Imaging of fluorescent signals is commonly carried out in biological sciences for investigating localization of proteins, organelles, ions, and cells. A common question of researchers is whether two molecules of interest are located in the same region, (i.e., whether colocalization occurs). Colocalized of signals occurs when they are observed in overlapping areas. Colocalization is defined as the presence of two or more different molecules in close spatial positions in a sample (Smallcombe, 2001).
Conventional colocalization is where two (Fig. 1) or more antigens (Figs. 2, 3) are viewed in the same section by secondary antibodies labeled with fluorescent compounds that have different excitation spectra. This is based on a dye-overlay method, where one antibody is one color (e.g., red) and a second antibody is another color (e.g., green; Li et al., 2004). Each color is separately imaged, and molecules are considered as colocalized in areas stained with the combined colors when the two images are overlaid (e.g., red + green = yellow). This method allows the researcher to determine whether the two different signals are present in the same or different areas. The main benefits of this method are that it is easy and spatial information (shape and contours) is preserved. Although this technique is simple and widely used, it has some disadvantages. The images are affected by the characteristics of the display method and the investigator's perception (preoverlay images are usually adjusted for brightness, contrast, and threshold). In addition, conventional colocalization only indicates the coexistence of two molecules and does not allow any quantification. This method does not address the issue of whether the intensity of staining for the two molecules varies in synchrony, which is expected if they are components of the same complex. Various methods have been developed to solve these issues and are discussed below.
Uses of Colocalization
Colocalization is a commonly reported visual event in the cellular and molecular fields. Researchers can obtain quantitative information on interactions on the molecular scale or localization from two or more images. Findings from colocalization are useful for determining the precise area of structures of interest. Colocalization is important for investigating metabolism, signaling events, and transcriptional control (Zinchuk and Grossenbacher-Zinchuk, 2009).
IMAGING TOOLS USED FOR COLOCALIZATION
Standard epifluorescence microscopy is not able to be used to examine whether two molecules are actually colocalized or whether they overlap in the z dimension. Optical sectioning devices, including confocal or multiphoton microscopes, are used to solve this problem. Multiphoton microscopy is better suited to studying colocalization than confocal systems because both of the fluorescent labels can be excited by a single wavelength. The reason one wavelength can be used is because of the use of extended excitation spectra in multiphoton absorption. Additionally, fluorescence excitation is restricted to a small volume in the sample, and therefore, any fluorescence present must originate from that particular optical plane. Images from confocal microscopes are affected by a high level of background noise, which can be as high as 30% of maximum image intensity (Landmann and Marbet, 2004; Zinchuk et al., 2007). Therefore, the background of images needs to be assessed and corrected before analysis (see section below “Adjustment of images”).
Fluorescence resonance energy transfer (FRET) can be used to examine the close proximity of two fluorescent molecules. FRET is potentially a useful technique for studying colocalization, but the parameters under which FRET occurs must be known (Smallcombe, 2001).
Estimating colocalization requires specialized algorithms (e.g., CoLocalizer Pro software) executed by computer software. Various methods that use different algorithms will be discussed below. For the purpose of this article, techniques and issues associated with confocal microscopy will be discussed.
ISSUES OF COLOCALIZATION
One important issue of colocalization is resolution (Smallcombe, 2001). Before designing a fluorescent staining protocol to evaluate colocalization, the researcher must define the size of the cellular domain being investigated. A low-power lens might be useful for identifying the general location of two fluorescent probes within a cell, but might not be able to conclusively demonstrate whether they are present in the same or different subcellular compartments. However, a high-power lens might clearly show that the two antibodies are bound to different subcellular structures.
One of the major problems with the determination of colocalization is that fluorescence bleed-through frequently occurs between the red and green fluorescent channels. This problem can occur with the red fluorochrome leaking into the green channel and/or the green fluorochrome leaking into the red channel. This can also be the case for autofluorescence or if the fluorescence antibodies cause nonspecific background staining. This bleed-through causes pixels to appear yellow/orange, but does not represent colocalization. Therefore, bleed-through and nonspecific staining must be removed to determine precise colocalization. To avoid this problem, researchers can perform sequential excitation and use narrow band emission filters. Single-labeled controls are used for evaluating any bleed-through and unstained negative controls should be used to determine the degree of background autofluorescence (Zinchuk and Grossenbacher-Zinchuk, 2011).
Appropriate sample preparation is crucial for reliably determining the degree of colocalization. To avoid the issue of bleed-through, the two fluorochromes selected by the researcher should have the least emission overlap possible, with well-separated excitation and emission spectra. The two fluorochromes used should also have the same intensity of brightness. Researchers should also select antibodies that are specific for the antigen and do not have cross-reactivity (Zinchuk and Grossenbacher-Zinchuk, 2011).
Adjustment of Images
Image quality is important for achieving good results. Confocal images need to be adjusted because they have out-of-focus light that comes from adjacent scanned planes. This issue is caused by background noise (pixels of unknown values). If these pixels are not removed, they may hinder the execution of the coefficient algorithms used and lead to inaccurate coefficient readings. Background, which is found to different degrees in all images, is an important factor that restricts the power of colocalization analysis. Background impairs the quality of images by affecting resolution and concealing image detail in the low-intensity range. Many reports that use confocal images that are unadjusted are still published, which raises concern regarding the accuracy of the reported colocalization results, and prevents their use for comparison. Various methods, such as deconvolution, filtering, and background correction, are used to improve the quality of images before analysis of colocalization (Landmann and Marbet, 2004; Zinchuk and Grossenbacher-Zinchuk, 2009).
Selecting a Region of Interest
Selecting the region of interest is important for analyzing colocalization. The region of interest is the part of the image that contains colocalization. Reducing the amount of pixels outside the region of colocalization improves the accuracy of calculations. However, this accuracy depends on the quality of the images. Images that are obtained with correction of background are less affected by selection of the region of interest.
Quantitative analysis of colocalization is important for determining how molecules are spatially organized in immunofluorescence images (Zinchuk and Grossenbacher-Zinchuk, 2009). The main approach used for quantifying colocalization is based on calculation of Pearson's correlation coefficient (“r”) (Manders et al., 1992). This calculation is dependent on how much colocalized signal there is in both channels in a nonlinear manner. This index of correlation is helpful for quantification but no information is provided on the localization of the signals of interest. Estimation of colocalization can also be achieved by calculation of a number of values that represent the proportion of colocalized pixels (Manders et al., 1993). These values are termed colocalization coefficients. Information regarding coefficients, including their meaning and how they should be used, is shown in the review by Zinchuk and Grossenbacher-Zinchuk (2009). Colocalization coefficients quantify the colocalized fraction of each type of molecule. However, a threshold value for each channel is also required for these coefficients. This value is used as a cut-off between specific staining and nonspecific staining. A disadvantage of this method is that the thresholds are usually created by visually estimating the images or performing a segmentation algorithm, which can lead to varying results that are not reproducible.
Intensity Correlation Analysis
Li et al. (2004) developed an approach called “intensity correlation analysis” for quantifying colocalization. Their method is based on measuring the deviation from the mean intensity for each pixel, as well as for both signals in paired images. This method produces a global correlation index for the two images (similar to Pearson's correlation index “r”). Quantification of the correlation between the two original images is obtained using this value. However, this technique does not take into account spatial considerations of the correlation between the two signals.
Modification of the “Intensity Correlation Analysis” Method
Jaskolski et al. (2005) developed a new method called the normalized mean deviation product (nMDP) method, combining quantification and imaging of colocalization. As well as being useful for quantifying and comparing colocalization of two fluorescent stains, their approach is useful for measuring the distribution of proteins, organelles, ions, and cells using a variety of methods. This method allows quantification of correlation, and preserves spatial data by examining the correlation between pairs of individual pixels instead of between global images. Correlation images reveal what cannot be seen by the naked eye. The nMDP method defines areas of interest in the paired images, and creates one area to examine correlation of signals. This method involves analysis of the fraction of staining that is overlapped and the correlation index. The fact that the correlation is calculated for paired pixels, which share the same spatial area, is the main advantage of the nMDP method. This useful automated tool reduces bias caused by the user and colocalized fluorescent signals are able to be quantified and imaged.
Automated Method to Quantify Colocalization Based on Spatial Statistics
The approach by Costes at al. (2004) solves the problem with the method by Manders et al. (1993) that is mentioned above, by accounting for the amount of correlation in different areas of a two-dimensional histogram for automatic estimation of the thresholds. They designed an algorithmic approach for measuring how much colocalization there is in two-color three-dimensional images. This procedure is based on spatial statistics. In the first step, this approach examines the probability (P value) that colocalization exists in a designated area of the image. In the next step, pixels that are colocalized in the area of interest are determined using a statistical criterion established from the two-dimensional histogram for both channels, enabling computation of the overall fraction of each protein being colocalized (i.e., colocalization coefficients). This algorithm has been commercialized by Bitplane AG (Zurich, Switzerland).
Fuzzy System Model
A limitation of quantifying colocalization is that there is no unified approach for interpreting results. After numerical values of colocalization coefficients are obtained, researchers often wish to define the degree of colocalization using subjective terminology, such as ‘‘weak'' and ‘‘strong.'' However, this terminology can vary in meaning between researchers. Therefore, a solution to this issue is required that relates the numerical values of colocalization coefficients to their qualitative estimation, while not being subject to personal prejudice.
Rivas-Perea et al. (2010) performed subjective colocalization analysis using fuzzy linguistics variables. They used a set of rules to map coefficient values to a linguistic interpretation. However, nonmatching and inconsistent variables were used for different coefficients, and no controls were included in their study. The term “fuzzy” in this context means “approximate” and not fixed and exact. The fuzzy system relates numerical values of colocalization coefficients to fuzzy propositions that use fuzzy values (e.g., weak and strong). Zinchuk et al. (2013) claim to have a more advantageous approach than that used by Rivas-Perea et al. (2010), using a model of a fuzzy linguistic system to interpret quantitative colocalization study results. They used five linguistic variables related to colocalization coefficient values ranging from “very weak” to “very strong.” Zinchuk et al. (2013) applied these variables to a range of coefficient values produced by means of images from computer simulation in which the degree of colocalization was precisely known. Therefore, these variables can be used with an accurate understanding of colocalization results. The details of this model are beyond the scope of this article. Ultimately, this method helps bridge the gap between quantitative and qualitative aspects of detecting colocalization.
The author thanks Monica Acosta for providing the confocal image with double labeling, and Junru Song and Birger Dieriks for providing the triple-immunostained images.