Journal of Microscopy, Vol. 169. Pt 3 , March 1993, pp. 375-382.

Received 1 August 1992; accepted 4 January 1993

Measurement of co-localization of objects in

dual-colour confocal images
E. M. M. MANDERS*, F. J . V E R B E E K t S & J . A . ATEN*
*Laboratoryfor Radiobiology, University of Amsterdam. AMC, FO-2 12, Meibergdreef 9, 1105 AZ
Amsterdam, The Netherlands
fDepartment of Anatomy and Embryology, University of Amsterdam, AMC, K2-266,
Meibergdreef 9, 1105 AZ Amsterdam, The Netherlands
#Pattern Recognition Group, Delft University of Technology, PO Box 5046, 2600 G 4 Delft,
The Netherlands

Key words. Co-localization, confocal microscopy, image analysis,

image reconstruction, correlation, double labelling.

Summary memory arrays (Dean et al.. 1990). a dual-colour image.

One of the areas of special interest in dual-colour imaging is
A method to measure the degree of co-localization of objects
the frequency of coincidental objects-the degree of
in confocal dual-colour images has been developed. This
co-localization. For example, if the localization of the
image analysis produced two coefficients that represent the
fluorochromes represents the distributions of different
fraction of co-localizing objects in each component of a
biochemical processes, it is of interest to analyse whether
dual-channel image. The generation of test objects with a
these processes take place at the same spot or are not
Gaussian intensity distribution, at well-defined positions in
correlated.
both components of dual-channel images, allowed an
Due to the recent development of dual-colour confocal
accurate investigation of the reliability of the procedure.
microscopy, co-localization studies of objects or processes in
To do that, the co-localization coefficients were determined
doubly labelled specimens have become feasible (Akner et
before degrading the image with background, cross-talk
al., 1990; Draeger et al., 1990; Kitamura et al., 1990;
and Poisson noise. These synthesized sources of image
Mossberg et al., 1990; Sato et al., 1990; Stamatoglou et al.,
deterioration represent sources of deterioration that must
1990; Fox et al., 1991; Merdes et al., 1991). Akner et al.
be dealt with in practical confocal imaging, namely dark
(1991) presented a method to visualize the total overlap of
current, non-specific binding and cross-reactivity of
two patterns by subtracting one component of an image
fluorescent probes, optical cross-talk and photon noise.
from the other. This method provides a clear representation
The degraded images were restored by filtering and cross-
of complete overlap. A more quantitative method of
talk correction. The co-localization coefficients of the
measuring the degree of co-localization can be realized by
restored images were not significantly different from
using binary masks of both components of a dual-colour
those of the original undegraded images. Finally, we
image (Lynch et aI., 1991). A method that does not use
tested the procedure on images of real biological
global intensity thresholding procedures is based on the use
specimens. The results of these tests correspond with data
of Pearson's correlation coefficient (Manders et al.. 1992).
found in the literature. We conclude that the co-localization
By calculating the correlation between both grey values of
coefficients can provide relevant quantitative information
the voxels, an indication of the amount of overlap can be
about the positional relation between biological objects or
obtained. However, the interpretation of this quantity is not
processes.
unambiguous.
In this paper we introduce an alternative method, the
Introduction
calculation of two different quantities which are required to
Dual-colour confocal microscopy is based on the simul- express the fraction of co-localizing objects in each
taneous or successive detection of two fluorochromes in the component of a dual-colour image, named the
same specimen (Amdt-Jovin et al., 1990; Carlsson, 1990; co-localization coefficients. The method was tested on
Mossberg & Ericsson, 1990). The data of the detected several synthesized images to obtain insight into the
signals are stored in three-dimensional (3-D) dual-channel significance of the quantities of the co-localization

[c 199 3 The Royal Microscopical Society 375

376 E. M . M . M A N D E R S . F . J . VERBEEK & J . A . A T E N

coefficients. The influence of image deterioration and image Ri = a . ~ ;and Gi = b . ~ : . This independence is a necessity
restoration on the result of the procedure was investigated. for all overlap and co-localization measurements. A
Finally, the method was tested on images of biological disadvantage of this method is that the result of the
specimens to determine the utility of the procedure. calculation is not unambiguous because of the strong
influence of the ratio of the number of objects in both
components.
Theoretical background
To cancel out this effect the overlap coefficient r should be
Pearson's correlation coefficient is one of the standard divided into two different coefficients by using for example
procedures in pattern recognition (Gonzalez & Wintz,
1987) for matching one image with another and can be
used to describe the degree of overlap between two patterns. with
It provides information about the similarity of shape
without regard to the average intensity of the signals.
Pearson's correlation coefficient (r,) can be calculated from

and

where Ri and Gi are the grey values of voxel i of the red

component and the green component of a dual-colour
image, respectively. We called the two components of a Using these coefficients the degree of co-localization is
dual-colour image the red component and the green expressed by two separate parameters. In these formulas kl
component since Texas Red (or rhodamine) and FITC are and k2 depend on the sum of the products of the red and the
often used in dual-colour confocal microscopy (Titus et al., green intensities. Therefore, the coefficient kl is sensitive to
1982; Mossberg & Ericsson, 1990). Rave, and Gave, are the differences in the intensity of the green signal (which can be
average values of Ri and Gi, respectively. In Pearson's demonstrated by substituting Gi = b . ~ ; ) . Analogously, k2
correlation the average grey values Rave, and Gave, are depends linearly on the intensity of the red signal.
subtracted from the original grey values. Therefore, the Therefore, based on the split overlap coefficients (Eqs. 4
value of this coefficient ranges from - 1 to 1. The negative and 5), two other coefficients were defined that are not
values of the correlation coefficient are di£Ecult to interpret dependent on the intensities of the signals. These
when the degree of overlap is the quantity to be measured. co-localization coefficients M 1 and M2 are defined by
When the average grey values of the voxels are not
subtracted from the original grey values, a new coefficient
can be defined as the overlap coefficient:

where Ri,coloc= Ri if Gi > 0 and Ri,coIoc= 0 if Gi = 0 and

The value of this coefficient ranges from 0 to 1. The product

(Ri-Gi) in the numerator of the equation brings in a
significant value only when Ri and Gi belong to a voxel of where Gi,coloc= Gi if Ri > 0 and Gi,coloc= 0 if R, = 0. These
one of the co-localizing objects (with Ri > 0 and Gi> 0). coefficients, M I and M 2 , are proportional to the amount of
Thus, the numerator is proportional to the number of co- fluorescence of the co-localizing objects in each component
localizing objects. In the same way, the denominator is of the image, relative to the total fluorescence in that
proportional to the number of (co-localizing and not- component. M I and M 2 , defined as the co-localization
co-localizing) objects in both components of an image. coefficients, can be determined even when the signal
A major advantage of the overlap coefficient is that it is intensities in the two components differ strongly.
not sensitive to differences in signal intensities between the
components of an image caused by different labelling with Material and methods
fluorochromes, photo-bleaching or different setting of the
amplifiers. This can be demonstrated by substitution of Applying the procedures presented in this paper, the
 CO-LOCALIZATION M E A S U R E M E N T 377

Fig. 1. Images containing test patterns of

Gaussian objects. Simulated dual-colour
images were obtained by combining image
(A) with itself and with the other images.
representing the components of a dual-
colour image, e.g. a red and a green
component. The degree of co-localization
differs for each combination.

positional information of a 3-D image disappears because down. All other test images (Fig. IB-1D and IF-11) were
we are summing over all voxels of a 3-D image. This means constructed by copying parts of the images of Fig. 1(A) and
that the result of the co-localization measurement algor- (E). Simulated dual-colour images were obtained by
ithm applied to all data from a 3-D image is equivalent to combining two test images. The test images represented
the result of the algorithm applied to a representative part of the parts of a dual-colour image. e.g. a red and a green
the same 3-D data set (e.g. one optical section of a 3-D component.
image). Therefore, it was acceptable to carry out all
measurements on 2-D dual-colour test images (simulated
Measurement
optical sections). However, for testing the method on real
biological specimens 3-D images were used in the To measure the degree of overlap and co-localization
calculations. For image processing and analysis the between the two components of a simulated dual-colour
software package SCIL-IMAGE was used on a SUN Sparc image, Pearson's correlation coefficient r,, the overlap
workstation. coefficient r and the co-localization coefficients MI and M2
were used, as mentioned in the section on the theoretical
background. The correlation, overlap and co-localization
Test images
coefficients were tested on the synthesized images shown in
In our experiments we used several 2-D test images Fig. 1.
containing patterns of objects with Gaussian intensity
distributions. The number of objects and the spatial
Influence of background, noise and cross-talk
distribution of the objects differed for each image. We
constructed a Gaussian object with a diameter of 12 pixels Images were degraded with background, cross-talk and
full width at half maximum (FWHM). We then generated a Poisson noise to simulate the sources of deterioration that
2 5 6 x 2 5 6 image with 3 6 objects at intervals of 3 2 pixels must be dealt with in practical confocal imaging, namely
(Fig. 1A). Figure 1 ( E ) was constructed by shifting all dark current, non-specific binding and cross-reactivity of
objects of Figure 1(A) 1 6 pixels to the right and 1 6 pixels fluorescent probes, optical cross-talk and photon noise.
 Simulated background caused by 'dark current' of the The S/N was calculated using

I
photo multipliers was added to the original images using
C ( ~ noise)
i,

and
R i , =
~ Ri
~ +~ B~~ @I = 20' log
/ [=i~~,noise -G ~ ) ~ (15)

Cross-talk caused by cross-reactivity of the fluorescent

probes (van den Engh et al.. 1985) and signal leak of the
where Boc is an integer value that represents the uniform optical components (Carlsson & Mossberg. 1992) was
background. simulated by using
In addition to background caused by 'dark current',
which is evenly distributed over the image, there is a further
source of background in confocal images of biological and
subjects, namely background caused by non-specific Gi,cross = (1- P).Gi + a.Ri, (17)
binding of the fluorescent probes. Earlier experiments have
shown that a typical intensity distribution of this type of where rr and p represent the cross-talk factors, and Ri,cross
background has high values at the central region of optical and Gi,c,ossthe new grey values of the image with simulated
sections of biological specimens. This was simulated as cross-talk. For simplicity, we simulated equal cross-talk
follows: factors in both directions, i.e. a = a.
Reconstruction
and The reliability of co-localization measurement was investi-
gated for several simulated dual-colour images. For this
purpose first the coefficients of co-localization were
with calculated and the images were degraded with uniform
and non-uniform background, cross-talk and noise. We first
added 6% uniform and 4% non-uniform background and
then simulated cross-talk with cross-talk factors rr and
equal to 8%. Finally, we added photon noise with an SIN of
5 dB, which corresponds to 0.3 photons per grey-value unit.
which is an intensity distribution that has its maximum in These values are representative of real confocal images.
the middle of the image and decreases linearly to zero at the From these degraded images we tried to determine the
borders of the image. The amount of background was coefficients of co-localization again.
expressed as the ratio of the average grey value of the added Subsequently, the reliability of image reconstruction
background and the maximum value of the signal. techniques was investigated. To do so, the degraded images
To measure the influence of noise on the co-localization were restored optimally, without taking into account the
coefficients, noise was added to both components of the information on the input values used in the degrading
simulated dual-colour images. Because the major contribu- procedure. Noise was reduced by filtering with a uniform
tion to the noise in a confocal image was assumed to 3 x 3 filter. Background was removed by applying a
originate from quantum noise we used the Poisson 31 x 31 uniform filter on each component of a dual-
distribution of probability: colour image, subtracting the result from the originals and
setting the negative pixels to zero. Finally, 2-D intensity
histograms of the dual-colour images were made. The
angles of the distributions representing background pixels
and with the axes were measured. From these angles the cross-
talk factors a and /? were determined. Using the inverse of
Gi.c. ( G ~ - c ) ( ~ ' ~ )
P{Gi, noise = X ) = e the transformation (Eqs. 1 6 and 17) that was used to
(X.C)! (14)
simulate cross-talk, we corrected for cross-talk as optimally
to simulate an image with noise (Ri,noiseand Gi,noise)In as possible.
these formulas P is the probability that the grey value of the The coefficients of co-localization were then calculated
new image equals X and the number of simulated photons from these restored images. By comparing the results of these
per grey-value unit is represented in the multiplication calculations with those of the undegraded images the reliability
factor C, which influences the signal-to-noise ratio (SIN). of the restoration procedure was assessed.
 CO-LOCALIZATION MEASUREMENT 3 79

Table 1. Pearson's correlation (rp).the overlap coefficient (r) and the co-localization coefficients (MI and M 2 ) performed on a set of generated
dual-colour test images. These images were made by combining the image in Fig. 1(A) (red component) with the other images of Fig. 1
(green component). The co-localization coefficient clearly reflects the fraction of co-localizing objects in the red and green components of an
image.

Number of objects

Figures Red Green Co-localization TP r MI

AA 36 36 36 1.00 1.00 1.00

AB 36 36 27 0.72 0.75 0.75
AC 36 36 18 0.44 0.50 0.50
AD 36 36 9 0.16 0.25 0.2 5
AE 36 36 0 -0.12 0.00 0.00
AF 36 27 9 0.22 0.29 0.2 5
AG 36 18 9 0.30 0.35 0.2 5
AH 36 9 9 0.48 0.50 0.2 5
AI 36 4 3 0.2 3 0.2 5 0.08

Application the co-localization coefficients (M1 and M z ) These

measurements were performed using the combination of
To test the procedure on real biological specimens we used
the image shown in Fig. 1(A) with itself and the other
3-D images of DNA replication patterns in doubly labelled
images of Fig. 1. Pearson's coefficient (Eq. 1 ) provides a
nuclei of V79 Chinese hamster cells (Manders et al.. 1992).
value for the correlation between the intensity distributions
Cells were grown in three tissue culture flasks. At T = 0 ,
of the components, but the overlap of the signals is better
iododeoxyuridine (IdUrd) was added to the medium of all
represented by the overlap coefficient r (Eq. 2). This is
cultures. At T = 5 min the IdUrd was washed away. DNA
illustrated in the case of the combination of Fig. 1(A) and
replicated during this 5-min period contains IdUrd. At
(E). Pearson's correlation coefficient is negative (Table
T = 0, T = 60 min or T = 180 min chlorodeoxyuridine
1: rp = -0.12). which indicates an inverse correlation of
(CldUrd) was added to the medium of one of these three
the signals. The fact that there are no overlapping objects in
cultures. Fiveminutes after the addition of CldUrd the
this example is expressed in a more perceptive way by the
cultures were washed again. During the second pulse label,
overlap coefficient r, which in this case has a value of 0.0.
CldUrd was incorporated in replicating DNA. The cells were
In the other cases shown in Table 1 with an equal number
then fixed and spun down on slides. Applying an immuno-
of objects in each component of an image, the overlap
chemical staining procedure (Aten et al., 1992), cells with
coefficient (r) exactly reflects the fraction of co-localizing
the incorporated deoxyuridines (IdUrd and/or CldUrd) were
objects in the images. When 1 8 objects out of 36 objects
stained with Texas Red and/or FITC, respectively. The
co-localize, the overlap coefficient (r) is equal to 0.5.
fluorescence signals were recorded by CLSM in
However, when 9 out of 9 green objects colocalize with 9 of
2 x 256 x 256 x 32 data arrays. In this way two repli-
36 red objects (the combination Fig. lA/H), the overlap
cation patterns originating from two different periods of the
coefficient also equals 0.5. Consequently, the overlap
S-phase were visualized in the same cell. From each culture
coefficient provides useful information only when the
one cell was analysed. First, the coefficients M1 and M2
number of objects in the red and green components are
were calculated. Then, the images were reconstructed
equal.
(noise reduction, background removal and cross-talk
When the numbers of objects in the two components of
correction) and the coefficients were again calculated.
the image are not equal, the co-localization coefficients M1
and M2 should be used. The co-localization coefficients M I
and M 2 reflect the difference between the above-mentioned
Results
examples (Fig. 1A/C and lA/H). This is illustrated by the
values of the co-localization coefficients which were found
Measurement
to be 0.50 and 0.50 in the case of the combination of Figs.
Table 1 shows the results of measurements of Pearson's l(A)/(C) and 0.25 and 1.00 in the case of the combination
correlation coefficient (rp), the overlap coefficient (r) and of Fig. l(A)/(H), while the overlap coefficient r cannot
 Red value

Fig. 2. The red (A. D and G) and green

components (B, E and H), and the 2-D
grey-value histograms (C, F and I) of a
dual-colour image before image deteriora-

lg2p1
tion (A-C), after deterioration (D-F) and
after image- restoration (G-I).
~, The co-
Red value localization coefficients MI and M2 of the
original image (A, B) equal 1/12 and 314,
256
respectively. After degradation of the

:
-
images with noise, background and cross-
talk the values of the coefficients no longer
reflect the amount of co-localization in the
128
O) original image. Restoration of the degraded
5 64 image resulted in a accurate determination

0 64
-
128 192 256
of the coefficients MI and M2, which were
found to be equal to 0.09 and 1177.
Red value respectively.

discriminate between these two cases and is equal to 0.50 [MI = 0.083 (= 1/12) and M2 = 0.751 implies that three
(Table 1) in both cases. The interpretation is simple: one of the four objects in Fig. 2(B) co-localize with objects in Fig.
quarter of the red objects (Fig. 1A) co-localizes with green 2(A). Figure 2(C) shows the 2-D histogram of this dual-
objects (Fig. 1H) and all green objects co-localize with red colour image. Near the axes the populations of pixels of the
objects. Starting from the knowledge that the red non-colocalizing objects are visible and the dots represent-
component of the image contains 36 objects, it can be ing the pixels of co-localizing objects are located on the line
concluded that the green component contains only nine y = x. After degrading the image with a 6% uniform and
objects, which all co-localize with one of the red objects. 4% non-uniform background, 8% cross-talk and 5 dB
The co-localization coefficients are very sensitive for Poisson noise (Fig. 2D and E), the co-localization
uniform and non-uniform background and cross-talk. coefficients M , and M2 were calculated again. The
When an image was deteriorated by one of these sources degradation of the signals is clearly visible in the bivariate
of degradation, the co-localization coefficients very rapidly histogram (Fig. 2F). The values of MI and M2 both equal
converged to 1.0, as the number of non-zero voxels rapidly 1.0 and are not meaningful. After restoration of the image
decreased. The coefficients were insensitive to Poisson noise (Fig. 2G and H) the histogram (Fig. 21) is very similar to
as long as the S/N exceeded 1 0 dB. When the image was too that of the undegraded image. Finally, the values of M, and
noisy the coefficients diminished. Mz were found to be 0.09 and 0.77, respectively. When we
depart from the knowledge that Fig. 2(G) contains 36
objects, then these results would imply that the image of
Reconstruction
Fig. 2(H) contains 4.2 objects of which 3.2 co-localize with
We carried out an experiment to investigate the reliability of objects of Fig. 2(G).
the image restoration procedures described above. We
combined the images from Fig. 1(A) and ( I ) (which are
Application
shown again in Fig. 2A and B) and calculated the
co-localization coefficients MI and M2. The result The co-localization analysis procedure was tested on three
 C O - L O C A L I Z A T I O N MEASUREMEN'I' 38 1

1.0, even though the image shows fewer co-localizing

objects than the combination Fig. 5(A)/(B). After image
restoration the co-localization coefficients showed that there
were fewer co-localizing objects: MI = 0.27 and M2 =0.28.
When the interval between the addition of each of the two
labels was 3 h (Fig. 3E/F), very few overlapping objects
could be identified (MI = 0.05 and M2 = 0.06).

Discussion
To measure 3-D intensity distributions in dual-colour
confocal images, the problem of a rigid data reduction
must be dealt with. The reduction of a data set of four
million image elements (2 x 2 56 x 2 56 x 32) to one or
two values implies a considerable loss of information.
Therefore, it is important to define the information that
should be derived from the images and to investigate to
what extent the quality of the information is retained
when it is expressed in terms of one or two variables. In this
paper we have introduced a method to measure the
co-localization of objects in dual-colour confocal images.
The degree of co-localization has been defined as the ratio of
the integral of the intensity distribution of co-localizing
objects and the total intensity of the respective components
of the image. We have investigated the quality of the results
obtained with this method and determined its limitations.
Pearson's correlation coefficient has already proved to be
a useful value for measuring the degree of overlap (Manders
et al., 1992). An accurate interpretaion of this coefficient.
however, has not been given previously. Interpretation of
the overlap coefficient as defined in this paper can be
Fig. 3. The red (A. C and E) and green components (B, D and F) of understood more easily. An image with an overlap
optical sections of 3-D dual-colour confocal images of DNA coefficient equal to 0.5 implies that 50% of both
replication patterns in three different cells. These cells were components of the image overlap with the other part of
labelled with two different halogenated nucleotides (IdUrd and
the image. We must make one condition: the number of
CldUrd) at the same time (A/B), with 1 h (C/D) and with 3 h (E/F)
objects in both components of an image has to be more or
between the addition of the labels. Before image restoration the co-
localization coefficients were all equal to 1.00.After restoration the less equal. When the number of objects is not equal.
co-localization coefficients MI and M2 provide clear information determination of the two co-localization coefficients MI and
about the amount of co-localizing objects in the components of an M2 is the proper method. A number of possible applications
image. A/B: M1 = 0.93 and M2 = 0.91;C/D: M1 = 0.27 and can be mentioned. For example, specific organelles in a cell
M2 = 0.28; E/F: M1 = 0.05 and M2 = 0.06. can be labelled and a specific process in this organelle can
be detected with another label. Using this method, the
fraction of organelles that carry out the process can then be
images of three different doubly labelled cells. Figure 3(A, B) calculated.
show the red and the green components of the image of a We have demonstrated that restoration of an image
cell that was labelled with both deoxyuridines at the same degraded with background, cross-talk and noise makes
time (both at T = 0). The co-localization coefficients MI possible an accurate assessment of the values of the co-
and M2 of the original image both equal 1.0. After the localization coefficients in the undegraded image. This
reduction of background, cross-talk and noise, the coef- provides confidence that the result of the calculations
ficients equalled 0.93 and 0.9 1, respectively, which means reflects the biological reality of the image. In addition,
that the images are almost identical. In Fig. 3(C) and (D) different intensities of the components of an image, caused
the components of a dual-colour image of a cell labelled at by, for example, different amplifier settings, did not
T = 0 (IdUrd) and T = 60min (CldUrd) are shown. Again, influence the values of the co-localization coefficients MI
the co-localization coefficients of the original image equalled and M,.
382 E. M. M. M A N D E R S . F. J . VERBEEK & J . A. A T E N

We have tested the method on images of cells doubly contraction bands and segmentation of the contractile ele-
labelled with IdUrd and CldUrd at different periods of the ments. 1. Cell. Biol. 111, 2463-2473.
S-phase. The results of the co-localization coefficients are in van den Engh, G.J., Trask, B.J. & Grey, J.W. (1985) The binding
agreement with results of earlier studies (Manders et al., kinetics and interaction of DNA fluorochromes used in the
analysis of nuclei and chromosomes by flow cytometry.
1992). The images of these replication patterns contain
Histochemistry, 84, 501-508.
objects (labelled DNA) of quite different intensities, shapes
Fox, M.H.. Arndt-Jovin. D.J.. Jovin, T.M., Baumann, P.H. & Robert-
and positions, in contrast with the objects in the synthesized Nicoud. M. (1991) Spatial and temporal distribution of DNA
images. replication sites localized by immunofluorescence and confocal
microscopy in mouse fibroblast. J. Cell Sci. 99, 247-253.
Acknowledgments Gonzalez. R.C. & Wintz, P. (1987) Digital Image Processing. 2nd
edn., Addison Wesley Publication Company, Mass.
We thank C. Hersbach and C. Gravemeijer for the Kitamura, T., Gatmaitan, Z. & Arias, I.M. (1990) Serial
photography work, R. J. A. M. Baeten and J. Stap for quantitative image analysis and confocal microscopy of hepatic
experimental assistance and Dr A. C. Schoolwerth, uptake, intracellular distribution and biliary secretion of a
Professor G. W. Barendsen, Professor J. Strackee and the fluorescent bile acid analog in rat hepatocyte doublets.
anonymous referees for critical reading of the manuscript. Hepatology, 12, 1358-1364.
This work has been partially sponsored by SPIN, project 3-D Lynch, R.M., Fogarty, K.E. & Fay, F.S. (1991) Modulation of
analysis and The European Communities, proposal No. hexokinase association with mitochondria analyzed with
0104. quantitative three-dimensional confocal microscopy. 1. Cell
Biol. 112, 385-295.
References Manders, E.M.M., Stap, J.. Brakenhoff, G.J., van Driel, R. & Aten.
J.A. (1992) Dynamics of three dimensional replication patterns
Akner, G.. Mossberg, K., Wilkstrom, A., Sundqvist, K. & during the S-phase, analyzed by double labelling of DNA and
Gustafsson, J. (1991) Evidence for colocalization of glucocorti- confocal microscopy. 1. Cell. Sci. 103.3, 85 7-862.
coid receptor with cytoplasmic microtubiles in human gingival Merdes, A., Stelzer, E.H. & De Mey, J. (1991) The three-
fibroblasts, using two different monoclonal anti-gr antibodies, dimensional architecture of the mitotic spidle, analyzed by
confocal microscopy and image analysis. I. Steroid Biochem. confocal flurorescence and electron microscopy. 1. Electron.
Molec. Biol. 39, 419-432. Microsc. Tech. 18, 61-73.
Akner, G., Sundqvist, K., Denis. M., Wilkstrom, A. & Gustafsson, J. Mossberg, K.. Arvidsson, U. & Ulfjake. B. (1990) Computerized
(1990) Immunocytochemical localization of gIucocorticoid quantification of immunofluorescence labeled axon terminals
receptor in human gingival fibroblasts and evidence for and of co-localization of neurochemicals in axon terminals with
colocalization of glucocorticoid receptor with cytoplasmic a confocal scanning laser microscope. I. Histochem. Cytochem.
microtubiles. Eur. 1. Cell Biol. 53, 390-401. 38, 179-190.
Arndt-Jovin, D.J.. Robert-Nicoud, M. & Jovin. T.M. (1990) Probing Mossberg, K. & Ericsson. M., (1990) Detection of doubly stained
DNA structures and function with multi-wavelength fluores- fluorescent specimens using confocal microscopy. J. Microsc.
cence confocal laser microscope. 1. Microsc. 15 7, 6 1- 72. 158, 215-224.
Aten. J.A.. Bakker, P.J.M.. Stap, J.. Boschman. G.A. & Veenhof, Sato, M.. Sardana, M.K., Grasser, W.A., Garsky, V.M., Murray. J.M.
C.H.N. (1992) DNA double labeling with IdUrd and CldUrd for & Gould, R.J. (1990) Echistanin is a potent inhibitor of bone
spatial and temporal analysis of cell proliferation and DNA resorption in cultures. 1. Cell Biol. 111, 1713-1 723.
replication. Histochem. 1. 24, 2 5 1-2 59 Stamatoglou. S.C., Sullivan. K.H., Johansson, S.. Barley. P.M..
Carlsson, K. (1990) Scanning and detection techniques used in a Burdett, I.D. & Hughes. R.C. (1990) Localisation of two
confocal scanning laser microscope. 1. Microsc. 157, 2 1-2 7. fibronectii-binding glycoproteins in rat liver primary hepato-
Carlsson. K. & Mossberg, K.. (1992) Reduction of cross-talk cytes. Co-distribution in vitro of integrin (alpha 5 beta 1 ) and
between fluorescent labels in scanning laser microscopy. 1. non-integrin (AGpllO) receptors in cell-substratum adhesion
Microsc. 167, 23-37. sites. 1. Cell Sci. 97, 595-606.
Dean. P., Mascio, L., Ow, D., Sudar, D. & Mullikin, J. (1990) Titus, J.A., Haugland, H., Sharrow, S.O. & Segal, D.M. (1982)
Proposed standard for image cytometry data files. Cytometry, 11, Texas Red. a hydrophilic, red-emitting flurophore for use with
561-569. fluorescein in dual parameter flow microfluorometric and
Draeger, A., Amos, W.B., Lkebe, M. & Small, J.V. (1990) The fluorescence microscopy studies. 1. Immunol. Methods, 50,
cytoskeletal and contractile apparatus of smooth muscle: 193-204.