Microscopy by default is a technique that allows us to observe, rather than measure, biological events and make conclusions based on what we see, rather than based on some calculations. The topic of this article may therefore seem a little surprising since it deals with the measurements of colocalization. While there are situations where you can determine protein colocalization visually, a more accurate confirmation of a mutual distribution of two probes will often be necessary.
Why Is a Visual Determination of Colocalization Insufficient?
Only if you have collected images following a controlled protocol for colocalization analysis (sufficient signal in each channel, no autofluorescence or signal bleed-through) is it safe to say that the two probes colocalize solely based on observation. In this case, if the green signal colocalizes with the red signal and the picture is mostly yellow, no statistical measurement is needed. But if you don’t see an overlap, it doesn’t mean it isn’t there! It can be that the intensities of the two channels are not the same. Namely, the intermediate color that we see can only appear if both probes are of the same intensity. Therefore conclusions based on visual determination can often lead to false negative results.
What Methods for Overlap Measurement Exist?
While there is no one standard approach to this matter, there are two methods that are widely used and generally accepted. These involve either Pearson’s correlation coefficient (PCC) or Mander’s colocalization coefficient (MCC).
These two methods relate to two different aspects of colocalization – correlation and co-occurrence. Pearson’s coefficient is related to the correlation of the pixel intensities in the two channels. It measures the relationship between signals – whether the signal values in one channel rise simultaneously with the other, or one signal falls when the other rises. Correlation is distinct from co-occurrence, which is mathematically expressed through Mander’s coefficient. This represents the coverage of one signal over the other, which reveals the extent to which two probes occupy the same place. Let’s explore this a bit further.
Pearson’s Correlation Coefficient (PCC)
Definition: PCC reflects the linear relationship between signal intensities. The values can range between 1 (perfect positive correlation) and -1 (perfect negative correlation), while 0 means that there is no correlation. It is possible to explore PCC visually through a scattergram where the coordinates on the plot represent pixel values (signal) in both channels. The more dots that cluster around a straight line, the better the correlation between the two signals
Requirements: The PCC should be measured in the region of interest (ROI) to avoid false positives or negatives. If measuring PCC over the entire image, pixels of the background will correlate perfectly and inflate the PCC. In contrast, if the measurement is performed in a region of no interest where there is a heterogeneous distribution of both channels, the PCC will be depressed.
You can select ROI by hand or by thresholding to exclude background. Whichever way you do it, be careful, especially when thresholding. The selection of the ROI needs to include all the relevant regions of the cell(s) i.e., every place where a probe can be expected to distribute. If you are using an intensity-based method for selecting ROI (thresholding), you might inadvertently exclude relevant results. How can this happen? You could have a region of mutual exclusion, where neither label appears, and this can be a biologically relevant result (both molecules are not expressed on that place in the cell). However, thresholoding will not include this region in the ROI, putting you in danger of losing relevant results.
Recommended for: PCC should be used on images were there is a linear relationship between intensities. If the data fits to a more complex model, PCC will not perform well. A different method should also be chosen if there is an uneven overlap, where probes co-distribute but in different proportions. This may occur when GFP is used as one probe. Its expression level may differ between cells, and potentially cause depression of PCC due to high intercell variability.
Mander’s Colocalization Coefficients
Definition: MCC is a metric that describes co-occurrence – the fraction of one protein that colocalizes with the other. MCC will give you a good measure of colocalization when you label one protein in a vesicle and want to see how it colocalizes with a certain structure in a cell, say a microtubule. If we assume that all vesicles colocalize with microtubules but only a proportion of microtubules colocalizes with the vesicle, you can calculate MCC for each channel and get a metric that quantitatively describes this fractional overlap.
Requirements: The catch with MCC is that it requires the elimination of background. The trickiest part here is setting a cutoff point for intensity that will enable background subtraction. MCC measures the complete fluorescence of one probe in every above-zero pixel. However, above zero pixels are extremely rare, due to factors such as: autofluorescence, light leakage, non-specific labeling, or fluorescence from out of focus image plains. The measurement of MCC therefore requires a careful selection of the threshold (cutoff).
The first way to do this is global thresholding, where you subtract a threshold value from each pixel, so that every level below the selected cutoff will be background and every pixel above will fall into a region of interest. While this is very intuitive, global thresholding is rather crude and may lead to unwanted situations like the exclusion of the low value pixels that are close to background but are in fact positive.
A more automatic and less subjective option is Costes method where threshold is estimated by calculating PCC multiple times. This serves to define the range of pixel values that are positive and therefore should not be excluded. PCC is calculated for different groups of pixels, and the pixel values for which PCC is equal to or close to zero are taken as the threshold values. Nevertheless, it should also be checked visually, since in images with lower signal-to-noise ratio it may identify a very low threshold level so that it doesn’t distinguish labeled structures from the background.
Recommended for: When the biological question of your experiment concerns the extent that protein/structures overlap, MCC should be the measure of choice. These coefficients are more intuitively interpreted than PCC, and are independent of signal proportionality (the differences in the number of structures labeled by each probe).
Before You Begin Your Analysis
Whichever choice for measuring colocalization you opt for, it is very important that the image collection is performed in a correct way. Try to control as many factors as you can, and prepare a set of control samples that allow you to monitor as many variables as possible.
Keep in mind that visual colocalization determination is influenced by many factors, including the subjectivity of the observer, the fact that each individual’s brains may see colors differently, a possible color blindness of the observer, and the spatial resolution that determines the pixel size, amongst others. Be sure to process the images you are about to analyze in a uniform manner, and bear in mind that any uncontrolled manipulation may cause you to lose relevant information.
And Last but Not Least
These calculations are implemented in most image analysis software packages e.g., ImageJ and Volocity. But, given that measuring colocalization is still a somewhat confounding field, improvements are required, so keep track of the new versions and packages for the software.
How do you measure colocalization? Let us know by writing in the comments section!
Dunn KW, Kamocka MM, McDonald JH. A practical guide to evaluating colocalization in biological microscopy. American Journal of Physiology – Cell Physiology (2011), 300 (4) C723-C742
Adler, Jeremy, Parmryd, Ingela: Colocalization Analysis in Fluorescence Microscopy. In: Taatjes, Douglas J., Roth, Jürgen (ed.), Cell Imaging Techniques: Methods and Protocols, New York: Humana Press, 2012, pp. 97-109