Colocalisation

Colocalisation and Fluorescence Microscopy

In fluorescence microscopy, colocalisation is widely used method to compare the distibution patterns of cellular components. As the resolution of optical systems is improving, colocalisation analyses will become both increasingly popular and meaningful.

The size of a detectable biological change depends upon the accuracy of the method used to detect the change.

Fluorophores are frequently described as being colocalised, which may mean that the fluorophores appear to be in the same cellular compartment(s) or that there is a relationship between the concentrations of the two fluorophores.

Co-occurrance

Unless access to the cellular compartment is restricted or unless binding sites exist, then this colocalisation may simply reflect a physico-chemical similarity between the two molecules, which must include the contribution of the fluorophores. Potentially quite a trivial observation that might usefully be called co-occurrance.

Spatial Corrleation

If there is a direct interaction between the molecules or an interaction with a secondary molecule or structure then a matching variation in intensities would be expected. Therefore a relationship between the intensities in individual pixels suggests a degree of biological interaction - a more interesting observation than mere codistribution. We suggest the fluorophores might be described as being spatially correlated, which could be abbreviated to correlated if the context is clear.

Colocalisation: Differentiating Between Co-occurrance and Correlation

1) Overlay Picture

Overlaying the images of the individual fluorophores using different layers of an RGB picture - red plus green will display as yellow, or red and blue as magenta. BUT the third colour only appears if the intensities of two images are matched - manupulations in for instance Photoshoptm can radically alter which pixels appear to show colocalization of the fluorophores. The link to matching intensities, and therefore the potential for manipulation, is greatly reduced by using thresholded images for each fluorophore. Thresholding is usually based simply on an intensity cutoff but becomes problematic if the images are noisy and there is an overlap between the intensity range of the background pixels and the intensities of the fluorophore containing pixels.

2) Scattergram.

A relationship between the intensities is difficult to estimate by eye but can be shown using a scattergram, which shows the occurrence of different combinations of intensities in homologous pixels. There are two forms of scattergrams (a) that shows that a particular combination of intensities exist and (b) that shows the frequency of each combination of intensity.

Quantifying Correlation

The Pearson correlation coefficient measures the relationship between two sets of data using a range of 1 perfect (match) through 0 ( no relationship) to

-1 (inverse relationship).

Biological Meaning of Correlation

A perfect correlation would require a tight stochiometric interaction between the fluorophores and a match in the number of molecules. An inverse relationship could occur between a enzyme and a substrate. A negative correlation could also reflect avoidance by the two fluorophores.

Problems in Correlation Measurements

The biggest problem arises from image noise. A fluorescent image must contain Poisson noise, due to the nature of fluorescence, and usually also some background noise. Noise of either type will reduce the apparent correlation and we can differentiate between the measured, or apparent, correlation and the true, or underlying, correlation, that could be measured if the images were of perfect quality and free from all forms of noise. In practice it is very difficult, if not impossible, to obtain images that are free from noise. When the image quality is poor even a perfectly correlated pair of images will have a low apparent correlation.

The size of a detectable biological change depends upon the accuracy of the method used to detect the change.

Correlation measurements are critically dependant upon image quality and since it is very difficult to obtain perfect images, or even images of consistent quality, correlation measurements may confuse rather than enlighten. Importantly the accuracy of correlation measurements made from noisy data is not improved by averaging repeated measurements or by increasing the number of pixels used in the calculation. This can improve the precision but the measured correlation may still be wildly inaccurate.