Variation in PAH-related DNA adduct levels among non-smokers: the role of multiple genetic polymorphisms and nucleotide excision repair phenotype

Abstract

Polycyclic aromatic hydrocarbons (PAHs) likely play a role in many cancers even in never-smokers. We tried to find a model to explain the relationship between variation in PAH-related DNA adduct levels among people with similar exposures, multiple genetic polymorphisms in genes related to metabolic and repair pathways, and nucleotide excision repair (NER) capacity. In 111 randomly-selected female never-smokers from the Golestan Cohort Study in Iran, we evaluated 21 SNPs in 14 genes related to xenobiotic metabolism and 12 SNPs in 8 DNA repair genes. NER capacity was evaluated by a modified comet assay, and aromatic DNA adduct levels were measured in blood by ³²P-postlabelling. Multivariable regression models were compared by Akaike’s information criterion (AIC). Aromatic DNA adduct levels ranged between 1.7 and 18.6 per 10⁸ nucleotides (mean: 5.8±3.1). DNA adduct level was significantly lower in homozygotes for NAT2 slow alleles and ERCC5 non risk-allele genotype, and was higher in the MPO homozygote risk-allele genotype. The sum of risk alleles in these genes significantly correlated with the log-adduct level (r=0.4, p<0.001). Compared with the environmental model, adding phase I SNPs and NER capacity provided the best fit, and could explain 17% more of the variation in adduct levels. NER capacity was affected by polymorphisms in the MTHFR and ERCC1 genes. Female non-smokers in this population had PAH-related DNA adduct levels 3-4 times higher than smokers and occupationally-exposed groups in previous studies, with large inter-individual variation which could best be explained by a combination of phase I genes and NER capacity.

Introduction

Polycyclic aromatic hydrocarbons (PAHs) are a class of carcinogens which are produced mainly by incomplete fuel combustion ¹. There are several studies documenting a possible role for PAH exposure in the etiology of esophageal squamous cell carcinoma (ESCC) ^2-7. In a recent report from Golestan Province, Iran, a high risk area for ESCC, Abedi-Ardekani and colleagues showed a striking dose-response relationship between the PAH content of non-tumoral esophageal biopsies, measured by immunohistochemical staining with antibodies raised against benzo[a]pyrene diol epoxide-I-modified guanosine, and ESCC case status ². Studies in this area and other high risk areas for ESCC in Brazil and along the central Asian esophageal cancer belt have also shown high exposures to PAHs ^{3-4, 6}. This high exposure has been observed among non-smokers as well as smokers, and food and coal-burning stoves have been suggested as potential sources ^{3, 5, 7-8}. Similarly, exposure to PAH has also been implicated in lung cancer pathogenesis in non-smokers ⁹.

Many PAHs do not cause biological changes by themselves, but do so only after being converted to active metabolites in the body ¹⁰. PAHs are first metabolized to more polar compounds by phase I enzymes. As a result of these reactions, reactive electrophilic intermediates may be formed that can bind to DNA to form DNA adducts, which are responsible for the mutagenic effects of PAHs; or they may be further metabolized by phase II enzymes that conjugate these electrophilic metabolites to form water-soluble compounds that can be excreted. DNA adduct formation is, in fact, an early biological event in the carcinogenic process ¹¹. DNA repair mechanisms, including nucleotide excision repair (NER), can restore the integrity of the damaged parts of DNA ¹².

While differences in environmental exposures are partly responsible for variations in DNA adduct formation, inter-individual variations among people with similar exposures may be explained partially by polymorphisms in genes that code for metabolizing or repair enzymes ¹⁰, and accumulating evidence suggests a role for genetic polymorphisms in PAH-related DNA adduct formation ¹³. So far, however, the results have been inconsistent regarding the roles of single polymorphisms, and many researchers now suggest that, due to the complex exposure and nature of PAH metabolism, a combination of genes in this metabolic pathway should instead be considered ¹⁰. For example, it has been shown that the sum of risk alleles in multiple PAH-metabolizing genes can affect adduct levels in smokers ¹⁴.

In this study, we evaluated a large number of polymorphisms in genes affecting DNA adduct formation and repair, along with NER phenotyping, in Golestan Province, a high-risk area for ESCC in northeastern Iran. Our main purpose was to identify the best models to explain inter-individual variations in adduct levels. Women in this area are not likely to be exposed to PAHs in an occupational setting, so non-smoking women constitute a relatively homogenous group in terms of PAH exposure. We restricted this study to female non-smokers in order to minimize the variation in adduct levels due to differences in environmental exposures.

Materials and Methods

Sampling

One hundred and eleven female non-smokers were randomly selected from subjects recruited in the Golestan Cohort Study (GCS), details of which have been published before ¹⁵. Briefly, 50,045 healthy adults from eastern Golestan Province were recruited into this cohort, including 48% men and 52% women, 80% from rural villages, and 74% of Turkmen ethnicity. Questionnaire data were collected about exposure to second-hand smoke at work or home; cooking methods in the household, including frying, boiling, baking, and barbequing; as well as the frequency of coal use for cooking. A standard food frequency questionnaire was administered, and the amounts of total and processed red meat consumption were calculated as potential sources of PAH.

Venous blood was collected from study participants, centrifuged at 800 g for 20 min, and the buffy coat was separated and stored at −70°C. After destroying RBCs using a lysis buffer (155 mmol/L NH₄Cl, 10 mmol/L KHCO₃, and 10 mmol/L EDTA), genomic DNA was extracted from the buffy coat by standard phenol extraction and stored in −70°C until genotyping and ³²P-postlabelling. For the NER assay, lymphocytes were isolated from venous blood using a standard density gradient centrifugation method¹⁶ and stored at −70°C. A single spot urine sample was collected from each participant and stored at −20°C.

The study was approved by the Institutional Review Board of the Digestive Disease Research Center of Tehran University of Medical Sciences.

Genotyping

Twenty-one SNPs in 14 genes related to xenobiotic metabolism and 12 SNPs in 8 DNA repair genes were studied (Table 1). SNP’s were selected on the basis of (a) their association with PAH metabolism and/or cancer development or (b) their expected influence on DNA repair activity¹⁷. These included genes for phase I metabolizing enzymes: cytochrome P450s (CYP1A1, CYP1A2, CYP1B1, CYP2E1, and CYP3A4), microsomal epoxide hydrolase 1 (mEH), and myeloperoxidase (MPO); genes for phase II enzymes: glutathione peroxidase 1 (GPX1), glutathione S-transferases (GSTM1, GSTP1, GSTT1), N-acetyltransferase 2 (NAT2), NAD(P)H quinone oxidoreductase 1 (NQO1), and manganese superoxide dismutase 2 (MnSOD2); and genes for DNA repair: apurinic-apyrimidinic endonuclease 1 (APEX1), DNA excision repair cross-complementing proteins (ERCC1, ERCC5, ERCC6), methylenetetrahydrofolate reductase (MTHFR), RAD23B, and the xeroderma pigmentosum group (XPA, XPC).

Table 1.

Details of the genes and polymorphisms evaluated in this study.

Set	gene	SNP ID	mutation	Putative risk allele* (frequency)	HWE p value	PCR multiplex
Phase I	CYP1A1	rs1048943	Ex7+131A>G	G (0.08)	0.4	2
		rs1799814	Ex7+129C>A	A (0.05)	0.2	2
	CYP1A2	rs762551	IVS1-154C>A	A (0.59)	0.03	1
	CYP1B1	rs1056836	Ex3+251G>C	C (0.27)	0.6	2
		rs1800440	Ex3+315A>G	G (0.20)	0.5	2
	CYP2E1	rs6413420	-70T>G	G (0.10)	0.3	2
	CYP3A4	rs2740574	-391A>G	G (0.02)	0.8	2
	mEH	rs2234922	Ex4+52A>G	A (0.55)	0.1	3
		rs1051740	Ex3-28T>C	C (0.49)	0.7	3
	MPO	rs2333227	-642G>A	A (0.36)	0.07	2
Phase II	GPX1	rs1050450	Ex1-226C>T	T (0.24)	0.2	3
	GSTM1		Ex4+10+>-	-/- (0.52)	-	1
	GSTP1	rs1695	Ex5-24A>G	G (0.41)	0.5	1
		rs1138272	Ex6+5C>T	C (0.25)	0.9	1
	GSTT1		Ex5-49+>-	-/- (0.89)	-	1
	NAT2	rs1801280	Ex2+347T>C	C (0.57)	0.9	1
		rs1799931	Ex2-313G>A	A (0.31)	0.9	1
		rs1799930	Ex2-580G>A	A (0.31)	0.5	1
	NQO1	rs1800566	Ex6+40C>T	T (0.09)	0.4	3
		rs1131341	Ex4-3C>T	T (0.27)	0.6	3
	MnSOD2	rs4880	Ex2+24T>C	C (0.18)	0.8	3
Repair	APEX1	rs1130409	Ex5+5T>G	G (0.34)	0.07	3
	ERCC1	rs11615	Ex4+33T>C	C (0.29)	0.6	4
		rs3212948	IVS3+74C>G	G (0.46)	0.4	4
		rs3212986	196+STP T>G	G (0.27)	0.8	4
	ERCC5	rs1047768	Ex2+50T>C	C (0.33)	0.5	4
	ERCC6	rs2228526	Ex18+219A>G	G (0.24)	0.7	4
	MTHFR	rs1801131	Ex8-62A>C	A (0.57)	0.4	1
		rs1801133	Ex5+79C>T	T (0.18)	0.2	1
	RAD23B	rs1805329	Ex7+65C>T	T (0.04)	0.8	4
	XPA	rs1800975	Ex1+62A>G	G (0.34)	0.6	4
	XPC	rs2228001	Ex16+211C>A	A (0.50)	0.8	4
		rs2228000	Ex9-377C>T	T (0.23)	0.7	4

SNP: single nucleotide polymorphism; PCR: polymerase chain reaction; MAF: minor allele frequency; HWE: Hardy-Weinberg equilibrium;

^aalleles which can increase adduct level as evidenced by previous studies or based on their effect on the enzyme activity.

Genotyping was performed by a SNaPshot procedure as described earlier ¹⁸ at the Department of Toxicology, Maastricht University (Maastricht, The Netherlands). Polymerase chain reaction (PCR) primers were designed using Primer 3 and Netprimer software, as detailed before ^{14, 19}. PCR was done in four separate multiplex reactions: two 10-plex and two 7-plex (shown as 1-4 in table 1) on a Tgradient 96-well thermal cycler (Biometra, Goettingen, Germany), each in a 10-μL volume using 96-well plates. The annealing temperatures were optimized for each multiplex PCR: 56°C for multiplex 1, 58°C for multiplex 2, 60°C for multiplex 3, and 60°C for multiplex 4. The PCR products were then incubated (37°C for 45 minutes) with 4 μ@L Exo-SAP-IT (Amersham, Roosendaal, The Netherlands) to digest contaminating deoxynucleotide triphosphates and PCR primers. Enzymes were deactivated at 75°C for 15 minutes. Subsequently, the single base extension (SBE) method by SnaPShot (Applied Biosystems, Nieuwekerk, a.d. IJssel, The Netherlands) was used for genotyping. A different multiplex SBE reaction was conducted on each multiplex PCR product ¹⁸. After genotyping, SBE products were analyzed on an ABI Prism 3100 genetic analyzer using Genscan Analysis software (version 3.7). Ten percent of the samples were randomly genotyped in a separate experiment and results were in 100% agreement with the first analysis.

Nucleotide Excision Repair

The nucleotide excision repair (NER) assay was performed at the Department of Toxicology, Maastricht University, by a modified single-cell alkaline gel electrophoresis (Comet) assay, which is based on the fact that the rate limiting steps in NER are the recognition and incision of the damaged DNA ²⁰. The capacity of the cells in performing these rate limiting steps can be quantified through formation of single-strand breaks measured by the comet assay. The NER capacity of the cell extracts is reflected by increased percentages of fluorescence in the tail (“comets”) ²¹. A549 cells (human epithelial lung carcinoma cells) were trypsinized at 80% confluency and diluted to a concentration of 2×10⁶ cells/ml one day before the assay. Aliquots (25 μl) of untreated A549 cells were mixed with 75 μl low melting point agarose and transferred to microscope slides, which were pre-coated with 1.5% normal electrophoresis grade agarose (Sigma–Aldrich, Germany). Slides were lysed overnight in cold (4°C) lysis buffer. On the day of the assay, the slides were washed and the resulting nucleoids were exposed to 1μM benzo[a]pyrene diol expoxide (BPDE) (NCI Chemical Carcinogen Reference Standard Repository, Midwest Research Institute, Kansas City, MO, USA) or vehicle control (DMSO, 0.5%) for 30 minutes.

The lymphocyte extracts were prepared using the method developed by Redaelli et al. ²² Fifty μl of this extract was added to each slide containing BPDE-exposed gel-embedded nucleoids, and incubated for 10 minutes at 37°C. Immediately after the incubation, slides were put on ice to stop the enzymatic reaction. The slides were further processed according to the conventional comet assay ²³. In brief, for DNA denaturation the slides were immersed in electrophoresis buffer (0.3 M NaOH, 1 mM EDTA, approximately pH 13) for 20 minutes, followed by 20 minutes of electrophoresis (25 V and 300 mA). The slides were neutralized to a pH of 7.4 and dried. The dried slides were stained with ethidium bromide (10 mg/ml), and comets were visualized using a Zeiss Axioskop fluorescence microscope. On each slide 50 random cells were analysed using Comet assay III software (Perceptive Instruments, Haverhill, UK). After subtracting background levels from all data, the final NER capacity was calculated by subtracting tail moments of the BPDE-/extract+ (non-adduct-containing nucleoids incubated with protein extract) and BPDE+/extract-(adduct-containing nucleoids exposed to BPDE only) slides from the BPDE+/extract+ (adduct-containing nucleoids incubated with protein extract) slides.

The NER capacity could be measured in only 81 subjects, because in 30 samples, either the amount of material provided was not sufficient to reliably measure DNA repair, or there was visible contamination with red blood cells, which made it impossible to make a representative cell extract.

Urinary 1-hydroxypyrene glucuronide concentration

1-hydroxypyrene glucuronide (1-OHPG) was measured in spot urine specimens at the Department of Environmental Health Sciences, Johns Hopkins Bloomberg School of Public Health (Baltimore, MD, USA) ²⁴. Synchronous fluorescence spectroscopy (Perkin Elmer LS50B Luminescence spectrometer, Norwalk, CT, USA) was used with a wavelength difference of 34 nm between excitation and emission. The limit of detection was 0.01 ng 1-OHPG/ml urine and the assay recovery was 95-100%.

Adduct ³²P-postlabelling

³²P-postlabelling analysis was carried out using the nuclease P1 enrichment with some modifications at the Institute of Cancer Research (Sutton, UK), as described previously ²⁵. Briefly, DNA (4 μg) was digested using micrococcal endonuclease (0.25 U/ml) and spleen phosphodiesterase (2 mg/ml) for 3.5 hours at 37°C. DNA digests were treated with nuclease P1 (0.96 μl, 1.25 mg/ml) for one hour at 37°C. To stop the nuclease P1-reaction, 1.92 ml Tris base (pH 9.6) was added. Labelling of PAH-modified nucleotides was done with [γ-³²P]ATP plus polynucleotide kinase (PNK) (6 U) for 30 min at 37°C. The radiolabelled adducted nucleotide bisphosphates were separated on PEI-cellulose sheets (Machery Nagel, Duren, Germany) by multidirectional thin layer chromatography (TLC). A standard with known levels of benzo[a]pyrene diol-epoxide(BPDE)-DNA adducts was analyzed in parallel for quantification purposes. Quantification was performed using an InstantImager (Canberra Packard, Dowers Grove, IL, USA). Assays were done in duplicate (or in triplicate when necessary) and the averages are reported. The assay coefficient of variation (CV) was 11.7% for the raw data and 10.2% for the log-transformed values.

Statistical analysis

Descriptive data are presented as mean ± standard deviation, and median (interquartile range: IQR). Allele and genotype frequencies were calculated for all SNPs and tested for Hardy-Weinberg equilibrium (Table 1). Since adduct levels, NER capacity, and the urinary 1-OHPG did not have normal distributions and were positively skewed, they were logarithmically transformed before being used in the analyses.

SNPs in the same gene were in linkage disequilibrium, as most D’ statistics were >0.9 (Supplementary Table). Polymorphism data were coded using a gene-based approach. First, for each SNP, using the functional information in the literature ^{17, 26}, the alleles potentially enhancing adduct formation (risk alleles) were determined (table 1). The “risk-allele score” was calculated for each gene in the following way: if at least one of the SNPs in a gene was homozygous for the risk allele, the gene was coded 2; if at least one of the SNPs was heterozygous and none were homozygous for the risk allele, it was coded 1; and if all the SNPs in the same gene had low-risk alleles, it was coded as having 0 risk alleles. For the NAT2 gene, the same scoring was used based on having a genotype with zero, one or two slow acetylation alleles in any of the three SNPs studied ²⁷. For GSTM1 and GSTT1, deletion, which results in the absence of the enzyme, was coded 2 (as opposed to 0 for those with the enzyme present).

In order to analyze the effect of individual genes, both univariate and multivariate methods were used. For univariate analyses, we used one-way analysis of variance to compare log-transformed adduct levels and NER capacity between individuals with different numbers of risk alleles. For multivariate analyses, a backward stepwise regression analysis was used. Log-transformed adduct levels were modeled against xenobiotic and repair genes (coded as described above) and the environmental factors (age, ethnicity, place of residence, passive smoking, red and processed meat intake, cooking method and coal use). Each group of genes (phase I, phase II, and repair) was also adjusted for the sum of risk allele scores in the other groups.

Full models were compared to determine the best model to explain inter-individual variation in DNA adduct levels. The base model included only the selected environmental factors. Other models were built by adding one component at a time to the base model (1-OHPG, phase I genes, phase II genes, repair genes, and NER capacity). Models were compared using Akaike’s information criteria (AIC):

AIC=−2 log L(θ)+2k

where L(θ) is the maximized likelihood function and k is the number of parameters in the model. In other words, AIC adds twice the number of parameters to the likelihood of the model as a penalty for overfitting. Lower AIC indicates a better fit for the model ²⁸.

Spearman correlation was used to assess the linear correlation between the number of risk alleles and the log-adduct levels. HaploStats R package was used to build regression models with haplotypes as predictors and log-adducts or log-NER as outcomes. Haplotypes which met these two criteria were used in the models: high linkage disequilibrium (D’>0.9 or r²>0.8) among the alleles, and a haplotype prevalence of at least 2%. We used Stata 11.2 (StataCorp LP, Tx) and snpStats R package version 1.2.1 for analyses. A two-sided p-value of less than 0.05 was considered to be significant.

Results

Sixty-seven (60%) of the 111 non-smoking women selected for this study were from rural villages, and 44 (40%) lived in the city. The mean age of the study population was 48.5±6.7 years, and 86 (78%) were of Turkmen ethnicity.

Adduct levels ranged from 1.7 to 18.6 adducts per 10⁸ nucleotides (Figure 1) with a median of 5.1 adducts per 10⁸ nucleotides (IQR: 3.7, 7.1). The mean level of aromatic DNA adducts was 5.8±3.1 adducts per 10⁸ nucleotides.

Figure 1.

Distribution of PAH-related DNA adduct levels among 111 female non-smokers in the Golestan Cohort Study

As Table 2 shows, in the univariate analyses, the levels of aromatic DNA adducts were significantly lower in homozygotes for at least one N-acetyltransferase 2 (NAT2) slow allele and in individuals with the excision repair cross-complementing 5 (ERCC5) non-risk-allele genotype, and were higher in individuals with the myeloperoxidase (MPO) homozygous risk-allele genotype (homozygous carriers of the −463G allele). The multivariate model to predict adduct level as the outcome included all of the genes in one group (phase I, phase II, or repair), and was adjusted for the total of risk-allele scores for the other groups, plus the environmental variables. None of the environmental factors showed a significant association with adduct levels in the multivariate model. Creatinin-adjusted 1-OHPG, a short-term marker of environmental exposure, also did not show a significant association with adduct levels (data not shown). In the adjusted model, the same genes had a significant effect on the adduct level (Table 2): MPO (β=0.21; p=0.01), NAT2 (β=-0.24; p=0.01), and ERCC5 (β=0.16; p=0.04). The sum of the risk allele scores in these three genes also had a significant linear correlation with the log-adduct level (r=0.4, p<0.001).

Table 2.

Univariate and multivariate analyses of the association between the combined risk allele score in each gene and PAH-DNA adduct levels.

Set	gene	Median adduct (IQR) by risk allele scorest†			adjusted model ††
		0	1	2	β	P value
Phase I	CYP1A1	5.0 (3.7,7.1)	5.3 (4.0,6.5)	9.2 (9.2,9.2)	−0.04	0.73
	CYP1A2	4.4 (3.6,6.0)	5.4 (4.4,7.1)	5.2 (3.3,7.3)	−0.05	0.54
	CYP1B1	5.4 (4.1,7.0)	5.2 (3.7,7.2)	4.4 (3.6,5.3)	−0.13	0.15
	CYP2E1	5.1 (3.8,7.1)	5.7 (3.7,7.1)	4.3 (3.3,5.4)	0.10	0.48
	CYP3A4	5.2 (3.8,7.1)	4.4 (4.0,5.5)	−	−0.20	0.43
	mEH	5.5 (3.0,7.1)	4.5 (3.9,6.1)	5.2 (3.3,7.6)	0.01	0.91
	MPO	4.3 (2.7,5.9)	4.4 (3.2,6.0)	5.3 (4.2,7.3)*	0.21	0.01
Phase II	GPX1	5.1 (3.8,7.4)	5.2 (3.7,6.3)	4.3 (2.3,6.7)	−0.08	0.36
	GSTM1	5.1 (3.3,7.1)	−	5.1 (4.2,7.1)	0.01	0.84
	GSTT1	5.2 (3.7,7.0)	−	5.1 (3.8,7.1)	0.01	0.88
	GSTP1	5.3 (3.7,7.1)	5.0 (3.8,7.1)	5.5 (3.9,6.0)	−0.09	0.29
	NAT2	6.9 (2.9,7.6)	5.3 (4.1,7.2)	4.2 (2.7,5.3)*	−0.24	0.01
	NQO1	5.3 (3.8,7.2)	5.1 (3.7,6.7)	4.6 (2.3,7.0)	−0.12	0.20
	MnSOD2	4.2 (3.7,5.3)	6.0 (4.4,7.7)	4.4 (3.3,5.6)	−0.04	0.65
Repair	APEX1	5.1 (3.7,5.7)	5.2 (4.0,7.1)	5.1 (3.2,7.1)	−0.01	0.97
	ERCC1	5.0 (4.5,5.5)	5.1 (3.6,7.1)	5.3 (4.1,7.1)	0.02	0.86
	ERCC5	4.4 (2.8,6.1)*	5.2 (4.0,7.1)	6.0 (4.1,7.3)	0.16	0.04
	ERCC6	5.0 (4.0,7.0)	5.3 (3.6,7.1)	5.0 (2.7,7.7)	−0.09	0.31
	MTHFR	5.3 (3.3,7.1)	4.8 (3.7,6.7)	5.4 (3.9, 7.1)	0.10	0.18
	RAD23B	5.2 (3.8,7.1)	4.5 (3.6,6.7)	5.1 (3.7,8.9)	−0.12	0.26
	XPA	4.5 (3.5,6.4)	5.5 (4.0,7.6)	5.4 (3.7,6.7)	0.06	0.50
	XPC	4.7 (4.1,5.7)	5.1 (3.9,7.1)	5.3 (3.6,7.0)	−0.01	0.90

IQR: inter-quartile range.

^†2: at least one SNP in the gene was homozygous for the risk allele; 1: at least one SNP in the gene was heterozygous and none were homozygous for the risk allele;0: all of the SNPs in the gene had low-risk alleles.

^††The model for polymorphisms in each group was adjusted for environmental exposures and the sum of risk allele scores for other groups;

^*p<0.05 comparing log-transformed adduct levels in one risk score category versus the other two.

NER capacity could be measured in 81 individuals. In these individuals, mean NER capacity was 0.83±0.68 (arbitrary units) and ranged between 0 and 2.49. As Table 3 shows, the homozygote non-risk-allele MTHFR genotype was significantly associated with lower NER capacity. MTHFR CA haplotype also showed a significant association with NER capacity (Supplementary table 2). Also, homozygotes for risk-allele genotype in ERCC1 had a higher NER capacity. NER capacity itself did not have a significant association with the DNA adduct level (data not shown).

Table 3.

Association between the combined risk allele score in repair genes and nucleotide excision repair (NER) capacity.

gene	Median NER capacity (IQR) by risk allele scorestδ
	0	1	2
APEX1	1.13 (0.73,1.26)	0.50 (0.29,1.04)	0.87 (0.42,1.42)
ERCC1	0.80 (0.56,2.03)	0.49 (0.17,1.10)	0.99 (0.60,1.20)*
ERCC5	0.56 (0.07,1.10)	0.75 (0.33,1.27)	0.57 (0.49,1.04)
ERCC6	0.61 (0.27,1.04)	0.61 (0.29,1.26)	1.19 (0.52,2.03)
MTHFR	0.36 (0.17,0.86)*	0.80 (0.44,1.20)	0.81 (0.33,1.42)
RAD23B	0.80 (0.27,1.42)	0.61 (0.33,1.10)	0.78 (0.37,1.19)
XPA	0.50 (0.14,0.99)	0.88 (0.42,1.42)	0.85 (0.40,1.04)
XPC	1.33 (0.44,2.21)	0.59 (0.32,1.20)	0.81 (0.30,1.12)

IQR: inter-quartile range.

^δ2: at least one SNP in the gene was homozygous for the risk allele; 1: at least one SNP in the gene was heterozygous and none were homozygous for the risk allele; 0: all of the SNPs in the gene had low-risk alleles;

^*p<0.05 comparing log-transformed NER capacity in one risk score category versus the other two.

In general, the models including phenotypically assessed NER capacity had a considerably better fit than the models without it. As Table 4 shows, the model that best fit the data (as evidenced by the lowest AIC) was the one including the environmental factors, phase I enzymes and NER capacity. This model had an r² of 0.24, and compared with the 0.07 r² of the model including environmental model, it could explain 17% more of the variation in DNA adduct levels. The model deteriorated slightly when the phase II genes were added. Also, the models had a better fit when phenotypically assessed NER was used instead of the polymorphisms in repair genes.

Table 4.

Comparison of different models in predicting DNA adduct levels

Variables in the Model*	AIC
a. Environmental	156.33
b. Environmental + 1-ohpg	157.84
c. Environmental+ phase I polymorphisms	151.19
d. Environmental+ phase II polymorphisms	157.10
e. Environmental + phase I and phase II polymorphisms	154.57
f. Environmental + repair genes	160.68
g. Environmental + phase I polymorphisms + repair genes	157.49
h. Environmental + phase I and phase II polymorphisms + repair genes	158.79
i. Environmental + NER capacity	102.66
j. Environmental + phase I polymorphisms + NER capacity	100.07
k. Environmental + phase II polymorphisms + NER capacity	105.02
l. Environmental + phase I and phase II polymorphisms + NER capacity	106.50

AIC: Akaike's information criterion (lower AIC indicates a better fit of the model); NER: Nucleotide excision repair;

^*The base model included only environmental factors; other models were built by adding one group of variables at a time to the base model.

Discussion

In this study we observed high levels of PAH-related DNA adducts, with substantial inter-individual variation. The model including phase I gene variant scores and measured NER capacity seemed to explain this variability better than other models, although individual SNPs from phase I, phase II and repair genes also showed significant associations with adduct levels.

In our study, mean aromatic DNA adduct level was 5.8±3.1 adducts per 10⁸ nucleotides. These findings are in line with previous findings of a very high level of exposure to PAH in this region ³. The source of this high PAH exposure is not very well understood, but studies suggest that food, and its method of preparation might play a role⁸. In any case, we ruled out external contamination by careful quality control of sample collection and processing. In Dutch smokers, the mean DNA adduct level has been reported to be 1.40±0.79 adducts per 10⁸ nucleotides, with a range between 0.25 and 3.90 ¹⁴. Binkova et al. reported PAH adducts in several occupationally-exposed groups, such as Czech and Slovak coke-oven workers, and even among the smokers in these groups, they observed levels much lower than those in the present study ^29-30. In another study of occupationally-exposed workers from China, another country with a high rate of esophageal squamous cell carcinoma, the highest mean PAH-related DNA adduct level, among coke oven workers, was 1.66±0.21 adducts per 10⁸ nucleotides ³¹. All of these adduct levels were measured using ³²P-postlabelling, and variations in postlabelling results from different laboratories may contribute to some of these differences. Nevertheless, both our measurements and those in the Dutch study were done in laboratories which participated in a study to make postlabelling protocols more uniform across Europe, which significantly improved assay reliability in the participating laboratories ³². In general, ³²P-postlabelling has proven to be a relatively reliable method for measuring aromatic DNA adducts, and has improved over the years ³³, and in our study, we observed a good assay CV (10.2%) in the duplicate samples.

Measurement of DNA adducts can be considered as an intermediate marker of cancer risk, which reflects the biologically effective dose of PAH exposure ^{30, 34}. As in our study, however, SNPs in some individual genes may not show significant associations with DNA adduct levels, in spite of the crucial role of these genes in coding enzymes which metabolize PAHs. This low level of association may be caused by several factors including a large degree of variation in environmental exposures and unmeasured factors which affect adduct levels, and insufficient power of the study due to the low frequency of the studied polymorphisms. In this setting, model comparison using information criteria such as AIC offers two advantages over traditional hypothesis-testing: firstly it does not rely on a predefined significance level, and its power in determining the best model is not dependent on study power or sample size ³⁵. In general, if the difference between the AIC of the “best” model and other models is more than 2, the model has considerable support, and if this difference is above 10, this support is very strong ³⁵. In our study models with NER capacity had the highest AIC difference, followed by those including phase 1 enzymes. To the best of our knowledge, this is the first study showing the importance of NER phenotyping in predicting aromatic DNA adduct levels. Compared to genotyping, repair phenotyping provides a better indication of the cells’ enzymatic repair capacity ¹⁷.

The second advantage of AIC, is that it assesses the whole model rather than its individual components. In the present study we found significant results for only one of the phase I genes, and the effect of NER capacity on DNA adduct levels was not significant by itself, but these two together comprised the best model. This underlines the importance of a comprehensive approach to evaluating pathways affecting cancer risk or carcinogenic metabolism. One of the approaches proposed for such analysis is the consideration of simultaneous genetic variations in multiple genes ³⁶. In a study on 170 healthy volunteers, Butkiewicz et al. ³⁷ showed that although a GSTP1 polymorphism alone was not associated with a change in DNA adduct level, when combined with GSTM1 deletion, it was associated with increased adduct formation. Georgiadis and colleagues ³⁸ also used a similar combination method, and showed a synergistic effect between phase I and phase II genes. The limitation of such an approach is that with increasing numbers of genes, and polymorphisms within each gene, interpretation becomes very complicated as the number of different possible combinations increases dramatically.

In a different approach, the total number of risk alleles can be used to summarize the genetic risk profile of an individual. Ketelslegers et al. used this method and observed an association between the number of risk alleles and DNA adduct levels in smokers ¹⁴. The sum of three risk alleles in their study (GSTM1, mEH and GPX) was associated with a two-fold increase in adduct level. We used the same approach for three different genes (MPO, NAT2 and ERCC5) and observed a linear association. We also adopted a gene-based approach by combining the polymorphic alleles in each gene, which were known to act in the same direction and were in linkage disequilibrium, and used modeling to identify the combination of different alleles that could best explain variability in adduct levels. In this way, SNP’s in phase I genes appeared to influence DNA adduct formation more than SNP’s in phase II genes. Agudo and colleagues ³⁹ also used a multivariate regression model in the EPIC-Spain cohort and defined “imputed phenotypes” based on the combinations of genotypes in defined pathways. They observed significant changes in DNA adduct levels caused by combined oxidation-hydrolysis (phase I enzymes), acetylation (NAT2), and oxidation-acetylation pathways, which is in line with our findings. Abnet and colleagues ⁴⁰ studied a group of individuals from another ESCC high risk area in Brazil, using a multivariate approach. They showed that adding phase I polymorphisms to environmental exposures increased the ability of the model to predict 1-OHPG concentration by 12%. 1-OHPG is a biomarker that has been extensively used to evaluate PAH exposure in general population and occupational groups. This marker, however, has a short half-life and may not be suitable for studying the long-term carcinogenic effects of PAHs and their metabolites ⁴⁰. In our study, 1-OHPG was not associated with the adduct level and did not improve adduct formation model fit.

We found several interesting associations with individual genes. The finding that MPO genotype influences PAH-related DNA adducts is consistent in the literature and is related both to its important role in the benzo-[a]pyrene (BaP) activation as a result of oxidation of B[a]P to BaP-7,8-diol and ultimately BPDE^41-42, and to its ability to diminish DNA repair via the local production of hypochlorous acid (HOCl) at inflammation sites^43-44. Rojas et al.⁴⁵ showed that BPDE-DNA adduct formation was decreased in the skin of coal-tar treated patients carrying the MPO mutant allele. In another study, this polymorphism was shown to reduce DNA adduct levels in bronchoalveolar lavage specimens from smokers ⁴⁶.

Studies on the effects of NAT2 polymorphisms on adduct formation have had different results. Godschalk et al. ¹⁰ showed a lower level of DNA adducts in the fast-acetylation group. In contrast, in the EPIC-Spain cohort, slow acetylation by NAT2 was associated with a significantly decreased adduct level in healthy adults ³⁹, similar to our results. Lee et al. ⁴⁷ showed that slow acetylation was associated with increased DNA adducts in the lung, but a non-significant reduction of adducts in the blood monocytes of lung cancer patients. Substrate-specific action of NAT has been proposed as a possible reason for such differences ³⁹.

Some of the repair genes also showed associations with either DNA adduct level or NER capacity. ERCC-5 was the only repair gene in our data which had a significant effect on DNA adduct level. The ERCC-5 C/C genotype which was associated with increased adduct level in our study, has been shown to increase the risk of lung cancer ⁴⁸. We also showed that NER capacity itself is increased in individuals with MTHFR and ERCC1 polymorphisms. Two of the SNPs we studied in the ERCC1 gene (rs11615 T allele and rs3212986 C allele) have been shown to increase DNA adduct levels ⁴⁹, but this is the first study to show the association between this gene and NER capacity.

Although the role of MTHFR in DNA repair may not be straightforward, it has been shown that the rs1801133 MTHFR T allele could increase 5,10-methylenetetrahydrofolate levels, which are necessary for DNA repair ⁵⁰. Choi and colleagues⁵¹ have shown reduced excision repair in folate-depleted rats. However, unlike our assay, which quantifies the comet formation due to single-strand breaks, this reduced repair was reflected by the slowed recovery of the damage-induced comets as a result of insufficient thymidylate and purine production. Besides it is not clear how much dietary intake, in the ranges higher than the depletion investigated in Choi’s study, can affect DNA repair. We did not measure folate intake in our study to further investigate the possibility of such an association, and it is unlikely that our study population had a very heterogeneous pattern of folate intake.

This study has one of the largest sets of PAH metabolizing and repair genes that have been studied together, and it combines repair phenotyping and genotyping in one study. Also, almost all previous studies, unlike the present one, have included only smokers or groups with occupational exposures. Given the importance of PAH exposure in non-smokers and its relevance to ESCC and lung cancer in this group of individuals, this study can contribute to our understanding of PAH-related carcinogenesis in this group. One of the limitations of this study is its relatively small sample size. So, as in many similar studies, the sample size did not allow more sophisticated pathway analyses, and may have been the reason for the lack of association between NER capacity and adduct levels. We also did not study more genes in the pathway, which might explain some of the remaining 76% of the variation in the adduct level, and had to assume linearity of effects when using regression models, which is not necessarily the case.

In conclusion, we have observed high levels of PAH-related DNA adducts in female non-smokers in this population, and found interesting associations between MPO, NAT2 and ERCC5 polymorphisms and DNA adduct levels. We have also shown the importance of studying combined genetic polymorphisms and NER phenotyping when studying cancer risk in relation to PAH exposure. Finally, we have shown that inter-individual variation in adduct levels could best be explained by a combination of polymorphisms in phase I genes and NER capacity.

Novelty and Impact.

In a population at high risk of esophageal squamous cell carcinoma (ESCC), female non-smokers had exceptionally high aromatic DNA adducts, with large inter-individual variation best explained by a combination of phase I genes and nucleotide excision repair capacity. This study has one of the largest sets of PAH metabolizing and repair genes studied together along with repair phenotyping, and uses novel methods in modeling an important biomarker in the carcinogenesis of ESCC and lung cancer.