Measuring Intensity of Junctional Proteins

Hi All!

I am working with images that show groups that differ in the intensity of junctional proteins. The set of images I have with high junctional intensity I can get secondary objects from by inverting the image with ImageMath, but this doesn’t work with the other set of images as there are no clear dividing lines between cells. To try and quantify this, I created a pipeline (attached below) that just expands the nuclei by a few pixels to create secondary objects for both sets of images and compares the intensity in the junctional image channel within those borders. This pipeline shows minor differences between the groups, but not nearly to the degree I would expect when looking at the images by eye. I thought perhaps this could be due to the region that is measured still being mostly low intensity (as only the very border of the cells is stained) which would bring the mean intensity down in both groups and limit the difference between the groups in my results, but I am unsure how to combat this. I tried using an upper quartile intensity measurement instead of a mean intensity measurement, but it didn’t seem to help much. I included screenshots of one image in each group so you can get a better idea of what I’m talking about.
image image
Thanks for any help, I really appreciate it!
TestPipeline.cpproj (495.2 KB)

Hi @NickSeyler,

Could you provide some example image sets?

Right now I have most of the images as .czi files which I can’t attach here, I have 1 from each set as jpgs for now I have attached them below, the pipeline is also set to accept images with 3 channels so that may take some tweaking, haven’t tried to run with jpgs so not sure. Thank you, and let me know if this isn’t helpful or if there’s anything else I should provide!

Well I think your nuclei stained image it’s probably needed as well so the primary objects can be first obtained in the pipeline.

People sometimes link their images hosted outwith the forum (GoogleDrive, etc.). That might be an option for you?

Good Idea! Here are 6 images, two in the high junctional group and four in the low junctional group.
https://drive.google.com/open?id=1qPr_hK96OQL2XxYbbeawJTnRqSs5qdUL

Just wanted to bump this thread in case anyone might come across it, I think the link to the trial images should work but if not let me know! Thanks!

HI Nick,

I’m sorry for not answering until now.

The main thing I would say is that I added an “OverlayOutlines” to look at what’s actually being measured as your cell. I’ve put two images below to show your original images with the outlines on top and I think it’s clear that they can quite often miss a lot of the highly stained areas.

Example Example2

I would attribute this to the fact that you are using the Distance - N method in IdentifySecondaryObjects. This means that the input image is not used at all and that your Primary Objects are just extended by 7 pixels. I suggest you look look at the other methods within IdentifySecondaryObjects to work on this and maybe add your own OverlayOutlines and SaveImages module to really look at what’s being measured per image.

Hi Nick, Try staining your samples with Phalloidin and then use the “Propagate” algorithm to find the cell cytoplasm and the boundaries between cells.

Thanks for the help! I will try that.

Thanks! I will look into this as well.

Good luck! What kind of cells are these? Beta-catenin staining might also be helpful in localizing the cell-cell junctions. You’ll need to stain with dapi and use that as a landmark to initiate the propagation segmentation. My screenshot of CP isn’t a full pipeline.