparisons between the group pairs (ANOVA). The mean TF was 6.3 Hz (±0.2 standard error mean, SE) in the Nold subjects, 5.4 Hz (±0.2 SE) in the ADMCI subjects, and 4.7 Hz (±0.2 SE) in the DLBMCI subjects. The mean IAF was 9.4 Hz (±0.1 SE) in the Nold subjects, 8.8 Hz (±0.3 SE) in the ADMCI patients, and 7.8 Hz (±0.3 SE) in the DLBMCI patients. ANOVAs were computed to evaluate the presence or absence of statistically significant differences between the three groups for both TF and IAF (p < 0.05). The results showed the following statistically significant effects: 1) the mean TF was greater (F = 16.6, p < 0.00001) in the Nold than the ADMCI (p < 0.005) and DLBMCI (p < 0.00005) groups; it was also higher in the ADMCI than the DLBMCI group (p < 0.01); 2) the mean IAF was greater (F = 16.6, p < 0.00001) in the Nold than the ADMCI (p < 0.05) and DLBMCI (p < 0.0005) groups; it was also higher in the ADMCI than the DLBMCI group (p < 0.005). These results confirm the importance of the determination of the individual frequency bands in the comparison of rsEEG biomarkers in ADMCI and DLBMCI patients.

Fig. 1 shows the grand average of the regional eLORETA solutions for the rsEEG source estimation relative to a statistically significant ANOVA 3-way interaction effect (F = 6.85; p < 0.00001) among the factors Group (Nold, ADMCI, and DLBMCI), Band (delta, theta, alpha 1, alpha 2, beta 1, beta 2, and gamma), and ROI (frontal, central, parietal, occipital, temporal, and limbic). In the figure, the eLORETA solutions had the shape of typical rsEEG relative power spectra. Notably, the profile and magnitude of the rsEEG source activity spectra in the Nold, ADMCI, DLBMCI groups differed across the ROIs, supporting the idea that the scalp EEG rhythms were generated by a distinct pattern of cortical source activity in those groups. The Duncan planned post-hoc testing showed that the discriminant source pattern ADMCI <DLBMCI<Nold was fitted by the occipital, temporal, and limbic alpha 2 sources as well as the occipital alpha 3 sources (p < 0.01 to p < 0.000001). Compared to the Nold group, these posterior alpha source activities showed an abnormal reduction in the ADMCI and DLBMCI groups (p < 0.005 to p < 0.000001). Furthermore, these activities they were lower in the ADMCI than the DLBMCI group (p < 0.01 to p < 0.000005). Of note, 3 ADMCI and 1 DLBMCI subjects exhibited an asymptotic rsEEG power spectra without alpha peak. On the contrary, the pattern DLBMCI >ADMCI >Nold was fitted by the frontal, parietal, and temporal delta sources (p < 0.05 to p < 0.000001). Compared to the Nold group, these delta source activities pointed to an abnormal increment in the ADMCI and DLBMCI groups (p < 0.05 to p < 0.000001). Furthermore, they were greater in the DLBMCI than the ADMCI group (p < 0.001 to p < 0.00005).

OPEN IN VIEWER

Fig.1

Regional normalized eLORETA solutions (mean across subjects) of the rsEEG rhythms relative to a statistical 3-way ANOVA interaction between the factors Group (Nold, ADMCI, and DLBMCI), Band (delta, theta, alpha 1, alpha 2, alpha 3, beta 1, beta 2 and gamma), and ROI (central, frontal, parietal, occipital, temporal, and limbic). This ANOVA design used the regional normalized eLORETA solutions as a dependent variable. Subjects’ transition frequency (TF) and individual alpha frequency peak (IAF) were used as covariates. Regional normalized eLORETA solutions modeled the cortical sources of the rsEEG relative power spectra at the scalp electrodes. These sources can be considered as a sort of “virtual” intracranial macro-electrodes located on the macro-cortical regions of interest. See the Methods for a definition of the TF and IAF. The rectangles indicate the cortical regions and frequency bands in which the eLORETA solutions presented statistically significant eLORETA patterns as in the following: Nold¬ =ADMCI¬ =DLBMCI (Duncan post hoc test, p < 0.05).

These results were confirmed in the following ANOVAs performed for control purposes. The first control analysis was performed in matched subgroups of ADMCI (n = 15) and DLBMCI (n = 13) patients who had never taken psychoactive drugs before EEG recordings. A matched Nold group (n = 17) served as a control reference (Table 6 reports all demographic data and global cognitive status as revealed by the MMSE score). The hypothesis was that the main findings of the present study may be observed even in these subgroups of ADMCI and DLBMCI patients free from the effects of psychoactive drugs.

These results were cross-validated by the following three statistical sessions performed for control purposes. In the first statistical session, the TF and IAF were compared across the ADMCI, DLBMCI, and Nold subgroups. The subgroups of ADMCI (n = 15) and DLBMCI (n = 13) patients had never taken psychoactive drugs before EEG recordings. The matched Nold subgroup (n = 17) served as a control reference. The hypothesis was that the main findings of the present study may be observed even in these subgroups of ADMCI and DLBMCI patients free from the effects of psychoactive drugs. Data analysis showed that the mean TF was 5.4 Hz (±0.3 standard error, SE) in the ADMCI patients and 4.6 Hz (±0.2 SE) in the DLBMCI patients. Furthermore, the mean IAF was 8.8 Hz (±0.4 SE) in the ADMCI patients and 7.7 Hz (±0.4 SE) in the DLBMCI patients. In the Nold group, the mean TF and IAF were 6.4 Hz (±0.3 SE) and 9.3 Hz (±0.2 SE), respectively. These mean values were statistically compared between the groups by ANOVA designs (p < 0.05). Results showed that the pattern Nold > ADMCI and DLBMCI was confirmed with both variables, namely TF (F = 9.38, p < 0.0005) and IAF (F = 5.75, p < 0.01). These findings confirmed those observed in the main analysis performed in the whole ADMCI and DLBMCI groups including some patients assuming psychoactive drugs.

In the second statistical session, the clinical diagnosis of MCI to the EEG recording (T-test, p < 0.05), the time from the clinical diagnosis to the last follow-up with the diagnosis of dementia or MCI (T-test, p < 0.05), and the conversion from MCI to dementia (Fisher test, p > 0.05) were compared across the mentioned ADMCI and DLBMCI subgroups. The mean time from the clinical diagnosis of MCI to the EEG recording was 1.5 months (±0.5 SE) in the ADMCI patients and 17 months (±5.8 SE) in the DLBMCI patients. Furthermore, the mean time from the clinical diagnosis to the last follow-up with the diagnosis of dementia or MCI was 18.9 months (±2.5 SE) in the ADMCI patients and 42.4 months (±10 SE) in the DLBMCI patients. Finally, 4 out of 15 (27%) ADMCI subjects and 8 out of 13 (62%) DLBMCI patients converted to dementia at the last follow-up. Results of the statistical comparisons between the subgroups showed no statistically significant difference between the two patients’ subgroups (p > 0.05).

In the third statistical session, an ANOVA design compared the eLORETA source solutions in the ADMCI (n = 15), DLBMCI (n = 13), and Nold (n = 17) subgroups (p < 0.05). The ANOVA factors were Group (ADMCI, DLBMCI, Nold), Band (delta, theta, alpha 1, alpha 2, alpha 2, beta 1, beta 2, and gamma), and ROI (frontal, central, temporal, parietal, limbic, and occipital). Results showed a statistically significant 3-way interaction effect among Group, Band, and ROI (F = 4.06, p = 0.00001) as in the main statistical analysis. The Duncan planned post-hoc testing pointed to: 1) a discriminant source pattern ADMCI < DLBMCI <Nold fitted by the occipital and temporal alpha 2 sources as well as the occipital alpha 3 sources (p < 0.05 to p < 0.000001); 2) the discriminant source pattern DLBMCI >ADMCI >Nold was fitted by the parietal and occipital delta sources (p < 0.05 to p < 0.000001). Again, these findings confirmed those observed in the main analysis performed in the whole ADMCI and DLBMCI groups including some patients assuming psychoactive drugs (see Fig. 2).

OPEN IN VIEWER

Fig.2

Regional normalized eLORETA solutions (mean across subjects) of the rsEEG rhythms relative to a statistical 3-way ANOVA interaction between the factors Group (17 Nold, 15 ADMCI, and 13 DLBMCI, who had never taken drugs before EEG recordings), Band (delta, theta, alpha 1, alpha 2, alpha 3, beta 1, beta 2 and gamma), and ROI (central, frontal, parietal, occipital, temporal, and limbic). The rectangles indicate the cortical regions and frequency bands in which the eLORETA solutions presented statistically significant eLORETA patterns as in the following: Nold¬ =ADMCI¬ =DLBMCI (Duncan post hoc test, p < 0.05).

The last three ANOVAs were performed in other subgroups of ADMCI and DLBMCI patients. In this case, the subgroups of ADMCI and DLB patients were paired as the conversion from MCI to dementia (i.e., DLB or AD), the time from the clinical diagnosis of MCI to the EEG recording, and the time from the clinical diagnosis to the last follow-up with the diagnosis of dementia or MCI. In all 3 ANOVAs, the Nold group of 30 seniors was used as a control reference.

The ANOVA in the subgroups paired as the conversion from MCI to dementia (16 ADMCI, 16 DLBMCI; 6 out of 16 ADMCI and DLBMCI patients converted to the demented patients) showed a statistically significant 3-way interaction effect among Group, Band, and ROI (F = 12.2, p < 0.0001). The Duncan planned post-hoc testing showed that: (i) the discriminant source pattern ADMCI <DLBMCI <Nold was fitted by the parietal, occipital, and limbic alpha 2 sources as well as the occipital alpha 3 sources (p < 0.05 to p < 0.000001) and (ii) the discriminant source pattern DLBMCI >ADMCI >Nold was fitted by the frontal, parietal, and temporal delta sources (p < 0.05 to p < 0.000001).

Moreover, the ANOVA in the subgroups paired as the time from the clinical diagnosis of MCI to the EEG recording (12 ADMCI, 12 DLBMCI; the mean time was 2.6 months in the ADMCI patients and 3.4 months in the DLBMCI patients) unveiled a statistically significant 3-way interaction effect among Group, Band, and ROI (F = 8.1, p < 0.0001). The Duncan planned post-hoc testing showed that: 1) the discriminant source pattern ADMCI <DLBMCI <Nold was fitted by the parietal, occipital, and limbic alpha 2 sources as well as the occipital alpha 3 sources (p < 0.05 to p < 0.000005); 2) the discriminant source pattern DLBMCI >ADMCI >Nold was fitted by the parietal and temporal delta sources (p < 0.05 to p < 0.000005).

Finally, the ANOVA in the subgroups paired as the time from the clinical diagnosis to the last follow-up with the diagnosis of dementia or MCI (15 ADMCI, 15 DLBMCI; the mean time was 26.9 months in both ADMCI and DLBMCI patients) pointed to a statistically significant 3-way interaction effect among Group, Band, and ROI (F = 10.1, p < 0.0001). The Duncan planned post-hoc testing showed that: 1) the discriminant source pattern ADMCI <DLBMCI <Nold was fitted by the parietal, occipital, and limbic alpha 2 sources as well as the occipital alpha 3 sources (p < 0.05 to p < 0.000005); and 2) the discriminant source pattern DLBMCI >ADMCI >Nold was fitted by the frontal, temporal, and limbic delta sources (p < 0.05 to p < 0.000005).

A control statistical analysis (Grubbs’ test, p < 0.0001) was performed to verify that the inter-group differences in the above seven rsEEG source activities (i.e., frontal, parietal, temporal delta; occipital, temporal, and limbic alpha 2; occipital alpha 3) were not merely due to the presence of some outliers in the individual eLORETA solutions. No outlier was detected (see Fig. 3), thus confirming the results of the main statistical analysis.

OPEN IN VIEWER

Fig.3

Individual values of the eLORETA cortical source activity showing statistically significant (p < 0.05) differences between the Nold, ADMCI, and DLBMCI groups in the main statistical analysis (i.e., frontal, parietal, and temporal delta; occipital, temporal, and limbic alpha 2; occipital alpha 2; see Fig. 1 for the specific rsEEG source activities showing statistically significant results). Noteworthy, the Grubbs’ test showed no outliers from those individual values of the eLORETA solutions (arbitrary threshold of p < 0.0001).

Correlation between the rsEEG source activity and MMSE

As a first analysis on the clinical relevance of the main results, Spearman test evaluated the monotonic correlation between the above seven rsEEG source activities (i.e., frontal, parietal, temporal delta; occipital, temporal, and limbic alpha 2; occipital alpha 3) and the MMSE score, as a rough index of global cognition status. The correlation analysis was performed using all Nold, ADMCI, and DLBMCI individuals as a whole group for two reasons. On the one hand, the hypothesis was that those rsEEG source activities were correlated with the global cognitive status in humans in general, namely including cases with both normal and impaired cognitive functions. On the other hand, the correlation study would have had a low statistical sensitivity if performed only in the separate groups, due to the very limited scatter of the MMSE scores within a given group (e.g., MMSE score from 30 to 28 in Nold subjects). To take into account the inflating effects of repetitive univariate tests, the statistical threshold was set at p < 0.007 to obtain the Bonferroni’s correction at p < 0.05. A statistically significant negative correlation was found between the activity of the parietal delta source and the MMSE score (r = –0.32, p = 0.003). The higher the parietal delta source activity, the lower the MMSE score. A statistically significant positive correlation was found between the MMSE score and the activity of the occipital alpha 2 (r = 0.38, p = 0.0003), limbic alpha 2 (r = 0.35, p = 0.001), and occipital alpha 3 (r = 0.37, p = 0.0005) sources (see Fig. 4). The higher the posterior alpha source activity, the higher the MMSE score. As a control analysis, the same correlation test for any single group showed no statistically significant result, possibly because of the very limited range of the MMSE score within the single groups (p > 0.05).

OPEN IN VIEWER

Fig.4

Scatterplots showing the correlation between (eLORETA) source activity of the rsEEG rhythms and the MMSE score in the Nold, ADMCI, and DLBMCI subjects as a whole group. The Spearman test evaluated the hypothesis of a correlation these rsEEG and MMSE variables (Bonferronio corrected p < 0.05). The r and p values are reported within the diagram.

The present results suggest that even if statistically significant (p < 0.005), the absolute values of the monotonic (Spearman) correlations between rsEEG source activities and the MMSE score in the Nold, ADMCI, and DLBMCI individuals were relatively low as data variance explained (r = 0.32–0.38). To extend the above correlation analysis, we also computed the simple non-linear correlations between the significant rsEEG source activities (i.e., parietal delta, occipital alpha 2, limbic alpha 2, occipital alpha 3) and the MMSE score using a procedure that has been successfully used in one of our previous rsEEG studies in patients with neurodegenerative disorders [12]. Specifically, we calculated the coefficient of determination r² for exponential, logarithmic, and power functions as follows: r 2 = 1 - ( SSE / SST ) where SSE = Σ ( y i - y = i ) and SST = ( Σ y i 2 ) - ( ( Σ y i ) 2 / n ) where n is the number of samples (i.e., subjects), y_i is the real value and y      = _i is the approximated value calculated with the following formula: y = i = c ln x i + b for logarithmic functions ( c , b constant ) y = i = c e i bx for exponential functions ( c , b constant ) y = i = c x i b for power functions ( c , b constant )

Table 4 reports the coefficient of determination r² between rsEEG source activities (i.e., parietal delta, occipital alpha 2, limbic alpha 2, occipital alpha 3) and the MMSE score using all Nold, ADMCI, and DLBMCI individuals as a whole group. Results showed that the r² values for the monotonic, exponential, logarithmic, and power functions were relatively low as explained data variance (i.e., r² = 0.10–0.16). However, these values exhibited statistical significance (p < 0.005), confirming the relevance of those rsEEG source activities for global cognition in humans. Of note, those values for the linear and non-linear functions were quite similar for the correlation between rsEEG alpha source activities and the MMSE score. Instead, the correlation between the rsEEG delta source activities and the MMSE score was characterized by a higher r² value for the monotonic than the exponential, logarithmic, and power functions.

Classification among Nold, ADMCI, and DLBMCI individuals based on rsEEG source activity

As a second analysis on the clinical relevance of the main results, the above seven rsEEG source activities (i.e., frontal, parietal, temporal delta; occipital, temporal, and limbic alpha 2; occipital alpha 3) served as discriminant input variables for the computation of the AUROC curves. These AUROC curves aimed at indexing the classification accuracy in the discrimination among the Nold, ADMCI, and DLBMCI individuals. The results were reported in detail in Table 7 and Fig. 5.

OPEN IN VIEWER

Fig.5

Top: Receiver operating characteristic (ROC) curves illustrating the classification of the ADMCI and Nold individuals based on the occipital alpha 2 (eLORETA) source activity. The area under the ROC (AUROC) curve was 0.86 indicating a good classification accuracy of the ADMCI and Nold individuals. Middle: ROC curves illustrating the classification of the DLBMCI and Nold individuals based on the parietal delta (eLORETA) source activity. The AUROC curve was 0.89 indicating a good classification accuracy of the DLBMCI and Nold individuals. Bottom: ROC curves illustrating the classification of the DLBMCI and ADMCI individuals based on the parietal delta (eLORETA) source activity. The AUROC curve was 0.72 indicating a moderate classification accuracy of the DLBMCI and Nold individuals.

Regarding the classification of the Nold versus ADMCI subjects, the following 4 rsEEG markers overcame the threshold of 0.7 of the AUROC curve, defined as a “moderate” classification rate (Table 8): occipital alpha 2, temporal alpha 2, limbic alpha 2, and occipital alpha 3 source activities. Among these rsEEG markers, the occipital alpha 2 source activity reached the following best classification rate (Fig. 4 top): a sensitivity of 90%, a specificity of 73.3%, an accuracy of 81.7%, and 0.86 of the AUROC curve.

Concerning the classification of the Nold versus DLBMCI subjects, the following 6 rsEEG markers overcome the threshold of 0.7 of the AUROC curve (Table 8): frontal delta, parietal delta, temporal delta, occipital alpha 2, limbic alpha 2, and occipital alpha 3 source activities. Among these rsEEG markers, the parietal delta eLORETA source activities reached the following best classification rate (Fig. 4 middle): a sensitivity of 82.6%, a specificity of 80%, an accuracy of 81.3%, and 0.89 of the AUROC curve.

Finally, regarding the classification of the ADMCI versus DLBMCI subjects, the following two rsEEG markers overcame the threshold of 0.7 of the AUROC curve (Table 8): parietal delta and temporal delta source activities. Among these rsEEG markers, the occipital alpha 2 source activity reached the following best classification rate (Fig. 5 top): a sensitivity of 78.3%, a specificity of 66.7%, an accuracy of 72.5%, and 0.72 of the AUROC curve.

The rsEEG markers exhibiting differences between ADMCI and DLBMCI groups

As an important aspect of the present methodology, the frequency bands from delta to alpha were defined on an individual basis using the TF and IAF as landmarks (for further details see the “Methods” section). This choice allowed us to take into account the substantial differences in the IAF in the present cohorts of ADMCI and DLBMCI patients. The IAF was lower in the ADMCI (8.8 Hz) and DLBMCI (7.8 Hz) groups than the Nold group (9.4 Hz). Furthermore, it was also lower in the DLBMCI than the ADMCI group. If we were used the standard fixed alpha 1 (8–10/10.5 Hz) and alpha 2 (10–12/13 Hz) frequency sub-bands, DLBMCI would have shown low source activity in these sub-bands merely due to the slowing of the IAF.

Compared with the Nold group, the posterior source activity of the individual low-frequency alpha (i.e., alpha 2) and the occipital source activity of high-frequency alpha (i.e., alpha 3) were abnormally lower in both ADMCI and DLBMCI groups. Remarkably, this source activity was lower in the ADMCI than the DLBMCI group. Furthermore, a widespread delta source activity exhibited an abnormally higher activity in both ADMCI and DLBMCI groups. Another novel finding was that this delta source activity was higher in the DLBMCI than the ADMCI group.

The present results extend to source space and individually-determined frequency bands previous EEG evidence reported in groups of ADD and DLB patients [20–23, 56–60]. In those previous investigations, widespread delta and theta (e.g., also termed “pre-alpha”) rhythms were greater in power in DLB than ADD patients [20–22]. Finally, the present results fit previous evidence documenting more abnormal occipital delta and theta rhythms in DLB than ADD patients, even at the prodromal stage of MCI [23].

Finally, a clinically relevant evidence was that the delta and alpha source activities demonstrated a linear and non-linear correlation with the MMSE score (roughly reflecting global cognitive status) across all Nold, ADMCI, and DLBMCI subjects as a whole population. However, it should be remarked that even if statistically significant (p < 0.005), the absolute values of the monotonic (Spearman) and simple non-linear correlations were relatively low as variance explained (i.e., r = 0.10–0.16). The present findings suggest that the neurophysiological mechanisms of cortical neural synchronization underpinning brain arousal and low vigilance (as reflected in the rsEEG source solutions of this study) are only one of the determinants of global cognitive functions in human subjects. Other relevant neurophysiological mechanisms involved in cognitive information processes may be those related to selective attention, encoding and retrieval of information in long-term memory, frontal executive functions (some assisted by internal language), and others. Therefore, future studies may measure cortical EEG rhythms not only during the resting state condition (i.e., low vigilance) but also during attention, episodic and working memory, and other cognitive tasks. The derived multiple EEG markers may be used as an input for linear (logistics regression) and non-linear (artificial neural networks or support vector machines) predictors of MMSE score as a measurement of the global status of cognitive functions in Nold subjects and patients with dementing disorders. The expected results may show high correlation values and remarkable insights about the derangement of brain functions in the evolution of dementing disorders.

The effect of ADMCI and DLBMCI on neurophysiological mechanisms generating resting state delta and alpha rhythms

Neurophysiological mechanisms underlying the above abnormalities in delta and alpha rhythms are poorly known. It can be just speculated that in the quiet wakefulness, the abnormal increase in magnitude of cortical delta rhythms in DLBMCI and (to lesser extent) ADMCI is caused by an altered interaction between thalamic and cortical pyramidal neurons. This effect might cause an abnormal functional connectivity, inducing a partial isolation of cortical delta sources [61–63]. This speculation is based on two lines of evidence in AD patients. Firstly, there is a concomitant increase of resting state delta rhythms and a decrease of regional cortical blood perfusion and metabolism [64–69]. Secondly, there is a clear relationship between expression of these rhythms and the atrophy of cortical gray matter and hippocampus [28, 71]. A new fascinating avenue is the study of the relationship between the abnormal cortical delta and alpha rhythms in those patients and the alterations of the default mode, frontoparietal attentional, and frontal executive networks as revealed by resting state functional MRI [72, 73].

Concerning the present abnormalities in patients’ posterior alpha sources (prominent in ADMCI), it can be speculated that they might reflect an alteration of a complex neurophysiological network regulating the continuous sensory flow to the cortex, as well as cortical arousal, and vigilance in the quiet wakefulness [74–76]. This speculation is based on the fact that oscillations of post-synaptic potentials at the alpha frequency control the nodes of such neurophysiological network, which spans thalamocortical high-threshold and relay-mode neurons, GABAergic interneurons, and cortical pyramidal cells in sensory areas [74–76]. As a result of those oscillations, cycles of excitation and inhibition around 70–100 ms can be observed in the local field potentials recorded in mammalian thalamic and cortical neural populations [74–76]. During the active processing of sensorimotor information, these cycles might frame perceptual events in discrete snapshots and would ensure the selectivity of that processing [74–76]. During the quiet wakefulness, they might regulate the level of general arousal in sensory, cognitive, and motor cortical areas [52, 63]. According to the above neurophysiological model, the prominent posterior alpha abnormalities in the present ADMCI and (to a less extent) DLBMCI patients might predict an alteration of visuospatial working and episodic memory processes, which may possibly relate to altered inputs from cholinergic basal forebrain to the visual cortex [77–79].

The rsEEG markers showing classification accuracy between Nold versus patients

Another clinically relevant output of the present study was the good classification accuracy between Nold and ADMCI/DLBMCI individuals based on rsEEG sources. We found an accuracy (e.g., AUROC curve) of 86% in the classification of the Nold and the ADMCI individuals, using occipital alpha 2 source activity as an input. Furthermore, the parietal delta source activity allowed an accuracy of 89% in the classification of the Nold versus the DLBMCI individuals. These findings build on previous variable evidence obtained from scalp rsEEG rhythms and fixed frequency bands. Discrimination accuracy has varied considerably in AD with values of 94–45% between Nold and ADD individuals, 92–78% between ADMCI versus ADD individuals, 87–60% in the prediction of the conversion from ADMCI to ADD status [14, 80–88]. More robust discrimination accuracy of higher than 90% has been reported in the classification of Nold and DLB individuals [29, 89].

In the present study, we report just a moderate accuracy of 72% in the classification of the ADMCI versus the DLBMCI individuals (parietal delta source activity as an input). This percentage was clearly lower than the range of classification accuracies reported in the literature about the discrimination between ADD and DLB individuals. In previous studies, a visual analysis of rsEEG traces allowed a discrimination accuracy of 78% between DLB and ADD individuals [90]. In a more recent study, the same kind of visual analysis allowed a discrimination accuracy of 77% between DLB and ADD individuals [91]. Compared with the visual analysis of rsEEG traces, the spectral analysis returned a greater discrimination accuracy of 80–91% between DLB and ADD individuals [25, 92].

At the present stage of the research, it is unclear why the discrimination accuracy is just moderate between ADMCI and DLBMCI ones. It can be speculated that at the prodromal stage, neuropathological features have more similar features in ADMCI and DLBMCI as compared to ADD and DLB status. Regarding the neuropathology factor, not only DLBMCI but also ADMCI patients may suffer from some depletion of cerebral tegmental dopamine [93] while individuals of both diseases may show a loss of basal forebrain cholinergic cells, Aβ neuritic plaques, and cortical tauopathy [94, 95]. Noteworthy, an elevated deposition of Aβ proteins in the brain was correlated with indexes of cognitive impairment in DLB patients [96]. In the same line, clusters of ADMCI and DLBMCI patients can share some progressive impairment of clinical variables such as visuospatial construction, visual conceptual reasoning, and speed of processing, even if these functions are differently impaired in the two groups [97–99]. Namely, there is greater impairment on visual tasks, speed of processing, and all-type hallucinations in DLB compared to ADD and more verbal memory impairment in ADD than DLB [97–99]. An alternative explanation is that some ADMCI and DLBMCI belonged to mixed cases of DLB-AD, thus reducing the classification accuracy between ADMCI and DLBMCI. Furthermore, another factor might be the effect of psychoactive treatment (cholinergic and/or dopaminergic) that may mitigate inter-group differences in the EEG rhythms at this stage of the disease. Future prospective studies should use appropriate structural MRI, CSF diagnostic biomarkers of AD, and DATsSCan in de-novo (i.e., no treatment) ADMCI andf DLBMCI groups of MCI patients to quantify those with mixed dementing disorders.

Methodological remarks

In the interpretation of the present findings, the following methodological limitations should be considered. The groups of ADMCI (n = 30) and DLBMCI (n = 23) patients were relatively small, so we cannot stratify them in relation to disease severity. Furthermore, this retrospective study was based on data of several clinical units that did not follow a harmonized protocol. As a result, there was a jeopardized availability of potentially relevant genetic and neuroimaging biomarkers, clinical measurements, criteria of MCI status, and neuropsychological scores across the clinical units. The only measurement of the global cognitive status common to all subjects was the MMSE score and the evaluation of functioning in the daily life activities. Furthermore, the clinical outcome of the two groups is not available for most of the ADMCI and DLBMCI patients.

Conclusions

This retrospective and exploratory study on archive data evaluated the preliminary hypothesis that cortical sources of rsEEG rhythms might characterize peculiar neurophysiological mechanisms of brain arousal in ADMCI and DLBMCI patients. Results showed that both MCI groups exhibited an IAF slower in frequency (especially the DLBMCI group) compared with the Nold group. Furthermore, both MCI groups exhibited abnormal lower posterior alpha source activities, especially the ADMCI group. Finally, they showed abnormally higher widespread delta source activities, especially the DLBMCI group. As a possible sign of clinical relevance, these rsEEG markers correlated with the MMSE score (i.e., global cognitive status) and allowed good classification accuracies (AUROC >0.85) between the Nold and MCI individuals regardless the diagnosis. These rsEEG markers allowed just a moderate classification accuracy (AUROC >0.7) between the ADMCI and DLBMCI individuals.

These preliminary results suggest that ADMCI and DLBMCI patients might be characterized by different spatial and frequency features of the rsEEG sources at the group level, possibly reflecting cortical neural synchronization underpinning brain arousal in quiet wakefulness. The abnormalities of these neural synchronization mechanisms could be also observed at the individual level.

The preliminary results of this study motivate future prospective, longitudinal, multi-center studies using harmonized protocols and hardware for a detailed evaluation of the patients’ cognitive status and brain function. If cross-validated, the present rsEEG biomarkers described may have utility in clinical practice and drug discovery (e.g., patients’ stratification based on abnormalities of rsEEG biomarkers).