Multi-Image Colocalization and Its Statistical Significance

Abstract

Accurately localizing molecules within the cell is one of main tasks of modern biology, and colocalization analysis is one of its principal and most often used tools. Despite this popularity, interpretation is often uncertain because colocalization between two or more images is rarely analyzed to determine whether the observed values could have occurred by chance. To address this, we have developed a robust methodology, based on Monte Carlo randomization, to measure the statistical significance of a colocalization. The method works with voxel-based, intensity-based, object-based, and nearest-neighbor metrics. We extend all of these to measure colocalization in images with three colors. We also introduce three new metrics; blob colocalization, where the blob consists of a local maximum surrounded by a three-dimensional group of voxels; cluster diameter, to measure the clustering of fluorophores in three or more images; and the intercluster distance to measure the distance between these clusters. The robustness of these metrics was tested by varying the image thresholds over a broad range, which produced no change in the statistical significance of the colocalizations. A comparison of blob colocalization with voxel and Manders colocalization metrics shows that the different measures produce consistent results with similar values for significance and nonsignificance. Using our methodology, we are able to determine not only whether the labeled molecules colocalize with a probability greater than chance, but also whether they are sequestrated into different compartments. The program, written in C++, is freely available as source, as well as in a Linux version.

Steady-State Antigen Scavenging, Cross-Presentation, and CD8+ T Cell Priming: A New Role for Lymphatic Endothelial Cells

“…After image deconvolution (Huygens Deconvolution software; Scientific Volume Imaging, The Netherlands), Fiji software (National Institutes of Health, Bethesda, MD) with the Image 5D plugin was used to generate the figures. To quantify fluorescent NP or OVA colocalization within clathrin-positive vesicles, single z-planes from deconvolved images were analyzed using a script that determines the statistical significance of object-based colocalization by comparison of the colocalization occurrences on actual images with colocalization by chance (23). The ImarisXT MATLAB plugin "split spots onto surface objects" was used to determine whether NP + or OVA + spots were within or outside of LAMP-1 + surfaces.…”

Couplons in rat atria form distinct subgroups defined by their molecular partners

“…Our analysis technique (Fletcher et al, 2010) allowed us to derive three useful metrics for describing the positioning of the molecules within the atrial cell; these are the median nearestneighbour distance of the individual molecules (Table 5); the cluster diameter, a measure of how closely associated the colocalized molecules are (Table 6); and the inter-cluster distance, a measure of how separated the colocalizations are. The values in Table 6 include only those colocalizations that were statistically significant (see Tables 1 and 2).…”

“…A with B and C) were counted as triply colocalized and separate from the dual colocalizations (e.g. A with B, B with C, and C with A) given that the triple grouping was thought to be functionally different from the doublets (Fletcher et al, 2010).…”

Quantifying molecular colocalization in live cell fluorescence microscopy

“…It turns out that the most commonly used parameter, the so-called Pearson's Correlation Coefficient (PCC) is superior to the Manders Overlap Coefficient [13,14] (MOC) which is currently implemented in most commercial software packages for image processing [15]. A closer look at the PCC, however, reveals that it has some frequently cited major drawbacks which significantly limit its utility for colocalization analysis in certain cases [16][17][18][19][20]. For example, it was found that the PCC, which can adopt values between þ1 and À1 as a measure of the similarity of channels (+1: perfect colocalization; values between 0 and À1: no colocalization), is very sensitive to strong intensity fluctuations or threshold variations.…”

“…Similarly, the PCC delivers erroneous results if the noise level of the different color channels is significantly different. Lastly, the PCC does not allow for colocalization analysis of images with more than two color channels [20].…”