Colocalization analysis of small vesicles

Hello everybody,

I´m kind of stuck at my image analysis. I want to quantify the colocalization between two signals in different mutant Backgrounds to see if there is any change. Problems are the pretty small size of the vesicles and the noise because of the weak signal from one of the proteins. I can´t enhance the expression to get a better signal because I have to use the natural promotor. I tried with JaCoP but I get values around 10 % colocalization in wildtype when from my optical impression it has to be around 80-90 %. I also tried to reduce the noise before the calculation which brings me to 20 % colocalization. Has anybody an idea how to solve this? Some pictures are attached
Thanks a lot!
UKS273_F_t001c1|536x169 UKS273_F_t001c1+2 UKS273_F_t001c2

Hi schneika,

I assume that you are working with confocal images.
Are they 8-bit or 16 bit? Stacks?
In my experience, working with weak signals in 16 bit images (or 12-bit) makes it much easier already.
Also, averaging a stack might improve signal to noise, but be careful not to merge too many layers since you might influence the “colocalization”.
Also be aware that you are dealing with 2 different background levels, 1 form the “no tissue” and 1 from the “tube” background were the vesicles are in. Depending on the method you are using, this might also influence the threshold for vesicle detection.

I quickly tried the following:

run("Median...", "radius=1");
run("Subtract Background...", "rolling=5 sliding");
run("Auto Local Threshold", "method=Bernsen radius=15 parameter_1=0 parameter_2=0 white");

Best regards,
Mario

Hi Scheika,
Are you trying to do pixel-based colocalisation analyses or use object-based colocalisation? From your question, it sounds like you are trying to do pixel-based colocalisation analysis. As Mario points out, setting the thresholds correctly will have a significant effect on the analysis results.
Looking at the images, I would probably try an object-based colocalisation, i.e. first segment both images into puncta of interest and then analyse overlap between puncta.
As for the noise level, have you perhaps considered using N2V (https://imagej.net/N2V) or something similar to clear up/reduce the noise? It might help with puncta detection.
Cheers,
Volko

Hi Mario,
Thanks for your input! I have 16 bit movies with 1 min duration taken with a confocal microscope (LSM900 with Airyscan2 from Zeiss if thats important). I used the JaCOP Plugin with M1&M2 coefficient. The Plugin used the whole movie for calculation. Do you recommend to use a Picture from a single timepoint instead or export the movie to single Pictures and create a stack? I tried to reduce the background with gaussian blur but that was not really helpful…
I will try your workflow, thanks a lot!

Hi Volko,
I used pixel-based analysis because I assumed the object-borders are not sharp enough and the size might be too small, too for object-based analysis. I tried to count similar vesicles for a different analysis with object-based analysis and wasn´t able to set a threshold so that imagej could clearly distinguish the individual vesicles.

Ah, you have movies. Interesting.
If all your vesicles are moving, you could try to calculate a mean average image from the movie and subtract this from each frame to get rid of the background.
Also, gaussian blur will not keep object borders. So I prefer using the median filter if borders are important.
I guess there is a reason why you acquired movies instead of images?

I recorded movies because all of my vesicles are moving and I wanted to create kymographs, too

You can try ComDet (object, distance-based colocalization), it should be able to handle this signal-to-noise level

1 Like

Thanks for your Suggestion! Nice tool, but in my case it included only about 50% of the vesicles in the analysis so I guess thats not working for me. What Mario suggested works good so I think thats the solution for me. Thank you all for your time :slight_smile:

I get 47.1% Colocation with 3D MultiColoc