Animals in vivo
The study included 216 healthy common marmosets (88 male and 128 female, mean weight: 357.1 ± 60.2 g) aged 0.8–10.3 (mean: 4.34 ± 2.56) years. Healthy marmosets were selected from individuals without weight loss or viral infection in the previous two weeks. The common marmosets were anesthetized, and their heads were immobilized prior to imaging. The in vivo MRI scan was performed with each animal in the supine position on an imaging table under anesthesia with 2.0% isoflurane (Abbott Laboratories, Abbott Park, IL, the USA) in an oxygen-air mixture. Heart rate, SpO2, and rectal temperature were monitored regularly during imaging to manage the physical condition of the animals. This study was approved by the Animal Experiment Committees of the RIKEN Center for Brain Science (CBS) and conducted in accordance with the Guidelines for Conducting Animal Experiments of RIKEN CBS.
In vivo image acquisition
The marmoset brain MRI dataset (NA216) contains multimodal neuroimaging data including in vivo T1w, T2w, DWI, and rsfMRI images. MRI was performed using a 9.4-T BioSpec 94/30 unit (Bruker Optik GmbH, Ettlingen, Germany) and a transmit and receive coil with an inner diameter of 86 mm. For T1w imaging, a magnetization-prepared rapid gradient echo (MP-RAGE) was used, with the following parameters: repetition time (TR) = 6000 ms, echo time (TE) = 2 ms, flip angle = 12°, number of averages (NA) = 1, inversion time = 1600 ms, voxel size = 270 × 270 × 540 μm, and scan time = 20 min. For T2w imaging, rapid acquisition with relaxation enhancement (RARE) was used with the following parameters: TR = 4000 ms, TE = 22 ms, RARE factor = 4, flip angle = 90°, NA = 1, voxel size = 270 × 270 × 540 μm, and scan time = 7 min, 24 s. For DWI, spin-echo echo-planar imaging was used, with the following parameters: TR = 3000 ms, TE = 25.6 ms, δ = 6 ms, Δ = 12 ms, b-value = 1000 and 3000 s/mm2 in 30 and 60 diffusion directions, respectively (plus 4 b0 images), number of segments = 6, flip angle = 90°, NA = 3, voxel size = 350 × 350 × 700 μm, and scan time = 90 min. Diffusion metrics were generated using the diffusion tensor imaging (DTI) model, and the diffusion fiber connectome was generated using constrained spherical deconvolution15. For rsfMRI, gradient-recalled echo-planar imaging was used, with the following parameters: TR = 1500 ms, TE = 18 ms, number of shots = 1, flip angle = 40°, NA = 1, number of repetitions = 400, voxel size = 500 × 500 × 1000 μm, and scan time = 10 min.
Treatment of ex vivo imaging animals
Each animal was perfusion-fixed with 4% paraformaldehyde (PFA), and the brain was dissected from the skull and immersed in PFA for ex vivo imaging. During ex vivo imaging, the brain was wrapped in a sponge and soaked in fluorine solution, which does not show signal on MRI images, in a plastic container. Vacuum degassing was performed to reduce artifacts. The PFA solution used for fixation was replaced with fresh solution weekly to maintain fixation.
Ex vivo acquisition
MRI was performed using a 9.4-T BioSpec 94/30 unit (Bruker Optik GmbH) and a transmit and receive solenoid type coil with inner diameter of 28 mm. RARE was used for T2w imaging with the following parameters: TR = 10,000 ms, TE = 29.3 ms, RARE factor = 4, flip angle = 90°, NA = 16, voxel size = 100 × 100 × 200 μm, and scan time = 3 h, 20 min. For DWI, spin-echo echo-planar imaging was used, with the following parameters: TR = 4000 ms, TE = 28.4 ms, δ = 7 ms, Δ = 14 ms, b-value = 1,000, 3,000, and 5,000 s/mm2 in each of 128 diffusion directions (plus 6 b0 images), number of segments = 10, flip angle = 90°, NA = 2, voxel size = 200 × 200 × 200 μm, and scan time = 6 h, 39 mins.
Data processing pipeline
Structural image
The schematic of the processing pipeline for structural, diffusion and function MRI is shown in Fig. 1. To correct T2w images, whole brains were extracted from the image data using BrainSuite18a (David W. Shattuck, Ahmanson-Lovelace Brain Mapping Center at the University of California). Mask images were generated and a registration process was performed to align the standard brain images by mapping brain region data to the structural images of each animal. The analysis software ANTs (Brian B. Avants, University of Pennsylvania) was used for this process16.
To locate brain regions, we digitized the Atlas17 proposed by Hashikawa et al.18 in a 3D setting. Since the Hashikawa atlas was segmented by histology, whose resolution scale is extremely high for MRI data analysis, we merged the regional labels into 6 and 52 and 111 anatomically validated regions defined by the anatomist, which are suitable for both structural and functional MRI analysis.
The migration information from the standard brain image to the structural image of each animal was calculated. This information was superimposed on the brain region data to generate information corresponding to the structural image of each animal. The T1w/T2w approach was proposed by Glasser et al. in 2011, and it showed how to increase the contrast related to myelin content by calculating a simple ratio between T1w and T2w images19. Since this can be calculated from the ratio of T1w and T2w images, there was no need for novel imaging20.
Diffusion MRI
Pre-processing steps, such as artifact removal, were performed. These processes were performed using the brain image analysis tool Mrtrix3 version 3.0.3.12 (J-Donald Tournier, School of Biomedical Engineering & Imaging Sciences, King’s College London)21. The following commands were used in various processes: dwidenoize, mrdegibbs, dwipreproc, and dwibiascorrect. After image pre-processing, diffusion metrics were created using the DTI model. The diffusion fiber connectome was created using constrained spherical deconvolution (Tournier et al., 2004), and the ex vivo diffusion fiber connectome was created using high angular resolution diffusion-weighted MRI22. We used the MRTrix3 software for tensor analysis and fiber construction (dwi2tensor, tensor2metrics, dwi2response, dwi2fod, taken, and SIFT). We constructed diffusion metric images, axial diffusivity (AD) images, radial diffusivity (RD) images, fractional anisotropy (FA) images, and connectome matrices based on the number of fibers.
Resting-state functional imaging
Data pre-processing was performed using the SPM12 software package (Wellcome Department of Cognitive Neurology, London, UK) running under MATLAB (MathWorks, Natick, MA, USA). We then performed denoising steps using the functional connectivity toolbox (CONN). The empirical blood oxygenation level-dependent signals were band-pass filtered within a narrow band of 0.01–0.08 Hz. The analysis used fMRI data from a 20-min scan (initial 40 volumes discarded; subsequent 560 functional volumes) for awake data and a 10-min scan (initial 20 volumes discarded; subsequent 380 functional volumes) for anesthetized data. The empirical FC matrix was calculated using Pearson correlation between the average time courses of 104 brain regions for 3 awake and 31 anesthetized healthy common marmosets at rest and was averaged across the marmosets.
Brain regions
In this study, we used the anatomically segmented atlas of the common marmoset brain reported by Hashikawa et al.18, The atlas was applied to the data from 111 regions of one brain and 52 regions of another brain created by combining several of the 111 regions. The regions of interest (ROIs) are listed in ROImerge_data_v2.xlsx. In addition, the data were divided into six regions (cerebrospinal fluid, gray matter, deep gray matter, white matter, cerebellum, and brainstem) were obtained for large-scale segmentation, and were fitted to individual brain data. ANTs software was used to align the brain atlas to the individual brains.