There are lots of ways to mess up colocalization measurements - the  
coefficients still have reassuring numbers of decimal places but their  
value shrinks.

ROIs and thresholding - we recommend both. The ROI should have  
biological relevance and thresholds should be set for both channels  
that exclude background pixels. This is illustrated in Fig 1 of our  
hot off the presses article (Cytometry August 2010, Adler & Parmryd).  
The article also shows that including background pixels can actually  
increase the measured correlation and also considers how the  
correlation measurement is confused when the there is more than 1  
relationship present (always look at the scattergram). We also argue  
strongly that the Manders Overlap Coefficient (not M1 and M2) is, at  
best, a highly confused measurement and should be abandoned - use the  
Pearson, Spearman or ICQ correlation coefficients instead.

Setting thresholds for both channels then measuring the colocalization  
between pixels that contain both fluorophores means that the remaining  
fluorescence is excluded from whichever correlation coefficient you  
use. Therefore we recommend qualifying the correlation with the M1 and  
M2 coefficients which give the fraction of the total fluorescence for  
each fluorophore to which the correlation applies. M1 and M2 are not  
measures of correlation but are measures co occurence. You might also  
report the fraction fraction of the pixels that contain both  
fluorophores in your chosen ROI.

Adler, J. and Parmryd, I. (2010), Quantifying colocalization by  
correlation: The Pearson correlation coefficient is superior to the  
Mander's overlap coefficient. Cytometry Part A, 77A: 733?742. doi:  
10.1002/cyto.a.20896

Jeremy Adler



> Hi all,
>
> Pardon me for opening up a can of worms of colocalization issue.
>
> I wonder if anyone has experience in thresholding a biological
> object using a channel (say, green actin in a contractile ring)
> and use only that selected ROI to quantify Pearson coeefficient
> between green and a red-labeled protein? Bottom line question
> is: how much of the red protein co-localizes to green within
> the object and not the whole cell?
>
> We have Volocity, MetaMorph and ImageJ.
>
> Thanks in advance for your help.
>
> Regards,
> Leong
>
>
> --
> Teng-Leong Chew, Ph.D.
> Director, Cell Imaging Facility & Nikon Imaging Center
> Director of University Imaging Resources
> Feinberg School of Medicine & Office of Research
> Northwestern University
> 303 E. Chicago Avenue
> Chicago, IL 60611
> (312) 503-2841
> (847) 491-7091 (W, F)
>



Jeremy Adler
Genetics & Pathology
Rudbeck Labs
Uppsala U
Sweden

0046 (0)18 471 4607