Turone Sheep Brain Atlas and Template toolkit. (A) Axial, coronal, and sagittal view CT template (Top) and MRI template (Down). Cross hairs showing the position of the lambdoid suture which is used as reference point within the TSBTA space. (B) 3D representation of the TSBTA space. Yellow shaded mesh represents the skull and crosshair represent the reference point of the skull (lambdoid suture position). (C) Sagittal view of TSBTA brain atlas. (D) Anterior (top) and posterior (down) 3D view of the thalamus segmentation.
To cope with seasonal modifications of the environmental resources, brain anticipates and changes both its structural organisation and functioning. However, the tempo and mode of these central changes have been poorly investigated in mammalians. Here we describe a longitudinal morphometric neuroimaging study in a well know animal model to study seasonality: sheep. Using new magnetic resonance imaging (MRI) resources comprising a high-resolution brain template, its associated tissue priors (500-µm isotropic resolution) and a corresponding sheep brain atlas (202 regions of interest) we investigate the impact of seasonal transitions between winter and summer season on brain microstructure using voxel-based morphometry. We observed significant modifications of grey matter concentration (GMC) in pivotal brain areas involved in circadian rhythms (pineal, hypothalamus) and light processing (suprageniculate nucleus) but also within regions related to sensory processing, learning, memory, behavior control, and social cognition. These findings provide new insights into mammalian brain functioning revealing its flexibility and its adaptability to cope to environmental changes.
Seventeen adult, sexually mature multiparous ewes (Ovis aries) of 3.4 ± 0.3 years old (weight = 64.5 ± 5.5 kg) have been included in this protocol and have been scanned at the PIXANIM platform (INRAE, French National Research Institute for Agriculture, Food & Environment, Nouzilly, France). Females were ovariectomized and implanted with an oestradiol silastic implant (2 cm) at the end of october, two months before the first scan session. The ewes were kept permanently indoors and fed ad libitum with dehydrated lucerne, maize, straw, and a supplement of vitamins and minerals, and had free access to water. All the procedures were conducted in accordance with the European directive 2010/63/EU on the protection of animals used for scientific purposes and the experimental protocol was approved by the local ethical committee (comité d’éthique en expérimentation animale Val-de-Loire) under reference number 00510.02.
First, ewes were fasted 24 h before surgery. Day of surgery, an i.v. injection of thiopental (14 mg/kg body weight, BW; Nesdonal, Merial, Villeurbanne, France) was done to induce analgesia, then animals were intubated and maintained under anesthesia by a mixture of 3 to 4% isoflurane (Vetflurane, Virbac, Carros, France) vaporized in 100% oxygen. Ovariectomy were conducted under sterile surgical conditions. Local anaesthesia with lidocaine (4%, Lurocaïne, Vétoquinol, Luré, France) was given prior laparotomy. Ovaries were surgically extracted, and the tissues were sutured. The entire procedure was performed within 20 min. Postoperative ventilation with oxygen was maintained until the first signs of awakening appeared. Animals were then housed individually for 6 h in a padded stall before being put back with congeners. They received an anti-inflammatory drug for 2 days (2 mg/kg, flunixin meglumine, Finadyne®, Intervet, Beaucouzé, France) to relief pain and an antiedema medication: (1 mg/kg BW of furosemide, Dimazon®, Intervet, Beaucouzé, France) at the end of surgery. Animals were treated with a diuretic medication associating 3 mg/kg of hydrochlorothiazide with 0.03 mg/kg dexamethasone (Diurizone®, Vétoquinol, Luré, France) for 2 days.
Two scanning sessions were performed for each animal at opposite moment of the circannual cycle: a first session was performed between mid-January/February (from 13/01/2014 to 17/02/2014), during the sexual season and a second session was performed between mid-June and mid-July (from 16/06/2014 to 22/07/2014) during the season of sexual rest. As 4-5 weeks were necessary to scan the entire group, animals were housed two weeks before and during each scan session (2-3 weeks) in photoperiodic facilities to ensure similar light duration for each subject and to avoid a photoperiodic shift over the scanning sessions (winter scanning session: mean daylight 556.1 min ± 22.26 min; summer scanning session: daylight 943.3 min ± 13.90 min). Therefore, ewes were housed under a 9h light/15h dark photoperiod (light on at 8AM, daylight duration 540 min) during the winter scan session and under a 15h light/9h dark photoperiod (lights on at 6h AM, daylight duration 960 min) during the summer scan session (Fig. S1).
To ensure that animals were responsive to the photoperiodic treatment, one week before the start of each MRI scan session, serial blood samplings were performed overnight to measure the pattern of blood melatonin secretion. This was done following previous procedures used in our laboratory (Tricoire et al., 2002). Briefly, animals were housed individually 24h hours and implanted with a catheter into the jugular before blood sampling to limit potential stressful response of the animals. Blood sampling was performed under dim red light once per hour from one hour before the lights turn off until 2 hours after they turn on. Blood samples were then centrifugated and plasma stored at -80°c until assay. Melatonin assay was performed as previously documented in our laboratory (Tricoire et al., 2002) (Fig. S1).
Animal preparation for MRI data acquisitions were performed as previously described (Ella et al., 2015, 2017). Briefly, animals were anesthetized with an intramuscular injection of ketamine just before the scan, intubated, maintained during the scan on 3% isoflurane vaporized in oxygen and continuously monitored by a MR-compatible Aestiva®/5 systems (Madison, USA). Three MR acquisitions were performed (see Ella et al., 2015 for the details of each acquisition) on each animal secured in a prone position in 3 Tesla VERIO Siemens systems (Erlangen, Germany), front legs apart and bent towards the abdomen, using a flexible coil (Siemens FLEX Large 4 elements) tied around the head. The sequences have been optimized to be perform in time compatible with anaesthesia (≤1h), to reduce artefact (folding, truncation, etc.) and to optimize SNR. For each acquisition parameters have been set as previously describe in Ella and Keller (2015):
- Three dimensional SPC-IR acquired in the sagittal plane (Echo Time/Repetition Time = 413 ms/4000 ms, Flip Angle = 120°, Inversion Time = 380 ms, Number of Excitation = 10, Partial Fourier = 1, Slice Thickness = 0.35 mm, Slice Number = 208, Field of View = 179.2x179.2 mm, matrix = 512x512, final resolution 0.35 mm3).
- Three dimensional T1 MPRAGE acquired in the sagittal plane (Echo Time/Repetition Time = 3.18 ms/2500 ms, Flip Angle = 12°, Inversion Time = 900 ms, Number of Excitation = 8, Partial Fourier = 1, Slice Thickness = 0.5 mm, Slice Number = 288, Field of View = 192x192 mm, matrix = 384x384, final resolution 0.5mm3).
- Three dimensional T2 MEDIC acquired in the sagittal plane (Echo Time/Repetition Time = 2.1 ms/38 ms, Flip Angle = 8°, Number of Excitation = 6, Partial Fourier = 0.75, Slice Thickness = 0.4 mm, Slice Number = 256, Field of View = 179.2x179.2 mm, matrix = 448x448, final resolution 0.4 mm3). At the end of the acquisitions, animals were transferred from the scanner to a recovery room, anaesthesia was stopped, and mechanical ventilation was maintained. From the first signs of awakening, animals were extubated and placed under constant surveillance in a dedicated box. When fully awake, the animal is returned to the photoperiodic facilities with its conspecifics until the end of the data acquisition campaign. DICOM data from the scanner were converted to NIFTI format and organized as a standardized data sets accordingly to the Brain Imaging Data Structure (BIDS) using BIDScoin and are downloadable on Zenodo.
The first version of the TSBTA was created from 18 scans of ovariectomized Iles de France ewes (3.4 years) and implanted with an oestradiol silastic implant (2 cm) at least two months before the scan session. Similar MRI acquisitions have been performed and a multimodality template (T1w and T2w) have been created using linear and nonlinear normalisations procedures scripted with FSL5.0 software library (FMRIB, University of Oxford, UK). Additionally, three prior probability maps (grey matter = GM, white matter = WM, and cerebrospinal fluid = CSF) were generated using FSL-FAST for automatic segmentation of the ovine brain (Ella and Keller, 2015). Finally, an atlas of the entire brain have been created delimiting twenty-five cortex gyri and twenty-eight inner sheep brain structures within the cerebrum (Ella et al., 2017). From this previous environment we updated both the templates and prior probability maps using cutting edges methods proposed by the ANTs environment (Advanced Normalization Tools). Indeed, the accuracy and the specificity of probabilistic maps (GM, WM, CSF) had the potential to be improved. Additionally, within the first version of the brain atlas, the thalamic region was not fully segmented, and this gap had to be filled.
Before post-processing steps, SPC-IR, T2 MEDIC and T1 MPRAGE acquired for each animals (34 scans by contrast) were noise and signal bias corrected using Ginkgo and N4BiasFieldCorrection respectively and coregistered to the first version of the TSBTA using antsRegistrationSyNQuick. Hence, both SPC-IR and T2 MEDIC images were resampled to a voxel size of 0.5×0.5×0.5 mm3. Together, all coregistrated data (T1w, IR, T2w) were used to segment each brain using antsAtroposN4 to create GM, WM and CSF probabilistic maps of each subject. In parallel, we used modelbuild, an optimized pipeline using antsMultivariateTemplateConstruction2, an unbiased template building method developed in ANTs package. This method has been integrated to our local cluster (ISLANDe Facilities) and require qbatch and slurn. For computation, our cluster comprises 240 physical cores, each associated with 9.6GB of memory (2.3TB of memory in total), with 90TB of long-term storage and 72TB of high-performance storage. Once both linear and non-linear (flowfield maps) were calculated for each animals, we compilate all the transformations calculated for each images and applied them once to denoised and signal bias corrected images (SPC-IR, T2 MEDIC and T1 MPRAGE) using antsApplyTransforms to limit interpolation effects. SPC-IR, T2 MEDIC and T1 MPRAGE images have been used to create the second version of the TSBTA templates by calculating the mean image of each normalize contrasts using Ginkgo. Probabilistic maps (GM, WM, CSF) have been normalized using the linear and non-linear (flowfield maps) purchased by modelbuild and the antsApplyTransforms command and the second version of the GM, WM and CSF priors have been created by calculating the mean image of each normalized map using Ginkgo.
The first version of the TSBTA T1 MPRAGE atlas have been linearly and non-linearly coregistrated to the second version TSBTA T1 MPRAGE template using antsRegistrationSyNQuick and both linear and non-linear transformations have been apply to the first version of the TSBTA atlas using antsApplyTransforms to create the second version of the TSBTA atlas (Fig. S2). Each ROIs have been visually inspected to check boundaries and accuracy of registration. Some regions of interest (ROIs) such like Corpus Callosum, arbor vitae, and ventricular systems have been updated using WM and CSF priors for a best fit to the new version of the templates. Eventually, the thalamus was fully segmented manually using fsleyes and itksnap platforms and ROIs have been labelled according to the work previously published in humans (Iglesias et al., 2018)
Tomographic data of nine ewes’ skulls have been acquired on our CT-scan (Siemens Somatom Definition AS, Siemens Corp., Germany). The X-ray source was set at 100 kV and 120 mA/s. A total of 800 slices were acquired using the following parameters: Thickness = 0.4 mm, Slice Number = 800, Field of View = 204,8x 204,8 mm, matrix = 512x512, final resolution 0.4 mm3) reconstructed using a filter Safire I26. DICOM data were converted to NIFTI format and organized as a standardized data sets that accordingly to the Brain Imaging Data Structure (BIDS) using BIDScoin and are downloadable on Zenodo. To create a CT template from our data, we first performed a linear coregistration of all the data to the first animal using antsRegistrationSyNQuick. Then we used modelbuild, to create both linear and non-linear transformations for each animal. All the transformations calculated previously were compilated and applied once to raw CT images using antsApplyTransforms to limit interpolation effects. Spatially, normalized CT images were used to create a temporary version CT template by calculating the mean image of the normalized dataset using Ginkgo. Finally, the CT template was coregistrated to the second version of the Turone T1 MPRAGE template manually using the Anatomist tool of the BrainVISA suite and resampled at the same spatial resolution (0.5×0.5×0.5 mm3). The manually coregistrated CT template was then linearly and non-linearly coregistrated to the second version of the Turone T1 MPRAGE template antsRegistrationSyNQuick. Eventually, all the transformations calculated from raw to now were compilated and applied once to raw CT images using antsApplyTransforms and the final version CT template was created by calculating the mean image of the normalized dataset using Ginkgo (Fig. S3).
For voxel-based morphometry analysis (VBM), we used the T1 MPRAGE data (T1) which have the higher contrast to noise ratio in our dataset and which have been acquired at the same resolution than the second version of the TSBTA T1 template. For each animal (n = 17) two images are available (Winter and Summer). Data were first denoised and signal bias corrected using Ginkgo and N4BiasFieldCorrection respectively and linearly coregistered to the TSBTA T1 weighted template using antsRegistrationSyNQuick. Coregistrated data were then pre-processed with SPM. Each image was segmented into probability maps of GM, WM, and CSF using the default settings in the SPM8 toolbox and the new version of GM, WM, and CSF probability maps. The transformation matrices obtained were used to normalize GM, WM and CSF probability maps obtained for each subject. GM and WM probability map of each scanning sessions were normalized to our stereotaxic space using transformations matrices obtained herein and resampled. Normalized GM and WM images were used to create a more population-specific template using diffeomorphic anatomical registration using exponentiated lie algebra (DARTEL) (Ashburner, 2007; Ashburner and Friston, 2005, 2000). Each normalized GM image was then warped using deformation parameters calculated by the DARTEL routine of SPM and then modulated to correct the volume changes that may have occurred during the deformation step. Finally, normalized-warped-modulated GM images were spatially smoothed by convolving with a 4 mm full width at half-maximum (FWHM) isotropic Gaussian kernel to create GMC maps (Fig. S4).
To assess regional GM changes between the two seasons over all the animals, GMC maps obtained during the winter period were compared to those obtained during the summer period for each animal using a paired t test proposed by SPM. A brain mask was used to constrain the analysis to brain. For each cluster, the significance of the peak voxel was set as p < 0.005 (t-score = 2.92, degree of freedom = 16). The results are presented on an axial, coronal and sagittal brain slice series generated with nilearn.
VBM significant cluster revealed SPM analysis data were identified using our atlas and a personal procedure developed in MATLAB. For each comparison, ROI masks from the second version of the TSBTA atlas were used to extract GMC values of corresponding regions within the GMC map using REX plugging. Group comparison (Winter versus Summer) of endocrine, and GMC data were compiled and analysed using GraphPad Prism 10.2.0 software, and were compared using a Wilcoxon matched pairs signed rank test. Statistical significance was defined as p < 0.05 for these analyses.
Seasonal grey matter modifications. Sagittal (A) Coronal (B) and axial (C) slices showing grey matter concentration (GMC) differences between winter and summer. SPM paired Student t-test analysis. Voxel-level threshold p < 0.005, t (16) = 2.92. BNST = bed nucleus of the stria terminalis ; Hy = hypothalamus ; Hip = hippocampus ; pHipp = parahippocampal cortex ; PAG = periaqueductal grey substance ; Cg = cingulate cortex ; sGen = suprageniculate nucleus ; Occ = occipital gyrus ; Clau = claustrocortex (insula) ; Rec = rectus gyrus ; OB = olfactory bulb ; Pir = piriform cortex ; Amyg = amygdala ; Te = temporal gyrus; Hb = habenular nucleus.