Looking for a way of automating the measurement of mitochondrial degradation by autophagy

Hi there,

I am trying to make a cellprofiler pipeline that allows me to analyze the degree of mitochondrial clearance by autophagy in a high throughput format.

As a readout the cells express a double-tagged (mCherry-EGFP) mitochondrial localization signal (seen as yellow). Once the mitochondria reaches the lysosome - and due to the acidic environment- EGFP gets quenched (and degrading structures can be seen as mCherry only).
I am trying to measure these mCherry only structures in an automated way that avoids having to manually feed it any parameters in the process (as I am doing now).

In the attached image you can find two example images (right and left panels show the same images, I just changed red to magenta in case someone is colorblind). The upper row shows active degradation whereas the bottom row shows no degradation.

I have some working cellprofiler pipeline already, but I am not completely happy with the result and I was wondering if you have any suggestion that could help me improve the pipeline.

What I currently do is:

  • to make a mask image that combines all structures in mCherry and EGFP channels.
  • to measure the intensities of mCherry and EGFP within the segmented structures.
  • to calculate the EGFP/mCherry ratio (I avoid mCherry/EGFP as it would give NaN in case EGFP is quenched) within the structures
  • THIS IS THE STEP I WOULD LIKE TO AVOID: by using the DisplayDataOnImage module on cellprofiler I check what ratio value corresponds to mCherry-only or yellow structures and I use this value to filter the objects, binning them between mCherry-only or yellow.
  • I can then count and measure objects in each category.

The fact that in both situations there is a vast extension of mitochondrial network that is not being degraded makes the analysis difficult… It would be great if you could suggest a way of measuring the amount of mCherry-only (either red or magenta in the image) structures.

Thank you very much in advance for your time and help!

Cheers,

Laura

I would like to thank @Christian_Tischer for his help. He has also been the one suggesting me to post this question here and see if you guys could give us some nice suggestion to approach the problem.

Thanks again!

Hi Laura,

First of all, it sounds like a good pipeline so far- well done!

The simplest solution would be to just measure all of your objects, then bin them after the fact, which could be done in semi-fancy ways (such as classification in CellProfiler-Analyst) or super simple ways (making a hist plot of the ratio measurement in Excel or your favorite graphing software, seeing if you distribution is indeed bimodal, and then sorting based on that measurement and splitting the data into halves at the point you determine from your histogram. I would personally do it in CPA, but that’s just me.

If for whatever reason it’s critical the split happens in-pipeline, I would

  • Use the Threshold module to threshold the EGFP channel
  • Use the MaskObjects module with your AllObjects and the EGFPThresholded image to call YellowObjects based on the “Keep depending on overlap” option set to something you determine makes sense (ie, that 75% of the object must be green-positive to be called yellow)
  • Use the MaskObjects module with AllObjects and YellowObjects and the “Remove” option to filter the YellowObjects from the AllObjects to keep RedOnlyObjects
1 Like

Hi Beth,

Thanks a lot for your quick reply!
I will give a try to what you suggest for keeping the split happening in-pipeline… (my preferred option).
Otherwise I could try CPA again. I did try it in the past to try to get rules that would allow a proper filtering back in CP and the result I got wasn’t great… (but that might have been me and my lack of experience when I first tried it!).

Thanks again.
Cheers,

Laura

Hi Larual

Sounds like your pipeline is mostly what you want but I can suggest a few changes.
My first change would be with any cellular morphology question, I would want to measure them per cell, just incase any of my treatments reduces/changes mitochondrial morphology/density. A simple way would be to identify the nuclei count, or propagate cytoplasm from the nuclei to get objects in each cell.
I would not mask my mCherry and EGFP structures, but I would identify every object that is mCherry positive. Then I would measure the intensity of that object with both mCherry and EGFP. Then I would use the object filtering to split these objects to only have mCherry single or both. Now that you have two object types, then I would do all my measurments.

Happy to chat more
Best lee

Hi Lee,

Thanks a lot for your answer!
I am sorry I didn’t explain my pipeline better. I do the measurements cell based, using as you say the nuclei staining to propagate cytoplasm.
Before I was also doing as you suggest: identifying every object and filtering them, but the segmentation I could get was not great, plus my main problem is to find a way to automatically do the filtering that allows separate red only or both.

I might have found a way to do so with suggestions from Beth and Tischi. I am currently trying if this approach work, if it does I can report it here.

Thanks a lot again for your time and help.

Best,

Laura