Background

Quantifying organelle abundance by determining how many organelles are present in the cell and the area of the cytoplasm they cover provides a useful measure of organelle dynamics and offers insight into organelle function. For example, stress granules are membrane-less organelles that form from stalled translation pre-initiation complexes and other proteins in response to stress [1–3], with their functional roles emerging upon their assembly in the cytoplasm. Similarly, peroxisomes are membrane-enclosed organelles that are essential for lipid and oxidative metabolism [4–6]. They undergo division and increase in number when their metabolic activity is required. In both cases, quantifying organelle density in the cytoplasm correlates with organelle function in healthy cells.

Organelle abundance can be estimated from confocal imaging by visualizing resident proteins or established organelle markers [7]. An organelle marker is any fluorescently labeled molecule that localizes to an organelle. Examples include overexpressing a fluorescent protein fused to an organelle localization signal or an endogenously tagged protein. Examples of endogenously tagged proteins include the CRISPR/Cas9 genomic fusion of polyA-binding protein PABPC1 to Dendra2, which is used as a marker of stress granules [8], and ATP-binding cassette sub-family D member 3 (ABCD3) fused to GFP as a marker of peroxisomes [9]. Other examples include using a fluorescent probe that selectively targets the organelle, such as PeroxiSPY, which is used to label peroxisomes [10], or using antibody staining of organelles post-fixation [9]. The quantification requires segmentation, or defining the boundaries of compartments, which is often based on visual inspection. Manual segmentation is prone to error; for example, it is difficult to consistently outline highly variable, punctate, or overlapping patterns. To improve the reliability of quantification, thresholding methods set an intensity value independent of the user, improving consistency [11]. Machine learning approaches further reduce manual segmentation errors by providing consistent labeling based on user-defined parameters [12–14].

Here, we present a protocol for quantifying the abundance of membrane and membrane-less organelles based on a single marker using Fiji software [15,16]. This protocol combines the WEKA Segmentation [12] and watershed algorithms into a user-friendly workflow. The quantification is based on trainable organelle recognition, which limits user-defined thresholding bias and improves the consistency of image processing [14,17,18]. Recognition requires reliably outlining organelles for initial training, a task that can be done by experienced researchers. The trained classifier can then be applied to any number of images obtained with the same microscopy settings. Alternatively, a beginning researcher can train the classifier and include controls to verify whether the recognized area corresponds to an organelle area.

As a proof-of-concept, we validate that machine learning–assisted recognition of cellular compartments corresponds to an independent marker of cellular cytoplasmic and peroxisome areas using a control marker for both the organelle and the cytoplasm. In many experimental setups, adding additional markers is not feasible, and manually extracting the cytoplasmic area from a single organelle marker is biased and time-consuming. Our results demonstrate that machine learning assistance enables the consistent and reliable identification of these regions. Finally, we apply the macro to quantify the non-membrane compartment density of stress granules using a CRISPR/Cas9-tagged PABPC1 cell line [19]. The algorithm recognizes conditions in which there are no stress granule compartments and extracts areas based on a single marker.

This protocol can be modified further to estimate organelle coverage and the area of the cell occupied by the cytoplasm, or to extract abundance in 3D using stacks of 2D images obtained from tissues. This provides a useful tool for quantitative cell biology experiments.