On Thu, 27 Aug 1992, James Slattery wrote:
 
>    We have recently been using the confocal microscope to examine
> colocalization of two proteins using MAb techniques. The signal level from
> one of the proteins is lower than the other by ca. a factor of 2-3.
> I am sure that many of you have observed that using such data it is possible
> by judicious selection of the intensities of the display colors (one for
> each protein) it is possible to make it appear that the proteins are
> colocalized or are not colocalized. The question is how does one establish
> 'reality'?
>     At first we assumed that if you could adjust the balance to show that
> they were colocalized then they were colocalized since for features that are
> not colocalized in our controls it is not possible to adjust the color scales
> to show colocalization. The problem with this assumption is that in more
 highly
> stained and thicker material, such as that with which we are working, there
> may be a sufficient level of out of focal plane signal contributing to our
> structures' signal. This structure might then be interpreted to be colocalized
> when it is not. Any ideas?
>
> Jim
 
Jim-
 
We have used analytical methods to create a mathematical description of
the colocalization. Depending on the specific question, we have used a
couple of different techniques.
 
On a lymphocyte, we may ask the question:
Of all of the points on the screen, what is the correlation of the
red:green images. This is done by simply doing a regression analysis with
the individual points at each location as the inputs. This generates a map
relative to intensities.
 
In a different situation, we may care more about the localization and not
the relative intensity. In this case, we would set a threshold for
"positivity" in each channel and use boolean functions to ask the
question: of all the green pts, how many are red, of all the red points
how many are green, and of all points, how many are red and green.
 
If I care more about exactly the alignment of the fluorochromes, you can
take a mean profile through a structure (eg across a membrane) and produce
a mean cross sectional map of the red vs. green channels.
 
A couple of caveats:
 
1) We can do this because we trust the alignment of our two images. We
trust it beacause we only use a single excitation wavelength and often
collect the images simultaneously.
 
2) We do this with green (FITC) and red (Texas Red or PI) probes but it
will work regardless of the color if condition 1) is met.
 
3) Do not trust what you see on the screen of an RGB monitor for
colocalization. This is because of the way R, G, and B are created on the
screen. Take the testcard image and merge it onto itself in R-G, G-B, and
R-B and you will see the severity of the misalignment. We only trust a
film recorder for pictures and the above methods for numerical analysis.
 
Paul Goodwin
FHCRC