Background

Since the invention of the microscope, quantification of specific cell types in tissue sections has played a crucial role in biological discovery. Advances in cell labeling, including immunohistochemistry, in situ hybridization, and the use of genetically encoded fluorescent labels, have enabled cells to be identified and distinguished with increasing specificity. However, the quantification of labeled cells has largely remained a tedious manual or semi-automated process. The time-consuming nature of this process means it has only been practical to quantify small sample areas of tissue, leaving the majority of tissue uncounted. Even when an entire brain can be imaged, computer-based semi-automated image analysis tools available in programs such as ImageJ are not able to handle the large files produced; therefore, analysis has largely remained limited to small sub-regions. In a complex and heterogeneous organ, such as the brain, these restricted processes can lead to experimenter bias, due to the subjective nature of manual counting and sampling bias, if a selected subset of tissue is not an accurate representation of a whole region.

Recent advances in computational image analysis and whole slide scanning technology have made it possible to quantify every single cell in a whole tissue section on a regular desktop computer. It is now practicable to quantify and further analyze the characteristics of hundreds of thousands of cells in a few minutes—a process that would take days or weeks to do manually. However, this process is not perfect. Software programs that examine digitized images to determine what is, and is not, a cell, are only as good as their underlying algorithms (which are hard-coded by the engineers of the software) and the validity of any user-determined input parameters. It remains important to carefully assess the validity of automated detection algorithms manually, as small differences in tissue processing, staining, and imaging can have significant impacts on the performance of automated algorithms when they are applied across whole projects. Despite this, many studies using automated cell counting approaches do not present evidence of optimization, or justification for the selection of specific parameters.

Here, we present a semi-automated method for counting and classifying two fluorescently labeled cell types—pericytes and microglia—in whole mouse brain tissue sections using the open source software QuPath (Bankhead et al., 2017) as used in Courtney et al. (2021). Our method uses extensive optimization to objectively determine the specific automated cell detection parameters required for each different brain region.

Although this method is written specifically for quantifying pericytes labeled with the fluorophore DsRed and microglia labeled with the fluorophore GFP in mouse brain tissue, it could easily be adapted to quantify any type of cell in any type of tissue, so long as the stain is localized within or immediately around the nucleus. In addition, we present the optimization of just one specific parameter important for automated cell detection, the fluorescence intensity threshold. However, we provide the framework for adding optimization steps for additional cell detection parameters, if desired.

Collectively, the method presented here offers a user-friendly and cost-effective framework for the automated quantification of cell numbers in whole tissue sections, without the need for extensive and time-consuming manual counting.