Alzheimer’s Disease Neuroimaging Initiative
Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database ( http://adni.loni . usc.edu). The ADNI was launched in 2003 as a public-private partnership by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private pharmaceutical companies, and nonprofit organizations, as a $60 million, 5-year public-private partnership. The primary goal of ADNI has been to test whether serial MRI, positron emission tomography, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of MCI and early AD. The initiative is headed by Michael W. Weiner,MD,VA Medical Center and University of California, San Francisco. ADNI is the result of efforts of many co-investigators from a broad range of academic institutions and private corporations, and was approved by the institutional review board at each site and was compliant with the Health Insurance Portability and accountability Act. Written consent was obtained from all participants.
The study selected participants matched in gender, age and education level from the ADNI2 and ADNI-GO. In the present study, a total of 224 participants with complete resting-state functional magnetic resonance imaging, neuropsychological tests and plasma NFL at baseline and a mean follow-up period of 17 months were eligible for inclusion. In details, an imaging analysis approach was utilized to investigate disease-related differences in test data (baseline, n = 112) and validation data (follow up, n = 112), including 29 cognitively normal (CN) subjects, 35 early mild cognitive impairment (EMCI), 30 late mild cognitive impairment (LMCI) and 18 AD patients both in baseline and follow up, respectively.
Clinical categorize was performed by the the central of ADNI based on diagnostic procedures from the ADNI protocol. Neuropsychologic assessments were used in the study including general cognition and episodic memory. General cognition containing Mini Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), Clinical Dementia Rating (CDR), Alzheimer's Disease Assessment Scale-Cognitive section (ADAS) and Functional activities questionary (FAQ), while Ray Auditory Verbal Learning Test (RAVLT) can be regarded as a marker of episodic memory. Briefly, ADNI required participants with CN to have MMSE scores of 24–30, CDR of 0, and no subjective or informant-reported memory decline and normal performance. The MCI are defined as including MMSE scores of 24–30, CDR of 0.5 and objective memory loss on the education adjusted Wechsler Memory. According to the degree of memory impairment of Wechsler memory scale (WMS), MCI is divided into EMCI and LMCI. AD participants showed memory decline with MMSE scores of 19–24, CDR of 0.5-1.0. For a detailed overview of full inclusion, exclusion and diagnostic criteria described are available in the ADNI procedure manual (http://www.adni-info.org/).
Plasma Nfl Markers
All subjects in this study had plasma NFL both in baseline and follow-up. Plasma NFL concentrations were analyzed using an in-house immunoassay on the Single molecule array (Simoa) HD-1 Analyser (Quanterix) as previously described.26 Analytical sensitivity of the assay was < 1.0 pg/mL, which used a combination of monoclonal antibodies and purified bovine NFL as a calibrator.
Mri Data Acquisition
All participants were scanned using a 3.0-Tesla Philips MRI scanner. Resting-state fMRI scans images were obtained with an echo-planar imaging sequence covering the entire brain with the following parameters: repetition time (TR) = 3000ms; echo time (TE) = 30ms; flip angle = 80◦; acquisition matrix = 64 × 64; number slices = 48; slice thickness = 3.3 mm; spatial resolution = 3.31 × 3.31 × 3.31mm3. All imaging data were downloaded in the ADNI cohort and a detailed description of the resting-state fMRI data acquisition protocols also can be found at ADNI website (http://www.adni-info.org). The whole pipeline of four main sectors of image processing (a flowchart see Fig. 1), and the followings were the details.
The resting-state fMRI data were preprocessed using the Data Processing Assistant for Resting-State fMRI v2.2 (DPARSF) and Resting-State fMRI Data Analysis Toolkit (REST, http://restfmri.net), which is based on Matlab platform (Matlab R2013b). The data preprocessing steps were as follows. The first ten volumes of each data was discarded to uniform the signal equilibrium and eliminate subject's inadaptation to the scanning noise. Then remaining volumes were corrected to compensate for differences in acquisition time between slices and realignment to rectify head motion between scanning. The acquired functional images normalized to the Montreal Neurological Institute (MNI) space, and resampled into 3 mm × 3 mm × 3 mm. Gaussian kernel (6 × 6 × 6 mm) was applied to smooth the data. Band-pass filtering (0.01–0.08 Hz) was performed to reduce low frequency drift and high frequency noise due to physiological respiration and cardiac noise.
Brain Networks Construction And Analysis
Network interaction based on the functional connectivity of core subsystem
Our previous study has shown that core-centered connection abnormalities inside DMN subsystems at an early stage of AD . In this study, we focused on the interactions of core subsystem associated with the outside of DMN subsystems and their clinical significance. First, the "Image Calculator" of REST was used to calculate the regions of core subsystem. Second, core subsystem served as seed for the functional connectivity of a voxel-wise analysis was used to explore the interactions between brain networks. For each subject, an average time series for core subsystem was computed as the reference time course. Pearson cross-correlation analysis was then conducted between the average signal change in the core subsystem and the time series of whole-brain voxels. Third, Fisher's z-transform (z = 0.5×ln(1 + r)/(1-r)) improved the normality of the correlation coefficients, and the individual maps of each network was constructed. Finally, a voxel-wise analysis of variance (ANOVA) was performed in the baseline four groups. The statistical thresholds were set at a corrected p < 0.05, determined by Monte Carlo simulation for multiple comparisons (Parameters: single voxel p value = 0.05, a minimum cluster size of 6156 mm3, FWHM = 6 mm, with mask. See program AlphaSim by D. Ward, and http://afni.nimh.nih.gov/pub/dist/doc/ manual/AlphaSim.pdf).
Association analysis between core-related networks and plasma NFL
We were particularly interested in these regions which were associated with statistically significant interactions of core-related networks and plasma NFL. Therefore, the regions of aforementioned ANOVA was extracted from each baseline participant’s image. And then, a voxel-level correlation analysis was performed between these regions and plasma NFL in baseline AD-spectrum subjects (i.e., EMCI, LMCI and AD). To remove possible effects of age, gender and education on the results, these parameters were introduced as covariates. To further remove spurious correlations, only those correlation coefficients were retained which corresponding p values were lower than a statistical threshold (p < 0.05, determined by Monte Carlo simulation for multiple comparisons).
Significance of plasma NFL coupling networks on groups identification
To further explore the classification by the association of core-related networks and plasma NFL, these regions identified via baseline ANOVA and association analysis was extracted as region of interest (ROI). To avoid circular analysis, the mean Z values of every ROI were then calculated in all follow-up groups, respectively. Finally, the ability of ROIs to separate these four follow-up groups (i.e., CN, EMCI, LMCI and AD) was computed using Receiver Operating Characteristic (ROC). Area under the ROC curve (AUC) values were extracted; the values distinguishing these follow-up four groups were examined. The thresholds were set at a p < 0.05.
Significance of plasma NFL coupling networks on cognitive impairments
To determine the hypothesis of whether these aforementioned ROIs mediated the relationships between plasma NFL and cognition, the mediation analysis was further performed. The primary estimates of interest were the behavioral significance in the extracted ROIs both in baseline and follow-up data, respectively. We computed the bias-corrected 95% confidence intervals (CI) for the size of the mediating effects with bootstrap method (k = 1000 samples). The significance of the indirect effects was confirmed when 95% CIs were not contain zero. All the data were processed with the PROCESS for the Statistical Statistical Package for Social Sciences (SPSS) framework.
Statistical Analysis Involving Demographic And Neuropsychological Data
For demographic and neuropsychological information, the Kruskal-Wallis tests were used for the variables (i.e., age, education and behavioral scales) and the dichotomous variables (i.e., gender) while post-hoc tests were also performed by Mann-Whiney U Tests. The significance level was set at p < 0.05 by SPSS version 22.