Problem with distinguishing very small nuclei from cell debris in genome-wide RNAi screen

I am very new to using Cellprofiler.

We perform big genome-wide RNAi screen in Drosophila S2R+ cells in 384-wells plates. Cells were stained with DAPI (channel2 in my images names) and anti-tubulin-FITC antibodies (GFP channel, channel1 in my images names).

We want to analyse the images taken with 20x magnification.
The thing is that some of our cells at some experimental are very big cells, and some at other conditions are very small and crowded. But for this I was able to create a pipeline which allows me to find both, small and large cells, so it was not such a big problem.

The problem which I have is to distinguish the very small nuclei from the dirt and cell debris, which I have in wells in which cells totally died out (example: positive control wells, images channel1_2 and channel2_2). I tried to rule out them by playing with size limits and threshold, but I was unsuccessful so far, when it was getting better with excluding debris, I started also exclude the proper (but very small) nuclei from other images. Do you have any ideas what should I do?



We’re glad that you can manage to tune the program to catch both big & small nuclei in a mixture.
As for the second issue, I don’t think we can digitally handle the problem, it’s even difficult by eyes.

I rather suggest to improve wet lab manipulation: first filter your DAPI vial to remove precipitation of the staining dyes. Then stain your nuclei with a lower concentration of DAPI. Then wash the samples thoroughly to remove debris. When capturing picture, use a low excitation intensity, I think you can still shine up your DAPI signal, while keep the debris faint as background.

Good luck.

Hello Minh,
thank you very much for your reply… the problem is that the whole experimental part for our RNAi screen was already performed (240 384-wells plates) and we can not repeat the experiment… It consumed large amount of time and money.
So I have to find some solution at digital level…

When I use my threshold settings (pipeline attached) I got nice primary objects in wells with both small and large cells, but when it comes to the wells in which almost 100% died I get some weird objects, which actually do not look like real nuclei. I attach images. So maybe there is some way to filter out those fake objects? I was thinking for example about using object intensity distribution measurements - but I am quite new to the image analysis, so I am not sure it it does make any sense? What do you think?

pipeline RNAi screen.cppipe (7.4 KB)

Do you have any other ideas, how to deal with my problem? Anyone?


I see.

While waiting for other opinion, I suggest let’s first adjust the noticeable illumination difference of the last picture, it has a very bright and uneven background.
You may use CorrectionIllumination calculate and apply module for this purpose.
After illumination correct, in your IdentifyPrimaryObjects, try adjust “Lower bound of threshold” to cut off more background.
And also I think the range 5-18 of diameter for nuclei is a bit small, have you tried 10-40 instead?

thank you for your suggestion. Yes, I also thought that illumination correction should be first step. But I have to wait for the rest of images to do this within a whole plate for each plate separately, I think it would the best way to get Illumination correction function for each plate. Ok, lets see if it is going to help.
Thank you again!