Measure the intensity and rate of GFP’s expression


#1

Hi,
I’m an old user of Cellprofiler for some months. Nowadays, I use it analysis the intensity and rate of GFP’s expression in chicken myoblast cells and whole embryonic fibroblast cells.
We have blue photos in which nuclei are stained with Hoechst33258 and green photos that are GFP’s expression. But there are only some cells expressing GFP rather than all. Besides, the cells’outlines are long and irregular, the intensity of cells are discrepant, so we don’t get perfect cell’s outline although we adjust the Threshold correction factor in IdentifySecondary.

Here is an example of pictures:
picasaweb.google.com/chenna666

I have created a pipeline about it. Here is my current pipeline:
LoadImages
ColorToGray
ColorToGray
IdentifyPrimAutomatic
IdentifySecondary
MeasureObjiectIntensity
ExportToExcel

For ColorToGray
Call the resulting greyscale image: BlueGray, gfpGray
For IdentifyPrimAutomatic
Otsu Adaptive
Threshold: 1.4
Size of smoothing filter: 9
Suppress local maxima within this distance: 10
For IdentifySecondary
Propagation
Set interactively (in intensity distribution, I choose 0.08 of image 1, 2 and 0.195 of image 3, 4)
For MeasureObjiectIntensity
Call the grayscale images: gfpGray
Call the objects: cells

In the Excel of the cells’intensity, there are some cells whose mean intensity is higher than 0.08 in image of 1, 2 and 0.195 in image of 3, 4, others are lower. So I choose the former as GFP-positive cells. I want to know whether the way is right.
And there is some diversity about the intensity of different images, even though the backgrounds’ intensity is different, so I can’t compare with the mean intensity of many images. I think I can adjust them through some coefficient, but I don’t know. How shall I do?

I look forward to your prompt help.
Thanks a lot.
ChenNa


#2

I am eager to get your help.

Thanks a lot.


#3

Hi,
Sorry for the delay. Have you tried changing the lower bounds for the threshold? also, is the issue with not identifying the entire cell properly, or a ‘declumping’ issue (where a group of cells are not properly segmented into individual cells)?

thanks,
martha


#4

Hi,
I have changed the lower bounds for the threshold, but it is useless. My issue is that it can’t identify the entire cells properly, especially the cell whose intensity is weak. I have changed the threshold but the result can’t also contain the whole cells.
Although I want to get the whole cell’s outline, it is not my emphasis. I want mostly to know whether the way that I analysis the intensity of cell is right.

Thanks,
ChenNa


#5

Hello,
I think we still aren’t clear about what you are asking… are you (1) having trouble identifying the cells themselves properly (using IdentifyPrimaryAutomatic), or (2) are you having trouble determining which cells to call ‘GFP-positive’ and which cells to call ‘GFP-negative’, based on a MeasureObjectIntensity module? Could you explain a bit more?

A couple of tips in the meantime: For case (1), you might try an Adaptive threshold method. I recommend you do not type in a number for the threshold but instead try something like Otsu Adaptive. If it is too stringent or lenient, then look at the help for how to adjust the Threshold Correction Factor to make it work better. For case (2), it might help you to use the Data Tool called “Show Data on Image” once you’ve run your pipeline. This would allow you to show on each cell its intensity value. This might help you choose a good threshold to use to define GFP-positive vs negative cells. Once you pick a threshold, you might use ClassifyObjects to count how many cells are positive or negative.

Let us know how you are doing!

Anne


#6

Hi,
Yes. Both of them are my questions. For case (1), I can identify the nuclei using IdentifyPrimaryAutomatic exactly, but have trouble in identifying the cells using IdentifySecondary. In it, I tried the Otsu Adaptive, but it is not suitable. I know I can adjust the Threshold correction factor through the help. But I adjusted manually the threshold according to the Set interactively before, is it OK? And is the intensity that I choose the threshold of cell’s mean intensity?

For case (2), I chose the GFP-positive cells in the Excel of the cell’ s intensity before. I think the cells whose mean intensity is higher than the threshold that I choose in IdentifySecondary are the GFP-positive cells, is it OK? Is the way identifying the GFP-positive cells right? And you tell me to use ClassifyObjects to count how many cells are positive or negative. I don’ t understand all of them. I can see the date on image, but I don’ t understand how I use the ClassifyObjects to count the positive cells.I see the manual, there is no help about the meaning of “To create evenly spaced bins”, “enter the bin labels” and I can’ t mean the display of ClassifyObjects.

Thanks a lot.

Chen Na


#7

I have thousands of photoes to wait for analysis. Thanks for your help quickly.

Chen Na


#8

Hi,
I took a look at the photos you posted on the web, and it doesn’t seem to me that every cell is visible in the GFP channel, so certainly there is no way you can set the settings to find the outlines of the cells using GFP. The cells vary too much, and the dim ones are so dim that I am not surprised you are having trouble. Thus, I recommend using the Distance option in IdentifySecondary to simply identify a small region around each cell to define (artificially) as the cytoplasm. I think it’s your only option, unfortunately, given that it’s very difficult to outline the cells even by close examination by eye.

For (2), please just try out the modules and experiment with the different settings, in addition to reading the manual. Download the Classify colonies pipeline from the Example page at cellprofiler.org and see how it works there to learn about Classification of objects. To answer your specific question, you want to classify your cells into two bins (positive, negative are the two bin labels). You don’t want evenly spaced bins but instead you want to set a threshold and count the percentage of objects above that threshold. See the example classify colonies pipeline to learn more. I would not choose the threshold for positive-counting cells based simply on whatever threshold was used to identify the outlines of the objects. Use ShowDataOnImage to display the measurement of interest (e.g., MeanIntensity of GFP in cells) and pick a threshold in that measurement that distinguished positive from negative cells. For this, you must use your own judgment to select which feature and which threshold, according to the biological question you are trying to answer.

Martha (mvokes@broad.mit.edu) can send you a tutorial on the classify colonies pipeline that explains more; it’s not yet published.

Best wishes, and keep us posted on your progress.

Anne


#9

Hi Anne,
Thanks for your help. I understand your meanings and will analysis the cell according to the way you told me. I will trouble you in the across of analysis and writing the manuscript.

Chen Na