Automated analysis of gray matter damage in aged mice reveals impaired remyelination in the cuprizone model

Abstract

Multiple sclerosis is a chronic autoimmune disease of the central nervous system characterized by myelin loss, axonal damage, and glial scar formation. Still, the underlying processes remain unclear, as numerous pathways and factors have been found to be involved in the development and progression of the disease. Therefore, it is of great importance to find suitable animal models as well as reliable methods for their precise and reproducible analysis. Here, we describe the impact of demyelination on clinically relevant gray matter regions of the hippocampus and cerebral cortex, using the previously established cuprizone model for aged mice. We could show that bioinformatic image analysis methods are not only suitable for quantification of cell populations, but also for the assessment of de‐ and remyelination processes, as numerous objective parameters can be considered for reproducible measurements. After cuprizone‐induced demyelination, subsequent remyelination proceeded slowly and remained incomplete in all gray matter areas studied. There were regional differences in the number of mature oligodendrocytes during remyelination suggesting region‐specific differences in the factors accounting for remyelination failure, as, even in the presence of oligodendrocytes, remyelination in the cortex was found to be impaired. Upon cuprizone administration, synaptic density and dendritic volume in the gray matter of aged mice decreased. The intensity of synaptophysin staining gradually restored during the subsequent remyelination phase, however the expression of MAP2 did not fully recover. Microgliosis persisted in the gray matter of aged animals throughout the remyelination period, whereas extensive astrogliosis was of short duration as compared to white matter structures. In conclusion, we demonstrate that the application of the cuprizone model in aged mice mimics the impaired regeneration ability seen in human pathogenesis more accurately than commonly used protocols with young mice and therefore provides an urgently needed animal model for the investigation of remyelination failure and remyelination‐enhancing therapies.

In the current study, we use a modified cuprizone model of delayed demyelination and limited remyelination to characterize de‐ and remyelination processes, glial reactions, as well as synaptic, axonal and neuronal damage in different gray matter regions in‐depth by utilizing novel bioinformatic image analysis approaches. Upon cuprizone administration, synaptic density and dendritic volume in the gray matter of aged mice decreased. The intensity of synaptophysin staining gradually restored during the subsequent remyelination phase, however the expression of MAP2 did not fully recover.

1. INTRODUCTION

Multiple sclerosis (MS) is a chronic inflammatory disease of the central nervous system (CNS) that results in primary demyelination of white and gray matter regions [1, 2, 3]. During recent decades it has been increasingly recognized that gray matter damage is frequent and extensive in patients with MS [4, 5]: Demyelination, alteration or reduction of synapses, axons, and neurons, as well as atrophy have been demonstrated for different gray matter structures such as the cerebral cortex and the hippocampus [5, 6, 7, 8]. Importantly, gray matter damage has been shown to correlate with cognitive impairment and the severity and progression of overall clinical disability [9, 10, 11, 12]. Remyelination is considered a potent neuroprotective repair mechanism following demyelination. It appears to be more pronounced in gray matter lesions compared to white matter lesions but often remains insufficient [13, 14]. Therefore, appropriate animal models are crucial to examine the process of remyelination and remyelination‐promoting therapies with the goal to advance regenerative therapeutic approaches for people with MS [15, 16]. For the accurate characterization of histopathological processes and MS research in general, the importance of AI‐assisted automation processes is increasingly being recognized, as these can allow for high objectivity and increased precision in research by assuring reproducibility across large sample sizes and avoiding human bias [17, 18].

We established a protocol for the toxic demyelination cuprizone model in aged mice, characterized by impaired remyelination in white matter to enable the investigation of the causes of remyelination failure and to study the effect of potential remyelination‐supporting agents [19]. In the current study, we use this modified cuprizone model of delayed demyelination and limited remyelination to characterize de‐ and remyelination processes, glial reactions, as well as synaptic, axonal and neuronal damage in different gray matter regions in‐depth by utilizing novel bioinformatic image analysis approaches.

2. MATERIALS AND METHODS

2.1. Animals

Male C57BL6/J mice were obtained from Charles River (Sulzfeld, Germany). The mice underwent routine cage maintenance once a week and were microbiologically monitored in accordance with the Federation of European Laboratory Animal Science Associations (FELASA) recommendations [20]. Water and food were freely available. All animal procedures were approved by the Review Board for the Care of Animal Subjects of the district government of Lower Saxony (LAVES), Germany; approval number: 15/1762) and carried out according to international guidelines on the use of laboratory animals.

2.2. Induction of de‐ and remyelination

Six‐month‐old male C57BL6/J mice with a mature adult phenotype [21] were fed with a concentration of 0.4% (w/w) cuprizone (bis‐cyclohexanone oxaldihydrazone, Sigma‐Aldrich) mixed into ground standard rodent chow (maintenance diet, rats/mice, Altromin, Lage, Germany). Cuprizone was fed for up to 6.5 weeks for the induction of demyelination. Optimum cuprizone concentration and treatment duration for complete demyelination of the white matter structure corpus callosum were previously established [19]. Demyelination status was assessed at weeks 5, 6, and 6.5. After the treatment period of 6.5 weeks, mice were fed with normal rodent chow and remyelination was analyzed 0.5, 1.5, 2.5, and 3.5 weeks after the end of cuprizone feeding (week 7, 8, 9, and 10). Age‐matched male C57BL6/J control mice were fed normal rodent chow without cuprizone. For each time point during de‐ and remyelination, 3–6 animals were examined with up to 4 replicates per animal, depending on tissue and image availability and quality.

2.3. Tissue processing

At each time point, mice were perfused with 4% paraformaldehyde in phosphate buffer through the left cardiac ventricle, as established previously [22, 23]. Next, the brains were removed and post‐fixed in 4% paraformaldehyde before paraffin embedment. 7 μm serial coronal paraffin sections were cut with a rotary microtome (RM2245, Leica) for light microscopy. Sections between bregma −0.82 mm and bregma −1.94 mm in accordance with the Paxinos and Franklin mouse brain atlas [24] were analyzed.

2.4. Immunohistochemistry

Paraffin‐embedded slices were first dewaxed and then heat‐unmasked in 10 mM citrate buffer (pH 6.0). For immunostaining, the following primary antibodies were applied: for myelin, anti‐proteolipid protein (PLP, mouse monoclonal IgG2a, 1:500, BIO‐RAD); for microglia, anti‐ionized calcium‐binding adaptor molecule 1 (IBA1, rabbit polyclonal IgG, 1:200, Wako); for mature oligodendrocytes anti‐adenomatous polyposis coli (APC, mouse IgG2b, 1:200, Calbiochem); for astrocytes, anti‐glial fibrillary acidic protein (GFAP, polyclonal rabbit IgG, 1:200, Dako); for axonal damage, anti‐Synaptophysin (mouse monoclonal IgG1, 1:200, BIO‐RAD); for dendrites, anti‐microtubule‐associated protein 2 (MAP2, rabbit polyclonal, 1:1000, Millipore); for neuronal nuclei, anti‐NeuN (1:200, Millipore). Antibodies were diluted in PBS containing 0.3% Triton‐X100. Sections were then incubated with biotinylated secondary antibodies and subsequently with peroxidase‐coupled avidin‐biotin complex (ABC Kit, Vector Laboratories). For immunofluorescence double staining, sections were incubated with secondary antibodies Alexa Fluor 555 goat anti‐rabbit IgG (H + L) and Alexa Fluor 488 goat anti‐mouse IgG (H + L) (all 1:500, Invitrogen). Staining of cell nuclei was carried out using either Mayer's hemalum solution (Merck) for hematoxylin–eosin staining for 3, 3′‐Diaminobenzidine (DAB)‐based cell analysis or 4′, 6‐diamidino‐2‐phenylindole (DAPI, Invitrogen) for immunofluorescence staining.

2.5. Determination of demyelination and quantification of glial reaction using QuPath software

2.5.1. Image processing was standardized and based on image metadata

Image processing was carried out in accordance with digital imaging guidelines for pathology [25, 26]: During image capturing, the brightness and contrast of each fluorescence image was normalized to account for image comparability. In this manner, brightfield images were normalized against a white point in the background of each image, respectively. Both fluorescence and brightfield image acquisition were performed by script with fixed parameters. Files were exported in native .tif (fluorescence) or .svs (brightfield) format and imported directly into the respective analysis software to conserve metadata. For brightfield images, stain vectors were adjusted globally for each image in order to provide a staining intensity normalization for comparable cell detection—a process known as color deconvolution [27]. No other preprocessing was performed (no contrast enhancement, smoothing or down sampling was applied), thus all measurements were carried out based on the original metadata. See the data availability statement for a link to representative example images and the detection and classification models.

2.5.2. Myelin assessment must incorporate multiple factors

The extent of cuprizone‐induced demyelination of the cerebral cortex was first assessed manually [23]: The cortical area of sections stained for PLP was scored by three independent and blinded observers using a light microscope with a magnification of 200× (Olympus BX61, Olympus). The different grades of myelination were assessed using a scale from 0 (complete loss of myelin staining), to 4 (normal, fully myelinated area) as established previously [23].

Independently, a pixel‐based classification algorithm was trained. After whole slide scanning of the tissue specimens (Leica Aperio AT2, Leica), images were imported into QuPath software (version 0.2.2) [28] and preprocessed: Stain vectors were adjusted for each image to account for comparability. Each section was then annotated with separate polygons encircling the left and right dentate gyrus (DG), the Cornu Ammonis (CA) 3 and 1 region, as well as layer I–IV and V–VI of the cerebral cortex, respectively, each as a separate annotation. See Appendix B1 for a representative annotation image.

Separately for cerebral cortex and hippocampus, the images were split in training, test and validation samples, and were manually annotated with multiple positive and negative area samples. Using a feedback loop, the quality of detection was iteratively increased by further annotations on heterogenous images. For sensitive detection, a full resolution (0.50 μm/pixel) of input images and a morphological closing prefilter with a smoothing sigma of 0.5 μm proved to achieve optimal results in general. Because of high heterogeneity of myelin morphology between de‐ and remyelination, some alterations of the input and the smoothing prefilter of the classifiers were required in order to precisely depict myelin structures during de‐ and remyelination, respectively. Also, two classifiers were trained separately for hippocampus and cortex because of differences in myelin‐structure and distribution. The resulting, region‐specific classifiers were then validated by three independent observers using masked images to ensure a concordance ≥0.95. See Appendix A1 for a detailed description of the workflow.

2.5.3. Cell‐by‐cell workflows for APC, GFAP, IBA1

For the analysis of cellular reactions, image files were acquired and preprocessed as described above. For segmentation of DAB images, a representative subset of images (n = 27) with different markers and staining intensities was randomly split into training, validation and test samples, on which rectangles were drawn randomly in the hippocampus, cerebral cortex or corpus callosum (n = 79). For transfer‐learning, a pre‐existing H&E versatile model [29] was applied to the pictures, which itself was trained on images from the Multi‐Organ Nuclei Segmentation Challenge (MoNuSeg) 2018 training data [30, 31] and the TNBC dataset [32]. On this basis, all un‐ or mistakenly recognized cells were annotated by hand. The resulting annotated dataset is publicly available for further cell segmentation research under https://doi.org/10.7910/DVN/JGEQPS [33]. Resulting images and annotations were then exported and used for training of a vector‐based cell detection via StarDist [29], a TensorFlow‐based machine‐learning software relying on star‐convex polygons for cell detection. After validation, the resulting neuronal network was applied to the annotations inside the original data set to detect all cells inside. After precise cell detection was established, an object classifier had to be trained for each stain respectively to distinguish between positive and negative cells. Each classifier was trained using a random‐tree classification algorithm, based on a total of 67 parameters (e.g., nucleus and cell hematoxylin and DAB‐staining intensity, cell size, ‐morphology, and ‐expansion). This training process was carried out on a subset of the original data containing random image‐regions (n = 134). The data‐subset was split into training (n = 67), validation (n = 34) and test samples (n = 33) for each stain respectively, each with an input of around 15.000 cells. For validation, three experienced researchers assessed the performance of the cell classifier in various conditions on a cell‐to‐cell basis to assure a concordance of ≥0.95. An overview of the analysis‐workflow is depicted in Appendix A2. Results of cell counting refer to the number of cells per mm².

2.5.4. Alternative approaches for NeuN neuronal staining

The approach described above is not suitable for a neuronal stain in the hippocampal formation, as neurons are mostly located densely inside the stratum pyramidale and the granular layer of the dentate gyrus. Thus, other layers need to be excluded from the measurements in order to reduce area variability that would interfere with quantification.

For training of a robust area‐classification model, the original dataset was used as a template on which annotations around the hippocampal neuron layers were drawn manually until a stable prediction tool was established, as independently confirmed by three experienced researchers. The resulting objects were then used to define regions on the original images, in which subsequent cell detection was carried out. As neurons are more evenly distributed in the cortical layers of the cerebral cortex, annotations were drawn manually as done for the myelin and cell‐by‐cell workflows and afterwards analyzed by script. An example image with annotations for both hippocampal and cortical regions is given in Appendix B2. After cell detection, advanced cell measurements were performed, for example, for assessment of staining intensity, cell area and neighborhood analysis. In general, the cell numbers are given in cells per square millimeters. To obtain these measurements, the digital pathology approach allowed annotating large regions for cell detection and classification. For accurate validation, further analysis of the spatial distribution was performed. A Delaunay triangulation algorithm was used to compute numerous metrics for spatial distribution analysis, from which two parameters are especially valuable for density evaluation: The mean triangle area quantifies the average size of triangles formed by connecting neighboring cells within a given cluster. It is computed by dividing the total area of all triangles within the cluster by the number of triangles. The mean distance quantifies the average interpoint distance among data points within a given cluster. It is computed by summing the distances between all pairs of data points within the cluster and dividing this sum by the number of pairs. This metric facilitates the assessment of spatial compactness or dispersion within clusters [34]. To selectively compute distance metrics for neuronal nuclei, all NeuN‐negative cells were removed after cell classification. Then, Delaunay triangulation was computed for all remaining cells with a distance threshold of 100 μm.

2.5.5. Synaptophysin

In order to measure synaptic density as a marker closely linked to neuronal damage, brain sections were stained and fluorescence images captured using an automated whole‐slide scanner (Vectra Polaris, Akoya Biosciences). Standardized exposure protocols were set up to gain images with intercomparable metadata. This was further assured by the selection of a stable fluorophore, a standardized cleaning before slide‐loading, standardized exposure protocols as commanded by the scanning‐software and finally the exclusive use of metadata‐preserving image formats. After annotation of the resulting images, the local background was subtracted by batch analysis and fluorescence intensity was measured with a resolution of 0.5 μm/pixel, given as mean inside each annotation. Results were compared as intensity measurements in arbitrary units (AUs).

2.6. Statistical analysis

For statistical analysis, GraphPad Prism Version 9.4.1 (Graphpad Prism Inc.) was used. For each time point, the mean value of all brains analyzed was calculated across all samples and replicates. First, normal distribution was assessed using the Kolmogorov–Smirnov test. As there was no matching or pairing, statistical analysis was then carried out using analysis of variance one‐way ANOVA, followed by Bonferroni's multiple comparison test or Kruskal‐Wallis test followed by Dunn's multiple comparison test when appropriate. The comparison of manual myelin scoring with the myelin‐classification algorithm was conducted in R version 4.2.1 (June 23, 2022) [35]. To obtain comparability between the two measurements, the data was first standardized using a Z‐transformation. Subsequently, a two‐sample t‐test was performed under the assumption of equal variance separately for each time point. For comparison of fluorescence and DAB‐staining, unpaired t‐tests for each time point were performed after assessment of normal distribution.

Data are presented as arithmetic means with standard error of the mean (SEM). Significant effects are shown by asterisks (compared to another time point) or hash marks (compared to controls) (ns = not significant, */#p < 0.05; **/##p < 0.01; ***/###p < 0.001) and are shown in the respective figures. The significance bars with asterisks represent the level of significance between a reference timepoint and its respective counterparts. If multiple counterparts show significance at the same level compared to the reference, they are marked with downward‐sloping ticks, while the reference timepoint itself has no tick mark.

3. RESULTS

3.1. Feeding 0.4% cuprizone results in severe but incomplete demyelination and impaired remyelination in the cerebral cortex of aged mice

To assess the effect of a 6.5‐week cuprizone feeding protocol in inducing severe demyelination in the cortex of aged mice, we first examined the appearance of PLP‐positive myelin fibers in this area. Initially, three independent and blinded researchers manually assessed the cortex on histological sections stained for PLP, using a scoring system previously established (Figure 1A, light grey) [23, 36]. Demyelination was evident after 5 weeks of cuprizone administration with a mean myelin score of 3.0. Myelin loss continued to progress during the feeding period until it reached its maximum, though remaining incomplete, after 6.5 weeks of cuprizone treatment (mean myelin score of 0.83). After the end of cuprizone feeding at 6.5 weeks gradual remyelination was detectable but remained incomplete until the end of the observation period 3.5 weeks after cuprizone cessation, as reflected in the mean myelin score of 1.92.

FIGURE 1.

Proteolipid protein (PLP) staining during de‐ and remyelination in the cerebral cortex and hippocampus of aged mice. (A) Compares conventional manual scoring of the cortex (light grey) as gold standard with automated analysis (dark grey = control, red = de‐/blue = remyelination). (B) Compares the automated analysis of fluorescence images (light grey) with DAB images (dark grey = control, red = de‐/blue = remyelination). As seen in (C), DAB sections exhibit a significant decrease of PLP‐positive myelin fibers during demyelination (red; 5, 6, 6.5 weeks), and slow and incomplete restoration during remyelination (blue; 7, 8, 9, 10 weeks. (C, upper left) shows the areas which were analyzed. (D) Provides representative DAB and fluorescence images. Bars represent mean + SEM. Significant effects in comparison to control are indicated by hashtags, significance between different time‐points is indicated by asterisks (*/#p < 0.05; */##*p < 0.01; ***/###p < 0.001). Control = no treatment. 5, 6, 6.5 weeks = cuprizone treatment period. 7, 8, 9, 10 weeks (0.5, 1.5, 2.5, 3.5 weeks after cuprizone cessation) = remyelination period after treatment. N = 3–6 animals per group with 1–4 replicates.

3.2. Bioinformatic measurement of cortical myelinated areas precisely detects de‐ and remyelination in aged mice

To further corroborate our findings, a pixel‐based classifier was trained to measure the PLP‐positive area in the cerebral cortex for each time point (Figure 1A, dark grey/red/blue) using QuPath software. Whereas in control mice 20.8% of the examined cortical region were classified as PLP‐positive in the cortex, after administration of cuprizone for 6.5 weeks only 5.3% of the area were positive for PLP. Even 3.5 weeks after the end of cuprizone treatment (“10 weeks”), only 9.2% of the area were classified as PLP‐positive, with no tendency to further increase after 1.5 weeks after the removal of cuprizone.

The course of de‐ and remyelination, as assessed by pixel classification of the whole cerebral cortex, was congruent to manual scoring (Figure 1A), with a score of 4 (full myelination) corresponding to 20% of the cortex being classified as myelinated, a score of 3 corresponding to 15%, a score of 2 measuring up to 10%, and a score of 1 conforming to 5% myelinated cortical area. Besides, there were no significant differences between the size of the detected PLP‐positive area in fluorescence or brightfield images (Figure 1B).

Upon further subdivision of the cortex by examining layers I–IV and layers V–VI separately, the time‐dependent pattern of PLP‐expression in both regions was found to be similar (Figure 1C, upper row). With only 14.4% of the area in layers I‐IV classified as PLP‐positive in control mice, the relative positive area in the first four layers was lower, as compared to 39.6% in layers V/VI. During maximum demyelination at 6.5 weeks, both regions displayed a reduction of 75.9 and 83.7% of PLP‐positive myelinated area. 3.5 weeks after the end of cuprizone administration myelination remained incomplete reaching 7.2% of the examined area in layers I–IV and 16.0% in layers V–VI compared to control.

3.3. Cuprizone administration leads to extensive and lasting demyelination of the hippocampus

The hippocampus of aged mice exhibited severe demyelination already after 5 weeks of cuprizone feeding (Figure 1C, lower row): Whereas in control mice 13.3% and 16.9% of the DG and CA3 region were classified as PLP‐positive, the PLP‐positive area after 5 weeks of cuprizone administration amounted to 0.02% and 0.01%, respectively. In the CA1 region, there was a detectable and significant decrease of the PLP‐positive area from 6.1% to 1.1% at 5 weeks with an even more pronounced reduction after 6 and 6.5 weeks. After cessation of cuprizone feeding, gradual remyelination was observed. At 10 weeks, 6.0% and 7.6% of the originally myelinated area in the DG and CA3 region were remyelinated, whereas in the CA1 a region of 3.5% was myelinated. The new myelin appeared to be of a different structure and unevenly distributed (Figure 1D, 10 weeks) as compared to the myelinated fibers in control tissue. In summary, remyelination after cuprizone‐induced demyelination of the cerebral cortex and hippocampus of aged mice occurs slowly and is incomplete with slight regional differences between different gray matter structures.

3.4. Cuprizone administration results in significant loss of oligodendrocytes in the cerebral cortex and hippocampus of aged mice while reconstitution displays high regional heterogeneity

To analyze the effect of cuprizone‐induced demyelination and the subsequent remyelination period on mature oligodendroglial cells in different gray matter structures in aged mice, APC‐positive mature oligodendrocytes were examined. An object classifier was used to objectively assess the loss of APC‐positive cells (as is visualized in Figure 2B, left half of each example image).

FIGURE 2.

Depletion and repopulation of adenomatous polyposis coli (APC)‐positive mature oligodendrocytes in the cerebral cortex and hippocampus of aged mice during de‐ and remyelination. (A, upper left) shows the areas, which were analyzed. Graphs (A) depict numbers of APC‐positive cells in the cortex and hippocampus of aged mice during cuprizone‐induced de‐ (red; 5, 6, 6.5 weeks) and remyelination (blue; 7, 8, 9, 10 weeks, corresponding to 0.5, 1.5, 2.5, and 3.5 weeks of remyelination) compared to control (gray; no treatment). (B) Shows cells as categorized by an object‐classification algorithm (left half: green = APC‐positive, red = APC‐negative) and the corresponding raw image (right half, respectively). Bars represent mean + SEM. Significant effects in comparison to control are indicated by hashtags, significance between different time‐points is indicated by asterisks (*/#p < 0.05; **/##p < 0.01; ***/###p < 0.001). N = 3–6 animals per group, 1–4 replicates.

A profound reduction of APC‐positive oligodendrocytes was apparent in the cerebral cortex after 5 weeks of cuprizone treatment and remained until 6.5 weeks of cuprizone administration. Already 0.5 weeks after cuprizone cessation, the number of oligodendrocytes had recovered back to levels in control animals in all cortical areas and persisted throughout the remaining remyelination period (Figure 2A, upper row). In the hippocampal regions assessed, the decrease of APC‐positive oligodendrocytes progressed throughout cuprizone administration, with a minimum amount observed at 6.5 weeks for all areas (Figure 2A, 2nd row). The loss of oligodendrocytes was especially pronounced in the DG and CA3. In the CA1 area, considerable repopulation was already evident 0.5 weeks after cuprizone cessation. In the DG and CA3 region, oligodendrocyte numbers slowly increased but remained reduced throughout the remyelination period until the end of observation 3.5 weeks after the end of cuprizone treatment (“10 weeks”).

It is notable that regional differences in oligodendrocyte recovery corresponded with the respective regional remyelination capacity in the hippocampus. There, fast oligodendrocyte repopulation followed by considerable remyelination was evident in the CA1 region, whereas persistently low oligodendrocyte numbers and lasting demyelination was observable in the CA3 and DG region. In contrast, the cerebral cortex showed insufficient remyelination despite rapid and near complete oligodendrocyte recovery.

3.5. A significant transient reduction in the volume of the cerebral cortex and the stratum pyramidale occurs during demyelination without obvious loss of neurons

To examine the cerebral cortex in respect to changes in size during de‐ and remyelination, brain sections were annotated with a rectangle spanning 1.5 mm laterally from the midline of the corpus callosum, which was then trimmed to exactly match the respective extent of the cerebral cortex of each section as illustrated in Figure 3A. Additionally, the stratum pyramidale of the hippocampal CA region was analyzed. Both gray matter areas displayed a significant reduction in volume at 6 and 6.5 weeks of cuprizone treatment (see Figure 3B). However, this effect was only transient with a normalization of gray matter volume during remyelination. To assess the effect of cuprizone administration on neurons in the cerebral cortex and hippocampus of aged mice, histological sections stained for NeuN were analyzed with regard to cell count/mm² and size of NeuN‐positive nuclei. In the cortex, no significant changes in number per area or size of NeuN‐positive cells occurred during cuprizone‐induced de‐ and subsequent remyelination (see Figure 3D, E, left half, respectively). In brightfield sections of the hippocampus, the high number of NeuN‐positive cells in the stratum pyramidale of the CA region resulted in a compact structure too dense to allow for sufficient segmentation, so fluorescence images were acquired. Here, a transient reduction in NeuN‐positive neuronal nuclei size was measurable in the CA during demyelination at weeks 6 and 6.5 (Figure 3E, right half). Simultaneously, the numbers of NeuN‐positive neurons/mm² did not change significantly during de‐ and remyelination (Figure 3D, right half). According to Delaunay triangulation cluster analysis of the hippocampal CA region, both parameters of intercellular distance (mean distance and in mean triangle area) exhibited a significant reduction at 6.5 weeks, while normalization was visible shortly thereafter (Figure 3F, right half; data not shown for mean triangle area). In the cortex, however, Delaunay triangulation cluster analysis revealed no significant changes in mean distance or in mean triangle area (data not shown) in response to cuprizone treatment (Figure 3G, left half).

FIGURE 3.

Area analysis and NeuN‐positive neuronal cell‐measurements in the cerebral cortex and hippocampus of aged mice during de‐ and remyelination. (A) Shows the regions, which were used for area analysis. (B) Depicts changes in cortical area (left) and NeuN‐positive area as a parameter for size of the hippocampal stratum pyramidale region (right) over the course of the experiment. (C) visualizes the regions as used for neuron size and number calculations. In (D), the number of NeuN‐positive neurons/mm² is depicted over time for the cortex (left) and hippocampus (right). (E) Illustrates the size of NeuN‐positive neuronal nuclei for respective time points in the two regions assessed. In (F), mean distance between neighboring NeuN‐positive cells over time is shown as calculated by Delaunay Triangulation, separated by region. In (G), representative areas of the cerebral cortex (V–IV) and the stratum pyramidale of the CA3 region from DAB‐ (right half) and fluorescence‐images (left half, respectively) are displayed. On each fluorescence image, gray lines render the Delaunay triangulation measurements. Bars represent mean + SEM. Significant effects in comparison to control are indicated by hashtags, significance between different time‐points is indicated by asterisks (ns = not significant, */#p < 0.05; **/##p < 0.01; ***/###p < 0.001). N = 3–6 animals per group, 1–4 replicates.

Taken together, the transient decline of the stratum pyramidale in the hippocampal CA area was accompanied by a reduction of the size of NeuN‐positive neuronal nuclei and signs of higher neuronal cell density seen with a reduction of distance between NeuN‐positive cells and a small trend towards a higher cell count per area at 6.5 weeks of cuprizone treatment, although this effect was not significant. In contrast to this, the cerebral cortex showed no significant changes in the density or size of NeuN‐positive neuronal nuclei despite transient volume reduction.

3.6. During cuprizone‐administration, synaptic density decreases in line with a loss of dendritic volume in the cerebral cortex and hippocampus of aged mice

To assess the effects of cuprizone treatment on synapses and neuronal dendrites, the gray matter areas of aged animals were examined for synaptophysin and MAP2‐expression. First, synaptic density was quantified by measurement of synaptophysin staining intensity in AU, corresponding to the fluorescence intensity of the respective images that were captured via a standardized exposure protocol to account for comparability. These differences in intensity as compared to control were visualized using a standardized heatmap (Figure 4B, respective left half). The synaptic density as measured by synaptophysin intensity continuously decreased during cuprizone treatment, reaching a minimum shortly after cessation of cuprizone treatment in week 7. During the subsequent remyelination period, synaptophysin staining intensity increased gradually and remained incomplete for 2–3 weeks after the end of treatment, with slight regional variations.

FIGURE 4.

Synaptophysin intensity in the cortex and hippocampus of aged mice during cuprizone‐induced de‐ and remyelination. (A, upper left) depicts the areas, which were analyzed. Graphs (A) show analysis of mean fluorescence intensity during demyelination (red; 5, 6, 6.5 weeks) and subsequent remyelination (blue; 7, 8, 9, 10 weeks) compared to control (gray; no cuprizone treatment). Differences in intensity of representative fluorescence images (B right half, respectively) visualized by a heat map (B left half, respectively; green = high intensity, blue = low intensity). Bars represent mean + SEM. Significant effects in comparison to control are indicated by hashtags, significance between different time‐points is indicated by asterisks (*/#p < 0.05; **/##p < 0.01; ***/###p < 0.001). N = 3–6 animals per group, 1–4 replicates.

To further assess neuronal and especially dendritic damage caused by cuprizone‐induced demyelination in aged mice, the MAP2‐positive area was analyzed (Figure 5). Upon cuprizone administration, a reduction of the MAP2‐positive area in the cerebral cortex as well as in the hippocampus was observed. (Figure 5A) In the hippocampal regions, the decrease was evident slightly earlier than in the cortex, and a significant reduction in MAP2‐positive area also beyond the cuprizone exposure period was observed in all regions studied. During the subsequent remyelination period, a slow recovery of the MAP2‐expression was observed across all gray matter regions with slight regional differences (Figure 5A, blue bars).

FIGURE 5.

MAP2‐positive area in the cerebral cortex and hippocampus of aged mice during cuprizone‐induced de‐ and remyelination. (A, upper left) indicates the areas, which were analyzed. Graphs (A) show analysis of MAP2‐positive area as assessed by pixel‐classification during demyelination (red; 5, 6, 6.5 weeks) and subsequent remyelination (blue; 7, 8, 9, 10 weeks) compared to control (gray; no cuprizone treatment). Fluorescence images (B) visualize the morphology exemplarily for three time points. Bars represent mean + SEM. Significant effects in comparison to control are indicated by hashtags, significance between different time‐points is indicated by asterisks (*/#p < 0.05; **/##p < 0.01; ***/###p < 0.001). N = 3–6 animals per group, 1–4 replicates.

3.7. Microgliosis persists in gray matter of aged animals after cuprizone‐induced demyelination

Staining for IBA1 was used to investigate microglia accumulation in the cerebral cortex and hippocampus during cuprizone‐induced demyelination and following remyelination. In aged mice, the different gray matter areas showed distinctions regarding the speed and magnitude of microglia recruitment as measured by the number of IBA1‐positive cells (Figure 6A). However, around the time point of maximum demyelination (6.5 weeks), all areas studied revealed significant microgliosis compared to control. Interestingly, the extent of the microglial reaction tended to either persists or increase even after termination of cuprizone administration. Only after 1.5 weeks after cuprizone cessation the extent of microgliosis in the cerebral cortex and hippocampus began to decline. Even after 3.5 weeks of remyelination, an elevated number of microglia seemed to persist in the different gray matter regions compared to control, although this elevation proved not to be significant.

FIGURE 6.

Accumulation of IBA1‐positive microglia in the cortex and hippocampus of aged mice at different time points compared to control. (A, upper left) shows the areas, which were analyzed. Graphs (A) depict numbers of IBA1‐positive cells in the cortex and hippocampus of aged mice during demyelination (red; 5, 6, 6.5 weeks) and remyelination (blue; 7, 8, 9, 10 weeks) compared to control (gray; no treatment). Representative images (B) show cells as categorized by an object‐classification algorithm (left half, green = positive, red = negative) and the corresponding raw image (right half, respectively). Bars represent mean + SEM. Significant effects in comparison to control are indicated by hashtags, significance between different time‐points is indicated by asterisks (*/#p < 0.05; **/##p < 0.01; ***/###p < 0.001). N = 3–6 animals per group, 1–4 replicates.

3.8. Short‐lasting astrocytosis is most pronounced at maximum demyelination in gray matter of aged mice

In the cerebral cortex as well as in CA3 and CA1 hippocampal regions, the number of GFAP‐positive astrocytes was constantly progressing during the cuprizone treatment period, reaching the peak at 6.5 weeks (Figure 7A), concurrent with maximum demyelination (Figure 1C, 6.5 weeks). In the DG, a near 5‐fold increase was already evident after 5 weeks of cuprizone exposure and reached the maximum at 6.5 weeks, as did all cortical areas considered. During remyelination phase, astrogliosis rapidly declined in both the cortex and hippocampus of aged mice, almost reaching control levels at 3.5 weeks after cessation of the cuprizone diet. These results differ from those of young mice, in which astrogliosis was previously shown to peak in the early demyelination phase and to remain at an elevated level during remyelination, especially in the cerebral cortex [23, 37].

FIGURE 7.

Reaction of glial fibrillary acidic protein (GFAP)‐positive astrocytes in the cortex and hippocampus of aged mice for different timepoints compared to control. (A, upper left) shows the areas, which were analyzed. Graphs (A) depict numbers of GFAP‐positive cells in the cortex and hippocampus of aged mice during demyelination (red; 5, 6, 6.5 weeks) and remyelination (blue; 7, 8, 9, 10 weeks) compared to control (gray; no treatment). Representative images of GFAP‐stained sections (B) show cells as categorized by an object‐classification algorithm (left half, green = positive, red = negative) and the corresponding raw image (right half, respectively). Bars represent mean + SEM. Significant effects in comparison to control are indicated by hashtags, significance between different time‐points is indicated by asterisks (*/#p < 0.05; **/##p < 0.01; ***/###p < 0.001). N = 3–6 animals per group, 1–4 replicates.

4. DISCUSSION

In the last two decades, gray matter pathology in MS and its crucial impact on cognitive impairment and physical disability has increasingly moved into the focus [12]. Notably, remyelination as a natural repair process occurs in gray matter lesions but often remains incomplete and is associated with reduced efficacy during disease chronicity [38, 39, 40]. Although mouse models constitute a crucial element in the study of MS pathology and repair processes, current protocols are either limited to the analysis of mainly the peripheral inflammatory components of the disease, or fail to adequately describe the impact of an agent on remyelination caused by the high natural regenerative capabilities of young mice, which are used almost exclusively [16]. To adequately study de‐ and subsequent remyelination in different gray matter areas we fed 6‐months‐old mice with 0.4% cuprizone for 6.5 weeks, with the goal to further validate a previously established optimized cuprizone mouse model. [19]. We furthermore aimed to establish protocols for reproducible and automated data analysis, reliant on bioinformatical image analysis.

Using the methodology described above, we confirmed severe demyelination in the cerebral cortex and hippocampus of aged mice after 6.5 weeks of 0.4% cuprizone administration. However, demyelination in the cerebral cortex did not reach the same extent as seen in the corpus callosum of animals treated with the same protocol [19]. The observed decrease in APC‐positive oligodendrocytes in all areas studied is known to be caused directly by cuprizone toxicity [15, 41], but may in part also be caused by a loss of APC expression in the surviving oligodendrocytes. An underlying mechanism could be ER stress leading to the accumulation of unfolded or misfolded proteins in the ER lumen [42]. ER stress and unfolded protein response (UPR) have been identified to play a critical role in several neurological diseases, including MS [43, 44, 45]. Interestingly, the same downregulation was observed for the oligodendroglial marker Nogo‐A in the corpus callosum of aged and young mice [19, 46, 47]. Similar patterns for both markers were also observed in the hippocampus and cerebral cortex of young mice [46, 48, 49].

Gray matter remyelination in aged mice occurred slowly and remained incomplete even after a remyelination period of 3.5 weeks in all gray matter areas analyzed, thus more precisely mimicking human pathology. Interestingly, the time course of de‐ and remyelination in gray matter closely resembled the situation in the white matter (corpus callosum) of aged mice [19] in contrast to the situation in young mice, in which white matter demyelination precedes gray matter demyelination [48]. In contrary to the homogenous course of impaired remyelination in the different gray matter areas, repopulation with mature oligodendrocytes displayed considerable regional differences following ubiquitous severe depletion after 6.5 weeks of cuprizone treatment. In the hippocampus, a continuous repopulation with mature oligodendrocytes was followed by a steadily progressing, yet incomplete remyelination after termination of cuprizone administration, with numbers of mature oligodendrocytes not reaching control levels. Strikingly, oligodendrocyte numbers in the cerebral cortex reached the cell count of control animals already 0.5 weeks after cuprizone cessation, though remyelination progressed slowly and remained incomplete during the observation period of 3.5 weeks. These findings suggest that impaired remyelination in the cerebral cortex may be mainly caused by either hampered final differentiation of mature oligodendrocytes or the myelination process itself, whereas in the hippocampus a reduced number of oligodendrocytes might additionally account for impaired remyelination. These results are in line with data from other experimental remyelination models using aged animals, which traced restricted remyelination back to disturbed recruitment and differentiation of oligodendroglial cells [50] as well as with the situation in chronic MS lesions [39, 40]. Based on animal studies, remyelination has been proposed to be a highly orchestrated process consisting of sequential migration, proliferation, and differentiation of oligodendrocyte progenitor cells into mature oligodendrocytes that could then successfully sheath axons [51, 52, 53, 54]. However, recent studies suggest that this is an oversimplified narrative and adult oligodendrocytes that have survived the demyelinating insult may also promote remyelination [55, 56]. This may explain the rapid recovery of oligodendrocytes after cuprizone removal. However, Hesse et al. have demonstrated that oligodendrocytes begin to perish within a few days after starting the cuprizone diet [47]. Moreover, numerous studies have shown that rapid recovery after cuprizone withdrawal is caused by the proliferation of oligodendrocyte progenitor cells, which proliferate intensively between the third and fifth week and differentiate into new remyelinating oligodendrocytes [48, 57, 58]. Therefore, we believe that OPC are the main source of remyelination, while surviving oligodendrocytes and neuronal progenitor cells might also contribute to this process.

A transient atrophy of the cerebral cortex and the stratum pyramidale was evident around maximum demyelination in aged mice. However, no significant changes of the amount of NeuN‐positive neurons were observable during de‐ and remyelination in the different gray matter areas when the number of NeuN‐positive cells was determined per mm². In the stratum pyramidale a reduction of the size of NeuN‐positive neuronal nuclei in combination with a higher neuronal cell density measured by Delauny triangulation was observable. These effects might therefore explain at least in part the volume loss of the stratum pyramidale. Conversely, the cerebral cortex displayed no changes in the density or size of NeuN‐positive neuronal nuclei despite transient volume loss. This discrepancy could be explained by the small, albeit significant, transient reduction in cortical volume, which may not immediately translate into detectable higher neuron density. Alternatively, simultaneous volume loss and apparently unchanged neuronal cell density suggest possible neuronal loss or loss of NeuN positivity during maximal demyelination. In conclusion, a slight loss of NeuN positivity or loss of NeuN‐positive neurons especially in the cerebral cortex cannot be excluded, but the transient volume reduction could be well explained by the observed changes in synapses and dendrites. This is in line with previous findings in the cuprizone model of chronic demyelination in young mice, which found that brain volume loss is explained mostly by axonal damage, rather than by changes in the amount or the size of neurons themselves [59]. Furthermore, in the cortex, myelin volume decrease could also contribute to the overall volume reduction. Similarly to gray matter damage in MS [6, 60], a decrease in synaptic density as measured by reduced synaptophysin intensity was evident around maximum demyelination at 6.5 weeks and showed gradual recovery during remyelination. Dendritic damage, visualized by decreased MAP2‐staining, was already detectable during ongoing demyelination and also displayed partial regeneration during remyelination. Recent work has shown that the nonselective chopper chelator activity of cuprizone leads to reversible mitochondrial impairment in neurons [61, 62, 63, 64]. Thus, the neuronal changes observed here may not only be indirect effects of the processes of demyelination and remyelination, but also a direct consequence of both mitochondrial dysfunction and metabolic deficit by cuprizone in the neurons themselves. Neurons, including their synapses and dendrites, could undergo immediate regeneration after cuprizone withdrawal independently of oligodendrocyte‐driven interactions.

Overall, our data suggest that cuprizone‐induced acute demyelination in aged mice affects neuronal processes and results in synaptic and dendritic damage but does not lead to substantial neuronal loss and limited atrophy of gray matter regions, which is in line with findings from young mice [59]. These results partially reflect the situation in MS lesions in different gray matter structures in which the decrease of synaptic density is the most pronounced change [6, 8, 60]. However, gray matter regions in MS patients also display neuronal loss and significant gray matter atrophy [6, 7, 10], which might be attributable to longer lasting demyelination compared to the short‐lasting demyelination period in our model. One might speculate that during longer lasting cuprizone‐induced demyelination neuronal damage and loss might become apparent, as has also been observed in young mice [59].

The different gray matter regions displayed a pronounced microgliosis that persisted after termination of cuprizone treatment during the remyelination period. This contrasts with findings in young mice, in which microglia accumulation and activation decreases quickly after cessation of cuprizone treatment [15]. Overall, despite their potential detrimental inflammatory effects, microglia are increasingly recognized to possess pro‐remyelination properties by clearing myelin debris after demyelination as prerequisite for successful remyelination as well as by producing growth factors for OPC proliferation and differentiation [36, 65, 66]. Prolonged microgliosis after demyelination in aged animals might be caused by impaired phagocytotic capabilities of aged microglia, thus hindering efficient remyelination [66, 67, 68, 69]. Future studies are needed to further characterize the microglia phenotype in aged mice during remyelination. The number of astrocytes, which was increased during cuprizone‐induced demyelination quickly declined during remyelination. This contrasts with the astrocytic reaction in the corpus callosum of aged mice, where the number of astrocytes remained unchanged and significantly elevated during remyelination [19]. The different astrocyte reaction in white and gray matter during remyelination might be attributable to the regional heterogeneity of cerebral astrocytes [36, 70, 71]. In young mice, current studies report that astroglia accumulation and activation in the cerebral cortex remain prominent for at least 2 weeks after cessation of cuprizone treatment [37]. In general, the role of astrocytes for remyelination remains controversial since crosstalk with other cell populations and secretion of various pro‐ or anti‐remyelination factors can have both supportive as well as detrimental effects on remyelination [36, 72, 73]. Accordingly, a clear statement on the role of astrocytes regarding remyelination in aged mice cannot be made based on the divergent astrocytic reaction in white and gray matter despite comparable remyelination impairment in both regions. Taken together, the reaction patterns of microglia and astrocytes after cuprizone‐induced demyelination in the gray matter of aged mice might contribute to remyelination failure by creating a remyelination‐hampering environment as it is discussed for non‐remyelinating human MS lesions [74, 75].

In summary, the remyelination process, glial cell reactions and neuronal alterations of the presented different gray matter regions in aged mice after cuprizone‐induced demyelination exhibit many parallels to findings of gray matter pathology in MS [3]. This modified cuprizone mouse model therefore offers a suitable model to study impaired gray matter remyelination and synaptic and dendritic damage with the potential to investigate remyelination‐enhancing therapies. The bioinformatical protocols introduced in this paper further present an opportunity to reproducibly analyze larger numbers of data, to better characterize various effects on cell composition and morphology. As these results fit well with previous observations in white matter areas of aged mice, in which a long‐lasting effect of cuprizone treatment has been described [19], we extend these previous findings and conclude that this protocol is also feasible for the study of gray matter areas, namely the cortex and hippocampus, which are of high clinical relevance.

In the field of digital pathology, QuPath has become increasingly influential caused by its capability to be used in cell classification as well as in other image analysis tasks [76]. Here, we were not only able to confirm its suitability for the measurement of cellular counts, cell morphology and respective staining patterns, but also established a novel workflow for the quantification of myelination status in different gray matter areas of the mouse brain. We show that the use of open‐source software like QuPath and StarDist is not only a reliable, but also a versatile tool for image analysis, caused by the possibility of creating and modifying workflows by script. In the context of myelination research, the software holds the potential to standardize myelination quantification and to further elaborate differences in myelin distribution and structure between de‐ and remyelination, for which evidence was found in the present study as well as in previous research [73].

The issue of reproducibility of scientific findings represents a constant challenge [77]. In this context, image analysis has proven to be particularly prone to misuse, both unknowingly and intentional [78]. Therefore, numerous guidelines have been introduced with the goal to unify the use of data and its reporting [25, 26], but with whole‐image analysis requiring large amounts of data and neuronal networks relying on a vast number of different parameters, transparency is difficult to maintain. In addition to the preexisting guidelines, this paper therefore aims to establish a basic workflow for researchers in image analysis to carry out measurements on randomized datasets and with reliable methods at hand (Appendix A1). The cell detection and classification models trained in the course of this work are available publicly and aim to encourage scientists to apply reproducible image analysis methodology to further enhance myelination research.

AUTHOR CONTRIBUTIONS

Stefan Gingele, Thiemo M. Möllenkamp, Martin Stangel, and Viktoria Gudi conceptualized and planned the experiments. Stefan Gingele, Thiemo M. Möllenkamp, Florian Henkel, and Viktoria Gudi performed the experiments. Thiemo M. Möllenkamp, Stefan Gingele, and Viktoria Gudi established the analysis protocols. Stefan Gingele, Thiemo M. Möllenkamp, Florian Henkel, Lara‐Jasmin Schröder, Martin W. Hümmert, Thomas Skripuletz, Martin Stangel, and Viktoria Gudi analyzed the data. Stefan Gingele, Thiemo M. Möllenkamp, and Viktoria Gudi drafted and wrote the manuscript. All authors read and approved the final version of the manuscript.

FUNDING INFORMATION

This research received no external funding.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Appendix A1. Example workflow for the setup of a pixel‐based analysis. The images are first split into training and data set. The training set is used for the development of a classifier, which may then be independently applied to any data set. Using a feedback loop, results are improved repeatedly before validation. Different settings may also be applied during this phase. It is important that different images, or at least different parts of the image, are used for training and validation, since otherwise the concordance cannot be assessed objectively. The data set itself becomes masked upon import, to avoid any bias during annotation. After this, batch image preprocessing may be considered, but must not be changed after cell segmentation.

Appendix A2. Example workflow for the setup of an object‐based analysis. The images are also first split into training and data set. The training set may then be used to build a cell detection neural network (NN) in StarDist, if required. The resulting algorithm may afterwards be applied to the training as well as to any other data set. For the training set, a feedback loop is used to classify positive and negative cells manually, or based on a preexisting classifier. Different settings may also be applied during this phase. After validation, the resulting classifier is ready for application on any data set. The data set itself should be randomized and masked before image processing or annotation, with demasking only during statistical analysis.

Appendix B1. Example image with representative annotations. Inside the hippocampus, CA3 (navy blue), CA1 (teal blue) and hilus of dentate gyrus (cyan blue) are each outlined for separate analysis. The cortex is split in two regions, one comprising the first four cortical layers (dark pink), the other containing cortical layer V and VI (purple). All annotations are drawn in accordance with the Paxinos and Franklin mouse atlas.

Appendix B2. Example image with representative annotations. The hippocampal stratum pyramidale and the granular layer of the dentate gyrus are encircled in cyan blue, as annotated by a pixel‐based classification algorithm. The cortex is split in two regions, one comprising the cortical layers I‐IV (dark pink), the other containing cortical layer V and VI (purple). All annotations are drawn in accordance with the Paxinos and Franklin mouse atlas.

Gingele S, Möllenkamp TM, Henkel F, Schröder L‐J, Hümmert MW, Skripuletz T, et al. Automated analysis of gray matter damage in aged mice reveals impaired remyelination in the cuprizone model. Brain Pathology. 2024;34(2):e13218. 10.1111/bpa.13218

DATA AVAILABILITY STATEMENT

The algorithms used for cell detection as well as cell and pixel classification, along with example images for comprehension of the workflow are publicly available at https://github.com/ThiemoMMoellenkamp/Gray-Matter-Damage-in-Aged-Mice.git. Further data presented in this study are gladly available on request from the corresponding author. For cell segmentation, the training dataset has been made available under “Möllenkamp, Thiemo, 2023, ‘Replication Data for: Gray Matter Damage – Training dataset for cell segmentation of DAB mouse brain images’, https://doi.org/10.7910/DVN/JGEQPS, Harvard Dataverse, V1”. This set was used in conjunction with the publicly available TNBC dataset (https://doi.org/10.5281/zenodo.1175282) and the MoNuSeq Data Segmentation Challenge: This data can be found under creative commons license CC BY‐NC‐SA 4.0.