The authors of the recent review on colocalization (Bolte & Cordelières, 2006) seem to misunderstand the Pearson correlation coefficient.
The authors conclude that ‘Values other than those close to 1 and especially in mid range coefficients (−0.5 to 0.5) do not allow conclusions to be drawn’. This transforms a very widely accepted graded measurement of the degree to which two populations are related into either a yes or a no.
Implicit is that correlations other than one or zero have no biological meaning. Clearly if two molecules have a strong one-to-one association and are present in equal number a correlation of 1 is possible. However, associations may not be exclusive, molecules may not be present in ratiometric numbers and they may have different spatial distributions. We have found intermediate values showing partial colocalization within the plasma membrane to be biologically relevant (Parmryd et al., 2003). The suggestion that the Pearson correlation coefficient provides a yes or no answer, with a transition at some unstated value, is wrong. Underlying this misinterpretation are two discordant views of colocalization: (a) the occurrence of pixels or objects that merely include both fluorophores or (b) the relationship between their intensities. The former is clearly binary but the latter is not. It might therefore be desirable to differentiate between these alternatives and call the former coappearance and reserve the term colocalization for the more subtle measurement.
The authors show that noise reduces the measured Pearson correlation, hence the requirement in the first study on quantitative colocalization that images should be of high quality (Manders et al., 1993) or that some compensation should be introduced. They also show that including pixels where the intensities of one fluorophore cannot be distinguished from background reduces the measured Pearson correlation. This seems to represent a failure to acquire images properly, rather than to a flaw in the analytical method. The accuracy of other measures of colocalization also depend upon the quality of the images and on the actual detection of the fluorophores, though measurements of coappearance, especially when confined to the centroid of objects, are likely to be less sensitive to noise.
A minor issue is the statement that ‘noise and background must be removed’, but the Pearson correlation coefficient is not affected by global shifts in the background, only by variability in the both the signal and background.