ACCEPTED: Meth. Mol. Biol., Protein‐Ligand Interactions: Methods and Applications, Second Edition, Mark A. Williams, Tina Daviter, eds. 

Quantitative Fluorescence Co-localization of Lipoproteins with the LDL Receptor

Shanica N. Pompey, Peter Michaely and Katherine Luby-Phelps*

Department of Cell Biology, UT Southwestern Medical School, Dallas, TX 75390-9039 *Corresponding author: kate.phelps@utsouthwestern.edu

Running Head: Fluorescence Co-localization

Keywords: fluorescence, co-localization, confocal, deconvolution, immunofluorescence, digital

imaging, LDL, VLDL, LDL receptor

Abstract

Fluorescence microscopy can be used to assess quantitatively the interaction between a ligand

and its receptor, between two macromolecules, or between a macromolecule and a particular

intracellular compartment by co-localization analysis. In general, this analysis involves tagging

potential interacting partners with distinct fluorophores-- by direct labeling of a small ligand, by

expression of fluorescent cDNA constructs, by immunofluorescence labeling or by some

combination of these methods. Pairwise comparison of the fluorescence intensity of the two

fluorophores at each pixel in a two channel digital image of the sample reveals regions where

both are present. With appropriate protocols, the image data can be interpreted to indicate where

the potential interacting partners are co-localized. Keeping in mind the limited resolution of the

light microscope, co-localization is often used to support the claim that two molecules are

interacting.

All quantitative methods for evaluating co-localization begin with identifying the pixels where

the intensities of both color channels are above background. Typically this involves two

sequential image segmentation steps: the first to exclude pixels where neither channel is above

background, and the second to set overlap thresholds that exclude pixels where only one color

channel is present. Following segmentation, various quantitative measures can be computed to

describe the remaining subset of pixels where the two color channels overlap. These metrics

range from simple calculation of the fraction of pixels where overlap occurs to more

sophisticated image correlation metrics. Additional constraints may be employed to distinguish

true co-localization from random overlap. Finally, an image map showing only the co-localized

pixels may be displayed as an additional image channel in order to visualize the spatial

distribution of co-localized pixels. Several commercial and open source software solutions

provide this type of co-localization analysis, making image segmentation and calculation of

metrics relatively straightforward. As an example, we provide a protocol for the time-dependent

co-localization of fluorescently tagged lipoproteins with LDL receptor (LDLR) and with the

early endosome marker EEA1.

1. Introduction

1.1. Fluorescence microscopy

Over the past two decades, technical advances have led to a revival of light microscopy as an

important tool for biomedical research. Development of sensitive, high resolution cameras with a

linear light response, and the availability of stable fluorescent tags with emission maxima

spanning the visible spectrum from near UV to far red, make it possible to visualize specific

molecules in a microscope specimen. Improved optical methods such as confocal, multiphoton

and deconvolution microscopy that decrease the contribution of signal from above and below the

focal plane permit the location of fluorescently tagged molecules with submicrometer accuracy

in three dimensions. Simultaneous visualization of two molecules whose fluorescent tags have

distinct emission spectra allows the investigation of molecular associations in situ, commonly

referred to as co-localization.

1.2. Image Analysis for fluorescence co-localization

The simplest and most commonly employed method of determining co-localization is to display

the image of one fluorophore in the red channel of an RGB image, display the image of the other

in the green channel, and look for yellow pixels in the merged image. While this approach yields

a qualitative assessment of co-localization, it often provides a poor estimation of the degree of

co-localization. Because the appearance of yellow depends on the relative intensities of red and

green, co-localization may be underestimated when there is wide disparity in the intensities of

the two fluorophores. Such qualitative assessments can also overestimate co-localization,

particularly in regions with dense populations of discrete structures that cannot be spatially

resolved in the fluorescence microscope. In these regions yellow color may give a false

impression of co-localization where none exists. For this reason, apparent co-localization in the

Golgi region of tissue culture cells should be treated with suspicion. Fortunately, because the

fluorescence intensity is a function of the number of fluorophores present, a more objective

evaluation of co-localization is possible.

Common quantitative measures of co-localization

For convenience, this discussion will refer to the images of the two potentially co-localized

species as the red and green channel, even though the original fluorescence tags may be any two

colors, and in any case only monochrome images should be used for co-localization images.

Finally, although we use the two-dimensional term, pixel, rather than the three-dimensional term,

voxel, to refer to the elements of the intensity array that comprises a digital image, the discussion

applies equally to both two- and three-dimensional datasets.

Cytofluorogram – Similar to how cells with different fluorescent tags are displayed in flow

cytometry, the intensity of the red and green signals at each pixel in an image may be plotted as a

two dimensional histogram, referred to as a cytofluorogram or scatter plot. In the absence of co-

localization, the scatter plot will be bi-lobed as shown (Fig 1A). Pixels with only green will be

spread out along the green intensity axis and pixels with only red will be spread out along the red

intensity axis. The center of the plot is empty because no pixels have both red and green. In the

ideal case of stoichiometric co-localization, the scatterplot will be a straight line (Fig 1B)

because all pixels have both red and green at a fixed intensity ratio. The slope of this line gives

the ratio of intensities in the two color channels, which may reflect the stoichiometry of the

molecular association between the two species. Multiple discrete stoichiometries are sometimes

apparent as distinct lobes of differing slope in the scatterplot. A tri-lobed scatterplot (Fig 1C)

indicates a mixture of co-localization with no co-localization. By choosing threshold values to

exclude pixels where only one of the two channels is present, (indicated by the solid black lines

in Fig 1C) the pixels where the two color channels overlap can be examined selectively to extract

quantitative information regarding the degree of overlap and the spatial pattern of overlap. In

practice, the actual scatter plot often resembles a cloud of pixels with no clear delineation

between co-localized and non-co-localized pixels (Fig 1D). This may arise from either co-

localization with no fixed stoichiometry or the effect of noise in the image or both. Objective

methods of obtaining the overlap threshold under these circumstances will be discussed later.

Fraction of Overlapping Pixels. Once thresholds have been set to exclude pixels lacking

significant overlap, the extent of overlap can be computed for each channel. This is usually

expressed as the fraction of pixels in each channel that overlap with signal from the other

channel, that is, the number of pixels above the overlap threshold normalized by the total number

of pixels in that channel. For example, a value of 0.5 for the red channel indicates 50% of the

pixels in the red channel also have green signal above the overlap threshold set for the green

channel. The fractional overlap for the two channels will not necessarily be the same. If the

binding partner in the red channel is widely distributed, while the binding partner in the green

channel is only present when red is present, the % overlap for the red channel will be low, while

that for the green channel will be very high.

Thresholded Manders’ Coefficients (M1 and M2). The variations of pixel intensities across a

fluorescence microscope image directly indicate the relative number concentration of the

fluorescent tag. With appropriate calibration standards, the fluorescence intensity at any given

pixel can be interpreted in terms of the absolute number concentration of the tagged molecule,

although this is seldom done. An intensity-based co-localization metric known as the Manders’

coefficient incorporates the relationship between intensity and concentration and thus provides

more meaningful biological information than the fraction of overlapping pixels for characterizing

the co-incidence of the red and green channels. The thresholded Manders’ Coefficient for each

channel (designated M1 and M2) is effectively the sum of the fluorescence intensities of the

overlapping pixels normalized by the sum of the intensities of all the pixels (1). As for the pixel

overlap coefficient, the values for the two channels will not necessarily be the same.

Making the assumption that the fluorescence intensities at each pixel are proportional to the

concentration of each species, the Manders’ coefficients can be interpreted as the fraction of

molecules of one species that overlap with the other. These values may differ dramatically from

the fraction of overlapping pixels – a relatively small fraction of very bright pixels may result in

a relatively high value for Manders’ coefficient.

Pearson’s Coefficient. The fraction of overlapping pixels is often referred to as % co-localization

but it does not distinguish between random overlap and true co-localization. A metric commonly

applied to identify truly co-localized pixels is the Pearson Coefficient (PC). This is a standard

statistical measure widely used in image correlation analysis and introduced to fluorescence

microscopy by Manders et al. (2). The PC tests whether there is a linear correlation between the

intensity of the red signal and the intensity of the green signal at every pixel. The PC may be

computed over all pixels in the image or in user specified regions of interest. It ranges from -1 to

1: a PC of 0 indicates random overlap, and a PC of 1 indicates perfect co-localization with a

constant stoichiometry of red and green. A PC of -1 suggests a mutually exclusive localization of

the two signals. It should be noted that in cases where the intensities of the two channels are not

related by a constant ratio, a low positive value of the PC does not necessarily indicate poor co-

localization, it may merely indicate that the intensities of the two color channels are not strongly

correlated.

Interpretation of PC. The PC is often used to compare the co-localization of two different pairs

of binding partners: for example, the binding of a ligand with a wild type protein vs. a mutant

lacking a functional binding domain. Although in principle a PC > 0 indicates co-localization,

significant noise in the images will make it difficult to determine at what value the PC becomes

significantly different from 0. A method developed by Costes (3) provides an objective

evaluation of PC significance. In this approach, the pixels in one channel are scrambled and the

PC is calculated. If 95 out of 100 randomized images give a PC less than the PC calculated for

the original image, the PC for the original image is considered significant of co-localization with

95% confidence. An alternative use of the data from randomized images is to plot the

distribution of PCs for all the randomized images and obtain the standard deviation from a

Gaussian fit to the data. A PC for the original image several fold higher than the standard

deviation for the random images is likely to indicate significant co-localization. For high quality

data, a PC as low as 0.1 may easily be as much as 50-fold higher than the standard deviation of

the distribution of PCs for randomized images.

Co-localization Threshold Selection – A method proposed by Costes et al (3) uses the PC to

determine automatic and relatively objective intensity thresholds for excluding pixels with no co-

localization. This is especially useful in cases like Fig 1D where the scatterplot does not clearly

indicate where the thresholds should be drawn. Beginning with the highest intensity values for

one channel, the Costes method progressively lowers the threshold values for each channel until

the PC for pixels below threshold in both channels reaches zero. At this point, the pixels below

threshold in either channel are considered not co-localized and are excluded from the calculation

of PC. The PC for the remaining, co-localized pixels can be dramatically better than for the

dataset overall in cases where true co-localization occurs in only a few regions of the image. One

caveat of the Costes method is that because PC assumes a linear relationship between red and

green intensities, the threshold values for the red and green channels are not set independently.

An additional issue is that dim pixels with true co-localization may be excluded by this method.

Other Measures of Co-localization --Additional measures of co-localization based on pixel

intensity correlations have been proposed and are available in some software packages. Two

common examples are van Steensel’s Cross Correlation Function (4) and Li’s Intensity

Correlation Analysis (5). A more detailed discussion of these and other intensity based co-

localization methods is available in two excellent reviews (6, 7).

Object based methods --An entirely different approach to co-localization analysis is to build

isosurface objects in each color channel and then calculate distance maps between the objects in

one channel and the objects in the other. The results can be used to find the nearest green

neighbor for each red object or to find all red objects within a certain distance of green objects.

This approach is very useful for co-localization when the color channels overlap only partially or

not at all – for example, when the two fluorescent tags are on opposite sides of the limiting

membrane of the same vesicle. Strategies for interpreting the results of this kind of analysis are

discussed in (6).

1.3. Limitations of Fluorescence Co-localization Analysis

None of the co-localization methods described above is completely unambiguous, especially in

the presence of noise and/or when the stoichiometry of red to green is variable. Automated

determination of thresholds may fail to find appropriate thresholds of co-localization or the

thresholds may be set too low, resulting in overestimation of the degree of co-localization. The

limitations of co-localization analysis have been rigorously investigated and are discussed in (1-

7). Development of more reliable co-localization metrics remains an active field of research e.g.

(8, 9).

Even under ideal conditions, the limited spatial resolution of fluorescence microscopy dictates

that fluorescence co-localization should never be regarded as definitive evidence of an

interaction between two species. Co-localization data should always be supported by genetic or

biochemical evidence of interaction, such as yeast two-hybrid, co-immunoprecipitation or

crosslinking studies. For standard fluorescence microscopy, the diffraction theory of image

formation according to Abbé and the Nyquist sampling theorem dictate that resolution is limited

by the wavelength of the light and the calibrated pixel size in the image. The latter depends on

the numerical aperture of the objective lens (NA), the image magnification, and the physical

dimensions of pixels on the camera chip. For visible light, the highest NA lenses, and cameras

with the smallest physical pixels, the resolution is limited to around 0.2 µm in xy and

approximately 0.4 µm along the z axis. Thus, each voxel has a volume ≥ 0.016 µm3. To put this

in perspective, each voxel is large enough to contain 100 synaptic vesicles. Other microscopic

methods with better resolution such as electron microscopy and Förster Resonance Energy

Transfer (FRET) allow more confident determination of actual interactions between two species.

Emerging super-resolution technologies, such as STED, STORM, PALM or SIM (10), may also

lead to improved resolution of co-localization.

1.4. Software for Co-localization

Most available software packages for analysis of fluorescence microscope images include some

form of co-localization analysis. Four commercial packages with very sophisticated co-

localization capability are Huygens (SVI, Inc), Imaris (Bitplane), Autoquant X (Media

Cybernetics) and Volocity (Perkin Elmer). In addition, co-localization plugins are available for

the open source java application ImageJ (http://rsbweb.nih.gov/ij/) and its various distributions,

including Fiji (http://pacific.mpi-cbg.de/wiki/index.php/Fiji). Two plugins that implement all the

co-localization methods described above are JaCoP (ImageJ) and Coloc_2 (Fiji). Different

implementations of the same co-localization algorithms by these different packages can yield

somewhat different values of PC and other metrics (Table I). In general, this seems to reflect

different criteria used for excluding background pixels and for setting intensity thresholds. The

documentation for all the co-localization analysis packages listed above is available online (12)

and offers a very useful resource for understanding the differences in how various features are

implemented.

1.5. The LDLR-EEA1 interaction as an example:

Defects in uptake of serum lipoproteins such as LDL and VLDL result in hyperlipidemia

disorders implicated in heart disease. Although it is well known that uptake and metabolism of

these lipoproteins depends on the LDL receptor (LDLR), the detailed regulation of uptake at the

cellular level remains an area of active investigation. Time-dependent co-localization analysis

can be used to map out the progress through the uptake pathway under different experimental

conditions. Here we present results showing co-localization of the LDL with LDLR at the

plasma membrane of the cell before uptake begins and the time course of association with the

early endosomal compartment as the LDL is internalized by endocytosis.

2. Materials:

1. Medium: Dulbecco’s MEM supplemented with 10% (v/v) fetal bovine serum, 20 mM

HEPES pH 7.5, penicillin G (100 units/ml) and streptomycin (100 µg/ml)

2. Cells grown in MEM on 12 mm round #1.5 (0.17mm) coverslips in 24-well microtiter

plates

3. Fluorescently labeled LDL and/or VLDL (11)

4. Delipidated medium: Ice cold medium with 10% (v/v) delipidated fetal bovine serum

substituted for the fetal bovine serum

5. Ice cold PBS

6. 3% paraformaldehye in PBS, chilled to 4°C

7. Digitonin (Sigma Cat# D141)

8. 5% Normal Goat Serum+1% BSA in phosphate buffered saline (PBS) (for blocking)

9. 5% Normal Goat Serum+0.1% BSA in PBS (for antibody dilution)

10. Primary/secondary antibodies (see Note 1): Rabbit polyclonal antibody to bovine LDLR

was a gift of Joachim Herz; mouse monoclonal antibody to EEA1 was from BD

Biosciences, (Cat# 610457); goat anti-rabbit Alexa 647 was from Invitrogen (Cat#

A21236)

11. Fluorescent nuclear stain: DAPI (Invitrogen, Cat# D3571)

12. Aquamount (Polysciences Inc., Cat# 18606)

13. Digital fluorescence imaging system (widefield or confocal)

14. ImageJ with JaCoP or Fiji with Coloc_2 or other image analysis software

3. Methods:

3.1. Overview of procedure

In general the protocol can be divided into three major steps: sample preparation, image

acquisition and image analysis. Sample preparation includes both the pulse chase labeling of live

cells with fluorescent lipoprotein and immunofluorescence labeling of other markers in cells

after uptake is stopped by fixation. Meaningful co-localization analysis requires careful attention

to critical details at each step as outlined in the Notes.

3.2 Pulse Chase Uptake of Lipoprotein (see Note 2)

1. Remove cells from 37°C incubator and replace MEM with ice cold MEM+FLPPS.

2. Chill cells, 10 min on ice to block endocytosis.

3. Dilute lipoprotein to 10 µg/ml in ice cold MEM+FLPPS.

4. Add lipoprotein solution to cells and place in 4°C cold room, 1.5 hrs on ice to allow

lipoprotein binding to receptors.

5. Rinse cells with ice cold MEM+FLPPS, 1x quickly.

6. Add warm MEM and immediately place in 37°C incubator for desired chase time. The

selection of time points will be determined by which stage of uptake is of interest.

3.3 Immunofluorescence Staining (see Notes 1, 3, 4)

1. At each time point, remove a dish of cells from the incubator, and quickly rinse 3x with ice

cold PBS to stop uptake.

2. Fix cells with 3% PFA in ice cold PBS, 20 min on ice.

3. Rinse with ice cold PBS, 2x quickly.

4. Permeablize cells with digitonin (10 µg/mL in PBS), 2 min on ice.

5. Rinse with ice cold PBS, 3x quickly.

6. Block cells with 5% Normal Goat Serum+1% BSA in PBS, 1hr at room temperature (500 µl

per well).

7. Add primary antibody diluted 1:100 in 5% Normal Goat Serum+0.1% BSA in PBS and

incubate 1 hr at room temperature (150-200 µl of antibody solution, dropwise per well) (see

Note 3).

8. Rinse with 0.1% BSA in PBS, 3x quickly or move coverslips to fresh wells with 0.1% BSA

in PBS.

9. Incubate with secondary antibody diluted 1:400 in 5% Normal Goat Serum+0.1% BSA in

PBS, 1 hr at room temperature (200 µl of antibody solution, dropwise per well) (see Note 3).

10. Rinse with 0.1% BSA in PBS, 1x quickly.

11. Wash with 0.1% BSA in PBS, 3x 5 min each.

12. Rinse with PBS, 1x quickly.

13. Incubate with DAPI in PBS (3 µl/mL) 10 min at room temperature (500 µl solution per

well).

14. Rinse with PBS, 3x quickly.

15. Add 10 µl Aquamount to a glass slide. With fine tweezers, remove coverslip from well and

invert on drop to mount (see Note 4).

16. Verify the specificity of your labeling (see Note 1).

17. Minimize background fluorescence on the coverslip (see Note 3).

3.4 Image Acquisition

1. Optimize imaging parameters (see Notes 5-9). Once imaging parameters have been

optimized, these conditions should be used for all samples that will be compared: for

example, comparison of control with experiment, wildtype with mutant, or monitoring

changes in co-localization over time (see Note 5).

2. Acquire wide-field z stacks or confocal slices (see Note 10).

3. Register the two color channels in x,y and z if necessary (see Note 11).

4. Deconvolve wide-field z stacks (see Note 12) (Fig 2).

3.5 Image Analysis

1. Choose an intensity threshold that excludes zero-zero pixels where no true signal is present in

either channel (see Note 13).

2. Apply Costes method of automated overlap threshold selection (Fig 3A).

3. Calculate the Pearson Coefficient for the pixels above the overlap threshold (see Table I)

(see Note 14).

4. Calculate the Manders coefficients for the pixels above the overlap threshold in each color

channel (see Table I).

5. Apply Costes randomization to check the significance of the Pearson Coefficient (Fig 3B)

6. Build colocalization channel (Fig 4).

Data obtained from different images is compared on the basis of the PC (see Note 5), e.g. a

change in PC over time indicates a change in co-localisation over time, which may be interpreted

judiciously to indicate association or disassociation of ligand and receptor. For example, Fig. 5

indicates colocalisation of LDL with the early endosome marker EEA1 peaks 10 min after

enodcytosis is started by warming the cells to 37 °C, consistent with the well established

pathway for uptake of LDL by endocytosis.

4. Notes

1. For immunofluorescence localization, the specificity of the antibodies/ligands should be

demonstrated on samples lacking the antigen, for example by gene silencing or using

tissue/cells from a knockout animal. Similar controls should be performed for fluorescently

tagged ligands. For expression tags, it is best to demonstrate that the tagged proteins are

functional and that their localization reflects the localization of the endogenous protein.

2. For optimal pulse chase experiments it is essential that the cells be kept on ice at 4°C at all

times before starting the chase, and also from the time when the chase is stopped until the

fixation is completed. If necessary, the results can be improved by adding salt to the ice.

3. Spinning both primary and secondary antibodies for 5 min at 4°C in a microfuge at high

speed before applying to the cells is recommended to prevent deposition of fluorescent

aggregates on the sample.

4. To minimize air bubbles in the sample, use only 10 µl of mounting medium and lever the

coverslip onto the drop of mounting medium slowly from the side opposite the tweezers,

allowing air bubbles to escape from the side where the tweezers grip the coverslip

5. Single cell image analysis is subject to high variability due to cellular heterogeneity. As a

result, the value of PC may vary widely from cell to cell. However, with high quality data

and sufficient sample size, quantitative comparison of co-localization under different

experimental conditions is possible as long as the image acquisition and processing

parameters are identical for all datasets (Fig 6).

6. Minimize imaging noise. Keep the gain of the detector as low as possible. Select excitation

intensities and exposure times to get the widest possible range of signal intensities without

saturating the detector. For best results, it is not advisable to use an offset on the bottom end

of the intensity range.

7. Avoid detector saturation (intensity clipping). Quantification of fluorescence intensities

assumes there is a linear relationship between the number of photons reaching the detector

and the electronic output of the detector. However, the light response of all detectors has a

maximum. Light intensities above this maximum will result in no further increase in

electronic signal from the detector. This is referred to as saturation or intensity clipping.

Extreme levels of saturation can also lead to spillover of signal from one pixel to another,

effectively increasing the size of objects in the image.

8. Verify absence of crosstalk between channels. Image a red-only and a green-only sample in

both channels using the imaging conditions you have established. Signal from the red-only

sample should not appear in the green channel and vice versa. If necessary, the imaging

conditions for each channel can be adjusted to minimize crosstalk.

9. Use a monochrome camera with 12-bit or 14-bit depth. Use of a color camera can introduce

channel bleedthrough depending on the particular technology it uses to acquire RGB images.

In addition, the RGB images acquired by color cameras are usually only 8-bits in each color

channel, limiting the range of pixel intensities to 256. Research grade monochrome CCD

cameras and photomultiplier tubes are inherently grayscale devices, and most support 12- or

14-bit images, increasing the range of possible intensities to 4096 or 16, 384 gray levels,

respectively. The image channels can be digitally pseudocolored red or green post-

acquisition if desired.

10. Be sure to save files in an uncompressed or lossless compressed file format. Usually, the best

choice is a 16-bit TIFF file. Be sure to choose an option that saves the raw intensity values.

The raw intensities will not be preserved when saving a 12- or 14-bit image as an 8-bit file,

saving the images in JPEG format, or saving a screen shot of what is displayed on the

monitor.

11. Some imaging systems may introduce a pixel shift between color channels that must be

corrected to ensure high quality results of co-localization analysis. In particular, objective

lenses that are not corrected for chromatic aberration across the entire spectrum of available

fluorophores can produce very large shifts in the z direction, especially for UV or far-red

emitting fluorophores. Because such shifts generally are systematic, once pixel shift has been

characterized for a set of imaging conditions, subsequent images can be aligned using image

analysis software. Z-stacks of beads that are fluorescent in multiple color channels, such as

Tetraspeck fluorescent microspheres (Invitrogen, Cat # T14792), can be used to determine

the degree of shift in x, y and z.

12. Co-localization analysis for a pair of wide-field epifluorescence images may give a false

impression of co-localization because each pixel contains signal from above and below the

plane of focus. In that case, true co-localization cannot be distinguished from coincidental

overlap of objects located in different focal planes. This limitation becomes increasingly

important as the thickness of the sample increases. For this reason, co-localization should

only be computed for pairs of confocal slices, confocal z stacks or wide-field epifluorescence

z stacks that have first been deconvolved to restore out of focus photons to their slice of

origin. Even confocal z stacks may benefit from deconvolution. Total Internal Reflection

Fluorescence (TIRF) microscopy images have better depth discrimination than any of these

methods and are also suitable for co-localization analysis.

13. Exclusion of background pixels may be accomplished by intensity thresholding, by drawing

a region of interest on one of the images or by masking the image pair with a third color

channel. For example, DAPI staining is often used as a mask to restrict the analysis to pixels

within the nuclear region. In Figure 4, the red channel was used to mask the dataset.

14. Table I illustrates how different implementations of the same co-localization algorithms may

give somewhat different results. One source of differences is that Colocalization Threshold

and JaCoP both require that the original 32-bit images be converted to 16-bit images, which

changed the intensity values in the images, affecting the results of the calculations. The

available documentation for each method (12) does not reveal why the thresholded Manders’

coefficients generated by Imaris Coloc are much lower than the others in Table 1. Some

algorithms report non-zero values for the PC of the pixels below threshold, suggesting they

have reached a local minimum that does not necessarily produce the optimal thresholds. For

example, according to its source code, Coloc 2 reports the best thresholds after 30 iterations,

rather than the threshold where PC becomes zero. In contrast, both Imaris Coloc and JaCoP

(which takes a lot longer to arrive at a solution) actually require PC ≤ 0 for the pixels below

threshold. In the case of Imaris Coloc, failure to reach this value within the maximum

number of iterations results in a failure to find automatic thresholds. While the

implementation of co-localization in open source software is very useful and more affordable

than a commercial package like Imaris, the user should be aware that code for ImageJ and

Fiji is under continuous development and updates may include changes and bugs in the way

algorithms are implemented.

REFERENCES CITED

1. Manders, E. M., Stap, J., Brakenhoff, G. J., van Driel, R., and Aten, J. A. (1992)

Dynamics of three-dimensional replication patterns during the S-phase, analysed by

double labelling of DNA and confocal microscopy, J Cell Sci 103 ( Pt 3), 857-862.

2. Manders, E. M. M., Verbeek, F. J., and Aten, J. A. (1993) Measurement of co-

localization of objects in dual-colour confocal images, Journal of Microscopy 169, 375-

382.

3. Costes, S. V., Daelemans, D., Cho, E. H., Dobbin, Z., Pavlakis, G., and Lockett, S.

(2004) Automatic and Quantitative Measurement of Protein-Protein Colocalization in

Live Cells, Biophysical journal 86, 3993-4003.

4. van Steensel, B., van Binnendijk, E. P., Hornsby, C. D., van der Voort, H. T., Krozowski,

Z. S., de Kloet, E. R., and van Driel, R. (1996) Partial colocalization of glucocorticoid

and mineralocorticoid receptors in discrete compartments in nuclei of rat hippocampus

neurons, J Cell Sci 109 ( Pt 4), 787-792.

5. Li, Q., Lau, A., Morris, T. J., Guo, L., Fordyce, C. B., and Stanley, E. F. (2004) A

syntaxin 1, Galpha(o), and N-type calcium channel complex at a presynaptic nerve

terminal: analysis by quantitative immunocolocalization, J Neurosci 24, 4070-4081.

6. Bolte, S., and Cordelieres, F. P. (2006) A guided tour into subcellular colocalization

analysis in light microscopy, J Microsc 224, 213-232.

7. Comeau, J. W., Costantino, S., and Wiseman, P. W. (2006) A guide to accurate

fluorescence microscopy colocalization measurements, Biophys J 91, 4611-4622.

8. Adler, J., Pagakis, S. N., and Parmryd, I. (2008) Replicate-based noise corrected

correlation for accurate measurements of colocalization, Journal of Microscopy 230, 121-

133.

9. RamÍRez, O., GarcÍA, A., Rojas, R., Couve, A., and HÄRtel, S. Confined displacement

algorithm determines true and random colocalization in fluorescence microscopy, Journal

of Microscopy 239, 173-183.

10. Toomre, D., and Bewersdorf, J. A New Wave of Cellular Imaging, Annual Review of

Cell and Developmental Biology 26, 285-314.

11. Zhao, Z., and Michaely, P. (2009) The Role of Calcium in Lipoprotein Release by the

Low-Density Lipoprotein Receptor, Biochemistry 48, 7313-7324.

12. Image documentation for JaCoP can be found here:

http://imagejdocu.tudor.lu/lib/exe/fetch.php?media=plugin:analysis:jacop_2.0:just_anoth

er_colocalization_plugin:jacop_ijconf2008.pdf

Documentation and a link to the source code for Coloc 2, Colocalization Threshold and

Colocalization Test can be found here:

http://pacific.mpicbg.de/wiki/index.php/Colocalization_Analysis

Figure Captions

Figure 1. Simulated cytofluorograms or scatterplots showing different cases of overlap between

two color channels. A. Only random overlap between red and green signal. B. Strong overlap

between the red and green channels. C. Mixture of overlapping with non-overlapping signal. D.

Actual data from an image with partial overlap, variable stoichiometry and random noise.

Scatterplots were generated in ImageJ using the JaCoP plugin and the data were re-plotted in

Prism (GraphPad).

Figure 2. Maximum intensity projections of a 15-slice wide-field epifluorescence z stack

showing LDL (red) and LDLR (green) before (A) and after (B) deconvolution. Image stacks

were deconvolved using the blind deconvolution algorithm with the default settings in Autoquant

X (Media Cybernetics).

Figure 3. A. Cytofluorogram of z slice number 7 from the deconvolved image stack in Figure 2.

Horizontal and vertical white lines show the overlap thresholds determined using Fiji

1.45k>Analyze>Colocalization>Colocalization Threshold to find the Costes automatic

thresholds. Zero-zero pixels were excluded from the analysis. Diagonal white line has a slope of

1.63 indicating the best fitting stoichiometry between LDL and LDLR intensity. B. PC

significance test. The frequency distribution of PCs for 1000 randomizations of the deconvolved

z slice was obtained using JaCoP plugin for ImageJ. Prism software (GraphPad) was used to fit

the data to a Gaussian function (solid line). The fitted curve had a mean of zero and a standard

deviation of 0.003 (R^2= 0.9965). Dashed lines represent the PCs determined for the z slice

using Fiji1.45k>Analyze>Colocalization>Coloc 2 before (PC=0.62) and after (PC=0.47)

applying Costes’ thresholds. The PC after thresholding is 150-fold greater than the standard

deviation of the distribution of PCs calculated for the randomized images, indicating highly

significant co-localization.

Table I. Co-localization statistics computed for the deconvolved z slice in Figure 3 using four

different methods. The PC before thresholding, the PC after applying Costes’ thresholds, the

Manders’ coefficients after thresholding (tM1 and tM2), the threshold values for each channel,

and the method used to exclude background pixels are shown.

Figure 4. Co-localized voxels in the deconvolved z slice used for Figure 3 and Table I visualized

as a separate channel. The co-localization channel was generated using Imaris Coloc. A. Red

channel B. Green channel C. Co-localization channel D. Merge

Figure 5. Timecourse of co-localization of LDL with the early endosome marker, EEA1. Ten z

stacks for each timepoint of a pulse-chase lipoprotein uptake experiment (except the zero

timepoint, for which n=5) were analyzed using Imaris Coloc (Bitplane). Mean PC ± SEM is

plotted vs. time. Camera gain, exposure times, binning, objective magnification, z stack spacing

and number of slices, deconvolution settings and co-localization analysis parameters were

identical for all datasets.

Table I. Comparison of colocalization metrics analyzed with different software implementations

of colocalization analysis. The overall PC for the raw images, the PC calculated after the images

were intensity thresholded, and the thresholded Manders’coefficients (M1 and M2) are shown

for the same pair of images analysed with Coloc 2 or Colocalization Analysis in Fiji, JaCoP in

ImageJ and Imaris Coloc. The threshold values for each channel and the method used to choose

the ROI are indicated for each result. The differences illustrate the dependence of the results of

on the algorithm used.

Figure 1

Figure 2

Figure 3

A. B.

Figure 4

Figure 5

Table I