The SCA34-KI rat model was generated on the Long-Evans background using CRISPER/Cas9 to knock-in the c.736T > G (p.W246G) mutation in Elovl4 that causes human SCA34 [20] as previously described [14, 37]. Rats were kept on a 12 h light:12 h dark cycle (light intensity of 25–40 lx at cage level) with food and water available ad libitum. All rats used in these studies were 120 days of age. Animal procedures were approved by the Institutional Animal Care and Use Committee of the University of Oklahoma Health Sciences Center and met National Institute of Health guidelines.
Tissue PreparationRats were anesthetized using ketamine (100 mg/kg body weight) and xylazine (5 mg/kg body weight), then perfused using 4% paraformaldehyde in 0.1 M phosphate buffered saline (PBS) through the left ventricle as described previously [12]. The brain was harvested and fixed in 4% PFA at 4 °C for an additional seven days. The brain was rinsed thoroughly with PBS, and stored in PBS until use. To prepare tissue sections, the cerebellum was isolated and hemisected along the midline, embedded in 7% agarose gel, and sectioned in the sagittal plane using a vibratome at 40–60 μm thick. All sections from within a given specimen were cut at the same thickness, based on the optimum for that specimen, and were stored in 1x PBS at 4˚C for up to 3 months (in the absence of preservatives). There was no systematic selection of sections for immunolabeling.
Analysis of Cortical Layers and Purkinje Cell DensityFree-floating vibratome sections were stained with 0.1 M Toluidine Blue for 10 min on a slide warmer at 59˚C. Sections were rinsed in PBS at room temperature with gentle agitation, then cover-slipped using Mowiol or Prolong Gold. Image montages of stained cerebellar sections were acquired using a Leica M205-MFC or a Nikon Ti2 inverted microscope empirically calibrated for scale. The width of the various layers of the cerebellar cortex were measured manually using ImageJ software as described previously [12]. To ensure consistent sampling, images of cerebellar cortex were acquired on linear portions of lobule 5 (inferior side of the lobule), lobule 8 (superior side of the lobule), and lobule 10 (superior side of the lobule) (Fig. 1). A line perpendicular to the white matter was drawn across each cortical layer to measure width. Linear density of Purkinje cells (PCs) in the Purkinje Cell Layer (PCL) was calculated by counting the number of PCs and dividing the number of cells by length of the region of the PCL counted [12]. Analysis was performed using well-oriented sections displaying Purkinje cells in a discrete monolayer to prevent any potential miscounting of Purkinje cells. These analyses were performed using sections prepared from female rats.
Fig. 1Sampling for quantitative analyses in sagittal cerebellar sections. boxes indicate the regions in lobules 5, 8, and 10 sampled for cell population analysis. Scale bar: 1 mm
ImmunohistochemistrySagittal vibratome sections were rinsed in 1x PBS, then incubated for 2 h at room temperature in blocker (10% Normal Goat Serum, 5% Bovine Serum Albumin, and 0.5% Triton in 1x PBS). Well established primary marker antibodies for labeling of specific cerebellar cell populations (see details in Table 1) were diluted in blocker. Nuclei were labeled using DAPI (5 mg/ml) applied together with the primary antibodies. Sections were incubated in primary antibodies in a humidified container for seven days at room temperature. Sections were rinsed in 1x PBS, then incubated in appropriate secondary antibodies (Table 1) for two hours at room temperature, then rinsed in 1x PBS with gentle agitation. Sections were mounted onto glass slides in Mowiol for imaging. All immunolabeling combinations were performed in all animals whenever possible, however, in some cases there were not sufficient sections available to perform all possible labeling combinations.
Workflow for Neuronal QuantificationQuantitative analysis of neuronal populations in WT, HET, and MUT rat cerebellum was based on a computer-assisted workflow to identify specific types of cerebellar neurons using immunolabeling for known markers for specific types of cerebellar neurons followed by confocal imaging, computer-assisted image segmentation, and cell counting (Fig. 2). Quantitation for each cell-specific marker was performed on minimum of 27 images from each animal in the analysis.
Fig. 2Workflow for cell quantitation. A. Images of neurons immunolabeled for cell-specific markers are obtained by confocal microscopy. B. Images are selected for analysis. Density of the immunolabeled cells of interest determines whether an entire cortical layer can be analyzed in a single image, or if analysis of a layer must be performed using multiple regions that encompass the entire layer. C. Multiple images are used to train a Cellpose 2.0 model to recognize the cells of interest. Once trained, the model is applied to the remaining images in the dataset. The trained model directs subsequent image segmentation to identify cells of interest or regions of interest (ROI), which are marked individually as a ROI that can be recognized by Image J. D. Following identification of individual cells of interest as independent ROIs in Cellpose 2.0, images are imported into Image J, and Image J is used to count the number of cells of interest (ROIs) in each image
Image AcquisitionTo minimize any effects of tissue laterality, all analyses were performed using sections obtained from the cerebellar vermis (between 0 and 1800 μm from the cerebellar midline). Sections for immunolabeling were selected at random from within this range to avoid any systematic bias. Sampling was performed in linear regions of lobules 5, 8, and 10 as described above (Fig. 1). Images were acquired from well-labeled sections using a Nikon W1-CSU Dual Spinning Disk Confocal microscope fitted with a Hamamatsu Orca Fusion BT sCMOS camera. Images of immunolabeling were collected as Z-stacks of 12 μm total depth at 0.3 μm step size, using a 60x objective lens with 1.4 numerical aperture. Image resolution was 0.10833 μm/px.
Image Preparation for QuantificationDue to the intrinsic differences in the density and location of the various types of cerebellar neurons in the layers of the cerebellar cortex, cell type-specific quantification required selection of regions of interest for cell counting from single X-Y image planes. Analysis was performed from multiple, non-overlapping Z-planes from within each image stack to prevent repeated sampling of individual cells. For granule cells, which are packed at very high density in the granule cell layer (GCL), images of labeling for NeuN and DAPI in the granule cell layer were cropped into 100 μm by 100 μm tiles for analysis. For analysis of molecular layer interneurons (MLIs) and displaced granule cells, which are present at low density in the molecular layer (ML), tiles spanning the entire width of the molecular layer were cropped from specimens labeled for FOX2, NeuN, and DAPI. For analysis of unipolar brush cells (UBCs), which are present at low to moderate density in the GCL, tiles spanning the entire width of the GCL were selected for analysis from specimens labeled for Calretinin and DAPI.
Selection of Image Sample Planes and Image CurationAn ImageJ macro was developed to select three non-overlapping image planes from each Z-stack for analysis: a plane from the top of the Z-stack, a plane from the middle of the Z-stack, and a plane from the bottom of the Z-stack, to prevent any repeated counting of individual cells. Out of focus image planes were excluded from analysis. Image tiles for analysis of GCs with less than 80% of the image area filled with GCs were excluded, to prevent under-sampling of GCs.
Model Training for Cell Quantification Using Cellpose 2.0Cellpose 2.0 is a semi-supervised machine learning algorithm for image segmentation through an iterative human-machine training process [40]. A cell-specific model for cell counting that required colocalization of neuron-specific marker labeling and nuclear DAPI labeling was developed for each type of neuron to ensure that all objects counted were genuine neuronal cell bodies, not fragments or processes of a cell. Accordingly, training the model for each cell type required a different number of training images to achieve accurate identification of the cell of interest. Training the GC model required 50 + images; the MLI model required 30 + images; the displaced granule cell model, required 60 + images; and the UBC model required 150 + images. A criterion of at least five cells of interest within the image tile was required for an image to be included in any model.
Image SegmentationThe cell-specific segmentation model was applied to each image in a data set using Cellpose 2.0 and a python script created by the developers of Cellpose 2.0 and run on Jupyter Notebook [40]. The developers of Cellpose 2.0 also provided an Image J macro to create ROI.zip files for each image analyzed by Cellpose 2.0 (available at: https://github.com/MouseLand/cellpose/blob/main/imagej_roi_converter.py [40].
QuantificationTo quantify the numbers of each type of cerebellar neuron, the ROI.zip files generated during segmentation by Cellpose 2.0 were opened using the ImageJ ROI.zip macro, which reported the number of segmented cells in each image into a result table that was then exported into a CSV file for quantitative analysis of cell density.
For analysis of each cell population, sampling for cell counting for each cell-specific marker included a minimum of three non-overlapping image planes, from three distinct regions of interest in each of at least three independent sections per rat (3 planes x 3 ROIs x 3 sections = a minimum of 27 images analyzed for each individual rat).
Data SummarizationFor each cell type of interest, the density of cells identified by colocalization of a neuron-specific marker and DAPI were averaged to calculate the cell density of that type of neuron in lobules 5, 8 and 10 for each animal. These data were then complied to determine the density of each type of neuron in each lobule for WT, HET, and MUT SCA34-KI rats.
Statistics and GraphingStatistical analysis and graphing were performed using GraphPad Prism. Data were analyzed using one-way ANOVA with Tukey’s Post Hoc test. A Bonferroni correction was applied to correct for multiple one-way ANOVAs per cerebellar lobule. Alpha was set to 0.017 and was calculated as 0.05/3 to adjust for the three lobules analyzed for each animal. Data were plotted as violin plots, with the mean cell density (cells/mm2) or mean cell size (area in µm2) shown as a solid line across the violin; dotted lines show the 25th to 75th percentile range. To show the distribution of the measurements, the mean value for each individual animal in the sample are shown as individual data points (solid for females, open for males).
Figure PreparationTo prepare immunolabeling figures, images were imported into Adobe Photoshop. Image scale was calibrated, and adjustments of brightness and contrast were made to highlight image features. Adjustments of brightness and contrast were applied equally to all pixels in the entire image. Graphs were prepared using GraphPad Prism.
Comments (0)