20066_ftp

7/30/2019 20066_ftp

1/10

Colocalization Analysis Yields Superior ResultsAfter Image RestorationLUKAS LANDMANN* AND PERMSIN MARBETDepartment of Anatomy, University of Basel, CH-4056 Basel, Switzerland

KEY WORDS 3D-imaging; confocal microscopy; image processing; median filtering; deconvolu-tion; signal-to-noise ratio; background; resolution; synthetic objects

ABSTRACT Colocalization analysis is a powerful tool for the demonstration of spatial andtemporal overlap in the distribution patterns of fluorescent probes. In unprocessed images, back-ground affects image quality by impairing resolution and obscuring image detail in the low-intensity range. Because confocal images suffer from background levels up to 30% maximumintensity, colocalization analysis, which is a typical segmentation process, is limited to high-intensity signal. In addition, noise-induced, false-positive events (dust) may skew the results.Therefore, suppression of background is crucial for this type of image analysis. Analysis of syntheticand biological objects demonstrates that median filtering is able to eliminate noise-induced colo-calization events successfully. Its disadvantages include the occasional generation of false-positiveand false-negative results as well as the inherent impairment of resolution. In contrast, imagerestoration by deconvolution suppresses background to very low levels (10% maximum intensity),which makes additional objects in the low-intensity but high-frequency range available for analysis.The improved resolution makes this technique extremely suitable for examination of objects of nearresolution size as demonstrated by correlation coefficients. Deconvolution is, however, sensitive tooverestimation of the background level. Conclusions for practical application are: (1) In raw images,colocalization analysis is limited to the intensity range above the background level. This means thehigher the RS/N the better. Unfortunately, images of most biological specimens have a low RS/N. (2)Filtering improves the result substantially. The reduction of background levels and the concomitantincrease of the RS/N are generated at the expense of resolution. This is a quick and simple methodin cases where resolution is not a major concern. (3) If colocalization in the low-intensity rangeand/or maximum resolution play a role, deconvolution should be used. Microsc. Res. Tech. 64:

103112, 2004.

2004 Wiley-Liss, Inc.

INTRODUCTION

Multichannel fluorescence microscopy makes use ofmultiple fluorochromes of different excitation andemission wavelengths for localizing different probes ina single specimen. The technique allows the demon-stration of spatial and temporal relationships in thedistribution pattern of various probes and has foundwidespread applications, including the tracing of dy-namic cellular processes. Colocalization of signal fromdifferent channels, i.e., the locations in an image wheretwo or more fluorochromes are found simultaneously, isof major interest in this type of work. Background,which is present in variable degrees in any image, is amajor factor restraining the power of colocalizationanalysis. Therefore, it is highly desirable to increasethe signal-to-noise ratio (RS/N), which, in a given image,can be achieved by filtering or image restoration. Thisreview shows that filtering and image restoration bydeconvolution substantially improve performance, ac-curacy, and resolution of colocalization analysis. Filtertechniques enhance the RS/N at the expense of resolu-tion by amalgamation of signal with background andnoise. In contrast, removal of background by imagerestoration allows for colocalization at very low inten-sities and at very high resolution.

Due to its spatial resolution power, confocal micros-copy is extremely well suited for colocalization analy-sis. However, wide-field microscopy in combinationwith appropriate deconvolution procedures is also suit-able and even superior to confocal microscopy in caseswhere optical properties are not the overriding crite-rion, e.g., in work with living cells that are susceptibleto light damage.

WHAT IS COLOCALIZATION ANALYSIS?

Colocalization analysis in digital images is a typicaldual-image pixel point process (Shotton, 1993), be-tween a pair of input images from two different chan-nels that generates a single output image, the colocal-ization map (two images in, one image out). This dataset selects and segments all voxels where both chan-nels have common events defined as signal intensityabove a chosen threshold or within a certain range.

*Correspondence to: Lukas Landmann, Ph.D., Dept. of Anatomy, University ofBasel, Pestalozzistr. 20, CH-4056 Basel. E-mail: [email protected]

Received 13 April 2004; accepted in revised form 14 May 2004

Abbreviations used: MIP, maximum intensity projection; PSF, point spreadfunction; RS/N, signal-to-noise ratio

DOI 10.1002/jemt.20066

Published online in Wiley InterScience (www.interscience.wiley.com).

MICROSCOPY RESEARCH AND TECHNIQUE 64:103112 (2004)

2004 WILEY-LISS, INC.

7/30/2019 20066_ftp

2/10

Therefore, colocalization analysis shares the followingsteps with other image analysis procedures: (1) (Op-tional) definition of a region of interest in order torestrict analysis to a spatially defined region of theimage or to reduce requirements on processing hard-ware. (2) Definition of a threshold intensity value to

separate features of interest from the rest of the imagein each channel. This crucial operation has great influ-ence on the outcome and is normally done interactivelyby adjusting the chosen intensity to a value separatingsignal from background. (3) Segmentation of the inputintensity images to generate an output binary image inwhich all voxels that fail to meet both threshold re-quirements are set to zero intensity and all voxels thatpass are color-coded by a look-up table. In the figures ofthis review, voxels with a colocalization event are con-sistently pseudeocolored at maximum intensity in theadditive color generated by the two original channels(e.g., green and red threshold yellow 255). (4)Computation of quantitative data such as co localizingvoxel number or intensity in each channel. Additionalcomputations can be performed by any tool provided by

an image analysis programs. (5) Finally, the colocaliza-tion map can then be added to the original data set asseparate channel(s) highlighting voxels meeting theconditions chosen for colocalization.

There are various colocalization software modulesavailable that help the user to analyze multichannelimages by displaying a 2D histogram of pixel intensi-ties for every singular image or for whole image stacks(Demandolx and Davoust, 1995). The histogram ar-ranges corresponding pixels from two images of iden-tical size but from different channels according to theirred and green intensities along a green x- and a redy-axis. Thus, every singular pixel of an image is char-acterized by a pair of intensities that can be regardedas coordinates in a Cartesian system (Fig. 1). Analysis

of the distribution pattern of the intensity pairs allowsfor the identification of colocalization as well as dis-crimination of background, crosstalk between chan-nels, fluorescence attenuation, and channel misregis-tration (Demandolx and Davoust, 1995, 1997). It isthen the task of the user to define in the histogram thearea reflecting colocalization by picking an adequateintensity threshold or range. The aim of colocalizationanalysis is, by definition, to include only those eventsgenerated by signal from the specimen and to excludeevents contributed by unspecific background and noise.It cannot be overemphasized that this is the singlemost critical step of the procedure. Choosing too low athreshold includes high background intensities and re-sults in the generation of false-positive colocalization

events, whereas a conservative estimate, i.e., thresh-olding at too high an intensity, misses colocalizationdisplayed in the specimen.

LIMITATION BY BACKGROUND

Colocalization analysis discriminates between sig-nal- and background-generated events by thresholdingat a chosen intensity level. Unfortunately, the image isnot a perfect copy of the distribution of fluorescentprobes in the specimen (Fig. 2). Additional intensitiesoriginating from the effect of a non-spherical pointspread function (PSF) or blur, from single photon hitsor stray light, and from photomultiplier offset or base-

line, all of which are subject to random fluctuations ornoise, add up to background. They impair image qual-ity by obscuring image detail in the low-intensity rangeand by affecting resolution. Because no informationcan be extracted from intensities below the background

Fig. 1. Colocalization softwares analyze topographically corre-sponding voxels in images or image stacks from two channels. Everysingular voxel is characterized by two (e.g., a green and a red) inten-sities that are used for arranging it in a 2D histogram along a greenx- and a red y-axis. Close ups (top) display voxels with their green and

red intensities. The highlighted voxel with an intensity of green168and red135 occupies the encircled position in the histogram (bottom),which gives intensities as percentages. The white area reflects thresh-olding at 25% maximum intensity in both channels. [Color figure canbe viewed in the online issue, which is available at www.interscience.wiley.com.]

104 L. LANDMANN AND P. MARBET

7/30/2019 20066_ftp

3/10

level, the discriminating intensity threshold, ideally,should separate signal from background. In low-con-trast biological specimens, background and noise canreach up to 30% maximum intensity even in confocalimages (Fig. 2). Inclusion of such intensities results inthe generation of unspecific, mainly single-voxel colo-calization events. Therefore, colocalization analysis isrestricted to the high-intensity range in raw images. In

order to extend the intensity range available for anal-ysis, image processing routines that improve RS/Narehighly recommended.

SUPPRESSION OF BACKGROUNDFiltering

High-intensity background is successfully sup-pressed by low pass, median, and Gaussian filters (De-mandolx et al., 1997). Filtering techniques recover avoxel value from the local neighborhood and compute aweighted average (see Fig. 2, intensity profiles). This,inherently, results in loss of resolution because theprocedure convolves the image with a smoothing ker-

nel. Therefore, background is merged with and distrib-uted over signal, which, in extreme cases, can result inartifacts. False-negative (dissipation of signal inten-sity below threshold value) as well as false-positive(merger of background with low-intensity signal re-sulting in above threshold intensities) results can begenerated. A great advantage of filtering techniques istheir ease of handling and their speed. Experience

demonstrates that median filtering can improve RS/N atleast 3-fold (see Fig. 5).

Deconvolution

Deconvolution (Shaw and Rawlins, 1991; Van derVoort and Strasters, 1995) uses the imaging propertiesof the optical system in the form of the point spreadfunction (PSF) for putting the light back where it iscoming from. Imaging of an entire 3D object may bedescribed as the convolution of this object with the PSF(Castleman, 1993). Therefore, the PSF can be used forcalculation of a likely model of the object from therecorded data set in an iterative process. This proce-

Fig. 2. Comparison of imaging modes and their intensity profilesin the same optical section from a data set showing microtubules in aRHCC. The image is presented in unprocessed form (left), after me-dian filtering (center), and after deconvolution (right). The graph(bottom) plots intensity against x-position (white line) for all threeimaging modes. Images are in false colors to show the power of imageprocessing. Structures that in raw images are barely recognizablebecome quite distinct after filtering and perfectly smooth with well-balanced internal intensity gradients after deconvolution. Structural

improvement is paralleled by reduction of background and noise. Notescattered appearance of the raw image profile (open circles) withsingle photon hits (arrows) occurring not only in the low-intensityrange but also distorting signal. Median filtering (blue dotted line)smoothes the profile at the expense of resolution. This is shown bythe highest peak, which has a smaller half-width after deconvolution(red line) than after filtering. Note almost complete suppression ofbackground by deconvolution.

105COLOC ANALYSIS IMPROVED BY DECONVOLUTION

7/30/2019 20066_ftp

4/10

dure eliminates effectively blur caused by distortion. Inaddition, by assuming a Poisson distribution of straylight (Sheppard et al., 1995), it suppresses backgroundto very low levels (Fig. 2). The technique yields imagesof appreciably increased contrast. Background levels atless than 10% maximum intensity increase the RS/N bya factor of at least 5 as compared to raw images. Thismakes additional objects in the low-intensity (but high-frequency!) range available to analysis. In addition,removal of noise and distortions induced by the opticalsystem results in improved resolution, which is of crit-ical importance in cases involving objects of near reso-lution size (see cytoskeleton below). With the incorpo-

ration of deconvolution algorithms into commerciallyavailable software modules (we used here the Huygensmodule, Scientific Imaging B.V., Hilversum, The Neth-erlands), a powerful tool for image restoration has be-come available. Disadvantages of the technique includethe requirement of computing power, the exact estima-tion of background and RS/N (see below) and time.

HOW TO CHOOSE A THRESHOLD?

Discrimination of signal from background is not al-ways easy and intensity distribution plots provided bymany softwares do not help a great deal in most cases.With experience, the thresholds of two or more chan-nels can be chosen interactively with great precision bypicking a threshold for volume or surface rendering

that is then applied to colocalization analysis. A moremethodical way includes the adjustment of all photo-multipliers to their respective channel intensities fol-lowed by the recording of control specimens probedwith one fluorochrome only as a multichannel data set.The histograms of this data set (Fig. 3) show a cloud ofpixels associated with the corresponding axis. Thethickness of the cloud yields the background inter-cept or threshold on the complementary axis. Process-ing of this data set results in threshold values suitablefor processed images. Care must be taken not to missthe seemingly few pixels lying high above the cloud inthe histogram.

EXAMPLESRat Hepatocyte Couplets

Isolated rat hepatocyte couplets (RHCC, Fig. 4)were prepared according to standard procedures(Coleman and Roma, 2000) and exposed for 15 minutesto TxR-conjugated dextran (red), a marker for fluid-phase endocytosis, andasialoglycoprotein linked toCy5 (blue) that labels specifically the lysosomal en-docytic route. The green channel shows microfila-ments probed with phalloidin Alexa488. Like mostbiological specimens, this sample yields noisy imageswith an intermediate RS/N (Fig. 5, left group of bars)that was less than 20 in all channels. Because decon-volution requires the recording of image s withouttruncation of the dynamic range, thus yielding im-ages with less contrast than usual, the differencebetween raw and processed images seems greaterthan usual. Treatment with a 3 3 3 median filterresulted in an improvement by roughly factor 3,whereas the same data set after deconvolution dis-played a RS/N improved 6-fold as compared to rawimages. These data quantitatively support the im-pression that background levels are decreased con-siderably by filtering and image restoration.

Therefore, intensity thresholds can be chosen at lowerlevels after image processing. In the raw image, thresh-olds were selected at 20% maximum intensity, whereas avalue of 15% was chosen for the filtered image and 10%for the restored data set. Colocalization events, indicatedby the pseudo-colors yellow, cyan, and magenta, respec-tively, are illustrated in a single optical section (Fig. 4, toprow). Comparison of the resulting colocalization eventsshows that raw data display channel overlap in largestructures of high intensity, the number and size of whichare increased considerably by median filtering and dras-tically by deconvolution. Whereas an increased number ofcolocalization events is caused by lower threshold set-tings (20 vs. 15 vs. 10% maximum intensity), enhancedimage detail is a benefit of image processing techniquesand the suppression of noise. This is demonstrated by

Fig. 3. Selection of threshold levels for colocalization. Histogramsof 2-channel recordings of calibration specimens stained for the green(left) or red (center) channel only. The width of the pixel clouds,which are associated with the green or red axis, respectively, is a

guideline for threshold selection in the final 2-channel specimen(right). Note that this test does not consider background caused byunspecific binding of primary antibodies. [Color figure can be viewedin the online issue, which is available at www.interscience.wiley.com.]


7/30/2019 20066_ftp

5/10

close-up views of a different optical section from the samedata set (Fig. 4, bottom row). Areas of intermediate in-tensity in raw images display colocalization events con-sisting predominantly of isolated (unconnected) voxels.

Although a conservative threshold above the noise levelapparent in the histograms has been selected, the occur-rence of noise-generated colocalization events is evident.This makes discrimination of artifact from real overlapdifficult. After processing with a median filter and appli-cation of a lower intensity threshold, most isolated colo-calizing voxels disappeared and channel overlap wasclustered into distinct patches. After deconvolution, thedata displayed enhanced image detail and were virtuallyfree of singular voxel events, although thresholds werechosen at even lower intensity. This results in additionalcolocalization events of low intensity that were not de-tected by filtering (Fig. 4, cyan and magenta colocaliza-tion signal). On the other hand, a major part of green-redoverlap (yellow), which is indicated by singular voxels inraw images and patches after filtering, can no longer bedetected after image restoration. This demonstrates thatthe colocalization pattern may differ positively as well asnegatively in a data set before and after image process-ing, although the general trend shows an increase ofcolocalization events in filtered and even more so in de-convolved images as compared to unprocessed data.

Fibroblasts

Rat SM2 fibroblasts (Fig. 6) were grown on coverslipsto half-confluence and were probed after fixation for actinwith phalloidin-Alexa488 (green), for microtubules by in-

Fig. 5. RS/N was calculated as ratio of maximum intensity dividedby average background as estimated by the Huygens software in anon-object volume (radius 0.5 m, axial size 0.3 m) of the data set,because the more accurate method of Sheppard (Sheppard et al.,1995), which is more adequate in low-contrast confocal images, cannotbe applied to processed images. The groups of bars at the left-handside are from the RHCC image presented in Figure 4, those to theright from the fibroblast in Figure 6. Note differential increase of R

S/N

after image processing (see text). Green, red, and blue channels areindicated as G, R, and B, respectively.

Fig. 4. Isolated rat hepatocyte couplets (RHCC) were incubated for15 minutes in medium containing 7.5 mg/ml dextran conjugated to TxR(red), a marker for fluid-phase endocytosis, and 0.01 mg/ml asialoglyco-protein linked to Cy5 (blue) that labels specifically the lysosomal endo-cytic route. After fixation and permeabilisation, cells were probed withphalloidin-Alexa488 (green) for microfilaments. A singular optical sec-tion (top) isshownas raw image (left), after filtering (center), and afterdeconvolution (right). Colocalization thresholds were selected at 20%

maximum intensity in the raw image, at 15% for the filtered, and at 10%for the restored data set. Colocalization maps are color-coded in additivecolors at maximum intensity. Bottom row: Close-ups from a differentimage of the same stack. Unconnected colocalizing voxels in the rawimage (overlap of green and red yellow) become patches after process-ing with a median filter. Deconvolution clearly displays overlap of greenand blue (cyan) and red and blue (magenta), which is not detected in thefiltered image.


7/30/2019 20066_ftp

6/10

direct immunofluorescence (Cy5, blue), and for interme-diate filaments with an antibody against vimentin that

has been conjugated to Cy3 (red). This specimen has asimilar contrast as the former one with a RS/N between10 and 20 in raw images (Fig. 5, right group of bars).Filtering improves contrast approximately 6-fold to levelstwice as high than those in RHCC, whereas deconvolu-tion results in a 2-fold increase as compared to filteredimages. Volume rendering by simulated fluorescence pro-cess (Fig. 6, top row) demonstrates that image processingresults in increased RS/N. Similar to RHCC, in unproc-essed images colocalization events (Fig. 6, bottom row,close-ups from a single optical section) are made up to alarge proportion of singular voxels the number of which isnot altered substantially after filtering but decreaseddrastically after deconvolution. In this specimen, how-ever, colocalization analysis of deconvolved images yields

results different from those in RHCC although thresholdswere lowered to a similar degree (30 vs. 20 vs. 10% max-imum intensity): Whereas channel overlap is increasedafter filtering, deconvolution results in a marked de-crease. This is shown by the close-up views that display aseemingly unambiguous colocalization in raw images be-coming coarser after filtering and disappearing over wideareas after deconvolution.

Quantitative Differences

Voxel Number, Sum of Intensities, and Numberof Objects. Quantitative data from cytoskeletal ele-ments in fibroblasts differ markedly from those obtained

in RHCC. Whereas the number of colocalizing voxelsshows only slight alterations after filtering and a strong

decrease after deconvolution, total intensities are de-creased slightly after filtering and strongly after decon-volution. In contrast, object numbers show a decreasesimilar to that in RHCC. This discrepancy is explained bythe size of objects displaying colocalization: Structures inthis specimen are smaller than the resolving power of theinstrument. Because F-actin (green, 6 nm), vimentin (red,10 nm), and microtubules (blue, 24 nm) are smaller thanthe sampling unit (voxel size 50 50 100 nm), theprobability is high that cytoskeletal elements occur in thesame voxel accidentally without necessarily being associ-ated. This occurrence generates false-positive colocaliza-tion events that even deconvolution cannot suppress. If,however, objects are present in closely associated voxels,the probability is high that straylight generates false-

positive events in raw and filtered but not in deconvolvedimages.Channel Correlation. Channel correlation func-

tions compute a number of coefficients that character-ize the degree of overlap between channels in an image.Pearsons linear correlation coefficient (Gonzalez andWintz, 1987; Manders et al., 1993) can be used toestimate the overlap of pixels. Its value is between 1and 1. A value of 1 represents perfect correlation, 0 nocorrelation, and 1 perfect inverse correlation. Be-cause average intensity in both channels is a factor, itis independent of image background. In contrast to the2D histogram, the coefficient does not take into account

Fig. 6. Rat SM2 fibroblasts were probed after fixation for actinwith phalloidin-Alexa488 (green), for microtubules by indirect immu-nofluorescence (Cy5, blue), and for intermediate filaments with anantibody against vimentin that has been conjugated to Cy3 (red).Unprocessed data (left), filtered (center), and deconvolved (right)images are presented with their colocalization maps in the simulated

fluorescence process (top row). Singular optical sections (bottomrow) show that irregular overlap of raw images is clustered by filter-ing but not always detected by deconvolution. This is caused by anobject size that is smaller than the resolution limit of the instrument(see text).


7/30/2019 20066_ftp

7/10

spatial relations. Therefore, it can be calculated in theentire data set volume or in the colocalized volumeonly. Figure 8 shows that channel correlation differsmarkedly in the two 3-channel images presented inFigures 4 and 6. In RHCC, correlation of the 3 channelpairs in the colocalizd volume decreases after filteringconsistently and, after deconvolution, increases in the

red-blue pair only. This demonstrates that dextran(red) and ASGP (blue) are the only channels displayinggenuine overlap in endocytic vesicles, whereas associ-ation of these probes with actin (green) is coincidental.In contrast, channel correlation of cytoskeletal ele-ments in fibroblasts is lower in raw images, increasedconstantly after filtering, and decreased after deconvo-lution to coefficients lower than those in raw images.Whereas the filtering-induced increase is due to dissi-pation of signal over more voxels generating morefalse-positive overlap, the decrease associated with de-convolution reflects the higher resolution of thismethod resulting in less false-positive overlap. Thus,channel correlation, too, demonstrates the superiorresolution of image restoration techniques as compared

to filtered or raw data.Effects of RS/N and Background on Deconvolu-tion. Although our data clearly demonstrate that decon-volution results in an improved resolution, it should beborne in mind that this technique depends crucially onthe settings of RS/N and background (Landmann, 2002).Over- and underestimation of both parameters alters theresults seriously. Overestimation of RS/N and backgroundcauses distortions that are greater than those caused byunderestimation, with overestimation of background re-sulting in the worst alterations in low-contrast speci-mens. This means for practical applications that correctestimation of background is more important than that of

Fig. 7. Effect of filtering and deconvolution on number (top), sum ofintensities (center) of colocalization events (voxels), as well as on num-ber of objects (bottom) in each channel pair (GR green and red; GBgreen and blue; RB red and blue) of the data sets shown in Figure4 (RHCC) and Figure 6 (fibroblasts). Values are normalized to rawimages as 100%. Numbers of colocalizing voxels and their intensities areincreased moderately after median filtering and strongly after deconvo-lution in RHCC whereas in fibroblasts intensities and voxel numbersshow little change after filtering, but are decreased after deconvolution.This discrepancy is explained by the size of colocalization events, whichin RHCC are larger and in fibroblasts smaller than the resolution limit.Consequently, image processing, which allows the recruitment of voxels

in the low-intensity but high-frequency range, results in an increase inRHCC. In contrast, overlap of cytoskeletal elements in fibroblasts, whichis due to diffraction-caused distortion, stray light, and blur, is apparentin raw and filtered images. Deconvolution avoids this artifact by itssuperior resolution power and its ability to suppress background. Con-sequently, voxel and object numbers as well as intensities are lowest inthis imaging mode. Analysis of object numbers, with values normalizedto objects 1 voxel (connected filled bars) in raw images, shows thatcolocalization maps of raw images contain many singular voxels (openbars) displaying an event that reflects high-intensity noise althoughthresholds were chosen conservatively. This artifact is suppressed to asatisfactory degree by filtering and almost completely by deconvolution.In RHCC, object numbers are decreased by filtering to a level that is notdecreased further by deconvolution. In contrast, object numbers in fibro-blasts that contain predominantly colocalization events of subresolutionsize are decreased additionally by deconvolution.

Fig. 8. Channel correlation coefficients in colocalized volume ofRHCC (left) and fibroblasts (right). In RHCC, correlation of the3 channel pairs decreases after filtering. After deconvolution, an in-crease is displayed only by the red-blue pair. This demonstrates agenuine overlap of dextran (red) and ASGP (blue) both of which areendocytic markers. Association of these probes with actin (green) iscaused predominantly by image defects as demonstrated by coeffi-

cients decreasing with increasing image quality. Cytoskeletal ele-ments in fibroblasts show coefficients that in raw images are lowerthan in RHCC. The consistent increase after filtering is caused bydissipation of signal over more voxels. All channel pairs show adecrease after deconvolution to values lower than those in raw im-ages, which indicates that colocalization effects are due predomi-nantly to image defects.


7/30/2019 20066_ftp

8/10

RS/N and that a conservative approach is less likely tointroduce artifactual alterations.

Synthetic Experiments

The conclusion that filtering but not deconvolutionimpairs signal by the merger of background and noiseis not easy to demonstrate. Objects yield distorted im-ages that do not faithfully reflect the real specimen. In

contrast, computer-generated objects have knownproperties and can be compared with their imagesafter appropriate processing. Therefore, virtual exper-iments were conducted (Huygens software package,SVI B.V., Hilversum, The Netherlands). Syntheticspheres with appropriate excitation and emissionwavelengths were generated in a matrix consisting of50 50 100 nm voxels, i.e., with a sampling density

Fig. 9. Virtual experiments with synthetic spheres. Computer-gen-erated spheres (left) with a radius of 800 nm (top row, 2D histograms)and 50 nm (middle row, colocalization highlighted in yellow) wereconvolved with a synthetic PSF followed by degradation with artificialPoisson noise (raw images, center left). Raw images were processed witha median filter (center right) or restored by deconvolution (right). Thegraph (bottom left) shows the number of voxels (black bars) and sum ofintensities (grey bars) of the 800-nm sphere. Colocalizing volume of50-nm spheres is shown in the graph at bottom right. The larger sphere(top) is characterized by two channels in perfect register that give rise toa diagonal line in the histogram (synthetic image). Artificial distortionand noise (raw image) widen the line slightly and give rise to twonoise-induced accompanying lines, 10% of maximum intensity (noise

level) apart. These lines are suppressed by filtering (median), a processdissipating not only background but also peak intensities of the object asdemonstrated by the lack of signal in the upper right-hand corner of thehistogram. Deconvolution, finally, is also able to remove background andyields a thinner and continuous line of overlap. A sphere with a radius of50 nm, the channels of which were shifted by 100 nm along the x-axis,shows no colocalization (yellow) in the synthetic image (center left) whileoverlap is present in degraded raw images, filtered and deconvolved datastacks. Colocalizing volume is greatest in filtered and smallest in decon-

volved images demonstrating that at near-resolution conditions, decon-volution is not able to restore the original situation completely. Thetechnique, however, has the capacity to improve image quality consider-ably.


7/30/2019 20066_ftp

9/10

corresponding to the RHCC and fibroblast specimens.The objects were convolved with a theoretical PSF,degraded by artificial Poisson noise assuming the real-istic values of 100 photons per unit and 10% maximumintensity, and joined to a 2-channel data set.

Images with complete channel overlap were filtered

and restored as before, and analyzed for colocalizationusing a threshold of 50% background-corrected maxi-mum intensity, i.e., half-maximum amplitude width.

The data for spheres with a radius of 800 nm (Fig. 9),which is well above the resolution limit of the instru-ment, demonstrate that synthetic spheres are charac-terized by a volume that compares favorably to theory(98%). In raw images, voxel numbers that are similarto those of convolved but smaller than those of virtual

spheres are obtained. This discrepancy is explained bythe convolution procedure that spreads light and af-fects low-intensity signal more than peak values, thusclipping the spheres at the periphery. Sum of intensi-ties in raw images corresponds fairly well to the addedvalues of convolved spheres and Poisson noise. Voxelnumbers as well as sums of intensities are moderatelyincreased after filtering and decreased after deconvo-lution, an effect of the constant threshold values thatprevent the recruitment of low-intensity signal in pro-cessed images. In contrast to raw images, filtered datasets show intensity maxima that are 10% lower, adifference that is slightly less than the chosen back-ground level. This demonstrates that removal of back-ground by filtering is achieved by dissipation of one-voxel noise over larger volumes, a process that inher-

ently decreases absolute maximum values. However,this is not true for deconvolution, which is independentof the original intensities and distributes its resultsover the full dynamic range.

Resolution power was tested on spheres of subreso-lution size with a radius of 50 nm, the two channels ofwhich were shifted by twice the radial distance alongthe x-axis in order to get apposed but not overlappingspheres. Using a threshold at half-maximum intensity,colocalization directly reflects impaired imaging qual-ity. Figure 9 shows that synthetic objects display nooverlap, whereas both convolution with a PSF andbackground sum up to substantial overlap in raw im-ages. Deconvolution improved this defect by more than50%, whereas filtering resulted in images of distinctly

lower resolution power. It should be emphasized that

Fig. 10. Endocytosis of dextran in RHCC. A 5-minute pulse of thefluid-phase marker dextran-TxR (red) (7.5 mg/ml medium) was fol-lowed by a maker-free chase of various time intervals. After fixation,specimens were immunostained for asialoglycoprotein- and immuno-globulin A-receptors (ASGP-R, pIgA-R). Panels show a singular opti-cal section (left) and a surface rendering (right). Endocytosis ofdextran for short time intervals (2.5 minutes, top) labels predomi-nantly a tubulo-vesicular network subjacent to the basolateral mem-brane domain. After prolonged endocytosis (30 min, center and bot-tom), dextran becomes concentrated in vesicles close to the nucleusand the apical membrane. Quantitative analysis of dextran-positive

volume (graph at bottom) demonstrates a rapid uptake during thefirst 2.5 minutes, which is followed by a plateau up to 10 minutes and

a subsequent decrease. After short time intervals, both receptors(blue) display a similar colocalization pattern with dextran (magenta)at the basolateral membrane and in the tubulo-vesicular network(top, pIgA-R). This pattern becomes weaker after longer time periodsand is changed in a receptor dependent mode. Overlap of ASGP-Rwith dextran is found predominantly in perinuclear vesicles (center),whereas colocalization with pIgA-R is shifted to the apical cell pole(bottom). This is paralleled by quantitative data (graph at bottom)that show that overlap of dextran and both receptors increases duringthe pulse phase and decreases thereafter. Differences between thereceptors include a slightly later peak for pIgA-R and different volumefractions after longer (15 minute) time intervals. Different kineticsreflect the quick internalization and recycling of ASGP-R, which me-diates uptake of compounds destined for lysosomal degradation, andthe vectorial itinerary of pIgA-R that stays associated with its ligandduring the transcytotic transport of pIgA, respectively.


7/30/2019 20066_ftp

10/10

in borderline situations even deconvolution is not ableto restore the original situation as shown by thesmaller but still persisting overlap. The data demon-strate that in high-resolution work, deconvolution hasthe capacity to improve image quality to some degreebut not completely.

Endocytosis in RHCCColocalization analysis in combination with deconvolu-

tion allows not only the elucidation of 3D relationships instatic specimens, it is also a powerful tool for the exami-nation of dynamic processes as well. In order to charac-terize endocytic pathways in RHCC, a 5-minute pulse ofthe fluid-phase marker dextran-TxR was applied fol-lowed by a maker-free chase of various time intervals.After fixation, specimens were immunostained for asialo-glycoprotein- and immunoglobulin A-receptors (ASGP-R,pIgA-R). Internalization of dextran for short time inter-vals (2.5 minutes) resulted in the labeling of a tubulo-vesicular network close to the basolateral membrane do-main (Fig. 10). After prolonged endocytosis (30 minutes),dextran became concentrated in vesicles close to the nu-

cleus and the apical membrane. Quantitation of dextran-positive volume showed a rapid uptake during the first2.5 minutes, which was followed by a plateau up to10 minutes and a subsequent decrease. ASGP-R medi-ates uptake of compounds destined for lysosomal degra-dation (Spiess, 1990; Steer and Ashwell, 1990) and has asteady-state distribution in the Golgi as well as in andbeneath the basolateral plasma membrane. Initially, itcolocalized with dextran at the basolateral membraneand in the tubulo-vesicular network. A shift into perinu-clear vesicles became prominent with progressing time.This is consistent with the view that ASGP-R dissociatesfrom its ligands gradually during its transport from theearly sorting to the late endosomal compartment and isrecycled back to the cell surface. In contrast, pIgA-R,

which mediates the transcytotic delivery of IgA into bile(Brown and Kloppel, 1989) and is expressed in variousvesicular compartments distributed throughout the cyto-plasm, showed colocalization patterns with dextran sim-ilar to ASGP-R at early time points. After longer timeintervals, overlap disappeared at the basolateral cell poleand became concentrated in an apical tubulo-vesicularnetwork. Quantitative analysis of overlap of dextran andboth receptors showed an increase during the pulse phasefollowed by a sharp decrease. The two receptors differedin their kinetics with pIgA-R peaking slightly later thanASGP-R and in the volume fraction preserved afterlonger (15 minute) time intervals. This discrepancy re-flects differences in the itinerary and rate of the receptorsand agrees well with the extremely high affinity of

ASGP-R for its ligands.CONCLUSIONS

The data show that image processing by medianfiltering and deconvolution successfully suppressesbackground and noise induced single voxel colocaliza-tion events that clutter raw images. Median filteringallows for the selection of lower thresholds by dissipa-tion of high-intensity background. The inherent atten-uation of signal may lead to failures in detecting smalland/or low-intensity objects as well as to false-positiveresponses if small objects of high intensity are exam-ined. Image restoration by deconvolution is able to

suppress noise more effectively than filtering, thusmaking low intensities available for analysis as well assmall objects. This results in a greater intensity rangeand superior resolution.

A take-home message can be summarized as follows:

In raw Images colocalization analysis requires ahigh RS/N and yields no information in the low-inten-sity range. Because most biological specimens do notbelong to this category, results will benefit from im-age processing techniques.Median filtering recovers voxel values from thelocal neighborhood and computes a weighted aver-age. The technique improves the results by decreas-ing background levels, thus increasing the RS/N. Thisadvantage is associated with loss of resolution. Fil-tering, therefore, is a quick and simple method forstudies not primarily concerned with high resolution.Deconvolution restores the effects of factors result-ing in impaired image formation and removes back-ground almost entirely. This method is indicated forcolocalization analysis at maximum resolution

and/or in the low-intensity range. It involves, how-ever, efforts in image acquisition and processing.

ACKNOWLEDGMENTS

We gratefully acknowledge the skillful technical as-sistance of Mireille Toranelli, Jean-Paul Boeglin, andEva Bryson.

REFERENCES

Brown WR, Kloppel TM. 1989. The liver and IgA: Immunological, cellbiological and clinical implications. Hepatology 9:763784.

Castleman KR. 1993. Resolution and sampling requirements for dig-ital image processing, analysis, and display. In: Shotton DM, editor.Electronic light microscopy. The principles and practice of video-enhanced contrast, digital intensified fluorescence, and confocalscanning light microscopy. New York: Wiley-Liss. p 7193.

Coleman R, Roma MG. 2000. Hepatocyte couplets. Biochem Soc Trans

28:136140.Demandolx D, Davoust J. 1995. Multicolor analysis in confocal immuno-fluorescence microscopy. J Trace Microprobe Tech 13:217225.

Demandolx D, Davoust J. 1997. Multicolour analysis and local imagecorrelation in confocal microscopy. J Micr 185:2136.

Demandolx D, Barois N, Davoust J. 1997. Guidelines for multifluo-rescence confocal imaging: Acquisition, processing and display. EurMicrosc Anal 7:57.

Gonzalez RC, Wintz P. 1987. Digital image processing. Boston, MA:Addison Wesley Publication Company.

Landmann L. 2002. Deconvolution improves colocalization analysis ofmultiple fluorochromes in 3D confocal data sets more than filteringtechniques. J Microsc 208:134147.

Manders EMM, Verbeek FJ, Aten JA. 1993. Measurement of co-localization of objects in dual-colour confocal images. J Microsc169:375382.

Shaw PJ, Rawlins DJ. 1991. The point spread function of a confocalmicroscope: its measurement and use in deconvolution. J Microsc163:151165.

Sheppard CJR, Gan X, Gu M, Roy M. 1995. Signal-to-noise in confocalmicroscopes. In: Pawley JB, editor. Handbook of confocal micros-copy, 2nd ed. New York: Plenum Press. p 363371.

Shotton DM. 1993. An introduction to digital image processing andimage display in electronic light microscopy. In: Shotton DM, editor.Electronic light microscopy. The principles and practice of video-enhanced contrast, digital intensified fluorescence, and confocalscanning light microscopy. New York: Wiley-Liss. p 3970.

Spiess M. 1990. The asialoglycoprotein receptor: a model for endocytictransport receptors. Biochemistry 29:1000910018.

Steer CJ, Ashwell G. 1990. Receptor-mediated endocytosis: mecha-nisms, biologic function, and molecular properties. In: Zakim D,Boyer TD, editors. Hepatology. A textbook of liver disease, vol. 1.Philadelphia: W.B. Saunders Co. p 1:137182.

Van der Voort HTM, Strasters KC. 1995. Restoration of confocalimages for quantitative image analysis. J Microsc 178:165181.


Documents

20066_ftp