Fluorescence intensity analysis with tissues

[spetry]

Hello there,

the quantative analysis of immunostained specimen troubles me and unfortunately the few things which I could find on google and in this forum did not match my concerns, so I deceided to open a new thread and ask for help.

I am working with a mouse model and got PCR results from isolating islets of Langerhans from the pancreatica. Now I want to confirm the results with immunofluorescence, to be precise: intensity measurement of target proteins on mouse pancreatic tissue, amongst which insulin.

Here is my method:

I started by taking greyscale pictures of the islets on my slides and used exactly the same settings (time, gain, gamma, ...). The same was done with blank slides (without primary antibody).

In Image J I used the image calculator to substract the picture from the blank slide from the picture I want to analyse in order to remove background.

Then I selected the islet area with the freehand tool and measured the mean grey value.


Now here are my questions:

1) Can you do it like this or did I miss something?

2) I read that people normalize the intensity with the area when working with cells. Should you also do that when working with tissue and if yes, how is the best way to do it?

3) What about using a threshold in order to get clearer results? I thought about it, but I got pictures with very low staining and others which very high intensity, so I could not use the same threshold for all pictures. I wonder whether it's a good idea or even scientifically correct to set the threshold to the mean grey value +/- 50 % on every picture or something like that.

Maybe someone can give me some inspiration, thank you in advance! :)

Best regards,

Sebastian

[Sean Burke]

Hi Sebastian,
         Have you thought about exporting to FCS Express Image Cytometry for analysis of the segmented data. There are some people doing this with histology sections already and there are some examples at https://www.denovosoftware.com/site/ImageApplicationExamples.shtml#histology.


Regards,
Sean Burke




-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of spetry
Sent: Wednesday, September 12, 2012 2:13 PM
To: [hidden email]
Subject: Fluorescence intensity analysis with tissues

Hello there,

the quantative analysis of immunostained specimen troubles me and unfortunately the few things which I could find on google and in this forum did not match my concerns, so I deceided to open a new thread and ask for help.

I am working with a mouse model and got PCR results from isolating islets of Langerhans from the pancreatica. Now I want to confirm the results with immunofluorescence, to be precise: intensity measurement of target proteins on mouse pancreatic tissue, amongst which insulin.

Here is my method:

I started by taking pictures of the islets on my slides and used exactly the same settings (time, gain, gamma, ...). The same was done with blank slides (without primary antibody).

In Image J I used the image calculator to substract the picture from the blank slide from the picture I want to analyse in order to remove background.

Then I selected the islet area with the freehand tool and measured the mean grey value.


Now here are my questions:

1) Can you do it like this or did I miss something?

2) I read that people normalize the intensity with the area when working with cells. Should you also do that when working with tissue and if yes, how is the best way to do it?

3) What about using a threshold in order to get clearer results? I thought about it, but I got pictures with very low staining and others which very high intensity, so I could not use the same threshold for all pictures. I wonder whether it's a good idea or even scientifically correct to set the threshold to the mean grey value +/- 50 % on every picture or something like that.

Maybe someone can give me some inspiration, thank you in advance! :)

Best regards,

Sebastian



--
View this message in context: http://imagej.1557.n6.nabble.com/Fluorescence-intensity-analysis-with-tissues-tp5000037.html
Sent from the ImageJ mailing list archive at Nabble.com.

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Olivier Burri]

Dear Sebastian,

> The quantative analysis of immunostained specimen troubles me and unfortunately the few things which I could find on google and in this forum did not match my concerns, so I deceided to open a new thread and ask for help.

First off thank you for the post and the details regarding your
experiment. Very often we see people using fluorescence quantification
without thinking too much about whether they are allowed to do it or
not.

> I started by taking pictures of the islets on my slides and used exactly the same settings (time, gain, gamma, ...). The same was done with blank slides (without primary antibody).
>
> In Image J I used the image calculator to substract the picture from the blank slide from the picture I want to analyse in order to remove background.
>
> Then I selected the islet area with the freehand tool and measured the mean grey value.
> 1) Can you do it like this or did I miss something?

From what you say in 3) you seem to have a very large difference
between your two conditions, so this should be doable. Some
considerations though.
1) How many different replicates do you have? (Of the same animal, of
other animals prepared in another batch). Ideally I think it would be
good if you assessed the variability within the same animal and the
variability between animals for both your WT and "treated" conditions.
That way, hopefully the difference you find between the conditions
will be larger than the uncertainty and you can be safer in your
interpretations.

2) Did you perform your stainings in batch, so as to minimize
variabiliy coming from the staining protocol?


> 2) I read that people normalize the intensity with the area when working with cells. Should you also do that when working with tissue and if yes, how is the best way to do it?

Usually when you normalize with the area it's to get a density
measurement, and is part of a different type of experiment where you
find your cell area (threshold) and you normalize that with the tissue
area to estimate the density of your sample. I am not sure about
measuring intensities inside cells (Using a threshold I guess) and
then normalizing the intensities. You'd get something like Pixel
Intensities / um2 which doesn't mean much more than just average
intensity...

When you normalize intensity measurements, you do it as ratios of
signals coming from within your own sample, like when you check for
the presence of actin on Western blots. Do you have some marker you
could use as an internal control? Some cytoplasmic marker for example,
that should be homogeneous no matter the condition, that would allow
for normalizing your signal to this one. It can also be a good
indicator that the staining protocol happened correctly.


> 3) What about using a threshold in order to get clearer results? I thought about it, but I got pictures with very low staining and others which very high intensity, so I could not use the same threshold for all pictures. I wonder whether it's a good idea or even scientifically correct to set the threshold to the mean grey value +/- 50 % on every picture or something like that.
A way we try it out here is to test out all the available automatic
thresholds available on Fiji/ImageJ using a macro. First the user sets
the threshold manually on a set of images (about 10 or more). Then the
plugin launches the auto-threshold methods on the images and recovers
their values. Then the plugin compares your values to the ones that
were automatically selected and recommends a threshold and a
correction factor that could be used (Something that is done in Cell
Profiler, for example).
However thresholds are usually not set on the data you want to measure
but on some other channel that acts as a mask to your data. Something
that only stains the tissue or cells that you are interested in.

>
> Maybe someone can give me some inspiration, thank you in advance! :)

Hope that's a bit of inspiration. If you'd like the little macro I was
mentioning, let me know. I'd have to comment it a bit to make it
useable but would be happy to share it.

Best

Oli
--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[spetry]

Hi,

first of all let me thank both of you for your input.

Sean Burke wrote

[..]
Have you thought about exporting to FCS Express Image Cytometry for analysis of the segmented data.
[..]

I will have a look at FCS Express, thanks for the link.

Olivier Burri wrote

[..]
From what you say in 3) you seem to have a very large difference
between your two conditions, so this should be doable. Some
considerations though.
1) How many different replicates do you have? (Of the same animal, of
other animals prepared in another batch). Ideally I think it would be
good if you assessed the variability within the same animal and the
variability between animals for both your WT and "treated" conditions.
That way, hopefully the difference you find between the conditions
will be larger than the uncertainty and you can be safer in your
interpretations.

2) Did you perform your stainings in batch, so as to minimize
variabiliy coming from the staining protocol?
[..]

... [show rest of quote]

Maybe some more information here which I forgot to mention, sorry.
Basically, I am interested in Insulin and two different antioxidant enzymes.
I stained in two batches, respectively to the latter two. Insulin was stained in both batches in order to be able to clearly identify the islets of Langerhans. Also, nucelus staining was done with Hoechst 33342.

The experiment includes two groups of mice with three corresponding time points each. I stained two sections from three mice each for each group and time point in order to have a good and statistically trustworthy average.

Most important, the huge difference in intensity appears between the two different enzyme stainings. Those do not have to be comparable, so I suppose my worries about the difference are softened.

Olivier Burri wrote

[..]
Usually when you normalize with the area it's to get a density
measurement, and is part of a different type of experiment where you
find your cell area (threshold) and you normalize that with the tissue
area to estimate the density of your sample. I am not sure about
measuring intensities inside cells (Using a threshold I guess) and
then normalizing the intensities. You'd get something like Pixel
Intensities / um2 which doesn't mean much more than just average
intensity...

When you normalize intensity measurements, you do it as ratios of
signals coming from within your own sample, like when you check for
the presence of actin on Western blots. Do you have some marker you
could use as an internal control? Some cytoplasmic marker for example,
that should be homogeneous no matter the condition, that would allow
for normalizing your signal to this one. It can also be a good
indicator that the staining protocol happened correctly.

... [show rest of quote]

... [show rest of quote]

As mentioned above I also stained nuclei. Maybe I can normalise the other stainings with pictures taken with the respective filter, for those should be stained equally. I think I will try this approach. Taking several images from random positions on every mouse / timepoint / group and calculating the average might be a good start!
I will study the density issue further, but for I am measuring the intensity in the whole islet area and not in single cells using the area does not seem like a reasonable approach. I tried to divide the measured intensities by the mean islet area, but that also seems not reasonable to me now.

Olivier Burri wrote

A way we try it out here is to test out all the available automatic
thresholds available on Fiji/ImageJ using a macro. First the user sets
the threshold manually on a set of images (about 10 or more). Then the
plugin launches the auto-threshold methods on the images and recovers
their values. Then the plugin compares your values to the ones that
were automatically selected and recommends a threshold and a
correction factor that could be used (Something that is done in Cell
Profiler, for example).
However thresholds are usually not set on the data you want to measure
but on some other channel that acts as a mask to your data. Something
that only stains the tissue or cells that you are interested in.

[..]

Hope that's a bit of inspiration. If you'd like the little macro I was
mentioning, let me know. I'd have to comment it a bit to make it
useable but would be happy to share it.

... [show rest of quote]

... [show rest of quote]

My thoughts about thresholds were based on the idea of further reducing background / low and high peeks in staining, but your point seems logic to me. If you don't mind I would like to try out the macro of yours anyway, just to see how the data behave.

Thank you very much for your detailed answer, I appreciate it a lot!

[spetry]

Hello,

I followed the idea of using nuclei stained with Hoechst 33342 as an internal control. Hoechst does not need secondary antibodies, it's a fluorochrome itself.
As mentioned above I've got slides without any primary antibody, two per timepoint and animal which means 36 samples in total (remember: three animals per timepoint, three timepoints, two groups of mice).

I took two pictures of random locations from each sample and calculated the average for every animal and timepoint which makes 18 average intensity values.
Then, I calculated the common average in order to see how much the staining intensities of the nuclei stained with Hoechst 33342 vary from it on each animal.

The most extreme average values are 5 % away from the total average, so I suppose that I can use Hoechst as the internal control.


Now I am thinking of the following in order to normalize the intensities of my target proteins:

Again, I will take pictures from nuclei stained with Hoechst, this time from the slides which were stained with the respective protein, separately for each protein. This way, only batches will be compared. I will make Hoechst intensity averages for each animal and normalize those by the total average of all animals in this batch.


Now I've got two questions:

1) Is my conclusion itself about the low variation in nucleus staining intensity as well as the method with which it was achieved (using slides w/o primary antibody) reasonable and correct?

2) Are you allowed to normalize the staining intensity the way I suggest (normalizing a protein in a batch to this batch's own nucleus intensity average)?

Thank you in advance for replies!

Best regards,

Sebastian

[Mbar]

Hello Sebastian

I followed this tread with great interest, because I also want to measure fluorescence intensities in tissue. I´m staining cell nuclei with DAPI, I think its similar to Hoechst, as DNA stain.
I don’t know, If I get you right: you want no normalize the intensity of your target protein to the average intensity of your nuclear staining?
If I got you right, then I´m unsure with the idea to get an internal control by creating the average of nuclear stains.´, because I made the  experience that immunohistochemistry can go wrong but DAPI staining still works on the specimen- independent of e.g.  age or pretreatment of the specimen.
For example, I treated specimens that were 40jears old-DAPI works great- others didn´t work.  So I think there´s no correlation between DAPI staining and other immunofluorescent signals- and I think for Hoechst it’s the same thing. In my opinion it gives you no more information about your intensities, if you normalize it to nuclear staining intensity.
But the question for me is, if you really need an internal control, I mean are your staining results per sample are so extremely different?
If its like that, perhaps you can find a protein, a typical Langerhans cell marker, then double-stain with your target protein and normalize it to the average intensity of this protein per Langerhans cell?


Best regards,
MBar