Mice
The number of mice handled for the presented work was approved by the Institutional and National General Veterinary Board (DGAV) and i3S internal Ethical Committee (approval reference number 003424), according to National and European Union rules. Three- to 12-month-old, male/female wild type C57BL6 mice were used. The animals were reproduced, maintained (regular rodents chow and tap water ad libitum) and experimentally manipulated under a 12h light/dark cycle in type II cages in specific pathogen-free conditions in the animal facility (microbiological heath status available). The method of euthanasia used was lethal anesthesia with 170 mg/kg ketamine/ 2 mg/kg medetomidine via (ketamine + medetomidine, intraperitoneal administration. Charles River Laboratories is the external animal facility used to acquire animals. Physical randomization for group selection was performed using a bioinformatics tool available from Graphpad Prism® QuickCalcs website: https://www.graphpad.com/quickcalcs/randomize1/. The order by which the animals from the different experimental groups were assessed was random during each experimental session. ARRIVE guidelines were taken into consideration in experimental reporting. This study was not pre-registered.
tMCAO
Mice were anaesthetized with isoflurane (4% for induction and 2-1.25% for maintenance) in a mixture of N2O:O2 (70:30) using a small-anesthesia system. Rectal temperature was maintained at 37.2°C throughout the surgical procedure making use of a rodent warmer with rectal probe (Stoelting, Wood Dale, IL, USA).
Cerebral blood flow (CBF) was monitored trans cranially using a laser Doppler flowmeter (LDF, PeriFlux System, Perimed, Stockholm, Sweden). Animals were placed under a stereo microscope (Leica S8 APO, Leica Microsystems, Wetzlar, Germany) and fur from the mice’s nape was shaved and disinfected. An incision of 1–2 cm was made on the clean skin, and the margins of the incision were pulled laterally to reveal the cranium. A 0.5 mm diameter microfiber Laser-Doppler probe (Master Probe 418-1connected to microtip: MT B500-0L240) was attached to the skull with cyanoacrylate glue 6 mm lateral and 2 mm posterior to bregma. While under general anesthesia, regional cerebral blood flow was monitored within the MCA territory. The surgical procedure was considered adequate if ≥ 70% reduction in blood flow occurred immediately after placement of the intraluminal occluding suture and reperfusion occurred after occlusion period; otherwise, mice were excluded.
Transient middle cerebral artery occlusion was induced by the intraluminal suture method 53 68 69 70. Briefly, after a laser-Doppler probe was attached to the skull, animals were placed in a supine position, under the stereo microscope. The surgical region of the neck was shaved and disinfected. A midline neck incision was made, followed by dissection of the underlying nerves and fascia to expose the right common carotid artery (CCA). The CCA was carefully separated from the vagus nerve, which lies laterally to it, taking extra care not to damage/stimulate or puncture it with the surgical tools. The right external carotid artery (ECA) and internal carotid artery (ICA) were also isolated by blunt dissection of the surrounding tissues (avoiding the rupture of the superior thyroid and occipital arteries). Afterwards, the CCA was temporarily ligated with a silk suture and the ECA was permanently ligated as distal as possible from the bifurcation of the CCA, and loosely ligated proximal to the bifurcation. A reduction in cerebral blood flow was observed. A microvascular clamp was applied to the ICA. A small incision was made in the ECA between the silk sutures, and a new silicone-rubber coated 6 − 0 nylon monofilament ([21–23] 6023PK10 Doccol Corporation, Massachusetts, USA) was introduced into the ECA and pushed up the ICA until the filament was ≈ 9 mm from the place of insertion (marked with a silver sharpie©) and a sudden drop was observed in the blood flow, effectively reaching the circle or Willis and blocking the MCA. The coating length of the monofilament lied between 2–3 mm, and the tip diameter depended on the animal’s weight. The loose suture around the ECA was then tightened around the inserted filament to prevent movement during the occlusion period, which lasted for either 45 or 10 minutes of occlusion. After designated occlusion time was over, the filament was removed and the proximal section of the ECA was permanently tied. The temporary tie around the CCA (only closed when necessary to avoid blood leakage through the ECA hole) was removed and reperfusion was established. This was observed in the Laser-Doppler as a rise in the regional CBF. Mice were administered 0.2mL of bupivacaine (2mg/kg in 0.9% NaCl) in the neck open wound. The Laser-Doppler probe was cut close to the skull and both the neck and head incisions were sutured. As a control, sham-operated animals were subjected to the same procedure without occlusion of MCA.
Immediately after surgery, a 0.5 mL saline (0.9% NaCl) and buprenorphine (50 µL/ 10 g of animal weight − 0.08 mg/kg) were given subcutaneously to each mouse as fluid replacement and as a form of analgesia, respectively. Buprenorphine was repeatedly administered at time intervals of 8 to 12 hours for the first 3 days post-ischemia; two daily administrations of 0.3 mL of both 5% glucose in 0.9% NaCl and 50/50% Duphalyte in NaCl were given until mice recovered weight. Right after surgery, mice were placed in a recovery box (Small Animal Recovery Chamber/Warming Cabinet, Harvard Apparatus) maintained at a humidified temperature of 36.8°C until fully recovered from anesthesia (~ 2h). To reduce stress, some elements such as paper from the home nest and roll papers were placed in the recovery box. When anesthesia completely wore off, recovery box temperature was decreased to 34.8° for the next 12 hours; afterward, home cages were half placed on a heating pad for the first day post-surgery, allowing mice to choose their environment during recovery, regulating body temperature and controlling post-stroke hypo/hyperthermia. When mice were returned to their home cage, significant care was taken into pairing mice with their pre-surgery cage mates. However, we separated sham-operated animals from tMCAO-operated animals to avoid stress and fighting within the cage, due to differences in the alertness and motile/cognitive deficits between the animals. Mortality rate was 10% (in tMCAO 45minutes only). After 3-, 24- or 72-hours of recovery, animals were euthanized.
Osmium tetroxide Stain
In the Osmium tetroxide cohort, mice were deeply anesthetized and were perfused transcardially (using 20ml syringes, through high-pressure manual delivery, keeping heart inflated) with 40 mL PBS, followed by 40mL of 2% PFA + 2.5% glutaraldehyde with 4% sucrose in PBS, pH 7.4. The whole brain was removed and fixed in 30 mL of the same fixation solution for at least 1 week, at 4°C, in a 50ml conical centrifuge tube.
After the post-fixation period, brains were washed in water to remove fixatives and PBS, and stained in aqueous 2% osmium tetroxide (OsO4) solution (15ml for each mouse brain) in 50ml conical centrifuge tube for two weeks, at 4°C in a horizontal shaker. Since osmium tetroxide is toxic if inhaled, tubes were sealed with Parafilm® to minimize evaporation and osmium tetroxide manipulation was always performed in a fume hood. When performing the staining, tubes were protected from direct light with aluminum foil. After staining, brains were washed, wrapped in Parafilm® (to prevent dehydration during micro-CT scanning) and transferred to Parafilm® sealed 5ml tubes (Sarstedt, 62.558.201) for µCT scanning, for short term storage. If long term storage is needed, brain dehydration followed by resin infiltration and embedding should be performed, as described by Masís et al 71. In this work, short term storing was the chosen option, mainly because one of the parameters to be analyzed was the brain edema after tMCAO. As such, the dehydration procedure could influence the measurement of this parameter, even though edema is always calculated in comparison with the contralateral hemisphere of the corresponding brain.
Detailed information regarding all the steps involved in ex-vivo mouse osmium tetroxide staining is available in Table 4.
Table 4
Detailed protocol for whole mice brain preparation for microCT with osmium tetroxide.
Step | Solution | Time (min) | Temp. (Cº) |
Perfusion |
Lethal anesthesia | Ketamine/ medetomidine, IP | 5 | 22 |
Blood perfusion | PBS | 10 | 22 |
Brain fixation | 2% PFA, 2.5% GA, 4% sucrose, in PBS | 10 | 22 |
Post-fixation |
Fixation | 2% PFA, 2.5% GA, 4% sucrose, in PBS | 7–30 days | 4 |
Washing | Distilled H2O | 1, 1, 1, 15 | 22 |
Staining |
Osmium | 2% OsO4, in H2O (15 mL/brain) | 15 days | 4 |
Micro-CT prep |
Washing | Distilled H2O | 1, 1, 1, 15, 15, 15 | 22 |
Preparation | Parafilm® wrapping, and 5 mL_ Tube accommodation | 5 | 22 |
Brain Storing |
Short-term | Protected from light; Sealed container | 3–12 months | 4 |
Long-term | Dehydration; Resin embedding | Years | 22 |
Micro-CT scanning for Osmium stained brains:
All osmium tetroxide stained ex-vivo mouse brains were scanned using Bruker micro-CT scanner (SkyScan1276, Bruker, Belgium), from the i3S Scientific Platform Bioimaging. The settings used were: voltage of 90 KV and an output current of 47 µA; 4 µm spatial resolution, 0.2º rotation step through 360 degrees, giving rise to 1801 projections; an Aluminum filter of 1 mm was used, together with a frame averaging of 4 The scanning of each brain took around 5h.
Reconstruction of scanned brain images was performed using the NRecon software (version 1.7.4.2, Bruker, Kontich, Belgium). The settings were normalized for all the brains, however fine tuning was performed for each individual brain to improve quality of deconvolution not achieved with normalized automatic deconvolution. Fine tuning parameters were: Smoothing of 0; Misalignment compensation between 52–60; Ring artifacts reduction between 4–6; and Beam-hardening correction between 20–30%.
Iodine staining
After the aforementioned recovery times, the animals from the iodine cohort were anesthetized and transcardially perfused (using 20 mL syringes, through manual delivery – keeping heart inflated) with 40 mL PBS, followed by 20 mL of 4% PFA in PBS, pH 7.4. Brains were carefully dissected and post-fixed in 20 mL of the same fixation solution for at least 3 days, at 4ºC, in a 50 ml conical centrifuge tube.
Brains were dehydrated in ethanol grade series (30%, 50%, 70%, 80%, 90%) for 2 h followed by a 24-hour staining in 90% methanol plus 1% iodine solution, in agitation, at room temperature. Since iodine solutions are not toxic, no protective measurements were not needed.
After staining, brains were quickly rehydrated with 30% and 70% ethanol, for 1 hour each, at room temperature with agitation. Finally, they were wrapped in Parafilm® (to prevent further dehydration during the micro-CT scan) for micro-CT scanning. For short-term storing, brains were kept in 5 mL tubes (Sarstedt, 62.558.201) at 4ºC. If long term storage is needed, brain dehydration, followed by resin infiltration and embedding should be performed, as described by Masís et al 16. In this work, short term storage was the chosen option.
Detailed information regarding all the steps involved in ex-vivo mouse iodine stain is available in Table 5.
Table 5
Detailed protocol for whole mice brain preparation for micro-CT with iodine.
Step | Solution | Time (min) | Temp. (Cº) |
Perfusion |
Lethal anesthesia | Ketamine/ medetomidine, IP | 5 | 22 |
Blood perfusion | PBS | 10 | 22 |
Brain fixation | 4% PFA, in PBS | 10 | 22 |
Post-fixation |
Fixation | 4% PFA, in PBS | | 4 |
Staining |
Dehydration | 30%, 50%, 70%, 80%, 90% ethanol | 2 hours, each | 22 |
Iodine | 1% iodine, in 90% metanol (15 mL/brain) | 24 hours | 22 |
micro-CT prep |
Preparation | Parafilm ® wrapping | 2 | 22 |
Brain Storing |
Short-term | Protected from light; Sealed container | 3–12 months | 4 |
Long-term | Dehydration; Resin embedding | Years | 22 |
Micro-CT scanning for iodine stain:
Iodine stained ex-vivo mouse brains were scanned using the Bruker micro-CT scanner (SkyScan1276, Bruker, Kontich, Belgium), from the i3S Scientific Platform Bioimaging, where the settings used were: voltage of 85 kV and an output current of 200 µA; 3 µm spatial resolution, 0.2° rotation step through 360°, giving rise to 1801 projections; an Aluminum filter of 1 mm was used, together with a frame averaging of 8. Scanning of each brain took around 3h.
Reconstruction of scanned brain images was performed using the NRecon software (version 1.7.4.2, Bruker, Kontich, Belgium). The settings were normalized for all the brains, however fine-tuning was performed for each individual brain to improve the quality of reconstruction not achieved with normalized automatic deconvolution. Fine-tuning parameters: Smoothing of 0; Misalignment compensation between 52–60; Ring artifacts reduction between 4–6; and Beam-hardening correction between 20–30%.
For higher resolution scans, used for striatum fiber automatic segmentation, stained samples were fixed in 1% agarose gel (Top-Bio s.r.o., Prague, Czech Republic) and placed in 1.5 mL centrifuge tube. GE Phoenix v|tome|x L 240 (Waygate Technologies GmbH, Huerth, Germany) was used for the acquisition. The micro-CT scan was carried out in an air-conditioned cabinet (21°C) at 60 kV acceleration voltage and 200 µA tube current. Exposure time was 700 ms and 3 X-ray projections were acquired in each angle increment and averaged for reduction of noise. The achieved resolution was 4.5 µm with isotropic voxels. Tomographic reconstruction was performed by using the software GE phoenix datos|x 2.0 (Waygate Technologies GmbH, Huerth, Germany). The data was then imported to VG Studio MAX 3.4 (Volume Graphics GmbH, Heidelberg, Germany).
Manual segmentation and quantification of total lesion/ core volume and edema using the CTAn software
Fine adjustment of the 3D volume images was performed using the DataViewer© software (version 1.5.6.2, Bruker, Kontich, Belgium), followed by volumetric analysis was performed on CTAn© software (version 1.20.3.0, Bruker, Kontich, Belgium). For edema extent calculation, brain hemispheres were manually annotated in the transaxial orientation every 10 slices. The software rendered the region of interest in every slice though interpolation and calculated object volume. In tMCAO 45-minutes brains, total lesion and core, when present, were manually delineated in the transaxial orientation of every 10 slices. Once again, the software adjusted the region of interest in every slice though interpolation and calculated object volume. Edema extent was calculated by applying the equation: Edema extent (% of the contralateral hemisphere) = ((V ipsilateral (µm3)- V contralateral (µm3))/ V contralateral (µm3))*100; and lesion volume was corrected for edema by applying the equation: Total lesion (% of brain volume) = (V lesion (µm3) x (V Contralateral (µm3)/ V Ipsilateral (µm3)*100)/ V brain (µm3); with V being the volume obtained during image quantification and V brain as the sum of V contralateral and Vipsilateral, based on 16.
Semi-automatic segmentation and quantification of white matter degeneration in the striatum
The first step required for the analysis of the white matter fibers was their segmentation in the whole CT volume. This segmentation was performed using the software Avizo 2020.2 (Thermo Fisher Scientific, Waltham, MA, USA). The tomographic slices were pre-processed by utilizing the non-local means algorithm for the reduction of noise. Then the caudate putamen was manually segmented in coronal tomographic slices utilizing the coronal Allen Mouse Brain Atlas as reference 72. Every 10th slice was manually annotated, and the in-between slices were computed by linear interpolation. These manually segmented volumes were used as a region of interest for subsequent segmentation of the white matter fibers. A semi-automatic approach was used for segmentation of the whiter matter: white top-hat transform. White top-hat transform detects light areas in the image using morphological operators. We utilized a ball-shaped structuring element 19 voxels in diameter. The top-hat-transformed image was then manually thresholded. This segmentation was performed simultaneously for both hemispheres, which allows accurate comparison. For the semi-quantification, a simple volume fraction analysis was performed in the Avizo 2020.2 (Thermo Fisher Scientific, Waltham, MA, USA), where the volume of the white matter was measured in relation with the volume of the manually segmented caudate putamen. The segmented striatum masks were then imported to VG Studio MAX 3.4 (Volume Graphics GmbH, Heidelberg, Germany) and wall thickness analysis was performed. For each voxel, the largest sphere inscribed to the white matter volume was determined, as containing the center position of the evaluated voxel. See Fig. 6A for the overview of the proposed workflow.
Automatic segmentation of the total lesion and core
An unsupervised deep learning-based approach for the segmentation of the total lesion and core volume was utilized for the automatic evaluation (See Fig. 6B). For the method to be applied, two conditions must be met: 1) the lesion should be isolated in one brain hemisphere, 2) the user must be able to distinguish the affected hemisphere from the healthy one. First, the brain midline is manually detected, and the brain scans are separated into two subvolumes along this midline. The subvolumes are all cropped to be of unified size of 1920 × 1280 pixels and downsampled 10 times to lower the computational requirements. The hemispheres unaffected by stroke form the training database for the deep learning model. We use a U-net-shaped CNN 73 (See Fig. 6C for the schematic representation of the CNN utilized in our work). The size of the CNN input and output is 192 × 128 pixels. The same as in the original U-net implementation, the CNN consists of an encoding and a decoding part. We use 3 × 3 convolutional kernels followed by 3 × 3 strided convolutions (stride 2) in the encoder to perform the repeated down-sampling of the extracted feature maps instead of the max pooling performed in the original implementation. In the decoding part, we apply 3 × 3 convolutional kernels and the up-sampling is performed by 3 × 3 transpose convolution layers with stride 2. The number of convolutional kernels in the first layer of the CNN is 64 and increases up to 1024 in the deepest part of the network. The feature maps from the encoder are concatenated with the decoder feature maps to preserve the spatial information from the input images during the training and inference. ReLU activation function is applied after each convolution. The input to the neural network is a coronal 2-D micro-CT cross-section of the brain hemisphere, where a lesion-like area is randomly simulated by subtracting two randomly generated concentric discs deformed by the elastic transform proposed in 74 (parameter alpha is set to 120 and sigma to 5) and blurred with by gaussian blurring with random parameter sigma (range from 5 to 20). The outermost ring simulates the whole lesion and the inner ring simulates the lesion core. The radius of the outer circle is defined randomly in the range from 20 to 80 pixels and its center is randomly set anywhere within the image. The radius of the inner circle is also defined randomly in the range from 0–50% of the outer circle radius. The decrease in intensity in the simulated lesion is random from 0–50% of the original intensity and is decreased further by up to 50% in the simulated lesion core.
The target of the CNN training is the original CT image of the healthy hemisphere without the simulated lesion area. The CNN thus learns to transform the simulated lesioned brain image into an approximation of images of a healthy brain. The CNN is trained on a total of 830 micro-CT images with randomly simulated lesion for 100 epochs with batch size 32 using the mean-squared-error loss and Adam optimizer 75 with initial learning rate of 0.001 (see Supplemental Fig. 6A). After the network is trained, it can be applied to real lesioned hemisphere image data. The CNN transforms the real 2D micro-CT cross-section of the stroke-affected brain hemisphere into an approximation of healthy brain hemisphere. By subtracting the lesioned image from the output of the CNN, lesion probability map is obtained. This probability map is then automatically thresholded in whole 3D volume by Li thresholding 76 to obtain the whole lesion area in 3D (see Supplemental Fig. 6B for visualization of the segmentation in selected tomographic cross-sections). Threshold for the lesion core is obtained as:
$${T}_{core }= \frac{3}{2}{T}_{lesion}$$
The largest connected component is selected for both the whole lesion area and lesion core thresholded image. The 3D segmented volumes are finally upsampled 10 times by bi-cubic interpolation and form the final segmentation masks and total lesion and core volume can then be computed. Nvidia Quadro P5000 with 16 GB of graphical memory was utilized for training the CNN on a system equipped with 512 GB of RAM and Intel® Xeon® Gold 6248R CPU. The proposed CNN was implemented in the programming language Python (version 3.7.9) using the library Keras 77 (version 2.3.1) and Tensorflow backend 78 (version 2.1.0). CUDA (version 10.1) and CUDnn (version 7.6.5) were used for GPU acceleration of the training and inference process. NumPy 79, scikit-image 80 and Pillow libraries were used for manipulating and transforming the image data.
TTC Stain
In the TTC cohort, infarct volume was evaluated as previously described in 58 with minor changes. After the mice were deeply anesthetized, the brain was removed, and the forebrain sliced into 1-mm-thick sections using a mouse brain slicer (Acrylic brain Matrix, Coronal, 40–75 g, Stoelting, USA) on ice. The sections were rinsed once in ice-cold 0.9% sodium chloride (NaCl) for 10 min and subsequently immersed in 25 mL of 0.01% 2,3,5-triphenyltetrazolium chloride (TTC) (Sigma-Aldrich, USA) in 0.9% NaCl at 37°C for 15 min. Slice images were acquired with a camera and analysed à posteriori. The unstained area in the section was regarded as an infarct area whereas the stained area was considered non-infarct area.
Infarct volume was determined using a semi-automated quantification method in Fiji-ImageJ 81. First, images were calibrated using a millimetric piece of paper as a reference, captured in all photos next to the slices. Next, images were batch processed and analyzed in a custom-made macro. Each image containing several slices of the same condition was pre-processed to isolate each slice, consisting in the manual selection of a colour threshold, followed by a Gaussian filter (sigma 10), thresholded with Huang’s method, and the morphometric operations fill holes and dilation. The output of the preprocessing step is the centroid of each slice. Finally, each slice was segmented alone by duplicating a rectangle, centered in the detected centroids, with size w x h (w = original image width, h = original image height/number of slices).
The infarct lesion measurement of each slice was performed by a blind experimenter with the following steps: i) manual delineation of the middle line to divide the slice into two hemispheres; ii) manual annotation of the stroke area (when present); iii) quality control of the resulting regions of interest. The macro output obtained was a results table with slice labels, the contra- and ipsilateral hemispheres area, and the infarct lesion area. Volumes were obtained by summing the area of the infarct brain segment from all brain slices (considering that each slice’s thickness is 1 mm). Edema extent and corrected lesion volume were calculated as described above and based on 82.
Image registration to Allen Mouse Brain Atlas
The whole image series from a tMCAO mice brain (stained with osmium) were registered to match the cutting plan and proportions of the atlas sections using the open-source serial aligner software QuickNII 83 (https://www.nitrc.org/projects/quicknii/). This software allows the assignment of spatial location (features such as angle, and size can be adjusted) so that atlas can match with section images. After, the whole image series segmented with the atlas was saved, and structures affected by ischemic lesion in this mouse were identified.
Brain slicing and immunolabeling
Brains were post-fixed in the same fixation solution, overnight, at 4°C and then left in 30% sucrose in PBS at 4ºC, until they sunk in the solution. Coronal sections (40 µm) were cut on a freezing microtome (Leica Cryostat CM 3050) and used for free-floating immunohistochemistry. Slices were incubated with 10% horse serum, 0.2% Triton X-100 in PBS, for permeabilization and blocking, for 1 hour, in agitation, at room temperature. Primary antibodies were always incubated for 48–72 hours at 4°C, in PBS. Secondary antibodies were incubated overnight at 4°C. The slides were mounted on gelatin-covered glass slides with a fluorescent mounting medium (DAKO) and imaging was performed on a laser scanning Confocal Microscope Leica SP8 (Leica mycrosystems, Wetzlar, Germany), using the ×10 air objective. In each set of experiments, the same batch of antibodies (primary and secondary) were used, and images were taken using the same settings. Primary antibodies used were anti-NeuN (1:500, Mouse, chemicon international, ab377), anti-MAP2 (1:500, Rabbit, abcam, ab246640), and anti-BrdU (TUNEL Assay Kit, Abcam, ab66110); as secondary antibodies Alexa Fluor 488 and 568 (1:500) were used. The fluorescent dye Hoechst 33342 (0.5 µg/mL–15 min room temperature) was used to stain nuclei.
Brains stained with iodine were rehydrated in two, 2-hours, PBS steps and then left in 25% sucrose in PBS at 4°C overnight. Coronal sections were cut and immunolabeled as described above.
Statistical Analysis
Statistical analysis was performed using GraphPad Prism Version 8.0/9.0 (GraphPad Software, Inc., San Diego, CA, USA). Data were collected from experiments and presented as mean ± SEM with individual points, “n” information (the experimental unit) is described in figure legends. The experimental unit was each single animal. Linear regression analysis was performed to correlate edema extent and total lesion values obtained either through microCT or TTC. For all statistical analysis: ****p < 0.0001, ***p < 0.001, ** p < 0.01, * p < 0.05, ns (not significant). Identification and removal of outliers was performed using automatic GraphPad prism software 8.0, using the ROUT (robust nonlinear regression) method, with a Q = 1% (recommended).