Cell morphology, actin organization, colocalization

Here I am again with a new project: I am looking at two cell types (A2 and M1) with a variety of treatments, each within the confluent monolayer and at the margin of a scratch wound (the treatments affect, among others, motility). I have stained nuclei (blue, channel _C001), actin (green, channel _C002) and several proteins, each in red (as in the uploaded examples, channel _C003), and I also have the corresponding DIC (channel _C004), all in confocal.

As you can see, the morphology of the cells are quite different, and even more so when they start migrating into the wound.
I also observed there is some kind of correlation (sometimes inverse) between:
a. The amount (I would say integrated intensity) of actin and that of some of the proteins;
b. Actin organization (more stress fibers – i.e., fibrillar – vs. more diffuse – i.e., granular, and the granules of varying size) and treatment;
c. Actin and some of the proteins cytosolic (at in one case nuclear) localization.

I am thus faced with the following problems:

  1. Detect correct cell boundaries; I would use both the DIC and the actin staining, but I don’t know how to use DIC images and how to combine them with the fluorescent ones in order to get as close to reality as possible secondary objects (cells). Also, would MeasureObjectSizeShape be able to tell me something about the highly polymorph A2 cells migrating into the wound? (that is, not just width/length or similar; this morphology changes with treatment).
  2. How to get measurements of actin organization? (fibrilar vs. granular?) I don’t know which of MeasureGranularity or MeasureTexture is more appropriate in this case, since I also need to know the amount of fibrilar actin (stress fibers).
  3. I guess that if I get integrated intensity measurements (per cell) for actin and my proteins, MeasureCorrelation would tell me if there is a direct/inverse/no correlation between these measurement in each cell; correct?
  4. However, I also need to see if I have, say, more of my protein when actin is more fibrilar, and if there is a co-localization or on the contrary, the proteins are to be found more in regions with less actin. I’m at a loss here…

All these are quite subtle, and that’s why I need numbers to support (or not) my hypotheses. Can you please help with suggestions on how to built a pipeline(s) for any or all of these?
Thank you!

PS. For some reason I could not attach the files, so here is a link to Dropbox:
dropbox.com/sh/7j1yxmr225w3 … nrIna?dl=0

Changing the method in IdentifySecondary to “Watershed-Gradient” using the green channel seems to get a result resembling the actual cell edges. But I found that enhancing the DIC image with EnhanceOrSupress and then adding the result to the green channel seems to do even better. A tentative version of the pipeline is attached.

MeasureObjectSizeShape would certainly provide a good suite of measurements to choose from. But which exact measurement would be most appropriate in this case is an assay-dependent issue, and hence, you would be most qualified to make that choice. It might be that one option is to make a variety of per-cell measurements, and then use the phenotype classification tool in CellProfiler Analyst to distinguish between fibrillar and granular.

This is hard for me to say, since the fibrillar organization is not terribly apparent to my eye in the images you submitted. My first guess would be that MeasureTexture would be more appropriate since MeasureGranularity is notoriously hard to optimize. But it’s not clear which of the 14 Haralick features would capture that difference best.

The MeasureCorrelation module measures the pixel-wise correlation between two images, so it’s not quite the same thing. But you could use CalculateMath to compute the ratio of the integrated intensities of the two channels, and see if there’s a trend one way or the other.

Here, MeasureCorrelation might be more useful since it doesn’t rely on detecting the features as such beforehand. If the bright proteins co-localize with the bright fibrils, you would expect a higher correlation than if the bright proteins were present in the dimmer regions.

assay.cppipe (14.4 KB)

Hi Mark,

My activity came to an abrupt stop last January, when I moved to a different university. I finally got back to try and finish this project, but now I cannot download the pipeline you added then - would it be possible to re-load it please?
Thanks you!

Can you please repost the pipeline? I’m working on a similar problem. Thank you!


Hello CP team,

I seem to have a problem downloading pipelines - I tried several (including this one) and on different computers, both at work and at home: I click on the pipeline, a new blank window opens and then…nothing. I had this problem for the last about two month, since I resumed this activity. Please help me since I just cannot figure out what’s going on.

Thank you!