Microscope

The principles of colocalization microscopy and image analysis

Written by:

Elisa Opriessnig and Fanny Dobrenova

Colocalization microscopy and image analysis are some of the principal and most commonly used tools in modern biology to accurately localize molecules, proteins and other morphological objects within a cell. We dive deep into the exciting subject of colocalization studies to provide you with an overview of what this technique entails, why it is important and how you can use it for the benefit of your research.

Keep reading to find out how colocalization analysis will help you determine whether two or more biomolecules are affiliated with the same cellular components. Discover our leading-edge software solution Sparkfinder and how data obtained with it can be used in colocalization analysis.

What is the purpose of colocalization?

Many processes on a cellular level depend on the spatial arrangements of proteins and their interactions. Identifying central interacting components within cellular processes is an essential step toward understanding the underlying mechanisms that regulate certain cellular functions.

Fluorescence microscopy has proven to be a reliable technique for in vitro and in vivo studies of biomolecular interactions. Up to this day, a common approach for measuring the various  mechanisms of interaction on a cellular level using fluorescence microscopy is the analysis of dual color images for the presence of a colocalized signal.

In biomedical sciences, colocalization is commonly understood as the overlap between certain labels in images. The term usually refers to the quantitative evaluation of the immunostaining in fluorescence microscopy (Pastorek et al., 2016). 

Quantitative Colocalization analysis is a digital imaging technique that examines antigens of interest in immunofluorescence images by employing algorithms to estimate the degree of overlap of fluorescence signals (Zinchuk et al., 2007). 

In the context of fluorescence microscopy colocalization in cell imaging (colocalization image analysis) refers to the spatial overlap and correlation between two or more fluorescent labels or the overlap between cells, thus it can be regarded as the analysis of spatial variations/proximity (Tameling et al., 2021; Dunn et al., 2011).

The most popular areas of use of this analysis method include:  

  • cell biology
  • colocalization of proteins
  • protein interactions
  • colocalization of cancer cells (see definition of colocalization above)
  • colocalization analysis in genetics
  • colocalization of enzymes
  • neurotransmitter colocalization
  • localization to organelles/sub-cellular structures (i.e. nuclear localisation, lysosome colocalization)

Colocalization in the context of different imaging modalities

The phenomenon of colocalization can readily be examined using images acquired with the help of a confocal microscope (image A). Colocalization is typically observed by using a selection of two images representing red and green fluorescent channels. (image B) A third image is created where the two channels are combined and overlapping pixels are displayed in yellow (image B and C). 

Preparing a Colocalization Sample

A: Cells were processed and imaged by fluorescence microscopy. Also seen here are subcellular localization of Vps28A and Vps28B . Image taken from Commons.wikimedia.org.

The Colocalization Process

B: Colocalization can usually be observed by using a selection of two images representing red and green fluorescent channels.The yellow color in the corresponding merged images indicates colocalization between co-expressed proteins; white arrowheads also indicate obvious protein colocalization. Also seen here is the co-expression of Vps28A or Vps28B. Image taken from Commons.wikimedia.org.

C:  Images of the triple-transformed cell are shown here. The yellow color in the corresponding merged images indicates colocalization between co-expressed proteins; white arrowheads also indicate obvious protein colocalization. Image taken from Commons.wikimedia.org.

For a successful analysis, it is vital to exclude any artifacts generated by background and noise.  Typically the colocalization analysis of fluorescence microscopy images is assisted by software, which estimates the degree of overlap in fluorescent channels based on an AI algorithm (Zinchuk et al., 2007). 

In order to assess colocalization properly, background correction needs to be performed. This entails defining pixel intensity values for the ROI(s) to separate features of interest. (Zinchuk et al., 2007).  

Prior to conducting a colocalization analysis, it is important to select the correct image acquisition mode, because accurate estimation of colocalization strongly depends on the quality of the images (Zinchuk et al. 2007). Here is what you need to consider so that your image is well-suited for subsequent colocalization analysis. 

Preparing a colocalization sample

  • Choose fluorophores that have well-separated excitation and emission spectra.
  • Confirm specificity and the absence of cross-reaction of the used antibodies. 
  • Keep the same mounting medium for all your samples. 
  • Determine the level of autofluorescence by using unstained samples. 

Setting confocal microscope

  • Use optimized emission filters (i.e. to maximize emission collection). 
  • Use plain chromatic lenses to reduce chromatic shift. 
  • Consider a proper setup of the microscope pinhole size. 
  • Keep the same objective lens when observing samples.

Acquiring images

  • Avoid too bright and too high contrast images. 
  • Use only sequential scanning when acquiring images (to minimise the bleed-through-effect).
  • Choose TIFF graphics file format when using a confocal microscope for image acquisition (avoid loss of image data)

Colocalization analysis methods

We offer you a review of the most common colocalization analysis methods.

Standard microscopy equipment offers a variety of techniques  to split single fluorescence channels. Scientists use fluorescence colocalization to see if a molecule localizes to a certain cell compartment or if two or more molecules are likely to interact with each other. 

However, this often results in an overinterpretation, as a simple co-occurrence in the same region does not necessarily mean interaction. This is where i.e. advanced signal detection methods like proximity ligation assays come into play. The underlying principle is to have only one signal generated by two neighboring  and most likely interacting structures. However, this is not the ultimate solution for all scientific questions, hence colocalization analysis methods are more commonly used. 

Many scientists try to assess the color overlay of two different fluorescent markers. To give you an example of such a visual method, green and red fluorescence would result in yellow color in case of colocalization. 

Visual evaluations require comparable fluorescence intensities of the two markers and cannot be regarded  as a reliable quantitative estimation of colocalization. Fortunately, these issues have been dealt with rather early in the developing stages of fluorescence microscopy, which resulted in the concept of calculating various statistical parameters to be able to evaluate the correlation of fluorescence intensities of multiple channels pixel by pixel. The degree of colocalization is measured by comparing pixel positions throughout the analyzed images. We describe some of the commonly used visual and statistical methods to measure colocalization  (Moser et al., 2017; Manders et al., 1993).

Color analysis in the context of colocalization: visual methods

Prior to starting the process, you have to acquire the images on which the colocalization analysis will be performed by choosing the right settings of your microscope source. First, you do a visual inspection as part of a fast preliminary analysis. To quantify this later you use co-occurrence and correlation coefficients. 

To conduct color analysis in the context of colocalization, you need to determine the “color pair” that will serve as a visual input for comparison. Typically, you have to select a set of red and green fluorescent channel images. 

Next, a scatterplot of the individual pixels is generated using image viewing software. This means that colocalization analysis can be performed on a pixel-by-pixel basis. Each pixel in the image is plotted in a scatterplot graph based on different color intensities and intensity levels for each channel. The color map in the scatterplot represents the number of pixels that are plotted in a specific region and represents an  intensity graph (Cordelières & Bolte, 2014). 

Depending on the degree of colocalization, colocalized pixels appear as orange or yellow and are located toward the middle of the plot, meaning correlations between signals emitted by different colors are visually displayed. 

Visual colocalization methods offer the possibility to quickly gather some spatial information of the colocalization at a glance. It is an essential  part of qualitative colocalization analysis and can be divided into two techniques: Dye-overlay and Threshold-overlap. Even though both of these methods provide a useful visual representation of colocalization, they need quantitative analysis to avoid certain biases (Manders et al., 1993; Cordelières & Bolte, 2014). 

The visual analysis process

The visual analysis process. Image created with biorender.com.

After you have generated  the scatterplot diagram, you can perform a  quantitative evaluation using different coefficients and statistical methods. Keep reading to find out how.

Measures of colocalization in microscopy image analysis: statistical methods 

Colocalization measurements aim to characterize the relative distribution of two molecules within a relevant area. This is mostly done by measuring two distinct features such as co-occurrence and correlation.

Co-occurrence refers to the extent to which i.e. molecules appear together, while correlation means how well variations in the concentration of the two molecules match. 

Correlation refers to a statistical method to estimate a possible linear association between two continuous variables. It is used to calculate and interpret (Mukaka, 2012). 

Calculations of the Pearson correlation coefficient (PCC) and Mander’s overlap coefficient (MOC) are supported by the majority of colocalization software packages (Adler & Parmryd 2021). 

The PCC and the MOC are mainly used to quantify the degree of colocalization between fluorophores, however, MOC was introduced to resolve some challenges associated with the use of  PCC like the lack of sensitivity to differences in the mean signal intensities or ranges. 

Both of these colocalization coefficients are mathematically very similar, but differ in their application. MOC relies on absolute intensities, while PCC uses the deviation from the mean. MOC can be considered a hybrid colocalization measurement as correlation is combined with co-occurrence. Further, it prioritizes high-intensity combinations and ignores blank pixels. In the case of PCC only correlation is measured (Adler & Parmryd, 2010). 

Pearson correlation coefficient (PCC) interpretation: PPC values range from -1 to 1. A value of  +1 stands for a perfect correlation, one of 0 signifies no correlation, while represents -1 an ideal anti-correlation  (imageJ.net). 

Manders overlap coefficient (MOC) interpretation: Values typically range from 0 to 1. They express the fraction of intensity in a channel that is located in pixels where there is above zero intensity in the other color channel (Manders et al., 1993). 

Colocalization coefficients m1 and m2: These coefficients represent the contribution of each one of two selected pixels of interest. The value of 1.0 for both channels means perfect colocalization (Zinchuk et al., 2007). 

Colocalization coefficients M1 and M2: These are identical to m1 and m2, but are used when analyzing a scattergram Region of Interest (ROI) (Zinchuk et al., 2007). 

Overlap coefficients k1 and k2: These coefficients strongly depend on the sum of the products of the intensities of the two channels. They have a high sensitivity when it comes to differences in signal intensities (Zinchuk et al., 2007). 

You can find a brief comparison of coefficients used to estimate colocalization (adapted from Zinchuk et al., 2007) in the Table below.

Coefficient

Description

Range of Values

Use

Pearsons's Correlation Coefficient (PCC)

Describes the correlation of the intensity distribution between channels.

-1.0 to 1.0; 0 means no significant correlation, -1 refers to a complete negative correlation.

Applicable in any colocalization experiment.

Mander's Overlap Coefficient (MOV)

Indicates an overlap of the signals and shows the true degree of correlation.

0 to 1.0; 0.5 indicates that 50% of both selected channels colocalize.

Applicable in any colocalization experiment, and when the fluorescence of one antigen is stronger than that of the other.

m1 and m2 coefficients

Describe the contribution of each one from two selected channels to the pixels of interest.

0 to 1.0; m1 and m2 of 1.0 and 0.2 for the red-green pair mean that all red pixels colocalize with green, but only 20% of green pixels colocalize with red.

Applicable in any colocalization experiment.

M1 and M2 coefficients

= m1 and m2, but applied to analyze a scattergram Region of Interest (ROI) 

0 to 1.0; m1 and m2 of 1.0 and 0.2 for the red-green pair means that all red pixels colocalize with green, but only 20% of green pixels colocalize with red.

Applicable in any colocalization experiment. 

Overlap Coefficients k1 and k2

Split the value of colocalization into two separate parameters; they allow to identify the contribution of each antigen to the area with colocalization. 

Vary. 

Applicable in any colocalization experiment. 

Table: Comparison of coefficients used to estimate colocalization (adapted from Zinchuk et al., 2007).

Tips and tricks: Choose the colocalization metric you want to apply based on the staining of your input images. 

What coefficient you need to use depends on the images you want to examine. The coefficients apply different approaches to evaluate colocalization and have different sensitivity and applicability. In general, Pearson’s coefficient is used in the majority of cases, but when an antigen is stained stronger, then the Manders overlap coefficient should be employed, as it gives more reliable results for such images (Zinchuk  et al., 2007).

Colocalization software

There are different types of colocalization software available. An example of that is the open-source image analysis software ImageJ with its life science panel Fiji. However, these software packages have to be extended with specialized colocalization finder plugins to be able to perform colocalization analysis.

ImageJ

ImageJ is a Java-based image processing program, which contains a list of colocalization analysis plugins depending on method type (imagej.net): 

One-to-one-pixel matching analysis: 

Spatial cross-correlation analyses: 

Temporal cross-correlation analyses: 

Object-based analyses: 

Fiji

Fiji is the bioimage analysis package of ImageJ. It can be extended with a lot of plugins that facilitate scientific image analysis tasks and is suitable for both ends of the spectrum – users and developers (Schindelin et al., 2012). 

Fiji contains a variety of add-on plugins, which can be installed, including one for colocalization analysis, named Coloc2. It calculates a number of colocalization parameters such as Pearson coefficient, Mander correlation, Spearman’s rank correlation and so on. However, the plugin does not perform object-based colocalization measurements (imagej.net; Li et al., 2004).

JACoP

JACoP is short for Just Another Co-localization Plugin and is a compilation of colocalization tools free for use. This plug-in will, depending on the ticked boxes, determine colocalization on two images according to the selected methods (i.e. Pearson’s Coefficient, Manders Overlap Coefficient, etc.). All these methods are further implemented to work on a 3D dataset. JACoP offers different features like calculating colocalization indicators, generating visualizations (Cytofluorogram) and/or performing methods like Coste’s automatic threshold, Li’s ICA, and Costes’ randomization (imagej.net; Bolte & Cordelieres, 2006).

Colocalization analysis example: the case of Sparkfinder

Sparkfinder has specifically been designed to help you segment cell nuclei in multichannel images, while also obtaining meaningful data about spatial arrangements of the (tissue) sample.

Measurement of channel intensities

Measurement of channel intensities in multiplexed images, including an optional segmentation with the help of Sparkfinder in IKOSA.

Sparkfinder supports multichannel fluorescence images and will help you gather information on  parameters associated with morphological objects like cell nuclei and virtual cytoplasm including: 

  • object type,
  • a total number of objects,
  • total area covered,
  • objects density,
  • mean object perimeter,
  • mean object circularity,
  • mean object solidity and
  • mean object eccentricity.

With the help of Sparkfinder, you can gather data on the spatial relationships between cellular components by means of instance segmentation. Instance segmentation is an analysis technique highly sensitive to spatial information due to the fact that object coordinates are already included per each identified object, e.g. nucleus, which allows the measurement of distance and other parameters relevant for subsequent statistical analyses. 

The Sparkfinder app available on the IKOSA Platform allows you to manually chose the channels of interest and helps you gather valuable insight into the intensity range of individual color channels of multiplexed fluorescent images.

To accommodate spatially-oriented analysis Sparkfinder supports the measurement of a number of spatial metrics like:

  • mean distance between detected objects,
  • mean distance between the detected objects and standard deviation of distances between detected objects, 
  • nearest neighbor analysis, 
  • t-SNE plotting,
  • PCA.

To top it all off, this cutting-edge application will help you gather valuable insights into the intensity range of individual color channels of multiplexed fluorescent images. In the results output, you will find quantitative information on the mean distribution and standard deviation of the intensity levels associated with single fluorescent channels and individual object labels.

All you need to do is download the CSV files with the Sparkfinder analysis outputs. Transform and into the desired format and upload it to the statistical software of your choice i.e. SPSS, Matlab or R. You can now run different statistical tests  on the gathered intensity data and calculate correlation coefficients like the Pearson Correlation coefficient or the Manders overlap coefficient. 

Sign up for our newsletter in order not to miss out on any of our feature articles on the latest developments in biomedical image analysis.

Use our IKOSA platform to generate the intensity data needed for subsequent colocalization analysis.

Our Sparkfinder app offers you everything you need to do your own, customized colocalization analysis. Extract valuable spatial data with the help of AI technology and speed up discovery. 

Sign up for our newsletter in order not to miss out on any of our feature articles on the latest developments in biomedical image analysis.

See references