262 individuals participated in a study aimed to characterize the neurophysiological features of healthy aging. Participants were recruited from local hospitals and through several dissemination talks, and a team of expert neuropsychologists assessed that they met inclusion criteria. A detailed list of exclusion criteria can be found in (12). The procedure was performed following current guidelines and regulations, and the study was approved by the Hospital Universitario San Carlos Ethics Committee. Every participant signed an informed consent.
We included participants who had available data regarding our main variables of interest (n = 158; Mini Mental State Examination, MMSE, score, genetic information and validated magnetic resonance imaging, MRI, MEG and actigraphy data). We then excluded anyone with an MMSE score less than 26 (n = 5), aged less than 50 years (n = 8) and participants carrying less frequent APOE genotypes (ε2ε3, n = 11; ε2ε4, n = 1; ε4ε4, n = 7). Among the remaining 127 participants, there were 36 APOE ε4 carriers and 91 non-carriers. We carefully selected 33 APOE ε4 carriers and 74 non-carriers so that both subsamples would match in PA levels (TPA and MVPA), age, sex, educational level, MMSE and body mass index. There were two main reason to match the sample according to all these relevant variables instead of using them as covariates in subsequent analyses. First, including several covariates in a cluster-based permutation test could have introduced a methodological pitfall in the permutation procedure. Second, using covariates only controls for linear influences on the data, dismissing any other possible non-linear confound.
The final sample was composed of 107 healthy older adults, aged 50-82 years. A detailed list of the sample characteristics can be found in Table 1, including scores extracted from the neuropsychological tests: Geriatric Depression Scale (13), the anxiety subscale from the Goldberg Anxiety and Depression Inventory (14) and the Digit Span Forward, Digit Span Backward and Logical Memory II (delayed recall, units and gist) subscales from the Weschler Adult Intelligence Scale IV (WAIS-IV, (15)).
Physical Activity Measurement
For PA measurement we used the ActiGraph GT3X+ accelerometer (LLC, Pensacola,FL). Participants were requested to wear the accelerometers on their right hip for 7 complete days, taking them off only during water-based activities (16,17). For cleaning and processing the data, we used ActiLife software(6.13.3) (LLC, Pensacola, FL). The validation criteria required each individual to wear the accelerometer during at least 3 weekdays and 1 weekend day for a minimum of ten hours per day (17).We considered ≥60 min of continuous zeroes while allowing for up to 2 min of counts ≤100 counts as non-wear time (18). To classify the PA, we categorised sedentary time as <100 counts/min, light activity as 100–1951 counts/min, and moderate to vigorous physical activity (MVPA) as ≥1952 counts/min (19).
In this study, two different measures of PA were incorporated: Total Time In Freedson Bouts, which is a standardized measure of PA volumes (total PA, TPA), and daily average of MVPA. TPA was normalized by total wear time.
As described in (12), we obtained genomic DNA from 10 ml blood samples in ethylenediaminetetraacetic acid (EDTA). Employing TaqMan assays on an Applied Biosystems 7500 Fast Real Time PCR machine (Applied Biosystems, Foster City, CA), single nucleotide polymorphisms (SNPs) rs7412 and rs429358 genotypes were determined. APOE genotype was established accordingly. In this study, only ε3ε3 and ε3ε4 individuals were considered.
MRI acquisition and Volumetric Analyses
To generate the T1-weighted MRI images from each participant, a General Electric 1.5 T system was employed. We applied a high-resolution antenna and a homogenization PURE filter (Fast Spoiled Gradient Echo sequence, TR/TE/TI = 11.2/4.2/450 ms; flip angle 12°; 1 mm slice thickness, 256x256 matrix and FOV 25 cm).
The resulting images were processed using Freesurfer software (version 5.1.0) and its specialized tool for automated cortical parcellation and subcortical segmentation (20). The measures that were included in further analyses were total grey matter, amygdala, precuneus and hippocampus (in mm3). The volumes of bilateral structures were collapsed in order to obtain a single measure for each region.
Diffusion Tensor Imaging
The same General Electric 1.5 T magnetic resonance scanner was also used to collect diffusion weighted images (DWI). The acquisition parameters for DWI were: TE/TR 96.1/12,000 ms; NEX 3 for increasing the SNR; 2.4 mm slice thickness, 128 × 128 matrix and 30.7 cm FOV yielding an isotropic voxel of 2.4 mm; 1 image with no diffusion sensitization (i.e., T2-weighted b0 images) and 25 DWI (b = 900 s/mm2). Data were recorded with a single shot echo planar imaging sequence.
DWI images were processed following the procedure previously published in (21). Probabilistic fiber tractography was run on the he automated tool AutoPtx (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/AutoPtx ) as reported in (21). Two bilateral tracts were later used for correlation analyses: the uncinate and the parahippocampal fasciculi. In order to reduce the number of tests, the weighted arithmetic mean of left and right structures was used. Likewise, a measure of global fractional anisotropy (FA) was calculated averaging all 27 original tracts provided by the system.
Data acquisition and signal preprocessing
MEG data was recorded using a 306-channel whole-head MEG system (Vectorview, ElektaNeuromag, Finland), placed in a magnetically shielded room at Center for Biomedical Technology in Madrid, following the protocol described in (12).
Raw data were first submitted to the Maxfilter software to remove external noise (22). Fieldtrip software (23) was used to automatically scan MEG data for artifacts, which were visually confirmed by an MEG expert. Artifact-free data were segmented in 4 seconds epochs. Then MEG time series were filtered into delta (2-4Hz), theta (4-8Hz), alpha (8-12Hz) and beta (12-30 Hz). This procedure has been reported in detailed in (24).
Source reconstruction and connectivity analyses
We used a regular 1cm grid in the Montreal Neurological Institute (MNI) template. The resulting model comprised 2459 sources distributed across the brain, which were transformed to each subject’s space following the methodology detailed in (24).
We used phase locking value (PLV) to calculate functional connectivity. The Automated Anatomical Labeling atlas (AAL, (25)) was applied to segment the source template with 2459 nodes excluding the cerebellum, basal ganglia, thalamus, and olfactory cortices. The resulting 78 regions of interest included 1202 nodes. Symmetrical whole-brain matrices of 1202x1202 nodes were obtained by averaging PLV values across trials for each participant and frequency band. Each node’s strength was computed by averaging its corresponding FC with the whole grid. Such averaging resulted in a source-reconstructed FC matrix of 1202 nodes by 4 frequency bands by 107 participants.
Functional Connectivity Strenght (FC-st) analyses
Network-based statistics (NBS) were carried out for each frequency band (26). Clusters consisted of several adjacent nodes that presented a significant partial correlation (age as covariate) between FC-st values and each PA variable (Spearman correlation, p< 0.01). To form a cluster, the correlation coefficients of all nodes were required to have the same sign. Only clusters including at least 1% of the grid (i.e. a minimum of 12 nodes) were considered. Spearman rho values were Fisher Z-transformed. Cluster-mass statistics were computed as the sum of all z values corresponding with all nodes within each cluster. To control for multiple comparisons, the whole process was repeated 5000 times, shuffling the correspondence between FC-st and each PA measure across all participants. At each repetition, the maximum surrogate cluster’s statistic was kept creating a maximal null permutation distribution. For each main cluster, cluster-mass statistics in the original and the randomized datasets were compared. In NBS, p-value represent the proportion of the permutation distribution with cluster-mass statistic values greater or equal than the cluster-mass statistic value of the original data. Only clusters which survived NBS (permutations p-value < 0.05) were considered in further analyses. For each main cluster, FC-st values were averaged across all nodes. These markers were used in subsequent correlation analyses with measures of specific AD signatures (the complete list is shown in Table 4). These were carried out taking the whole sample and following stratification of the cohort by APOE ε4 carriers and non-carriers. p-values were corrected using false discovery rate (FDR) to account for multiple testing. Statistical analyses were carried out using Matlab R2018b (Mathworks Inc).
In order to examine whether the FC-st results were caused by global or local effects, we performed corresponding seed analysis, using the previous clusters as seeds. The FC values assessed were the average FC between each node of the grid and corresponding cluster’s nodes. Then, a set of partial correlation (age as covariate) between these FC values and each PA variable (Spearman correlation, p< 0.01) were computed. Only clusters that did not overlap with the original seed-cluster were reported in this study.