Image analysis

Implemented as an ImageJ2 plugin in Java [59] and provided via a Fiji update site. The image analysis is separated into three core tasks: segmentation of nuclei, segmentation of cell area, and detection of organelles. The settings of the image analysis need to be fine-tuned for each individual imaging experiment, thus different settings for each individual image analysis component may apply. To ease the usage of OrgaMapper and to promote reproducibility the settings are saved by OrgaMapper into a XML file for each analysis run. This settings file can easily be opened again in OrgaMapper to apply the same settings with minimal effort. For the datasets in this publication the precise settings are provided for download (https://doi.org/10.5281/zenodo.10932803).

The nuclei segmentation is performed on the DAPI channel. First, a median filter was applied to level out in-homogeneities in the nuclei signal without smoothing of the nuclei edge. For background subtraction, a rolling ball background subtraction was applied. To segment the nuclei, an automatic global intensity threshold was applied. Optionally, the segmentation mask can be adjusted using binary erosions. The particle analyzer is used to reject nuclei at the edge of the field of view as well as apply an optional size and circularity filter (Additional file 3: Fig. S6A).

For cell segmentation, the CMFDA channel was filtered using a median filter. A rolling ball background subtraction was applied to the filtered image. A fixed global intensity threshold was applied to generate binary masks of the cell area. To separate touching cells a marker-controlled watershed was used. First, the signal of the nuclei channel and the CMFDA channel was added together. The composite image was then filtered using a large Gaussian blur. To determine the separation of the touching cells the find maxima algorithm was applied using the segmented particles option. This applies a watershed algorithm based on the intensity values of the combined and smoothed Nuclei and CMFDA channel and results in a binary mask containing the boundaries of touching cells. The cell area mask and the cell boundary mask were multiplied to generate a binary mask with individual cells separated. The cells were further filtered for size and circularity using the particle analyzer option in ImageJ. Further, the cells were filtered if they did not contain a nuclei segmentation or more than one nuclei segmentation (Additional file 3: Fig. S6B).

To detect individual blob-shaped organelles the organelle channel was filtered using an ImageJ implementation of the Laplacian-of-Gaussian filter [49]. Individual organelles were then detected using a maxima detection. The detections were filtered for excluding detections in the nuclei mask (Additional file 3: Fig. S6C).

We extracted key measurements per well such as total cell count and mean intensity of the background based on the area outside of the cell segmentation. For each cell, we further extracted parameters such as cell area, ferret diameter, and mean intensity of the organelle channel as well as an optional measurement channel within the cytoplasmic area (cytoplasmic area: cell mask minus nuclei mask). To determine the distance from the nucleus of each organelle detection, an EDM was computed per cell, which is a very fast computation as compared to the algorithm used by [35]. For each individual detection within each cell, the distance based on the EDM was extracted (Additional file 3: Fig. S6D). Further, the signal intensity at that location of the detection was measured in the organelle channel as well as an optional measurement channel. As an alternative detection-independent measurement, the distance of each individual pixel within the cytoplasmic mask of each cell was extracted as well as the corresponding intensity value in the organelle channel and measurement channel.