(108g) Enhancing Insight to Individual and Population-Based Microglial Reactivity with Image Analysis

Organotypic whole hemisphere (OWH) brain slices serve as a high-throughput intermediate between cell cultures and in vivo studies that still maintain 3D complexity, regional representation, and the multicellularity of the brain microenvironment [1]. To model ischemic pathology, we developed an oxygen-glucose deprivation (OGD) OWH brain slices model [2] that has demonstrated impact on neurons and microglia. Microglia are the resident immune cells of the brain and can become activated and drive neuroinflammation and ongoing injury after an injurious event. Therefore, microglia have become of interest for therapeutic targets. However, microglia are dynamic after injury and, when analyzed on a population-level, exhibit heterogenous phenotypes that complicate therapeutic strategies. For example, the ischemic hippocampus may exhibit five different phenotypes of microglia, including a resting phenotype not typically associated with injury or injury-driving microglia. We sought to provide a quantitative imaging-based assessment to categorize individual microglial features based on microglia shape features, including perimeter, area, convex area, and circularity, among others, while also analyzing larger populations of microglia to quantify phenotype heterogeneity across treatments, regions, and sex. To quantify entire populations of microglia, we combine our feature analysis with a principal component analysis (PCA) on coordinates positioned along the outside of the segmented cells [3]. The addition of PCA allows us to classify microglia into a number of groups with distinct phenotype representations. Combining the representative phenotypes analyzed on a population level with individual cell features obtained via non-destructive image processing techniques allows us in situ insight into comprehensive microglial morphology in the presence of injury.

Methods:

We developed a high-throughput cell quantification pipeline called Frameworks for Fluorescent neuroImage Based Experimental Routines (FFIBER). FFIBER is a seven-step process: develop a data awareness, design a data management plan, determine an optimal experimental pipeline, build out supporting data science infrastructure, perform primary and supplemental imaging, produce interpretable visualizations of results, and build educational modules for developed protocols. For each brain slice dataset, we use FFIBER to develop a routine to quantify cell morphology. For the data science infrastructure, we use Sci-Kit Image in Python to apply an Otsu threshold to segment and label our cells. We then apply geometric algorithms from Sci-Kit image to obtain more than ten morphological features. We manage these features using essential database management in Python. We then split our images into testing, training, and validation sets to feed into the Visually Aided Morphometric Image Recognition (VAMPIRE) package [3]. This package applies PCA to all of our cells and produces a dendrogram of representative cell morphologies. Dendrogram testing data is then fed into the algorithm labeled for key experimental characteristics – sex, treatment, brain region, and species. We then use data visualization concepts to produce interpretable results by generating heat maps (1) of all features by treatment group, and (2) by representative morphology from VAMPIRE. The heat maps of all features are then connected to representative morphology in order to determine correlation of individual cell features such as perimeter, area, convex area, and circularity with representative phenotype distribution across treatment-based, sex-based, and regional microglial populations.

Results:

We applied FFIBER to experimental OWH OGD brain slice data obtained from the rat and the ferret. In the rat OGD brain slice study, we examined microglia as a population and individually. Our population analysis with VAMPIRE showed more heterogeneity in microglia phenotypes after OGD compared to the non-treated control. While, OGD followed by treatment with an anti-inflammatory drug, azithromycin (AZ, an anti-inflammatory drug), had similar heterogeneity of microglia phenotypes compared to the non-treated control. The heterogeneity results are interesting because we observed microglial behavior population-wise and in situ. Across all treatment groups, frequency ranking – a ranking of how often a specific phenotype occurs - of each microglial phenotype remains almost exactly the same. However, the frequency difference between each rank is much smaller in the injured group than in the non-treated control and AZ group.

Additionally, we applied geometric analysis to individual microglia to analyze circularity. We observed that the microglia that underwent OGD without treatment were more circular than the microglia in the non-treated control and in the OGD with anti-inflammatory treatment. Microglia with high circularity are correlated with an ameboid phenotype and are associated with driving pathogenic inflammation in the brain. These results supported cytotoxicity data in the rat studies, which signified that OGD results in cell death, including microglia, and can be recovered when treating with AZ. We identified population-wide shifts in microglia phenotypes in the presence of injury that are reversed following treatment with AZ and observed that some of this shift is due to individual microglia taking on a more circular and less ramified structure.

We also applied FFIBER to OGD brain slice data from the ferret. We began by analyzing the representative morphologies of whole microglia populations. We visualized population-wide sex differences in microglia phenotype distributions, including higher reactivity in microglial shape shifts for the male ferret when compared to the female. The larger shift in microglia population phenotypes in the male could be correlated to the clinical differences in sex-based response to injury and treatment in neonatal brain injury. We also delineated and visualized the microglia populations by brain region. When populations were analyzed by region, we observed differences in the reactivity of microglia across the hippocampus, cortex, and corpus callosum – this data aligns with the clinical data showing regional responses to injury in neonates with ischemic brain injury. Finally, we used image processing to quantify circularity, perimeter, area, and convex area of individual cells and were able to correlate the population-based representative phenotypes that have the greatest frequency shifts between non-treated, injured, and treated groups with ramified and bushy microglia phenotypes. Changes in microglia ramification in situ correlate with microglia activation in presence of injury and possible deactivation in presence of injury with treatment. All results were obtained using an image processing method that does not alter the microglia or their environment during image procurement or measurement, supporting FFIBER as an in-situ cell quantification method.

Conclusions:

By developing quantitative cell morphology pipelines for fluorescent images, we can quantify population-level microglia phenotype presentation and individual microglia features for comprehensive understanding of the dynamic behavior of microglia. With an individual and population-based perspective we can measure sex and regional patterns in cell phenotypes in situ without destroying or altering the cells, which is not possible by other methods. These image processing pipelines combined with the high-throughput nature of brain slice models allowed us to study the effects of OGD across both the rat and ferret species. In the rat, we detected population-wide phenotypic changes in microglia in response to injury that were resolved using AZ. Large shifts in population-based phenotypes were correlated with circularity increases after injury that were not observed in the azithromycin group. In the ferret, we connected representative morphologies found using PCA with features of active microglia such as cell ramification. We were also able to use FFIBER and the brain slice model platforms to perform experiments across different sexes, brain regions, and multiple time points of OGD and treatment. In the future, the development of FFIBER will grow our image processing capabilities to include more algorithms for microglia shape feature analysis including fractal analysis for branching and increase robustness of our software so that cells, including other types of brain and non-brain cells, can be analyzed by users with minimal coding experience.

References:

[1] Liao, R., T.R. Wood, and E. Nance, Superoxide dismutase reduces monosodium glutamate-induced injury in an organotypic whole hemisphere brain slice model of excitotoxicity. Journal of Biological Engineering, 2020. 14(1).

[2] Joseph, A., et al., Nanoparticle‐microglial interaction in the ischemic brain is modulated by injury duration and treatment. Bioengineering & Translational Medicine, 2020. 5(3).

[3] Phillip, J.M., et al., A robust unsupervised machine-learning method to quantify the morphological heterogeneity of cells and nuclei. Nature Protocols, 2021.