Most experiments used only Mt-GFP; some used both Mt-GFP and Cyto-tdTomato. The Z-projected images were preprocessed using a custom Fiji macro. For preprocessing of Mt-GFP images, background was subtracted (rolling ball, radius = 3), images were median-filtered (radius = 1), and somatic mitochondria were removed. For removal of the somatic mitochondria, they were first segmented and then a somatic mitochondrial mask was overlaid on the image. Steps for somatic mitochondria segmentation were as follows: taking the square root of the image (equalization of the too bright and too dim somatic mitochondria), minimum filtering (kernel = 20) to erode small structures, using Gaussian filter (σ = 20) to ensure mitochondria in neurites were downscaled, Otsu thresholding, object filtering (5000 to 100,000 pixels), and binary dilation (15 cycles) of segmented somatic mitochondria. The size or shape of the mitochondria in neurites did not change during this preprocessing. For preprocessing of Cyto-tdTomato images, we applied a tubeness filter (39) and then removed the soma using the somatic mitochondrial mask from the green channel. For further details, see also (40).

Segmentation of the preprocessed images followed using GE Developer software. Images were equalized to the complete 16-bit dynamic range and segmented using object-based segmentation with a kernel size of 3. Segmented objects were classified as axonal mitochondria (green channel; length, 0.5 < Mtaxon ≤ 1.4 μm; intensity, >5000; area, >0.25 μm2; circularity, >0.6), dendritic mitochondria (green channel; Mtdend length, ≥ 2.4 μm; intensity, >5000), and neurites (red channel; intensity, >5000; circularity, <0.5). Count, length, area, and circularity were measured for axonal and dendritic mitochondria; area was measured for neurites; and the measurements were aggregated using the median for mitochondrial area, circularity, and length and a sum of the area for mitochondria (CA) and neurites.

Aggregated field values were further processed by a custom Python script. Field values containing less than 500 mitochondria were removed, along with outliers, identified by a custom-written machine learning algorithm (fig. S6, left), and field values were averaged to well values. Robust Z-scores were calculated using neutral wells (fig. S6, right) for the primary screen and DMSO-treated wells for the rescreen. Outliers were removed from the replicate plate well values, and the replicate plates were averaged. Replicate plate wells containing outliers were identified by selecting those where the SDs of the replicates’ robust Z-scores were larger than 3. In those cases, one value, which was furthest from the mean, was removed. In the rescreen, where 10 replicate plates were used instead of 4, this algorithm was run three times, allowing the removal of maximum three values from the 10 replicates. Finally, hits were selected on the basis of their robust Z-scores: for axonal mitochondrial content hits, axonal mitochondrial CA > 2.5 and dendritic mitochondrial CA between −2 and 2; for dendritic mitochondrial content hits, CA > 2.5 and axonal mitochondrial CA < 2; for elongation hits, dendritic mitochondrial length median > 2.5; and for health hits, axonal mitochondrial circularity median < −2.5.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.