AD mouse model
The mice were purchased from Beijing Vitalstar Biotechnology Co., Ltd. The 3xTg-AD mice were maintained on the C57BL6/129SvJ hybrid background. The study employed female 3xTg-AD mice and age-matched control female C57BL/6N mice, aged 22 weeks and 40 weeks, with ten mice per group (AD22, WT22, AD40, WT40). Only two time points were used to maximize statistical power and apply the reduction principle 27; these specific time points were chosen based on previous work that indicates that they are sufficient to examine the state prior to, and succeeding, the large increase in Aβ plaques28. Mice were cage-housed under a 12:12-h light/dark cycle in a constant 25℃ and 60–70% humidity environment where mice could access food and water freely.
Behavior test
Mice were acclimated to the test room for 1 hour before trials and testing. Open field tests (OFT) were carried out when the mice were awake to test their anxiety 29. Mice were put in the testing arena (70 × 70 × 70 cm) where they could move freely for 10 min. We analyzed the trace of each mouse’s movement in MATLAB R2020b (MathWorks, America). Novel object recognition (NOR) tests were carried out to test the mice's cognition and memory 30. Each mouse was first habituated to the testing arena (70 × 70 × 70cm) for 10 min on two consecutive days. Next, they were allowed to explore two identical objects for 10 min. After 3 hours, they were put in the area again, with a novel object with a different shape and material in place of one of the previous objects, and they were allowed to explore the two differing objects for another 10 min. The time that the mice explored the novel object was recorded 31.
MRI acquisition
The mice were anesthetized using intraperitoneal bolus injections of 25% urethane (Sigma-Aldrich, U2500) dissolved in saline, administered in two separate doses totaling 7 µl/g 32–34. The breath rate of mice under urethane was approximately 150–240, similar to previous studies 34, however it was significantly higher in WT22 than WT40 (p = 0.0021, two-sample t-test) but there was no significant difference between AD22 and AD40 mice, AD22 and WT22 mice, nor AD40 and WT40 mice.
Functional and diffusion tensor images were acquired on the 9.4 Tesla MRI scanner (BioSpec 94/30 USR; Bruker Biospin MRI, Ettilingen, Germany), with a 4-element surface coil and an 86-mm volume transmit coil. T2 Turbo RARE (Rapid Acquisition with Relaxation Enhancement) sequence was used for 2D T2-weighted brain anatomical imaging. The acquisition parameters were as follows: field of view (FOV) = 24.2×9.6 mm2, echo time/repetition time (TE/TR) = 35/2003.56 ms, number of slices = 23, slice thickness = 0.3 mm, resolution = 0.1× 0.1 mm2, matrix size = 242×96, total scan time = 2 min 36 sec.
Echo-planar imaging (EPI) sequence was used to acquire T2*-weighted free induction decay (FID) images for 2D functional imaging. The acquisition parameters were as follows: FOV = 24.2×9.6 mm2, TE/TR = 13/1000 ms, matrix size = 121×48, number of slices = 13, slice thickness = 0.6 mm, resolution = 0.2 × 0.2 mm2, repetition = 480, total scan time = 8 min.
To characterize the microstructure of axon tracts, in vivo Diffusion Tensor Imaging (DTI) was performed using spin-echo EPI. The acquisition parameters were as follows: FOV = 16×8 mm2, TE/TR = 19/2500 ms, matrix size = 80×40, number of slices = 60, slice thickness = 0.20 mm without gap, voxel size = 0.20×0.20×0.20 mm3, b-value = 1000 s/mm2, diffusion gradient duration/separation time (∆/δ) = 3.5/10.0 ms, with a total of 32 diffusion directions and five b = 0 s/mm2 images acquired. No. of segments = 4, averages = 4, total scan time = 24 min 40 sec.
For tractography analysis, ex vivo DTI was acquired using spin-echo echo-planar imaging (EPI) with the following parameters: a 3D acquisition, FOV = 12×7.5×16 mm2, TE/TR = 32.5/250 ms, matrix = 120×75×160, voxel size = 0.10×0.10×0.10 mm3, b = 3000 s/mm2, ∆/δ = 3.0/21.0 ms, 60 diffusion directions, 5 b = 0 s/mm2 images, No. of segments = 4, average = 4, total scan time = 11 h 33 min 20 sec. Due to animal death (see “Preparation of ex vivo samples” section), limitations in scanner availability, and the extremely long scan time, only 5 mice per group for AD22, WT22, and AD40, and 6 mice for WT40 were scanned using this protocol.
Preparation of ex vivo samples
Following the MR scan, mice were intracardially perfused with 5 mM Gadopentetic acid (Gd-DTPA) (Adamas 86050-77-3). The brain sample was stored in a 15 ml tube with 4% PFA solution at 4℃ overnight for fixation. After this, the sample was moved to a 15 ml tube with 5mM Gd-DTPA saline and stored at 4℃ for at least 2 weeks for rehydration. On the day of ex vivo imaging, the brain samples were transferred into a 5 ml tube filled with Galden® perfluoropolyether (Solvay Specialty Polymers). Air bubbles were removed by vacuum before scanning. Due to the long scan time under urethane anesthesia and stress of perfusion setup, 2 AD22, 2 WT22, 2 AD40, and 4 WT40 mice did not survive long enough to be able to successfully perfuse and thus were unable to be used for ex vivo DTI.
fMRI processing
The fMRI and DTI data processing are shown in “Supplementary methods.” Our methods comparing DTI, fMRI, and behavior are in part based on a previous study by Kaneko and colleagues 35
For resting-state fMRI (rs-fMRI) analysis, functional images were pre-processed by brain extraction, slice-timing correction, and motion correction in MATLAB R2020b (https://uk.mathworks.com/products/matlab) with SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/) to adjust as if all slices were acquired simultaneously, eliminating motion artifacts. Functional images were then aligned with anatomical images within the same subject by rigid transformation and then anatomical images were co-registered to the Allen mouse brain atlas 36 in MATLAB R2020b and BioImage Suite Web (https://bioimagesuiteweb.github.io/webapp/). All images were smoothed with FWHM (full-width-half-maximum) of 5 to allow for better comparison between mice and to facilitate the group-based, ROI (region of interest)-based analysis in this study.
Group independent component analysis (group ICA) took place in MATLAB R2020b using Group ICA of the fMRI Toolbox (GIFT) (https://trendscenter.org/software/gift/), and the components related to cerebral spinal fluid (CSF) noise were removed 37.
Preprocessed data were divided into 5 epochs, with each epoch including 96 repetitions. We removed epochs which had a Pearson’s correlation coefficient (pcc) of < 0.1 between the left and right brain regions of lower forelimb primary somatosensory cortex (S1FL). This was undertaken to improve signal quality, as the S1FL region normally has high correlation if the mouse is in a good physiological state 32, 33. This resulted in the complete removal of 3 AD22, 2 WT22, 1 AD40, and 2 WT40 mice. (Note that behavioral data from these mice were not removed for calculation of average behavioral parameters.)
Regions of interest (ROI) were created for 25 regions using a labeled atlas of the template brain 38. Abbreviations are used as per the Allen atlas. Functional connectivity matrices were calculated using the pcc of the mean BOLD signals between different brain regions. Student’s t-tests were calculated using MATLAB and resulting p values were corrected for multiple comparisons by sequential goodness of fit metatest (SgoF) for family-wise error rate (FWER) of 0.05 39.
Global functional connectivity density (gFCD) was calculated by creating a correlation matrix for all voxels in the brain, then counting each voxel’s number of connections with other voxels as the number of voxels it correlated with above a threshold of 0.1 as per 40. Thus, gFCD reflected the level at which each voxel in the brain was a functional connectivity hub. In addition, gFCD was used to perform outlier removal. Mice whose data included greater than 3 standard deviations from whole brain mean gFCD were removed from behavior-gFCD analysis, this was only one AD22 mouse. (Thus, the total number removed including prior removals was 4 AD22 2 WT22, 1 AD40, and 2 WT40 mice.)
DTI processing
Pre-processing: both in vivo and ex vivo DTI datasets were initially converted into 4D NIFTI images using DSI studio (2022, Jan, 3) as described by Yeh 41. The dataset underwent denoising using nlsam (https://nlsam.readthedocs.io/en/latest/autoapi/nlsam/index.html). Summed DWI was utilized for brain extraction mask delineation in BrainSuite (Version 21a) (https://brainsuite.org/). Motion and Eddy current correction procedures were conducted.
The in vivo DTI metrics, including Fractional Anisotropy (FA) as an indicator of axon integrity; Mean Diffusivity (MD) and Axial Diffusivity (AD) as an indicator of axon injury; and Radial Diffusivity (RD) as an indicator of myelination, were computed in DSI studio (Version 2022 Jul) using the denoised DTI dataset. Q-Space Diffeomorphic Reconstruction (QSDR) was performed based on the Allen Mouse Brain Common Coordinate Framework version 3 (ABA). Subsequently, the average parametric values for each 3D brain region were extracted using MATLAB scripts (Mathworks, R2021b). Prior to further analysis, mice were excluded if visual inspection of their in vivo DTI images could not clearly differentiate white matter from gray matter, which resulted in 3 AD22, 3 WT22, 1 AD40, and 3 WT40 mice being removed from in vivo DTI results.
The ex vivo DTI connectome analysis employed QSDR reconstruction with the ABA mouse brain template. The tracking threshold was set at qa = 0.01, angular threshold at 45°, and minimum/maximum length at 1.0/40.0mm, with terminal seeds of 1,000,000. Connectivity analysis utilized 15 ROIs derived from rsfMRI analysis. For each mouse brain, a connectivity matrix was calculated using these 15 brain regions. Group comparison was conducted through a two-sample t-test across ages and genotypes. To visualize the connectogram displaying significant group differences, the averaged connectivity matrix with significant differences was plotted on the website of circos (http://mkweb.bcgsc.ca/tableviewer/visualize/).
Statistical analysis
Statistical analysis regarding behavior and also regarding the relationship between FC and both ex vivo and in vivo DTI were calculated in GraphPad Prism (version 8, GraphPad Software, San Diego, CA, USA). For behavior, unpaired two-tailed Student’s t-tests were used to compare the significance between different groups. The relationship between FC and both ex vivo and in vivo DTI was calculated by Pearson correlation. (Only mice whose data were neither excluded for fMRI nor ex vivo DTI were used for the FC versus in vivo comparison, and only mice whose data were not excluded for fMRI and who survived long enough to perform brain perfusion were used for the FC versus ex vivo comparison.) For each group, the mean slope from all linear regressions within the group was calculated. Unpaired two-tailed Student’s t-tests were used to compare groups with the relationship between FC and in vivo DTI. No multiple comparison corrections were performed for the behavior test or for the relationship between FC and in vivo DTI. (Note that when matching data from the mice in the correlation between FC and in vivo DTI, 5 AD22, 5 WT22, 2 AD40, and 3 WT40 mice had data removed due to a combination of exclusion criteria described previously.)