Participants
A total of thirty-eight individuals were recorded, 22 elderly cognitively healthy controls and 16 early AD (eAD), where the mean age was (70.32 (+/-7.79); 6 male) and (77.06 (+/-6.97); 8 male) years for the control and eAD group, respectively. The exclusion criteria for both groups were evidence of non-degenerative dementia (e.g., inflammatory, metabolic, or vascular etiology); dementia or mild cognitive impairment of doubtful origin; severe medical conditions that limited their ability to participate in the study (e.g., uncompensated diabetes, severe hypertension). The Scientific and Ethical Committee approved all procedures involving participants of the Clinical Hospital of the University of Chile Protocol number: 26/2015. In addition, all participants signed an informed consent previously approved by this Ethics Committee.
Neuropsychological testing
Cognitive functions were examined with the Clinical Dementia Rating scale (CDR) [17] and CDR Sum-of-Boxes (CDR-SOB) [18], a numeric scale used to quantify the severity of dementia (e.g., its stages). Also, the Montreal Cognitive Assessment (MoCA) [19], the Montreal Cognitive Assessment Memory Index Score (MoCA-MIS) [20], and the Minimental State Examination (MMSE) were used as a rapid screening instrument for mild cognitive dysfunction. A complete clinical evaluation, including full neuropsychological tests, was performed by a neurologist from the Department of Neurology of the Clinical Hospital of the Universidad de Chile, who was blind to the performance of the subjects in the navigation task. Participants were classified in cognitively unimpaired (CDR0) and early AD (eAD) based on their cognitive status.
The study involved 18 volunteers: 9 participants (eAD group) and 9 participants (control group) aged between 61–88 years. The eAD group consisted of patients with cognitive impairment with apparent memory deficits, with a global CDR 0.5, a CDR-SOB ≥ 1.5, and a MoCA-MIS ≤ 10, with two or fewer words out of 5 of the MoCA tests recalled spontaneously. Also, eAD participants had a very mild loss of instrumental activities of daily living. Control participants were submitted to the same neurological and neuropsychological evaluations as the eAD group. The demographic data of the participants are shown in Table 1.
Table 1
Demographic characteristics and levels of cognitive function.
Characteristic
|
eAD
number (%)
or mean ± SD
|
Control
Number (%)
or mean ± SD
|
p value
|
Sample
|
|
|
|
Size
|
9
|
9
|
|
Age
|
76.67 ± 6.16
|
71.22 ± 8.48
|
0.2138
|
Range (min-max)
|
69–88
|
61–84
|
|
Gender
|
|
|
0.6199
|
Male
|
2 (22.22%)
|
4 (44.44%)
|
|
Female
|
7 (77.78%)
|
5 (55.56%)
|
|
Education (years)
|
12.33 ± 4.03
|
16.89 ± 4.31
|
0.0299
|
Neuropsychological measures
|
|
|
|
CDR-SOB scores
|
0.89 ± 0.22
|
0
|
< 0.001
|
MoCA scores
|
20.44 ± 3.32
|
29.22 ± 0.83
|
< 0.001
|
MoCA-MIS
|
9.56 ± 2.19
|
14.78 ± 0.44
|
< 0.001
|
MMSE scores
|
23.22 ± 2.05
|
29.78 ± 0.44
|
< 0.001
|
Continuous data are presented as mean (standard deviation), and categorical data are expressed as frequencies (%). Wilcoxon rank-sum test was used for age, education, CDR-SOB, MoCA, MoCA-MIS, and MMSE comparison between groups. The Chi-square test was applied for gender comparison (Fisher's exact test). Abbreviations: HC, Healthy controls; eAD, very early Alzheimer's Disease; CDR-SOB, Clinical Dementia Rating Scale Sum-of-Boxes; MoCA, Montreal Cognitive Assessment; MoCA-MIS, Montreal Cognitive Assessment Memory Index Score; and MMSE, Mini-mental State Examination. |
Spatial navigation paradigms: Virtual Morris Water Navigation Task
Spatial exploration was assessed by the VMWM task in both eAD and control participants. The task was executed through an open-use program with support provided by its author [10]. The virtual environment of the VMWM task simulates the traditional MWM, which consists of a circular pool located inside a room with visual cues on its four walls. The task is presented on a computer screen that the participant must navigate through the buttons of a keyboard to find a platform hidden under the water surface. The setup enables controlling a series of parameters to optimize the sensitivity of the task, related to characteristics of the visual cues and the location of the hidden platform, the maximum duration, and the number of trials. After finishing the task, the program stores in text files the information related to the route traveled, the latency to find the platform, and the relative percentages of the length of the path spent in each quadrant of the virtual pool.
All participants performed a well-established navigation task which was divided into three stages. Training (Stage I): this task involved finding the platform hidden under the water in a room furnished with visual clues on each of its walls. The platform became visible after one minute of exploration if the participant did not find it. Therefore, the participant needed to reach the platform to finish the trial. Before the end of the trial, the participant had 2 seconds to rotate in place to explore its position in the environment visually. This stage comprised four repetitions of the trial. The platform was kept in the same hidden position inside the pool, and the participant always started each repetition from a different position. Task I (Stage II): The participant had to find the platform hidden based on different visual cues to those presented in the training session as in the previous stage. The task included twenty trials divided into four groups. A two-minute break was considered among each group of trials. Similar to the training session, the platform was always kept in the same position inside the pool. The participant always started each repetition from a different position inside the pool. Task II (Stage III): in an equivalent room but including different visual cues, the participant had to select one of two visible platforms; only one represented the correct choice. A note on the screen indicated whether the platform chosen was correct. This task was used to control each participant's visual and psychomotor functioning, making it possible to rule out any deterioration in these parameters as plausible reasons for unsatisfactory performance in task I.
EEG recordings and eye-tracking
During the behavioral navigation task, the continuous acquisition of the electroencephalographic activity of each participant was carried out beneath adequate recording conditions. An EEG system of 32 + 8 channels was employed (8 external channels to measure electro-ocular activity and referential mastoid recording; BioSemi®). The acquired analog signal was filtered between 0 (real DC) and 1000 Hz before its conversion, sampled at 2048 Hz, and digitally converted with a precision of 24 bits. The recording system used a specific reference system (CRS-DRL) that allows unlimited data storage. After electrodes were located, the participant was installed in a suitable chair in front of the monitor where the VMWM task was displayed. The head was positioned on chin support to minimize movement during the task and allow optimal detection and recording of eye movements. Eye-tracking was performed using an Eyelink® 1000 system, which digitizes and stores eye-tracking data in a binary file convertible to text. The bi-dimensional position of the pupils was obtained at a frequency of 500 Hz. The system also automatically registers the occurrence of blinks, fixations, and ocular saccades based on user-defined initial setting parameters.
Data Analysis
Henceforth, all data analyses made in this study, including behavioral parameters of space navigation in the VMWM task, electroencephalographic signals, and eye-tracking data, were performed with MATLAB® (The MathWorks, Inc.). First, for behavioral data, the text files generated by the program after finishing the task were imported and analyzed using custom algorithms designed for these purposes, employing general functions of the basic software package. Second, the binary .bdf files generated were directly preprocessed using the Fieldtrip open-source toolbox for EEG signals. The eye-tracking binary data were initially converted to ASCII text files using the Eyelink® executable EDF2ASC.exe. Finally, the EYE-EEG MATLAB toolbox was used as a plugin of the EEGLAB package to import, visualize and verify the detected eye-tracking events and synchronize with the EEG signals [21].
EEG signals were visually inspected by a qualified clinical neurophysiologist to judge the collected signals' quality and reject abnormal recordings (i.e., epileptiform activity and abnormal basal rhythms). Then the segments of data presenting non-extractable artifacts were manually marked for the exclusion of the analysis. Offline filter settings were bandpass filter at 1–40 Hz (Butterworth, FIR). Next, artifact elimination was applied using ICA decomposition (fieldtrip toolbox) over the whole continuous record. Finally, noise components were identified in a semi-automated way, utilizing algorithms executed in the EYE-EEG package to detect ocular movement-related components.
Continuous EEG signals were segmented using the ocular tracking signal synchronized with the filtered EEG signal, between − 1000 and 3000 ms around the ocular fixations. To ensure that the data did not significantly bias, we evaluated the first 30 seconds of the records because the time of each participant was unsteady over the trials. Fixation-related epochs were consequently subjected to analyses in the time and frequency domains. First, the time-frequency decomposition was computed on the EEG epochs tapered by a sliding Hanning window, using the same fixed window length seconds for all frequency ranges of 1 and 40 Hz in steps of 1 Hz implemented at the time between − 750 and 1500 ms in steps of 10 ms. Second, we implement an analysis method multitaper based on multiplication in the frequency domain, obtaining finally output power-spectra (fieldtrip toolbox). First, for each time, frequency, and electrode, was calculated as the modulus of the mean across trials. Then, for each frequency, values of power-spectra were transformed to Z scores, normalizing by the corresponding mean and standard deviation of prestimulus time between − 750 and − 450 ms. Finally, these normalized values were compared for each time, frequency, and electrode between experimental groups.
We computed the coherence in two ways: first, performing time-frequency analysis on any time series trial data using the multitaper method, based on conventional tapers like Hanning to obtain the power and the cross-spectra, and in a second way using a method that performs frequency analysis on any time series trial data using a conventional single taper (e.g., Hanning) to obtain the complex Fourier-spectra (fieldtrip toolbox). The estimated coherence ranges from 0 to 1, whereby 0 means that the corresponding frequency components of both signals are linearly independent, and 1 means that the frequency components of the two signals give the maximum linear correlation. Thus, coherence estimation is a valuable tool to observe and quantify the synchrony property of two EEG series, mainly when they are limited to some particular frequency bands [5, 6]. Coherences for delta (2–4 Hz), theta (4–7 Hz), alpha (7–13 Hz), beta (13–30 Hz), and gamma bands (30–40 Hz) were calculated as the mean coherence values of the epochs between 0 and 300 ms. To estimate whether the changes were primarily related to the impairment of short or long-range, coherences were calculated using the following electrodes: (O1, O2, PO3, PO4, CP1, CP2, C3, C4, F3, F4, F7, F8, Fp1, Fp2) that were frequently reported to be affected in AD progression. The EEG coherence calculation for each electrode pair generates a 14 x 14 (EEG channels selected) matrix showing the connectivity between all possible functionally independent brain areas in each frequency band. To examine the regional difference in coherence values, we subdivided the EEG channels into the following four groups: frontal (Fp1, Fp2, F3, F4, and FZ), parietal (FZ, C3, C4, CZ, and PZ), occipital (O1, O2, P3, P4, and PZ), and temporal (F7, F8, P7, and P8). Finally, we measured the effect sizes (Cohen's d) and an independent-sample t-test between groups considering the Beta coherence spectrum.
Eye-tracking analysis
Fixations and saccades were automatically identified based on the velocity (30°/s) and acceleration threshold (8000°/s2). Saccades longer than 5 ms and smaller than 100 ms and fixations between 50 and 800 ms were picked for further processing. In addition, blinks were defined as the absence of pupil data.
Statistics
Demographics and neuropsychological performance were divided into continuous variables expressed as mean and standard deviation (SD), while the categorical variables were expressed as frequencies (%). Wilcoxon rank-sum test was used for age, education, CDR-SOB, MoCA, MoCA-MIS, and MMSE comparison between groups. A chi-square test was used for gender comparison (Fisher's exact test). Differences in ocular behavior as the frequency of ocular movements (fixations and saccades) were measured using a nonparametric Wilcoxon rank-sum test between the control and eAD group. Additionally, we obtained a heat map of the probability of differences and a map of significant differences between the exploration performed by both groups. Finally, we conducted a rank-sum test as a statistical approach pixel by pixel on the image, performed at the 0.05 significance level.
Statistical tests based on permutations were applied to evaluate the differences between groups of the oscillatory activity, such as the time-frequency power spectral decomposition and coherence. The Montecarlo method was considered an estimator of the permutation's significance probabilities and critical values (two-tailed, alpha: p < 0.01, cluster correction, cluster-alpha: p < 0.05). Given many corrections applied to the number of electrodes to be compared between groups, we used cluster as a correction method that solves the Multiple Comparison Problem (MCP) [22, 23]. Moreover, we calculated effect sizes (Cohen's d) as the difference of the means of groups (control and eAD) divided by the weighted pooled standard deviations of the groups. Cohen's d effect size of 0.2 to 0.3 could be a "small" effect, around 0.5 a "medium" effect, and 0.8 to infinity, a "big" effect (a Cohen's d greater than 1 as possible). We estimated significant differences intragroup for the baseline in a beta-band coherence region of interest (15–20 Hz and between 0–300 ms) using a statistical threshold criterion of effect size (> 0.5) and one-sample t-test (< 0.05). This threshold was applied to a chi-square test for intragroup ratios. The data were treated by MATLAB toolbox and Fieldtrip toolbox.