Association between frequency of chromosomal aberrations and cancer risk is not influenced by genetic polymorphisms in GSTM1 and GSTT1.

Abstract

The frequency of chromosomal aberrations (CA) in peripheral blood lymphocytes of healthy individuals has been associated with cancer risk. It is presently unclear whether this association is influenced by individual susceptibility factors such as genetic polymorphisms of xenobiotic-metabolizing enzymes.
To evaluate the role of polymorphisms in glutathione S-transferase (GST) M1 (GSTM1) and theta 1 (GSTT1) as effect modifiers of the association between CA and cancer risk.
A case-control study was performed pooling data from cytogenetic studies carried out in 1974-1995 in three laboratories in Italy, Norway, and Denmark. A total of 107 cancer cases were retrieved from national registries and matched to 291 controls. The subjects were classified as low, medium, and high by tertile of CA frequency. The data were analyzed by setting up a Bayesian model that included prior information about cancer risk by CA frequency.
The association between CA frequency and cancer risk was confirmed [OR(medium) (odds ratio)(medium) = 1.5, 95% credibility interval (CrI), 0.9-2.5; OR(high) = 2.8, 95% CrI, 1.6-4.6], whereas no effect of the genetic polymorphism was observed. A much stronger association was seen in the Italian subset (OR(high)= 9.4, 95% CrI, 2.6-28.0), which was characterized by a lower technical variability of the cytogenetic analysis. CA level was particularly associated with cancer of the respiratory tract (OR(high)= 6.2, 95% CrI, 1.5-20.0), the genitourinary tract (OR(high) = 4.0, 95% CrI, 1.4-10.0), and the digestive tract (OR(high) = 2.8, 95% CrI, 1.2-5.8).
Despite the small size of the study groups, our results substantiate the cancer risk predictivity of CA frequency, ruling against a strong modifying effect of GSTM1 and GSTT1 polymorphisms.

Full text

Page 1

Environmental Health Perspectives • volume 117 | number 2 | February 2009
203
Research
During the last few decades, evidence con-
cerning the role of chromosomal aberrations
(CA) in carcinogenesis has been enriched by
a number of epidemiologic studies showing
that high CA frequencies in peripheral blood
lymphocytes of healthy individuals are asso-
ciated with increased cancer risk (Boffetta
et al. 2007; Bonassi et al. 1995, 2000; 2004;
Brøgger et al. 1990; Hagmar et al. 1994,
1998, 2004; Liou et al. 1999; Rossner
et al. 2005). A case–control study nested
within ESCH (European Study Group of
Cytogenetic Biomarkers and Health) cohort
studies (Bonassi et al. 1995; Hagmar et al.
1994, 1998) found that the strength of such
association is not influenced by occupa-
tional exposure to carcinogens or tobacco
smoking (Bonassi et al. 2000). Contrasting
results were reported by other investiga-
tions, such as a cohort study on Czech work-
ers (Smerhovsky et al. 2002) describing a
stronger association between CA frequency
and cancer incidence in miners exposed to
radon. Whether the CA–cancer association
reflects (occasionally undetected) exposure
to carcinogens, individual susceptibility to
carcinogens, some form of chromosomal
instability, or a causal role of chromosomal
rearrangements in the carcinogenic process is
still an open issue.
A striking amount of data supports the
hypothesis that individual characteristics
associated with cancer risk, such as inherited
differences in metabolic enzymes or DNA
repair capacity, may also modify CA occur-
rence (Norppa 2004). These findings raise the
obvious question of whether the association
between CA and cancer risk depends on indi-
vidual metabolism and DNA repair capabil-
ity, so that CA would better predict cancer
risk in people with an unfavorable genotype.
This question can be assessed by incorporat-
ing genotype data to the CA–cancer studies,
but such an approach has not been applied
previously, as DNA samples have not been
readily available.
Among the most frequently studied poly-
morphisms are those concerning the metabo-
lism of xenobiotics, in particular glutathione
S-transferases (GSTs). GSTs catalyze the
conjugation between glutathione and reac-
tive xenobiotic compounds in a pathway that
leads to thioethers excreted in the urine (Bolt
and Thier 2006; Parl 2005). The major bio-
logical function of GSTs is considered to be
protection against electrophilic chemical spe-
cies, although metabolic activation involving
GST-mediated glutathione conjugation has
also been described (e.g., for some chlorin-
ated substrates) (Bolt and Thier 2006; Parl
2005). Detoxification by glutathione con-
jugation can represent a minor (e.g., styrene
oxide) or a major (e.g., benzo[a]pyrene) met-
abolic pathway for many genotoxic agents.
Given the importance of GSTs in the detoxi-
fication of electrophilic carcinogens, the pos-
sible influence of polymorphisms in GST
genes on cancer risk has been investigated
extensively (Bolufer et al. 2006; Bolt and
Thier 2006; Carlsten et al. 2008; Hiyama
et al. 2008; Parl 2005; Shi et al. 2008; Vineis
et al. 2007; White et al. 2008). The GST M1
(GSTM1; BC036805, GenBank) and theta
1 (GSTT1; X79389, GenBank) genes have
received much attention because of the high
prevalence of homozygous deletions result-
ing in null genotypes with a decreased ability
to detoxify carcinogenic compounds, plac-
ing null individuals at increased cancer risk
(Bolufer et al. 2006; Bolt and Their 2006;
Address correspondence to S. Bonassi, Unit of
Molecular Epidemiology, National Cancer Research
Institute, Largo Rosanna Benzi, 10, 16132 Genoa,
Italy. Telephone: 390 10 5600924. Fax: 390
10-5600501. E-mail: stefano.bonassi@istge.it.
*These two authors contributed equally to the
present study.
This work was supported by grants from the
European Union 5th FP (QLRT-2000-00628) and its
extension (QLK4-CT-2002-02831), the Associazione
Italiana per la Ricerca sul Cancro (AIRC), the Italian
Ministry of Health, the Italian Space Agency (ASI),
the Telemark Hospital (Research and Development),
the Fondazione Buzzi-Unicem per la Ricerca sul
Mesotelioma, Casale (Italy), the Danish National
Programme of Environmental Health Research
on Cancer (0-302-02-4/2), and the Finnish Work
Environment Fund.
The Cancer Registry of Norway is not responsible
for the analysis or interpretation of the data presented.
The authors declare they have no competing
financial interests.
Received 6 June 2008; accepted 16 September
2008.
Association between Frequency of Chromosomal Aberrations and Cancer
Risk Is Not Influenced by Genetic Polymorphisms in GSTM1 and GSTT1
Anna Maria Rossi,1* Inger-Lise Hansteen,2* Camilla Furu Skjelbred,2 Michela Ballardin,1 Valentina Maggini,1
Elena Murgia,1 Antonio Tomei,1 Paolo Viarengo,3 Lisbeth E. Knudsen,4 Roberto Barale,1 Hannu Norppa,5
and Stefano Bonassi 6
1Department of Biology, Pisa University, Pisa, Italy, 2Department of Laboratory Medicine, Section of Medical Genetics, Telemark Hospital,
Skien, Norway; 3Department of Aerospace Engineering, University of Naples Federico II, Naples, Italy; 4Environmental Health, Institute of
Public Health, University of Copenhagen, Copenhagen, Denmark; 5New Technologies and Risks, Work Environment Development, Finnish
Institute of Occupational Health, Helsinki, Finland; 6Unit of Molecular Epidemiology, National Cancer Research Institute, Genoa, Italy
Background: The frequency of chromosomal aberrations (CA) in peripheral blood lymphocytes of
healthy individuals has been associated with cancer risk. It is presently unclear whether this associa-
tion is influenced by individual susceptibility factors such as genetic polymorphisms of xenobiotic-
metabolizing enzymes.
oBjectives: To evaluate the role of polymorphisms in glutathione S-transferase (GST) M1
(GSTM1) and theta 1 (GSTT1) as effect modifiers of the association between CA and cancer risk.
Methods: A case–control study was performed pooling data from cytogenetic studies carried out
in 1974–1995 in three laboratories in Italy, Norway, and Denmark. A total of 107 cancer cases
were retrieved from national registries and matched to 291 controls. The subjects were classified as
low, medium, and high by tertile of CA frequency. The data were analyzed by setting up a Bayesian
model that included prior information about cancer risk by CA frequency.
results: The association between CA frequency and cancer risk was confirmed [ORmedium (odds
ratio)medium = 1.5, 95% credibility interval (CrI), 0.9–2.5; ORhigh = 2.8, 95% CrI, 1.6–4.6], whereas
no effect of the genetic polymorphism was observed. A much stronger association was seen in the
Italian subset (ORhigh= 9.4, 95% CrI, 2.6–28.0), which was characterized by a lower technical vari-
ability of the cytogenetic analysis. CA level was particularly associated with cancer of the respiratory
tract (ORhigh= 6.2, 95% CrI, 1.5–20.0), the genitourinary tract (ORhigh = 4.0, 95% CrI, 1.4–10.0),
and the digestive tract (ORhigh = 2.8, 95% CrI, 1.2–5.8).
conclusions: Despite the small size of the study groups, our results substantiate the cancer risk
predictivity of CA frequency, ruling against a strong modifying effect of GSTM1 and GSTT1
polymorphisms.
key words: Bayesian, biomarker, cancer risk, case–control study, chromosomal aberration, genetic
polymorphism, glutathione S-transferase, GSTM1, GSTT1, Monte Carlo Markov Chain. Environ
Health Perspect 117:203–208 (2009). doi:10.1289/ehp.11769 available via http://dx.doi.org/
[Online September 2008]
hether this associa-
technical vari-

Page 2

Rossi et al.
204
volume 117 | number 2 | February 2009 • Environmental Health Perspectives
Hiyama et al. 2008; Parl 2005; Shi et al.
2008; White et al. 2008).
The availability in the ESCH database of
a group of subjects genotyped for GSTM1
and GSTT1 genes in three national labora-
tories allowed us to address the question if
these genetic polymorphisms can modify the
strength of the association between CA and
cancer risk. In a group of subjects from Pisa,
Italy—not included in the original ESCH
database—cytogenetic slides could be retrieved
from freezers, stained, and scored all together,
minimizing the impact of protocol variation,
reagents drift, and scorer heterogeneity.
The small size of study groups is a com-
mon limitation of biomarker validation
studies. The presence of sparse data and miss-
ing covariates are conditions for which the
Bayesian paradigm presents some advantages
over the frequentist paradigm. In particular,
Bayes’ rule opens the possibility of improving
the estimation of the parameters of interest
by including the prior proba bility, based on
knowledge already available—in this case, the
association of CA with cancer risk.
This study aimed at a) confirming the
presence of the association between CA and
cancer risk in a partially new data set, and b)
evaluating the role of GSTM1 and GSTT1
polymorphisms as effect modifiers of this
association. Ancillary but still important aims
were to evaluate the impact of reduced labo-
ratory/technical variability on risk estimates
and to verify whether Bayesian statistics can
provide reliable estimates in small strata.
Material and Methods
Subjects. A nested case–control study was
performed pooling data from cytogenetic
surveillance studies carried out from 1974
to 1995 at the Department of Biology,
University of Pisa, Pisa, Italy; the Department
of Occupational Medicine, Telemark
Hospital, Skien, Norway; and the Danish
National Institute of Occupational Health,
Copenhagen, Denmark. These laboratories
analyzed GSTM1 and GSTT1 polymorphisms
in subjects screened for CA and followed up
in the framework of the ESCH (Bonassi et al.
2000; Hagmar et al. 1998, 2004) and the more
recent CancerRiskBiomarkers (Cytogenetic
Biomarkers and Human Cancer Risk) (Norppa
et al. 2006) collaborative projects.
The Italian cohort consisted of 1,650
healthy subjects selected from the general popu-
lation living in three areas in Tuscany (Pisa,
Cascina, and Navacchio) enrolled between
1991 and 1993 (Barale et al. 1998a, 1998b;
Landi et al.1999; Milillo et al. 1996). The
Norwegian cohort included 681 healthy sub-
jects, either occupationally exposed or referents,
collected between 1974 and 1990 (Brøgger
et al. 1990; Hansteen et al. 1978, 1984). The
Danish data arose from a biomonitoring study
sparse data and miss-
conducted in 1987 in a group of 226 male
stainless steel welders and referents (Knudsen
et al. 1992). Information was gathered on
demographic, occupational, and lifestyle factors
at the time of blood sampling.
In the Nordic countries, all malignant
tumors diagnosed from the date of CA test-
ing until the end of the follow-up (2006
in Norway, and 2003 in Denmark) were
retrieved through linkage from national can-through linkage from national can-
cer registries. In Italy, the cause of death was
obtained from the municipality of residence
(2006). At the end of the follow-up, a total
of 107 cancer cases with information about
GST polymorphism were tracked (105/107
cases genotyped for GSTM1 and 77/107 for
GSTT1). Within each national cohort, con-
trols were matched with their corresponding
case by sex, age (± 10 years), and year of CA
test (± 5 years). The overall cases to controls
ratio was around 1:3 (107 cases/291 controls).
In all participating laboratories, the stand-
ard cytogenetic protocol was applied using
heparinized whole blood and harvesting the
cells after 48 hr of culture (Buckton and Evans
1973). CA preparations were stained with
Giemsa. In Italy, the slides were banked at
–20°C at the time of the original cytogenetic
study (1991–1993) and stained in Giemsa
immediately before the new CA analysis. In
Italy, 150 and in Denmark, 100 metaphases
per donor were scored for CA by a single
scorer. In Norway, 100 or (in more recent
studies) 200 cells per subject were scored by
the same three microscopists. Savage’s clas-
sification criteria for CA were used (Savage
1976). Total CA was defined as the number
of cells with aberrations, excluding gaps, per
100 cells. To standardize for interlaboratory
variation, all subjects were classified as low
(1–33 percentile), medium (34–66 percen-
tile), or high (67–100 percentile) according
to tertiles of CA frequency distribution. More
details on the laboratory protocols can be
found in previously published reports (Barale
et al. 1998a, 1998b; Bonassi et al. 2000;
Hagmar et al. 1994, 1998, 2004; Hansteen
et al. 1978, 1984; Knudsen et al. 1992; Landi
et al. 1999; Milillo et al. 1996). The study
protocols were approved by the ethics com-
mittees and legal authorities in each country.
DNA extraction and genotype analysis. In
Italy, genomic DNA was extracted from frozen
whole blood or serum as described by Murgia
et al. (2007). In Norway, DNA was extracted
from fixed cell suspensions with the method
described by Skjelbred et al. (2006). When
cell suspensions were not available, DNA was
extracted from unstained slides kept at room
temperature for 3–26 years. The slides were
hydrated, dried carefully before adding 100
µL buffer (50 mM Tris, 1 mM EDTA, 0.5%
Triton 100), covered with plastic film, kept for
15 min at 56°C, scraped in 50–100 µL dH2O
from national can-national can-
he stand-
2–4 times (total volume 200 µL), and cen-
trifuged at 3,000 × g for 5 min; 50 µL buffer
(as above) and 5 µL proteinase K (20 mg/
mL) were added to the pellet. The samples
were kept on a Thermomixer (Eppendorf,
Hamburg, Germany) at 56°C, 1,000 rpm for
3 hr, before deactivating the enzyme on the
Thermomixer at 96°C, 1,000 rpm for 10 min.
The suspensions were stored at –20°C. The
success rate depended upon the amount of
cells, the age of the slides, and the genotype
tested. In Denmark, DNA from mononuclear
blood cells, originally analyzed for DNA repair
by the unscheduled DNA synthesis technique
and stored frozen in 0.4 M phosphate buffer
in 1987, was used for the genotyping. This
DNA had been isolated by using Millipore
(Bedford, MA, USA) Microcon YM-30 cen-
trifugal filter devices.
In Italy, genotype analysis was performed
by a specific multiplex GSTT1/GSTM1 PCR
assay as described by Murgia et al. (2007).
In Norway, polymorphisms of the GSTM1
and GSTT1 (and GSTP1) genes were ana-
lyzed simultaneously by multiplex polymer-
ase chain reaction using primers described
by Nedelcheva Kristensen et al. (1998). A
detailed report of the methods used for geno-
typing the Danish samples can be found in
Ko et al. (2000).
Statistical analysis. As the response vari-
able in the study was binary (cancer yes/
no), we used the standard logistic regression
approach in which the response variable is
the log odds (or logit). The purpose was to
estimate the probability that the ith person
will develop a cancer, conditionally on the
information about his/her CA frequency Ei
[1 = L (low), 2 = M (medium), 3 = H (high)],
age Ai, sex Ei, and smoking habit Si, which are
the design variables (M1):
logit (pi) = β0 + β1Ei2+ β2Ei3

+ β3Gi + β4Ai + β5Si.

[1]
To perform the analysis, we used the Gibbs
sampler, which is a Monte Carlo Markov Chain
(MCMC) sampling algorithm. MCMC is a
class of methods for sampling from probability
distributions based on constructing Markov
chains. The iterative procedure simulates a
Markov chain, which has the desired posterior
distribution as its stationary distribution.
We used the statistical software WinBUGS
(http://www.mrc-bsu.cam.ac.uk) (Lunn et al.
2000) to set up a Bayesian model specify-
ing an informative normal prior distribution
for the regression coefficients where existing
knowledge was available. More precisely, from
the pooled cohort analysis of 22,358 subjects
examined for CAs in 11 countries (Bonassi
et al. 2008), we knew that the overall odds
ratio (OR) for the medium tertile compared
with the low was 1.3 with a 95% confidence

Page 3

CA, cancer risk, and genetic polymorphism
Environmental Health Perspectives • volume 117 | number 2 | February 2009
205
interval (CI) of 1.07–1.60, and that the OR
for the highest tertile was 1.4 (1.16–1.72).
The assumption that the log OR is normally
distributed leads to a normal distribution with
mean value given by ln(1.3) = 0.26 and ln(1.4)
= 0.33, respectively. The standard deviation of
the normal distributions was found by solv-
ing the equations 1n(1.3) + 1.96σ = 1n(1.60)
and 1n(1.4) + 1.96 σ = 1n(1.72), which gave
a standard deviation on the log odds scale of
0.1 in both cases. As suggested by Birkes and
Dodge (1993) and Gelman et al. (1995), a
simple way to provide a prior variance on the
parameters β1 and β2 in the current study was
to inflate the historical variance. To take into
account the heterogeneity and other differ-
ences between current and historical data, we
considered the prior variance as a Gamma ran-
dom variable which, on average, fluctuated
around the inflating factor (60 in our case).
Further prior information was obtained from
International Agency for Research on Cancer
(IARC) mono graph on tobacco smoking
(IARC 1986): (weighted) mean value of OR
(current vs. never smokers) was considered 10
for lung cancer, 1.4 for oral cavity–digestive
tract cancers, and 2.1 for genitourinary organ
cancers. With regard to the Bayesian interval
estimation of OR, we adopted the 95% cred-
ibility interval (CrI) with the highest posterior
density, that is, an interval of the OR posterior
distribution such that the area under the curve
is 0.95 and the density at any point inside the
interval is greater than the density at any point
outside. The traditional CI based on frequen-
tist inference refers to repeated sampling, ran-
dom intervals, and true parameter values (OR)
and does not provide any estimates concerning
the study that was actually conducted. On the
contrary, CrI is a probability statement within
the data set studied concerning the random
variable OR and specifying the range within
which the OR lies with 0.95 probability.
To estimate if the GSTM1 null and
GSTT1 null genotypes were a risk factor for
all cancers, we used a similar multivariate anal-
ysis, adding to the model M1 described in
Equation 1 the variables GSTM1, GSTT1 (we
called this model M2) and their interactions
within CA (we called this model M3). The
analysis performed to choose between compet-
ing models was based on the BIC index, BIC =
–21nLm + m1nn, where n is the sample size, m
is the difference of the number of parameters
in the models, and Lm is the likelihood ratio
for the models. This criterion penalized mod-
els that improved the fit, increasing the num-
ber of parameters. The model with the lower
value of BIC was the one to be preferred.
Results
The distribution of cases and controls in the
whole database by country and by other vari-
ables considered in the analysis is given in
Table 1. The proportion of males and of cur-
rent smokers was higher in the databases from
Norway and Denmark. Smoking habit was
not different between the cases and controls.
The Italian subjects selected from the general
population had a higher mean age, a higher
proportion of women, and fewer current
smokers than the Nordic groups. The distri-
bution by tumor site of the 107 cancer cases
included in the study is reported in Table 2.
The distribution of GST polymorphisms
did not differ significantly in the three coun-
tries, and therefore all data were combined
before analysis (Table 3). No significant dif-
ferences were evident in the genotype distribu-
tion of cases and controls, with null genotype
prevalence of 52.4 and 50.5% for GSTM1
and 29.9 and 29.8% for GSTT1, respectively.
Accordingly, the mean frequency of CA did not
appear to be modulated by genotype (Table 3).
Table 1. Number of cancer cases and controls classified according to sex, age, and smoking status.
Italy Norway
Group Case/control Total Case/control
Sex
Male 21/48 69 64/165
Female 8/23 31 2/6
Age (years)
≤ 39 0/0 0 13/44
40–47 2/5 7 12/41
48–55 1/13 14 20/39
56–61 9/20 29 7/34
≥ 62 17/33 50 14/13
Smokinga
Never 13/28 41 25/61
Former 9/23 32 7/21
Current 7/20 27 33/81
Total 29/71 100 66 /171
Denmark Total
Total

229
Case/control

12/49
0/0

5/16
1/19
4/10
2/4
0/0

4/22
5/19
3/8
12/49
Total Case/control Total (%)
61
0
97/262
10/29
359 (90.2)
39 (9.8)8

57
53
59
41
27
21
20
14
6
0
18/60
15/65
25/62
18/58
31/46
78 (19.6)
80 (20.1)
87 (21.9)
76 (19.1)
77 (19.3)

86
28
114
237
26
24
11
61
42/111
21/63
43/109
107/291
153 (38.4)
84 (21.1)
152 (38.2)
398 (100)
aAt the time of CA sampling. For one case and eight controls from Norway, no data were available on smoking status.
Table 2. Distribution of cancer cases included in the nested case–control study by site and country.

Tumor site (ICD–9 code)a Italy
Oral cavity (140–149) 0
Esophagus (150) 1
Stomach (151) 1
Intestine, colon, and rectum (152–154) 4
Liver (155) 3
Pancreas (157) 4
Larynx (161) 0
Lung (162) 6
Bone, skin (170–173) 0
Breast (174) 2
Uterus (179,182) 1
Ovary (183) 0
Prostate (185) 1
Bladder (188) 2
Kidney (189) 1
Other sites 3
Total 29
No. of cases
Norway
1
0
3
8
2
1
1
5
19
0
0
1
6
4
2
13
66
Percent of all
cases
1.9
0.9
3.7
12.1
4.7
4.7
0.9
11.2
18.7
1.9
0.9
0.9
6.5
6.5
2.8
21.5
100.0
Denmark
1
0
0
1
0
0
0
1
1
0
0
0
0
1
0
7
12
Total
2
1
4
13
5
5
1
12
20
2
1
1
7
7
3
23
107
aInternational Classification of Diseases, 9th Revision (WHO 1975).
Table 3. Distribution of GSTM1 and GSTT1 genotypes in cases and controls and mean CA frequency.a




Genotype Cases (%)
GSTM1
Null
Positive
All
GSTT1
Null
Positive
All


















Two-sample
Kolmogorov–Smirnov
test of identical
distribution functions
(null vs. positive)
Total Mean CA %
Controls (%)

141 (50.5)
138 (49.5)
279

65 (29.8)
153 (70.2)
218
Total (%)

196 (51)
188 (49)
384

88 (29.8)
207 (70.2)
295
Cases (SE) Controls (SE) Total (SE)

55 (52.4)
50 (47.6)
105

23 (29.9)
54 (70.1)
77
2.13 (0.2)
1.83 (0.2)
1.42 (0.1)
1.47 (0.1)
1.62 (0.1)
1.57 (0.1)
p = 0.69

1.78 (0.36)
1.80 (0.17)

1.20 (0.15)
1.46 (0.12)
1.35 (0.2)
1.55 (0.1)
p = 0.67
aBecause of the limited amount of DNA retrieved from stored specimen, only 105/107 cases were genotyped for GSTM1
and 77/107 for GSTT1.

Page 4

Rossi et al.
206
volume 117 | number 2 | February 2009 • Environmental Health Perspectives
The univariate analysis comparing the
overall mean CA frequency (SE) in cases and
controls [1.98 (0.14) vs. 1.38 (0.14)] revealed a
highly significant difference (t-test, p = 0.006).
At the national level, cancer cases had a sig-
nificantly higher mean CA frequency in the
Danish and Italian databases (data not shown).
The results of the multivariate Bayesian
model linking CA tertile and cancer risk are
reported in Table 4. No effect of sex or age were
observed; smoking habit increased the overall
risk of cancer but did not confound or modify
the effect of CA level on cancer risk. The overall
analysis showed a borderline risk increase for
subjects in the medium tertile of CA distri-
bution (OR = 1.5; 95% CrI, 0.88–2.50) and
an increase for those in the high tertile (OR =
2.8; 95% CrI, 1.6–4.6) when compared with
the lowest tertile. Increased cancer risks for the
medium and high tertiles were found in all
national data sets, although only in the high-
est tertile of the Italian subset did the CrI not
include 1 (OR = 9.4; 95% CrI, 2.6–28.0).
The possible role of GSTM1 and GSTT1
genetic polymorphisms as effect modifiers of
the association between CA frequency and
cancer risk was tested in the Bayesian mod-
els (model M3 vs. model M2). The analysis,
based on the BIC index, favored the model
without interaction terms (M2), ruling against
the hypothesis that the cancer risk predictivity
of CA frequency could be modified by these
polymorphisms, i.e., BIC(M3) = 481.9 >
BIC(M2) = 473.1 for GSTM1 and BIC(M3)
= 365.5 > BIC(M2) = 357.7 for GSTT1.
A similar multivariate model was used to
estimate if the GSTM1 null and GSTT1 null
genotypes were a risk factor for all cancers
(model M2 vs. model M1). Again, the BIC
index did not show any increase in the null
genotypes, as anticipated by the results of the
univariate analysis, i.e., BIC(M2) = 473.1 >
BIC(M1) = 467.3 for GSTM1 and BIC(M2)
= 357.7 > BIC(M1) = 352.6 for GSTT1.
Finally, we fitted multivariate Bayesian
models to major cancer sites to test the hypoth-
esis that CA frequency could more specifi-
cally predict the risk by cancer type (Table 5).
Credible associations (see statistical methods
for a definition of credibility intervals) were
found with the high tertile of the CA distribu-
tion for cancers of the genitourinary tract (OR
= 4.0; 95% CrI, 1.4–10.0); respiratory tract
(OR = 6.2; 95% CrI, 1.5–20.0), and digestive
tract (OR = 2.8; 95% CrI, 1.2–5.8).
Discussion
The results of this case–control study provide
new information for the validation of CA as a
biomarker of cancer risk. Although the study
groups were rather small, our findings sug-
gest that the association between CA and can-
cer is not modified by GSTM1 and GSTT1
polymorphisms, that the association is higher
when the evaluation is based on uniform cyto-
genetic data rather than pooled historical data,
and that the use of Bayesian modeling is a
credible approach for risk estimation in case of
sparse data, a common condition in biomarker
validation studies.
Our study offered a rare opportunity to
assess the impact of genetic polymorphisms on
the relationship between chromosomal dam-
age and cancer risk with a longitudinal design.
We were able to evaluate only two genotypes,
that is GSTM1 and GSTT1, because of the
limited amount of DNA available for each
subject. Previous data have suggested that the
null genotypes of both GSTM1 and GSTT1
are associated with a slightly increased risk of
some forms of cancer, although this does not
appear to concern all cancers (Bolufer et al.
2006; Bolt and Thier 2006; Carlsten et al.
2008; Hiyama et al. 2008; Parl 2005; Shi et al.
2008; Vineis et al. 2007; White et al. 2008).
There is also evidence in favor of GSTM1 and
GSTT1 genotypes affecting the level of CA in
lymphocytes especially in smokers, although
most studies have suggested no clear effects
of either polymorphism on the baseline level
of CA (Norppa 2004; Tuimala et al. 2004;
Vodicka et al. 2004). Furthermore, the pos-
sible presence of heterogeneity in the genetic
background of national populations, as well
as the activation of carcinogens occasionally
caused by polymorphic GSTs (Hu et al. 2006;
Kligerman and Hu 2007), may have weak-
ened the associations studied. Thus, our nega-
tive findings on the modifying effect of the
two GST polymorphisms on the association
between CA and cancer risk seem to agree with
the existing information about their negligible
influence on total cancer risk and baseline CA
level. Our conclusions are, however, limited
by the small size of the study groups, which
did not allow assessing possible interactions
between genotype and smoking. Moreover,
the genotype data were obtained using dif-
ferent sources of DNA, and thereby different
methods, for DNA isolation and genotyping
in the participating laboratories.
In general, the present findings confirmed,
largely with new data, the results published by
the ESCH in 2000 (Bonassi et al. 2000)—
that cancer prediction based on CA frequency
is independent of smoking and exposure to
carcinogens.
The polymorphisms of other genes, such
as those involved in DNA repair, might have
been relevant as well (Tuimala et al. 2004;
Vodicka et al. 2004). However, polymor-
phisms in the DNA repair genes OGG1
(8-oxoguanine DNA glycosylase; U96710,
GeneBank), XRCC1 (X-ray repair comple-
menting defective repair in Chinese ham-
ster cells 1; M36089, GeneBank), XRCC3
(X-ray repair complementing defective repair
in Chinese hamster cells 3; AF035586,
GeneBank), ERCC2 (exicision repair cross-
complementing rodent repair deficiency,
complementation group 2; HGNC:3434,
GeneBank), and the folate metabolism gene
MTHFR (5,10-methylenetetrahydrofolate
reductase; BC053509, GeneBank) were eval-
uated in the Norwegian data set (Skjelbred
et al. 2006) and explained only 4–10% of the
variance in CA, suggesting that these polymor-
phisms would not have strong effects on the
association between CA frequency and cancer
risk. Although many genetic polymorphisms
have been observed to affect the level of chro-
mosome damage (Norppa 2004; Skjelbred
et al. 2006; Tuimala et al. 2004; Vodicka et al.
2004), single or a few genes may be expected
to have only a small effect on CA frequency.
Most studies of CA and cancer risk have
been multicentric, and the cytogenetic data
generated during many years in a number of
laboratories are subject to methodologic vari-
ability due to multiple scorers and differences
in cell culture, slide preparation, staining, and
analysis. A unique feature of the current Italian
data was that the cytogenetic slides of the can-
cer cases and selected controls, originating from
surveillance studies performed in Pisa in the
early 1990s (Barale et al. 1998a, 1998b; Landi
et al. 1999; Milillo et al. 1996), were retrieved
from the freezer, stained, and consecutively
scored for CA by a single microscopist spe-
cifically for the present study. Most probably,
such procedure reduced technical variability,
and this is the most likely explanation for the
very high risk estimates in Pisa, previously
reported only by another case–control study
from Taiwan (Liou et al. 1999). This observa-
tion raises an interesting point concerning the
Table 4. Multivariate Bayesian estimates of cancer risk by tertiles of CA frequency by country.a

CA level
Low
Medium
High
Total
Italy Norway Denmark Total
Case/control
4/40
8/19
17/12
29/71
OR (95% CrI)
1.00
2.9 (0.8–8.3)
9.4 (2.6–28.0)

Case/control
22/73
22/58
22/40
66/171
OR (95% CrI)
1.00
1.3 (0.7–2.3)
1.9 (1.0– 3.4)

Case/control
6/27
1/10
5/12
12 /49
OR (95% CrI)
1.00
1.0 (0.2–2.7)
2.0 (0.6–5.1)

Case/Control
32/140
31/87
44/64
107/291
OR (95% CrI)
1.00
1.5 (0.9–2.5)
2.8 (1.6–4.6)
aEstimates based on 20,000 (MCMC) updates.

Page 5

CA, cancer risk, and genetic polymorphism
Environmental Health Perspectives • volume 117 | number 2 | February 2009
207
use of CA as a biomarker of cancer risk. The
cancer risk associated with a high frequency
of CA, although consistently detected in all
studies thus far published (Boffetta et al. 2007;
Bonassi et al. 1995, 2000, 2004; Brøgger et al.
1990; Hagmar et al. 1994, 1998, 2004; Liou
et al. 1999; Rossner et al. 2005), has generally
been low and variable. Therefore, it has been
considered that CA can hardly be applied for
individual risk assessment. If the findings of
this study are confirmed, and a stronger asso-
ciation between CA frequency and cancer risk
can be obtained by reducing technical variabil-
ity, the perspective for the use of this biomarker
in cancer prevention may have to be rethought.
Uniform CA analysis is a routine practice in
cytogenetic surveillance studies, which are
mostly interpreted at the group level, with
the exception of radiation biodosimetry. For
group-level evaluation, the present practice of
scoring 100–200 cells per person is adequate,
but it is likely not sufficient for reliable indi-
vidual CA assessment. However, if the level of
CA had, in reality, a stronger association with
cancer risk than has thus far been assumed, rea-
sonable individual cancer risk estimation might
become feasible with extended CA analysis.
At any rate, our findings lend further support
to earlier considerations (Bonassi et al. 2005;
Norppa et al. 2006) that the CA assay could be
used more widely in estimating cancer risk at
the group level.
Another innovative feature of the present
study was the potential provided by the
Bayesian approach for improving the reli-
ability of the estimates based on data from
small groups, as often happens with biomarker
validation studies. Most recent results from
large CA cohort studies, such as the Czech
cohort (Rossner et al. 2005) and the new
cohort of Central and Eastern European coun-
tries (Boffetta et al. 2007), described stronger
associations with specific cancers. In smaller
cohorts and in case–control studies, risk esti-
mates based on classic frequentist modeling are
likely to produce unreliable estimates in small
strata, and any likelihood-based analysis for
small data, or even worse, missing data, usually
involves computationally intensive methods or
ad hoc adjustments. Even without epidemio-
logic biases, the presence of small counts indi-
cates that large statistical biases may affect the
point estimates (Greenland et al. 2000).
In contrast, the Bayesian analysis allows
one to efficiently manage extra information,
sparse data, and missing covariates. In the
present study, MCMC sampling enabled us
to make inferences for any sample size without
resorting to asymptotic calculations. In many
cases, a frequentist inference can be obtained
as a special case of the Bayesian inference, and
when vague prior information is used, point
and interval estimates are similar to the fre-
quentist counterpart. However, the fundamen-
tal difference is the interpretation of intervals
of variability associated with risk estimates.
This is the best possible description of risk vari-
ability within the study context. If the study is
correctly designed and the prior information is
correctly selected, this risk range provides the
most credible assessment of risk variability.
Interestingly, the present results showed
the strongest associations of CA with the same
cancer types reported by the cohort studies
mentioned above, that is, cancers of the diges-
tive (Boffetta et al. 2007; Rossner et al. 2005)
and respiratory tract (Bonassi et al. 1995),
although the numbers of specific cancer cases
in the present evaluation were small.
In conclusion, this nested case–control
study provided relevant new information for
interpreting the role of chromosomal dam-
age in carcinogenesis. In particular, it sug-
gested that reducing technical variability in
the cytogenetic analysis may increase the
strength of the association between CA and
cancer risk, with possible implications for the
results provided by cohort studies published
so far, which were affected by large discrepan-
cies in laboratory protocols and scoring. The
negative result concerning the influence of
GSTM1 and GSTT1 polymorphisms, despite
the intrinsic limitation due to small numbers,
is in agreement with the idea that individual
polymorphisms are not expected to have a dra-
matic influence on baseline CA level or overall
cancer risk. Our findings support the hypoth-
esis that CA frequency, although indirectly
measured in surrogate tissues, can predict the
risk of cancer by itself as a phenotypic mani-
festation of multiple carcinogenic processes
or as an intermediate step of a causal process.
This conclusion is of great value for the use of
CA as biomarker in cancer prevention policies.
RefeRences
Barale R, Chelotti L, Davini T, Del Ry S, Andreassi MG, Ballardin
M, et al. 1998a. Sister chromatid exchange and micronu- Sister chromatid exchange and micronu-
cleus frequency in human lymphocytes of 1,650 subjects
in an Italian population. II. Contribution of sex, age, and
lifestyle. Environ Mol Mutagen 31:228–242.
Barale R, Marrazzini A, Bacci E, Di Sibio A, Tessa A, Cocchi L,
et al. 1998b. Sister chromatid exchange and micronucleus
frequency in human lymphocytes of 1,650 subjects in an
Italian population. I. Contribution of methodological fac-
tors. Environ Mol Mutagen 31:218–227.
Birkes D, Dodge Y. 1993. Alternative Methods of Regression.
New York:John Wiley & Sons.
Boffetta P, van der Hel O, Norppa H, Fabianova E, Fucic A,
Gundy S, et al. 2007. Chromosomal aberrations and cancer
risk: results of a cohort study from Central Europe. Am J
Epidemiol 165:36–43.
Bolt HM, Thier R. 2006. Relevance of the deletion poly morphisms
of the glutathione S-transferases GSTT1 and GSTM1 in
pharmacology and toxicology. Curr Drug Metab 7:613–628.
Bolufer P, Barragan E, Collado M, Cervera J, López JA, Sanz
MA. 2006. Influence of genetic polymorphisms on the risk
of developing leukemia and on disease progression. Leuk
Res 30:1471–1491.
Bonassi S, Abbondandolo A, Camurri L, Dal Prá L, De Ferrari M,
Degrassi F, et al. 1995. Are chromosome aberrations in cir-Are chromosome aberrations in cir-
culating lymphocytes predictive of a future cancer onset
in humans? Preliminary results of an Italian cohort study.
Cancer Genet Cytogenet 79:133–135.
Bonassi S, Hagmar L, Strömberg U, Huici Montagud A, Tinnerberg
H, Forni A, et al. for the European Study Group on Cytogenetic
Biomarkers and Health (ESCH). 2000. Chromosomal aberra-
tions in lymphocytes predict human cancer independently of
exposure to carcinogens. Cancer Res 60:1619–1625.
Bonassi S, Norppa H, Ceppi M, Strömberg U, Vermeulen R,
Znaor A, et al. 2008. Chromosomal aberration frequency
in lymphocytes predicts the risk of cancer: results from
a pooled cohort study of 22,358 subjects in 11 countries.
Carcinogenesis 29:1178–1183.
Bonassi S, Ugolini D, Kirsch-Volders M, Strömberg U,
Vermeulen R, Tucker JD. 2005. Human population studies
with cytogenetic biomarkers: review of the literature and
future prospects. Environ Mol Mutagen 45:258–270.
Bonassi S, Znaor A, Norppa H, Hagmar L. 2004. Chromosomal
aberrations and risk of cancer in humans: an epidemio-
logical perspective. Cytogenet Genome Res 104:376–382.
Brøgger A, Hagmar L, Hansteen IL, Heim S, Högstedt B,
Knudsen L, et al. 1990. An inter-Nordic prospective study
on cytogenetic endpoints and cancer risk. Nordic Study
Group on the Health Risk of Chromosome Damage. Cancer
Genet Cytogenet 45:85–92.
Buckton KE, Evans HJ, eds. 1973. Methods for the Analysis of
Human Chromosome Aberrations. Geneva:World Health
Organization.
Carlsten C, Sagoo GS, Frodsham AJ, Burke W, Higgins JP. 2008.
Glutathione S-transferase M1 (GSTM1) polymorphisms
and lung cancer: a literature-based systematic HuGE
review and meta-analysis. Am J Epidemiol 167:759–774.
Gelman A, Carlin J, Stern H, Rubin D. 1995. Bayesian data
analysis. London:Chapman & Hall.
Greenland S, Schwartzbaum JA, Finkle WD. 2000. Problems
due to small sample and sparse data in conditional logistic
regression Am J Epidemiol 151:531–539.
Hagmar L, Bonassi S, Strömberg U, Brøgger A, Knudsen L,
Norppa H, et al. 1998. Chromosomal aberrations in lympho-Chromosomal aberrations in lympho-
cytes predict human cancer – a report from the European
Study Group on Cytogenetic Biomarkers and Health
(ESCH). Cancer Res 58:4117–4121.
Hagmar L, Brøgger A, Hansteen IL, Heim S, Högstedt B,
Lambert B, et al. 1994. Cancer risk in humans predicted
by increased levels of chromosome aberrations in lym-
phocytes: Nordic Study Group on the Health Risk of
Chromosome Damage. Cancer Res 4:2919–2922.
Hagmar L, Strömberg U, Bonassi S, Hansteen I-L, Knudsen
LE, Lindholm C, et al. 2004. Impact of types of lymphocyte
chromosomal aberrations on human cancer risk: results
from Nordic and Italian cohorts. Cancer Res 64: 2258–2263.
Hansteen IL, Hillestad L, Thiis-Evensen E, Heldaas SS. 1978.
Effects of vinyl chloride in man: a cytogenetic follow-up
study. Mutat Res 51:271–278.
Hansteen IL, Jelmert O, Torgrimsen T, Forsund B. 1984. Low
human exposure to styrene in relation to chromosome
breaks, gaps and sister chromatid exchanges. Hereditas
100:87–91.
Hiyama T, Yoshihara M, Tanaka S, Chayama K. 2008. Genetic
polymorphisms and head and neck cancer risk. Int J Oncol
32:945–973.
Hu Y, Kabler SL, Tennant AH, Townsend AJ, Kligerman AD.
2006. Induction of DNA-protein crosslinks by dichlo-
romethane in a V79 cell line transfected with the murine
Table 5. Multivariate Bayesian risk estimates
of cancer risk by tertiles of CA frequency and by
cancer site.
Cancer site (ICD–9 code)
CA tertile
Oral cavity–digestive tract (140–159)
Low
Medium
High
Larynx/lung (160–169)
Low
Medium
High
Genitourinary organs/bladder (179–189)
Low
Medium
High
Case/control OR (95% CrI)
9/140
8/87
13/64
1.00
1.4 (0.6–2.8)
2.8 (1.2–5.8)
1/140
4/87
8/64
1.00
2.2 (0.6–6.1)
6.2 (1.5–20.0)
4/140
5/87
10/64
1.00
1.6 (0.5–3.8)
4.0 (1.4–10.0)

Page 6

Rossi et al.
208
volume 117 | number 2 | February 2009 • Environmental Health Perspectives
glutathione-S-transferase theta 1 gene. Mutat Res
607:231–239.
IARC (International Agency for Research on Cancer). 1986.
Tobacco Smoking. Monogr Eval Carcinog Risk Chem Hum.
Vol. 38. 1–142.
Kligerman AD, Hu Y. 2007. Some insights into the mode of
action of butadiene by examining the genotoxicity of its
metabolites. Chem Biol Interact 166:132–139.
Knudsen LE, Boisen T, Christensen JM, Jelnes JE, Jensen GE,
Jensen JC, et al. 1992. Biomonitoring of genotoxic exposure
among stainless steel welders. Mutat Res 279:129–143.
Ko Y, Koch B, Harth V, Sachinidis A, Thier R, Vetter H, et al.
2000. Rapid analysis of GSTM1, GSTT1 and GSTP1 poly-
morphisms using real-time polymerase chain reaction.
Pharmacogenetics 10:271–274.
Landi S, Frenzilli G, Milillo PC, Cocchi L, Sbrana I, Scapoli C,
et al. 1999. Spontaneous sister chromatid exchange and
chromosome aberration frequency in humans: the familial
effect. Mutat Res 444:337–345.
Liou SH, Lung JC, Chen YH, Yang T, Hsieh LL, Chen CJ, et al.
1999. Increased chromosome-type chromosome aberra-
tion frequencies as biomarkers of cancer risk in a black-
foot endemic area. Cancer Res 59:1481–1484.
Lunn DJ, Thomas A, Best N, Spiegelhalter D. 2000. WinBUGS –
a Bayesian modelling framework: concepts, structure, and
extensibility. Stat Comput 10:325–337.
Milillo CP, Gemignani F, Sbrana I, Carrozzi L, Viegi G, Barale R.
1996. Chromosome aberrations in humans in relation to
site of residence. Mutat Res 360:173–179.
Murgia E, Maggini V, Barale R, Rossi AM. 2007. Micronuclei,
genetic polymorphisms and cardiovascular disease mor-
tality in a nested case-control study in Italy. Mutat Res
621:113–118.
Nedelcheva Kristensen V, Andersen T I, Erikstein B, Geitvik G,
Skovlund E, Nesland JM, et al. 1998. Single tube multiplex
polymerase chain reaction geneotype analysis of GSTM1,
GSTT1 and GSTP1: relation of genotypes to TP53 tumor
status and clinicopathological variables in breast cancer
patients. Pharmacogenetics 8:441–447.
Norppa H. 2004. Cytogenetic biomarkers and genetic polymor-
phisms. Toxicol Lett 149:309–334.
Norppa H, Bonassi S, Hansteen I-L, Hagmar L, Strömberg U,
Rössner P, et al. 2006. Chromosomal aberrations and SCEs
as biomarkers of cancer risk. Mutat Res 600:37–45.
Parl FF. 2005. Glutathione S-transferase genotypes and cancer
risk. Cancer Lett 221:123–129.
Rossner P, Boffetta P, Ceppi M, Bonassi S, Smerhovsky Z,
Landa K, et al. 2005. Chromosomal aberration in lympho-
cytes of healthy subjects and risk of cancer. Environ Health
Perspect 113:517–520.
Savage JR. 1976. Classification and relationships of induced
chromosomal structural changes. J Med Genet 13:103–122.
Shi X, Zhou S, Wang Z, Zhou Z, Wang Z. 2008. CYP1A1 and
GSTM1 polymorphisms and lung cancer risk in Chinese
populations: a meta-analysis. Lung Cancer 59:155–163.
Skjelbred CF, Svendsen M, Haugan V, Eek AK, Clausen KO,
Svendsen MV, et al. 2006. Influence of DNA repair gene
polymorphisms of hOGG1, XRCC1, XRCC3, ERCC2 and the
folate metabolism gene MTHFR on chromosomal aberra-
tion frequencies. Mutat Res 602:151–162.
Smerhovsky Z, Landa K, Rössner P, Juzova D, Brabec M,
Zudova Z, et al. 2002. Increased risk of cancer in radon-
exposed miners with elevated frequency of chromosomal
aberrations. Mutat Res 514:165–176.
Tuimala J, Szekely G, Wikman H, Järventaus H, Hirvonen A,
Gundy S, et al. 2004. Genetic polymorphisms of DNA repair
and xenobiotic-metabolizing enzymes: effects on levels
of sister chromatid exchanges and chromosomal aberra-
tions. Mutat Res 554:319–333.
Vineis P, Anttila S, Benhamou S, Spinola M, Hirvonen A,
Kiyohara C, et al. 2007. Evidence of gene gene interac-
tions in lung carcinogenesis in a large pooled analysis.
Carcinogenesis 28:1902–1905.
Vodicka P, Kumar R, Stetina R, Sanyal S, Soucek P, Haufroid
V, et al. 2004. Genetic polymorphisms in DNA repair
genes and possible links with DNA repair rates, chro-
mosomal aberrations and single-strand breaks in DNA.
Carcinogenesis 25:757–763.
White DL, Li D, Nurgalieva Z, El-Serag HB. 2008. Genetic vari-
ants of glutathione S-transferase as possible risk factors-
for hepatocellular carcinoma: a HuGE systematic review
and meta-analysis. Am J Epidemiol 167:377–389.
WHO. 1975. International Classification of Diseases, 9th
Revision. Geneva:World Health Organization.
Genetic vari-