Automated image analysis of cytokinesis-blocked micronuclei: an adapted protocol and a validated scoring procedure for biomonitoring.

Abstract

Micronuclei (MN) frequencies in peripheral blood lymphocytes have been used worldwide as a biomarker of chromosomal damage for genotoxicity testing and biomonitoring studies. Automation of MN analysis would provide faster and more reliable results with minimizing subjective MN identification. We developed an automated facility for the scoring of the in vitro MN cytokinesis-block assay for biomonitoring on Giemsa-stained slides, fulfilling the following criteria: applicable to the cytokinesis-block micronucleus methodology, discriminating between mono-, bi- and polynucleated cells, MN scoring according to HUMN scoring criteria, false-negative MN rate <10% and false-positive (FP) MN rate <1%. We first adapted the slide preparation protocol to obtain an optimal cell density and dispersion, which is important for image analysis. We developed specific algorithms starting from the cell as a detection unit. The whole detection and scoring process was separated into two distinct steps: in the first step, the cells and nuclei are detected; then, in the second step, the MN are searched for in the detected cells. Since the rate of FP MN obtained by the automatic analysis was in the range of 0.5-1.5%, an interactive visual validation step was introduced, which is not time consuming and allows quality control. Validation of the automated scoring procedure was undertaken by comparing the results of visual and automated scoring of micronucleated mono- and binucleated cells in human lymphocytes induced by two clastogens (ionizing radiation and methyl methane-sulphonate), two aneugens (nocodazole and carbendazim) and one apoptogen (staurosporine). Although the absolute MN frequencies obtained with automated scoring were lower as compared to those detected by visual scoring, a clear dose response for MNBN frequencies was observed with the automated scoring system, indicating that it is able to produce biologically relevant and reliable results. These observations, together with its ability to detect cells, nuclei and MN in accordance with the HUMN scoring criteria, confirm the usability of the automated MN analysis system for biomonitoring.

Full text

Page 1

Mutagenesis pp. 1–9, 2008 doi:10.1093/mutage/gen057
Automated image analysis of cytokinesis-blocked micronuclei: an adapted protocol
and a validated scoring procedure for biomonitoring
Ilse Decordier*, Alexander Papine1, Gina Plas,
Sam Roesems, Kim Vande Loock, Jennifer Moreno-
Palomo2, Eduardo Cemeli3, Diana Anderson3,
Aleksandra Fucic4, Ricardo Marcos2,
Franc xoise Soussaline1and Micheline Kirsch-Volders
Laboratorium voor Cellulaire Genetica, Vrije Universiteit Brussel, Pleinlaan 2,
1050 Brussels, Belgium,1IMSTAR, 60, rue Notre Dame des Champs, 75006
Paris, France,2Grup de Mutage `nesi, Departament de Gene `tica i de
Microbiologia, Universitat Auto `noma de Barcelona, Bellaterra, Spain,
3Department of Biomedical Sciences, University of Bradford, Bradford BD7
1DP, UK and4Institute for Medical Research and Occupational Health, 10000
Zagreb, Ksaverska c2, Croatia
Micronuclei (MN) frequencies in peripheral blood lym-
phocytes have been used worldwide as a biomarker of
chromosomal damage for genotoxicity testing and bio-
monitoring studies. Automation of MN analysis would
provide faster and more reliable results with minimizing
subjective MN identification. We developed an automated
facility for the scoring of the in vitro MN cytokinesis-block
assay for biomonitoring on Giemsa-stained slides, fulfilling
the following criteria: applicable to the cytokinesis-block
micronucleus methodology, discriminating between mono-,
bi- and polynucleated cells, MN scoring according to
HUMN scoring criteria, false-negative MN rate <10% and
false-positive (FP) MN rate <1%. We first adapted the slide
preparation protocol to obtain an optimal cell density and
dispersion, which is important for image analysis. We
developed specific algorithms starting from the cell as
a detection unit. The whole detection and scoring process
was separated into two distinct steps: in the first step, the
cells and nuclei are detected; then, in the second step, the
MN are searched for in the detected cells. Since the rate of
FP MN obtained by the automatic analysis was in the
range of 0.5–1.5%, an interactive visual validation step was
introduced, which is not time consuming and allows quality
control. Validation of the automated scoring procedure
was undertaken by comparing the results of visual and
automated scoring of micronucleated mono- and binucle-
ated cells in human lymphocytes induced by two clastogens
(ionizing radiation and methyl methane-sulphonate), two
aneugens (nocodazole and carbendazim) and one apop-
togen (staurosporine). Although the absolute MN frequen-
cies obtained with automated scoring were lower as
compared to those detected by visual scoring, a clear dose
response for MNBN frequencies was observed with the
automated scoring system, indicating that it is able to
produce biologically relevant and reliable results. These
observations, together with its ability to detect cells, nuclei
and MN in accordance with the HUMN scoring criteria,
confirm the usability of the automated MN analysis system
for biomonitoring.
Introduction
Analysis of micronucleus (MN) frequencies has since many
years been applied for in vitro genotoxicity testing of new
chemicals and for biomonitoring of human populations exposed
to different environmental, occupational or lifestyle factors (1).
MN are found in interphase cells as small, extranuclear
bodies resulting from chromosome breaks (leading to acentric
fragments) and/or whole chromosomes that did not reach the
spindle poles during cell division. At telophase, when the
nuclear envelope is reconstituted around the two daughter cells,
these lagging chromosomes or fragments are not incorporated
into the main nucleus but encapsulated into a separate, smaller
nucleus, a MN. MN represent therefore a measure of both
chromosome breakage and chromosome loss and can be used
to classify chemicals into clastogens or aneugens (2).
Since the formation of a MN requires a nuclear division, it is
necessary to be able to distinguish dividing cells from resting
cells. The cytokinesis-block micronucleus (CBMN) methodol-
ogy developed by Fenech and Morley (3) uses cytochalasin B
to identify cells that have divided in culture. Cytochalasin B is
an inhibitor of actin polymerization which is required for the
formation of the microfilament ring that constricts the
cytoplasm between the daughter nuclei during cytokinesis
(4). The CBMN methodology allows distinction between
a mononucleated cell, that did not divide, and a binucleated cell
that has divided once. micronuclei present in mononucleated
cells (MNMONO) may provide an indication of the genome
instability accumulated in vivo, while micronuclei in bi-
nucleated cells (MNBN) indicate the chromosome/genome
mutations accumulatedbefore
expressed during in vitro culture.
The International Collaborative Project on Micronucleus
Frequency in Human Populations (the HUMN project, http://
www.humn.org) established scoring criteria for MN using
isolated human lymphocyte cultures and used combined
databases to assess intra- and interlaboratory variation in MN
scoring, background MN frequencies and the influence of
culture conditions, age, gender and smoking on MN frequen-
cies (5–10). In addition, a large international cohort study
conducted within the frame of HUMN provided evidence that
baseline frequency of MN in peripheral blood lymphocytes is
a predictive biomarker of cancer risk (11,12).
Although the MN assay is well validated and easy to realize
and to score, applicability on a large scale for environmental
biomonitoring is hampered by lack of high-throughput
technology. Visual scoring of MN can be very time consuming
and large numbers of cells and/or donors have to be analysed to
obtain statistically relevant data, especially for biomonitoring
of populations characterized by low baseline MN frequencies
such as children (13). Moreover, even under optimized
laboratory conditions, often a high inter-scorer variation is
cultivation pluslesions
*To whom correspondence should be addressed. Tel: þ32 2 629 34 28; Fax: þ32 2 629 27 59; Email: idecordi@vub.ac.be
? The Author 2008. Published by Oxford University Press on behalf of the UK Environmental Mutagen Society.
All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org.
1

Mutagenesis Advance Access published October 14, 2008

Page 2

observed with visual MN analysis (9). Automation of the MN
analysis is still a requirement and would provide a faster and
more reliable analysis of MN frequencies with minimization of
subjective identification of MN.
To obtain reliable results, an automated system for MN
scoring used in biomonitoring studies should fulfil the same
requirements as those for visual scoring: it should be applicable
to the CBMN methodology and be able to perfectly distinguish
mono-,bi-andpolynucleatedcells; furthermore,thedetectionof
MN should be based on the scoring criteria described by the
HUMN project. Up to now, only a few studies reported on
the development of an automated image analysis system for the
detection of MN. Castelain et al. (14) developed an algorithm to
detect MN, which contained a sequence of grey operators and
binary operators. However, this automated system still showed
some limitations: errors occurred in the classification of
binucleated cells due to overlapping nuclei and in the detection
of some small MN. Moreover, it was developed at a time where
noscoringcriteriawereyetavailable.Morerecently,Vargaetal.
(15) established an automated scoring procedure for the CBMN
assaywithalgorithmstodetectbinucleatedcells,withorwithout
MN. A major drawback of this methodology is that binucleated
cells were scored starting from two similar nuclei as detection
unit and no information on cytoplasm was available.
In the present study, we describe an automated image
analysis system for the scoring of the in vitro MN cytokinesis-
block assay for biomonitoring on Giemsa-stained slides,
applying all scoring criteria defined by the HUMN project
and starting from the cell as a detection unit. For the latter, we
developed specific algorithms which in a first step detected
cells and nuclei, followed by a second step, in which the
system searched for MN in the detected cells. In this context,
we adapted the slide preparation protocol to obtain an optimal
density and dispersion of the cells, which is of major
importance for image analysis. Moreover, a standardized slide
preparation protocol ensures a high reproducibility when large
numbers of samples have to be prepared in parallel, which is
the case for biomonitoring. To validate the automated scoring
procedure, we compared the MN frequencies obtained by
visual and automated scoring in human peripheral blood
mononucleated cells (PBMC) treated with two clastogens:
ionizing radiation(IR)and
(MMS), two aneugens: nocodazole (NOC) and carbendazim
(CAR) and one apoptogen: staurosporine (STP).
The development of this automated facility for the scoring of
the in vitro MN cytokinesis-block assay for biomonitoring was
performed in the framework of the EU Integrated Project
NewGeneris. This research project aims at investigating the role
ofprenatalandearly-lifeexposuretogenotoxicchemicalspresent
in food and environment in the development of childhood cancer
and immune disorders. Our contribution to this project is (i) to
study the interaction between dietary carcinogens having also
genotoxic and immunotoxic properties and the induction of MN
as a validated biomarker for early mutagenic effects and (ii) to
comparemotherversuschildgenotoxicresponsestoasimilardiet
and environment using this biomarker.
methylmethane-sulphonate
Materials and methods
Preparation of slides suitable for automated image analysis of the CBMN assay
for biomonitoring
Blood cultures Human blood samples from a healthy donor of 30-year old
were drawn by venipuncture into heparinized tubes (Vacutainer; Becton
Dickinson, Oxford, UK). Whole blood cultures (5 ml) were set up in
RPMI1640/HEPES medium (Gibco BRL, Paisley, UK) supplemented with
15% foetal calf serum (Gibco BRL), 1% penicillin–streptomycin (Gibco BRL),
1%
L-glutamine (Gibco BRL), 2% phytohaemagglutinin (PHA; Remel,
Dartford, UK) and 7.5% whole blood and cultivated at 37?C. After 44 h,
cytochalasin-B (Sigma-Aldrich, Steinheim, Germany) was added at a final
concentration of 6 lg/ml.
Drug treatment of the blood cultures After 24 h PHA stimulation, the whole-
blood cultures were treated with three concentrations and a concurrent control
of the different mutagens (one culture per concentration). The choice of the
concentrations was based on previous studies from the laboratory—for NOC
(Acros Organics, Geel, Belgium): 0.01, 0.02 and 0.04 lg/ml (16); for CAR
(Aldrich Chemie, Steinheim, Germany): 2.5, 5.0 and 7.5 lg/ml (16); for MMS
(Sigma-Aldrich): 10, 20 and 30 lg/ml (16); for STP (Sigma-Aldrich): 0.029,
0.058 and 0.116 lg/ml (17) and for IR: 1, 2 and 3 Gy produced from a 60 Co
source (18). NOC, CAR and STP were dissolved in dimethylsulphoxide
(DMSO; Merck, Darmstadt, Germany) and MMS in phosphate-buffered saline
(PBS). The control cultures were treated with the solvents of the mutagens:
DMSO for NOC, CAR and STP and PBS for MMS. The final concentration of
DMSO did not exceed 0.5%.
Hypotonic shock At 72 h, the whole-blood cultures were harvested and cells
were centrifuged at 800 r.p.m./8 min at room temperature. After discarding the
supernatant with water pump until 500 ll remained, cells were re-suspended by
patting. Prior to fixation, cells were subjected to a cold hypotonic treatment at
several KCl concentration: 75, 90, 100, 110, 125 and 140 mM and three
hypotonic treatment times: 15, 30 and 45 min and were tested on untreated cell
cultures. KCl was added on vortex (800 l/min) and cells were incubated at 4?C
during the three tested times.
Fixation After hypotonic treatment, cells were centrifuged at 800 r.p.m./8 min
at room temperature and fixed with 5 ml of 3:1 freshly prepared methanol/acetic
acid which was added drop by drop on vortex (800 l/min), followed by three
additional drops of formaldehyde. After centrifugation at 800 r.p.m./8 min at
room temperature, the fixation was repeated twice without formaldehyde. After
each centrifugation, the supernatant was discarded by water pump until 500 ll
remained and cells were re-suspended by patting. After the last centrifugation,
the supernatant was discarded until 100 ll remained and cells were
re-suspended in 200–600 ll methanol/acetic acid, according to cell density.
Slide preparation The fixed cells were dropped, 2 ? 20 ll, on dry slides using
a micropipette on pre-marked positions, 13 mm from edges and frosted end,
resulting in two clear separated spots. Two slides per culture were prepared, and
slides were dried overnight.
Staining Slides were stained for 20 min in freshly prepared 5% Giemsa in
So ¨rensen buffer (pH 6.8) (Prosan, Merelbeke, Belgium), which was filtered
twice through Whatman 41 filter (Whatman International Ltd, Maidstone, UK).
Visual scoring At least 1000 binucleated cells were scored on one slide per
concentration of the different mutagens for the presence of one, two or more
MN and expressed per thousand (MNBN). The scoring criteria [http://
www.humn.org (1,10)] for MN included round or oval shaped MN with no
connection to the main nuclei, a size between 1/16 and 1/3 and similar staining
characteristics of the main nuclei. In addition, the percentage of BN,
polynucleated cells and mononucleated cells (MONO) with micronuclei
(MNMONO) or without were scored. The slides were coded and analysed on
a Zeiss transmitted light microscope at a magnification of ?400. From the data
of the MN analysis, the cytokinesis-block proliferation index (CBPI) was
calculated as follows: CBPI 5 (number of mononucleated cells þ 2 ? number
of binucleated cells þ 3 ? number of polynucleated cells)/total number of cells.
PathFinder? platform: high-content image cytometer for automated analysis
of the CBMN assay
The PathFinder? platform installed at the laboratory of Cell Genetics consists of
a PathFinder? CELLSCAN? capture station and two PathFinder? MN analysis
workstations. This configuration fulfils the requirement of the EU Integrated
Project NewGeneris to process up to 12 slides/day, while scoring a minimum of
2500 cells per slide, to ensure statistics of at least 1000 binucleated cells, and
limiting the human intervention time to ,1–2 min per slide.
The pathfinder? CELLSCAN? capture station The capture station comprises
an upright microscope (Nikon E50i) equipped for bright field and fluorescence
I. Decordier et al.
2

Page 3

with ?2, ?10, ?20, ?40 and ?100 objectives; motorized X-Y-Z stage
(ProScan?, Prior) with four-slide holder; an opto-electronic device for multi-
plan colour capture (BrightColor?, Imstar); a digital monochrome camera
(C8484, Hamamatsu) with 1280 ? 1024 resolution and 12 bits digitalization
and a Windows XP Pro? computer with Pathfinder? software platform,
including the following components: (i) Smart Capture Drivers for control of
motorized components and camera; (ii) Pathfinder? LightVision? for
automated slide scanning and capture in brightfield; (iii) Pathfinder?
CELLSCAN? for cells, nuclei and MN detection and multiparametric results
table output, presentation, validation and export and (iv) Pathfinder? Batch
Processor for the processing tasks distribution over the available processors
(analysis workstations).
The pathfinder? MN analysis workstation The MN analysis workstation
consists of a Windows XP Pro? computer with a Pathfinder? software
platform including the CELLSCAN? module.
Automated analysis procedure The procedure of automated MN slide analysis
as performed by the Pathfinder? platform includes the following steps: (i)
automated multiple-slide scanning and images capture; (ii) automated cells,
nuclei and MN detection; (iii) optional validation of detected MNs by the user/
expert and (iv) numerical results (scoring) reporting and export.
After the slides have been loaded in the holder, the following workflow is
performed, slide by slide:
(i) The predefined scanning zone, corresponding to the pre-marked two-drop
positions, is loaded. The software allows the user to modify the zone, this
feature not being used in routine scoring, where the goal is to minimize
the user interactions.
(ii) Landmarks for the automated focusing are set: this procedure consists of
taking tens of images at different focus positions and choosing the one
corresponding to the optimum focus. This procedure is too long to be
applied to every field of view to be captured. Thus, the automatic focusing
is performed on a limited set (usually 5) of positions within the zone to
scan and subsequent focus positions are calculated by interpolation based
on these five positions.
(iii) The zone is scanned (time of the order of 15 min), so that every field of
view inside the zone is placed under the nosepiece, captured with ?10
objective by the camera and saved in a file including image and capture
conditions. The ?10 magnification has been used for the performance
reasons. The nuclei we detected are .10 lm in diameter, which, given the
camera pixel size of 6.5 lm, corresponds to 16 image pixels. Thus, the
MN to detect (given the HUMN criteria of theoretical minimum of 1/16 of
the mean diameter of the main nuclei) exceed the image pixel size and
thus are detectable. The real MN were actually resolved as .6 pixel
objects because the diffraction limit of the ?10 objective is higher than
the pixel size (Rayleigh criterion of 0.6 ? wavelength/numerical aperture
(NA) gives the resolution limit of ?1 lM, for the 500 nm light and ?10
nosepiece with NA 0.3, which means that the object edges are .1 pixel).
(iv) When the slide scanning is completed, the slide images are queued for
processing.
(v) The processing is started as soon as a Pathfinder? Batch Processor is
available (i.e. terminated the previous task) on one of the analysis stations.
(vi) At the end of processing, the user can ask for interactive validation: the
cells containing detected MN are presented one by one; the user either
confirms or rejects each MN individually.
(vii) After the validation, the results are exported to the preset file, in the
format readable by spreadsheet or statistical analysis applications. The
results include total numbers of mono-, bi-, tri- and polynucleated cells
and number of cells of each type with one or more MN. In addition, the
CBPI is calculated and presented in the results table.
Specific algorithms for the detection of the cells, nuclei and MN Cell and
nuclei detection (see supplementary material, available at Mutagenesis Online):
(i) The cytoplasm regions are detected: optical density (OD) threshold is
automatically set based on the parameters of the main peak of OD
distribution in the image. Individual cells, cell clusters and various artefacts
are detected at this step.
The nuclei regions are detected inside these cytoplasm regions. A
threshold is set using the global analysis of the OD distribution, optimizing
the detection of the convex objects separately for each cytoplasm region.
(iii)The separation of touching cells is performed, avoiding cuts through the
nuclear regions. Thus, the cytoplasm regions corresponding to the cell
clusters are separated into individual cells.
(ii)
(iv) The nuclei are then detected inside the nuclei regions. They are then
filtered according to their shape, relative size and texture. Most of the
artefacts are thus eliminated.
Finally, the cells are filtered according to criteria of shape, size and number
of nuclei (cytoplasm regions without detected nuclei or of atypical cell
morphology are eliminated).
MN detection (see supplementary material, available at Mutagenesis Online):
(i) Large background variations are eliminated from the image.
(ii) The MN candidates are detected by OD thresholding as described above
for cytoplasm regions detection.
(iii) The candidates MN are, then, filtered according to the following criteria:
(a) Size relative to the nuclei (MN/nucleus size ratio) in the given cell.
(b) Position within the cell: candidates touching the cell contour are
discarded.
(c) Colour: candidates having a colour too different from the nuclei, such
as dust particles, are removed.
(d) Circular symmetry: the MN candidates below a level of circular
symmetry are removed.
(e) Local contrast: not sufficiently contrasted candidates are removed, the
contrast being measured as the ratio of the candidate’s OD to the OD
of a cytoplasm ring around it.
Results
Slide preparation
For image analysis, it is important to standardize the slide
preparation as much as possible to be able to analyse the
images in an unambiguous and consistent way and to obtain an
optimal reproducibility. Therefore, we first standardized the
slide preparation protocol in order to obtain an optimal density
and dispersion of the cells (Figure 1), avoiding overlapping of
the cells, which is of major importance for image analysis to
reach the most appropriate detection as possible. An important
step in the slide preparation protocol is the hypotonic treatment
which induces swelling of the cells and hence influences the
Fig. 1. Image of a suitable dispersion and swelling of cells for automated
scoring. Cells are well dispersed and not overlapping.
Automated image analysis of micronuclei
3

Page 4

detection of cells. Various concentrations of the hypotonic
treatment (KCl), all applied at 4?C, were tested on untreated
lymphocyte cultures to find the optimal detection which is not
sensitive to small variations over the hypotonic exposure times.
Figure 2 presents the total number of cells detected (Figure 2A)
and the distribution of the different sub-populations (mono-, bi-
and polynucleated cells) (Figure 2B), with the different KCl
concentrations (75, 90, 100, 110, 125 and 140 mM) and times
of hypotonic treatment (15, 30 and 45 min). At 140 mM of
KCl, the lowest variation in the distribution of the sub-
populations of cells was observed over the different times of
hypotonic treatment, indicating that the duration of the
hypotonic treatment did not have a major influence. However,
in regard to the total number of cells detected, a hypotonic
treatment of 110 mM showed the best detection capacity at 15
and 30 min. Since this KCl concentration of 110 mM also
showed a consistent repartition of the sub-populations at 15
and 30 min, it was decided to apply this concentration of
hypotonic treatment for 15 min at 4?C.
Besides changes to the hypotonic treatment, a few other
steps in the slide preparation protocol were slightly adapted. To
obtain an optimal spreading of the cells, without overlapping
too much, the cells were re-suspended in a higher volume of
fixative as compared to that used for the standard protocol
before spreading onto slides.
Slides were stained with 5% Giemsa (14) instead of 10 %
(19), in So ¨rensen buffer, to obtain an optimal contrast between
nuclei and cells. Giemsa solution was freshly prepared and
filtered twice to avoid artefacts in the detection due to debris
present in the Giemsa solution (data not shown).
Validation of automated MN scoring by image analysis
To validate the image analysis system for the automated
scoring of MN, we compared the results of automated and
visual scoring of MN induced by two aneugens, NOC and
CAR, showing a threshold in dose response; two clastogens, IR
and MMS, without threshold in dose response, and a non-
genotoxic apoptogen, STP. The latter was included to verify
A
0
200
400
600
800
1000
1200
1400
15 30 45 1530 4515 30 45 153045 153045 15 30 45
[ KCl ] = 75
mM
[ KCl ] = 90
mM
[ KCl ] = 100
mM
[ KCl ] = 110
mM
[ KCl ] = 125
mM
[ KCl ] = 140
mM
KCl concentration (mM) and time of the hypotonic shock (min)
total number of detected cells
B
0
10
20
30
40
50
60
70
80
90
100
1530 4515 30 4515 3045 1530 4515 30 4515 30 45
[ KCl ] = 75 mM [ KCl ] = 90 mM[ KCl ] = 100
mM
[ KCl ] = 110
mM
[ KCl ] = 125
mM
[ KCl ] = 140
mM
KCl concentration (mM) and time of the hypotonic shock (min)
mononucleated
binucleated
polynucleated
Fig. 2. (A) Total number of cells detected in function of KCl concentration and time of hypotonic treatment. (B) Distribution of the different sub-populations
(mono-, bi- and polynucleated cells) in function of KCl concentration and time of hypotonic treatment.
I. Decordier et al.
4

Page 5

whether the automated analysis system was able to discrim-
inate between MN and apoptotic bodies. For each compound,
three concentrations and corresponding controls were analysed.
The visual scoring was performed by an experienced scorer
(G). As performed with visual scoring, the automated image
analysis system detected mono-, bi- and polynucleated cells
allowing to calculate the CBPI, and MN were scored in
mononucleated (MNMONO) and binucleated (MNBN) cells.
The frequencies of these micronucleated cells and CBPI were
compared between the two scoring methods, for which the
same coded slides were used. Table I presents an overview of
the total number of cells, percentage of mononucleated,
binucleated and polynucleated cells detected by visual scoring
and by automated scoring, and the CBPI calculated from these
values for the two scoring methods. When comparing the CBPI
for the five compounds tested, very similar values were
obtained with the two scoring procedures.
Since the rate of false-positive (FP) MN obtained by the
automatic analysis was in the range of 0.5–1.5% (data not
shown), an interactive visual validation step was introduced.
MN scored by the automatic detection are presented one by one
on the screen (Figure 3), the scorer has to either validate or
invalidate (5FP) the MN found by the automated analysis. It is
also possible to label it for future review in case of doubt. This
interactive validation step was performed by the same scorer
(G) who performed the visual scoring.
Table II presents the comparison of MN frequencies in
mono- and binucleated cells (MNBN and MNMONO)
obtained by visual, automated and automated scoring after
visual validation. Comparison of the MN frequencies between
automated and automated scoring after visual validation
showed that the visual validation step is required to reduce
the FP MN. Nevertheless, comparison of the frequencies of
MNBN obtained by visual and automated scoring after visual
validation for the five compounds tested shows that both
scoring methods detect a dose response for the two clastogens
and aneugens analysed. However, in all cases, lower
frequencies of MNBN were detected with the automated
system as compared to the visual scoring. As far as the
frequencies of MNMONO are concerned, with both visual and
automated scoring with visual validation, a clear increase was
observed for the highest dose of NOC and CAR, which is
expected from aneugens inducing mitotic slippage at high
concentrations. Again, lower MN frequencies were found with
the automated scoring with visual validation as compared to the
visual scoring. For MMS and IR, as expected from clastogens,
no increase in the frequencies of MNMONO was observed,
neither with visual nor with automated scoring. Concerning the
apoptogen STP, as expected no significant increase in MN was
observed, neither in mononucleated nor in binucleated cells,
with both scoring methods.
In order to investigate the reproducibility of the automated
image analysis combined with the visual validation step, we
assessed its inter-capturing variability. Therefore, the same
coded slides used for the comparison between visual and
automated scoring with visual validation were captured and
analysed four more times and visually validated by the same
scorer (G), i.e. all slides were scored five times. With the four
additional capturing sessions, the results were similar to the
first capture session (Figure 4). A dose response in MNBN
Table I. Comparison of total number of cells, percentage of mononucleated, binucleated, and polynucleated cells detected by visual scoring (vis) and by automated
scoring (auto) and the CBPI calculated from these values for the two scoring methods
Total number of cells % Mononucleated cells% Binucleated cells % Polynucleated cellsCBPI
vis auto visauto vis autovis autovis auto
MMS
Untreated
PBS
10 lg/ml
20 lg/ml
30 lg/ml
IR
Untreated
1 Gy
2 Gy
3 Gy
NOC
Untreated
DMSO
0.01 lg/ml
0.02 lg/ml
0.04 lg/ml
CAR
Untreated
DMSO
2.5 lg/ml
5.0 lg/ml
7.5 lg/ml
STP
Untreated
DMSO
0.029 lg/ml
0.058 lg/ml
0.116 lg/ml
1000
1000
1000
1000
1000
5184
4826
5091
3974
1514
29.0
36.2
48.0
49.2
69.8
45.8
32.1
47.0
61.5
72.6
61.4
57.2
48.4
48.0
29.6
50.1
63.8
50.9
37.5
27.0
9.6
6.6
3.6
2.8
0.6
4.1
4.1
2.1
1.0
0.4
1.8
1.7
1.6
1.5
1.3
1.6
1.7
1.6
1.4
1.3
1000
1000
1000
1000
3860
5760
4127
4647
35.8
33.2
58.4
71.7
49.6
58.7
69.7
79.1
52.0
61.8
39.8
26.3
43.2
38.9
29.2
20.1
12.2
5.0
1.8
2.0
7.2
2.4
1.1
0.9
1.6
1.7
1.4
1.3
1.6
1.4
1.3
1.2
1000
1000
1000
1000
1000
5184
4776
3980
4135
4461
29.0
31.3
30.5
29.0
40.0
45.8
35.7
39.3
43.1
56.5
61.4
57.1
61.7
59.4
57.2
50.1
58.8
56.1
52.4
41.9
9.6
11.6
7.8
11.6
2.8
4.1
5.5
4.5
4.4
1.5
1.6
1.8
1.8
1.8
1.6
1.6
1.7
1.7
1.6
1.4
1000
1000
1000
1000
1000
5184
4776
2865
2286
5928
29.0
31.3
34.2
35.2
38.6
45.8
35.7
36.1
34.2
37.2
61.4
57.1
57.3
55.2
56.4
50.1
58.8
58.8
61.5
60.3
9.6
11.6
8.5
9.6
5.0
4.1
5.5
5.0
4.2
2.5
1.6
1.8
1.7
1.7
1.7
1.6
1.7
1.7
1.7
1.7
1000
1000
1000
1000
1000
5184
4776
2211
3397
4034
29.0
31.3
39.6
38.4
43.4
45.8
35.7
26.9
35.7
29.7
61.4
57.1
54.3
55.2
51.3
50.1
58.8
71.0
61.6
67.9
9.6
11.6
6.1
6.4
5.3
4.1
5.5
2.1
2.6
2.4
1.6
1.8
1.7
1.7
1.6
1.6
1.7
1.8
1.7
1.7
Automated image analysis of micronuclei
5

Page 6

frequencies was found for the two clastogens (Figure 4A and B)
and aneugens (Figure 4C and D) analysed, but lower
frequencies of MN, both in binucleated (MNBN) and mono-
nucleated (MNMONO) cells, as compared to the visual
scoring. Furthermore, no difference in CBPI was observed
for all compounds analysed (data not shown).
To investigate the inter-scorer variability of the visual
validation procedure, a second independent scorer (S)
performed the visual validation of the same five capturing
sessions. Comparison of the results obtained by the two
independent scorers is shown in Figure 4. The values obtained
for MNBN and MNMONO frequencies by the five capturing
sessions and the visual validation step performed by the two
different scorers were highly correlated with those of the visual
scoring for the two clastogens and aneugens studied, despite
the differences in absolute MN frequencies (Table III).
Discussion
For many years, the CBMN assay has been successfully
applied for human biomonitoring of in vivo genotoxin exposure
and provides a sensitive and relatively easy methodology to
assess mutations (1). Moreover, the fact that baseline MN
frequencies in cytokinesis-blocked lymphocytes have been
shown to be a predictive biomarker for cancer risk (11,12)
strengthens the importance of the CBMN assay as a reliable
methodology for human biomonitoring of early genetic effects.
Nevertheless, there is need for automation of MN analysis for
quicker and more reliable detection with minimizing subjective
judgement and individual scoring skills. Additionally, for
biomonitoring purposes, automation would increase the
sensitivity, which is of major importance when low MN
frequencies have to be detected. For this goal to be achieved,
we decided that the following requirements should be fulfilled:
the automated MN analysis system should be able to clearly
detect and distinguish mono-, bi- and polynucleated cells; for
MN scoring the same criteria should be applied as defined by
HUMN; the false-negative MN rate (ratio of number of
detected MN to number of all MN) should be 10% and the FP
rate (ratio of number of cells without MN where MN has been
Fig. 3. Validation screen for the interactive visual validation step, MN detected by the automated system are presented one by one; for each MN, the scorer has to
either validate it or invalidate it (5FP) or label it for future review.
Table II. Comparison of MN frequencies in bi- and mononucleated cells
(MNBN and MNMONO) obtained by visual (vis), automated (auto) and
automated scoring after manual validation (val)
& MNBN
& MNMONO
visautovalvisauto val
MMS
Untreated
PBS
10 lg/ml
20 lg/ml
30 lg/ml
IR
Untreated
1 Gy
2 Gy
3 Gy
NOC
Untreated
DMSO
0.01 lg/ml
0.02 lg/ml
0.04 lg/ml
CAR
Untreated
DMSO
2.5 lg/ml
5.0 lg/ml
7.5 lg/ml
STP
Untreated
DMSO
0.029 lg/ml
0.058 lg/ml
0.116 lg/ml
4.0
5.0
10.0
21.0
47.0
21.9
20.1
14.7
28.6
41.6
3.8
3.2
8.1
19.1
26.9
0.0
1.6
0.0
0.0
0.0
13.1
27.8
2.2
9.8
2.6
2.2
2.0
0.0
0.0
0.0
6.0 9.6
59.0
126.1
298.0
2.4
47.8
102.1
261.5
2.9
3.7
4.1
9.2
9.6
13.2
8.4
12.7
0.0
2.0
2.5
5.3
107.0
275.0
483.0
4.0
5.8
5.0
7.0
53.0
21.9
12.5
15.7
14.3
39.0
3.8
3.6
3.6
5.1
19.2
0.0
1.8
2.0
2.0
50.1
13.1
16.4
18.6
25.5
48.2
2.2
1.6
1.4
6.1
18.5
4.0
5.8
6.0
7.0
38.0
21.9
12.5
16.0
29.9
42.8
3.8
3.6
4.2
8.5
20.1
0.0
1.8
1.7
1.6
1.5
13.1
16.4
13.0
10.8
42.1
2.2
1.6
0.0
1.8
11.3
4.0
5.8
6.0
12.0
8.0
21.9
12.5
49.0
17.2
28.5
3.8
3.6
8.9
5.3
6.9
0.0
1.8
5.5
7.2
3.5
13.1
16.4
26.4
8.6
13.7
2.2
1.6
2.6
0.0
2.3
I. Decordier et al.
6

Page 7

detected to the total cell number) ,1%. Furthermore, it should
allow a faster and more efficient MN analysis and quality
control. Developing an automated image analysis for the
CBMN assay taking into account these considerations was only
possible through an intensive collaboration between cytologists
with a large experience in visual scoring and a high-technology
company highly experienced in imaging automated systems,
who could both understand the research question.
Compared to other automated image analysis systems to
score MN described up to now (14,15), the automated scoring
of MN developed in the present study provides several
advantages and improvements:
(i) We first standardized the slide preparation protocol by
adapting the hypotonic treatment in order to obtain
uniformity in cell size, which is important when different
laboratories are preparing slides that have to be analysed
by the same image analysis system.
(ii) The software protocol that was developed for the
automated scoring of MN started from the detection of
cells and hence the identification of the cytoplasm,
followed by the identification of the mono-, bi- and
polynucleated cells, and MN in these different sub-
populations of cells. We separated the whole detection
and scoring process into two distinct steps: in the first step,
A
0
10
20
30
40
50
60
70
MMS
µg/ml
untreated 102030 PBS
B
0
100
200
300
400
500
600
Gray
IR
untreated123
0
10
20
30
40
50
60
70
MMS
µg/ml
untreated102030PBS
Gray
0
100
200
300
400
500
600
IR
123
0
10
20
30
40
50
60
70
NOC
µg/ml
0.020.04 untreated DMSO0.01
0
10
20
30
40
50
60
70
CAR
µg/ml
untreated DMSO2.5 5.07.5
0
10
20
30
40
50
60
70
C
NOC
µg/ml
0.020.04 untreated DMSO0.01
0
10
20
30
40
50
60
70
D
CAR
untreated
µg/ml
untreated DMSO 2.55.0 7.5
0
10
20
30
40
50
60
70
E
µg/ml
0.058 untreated DMSO 0.029 0.116
STP
20
10
30
40
50
60
70
STP
0.058untreated DMSO 0.0290.116
µg/ml
Fig. 4. Inter-capturing session and inter-scorer comparison for the frequencies of MNBN, MNMONO for MMS (A), IR (B), NOC (C), CAR (D) and STP (E). Five
independent capturing sessions were performed. For the inter-scorer assessment two independent scorer (G and S) performed the visual validation of the same five
capturing sessions (Auto G1-G5, Auto S1-S5). Filled squares, Man; filled triangles, Auto_S1; inverted filled triangles, Auto_G1; filled diamonds, Auto_S2; filled
circles, Auto_G2; open squares, Auto_S3; open triangles, Auto_G3; open inverted triangles, Auto_S4; open diamonds, Auto_G4; open circles: Auto_S5 and cross,
Auto_G5.
Table III. Linear correlation coefficient (R) between MNCB frequencies obtained with manual scoring and automated scoring for the five independent capturing
sessions
Session 1Session 2Session 3Session 4 Session 5
Scorer GScorer S Scorer GScorer SScorer GScorer SScorer GScorer SScorer GScorer S
MMS
IR
NOC
CAR
STP
0.962
0.981
0.998
0.970
0.140
0.963
0.985
0.996
0.991
0.178
0.963
0.962
0.975
0.995
0.206
0.952
0.964
0.992
0.992
0.321
0.995
0.952
0.991
0.995
0.616
0.996
0.955
0.992
0.995
0.628
0.990
0.971
0.993
0.986
0.895
0.988
0.973
0.996
0.981
0.913
0.999
0.973
0.995
0.993
0.896
0.987
0.977
0.994
0.993
0.852
Automated image analysis of micronuclei
7

Page 8

the cells and nuclei are detected; then, in the second step,
the MN are searched for in the detected cells. The
algorithms were applied to every individual image. The
choice of the parameters for these algorithms was
performed following an iterative process with important
participation of the experienced visual scorers, who also
participated in the MN studies of the HUMN project (9):
the parameters were chosen to fit as closely as possible
with the scorer’s choice of viable cells, nuclei and MN to
score. The scoring criteria were based on the HUMN
programme [http://www.humn.org (1,9)]. To us, it was
essential to start from the cell as detection unit, in order to
obtain results on mono-, bi- and polynucleated cells and not
to restrict to the analysis of binucleated cells. At first,
information on these different cell classes allows the
assessment of cell proliferation through nuclearity index,
which is important for an efficient assessment of the genetic
damage in human biomonitoring (1,6,20,21). At second,
scoring of mononucleated cells also allows the detection of
MN in those cells (MNMONO). Information on the
frequencies of MNMONO may provide an indication of
the genome instability accumulated in vivo, while MNBN
indicate the chromosome/genome mutations accumulated in
vivo before cultivation plus lesions expressed during in vitro
culture. Moreover, micronucleated mononucleated cells can
also constitute tetraploid mononucleated cells arising after
mitotic slippage in presence of microtubule inhibitors. This
is important to be taken into account when investigating the
aneugenic potential of chemicals (reviewed in ref. 22).
Besides the fact that the acquisition starting at cell level
instead of nucleus level provides information on cell prolif-
eration,italso provides the possibility todetect apoptotic and
necrotic cells, nucleoplasmic bridges and nuclear buds (23).
(iii) The specific detection algorithms were developed to
identify cells, nuclei and MN on Giemsa-stained slides.
Giemsa staining has the advantage that the slides do not
have to be protected from the light which allows a more
efficient acquisition as compared to fluorescent dyes that
need protection from light. Moreover, this staining also
provides the advantage that the slides can be easily re-
examined visually if necessary without loss of quality of the
staining. In a study by Castelain et al. (14), an algorithm
was developed to detect MN on Giemsa and Feulgen-
Congo-Red-stained slides and also started from the cell as
detection unit, allowing the detection of mono-, bi- and
polynucleated cells. However, this system showed a mis-
classification of binucleated cells, mainly due to non-
separation of overlapping nuclei. Although the authors
compared different slide preparation procedures, no robust
standardized protocol was established. In a more recent
study by Varga et al. (15), an automated scoring procedure
was developed for the CBMN assay, with algorithms
developed to detect only binucleated cells, with or without
MN, using two similar non-overlapping nuclei, as unit of
detection, without acquiring information on nuclearity.
Also in this study, a standardized slide preparation protocol
was missing.
(iv) Comparison of MN frequencies in PBMC, exposed to five
well-known mutagens, obtained by visual scoring to those
obtained with automated scoring revealed that both
scoring methods are able to detect a dose response.
(v) Comparison of the CBPI obtained from the visual and the
automated scoring showed that the detection of the
different sub-populations of cells (mono-, bi- and poly-
nucleated cells) was very accurate.
(vi) ByMNanalysisinSTP-treatedcultures,weexcludedthatthe
detectedmicronucleatedcellsbytheautomatedimagesystem
correspond to apoptotic bodies, thus confirming that the
image analyser can distinguish MN from apoptotic bodies.
(vii) As far as the reproducibility is concerned, with the
automated image analysis system presented here very
similar results were obtained between different capturing
sessions combined with the visual validation step. For each
capturing and analysis session combined with the visual
validationbytwoindependentscorers,acleardoseresponse
in MN frequencies was observed. In addition, MNBN
frequencies found for the five capturing sessions and the
visualvalidationstepperformedbythetwodifferentscorers
were highly correlated with those of the visual scoring,
indicating the ability of the automated scoring system to
detect biological differences as observed by visual scoring.
Nevertheless, the automated image analysis system for MN
detection described here also shows some limitations:
(i) Since the current FP rate was not always below the
required 1%, we decided to include a visual validation
step, which is performed visually on screen by a scorer.
However, this human interaction is not time consuming,
provides a good quality control and allows to re-analyse
the same sample.
(ii) The MN frequencies obtained with the automated system
were lower as compared to the visual scoring, and in
particular, the automated system was less efficient in
detecting high frequencies of MN at the highest concen-
trations of the two clastogens and aneugens tested. A
similar phenomenon of a drop in MN frequencies detected
with automated analysis as compared to visual scoring was
also observed by Varga et al. (15) when analysing MN
frequencies in lymphocytes of control and breast cancer
patients. In that study, the differences between visual and
automated scoring were larger as compared to our data: on
average, only 34.5 % of the MN scored visually were
detected by the automated system, while in the present
study the mean detection rate was 68.5%. The automated
MN detection system described by Castelain et al. (14)
also missed a part of the MN detected visually; they found
an overall detection rate of ?51%. Nevertheless, it is
difficult to directly compare our results on the accuracy of
automated MN detection with these previously described
studies, since different slide preparation protocols were
used and low numbers of binucleated cells (,500) were
analysed (14). In addition, visual and automated scoring
was performed on different slides from the same cultures,
stained with Giemsa and 4#, 6-diamidino-2-phenylindole
(DAPI), respectively (15). There are several plausible
explanations for the lower MN detection rate obtained
with automated scoring described in the present study. The
differences may in part be explained by the very strict
scoring criteria that were applied for the automated image
analysis. When MN detected by the automated system
were doubtful according to the scorers performing the
visual validation step, these MN were not labelled as valid
MN. When performing the visual scoring using the direct
microscope observation, the scorers, in the case of doubt,
may slightly modify the focus, thus possibly validating
I. Decordier et al.
8

Page 9

a doubtful object as MN. Furthermore, one can assume
that the detection of MN, in particular small MN, is very
close to the detection limit of the image analysis system.
We verified whether the underestimation of small MN was
not due to the capturing at a low magnification (?10).
Therefore, some slides were captured with the ?20
objective and MN results were compared with those
obtained with a capturing session performed with the ?10
objective (data not shown). Similar results were obtained
with both magnifications, indicating that very small MN
observed on a ?20 image are still visible on a ?10 image.
Another plausible explanation for the lower MN frequen-
cies found with the automated scoring as compared to
visual scoring at the highest concentrations could be the
irregular shape of nuclei. At those high mutagen
concentrations, cells accumulate more genetic damage
which often results in a higher number of MN per cell and
more irregular nuclei. Nevertheless, with the automated
system, the number of cells to be analysed can easily be
increased (e.g. up to 5000 cells), thereby reducing the
number of errors in the detection of MN. Moreover, we
aimed at developing an automated image analysis system
for MN scoring specifically for biomonitoring for which
we do not expect very high MN frequencies.
In summary, in the present manuscript, we presented an
automated image analysis system for the scoring of the in vitro
MN cytokinesis-block assay for biomonitoring, starting from
thecellasa unitfordetection anddeveloped astandardizedslide
preparation protocol suitable for automated scoring. The
automated scoring application automatically scans slides,
detects cells, nuclei and MN, following the HUMN scoring
criteria. A visual validation of MN is still required, but provides
an additional quality control which is limited to only a few
minutes per slide. Although the absolute MN frequencies found
with automated scoring were lower as compared to those
obtained by visual scoring, the automated scoring system is able
to produce biologically relevant and reliable results, with low
inter-scorer variability in the visual validation step, and thus
which can be used for biomonitoring. In the near future, we aim
at further developing the automated scoring system for a more
efficient detection of high MN frequencies; the detection of
apoptosis/necrosis, nucleoplasmic bridges and nuclear buds,
which would also allow its usability for in vitro genotoxicity
testing.
Supplementary data
Supplementary material is available at Mutagenesis Online.
Funding
EU Integrated Project NewGeneris (acronym of the project
‘NewbornsandGenotoxic
newgeneris.org), 6th Framework Programme, Priority 5: Food
Quality and Safety (Contract no. FOOD-CT-2005-016320).
exposure risks’http://www.
Acknowledgements
Conflict of interest statement: None declared
References
1. Fenech, M. (2007) Cytokinesis-block micronucleus cytome assay. Nat.
Protoc., 2, 1084–1104.
2. Mateuca, R., Lombaert, N., Aka, P. V., Decordier, I. and Kirsch-
Volders, M. (2006) Chromosomal changes: induction, detection methods
and applicability in human biomonitoring. Biochimie, 88, 1515–1531.
3. Fenech, M. and Morley, A. A. (1985) Measurement of micronuclei in
lymphocytes. Mutat. Res., 147, 29–36.
4. Carter, S. B. (1967) Effects of cytochalasins on mammalian cells. Nature,
213, 261–264.
5. Fenech, M. (1998) Chromosomal damage rate, aging, and diet. Ann. N. Y.
Acad. Sci., 854, 23–36.
6. Fenech, M., Holland, N., Chang, W. P., Zeiger, E. and Bonassi, S. (1999)
The human micronucleus project—an international collaborative study on
the use of the micronucleus technique for measuring DNA damage in
humans. Mutat. Res., 428, 271–283.
7. Bonassi, S., Fenech, M., Lando, C. et al. (2001) Human Micronucleus
Project: international database comparison for results with the cytokinesis-
block micronucleus assay in human lymphocytes: I. Effect of laboratory
protocol, scoring criteria, and host factors on the frequency of micronuclei.
Environ. Mol. Mutagen., 37, 31–45.
8. Bonassi, S., Neri, M., Lando, C. et al. (2000) Effect of smoking habit on the
frequency of micronuclei in human lymphocytes: results from the human
micronucleus project. Mutat. Res., 543, 155–166.
9. Fenech, M., Bonassi, S., Turner, J. et al. (2003) Intra- and inter-laboratory
variation in the scoring of micronuclei and nucleoplasmic bridges in
binucleated human lymphocytes. Results of an international slide-scoring
exercise by the HUMN project. Mutat. Res., 534, 45–64.
10. Fenech, M., Chang, W. P., Kirsch-Volders, M., Holland, N., Bonassi, S. and
Zeiger, E. and Human Micronnucleus Project (2003) HUMN project: detailed
description of the scoring criteria for the cytokinesis-block micronucleus
assay using isolated human lymphocyte cultures. Mutat. Res., 534, 65–75.
11. Bonassi, S., Znaor, A., Ceppi, M. et al. (2007) An increased micronucleus
frequency in peripheral blood lymphocytes predicts the risk of cancer in
humans. Carcinogenesis, 28, 625–631.
12. Murgia, E., Ballardin, M., Bonassi, S., Rossi, A. M. and Barale, R. (2008)
Validation of micronuclei frequency in peripheral blood lymphocytes as
early cancer risk biomarker in a nested case-control study. Mutat. Res., 39,
27–34.
13. Neri, M., Ceppi, M., Knudsen, L. E., Merlo, D. F., Barale, R., Puntoni, R.
and Bonassi, S. (2005) Baseline micronuclei frequency in children: estimates
from meta- and pooled analyses. Environ. Health Perspect., 113, 1226–1229.
14. Castelain, P., Van Hummelen, P., Deleener, A. and Kirsch-Volders, M.
(1993) Automated detection of cytochalasin-B blocked binucleated
lymphocytes for scoring micronuclei. Mutagenesis, 8, 285–293.
15. Varga, D., Johannes, T., Jainta, S., Schuster, S., Schwarz-Boeger, U.,
Kiechle, M., Patino Garcia, B. and Vogel, W. (2004) An automated scoring
procedure for the micronucleus test by image analysis. Mutagenesis, 19,
391–397.
16. Elhajouji, A., Tibaldi, F. and Kirsch-Volders, M. (1997) Indication for
thresholds of chromosome non-disjunction versus chromosome lagging
induced by spindle inhibitors in vitro in human lymphocytes. Mutagenesis,
12, 133–140.
17. Decordier, I., Cundari, E. and Kirsch-Volders, M. (2005) Influence of
caspase activity on micronuclei detection: a possible role for caspase-3 in
micronucleation. Mutagenesis, 20, 173–179.
18. Godderis, L., Aka, P., Mateuca, R., Kirsch-Volders, M., Lison, D. and
Veulemans, H. (2006) Dose-dependent influence of genetic polymorphisms
on DNA damage induced by styrene oxide, ethylene oxide and gamma-
radiation. Toxicology, 219, 220–229.
19. Lucero, L., Pastor, S., Sua ´rez, S., Durba ´n, R., Go ´mez, C., Parro ´n, T.,
Creus, A. and Marcos, R. (2000) Cytogenetic biomonitoring of Spanish
greenhouse workers exposed to pesticides: micronuclei analysis in peripheral
blood lymphocytes and buccal epithelial cells. Mutat. Res., 464, 255–262.
20. Kirsch-Volders, M. and Fenech, M. (2001) Inclusion of micronuclei in non-
divided mononuclear lymphocytes and necrosis/apoptosis may provide
a more comprehensive cytokinesis block micronucleus assay for bio-
monitoring purposes. Mutagenesis, 16, 51–58.
21. Kirsch-Volders, M., Sofuni, T., Aardema, M. et al. (2003) Report from the
in vitro micronucleus assay working group. Mutat. Res., 540, 153–163.
22. Decordier, I., Cundari, E. and Kirsch-Volders, M. (2008) Survival of
aneuploid, micronucleated and/or polyploid cells: crosstalk between ploidy
control and apoptosis. Mutat. Res., 651, 30–39.
23. Fenech, M. (2006) Cytokinesis-block micronucleus assay evolves into
a ‘‘cytome’’ assay of chromosomal instability, mitotic dysfunction and cell
death. Mutat. Res., 600, 58–66.
Received on June 2, 2008; revised on September 17, 2008; accepted on
September 18, 2008
Automated image analysis of micronuclei
9