0000-2006 a Guided Tour Into Sub Cellular Colocalization Analysis in Light

8/6/2019 0000-2006 a Guided Tour Into Sub Cellular Colocalization Analysis in Light

1/20

2006 The Royal Microscopical SocietyNo claim to original US government works

Journal of Microscopy, Vol. 224, Pt 3 December 2006, pp. 213232

Received 13 April 2006; accepted 28 June 2006

BlackwellPublishingLtdTUTORIAL REVIEW

A guided tour into subcellular colocalization analysis in lightmicroscopy

S. BOLTE * & F. P. COR DEL IRE S*Plateforme dImagerie et de Biologie Cellulaire, IFR 87 la Plante et son Environnement, Institut desSciences du Vgtal, Avenue de la Terrasse, 91198 Gif-sur-Yvette Cedex, France

Institut Curie, CNRS UMR 146, Plateforme dImagerie Cellulaire et Tissulaire, Btiment 112,Centre Universitaire, 91405 Orsay Cedex, France

Key words. Colocalization, confocal microscopy, uorescence microscopy, imageanalysis, wide-eld microscopy.

Summary

It is generally accepted that the functional compartmentalizationof eukaryotic cells is reected by the differential occurrence of proteins in their compartments. The location and physiologicalfunction of a protein are closely related; local information of aprotein is thus crucial to understanding its role in biologicalprocesses. The visualization of proteins residing on intracellularstructures by uorescence microscopy has become a routineapproach in cell biology and is increasingly used to assess their

colocalization with well-characterized markers. However, image-analysis methods for colocalization studies are a eld of contentionand enigma. We have therefore undertaken to review the mostcurrently used colocalization analysis methods, introducingthe basic optical concepts important for image acquisition andsubsequent analysis. We provide a summary of practical tipsfor image acquisition and treatment that should precede propercolocalization analysis. Furthermore, we discuss the applicationand feasibility of colocalization tools for various biologicalcolocalization situations and discuss their respective strengthsand weaknesses. We have created a novel toolbox for subcellularcolocalization analysis under ImageJ, named JACoP, thatintegrates current global statistic methods and a novelobject-based approach.

Introduction

Colocalization analysis in optical microscopy is an issue thatis aficted with ambiguity and inconsistency. Cell biologists haveto choose between a rather simplistic qualitative evaluation of

overlapping pixels and a bulk of fairly complex solutions, mof them based on global statistic analysis of pixel intensdistributions (Manderset al., 2003; Costeset al., 2004; Liet al.,2004). The complexity of some of these different analysis tomakes it difcult to implement the appropriate method anreects the fact that the majority of colocalization situatiodemand customized approaches. All-round analysis tools dnot necessarily t all circumstances as cells contain a plethora structures of multiple morphologies, starting from lineaelements of the cytoskeleton, punctate and isotropi

compartments such as vesicles, endosomes or vacuoles, goito more complex anisotropic forms such as Golgi stacks athe network-like endoplasmic reticulum. The colocalizationtwo or more markers within these cellular structures may bdened as an overlap in the physical distribution of the molecupopulations within a three-dimensional volume, where thimay be complete or partial overlap.

The limits of resolution in optical microscopy imply uncertainty of the physical dimensions and location of smaobjects in the two-dimensional and even more in the threedimensional space. The frequent question is: are two uorochromlocated on the same physical structure or on two distincstructures in a three-dimensional volume? The answer dependon the denition of terms and limits, bearing in mind that thuorochrome distribution may be in the nanometre rangewhereas the optical microscopes resolution is closer to thmicrometre. The veracity of any statement concernincolocalization will thus be limited not only by a good undestanding of the three-dimensional organization of the cell anits subcellular compartments, the quality and reliability of thlabelling techniques or the faithfulness of the markers applito highlight and identify the different cellular addresses. will be equally limited by the dimensions dened by the optisystem and the image-acquisition procedure. The authenti

Correspondence to: S. Bolte. Tel: 0033 69863130; Fax: 0033 169 86 1703;e-mail : [email protected]. P. Cordelires. E-mail: [email protected]; accepted28June2006


2/20

214 S. BOLTE AN D F. P. CORDEL IRES

2006 The Royal Microscopical Society, Journal of Microscopy ,224 , 213232No claim to original US government works

visualization of this three-dimensional organization thusdepends on a good control of the optical system used and, as amatter of fact, on the mastery of some basics in optics, imageprocessing and analysis.

We therefore propose a guideline for the acquisition,qualitative evaluation and quantication of data used forcolocalization purposes. We give an overview on the state of the art of colocalization analysis by reviewing the mostimportant features available in standard imaging software.Finally, we introduce a novel tool for colocalization analysis,named JACoP (Just Another Co-localization Plugin), that combinesthese currently used colocalization methods and an object-basedtool named three-dimensional object counter as plugins to thepublic domain ImageJ software (Rasband, 19972006).

Before getting started

Basic optical principlesBefore using any microscope to collect images, one has to beaware of its limitations. One of these is closely linked to thedual nature of light, which is both a wave and particle phe-nomenon. The objective lens allows the collection of light that

is only partial and is quantied by a parameter called numeri-cal aperture (NA). It is linked to the angle of collection of lighemitted from the specimen and will determine the ability todistinguish between two adjacent punctate light sources. Undercritical illumination, the NA of the condenser illuminating thesample should be the same as that of the objective. In epiuo-rescence microscopy, the objective acts as the condenser andso this critical condition is met. Each point of a light waveexiting a lens can then be considered as a single light sourceemitting a circular wave front (Huygens principle). Thereforewhen placing a screen after a lens, a diffraction pattern can becollected, resulting from interferences between adjacent wavesThis pattern denes the two-dimensional diffraction gure,which consists of concentric rings alternating from light todark (Fig. 1A). The rst light disc is called the Airy disc (Inou1995). When tracing a line through this pattern, we obtain acurve (Fig. 1D) representing the uorescence intensity distributionof the particle along this line. The Airy disc then correspondto the area below the major peak of this curve and the fullwidth at half maximum of this uorescence intensity curve(Fig. 1D) is used to dene the resolution of the optical system

To be able to distinguish between two similar punctatelight sources through a lens, the corresponding Airy discs should

Fig. 1. An image of a point is not a point but a pattern of diffracted light. (AC) Two-dimensional diffraction patterns of the centres of 170-nmuorescent beads seen through a wide-eld microscope. (D) and (E) Corresponding uorescence intensity curves traced along a line passing throcentre of the beads in (A) and (B), respectively (I being the maximum intensity). (F) Three-dimensional projection of the z-stack representing the dipattern of the uorescent bead seen from the side. (A) and (D) Note the concentric light rings around the Airy disc of a single uorescent bead. Tdisc is the rst light patch in this diffraction pattern. Two characteristic dimensions may describe the bell-shaped curve: 1, Airy disc diameter, whicdistance between the two points where the rst light ring extinguishes; 2, full width at half maximum (FWHM), which is directly related to resolubelow). (B) and (E) Diffraction pattern of two beads. Two objects are resolved if their corresponding intensity curves at I/2 are distinct. The criticald between the centres of the intensity curves denes the lateral resolution (x, y) of the optical system. It is equal to FWHM. (C) Three-dimprojection of a z-series of a uorescent bead seen from the side (x, z) representing the diffraction pattern of the same uorescent bead. Note that tresolution (z) of an optical system is not as good as the lateral resolution (x, y). (F) The diffraction pattern is not symmetric around the focal planmore pronounced on the upper side proximal to the objective. Note that a bright 10-nm bead would produce patterns of the same dimensions as thinm bead.


3/20

GUIDED TOUR INTO SUBCELLULAR COLOCALIZATION ANALYSIS IN LIGHT MICRO215

2006 The Royal Microscopical Society, Journal of Microscopy , 224 , 213232No claim to original US government works

be apart from each other (Fig. 1B). The minimal distance (d )between their centres, which gives an integral energy distributionwhose minimum isI /2, is taken to dene the optical resolutionor separating power (Fig. 1E). This parameter may be calculatedaccording to the laws of Abbe (Table 1). It depends on the NAof the objective that, in turn, is dependent on the refractiveindex of the medium and on the wavelength of emitted light.Furthermore, the optical resolution depends on the type of microscope used. A wide-eld microscope may separate twodots 200 nm apart from each other (63oil immersion objective,NA=1.32, emission wavelength 510 nm). Introducing a con-focal pinhole of 1 Airy width (i.e. an aperture whose diametercorresponds to the diameter of the rst Airy disc for the currentwavelength) into the optical system will result in an improve-ment by approximately 30% of this lateral resolution becauseout-of-focus light is eliminated from the detector (Abbe, 1873,

1874; Minksy, 1961). As a rst approximation, only lightcoming from the rst Airy disc is collected. This means thatthe aperture of the pinhole will mainly depend on the objectiveused and on the refraction indexes of all media encountered bylight on its way to and away from the sample. It should be set to1 Airy unit to ensure confocal acquisition.

Biological samples are not two-dimensional limited. The useof stepper motors or piezo-electrical devices in wide-eld orconfocal laser scanning microscopes allows the collection of optical sections representing the three-dimensional volumeof the sample by moving the objective relative to the object orvice versa. As a consequence, the diffraction pattern of lightshould be considered as three-dimensional information andwill dene the point spread function (PSF) (Castelman, 1979).The Airy disc along the z-axis appears elongated, like a rugbyball (Fig. 1C), and the overall diffraction pattern of light hasaxial symmetry along the z-axis with a three-dimensionalshape of the PSF that is hourglass-like (Fig. 1F). The minimumdistance separating two distinguishable adjacent Airy discsalong the depth of the PSF will dene the axial resolution of the microscope (Table 1). The optical laws introduced hereimply that colocalization must be measured in the three-dimensional space. The imbalance between the lateral andaxial resolution of optical microscopes leads to a distortion

of a round-shaped object along the z-axis. Bear in mind thabrilliant nanometric object will nevertheless yield an imagwhose waist is at least 200 nm and whose depth is abou500 nm, as dened by the Airy disc. Therefore, any colocalizatanalysis must be carried out in the three-dimensional spaceFurthermore, it is self-evident that three-dimensional projectioof image stacks must not be analysed as they shrink volumetrinformation to two dimensions, leaving aside the deptcomponent.

Digital imaging

The limits of optical resolution depend on the PSF and direcinuence imaging parameters. Once an image has beeformed by the optical system, it will be collected by an electrodevice that will translate a light signal into an electronic sign

for further processing by the computer. Microscope images agenerally captured either by digital cameras (a parallel matrix) photomultipliers (a sweep of point measurements) thacompose the nal image as a matrix of discrete picture eleme(pixels). The denition of an image as pixels implies soprecautions in image acquisition. To resolve two points and avoid under- or over-sampling, the pixel size applied shouldequal to the lateral limit of resolution between the two poindivided by at least 2 according to the Nyquist samplintheorem (Oppenheimet al., 1983). In microscopy it is widelyaccepted that, according to this theorem, to reproducfaithfully formed images the detector should collect light 2.3the frequency of the original signal. Basically, this meathat the projected image of a single dot should appear on least two adjacent sensitive areas of the detector in a giveaxis, namely on four pixels (2 2 for x, y). Therefore, thesampling frequency should be at least twice greater than thresolution of the current dimension (x, y or z). For twdimensional acquisitions this means that the minimal justied pixel size is calculated by dividing the lateral resoluby at least 2. In three-dimensional imaging, the size of the z-srelies on the same laws, i.e. the axial resolution also has to divided at least by 2. The minimal justied pixel size and z-step size depend on the NA of the objective, e.g. a 6

Table 1. The laws of Abbe and their effect on optical resolution and pixel sizes in wide-eld and confocal microscopy.

Wide-eld Confocal

Lateral resolution dx, y Axial resolution dx, z Lateral resolution dx, y Axial resolution

Expression 0.61 em/NA 2 em/NA2

0.4 em/NA 1.4 em/NA2

Limit resolution of a 63oil 232 nm 574 nm 152 nm 402 nmimmersion objective withNA=1.32 at em =500 nmMinimal justied pixel sizefor this objective

101 nm 250 nm 66 nm 175 nm

NA, numerical aperture.


4/20



objective (oil immersion, NA= 1.32) collecting emittedlight of 500 nm with a lateral resolution of 232 nm and anaxial resolution of 574 nm implies a minimal justied pixelsize of 101 nm and a z-step size of 250 nm (see also Table 1).

It is important to note that image acquisition for colocalizationanalysis should always be carried out on several subsequentoptical sections, i.e. in three dimensions, and near to theresolution limit of the optical system, i.e. with the appropriate justied pixel size and z-step size.

A frequent mistake in microscopy is oversampling. Thishappens when a single subresolution light source is tted onmore than 2 (or 2.3) adjacent pixels on the detector, i.e. usingpixel sizes smaller that the minimal justied pixel size denedby optical resolution and the Nyquist theorem. The resultingimage looks larger but the signal looks dimmer as the light isspread out on more parts of the detector than required. Eventhough the sample seems to be highly magnied, there is nogain in resolution as the optical resolution limit cannot besurmounted. It is furthermore important to avoid saturationof images, as saturated pixels may not be quantied properlybecause information of the most intense grey level values in ahistogram gets lost. It is difcult to judge by eye if an imagecomposed of grey values, or green or red hues is saturated, asthe human eye is not sensitive enough. Our eye can, however,distinguish between hundreds of colours and therefore mostimage-acquisition software provides colour look-up tables withhues indicating saturated pixels and providing the possibilityof adjusting the dynamics of grey values on the detector side.

Choice of the acquisition technique

We have learned that optimal image acquisition for colocalizationanalysis relies mainly on the limits of optical resolution; it isthus important to adapt the optical system to the biologicalquestion and to choose the appropriate microscope. Confocalimaging gives high resolution, eliminating out-of-focus lightby introducing a pinhole on the detector side. Confocal imaging is

recommended when handling thick or highly diffusive samplessuch as plant tissue or brain tissue. It is important to note thatimage acquisition with standard confocal microscopes is fairlyslow (1 s image1) and thus has been more suited to three-dimensional imaging of colocalization in xed samples ratherthan in live samples. A disadvantage of excluding out-of-foculight from the detector by a confocal pinhole is that valuableinformation may get lost and low signals might not bedetected (Fig. 2A). The Airy disc in fact comprises only 10% the total energy from a point source. Wide-eld microscopesequipped with rapid charge-coupled devices might be a goodalternative if one wants to cope with these kinds of problemsas three-dimensional acquisition can be performed very rapidly(20 ms image1) and low-intensity information will not belost, as all information will be collected by the detector. Theadvantage of collecting all information, i.e. out-of-focus lightis a constraint at the same time as images are blurred anddifcult to analyse directly (Fig. 2B). This out-of focus lighinterferes with accurate colocalization analysis and makesimage restoration necessary. The image that is formed on adetector by a single particle (with a size below optical resolutionwill be dened by the PSF of the optical system used. Opticonvolute image information. This means that the hourglass-like shape of the PSF is a model for the three-dimensionaspread of light caused by the optical system. Reassigning theout-of-focus blurred light to its origin is performed by a procescalled deconvolution (Fig. 2C). This is a computationaltechnique that includes methods that help to reattribute thesignal spread in three dimensions according to the PSF toits origin. Deconvolution may restore the resolution of images

in both wide-eld and confocal microscopy and is the subjecof some excellent reviews (Wallace & Swedlow, 2001; Sibarita2005). Deconvolution in combination with wide-eld microscopyis restricted to thin objects (


5/20



each uorescent signal of the sample results in a diffractionpattern that is displayed on the detector. Moreover, PSFs arenot constant in the three-dimensional volume imaged, as the PSFsare degraded in the depth of the sample and appear to be disturbedat the interface of two media with different refraction indexes.

Further techniques have been developed that overcome theconstraints of acquisition rate or out-of-focus light. These includestructured illumination and rapid confocal devices and arediscussed in detail elsewhere (Brownet al., 2006; Gariniet al.,2005). In this work, however, we will focus on commonlyavailable standard confocal and wide-eld microscopy.

Incidence of uorochromes, light sources, lters and objectives

It has already been mentioned that the resolution capacity of an optical system depends on the angular properties of itsobjective, the composite refractive index of all media crossedby light and the emission wavelength of the uorochromesused (Table 1). A number of uorochromes may be used tolabel different proteins of interest. The ability to distinguishbetween individual emission spectra is a primary concern,reinforced by selective excitation of only one uorochrome at atime. This aim is achieved by optimizing: (i) the choice of uorochromes, (ii) the selectivity of excitation and (iii) themeans of emission discrimination.

Any uorescent reagent can be characterized by its excitationand emission spectra, which in turn may depend upon theuorophores environment (Valeur, 2002). These classicalcurves, respectively, represent the probability of making anelectronic transition from ground to excited state when

exposed to photon energy of a particular wavelength and torelease a photon at a particular wavelength when fullling theopposite transition. The rst value to be taken into account isthe Stokes shift, which is dened as the spectrum distancebetween the most efcient excitation (peak in the excitationspectra) and the maximum of emission. The ability to sortemission from excitation light depends partly on this value, asincident light is about 104 more intense than the signal being

recovered (Tsien & Waggoner, 1995). The width of excitatiand emission curves contributes to the practicality of uorescereagents for distinctiveness; the narrower the curves, the easithe uorochromes will be to separate. However, this is ontrue for uorochrome pairs with spectra far enough apart fromeach other.

A wide range of uorescent reagents is now available cover the spectrum from visible to near infrared. Fluorochrommay be coupled to primary or secondary antibodies for immnolabelling. Other uorescent compounds may accumulate ispecic cellular compartments, such as nuclei, endoplasmreticulum, Golgi apparatus, vacuoles, endosomes, mitochondror peroxisomes. Genetically encoded targeted uoresceproteins from jellysh or corals are readily available and ahelpful in live cell studies. Newly engineered semiconduccolloidal particles (Q-Dots) are adapted for single moleclabelling (Dahanet al., 2003; Gaoet al., 2004).

When choosing uorochrome combinations for colocalizatiostudies, their spectra must be unambiguously distinctive. Furthemore, it has to be considered that these spectra may be dependeon the physical environment (Bolteet al., 2004a, 2006).

We have to introduce here the terms bleed-through andcross-talk of uorochromes, as avoiding these phenomena crucial to colocalization analysis. Bleed-through is the pasage of uorescence emission in an inappropriate detectiochannel caused by an overlap of emission spectra (Fig. 3Cross-talk is given when several uorochromes are excitwith the same wavelength at a time because their excitatiospectra partially overlap.

Lets consider the uorochrome couple uorescein is

thiocyanate (FITC) and Cyanine3.18 (Cy3), which is frequenused for immunolabelling for colocalization analysis (Fig. The excitation spectra of these two uorochromes seem to well apart with FITC peaking at 494 nm and Cy3 with a minexcitation peak at 514 nm and a major excitation peak a554 nm. Even using the narrow laser line of 488 nm for FITexcitation, one may already observe a slight cross-talk betweFITC and Cy3, as Cy3 excitation spectra have slight but signic

Fig. 3. Denition of cross-talk and bleed-through with the uorochrome couple uorescein iso-thiocyanate/Cyanine3.18 (FITC/Cy3). (A) Espectra of FITC (broken line, max. 490 nm) and Cy3 (solid line, max. 552 nm). The grey arrow marks the position of the standard 488-nm lasconfocal microscopes. Note the overlap of the excitation spectra at 488 nm (cross-talk). (B) Emission spectra of FITC (broken line, max. 520 nm(solid line, max. 570 nm). The grey bar marks the typical detection window of Cy3. Note the overlap of FITC and Cy3 emission in this detecti(bleed-through).


6/20



absorbance at 488 nm (Fig. 3A). Moreover, even when excitingFITC and Cy3 sequentially with 488 and 543 nm, one maydetect a bleed-through of the lower energy (yellow) part of theFITC emission coinciding with the emission maximum of Cy3in the Cy3 detection channel (Fig. 3B). When using band-pass-ltered excitation light, such as in wide-eld microscopy,instead of laser lines or monochromatic light, the situationmay get worse. It is thus essential to apply some simple strategiesthat help to avoid cross-talk and bleed-through. Firstly, it isalways important to have single labelled controls for eachuorochrome used. In this way one may check for bleed-through between uorochromes on the detector side. Secondly,in laser scanning microscopy, it is highly recommended toperform sequential acquisitions exciting one uorochrome ata time and switching between the detectors concomitantly.

Another method of meeting the challenge is spectral unmixing,a quite simple mathematical operation that was originallydeveloped for satellite imaging. Spectral unmixing softwarepackages are often included in image-acquisition software of the microscope manufacturers. By this technique, which isa correction of spectral bleed-through, it is also possible toenhance the chromatic resolution of uorescence microscopy.Two general approaches may be distinguished. One is to performmicrospectrouorometry and to use the model (or measure)of separate uorochromes to perform spectral deconvolutionof the complex raw image (Zimmermannet al., 2003). Thisimplies curve tting and extrapolation. A second, simplerapproach is to experimentally determine the bleed-throughfactor for a given optical conguration and to use this to derivecorrected values for each pixel. This is analogous to pulse

compensation in ow cytometry.To unmix the spectra of uorochromes with stronglyoverlapping emission spectra, it is necessary to assign thecontribution of different uorochromes to the overall signal.This is done rst by determining the spectral properties of theindividual uorochromes under the same imaging conditionsused for the multilabelled samples.

We will again consider the two uorochromes FITC and Cy3seen through their respective lters A and B. Using a mono-labelled slide, FITC seen through A will give an intensity aFITCand bFITCthrough B. Analogous notations will be used for Cy3.Then imaging a dual-labelled FITC and Cy3 sample, the imagethrough A will be aFITC +aCy3; the image of FITC acquired usingthe appropriate lter is contaminated by a contribution fromCy3. The same phenomenon will occur for the image of Cy3collected through B (bFITC +bCy3). The use of mono-labelledslides allows the estimation of the relative contribution of FITCto the image of Cy3 and is used to give a more reliable image of FITC (aFITC +bFITC) and Cy3 (aCy3 +bCy3). The ratio FITC : Cy3 of the average intensities of single uorochrome-labelled struc-tures measured at the two excitation wavelengths for FITC andCy3, respectively, gives a constant that is specic for eachuorochrome under given experimental conditions and xedsettings. The intensity is then redistributed in order to restore

a corrected signal for each colour channel undisturbed byemission from the other uorochrome.

Fluorochromes may also transfer energy to each other byFrster resonance energy transfer (for review see Jares-Erijman & Jovin, 2003). This non-radiative energy transfermay occur when the emission spectrum of the rst uorochrome(donor) overlaps with the excitation spectrum of the seconduorochrome (acceptor) and if the donor and acceptormolecules are in close vicinity (10100 ). Frster resonanceenergy transfer causes a reduction of the emission of the donoruorochrome and an increase of the emission of the acceptoruorochrome, therefore resulting in a misbalanced intensityratio between the two image channels. It is thus also crucial toselect the rst uorochrome with an emission spectrum asdistinct as possible from the excitation spectrum of the secondfluorochrome in order to avoid Frster resonance energy transfereffects that would complicate the interpretation of colocalizationdata.

The choice of light sources and appropriate lters is the nexstep for appropriate discrimination between uorescencespectra. We have already learned that using monochromaticlight from a laser source in a confocal microscope lowers therisk of exciting several uorochromes at a time, even if it doenot exclude cross-talk. In wide-eld microscopy mercury orxenon lamps have spectral output spanning from UV toinfrared, with numerous peaked bands, notably in the case of mercury. They are used in combination with appropriatelters or as part of monochromators. As a consequence, whenusing ltered light the excitation is not monochromatic andthe risk of exciting several uorochromes at a time is high

This inconvenience may be partially circumvented by using amonochromator to generate a suitably narrow subrange of wavelengths that may be optimized for each situation. How-ever, care has to be taken as the monochromator may gener-ate a slight excitation leakage on both boundaries of thenarrowed excitation window, leading to possible cross-talk.

The choice of objectives used for colocalization analysis athe subcellular level is crucial to attain optimal resolution.Objectives used should be of high quality, with a high NA(>1.3) and magnications adapted to the camera in wide-eld microscopy. In both kinds of microscopy, the NA icritical, as z-resolution improves as a function of (NA)2 (seeTable 1). Objectives should be corrected for chromatic andspherical aberrations. Chromatic aberrations are due to thefailure of the lens to bring light of different wavelengths to acommon focus. Spherical aberrations come from the failure oa lens system to image the central and peripheral rays at thesame focal plane. Objectives corrected for both aberrations arecalled plan-apochromatic and confocal microscopes areusually equipped with these. For colocalization analyses itis recommended to use immersion objectives to reduce aberrationdue to the refraction index changes. This means oil immersionfor xed mounted specimens and aqueous immersion for livecell studies.


7/20



Checking the system

Before performing colocalization measurements, it is importantto check the microscopes integrity. This may be done bymeasuring the PSF of the optical system (Scalettaret al.,1996; Wallace & Swedlow, 2001), using objects whose sizesare just matching or below the microscopes resolution. Smalluorochrome-labelled polystyrene beads of 100170 nm areavailable for this. Remember that the resolution of the opticalsystem is closely linked to the NA of the objective used, refractionindex of the mounting medium, immersion medium (oil, glycerolor water), coverslip thickness and emission wavelength of theuorochrome. Individual PSFs should thus be measured onuorescent beads of the respective wavelengths mounted inidentical conditions to the sample and with the objectives thatare used for colocalization analysis.

The shape of the PSF of a uorescent bead gives an intuitivecharacterization of the image quality. It can also be used to testthe objective performance and integrity. A dirty objective or anon-homogeneous immersion medium will result in a deformedPSF (Sibarita, 2005). Returning to objective quality, one maybe surprised to observe that the maxima of intensity for alluorochromes may not be coincident in space. This observationis due to an imperfection in the lens design or manufactureresulting in a variable focalization of light as a function of wavelength. Even if most manufactured objectives areapochromatic, the refraction index of immersion oil isdependent on both temperature and wavelength, giving riseto this phenomenon. Likewise, glycerol is hygroscopic andits refractive index will in practice change with time. As a

consequence, and especially in the case of colocalizationstudies, the chromatic aberration may in this case be determinedand the shift between images corrected (Manders, 1997).

Pre-processing of images

As perfect as an optical system can be, we have already seenthat an image is an imperfect representation of the biologicalsystem. The illumination system used in wide-eld microscopywill impair the image, especially if it is not well aligned. As aconsequence, the eld of view may not be illuminated in ahomogeneous fashion. When trying to quantify colocalizationas a coincidence of intensity distributions, one may need tocorrect uneven illumination. This may simply be done bycorrecting the image of the sample using a bright image of anempty eld. This correction is achieved by dividing the formerimage by the latter. This operation may be carried out withImageJ using the Image Calculator function.

Noise is another major problem in digital imaging. However,before trying to correct images for it, we must rst address itspossible origins. Illumination systems such as mercury orxenon lamps are not continuously providing photons andmay be considered as blinking sources. As a consequence,even though all regions of a eld will statistically be hit by

the same number of photons over a long period, the numbeof photons exciting uorochromes is not the same whecomparing a region with its neighbours on a milliseconscale. Similarly, the emission of a photon by a uorochromedependent on its probability of returning to ground state. Thso-called photon noise will imprint a salt-and-pepper-likbackground on the image. As it is a stochastic function, it cabe partially overcome by increasing the exposure time ocharge-coupled device cameras or slowing the frequenc(increasing dwell time) of scanning on a confocal microscoOne may also collect successive images and average them.

Furthermore, noise originating from the detection devic(electronic noise or dark current) may be limited by coolinthe detection devices.

Intrinsic statistical noise follows a Poisson distribution. Tremove this kind of noise, images may be post-processusing adaptive ltering. This may be done by changing thpixel value to an intensity calculated on the basis of the locstatistical properties of both the signal and noise of neighbouripixels. This may, however, result in a loss of features such sharp contours. Out-of-focus light may be reassigned to origin by deconvolution as already mentioned (Wang, 1998)

Finally, imaging may be impaired by background cominfrom either natural uorescence of the sample or being generatwhen preparing the sample. In most cases, nothing can bdone after image acquisition unless a uniform background observed. In this special case, its mean intensity is determinand this value is subtracted across the full image. More subprocesses exist, such as spectral unmixing, that may givbetter results on specic problems and the reader may consu

appropriate image-processing handbooks (Gonzales & Woo1993; Pawley, 1995; Ronot & Usson, 2001).

Visualizing colocalization

When visualizing colocalization, the elementary method is present results as a simple overlay composed of the differechannels, each image being pseudo coloured using an appropriacolour look-up table. For example, it is commonly acceptthat the dual-channel look-up table for green and red will givrise to yellow hotspots where the two molecules of interest apresent in the same pixels. However, anyone who has beeusing this method knows its limits. The presence of yellospots is highly dependent on the relative signal intensitcollected in both channels; the overlay image will only givereliable representation of colocalization in the precise case whboth images exhibit similar grey level dynamics, i.e. when thistograms of each channel are similar. This is rarely the cawhen imaging two uorochromes with differential signastrength. As a consequence, image processing is required match the dynamics of one image to the other. This is oftedone by histogram stretching. However, histogram stretchinmay result in falsied observations because the resultanimage does not reect the true stoichiometry of the molecul


8/20



imaged. An alternative to histogram stretching is the useof specically designed look-up tables that will enhance thevisual effect of coincidental locations (Demandolx & Davoust,1997). These authors proposed a new pseudo-colourizationmethod in the form of a look-up table enabling visualization of the rst uorophore alone in cyan and the second alone inmagenta. As the colocalization event is generally difcult to visualizeand as the ratio of uorophores may vary locally, they usedgreen and red to highlight regions where one uorophore ismore intense than the other and yellow in the case where bothintensities are the same. This method improved the discrimi-nation of uorescence ratios between FITC and Texas Red.

Measuring colocalization

Overlay methods help to generate visual estimates of colocali-zation events in two-dimensional images; however, they neitherreect the three-dimensional nature of the biological probenor the restrained resolution along the z-axis. Furthermore,these overlay methods are not appropriate for quanticationpurposes because they may result in misinterpretation of relative

proportions of molecules. To overcome these problems imaganalysis is crucial. There are two basic ways to evaluatecolocalization events, a global statistic approach that performsintensity correlation coefcient-based (ICCB) analyses and anobject-based approach.

The theory behind some of these tools is rather complex andsometimes difcult to compile and the results obtained havebeen difcult to compare until now. Here, we introduce apublic domain tool named JACoP(http://rsb.info.nih.gov/ij/plugins/track/jacop.html) that groups the most importantICCB tools and allows the researcher to compare the variousmethods with one mouse-click. Furthermore, an object-based tool called three-dimensional object counter(http://rsb.info.nih.gov/ij/plugins/track/objects.html) is also availablethat may be used for object-based colocalization analysis. Thesetools process image stacks and allow an automated colocalizationanalysis in the three-dimensional space. To introduce thesetools and their utility in colocalization analysis we will give a generaoverview on the roots of ICCB and object-based methods.

For this purpose, we have compared four different possiblesubcellular colocalization situations (Fig. 4). A complete

Fig. 4. Reference images for colocalization analysis.Images for colocalization analysis were acquired fromxed maize root cells with Golgi staining (A) (Bouttet al ., 2006) or endoplasmic reticulum staining (B)(Klugeet al. , 2004) and on xed mammalian HeLacells with microtubule plus-end tracking proteins EB1and CLIP-170 staining (C) (Cordelires, 2003), andnuclear and mitochondrial staining (D). Scale bars,10 m. These images illustrate the four commonlyencountered situations in colocalization analysis. (A)Complete colocalization. (B) Complete colocalizationwith different intensities. (C) Partial colocalization.(D) Exclusion. Grey level images of the green and redimage pairs (AD) were used for subsequent treatmentswith ImageJ. A zoomed view of the insets is shown oneach side of the colour panels.
http://rsb.info.nih.gov/ij/http://rsb.info.nih.gov/ij/


9/20



colocalization situation has been modelled by duplicating araw image of a Golgi staining in a plant cell (as in Bouttet al.,2006) and assigning it to two different colour channels(Fig. 4A, Raw and Duplicated). Another situation, completecolocalization with different intensities, is given by the cola-belling of the endoplasmic reticulum with two endoplasmicreticulum-specic antibodies (as in Klugeet al., 2004; Fig. 4B). Apartial colocalization situation is shown by the colabelling of mammalian cells with different microtubule plus-end trackingproteins (Cordelires, 2003; for reviews, see Schuyler & Pellman,2001; Galjart, 2005) (Fig. 4C). Exclusion of uorescentsignals has been achieved by staining mitochondria andthe nucleus in mammalian cells (Fig. 4D). To investigate theinuence of uorescence background or photonic noise oncolocalization analysis with JACoP, we added different levels of random noise to the complete colocalization image pair (imagedata not shown). The signal-to-noise ratios in these imageswere calculated and varied from 12.03 to 3.52 dB.

Correlation analysis based on Pearsons coefcient

The ICCB tools mainly use statistics to assess the relationshipbetween uorescence intensities. A wealth of colocalizationanalysis software now available as part of basic image-analysistools or more specialized imaging-analysis software is basedon ICCB analysis. This is mainly due to the relative ease of implementing the software. In this case, statistical analysis of the correlation of the intensity values of green and red pixelsin a dual-channel image is performed. This is mostly doneusing correlation coefcients that measure the strength of the

linear relationship between two variables, i.e. the grey valuesof uorescence intensity pixels of green and red image pairs.

Pearsons coefcient. A simple way of measuring the dependencyof pixels in dual-channel images is to plot the pixel grey valuesof two images against each other. Results are then displayedin a pixel distribution diagram called a scatter plot (Fig. 5) oruorogram. The intensity of a given pixel in the green imageis used as the x-coordinate of the scatter plot and the intensityof the corresponding pixel in the red image as the y-coordinate.In some software the intensity of each pixel represents thefrequency of pixels that display those particular red and greenvalues in the uorogram image. Leaving aside noise and lowbackground, we will rstly examine the scatter plot to see if there are numerous pixels with only one signicant signal(Fig. 5E). Secondly, where both signals are present, we shalldescribe their relationship as a strong, lower, weak or non-existentcorrelation that may be positive or negative. If we considerthat the labelling of both uorochromes is proportional tothe other and the detection of both has been carried out in alinear range, the resulting uorogram pattern should be aline. The slope would reect the relative stoichiometry of both uorochromes, modulated by their relative detectionefciencies. In practice in a complete colocalization situation,

dots on the diagram appear as a cloud centred on a line (seFig. 5A). The spread of this distribution with respect to ttted line may be estimated by calculating the correlatiocoefcient, also called Pearsons coefcient (PC). As most Itools are based on the PC or its derivatives, we will introduchere in detail.

The linear equation describing the relationship between thintensities in two images is calculated by linear regressioThe slope of this linear approximation provides the rate association of two uorochromes. In contrast, the PC providan estimate of the goodness of this approximation. Its valucan range from 1 to1, with 1 standing for complete positivecorrelation and1 for a negative correlation, with zero standingfor no correlation. This method has been applied to measuthe temporal and spatial behaviour of DNA replication iinterphase nuclei (Manderset al., 1992). We used the JACoPtool to analyse the Pearsons correlation coefcients and visualize the corresponding scatter plots of the four differecolocalization situations described in Fig. 4. Figure 5(A) shothe scatter plot with the dots on the diagram appearing as cloud centred on a line in the case of complete colocalizatioThe PC approaches 1 in this case. A difference in the intensitof the green image with still completely colocalized structuresults in a rotation of the dotted cloud towards the red ax(Fig. 5B). As a consequence, the tted line changes its sloand comes closer to the axis of the most intense channel. We cstate that colocalization is observed whenever both signals asignicant but that a subpopulation of purely red pixels haappeared because of poor sensitivity in the green channel. the partial colocalization situation the dots of the scatter pl

form a rather uniform cloud with a PC of 0.69 (Fig. 5C). Mutexclusion of the uorescent signals shows scattered distributioof the pixels close to both axes (Fig. 5D) and a negative PC.

Scatter plots and PCs point to colocalization especialwhere it is complete (Fig. 5A and B); however, they rarediscriminate differences between partial colocalization exclusion, especially if images contain noise. The inuencenoise and bleed-through on the scatter plots and PCs is showin Fig. 5(A*) and (F) (black bars). Random noise has beadded to the image pairs of Fig. 4(A) and is recognizable the shapeless cloud of dots near the origin (Fig. 5A*). Aconsequence, the PC will decrease and nally tend to zero more noise is added (Fig. 5F, black bars). This demonstrathe sensitivity of PC to background noise and hence to threshoing. These results show that an evaluation of colocalizatioevents using PCs alone may be ambiguous, as values are highdependent on noise, variations in uorescence intensities oheterogeneous colocalization relationships throughout thesample (Fig. 5AC). Noise and background must be removMoreover, the coefcient will soon be dominated, not by tcentral phenomenon, but by the perimeter given to the analys(the near-threshold events). Values other than those close toand especially mid-range coefcients (0.5 to 0.5) do notallow conclusions to be drawn.


10/20



This also applies when looking at images corrupted by bleed-through. A thin cloud of correlated pixels will appear on thescatter plot, close to one or both axes (data not shown). As aconsequence, PC will tend to1 or 1 although not representing abiological correlation.

Although provided in most standard image-analysissoftware packages, scatter plots in combination with the PConly give a rst estimate of colocalization. They are especiallyuseful for initial identication of diverse relationships (correla-tions, bleed-through, exceptional coexpression of signals) andfor examination of complex overlays through the windows(regions of interest) so dened. However, they are not sufcientto evaluate colocalization events rigorously. The PC denes the

quality of the linear relationship between two signals but whatif the sample contains two or more different stoichiometries oassociation? The linear regression will try to t the segregateddot clouds as one, resulting in a dramatic decrease of the PCThe best alternative would be to t dot clouds by intervals,resulting in several PCs for a single pair of images.

Manders coefcient. Manders overlap coefcient is based onthe Pearsons correlation coefcient with average intensityvalues being taken out of the mathematical expression(Manderset al., 1992). This new coefcient will vary from 0 to1, the former corresponding to non-overlapping images andthe latter reecting 100% colocalization between both

Fig. 5. Colocalization analysis with JACoP; Pearson and Manders, scatter plots and correlation coefcients. Scatter plots (AD) correspondcolocalization events as shown in Fig. 4. (E) Model scatter plot explaining the effects of noise and bleed-through. (F) Pearsons and Manders coefthe different colocalization situations. A complete colocalization results in a pixel distribution along a straight line whose slope will depend uorescence ratio between the two channels and whose spread is quantied by the Pearsons coefcient (PC), which is close to 1 as red and green intensity distributions are linked (F, an0, black bar). (B) A difference in uorescence intensities leads to the deection of the pixel distribution towards t

axis. Note that the PC diminishes even if complete colocalization of subcellular structures is still given (F, b, black bar). (C) In a partial colocalizationpixel distribution is off the axes and the PC is less than 1 (F, c, black bar). (D) In exclusive staining, the pixel intensities are distributed along the axes of plot and the PC becomes negative (F, d, black bar). This is a good indicator for a real exclusion of the signals. (E) The effect of noise and bleed-throscatter plot is shown in the general scheme. (F) The inuence of noise on the PC was studied by adding different levels of random noise (n1n4)complete colocalization event (A=n0, no noise). (F) Note that the PC (black bar) tends to 0 when random noise is added to complete colocalizing structThe inset (A*) in (A) shows the scatter plot for the n2 noise level. Note that all of the mentioned colocalization events (AD) may only be detectedonce images are devoid of noise. (F) Manders coefcients were calculated for (AD). The thresholded Manders tM1 (cross-hatched bars) and tM2 (diagonalhatched bars) are shown. Compare complete colocalization (an0), complete colocalization with random noise added (an1 an4), and complete colocalization withdifferent intensities (b), partial colocalization (c) and exclusion (d). Note that the original Manders coefcients are not adapted to distinguish betweevents, as they stay close to 1 for all situations (not shown). *Signal-to-noise ratios are: n1=12.03 dB, n2=6.26 dB, n3=4.15 dB and n4=3.52 dB.


11/20



images. M1 is dened as the ratio of the summed intensities of pixels from the green image for which the intensity in the redchannel is above zero to the total intensity in the greenchannel and M2 is dened conversely for red. Therefore, M1(or M2) is a good indicator of the proportion of the green signalcoincident with a signal in the red channel over its totalintensity, which may even apply if the intensities in bothchannels are really different from one another. This denitioncould reveal both coefcients to be perfect for colocalizationstudies. Unfortunately, this is only true if the background is setto zero. Furthermore, it is not possible to distinguish betweencomplete and partial colocalization situations with the M1 andM2 coefcient. The Manders coefcient is very sensitive tonoise. To circumvent this limit, M1 and M2 may be calculatedsetting the threshold to the estimated value of backgroundinstead of zero (Fig. 5F, cross-hatched and diagonal hatchedbars). When noise or cross-talk are present, the automaticallyretrieved threshold may be too high, leading to the loss of valuable information. In this case, noise and cross-talk mustbe corrected before calculating the coefcients.

Costes approach. Recently, a statistical signicance algorithmbased on the PC has been introduced (Costeset al., 2004). TheCostes approach is performed in two subsequent steps. Firstly,the correlation in different regions of the two-dimensionalhistogram is taken into account to estimate an automaticthreshold and the PC of this thresholded image pair is calculated.To calculate this automatic threshold, limit values for eachchannel are initialized to the maximum intensity of each channeland progressively decremented. The PC is concomitantly

calculated for each increment. The nal thresholds are thenset to values that minimize the contribution of noise (i.e. PCunder the threshold being null or negative). As a second step,Costeset al. (2004) introduced a new statistical analysis basedon image randomization and evaluation of PC. The authorspointed out that a single image reects a particle distributionwith sizes above optical resolution. These particles appear as acollection of adjacent pixels with intensities correlated to theirneighbours. The intensity distribution depends on the PSF of the acquisition system and the approximate particle size maybe calculated using the full width at half maximum of theuorescence intensity curve. The full width at half maximumdenes the area over which a signal belonging to a singleparticle is spread out, given the fact that the particle size isconvolved by the PSF of the optical system. The authorscreated a randomized image by shufing pixel blocks with thedimensions dened by the full width at half maximum for theimage of the green channel. This process is done 200 times fora single image and the PC is calculated each time between therandom images of the green channel and the original image of the red channel. The PC for the original non-randomizedimages is then compared with the PCs of the randomized imagesand the signicance ( p-value) is calculated. The p-value, expressedas a percentage, is inversely correlated to the probability of

obtaining the specied PC by chance (i.e. on randomizimage pairs). This value is calculated as the integrated areunder the PC distribution curve, from the minimum PC valuobtained from randomization to the PC obtained from originimages (see Fig. 6). This method introduces for the rst tima statistical comparison that may exclude colocalization opixels due to chance.

We performed this two-step analysis with JACoP for the focolocalization events mentioned earlier. However, for clariwe only show the scatter plot and image pairs analysed for thpartial colocalization event (Fig. 6). We obtained a scatter plthat is divided into four differentially coloured zones bhorizontal and vertical lines that represent the borders of thautomatic thresholds for the red and green channel, respectively (Fig. 6A). The PC is 0.69. Subsequently, we created a of 200 randomized images (see Fig. 6B, randomized greimage) from the green image and calculated the colocalizatiomap and the p-value (Fig. 6B). An overlay of green and rechannels with the mask of the colocalizing pixels in whi(Fig. 6B, colocalization map) gives a topological map of localization distribution. The PC calculated earlier has a p-valueof 100%, suggesting that colocalization in the regions maskin white is highly probable.

Figure 6(C) and (D) show the condence interval, i.e. trange of PC variation obtained from randomized images (curve; D, grey bars), in comparison to the PCs obtained fthe initial set of images (red lines and bars). Surprisinglthe original PC is above the upper boundary of the condeninterval in the complete colocalization situation, in complecolocalization with different intensities and in partial coloca

zation (Fig. 6D, an0 to c). This means that all of those situationmay be considered as true colocalization cases. As expectedthe case of exclusion, the PC is below the lower boundarythe interval and the p-value is equal to 0% (Fig. 6D, d). It seemthat this method points out true colocalization even whenimages are corrupted by high levels of noise (Fig. 6D, an1 an4).However, the Costes approach may reach its limits wheincreasing the statistical parameters of noise and especiallthe SD of noise. The condence interval may encompass toriginal PC, which may impair a prognostic of a true colocazation, as the p-value is dependent on the distance between thelower boundary of the interval and the original PC value. Ithat particular situation, the colocalization diagnostic maynot give rise to a valid conclusion.

Although providing a rst statistical estimate of colocaliztion, Costes approach is also highly dependent on the waywhich the test is set up. The authors initially proposed 20randomization rounds to obtain a signicant statisticadistribution with more randomization leading to more reliabelimination of false positives.

Van Steensels approach. Another development based on PChas been proposed for colocalization analysis using, as aexample, glucocorticoid and mineralocorticoid receptors


12/20



the nuclei of rat hippocampus neurones (Van Steenselet al.,1996). These receptors are concentrated in punctate clusterswithin the nucleus that partially colocalize. The authors applieda cross-correlation analysis by shifting the green image inthe x-direction pixel per pixel relative to the red image andcalculating the respective PC. The PC is then plotted as the

function of x (pixel shift) and the authors thus obtained across-correlation function. We performed the analysis on thefour different colocalization situations with the following resultsCompletely colocalizing structures peak atx =0 and show abell-shaped curve (Fig. 7A). A difference in uorescence intensileads to a reduction of the height of the bell-shaped curve,

Fig. 6. Colocalization analysis with JACoP; Costes. (A) Scatter plot of a partial colocalization situation (such as Figs 4C and 5C). We distinguregions of interest (red, yellow, green and blue overlay); the yellow region represents all pixels above the dual automatic thresholds; the red represents all pixels with red channel intensities over the automatic threshold and the green channel represents intensities below the automthreshold. The green region represents pixels with green pixels over and red pixels below threshold and the blue region designates pixels undthreshold in both channels. (B) A green and red image pair (Green and Red channel) was used for image randomization, creation of a colocalizatiand subsequent p-value calculation. A set of 200 randomized images was created from the green channel image (randomized green image is one exout of 200). Co-localizing pixels are shown as a white overlay on the green and red channel merge (Colocalization map). (C) Plot of the distributioPearsons coefcients (PCs) of randomized images (curve) and of the green channel image (red line). The red line indicates the PC and the curve s

probability distribution of the PCs of the randomized images. Note that the p-value for this analysis was 100% indicating a high probability of colocalization. (D) Range of PCs obtained from randomized images (grey bars, mean valueSD) compared with the PC obtained for the initial set of images(red lines) in cases of complete colocalization events (a) with different levels of noise added (an0 an4), different intensities (b), partial colocalization (c) andexclusion (d). TheP-values were 100% for (ac) and 0% for (d).


13/20



whereas the peak is still atx=0 (Fig. 7B). Partially overlappingstructures show a peak aside of x =0 (Fig. 7C). Structuresthat are mutually excluded from each other show a dip ax =0 (Fig. 7D).

The cross-correlation function allows ready discriminatiobetween the different colocalization events. However, it hthe major drawback that it is only valuable for small anisotropic particles, as it may vary depending on their orientatiorelative to the selected shift axis. The cross-correlation functicalculation allows an estimation of the dimensions of thparticles, as the width of the bell-shaped curve at half maximureects the approximate particle size convolved by the PSFthe optical system.

Lis approach. The work of Liet al. (2004) is of particular interestin the search for an interpretable representation of colocalizatioto discriminate coincidental events in a heterogeneous situationThey rst assumed that the overall difference of pixel intenties from the mean intensity of a single channel is equal to ze

and with the upper-casecharacter being the current pixels intensity and the lower-cascharacter being the current channels mean intensity. As aconsequence, the product of the two equalities should tento zero. Now if we consider colocalizing pixels this prodshould be positive as each difference from the mean is of tsame sign. The differences of intensities between both channare scaled down by tting the histogram of both images to a 1 scale. The intensity correlation analysis results are thepresented as a set of two graphs, each showing the normalizintensities (from 0 to 1) as a function of the product (Ai

a)(Bi b) for each channel (Fig. 8). In this representation thx-axis reects the covariance of the current channel and the yaxis reects the intensity distribution of the current channeAs previously stated, in the case of colocalization the produ(Ai a)(Bi b) is positive and therefore the dot cloud is mostconcentrated on the right side of the x=0 line, althoughadopting a C shape (Fig. 8A, A* and E). Its spread is dependenthe intensity distribution of the current channel as a function o

=n pixels iA a( ) 0 =n pixels iB b( ) 0

Fig. 7. Colocalization analysis with JACoP; Van Steensel. (AD) Crocorrelation functions (CCFs) were calculated (with a pixel shift = 20) for complete colocalization (A), complete colocalization widifferent intensities (B), partial colocalization (C) and exclusion (Completely colocalizing structures peak at =0 (A), even if differentintensities of the two uorescent channels are present (B). Partiallcolocalizing structures show a shift away from 0 in the maximum of thCCF (C). When the region of interest is quite crowded, shifting one imwith respect to another may enhance the probability of obtainingcolocalization, therefore slightly increasing the Pearsons coefcie(arrowheads). Exclusion of structures leads to an inversion of the CCwhich shows a dip around =0 (D). (E) Effect of random noise (n1n4) othe CCF in comparison to A=n0. Random noise results in a decrease of the maximum while full width at half maximum increases; it is stipossible to identify the colocalization event.


14/20



Fig. 8. Colocalization analysis with JACoP; Li. (AD) Intensity correlation analysis (ICA) was performed for complete colocalization (A and A*)

colocalization with different intensities (B), partial colocalization (C) and exclusion (D). (AD) ICA of the green channel; (A*) and insets of (BDred channel. The x-value is dependent on covariance of both channels and the y-value reects the intensity distribution of the current channel. Pixevalues situated left of the x=0 line do not colocalize or have inversely correlated intensities, whereas pixels situated on the right side colocalize (seedetails). The horizontal line indicates the position of the mean intensity of the current channel allowing the visual estimate of the spread of intdistribution with respect to the mean value. (A and A*) Complete colocalization results in a C-shaped curve on the right side of both graphs. The adrandom noise leads to the expansion of the C-shaped curve (A and A*, insets, grey dots). (B) In the case of complete colocalization with different ithe pixel cloud is shifted up or down the ordinate axis, with most pixels situated on the positive side of the graph. (C) Partial colocalization results ivaluable information as the minority of colocalized pixels fail to form a strong identiable dense cloud. (D) Exclusion of the uorescent signals rpixel distribution with most of the pixels found on the left side of the plot. Pixels with low intensities that are found on the right side are due to noisF) Intensity correlation quotient (ICQ) values, which are dependent on the proportion of pixels on the left side of the x=0 line to the total number of pixels,are plotted for compete colocalization events (a) with different levels of noise added (an0 an4), different intensities (b), partial colocalization (c) andexclusion (d).


15/20



the covariance of both channels intensities. This becomesclearer when adding random noise to the completely colocalizingimages. Compare the C-shaped curve of complete colocalization(Fig. 8A and A*) with the expanded curve when noise is added(Fig. 8A and A*, insets). Note that the addition of noise mayalso result in the spread of dots to the left side of the graph. Inthe case of complete colocalization with different intensities,the pixel cloud in the red channel is shifted up the ordinateaxis (Fig. 8B). Non-colocalizing pixels are found on the left sideof the plot. Partial colocalization spreads the pixel cloud withinthe right side of the plot (Fig. 8C). Mutual exclusion of theuorescent signals results in a pixel distribution with most of the pixels found on the left side of the plot (Fig. 8D). Pixels withlow intensities that are found on the right side are due to noiserandomly coincident between the two channels.

For random distribution of uorescent signals, badly decon-volved images or, in the case of high contamination by noise, arather symmetrical hourglass-shaped distribution of dots isobserved (Fig. 8E). In these cases, the result is quite difcult tointerpret and therefore the intensity correlation quotientmight be calculated. This is dened as the ratio of positive (Ai a)(Bi b) products divided by the overall products subtractedby 0.5. As a consequence, the intensity correlation quotientvaries from 0.5 (colocalization) to0.5 (exclusion), whereasrandom staining and images impeded by noise will give avalue close to zero (Fig. 8E and F). The development of thisgraphical method interpreting image sets based on theirrespective intensities is a step forward compared with thepreviously described scatter plots as it allows a direct identi-cation of colocalization and exclusion. However, it is still a

global method that does not allow conclusions in intermediatecases.

Object-based analysis

The main disadvantage of the ICCB tools introduced so far isthat no spatial exploration of the colocalized signal is possible.All methods previously described rely on individual pixelcoincidence analysis, considering that each pixel is part of theimage and not part of a unique structure. Although giving aglobal estimation of colocalization, their numerical indicatorssuffer from the composite nature of the images, which is apatchwork of both structures and, even though minimized,background.

There are several possibilities for measuring and evaluatingsubcellular structures by object-based approaches. The methodsdepend on the nature of the colocalization event but alsoon the size, form and intensity distribution of the uorescentsignal. Concerning the nature of colocalization situations, wehave to distinguish between those with two markers occupyingthe same space on all subcellular structures (complete colo-calization, such as Fig. 4A) or on some subcellular structures(partial volumetric colocalization, such as Fig. 4C) and betweenincomplete colocalization situations with two markers

overlapping partially on all or some subcellular structure(partial topological colocalization, such as in Bolteet al., 2004b).It is recalled that any entity below optical resolution woccupy at least 2 2 =4 pixels (or even 33 = 9 pixels inthe case of sampling at 2.3 pixels per resolution unit) in thtwo-dimensional space so no discrimination can be expectbetween subresolution objects. However, respecting the Nyqusampling criterion, an object may be positioned with an erroof 70 nm (Webb & Dorey, 1995). Biological structures arthree-dimensional and it has already been mentioned that thdiscrepancy between lateral and axial resolution of opticmicroscopes leads to a distortion of the object along the z-axTherefore, object-based analysis needs to be carried out in tthree-dimensional space by taking account of the degree distortion by the optical device.

A method of choice to measure colocalization on structurwith a size close to or larger than the resolution limit anespecially in the case of partial volumetric colocalization relon a manual identication of structures and a subsequenmeasurement of their uorescence intensity curves. This done by drawing a vector through these structures andplotting the uorescence intensities for the green and rechannel against the length of the vector. This can be done iany image software and is basically a line scan through a twdimensional image of a uorescent object, representing thuorescence intensities along a vector traced across thobject. Colocalization is present when the true overlap distanof the uorescence intensity curves at mid-height is largthan the resolution of the objective used for image acquisiti(Fig. 9B). Fluorescence intensity proles of overlappi

subcellular structures should give similar overlap results ithose successive single sections from an image stack reprsenting the two structures and matching the z-resolution othe optical system used. This method has been applied to shothe partial colocalization of plant Golgi stacks and prvacuolar compartments (Bolteet al., 2004b). Although powerfulon colocalization estimation, this method is time consuminand will only be applicable to a limited number of structurespositioning of the vector is interactive. Furthermore, mispotioning of the vector may lead to underestimation of colocazation events. Moreover, this method is likely to work only isotropic, solid structures such as doughnut-shaped or elongatestructures.

One step forward in colocalization quantication relietherefore on its local estimation based on object identicatiand delineation. This challenging area of image processingknown as image segmentation. Although many techniqueexist, we will only describe segmentation procedures that haalready been used for colocalization analysis.

Looking for objects: basic image segmentation. In an optimal situation,pixels deriving from noise should have lower intensities thapixels deriving from structures. A rst step to identifying thstructural pixels as objects may be achieved by applying


16/20



threshold to the image; all pixels with intensities above a limitvalue (threshold) will be considered to be part of an object. Inmost cases, this threshold value may be dened manuallyfollowing visual inspection (Fig. 9C and D). It is also possibleto apply an automatic threshold as we have already seen(Costeset al., 2004). Noise is not fully eliminated as it remains

within structures but at least two main areas are now denedon the image, regions where structures (and noise) are presentand regions where only noise is present.

Although thresholding enables one to distinguish betweenbackground and objects, one more step is required to delineateeach structure. As a rst approximation, the limit of an object

Fig. 9. Object-based colocalization analysis by uorescence intensity proles and connexity analysis. The analysis was performed on grey level impartially colocalizing uorescent structures (as shown in Fig. 4C). (A) Raw images showing partial colocalization of uorescent subcellular strwith green (left panel) and red (right panel) channels. (B) Inset of overlay of raw images as shown in (A) and intensity curves measured along a

across two uorescent structures (white arrow). (C) Magnied view of the inset shown in (B). The segmentation process by connexity analysis reparticle (D) and centroid (E) detection. (F) Nearest-neighbour distance approach by merging green and red channel centroids. Colocalization is when centroids have distances below optical resolution (yellow arrowheads). (G) Merged view of centroids of the green image (E) and particles oimage (D) illustrates the overlap. Note that the overlap method doubles apparent colocalization events.


17/20



may be seen as a sudden variation of the pixel intensities whenperforming a line scan. The rst derivative of this line scan willbe zero as long as the intensities in the background area, orinside a uniformly labelled structure, are almost constant anddifferent from zero when passing from background to object(or from object to background). A new image may be createdusing these values to show enhanced edges. This so-called edgedetection may be achieved by the use of lters that are available inmost common imaging software, namely Sobel and Laplacianlters (Sobel, 1970; Ronot & Usson, 2001). It is, however,important to note that these lters have their limits. Structureswith non-uniform uorescence intensity distribution maylead to an artefactual detection of concentric edges. Moreover,such lters will highlight the outline of the structure but giveno information on the structural content.

Other methods may be used to separate structures frombackground while keeping information on their uorescenceintensities as intact as possible. The rst approach is based onthe topological relationship of adjacent pixels, a step namedconnexity analysis (implied in the three-dimensional objectcounter). Briey, this process consists of systematic inspectionof the neighbourhood (8 pixels in two-dimensions and26 voxels in three-dimensions) of the current pixel (referencepixel); all adjacent pixels with intensities above the thresholdlimit are considered to be part of the same structure as thereference pixel. Each pixel is then tagged with a number, withall pixels of the same structure carrying the same tag. A pixellacking at least one of its neighbours is considered to be at theedge of the structure. This procedure results in two images,one carrying the intensity information (Fig. 9C, raw image)

and the other representing individualized structures (Fig. 9D,particles). This method applies whatever the size and shape of the target structures are and requires noa-priori knowledge of those parameters. In the case where all structures have thesame shape and size, another approach may be used. Thetop-hat lter (Meyer & Beucher, 1990) is a morphological lterthat may be utilized to look for structures matching a preciseshape called the structuring element. The top-hat lter slightlyaffects the pixel intensities but has the advantage of correctinguneven illumination by bringing the foreground intensityinside the structuring element back to the minimum value. Itsselectivity on the structural features implies that part of theinformation may be left aside in the subsequent analysis.By performing connexity analysis or top-hat ltering, thesegmentation of structures may not be perfect. Structuresmay still stick together and may be individualized by a furtherstep called watershed ltering that will split apart the jointstructures by highlighting their common boundaries (for review,see Roerdink & Meijster, 2000).

After segmentation it is possible to determine centroidsand intensity centres from the structures. This process may becarried out automatically in the three-dimensional space (Fig. 9E).Centroids are the geometrical centres of objects including theglobal shape of the structures. Intensity centres take into

account the distribution of uorescence intensity of the objeIn the case of geometrically isotropic structures, both centroiand intensity centres may be coincident but this is not obligatoras uorescence distribution might be anisotropic. The abovmentioned segmentation procedures and the parametersretrieved may be used differentially to estimate the degreeobject-based colocalization of two markers as will be descriin the following.

Looking for coincidence of discrete structures: object-based colocalization. One way to measure colocalization is to comparthe position of the three-dimensional centroids or intensitcentres of the respective subcellular structures of the twcolour channels. Those positions may be displayed in an overlwindow (Fig. 9F) and their respective x, y, z coordinates wthen be used to dene structures separated by distances equto or below the optical resolution. As a consequence, we wconclude that both structures colocalize if their distance below optical resolution. This method has been applied prove the Golgi association of AtPIN1, the plant auxin efcarrier. Two objects were considered to colocalize if the distabetween their centres was less than the resolution of thmicroscope used (Bouttet al., 2006). A similar approach hasbeen used to study the complex formation among membranproteins underlying the plasma membrane of mammaliancells (Lachmanovichet al ., 2003). The authors includedtop-hat ltering and watershed processing to separate smaround-shaped vesicles. After segmentation, centroids wercalculated and the distances between objects from the greeand red channel images were measured. This process wa

called nearest-neighbour distance approach. As the numbeof objects may differ between two channels, the measuremehas to be set to select objects from the channel with fewobjects and to search for the nearest neighbour from thchannel with more objects. The degree of colocalization then calculated from the percentage of objects in the rchannel colocalizing with objects from the second channedivided by the total number of all objects from the rchannel.

Lachmanovichet al. (2003) tested the signicance of thecolocalization results against the degree of colocalization randomized images, produced as already described (Costet al., 2004). The use of randomized images as referencallowing statistical evaluation of the object-based approachindeed a step forward and adds to the validity of the resuHowever, the measurement of centroid distances by the nearesneighbour distance has two main limits. Firstly, the segmentatioprocedures select elements that meet pre-dened criteria. Tmethod is thus restricted to rather isotropic structures andmay lead to under-estimation of colocalization. Structurewith shapes deviating from the pre-xed criterion may bincorrectly discarded. Secondly, the use of centroids to deobjects may result in under-estimation of colocalization dueanisotropic intensity distributions within the structures if th


18/20



objects are larger than the optical resolution or if they differ insize between the two colour channels. The rst case can beruled out by calculating intensity centres rather than centroids.For the second case, Lachmanovichet al. (2003) developedanother approach called the overlap approach; objects in thegreen and red channels colocalize if the centroid of an object of the green channel falls into the area covered by an object of the red channel (Fig. 9G). The degree of colocalization is thengiven by the percentage of green objects colocalizing with redobjects in the area of interest. Counting the number of greencentroids matching red object areas and red centroids matchinggreen object areas resulted in two percentages of overlap.These percentages were compared with a random distributionobtained as described before and thereby allowed a statisticalevaluation of colocalization. The overlap method enhancesthe probability of matching structures, as matching a centroidto an object area is more probable than matching two centroids.This method may work on categories of objects and thereforegives information on a single class of structures rather thangiving an overall estimate of colocalization. By reiterating theanalysis on the same images with differential settings of top-hatltering or other means of segmentation, one may obtaininformation on different classes of objects. We have automatedthe analysis of centroids and intensity centres with the three-dimensional object counter plugin that may be combined withseveral image-segmentation and randomization proceduresto provide a rst step towards multilevel analysis.

Object-based colocalization implying intensity correlation coefcient-based analysis. Jaskolskiet al. (2005) proposed a new repre-

sentation of coincident pixels that has been elaborated afterimage segmentation based on Sobel ltering. As previouslydescribed, a Sobel lter will only highlight the edges of structures,based on detection of rapid intensity variations. The result of this process is a map of edges that will be translated to a binaryimage by lling the area outside the edges with black pixels(intensity=0) and the area inside the edges with white pixels(intensity=1). However, the position of uorescent structuresmay differ from one colour channel to the other. As a consequence,to keep track of both sets of structures, the binary imagesobtained from the green and red channels were combinedusing the Boolean operation OR. This creates a mask encom-passing the relevant structures of both images. By multiplyingthe original green and red image to the mask, the structuresfrom each colour channel were isolated. This step represents aview of the original image through the lled edge map. As aresult, a region of interest only composed of structural pixelspresent in both channels is obtained, which allows explorationof the correlation of both signals within this region of interest.

The correlation image is then calculated using the normalizedmean deviation product (nMDP). In principle this is done usinga modication of the intensity correlation analysis method (Liet al., 2004). The numerator is analogous to the abscissa value(Ai a)(Bi b) (see Correlation analysis based on PC above),

whereas the denominator is used to normalize the nMDP tothe product of differences between maximum (Amax, Bmax) tomean intensity (a, b) of both channels [(Amax a)(Bmax b)].This allows comparison of the values from one set of images tanother.

The numerator of the nMDP is positive for colocalizingpixels as we have previously seen (Liet al., 2004). Jaskolskiet al. (2005) provide a correlation image (nMDP image)designing non-correlated pixels with values between1 and 0with cold colours and correlated pixels with values between 0and 1 with hot colours. A new numerical indicator (Icorr) givesthe fraction of pixels with positive nMDPs.

This method of Jaskolski is of particular interest as itcombines a direct visualization of colocalization with correlationdata. It provides an overall statement based on the global analysisof a region of interest of the image containing the structure.The recapitulative correlation image may help to draw conclusionson structures in a particular region of interest. However, themethod is highly dependent on the applicability of the algorithmand the Sobel ltering. The reliability of the segmentation stepis crucial and has to be faithfully adapted to the structuresinvestigated. Finally, although this method does not offer anydirect statistical validation of the results, as do Costes andLachmanovich, it proposes a differential diagnostic thanks tothe normalization parameter included in nMDP.

Guidelines

We have provided an overview of the most currently usedcolocalization analysis methods. Although not exhaustive, it

points out the advantages and pitfalls of each approach thatthe cell biologist may use. To help in choosing a method, wewill now propose several guidelines for the reader to undertakecolocalization analysis.

To get started, colocalization of rather isotropic structurescan generally be analysed with the method of Van Steenselet al. (1996) thanks to its ability to distinguish betweencolocalization, exclusion and unrelated signals.

In the event of an evident complete colocalization devoid onoise, simple ICCB methods such as Pearsons approach areefcient at obtaining a numerical estimator from the image.Manders coefcients may be calculated simultaneously,keeping in mind that comparison of results between datasetsmay only be applicable if similar acquisition and thresholdingconditions are applied. Pearsons and Manders coefcientsare reliable as long as several sets of images have to be comparedhowever, it is difcult to draw a conclusion from a singledataset. Here, Costes approach using the creation of a randomizedimage is useful to evaluate the correlation coefcients obtainedin comparison to events occurring due to chance, although itmay need more computing time. Subsequent object-based analysiswith centroids or intensity centres will tend to amplify theconclusion because they only take into account that fractionof the image occupied by structures.


19/20


20/20


Jares-Erijman, E.A. & Jovin, T.M. (2003) FRET imaging.Nat. Biotechnol .21 , 13871395.

Jaskolski, F., Mulle, C. & Manzoni, O.J. (2005) An automated method toquantify and visualize colocalized uorescent signals. J. Neurosci. Meth .146 , 4249.

Kluge, C., Seidel, T., Bolte, S., Sharma, S., Hanitzsch, M., Satiat-Jeunemaitre,

B., Ross, J., Sauer, M., Golldack, D. & Dietz, K.-J. (2004) Subcellulardistribution of the V-ATPase complex in plant cells, and in vivo localisa-tion of the 100 kDa subunit VHA-a within the complex.BMC Cell Biol .5 , 29.

Lachmanovich, E., Shvartsman, D.E., Malka, Y., Botvin, C., Henis, Y.I. &Weiss, A.M. (2003) Co-localization analysis of complex formation amongmembrane proteins by computerized uorescence microscopy: applicationto immunouorescence co-patching studies. J. Microsc . 212 , 122131.

Li, Q., Lau, A., Morris, T.J., Guo, L., Fordyce, C.B. & Stanley, E.F. (2004) Asyntaxin 1, Galphao, and N-type calcium channel complex at a presynapticnerve terminal: analysis by quantitative immunocolocalization. J.Neurosci . 24 , 40704081.

Manders, E.M.M. (1997) Chromatic shift in multicolour confocal micro-scopy. J. Microsc . 185 , 321328.

Manders, E., Stap, J., Brakenhoff, G., van Driel, R. & Aten, J. (1992)Dynamics of three-dimensional replication patterns during the S-phase, analysed by double labelling of DNA and confocal microscopy. J. Cell Sci. 103 , 857862.

Manders, E., Visser, A., Koppen, A., de Leeuw, W., van Liere, R., Brakenhoff, G. &van Driel, R. (2003) Four-dimensional imaging of chromatin dynamicsduring the assembly of the interphase nucleus.Chromosome Res . 11 ,537547.

Meyer, F. & Beucher, S. (1990) Morphological segmentation. J. VisualComm. Image Rep . 1 (1), 2146.

Minsky, M. (1961)Microscopy Apparatus . United States Patent3,013,467, December 19, 1961 (led November 7, 1957).

Oppenheim, A.V., Willsky, A.S. & Young, I.T. (1983)Signals and Systems

2nd ed. Prentice Hall, Englewood Cliffs, New Jersey.Pawley, J.B. (1995)Handbook of Biological Confocal Microscopy . Plenum

Press, New York.Rasband, W.S. (19972006) ImageJ. US National Institutes of Health,

Bethesda, MD, U.S.A.http://rsb.info.nih.gov/ij/Roerdink, J.B.T.M. & Meijster, A. (2000) The watershed transform: deni-

tions, algorithms and parallelization strategies.Fundamenta Informaticae ,41 , 187226.

Ronot, X. & Usson, Y. (2001)Imaging of Nucleic Acids and Quantitatio

Documents

0000-2006 a Guided Tour Into Sub Cellular Colocalization Analysis in Light