Genetic variation in the bioactivation pathway for polycyclic hydrocarbons and heterocyclic amines in relation to risk of colorectal neoplasia

Abstract

Animal work implicates chemical carcinogens, such as polycyclic aromatic hydrocarbons (PAHs) and heterocyclic aromatic amines (HAAs) as contributing to the development of colorectal cancer (CRC). The epidemiologic evidence, however, remains inconsistent possibly due to intra-individual variation in bioactivation of these compounds. We conducted a case–control study of colorectal adenoma (914 cases, 1185 controls) and CRC (496 cases, 607 controls) among Japanese Americans, European Americans and Native Hawaiians to investigate the association of genetic variation in the PAH and HAA bioactivation pathway (CYP1A1, CYP1A2, CYP1B1, AHR and ARNT) identified through sequencing with risk of colorectal neoplasia, as well as their interactions with smoking and intakes of red meat and HAAs. The A allele for ARNT rs12410394 was significantly inversely associated with CRC [odds ratios (ORs) and 95% confidence intervals (CIs) for GG, AG and AA genotypes: 1.00, 0.66 (0.48–0.89), 0.54 (0.37–0.78), P_trend = 0.0008] after multiple comparison adjustment. CYP1A2 rs11072508 was marginally significantly associated with CRC, where each copy of the T allele was associated with reduced risk (OR: 0.72, 95% CI: 0.58–0.88, P_trend = 0.0017). No heterogeneity of genetic effects across racial/ethnic groups was detected. In addition, no significant interaction was observed after adjusting for multiple testing between genetic variants and pack-years of smoking, intake of red meat or HAAs (PhIP, MeIQx, Di-MeIQx or total HAAs) or NAT2 genotype (Rapid versus Slow or Intermediate). This study suggests that the genomic region around ARNT rs12410394 may harbor variants associated with CRC.

Introduction

Much experimental work points to the role of chemical carcinogens in colorectal cancer (CRC), such as polycyclic aromatic hydrocarbons (PAHs) and heterocyclic aromatic amines (HAAs). These compounds are present in tobacco smoke and are formed in meat when cooked on an open flame or at a high temperature for a long duration (1). Biotransformation by Phase I enzymes is required to activate these chemicals into ultimate carcinogens that can initiate carcinogenesis (2–4). Although suggestive, the epidemiologic evidence remains inconsistent possibly because of the difficulty in assessing exposure to these compounds through questionnaires and/or intra-individual variation in the bioactivation process.

CYP1A1, CYP1A2 and CYP1B1, the main enzymes that bioactivate PAHs and HAAs (1), are induced by PAHs and other toxicants (e.g. from smoking) via the effect of the transcription factor aromatic hydrocarbon receptor (AHR), which dimerizes with aromatic hydrocarbon receptor nuclear translocator (ARNT) before interacting with the promoters of the corresponding CYP genes (5,6). Although there exists a large inter-individual variation in the activity of these enzymes, in contrast to what has been found for CYP1A1 and CYP1B1, no polymorphism in the CYP1A2 gene has been shown to substantially affect CYP1A2 activity (7–10). There is also evidence that CYP1A1 induction is associated with specific AHR variants (11). Finally, NAT2 is a Phase II enzyme that is involved in the further bioactivation of HAAs in the colon through O-acetylation (12). Seven common (frequency > 1%) NAT2 genetic variants can be used to predict all common allelic variants responsible for the rapid, intermediate or slow acetylation phenotype (13).

Genetic polymorphisms in this biotransformation pathway have been hypothesized to potentiate the effects of PAHs and HAAs and contribute to individual risk for CRC. However, no study has comprehensively examined common genetic variation in CYP1A1, CYP1A2, CYP1B1, AHR and ARNT. We surveyed the genomic regions covering these five genes and assessed the association of common genetic variation at these loci with risk of colorectal neoplasia in two case–control studies—one of CRC and the other of colorectal adenoma. We also assessed the modifying effects of these genetic variants on the associations of HAA and PAH exposure (smoking and intakes of meat and HAAs) and NAT2 phenotype with risk of colorectal neoplasia. Since adenomas are regarded as precursor lesions for most CRCs (14), we studied both conditions with the same methodology to investigate whether these risk factors may be involved in the progression from dysplastic epithelium to invasive carcinoma.

Methods

The study design and data collection for the adenoma and CRC studies have been described in detail elsewhere (15,16). For the adenoma study, two flexible-sigmoidoscopy screening clinics were first used to recruit participants on Oahu, Hawaii. Adenoma cases were identified either from the baseline examination at the Hawaii site of the Prostate Lung Colorectal and Ovarian cancer screening trial during 1996–2000 or at the Kaiser Permanente Hawaii’s Gastroenterology Screening Clinic during 1995–2007. In addition, starting in 2002 and up to 2007, we also approached for recruitment all eligible patients who underwent a colonoscopy in the Kaiser Permanente Hawaii Gastroenterology Department. Cases were patients with histologically confirmed first-time adenoma(s) of the colorectum and were of Japanese, Caucasian or Hawaiian race/ethnicity. Controls were selected among patients with a normal colorectum and were individually matched to the cases on age at exam, sex, race/ethnicity, screening date (±3 months) and clinic and type of examination (colonoscopy or flexible sigmoidoscopy). We recruited 1016 adenoma cases (67.8% of all eligible) and 1355 controls (69.2% of all eligible); 889 cases and 1169 controls agreed to give a blood and 29 cases and 34 controls, a mouthwash sample.

For the CRC study (15), incident adenocarcinoma cases of Japanese, European or Native Hawaiian ancestry, residing on Oahu and diagnosed during 1994–99 were identified through the Hawaii Surveillance, Epidemiology and End Results cancer registry Controls were selected from participants in an ongoing health survey conducted by the Hawaii State Department of Health among an annual random 2% sample of households within the state. An additional source of controls ≥65 years of age was Medicare participants on Oahu. One control was individually matched to each case by sex, ethnicity and age (±2 years). Participation rates were 58% for case and 53% for controls. A total of 498 CRC cases and 609 population controls had blood DNA available for this study.

In both studies, the same interview-administered questionnaire was used to collect exposure information, including lifetime histories of tobacco smoking and alcohol drinking, lifetime history of recreational exercise, aspirin and dietary supplement use, family history of CRC and for females, hormone use history. The interview also included a food frequency questionnaire with >200 food items (17,18) and a meat module assessing frequency of consumption and degree of doneness for various meats cooked with high-temperature methods (broiling, barbecuing/grilling and pan-frying), from which HAA intakes (PhIP, MeIQx, Di-MeIQx) were calculated (19). Blood samples were processed within 2 h of collection and stored at −80°C until DNA extraction using QIAamp Blood Kits (Qiagen, Valencia, CA).

To infer the NAT2 acetylation phenotype, we genotyped seven variants in NAT2 in the adenoma study, [G191A (R64Q), C282T, T341C (I114T), C481T, G590A (R197Q), A803G (K268R) and G857A (G286T)], allowing for the detection of 26 of the more common alleles (NAT2*4; NAT2*5A,B,C,D,E,G,J; NAT2*6A,B,C,E; NAT2*7A,B; NAT2*11A; NAT2*12A,B,C; NAT2*13; NAT2*14A,B,C,D,E,F,G) (13). Individuals with two ‘rapid’ alleles (NAT2*4, NAT2*11A, NAT2*12A, B, C and NAT2*13), one rapid and one ‘slow’ allele (all other alleles) and two slow alleles were assigned to rapid, intermediate and slow NAT2 phenotypes, respectively. Genotyping in the CRC study was limited to the C481T, G590A and G857A variants that are diagnostic for the most common alleles (NAT2*5A,B, NAT2*6A,B, NAT2*7A,B, NAT2*5C) (15).

Because this study was initiated before the completion of Phase I of the HapMap project, genetic variation in CYP1A1, CYP1A2, CYP1B1, AHR and ARNT was first characterized through exon resequencing in 49 CRC cases of European, Japanese and Hawaiian ancestry from the CRC study and of African American and Latino ancestry from the Multiethnic Cohort Study (20). We also surveyed public databases for single nucleotide polymorphisms in each gene, including 20 kb upstream and 10 kb downstream from the coding regions. SNPs were selected approximately every 1 kb to provide dense coverage. Preference was given to SNPs previously reported to be associated with disease and to non-synonymous SNPs. Other considerations included heterozygosity with a minor allele frequency (MAF) of at least 5% and being in an evolutionary conserved region. The selected SNPs were then genotyped in a multiethnic panel of Japanese American (n = 70), European American (n = 70), Native Hawaiian (n = 70), African–American (n = 70) and Latino (n = 70) samples from the Multiethnic Cohort. These data were used to select tag-SNPs that captured ≥80% of total genetic variants identified through sequencing. The tag-SNPs were genotyped in the adenoma and CRC studies with the 5′ nuclease TaqMan allelic discrimination assay, where primers and probes were designed using the PrimerExpress software (Applied Biosystems, Foster City, CA).

Fifty-eight tag-SNPs were genotyped in both studies, including one insertion/deletion polymorphism. Among the 58 markers, nine were missense variants. Hardy–Weinberg Equilibrium(HWE) test was performed in controls of each study within each ethnic group using the χ² Goodness-of-fit test or the exact test implemented in Pedstats when MAFs were low (21). Among the 58 variants, 3 SNPs were in almost complete linkage disequilibrium with another variant (pairwise r² > 0.99), 6 had a MAF < 0.01 in either study, 2 SNPs showed deviation from HWE in at least one racial group (P_HWE < 0.001) and 1 SNP had a <95% call rate among the genotyped subjects—these 12 SNPs were excluded from further analysis. In addition, a single SNP with a <95% concordance rate among duplicate samples was excluded. After performing these quality control filters, the concordance rate across duplicates was >99% for the adenoma study and >96% for the CRC study. Subjects in both studies with a call rate of <60% across all genotyped loci were also excluded. After the exclusion, all subjects in the adenoma study and 1087 CRC subjects had a call rate >92%; the rest 16 CRC subjects had call rates between 68 and 90% (11 rates ≥ 75%).

Data from the adenoma and CRC studies were analyzed separately but with the same strategy. We computed odds ratios (ORs) and 95% confidence intervals (CIs) for each variant, using unconditional logistic regression under a log-additive genetic model. Variants with MAFs < 0.05 were also analyzed under a dominant model (i.e. pooling subjects carrying one or two minor alleles). In minimally adjusted models, we controlled only for study-specific matching factors—age, sex and ethnicity (and recruitment site and endoscopic procedure for the adenoma study). In fully adjusted models, additional risk factors were included: body mass index, pack-years of smoking, alcohol intake (using quartiles), calorie-adjusted total folate intake (≤400; 400 to ≤1000; ≥1000 Dietary Folate Equivalents) and lifetime hours of recreational physical activity in the adenoma study and body mass index 5 years ago, pack-years of smoking (0; 0 to ≤20; >20), ever use of aspirin for 3 months, years of schooling, lifetime hours of recreational physical activity (using quartiles), daily intakes of calcium and non-starch polysaccharides from vegetables (using quartiles) in the CRC study. The logarithm of daily energy intake was also included in fully adjusted models.

In both studies, interactions between each genetic variant and, successively, race, pack-years of smoking (0; 0 to ≤20; >20), total red meat intake, processed meat intake, red meat preference (well-done/very well-done versus no red meat/rare/medium), PhIP, MeIQx, Di-MeIQx and total HAA intakes and NAT2 genotype (Rapid versus Slow or Intermediate) were evaluated using log-likelihood tests comparing main effect models with models including cross product interaction effects. Continuous factors were categorized into three levels based on tertiles derived from the CRC study subjects, unless otherwise noted.

We adjusted for multiple comparisons using the online tool SNPSpD to estimate the effective number of independent tests (M_eff) with two related methods (22,23). Both estimates, denoted as M_{eff_N} and M_{eff_L} from Nyholt’s and Li’s methods, respectively, were calculated based on the reduction in the variance of the eigenvalues of the LD matrix among SNPs and rounded to the next largest integer. A significance threshold after Bonferroni correction on M_eff independent tests is then α/M_eff, where α is the type I error rate. As shown in (24), Nyholt’s correction (α_N = α/M_{eff_N}) tends to be conservative and Li’s correction (α_L = α/M_{eff_L}) anticonservative for correlated tests. Therefore, they can be considered as significance and non-significance boundaries, i.e. a test is not significant at type I error α after multiple comparison if the corresponding P-value is > α_L and significant if the P-value < α_N. We applied this strategy in single-variant analysis and in interaction analyses between all variants and each environmental factor. All statistical tests were performed at significance level (α) 0.05 (two sided) using SAS (version 9.1) unless otherwise noted.

Results

The characteristics of the participants are presented in Table I by study. There was no important difference between cases and controls by race and sex distribution in either study and in the age distribution in the CRC study (P-values > 0.05). In the adenoma study, the median age was 1 year older for controls, and the proportion of controls (81%) recruited from Kaiser Permanente Hawaii was lower than for cases (90.5%) (P-values < 0.01). When compared with adenoma cases, controls weighed less and smoked less often (P-values < 0.01, Table I). They also drank fewer alcoholic beverages and consumed more folate (data not shown). Similarly, in the CRC study, compared with cases, controls weighed less and smoked fewer cigarettes (P-values < 0.01, Table I) and were more educated, consumed fewer calories, participated in more recreational physical activity during their lifetime, consumed more calcium and non-starch polysaccharides from vegetables and were more probably to have used aspirin regularly (data not shown).

Table I.

Characteristics of the participants in the colorectal adenoma and colorectal cancer (CRC) studies

	Adenoma		CRC
	Cases (n = 914)	Controls (n = 1185)	Cases (n = 496)	Controls (n = 607)
Sex
Male, n (%)	550 (60.2)	745 (62.9)	305 (61.5)	353 (58.1)
Race/ethnicity
Japanese American, n (%)	291 (31.8)	385 (32.5)	285 (57.5)	368 (60.6)
European American, n (%)	428 (46.8)	564 (47.6)	140 (28.2)	159 (26.2)
Native Hawaiians, n (%)	195 (21.4)	236 (19.9)	71 (14.3)	80 (13.2)
Recruitment site (procedure)a
Kaiser Permanente Hawaii (colonoscopy), n (%)	392 (42.9)	359 (30.3)	—	—
Kaiser Permanente Hawaii (sigmoidoscopy), n (%)	435 (47.6)	601 (50.7)
PLCO (sigmoidoscopy), n (%)	87 (9.5)	225 (19.0)
Ageb	61 (55, 68)	62 (56, 68)	66 (57, 73)	67 (57, 74)
Smoking statusa
Never-smoker, n (%)	382 (41.8)	614 (51.8)	206 (41.5)	288 (47.4)
Past smoker, n (%)	406 (44.5)	484 (40.8)	211 (42.5)	259 (42.7)
Current smokers, n (%)	125 (13.7)	87 (7.4)	79 (16.0)	60 (9.9)
Pack-yearsbc	22 (8, 42)	18 (6, 36)	30 (17, 52)	22 (9, 42)
BMI (kg/m2)b	26.6 (24.0, 30.7)	26.1 (23.4, 29.4)	24.6 (22.2, 27.8)	24.0 (21.9, 26.6)

BMI, body mass index; PLCO, Prostate Lung Colorectal and Ovarian.

^aProportions were significantly different between cases and controls in both studies (P-values < 0.01, χ² test).

^bMedian (interquartile range). The distributions of age in the adenoma study and of pack-years and BMI in both studies were significantly different between cases and controls (P-values ≤ 0.008, Wilcoxon rank-sum test).

^cPack-years = pack-years of cigarette/cigar/pipe smoking among ever smokers.

Table II shows the main effects of red meat, processed red meat, MeIQx, Di-MeIQx, PhIP, total HAAs, red meat preference and NAT2 genotype in both the adenoma and CRC studies, after adjustment of the matching factors and additional risk factors. Total red meat consumption was not associated with either disease (P-values > 0.1). Among all comparisons, only Di-MeIQx was associated with adenoma risk (P = 0.01) but it was not associated with CRC risk (P = 0.85). Before adjusting for additional risk factors, processed red meat, MeIQx, Di-MeIQx, PhIP and total HAAs were associated with an increased adenoma risk (P-values ≤ 0.03) and processed red meat and MeIQx with an increased CRC risk (P_trend = 0.0002 and 0.02, respectively).

Table II.

Main effect ORs (95% CIs) for red meat and HAA intakes in the colorectal adenoma and cancer (CRC) study

	Mediana	Adenoma			CRC
		Na	OR (95% CI)b	Ptrendc	Na	OR (95% CI)b	Ptrendc
Total red meat (g/day)
<42	23	239/328	1.0		144/223	1.0
42–89	63	297/421	0.92 (0.72, 1.17)		179/189	1.15 (0.83, 1.60)
>89	127	375/433	1.11 (0.83, 1.48)	0.36	173/195	0.77 (0.50, 1.17)	0.14
Processed meat (g/day)
<11	5	236/359	1.0		129/239	1.0
11–27	18	310/385	1.20 (0.95, 1.52)		178/189	1.50 (1.09, 2.07)
>27	45	365/438	1.23 (0.94, 1.61)	0.20	189/179	1.33 (0.92, 1.93)	0.28
PhIP (ng/day)
<25	0	176/269	1.0		162/206	1.0
25–91	53	209/306	1.11 (0.84, 1.45)		148/220	0.85 (0.62, 1.17)
>91	173	529/610	1.26 (0.98, 1.62)	0.07	186/181	1.20 (0.86, 1.68)	0.16
MeIQx (ng/day)
<8	0.6	163/264	1.0		155/212	1.0
8–40	20	221/287	1.17 (0.89, 1.53)		155/213	0.90 (0.66, 1.24)
>40	79	530/634	1.19 (0.92, 1.54)	0.30	186/182	1.10 (0.78, 1.54)	0.41
Di-MeIQx (ng/day)
<0.25	0	198/336	1.0		164/204	1.0
0.25–2	0.9	252/333	1.15 (0.90, 1.48)		156/211	0.88 (0.64, 1.21)
>2	4	464/516	1.37 (1.08, 1.75)	0.01	176/192	0.99 (0.71, 1.37)	0.85
Total HAAs (ng/day)
<42	8	167/259	1.0		155/212	1.0
42–139	82	203/283	1.14 (0.86, 1.50)		162/207	0.97 (0.71, 1.32)
>139	247	544/643	1.23 (0.95, 1.59)	0.15	179/188	1.07 (0.76, 1.51)	0.64
NAT2 genotype
Slow		457/539	1.0		116/147	1.0
Intermediate		283/393	0.85 (0.69, 1.05)		203/250	1.13 (0.80, 1.60)
Rapid		166/238	0.85 (0.65, 1.11)	0.15	174/209	1.11 (0.75, 1.63)	0.65
Red meat preference
None/rare/medium		554/569	1.0		362/465	1.0
well-done/very well-done		115/93	1.25 (0.91, 1.71)	0.17	134/142	1.16 (0.86, 1.56)	0.34

^aNumber of cases/controls.

^bORs are adjusted for age, sex and ethnicity, daily energy intake, lifetime hours of recreational physical activity and additionally for recruitment site and examination procedure, BMI, pack-years of smoking, alcohol intake, folate intake in the adenoma study and BMI 5 years before diagnosis, pack-years of smoking, ever use of aspirin, years of schooling, daily intakes of calcium and non-starch polysaccharides from vegetables in the CRC study.

^cTrend test was based on medians within each category. Three categories were defined based on tertiles in the CRC study for continuous variables. P-values < 0.05 were considered statistically significant.

Supplementary Table 1 (available at Carcinogenesis Online) lists the allele frequency and P-value for the HWE test for each of the 58 SNPs genotyped in each study, by each ethnic group and overall. The 45 SNPs that passed quality control had an average MAF of 0.23 in the adenoma and 0.22 in the CRC study. The average inter-marker distance was 5.5 kb (range 0.01–24.8 kb).

The effective number of independent tests when considering the 45 correlated variants simultaneously was estimated to be 39 with Nyholt’s method (23) and 30 with Li’s method (22). The corrected significance levels α_N = α/M_{eff_N} = 0.05/39 = 0.0013 and α_L = α/M_{eff_L} = 0.05/30 = 0.00167 were used as the upper threshold of significance and lower threshold of non-significance, respectively. These compare to an uncorrected Bonferroni threshold of 0.05/45 = 0.0011.

There was no important heterogeneity in the associations with single SNPs across ethnic groups (P_interaction ≥ 0.01) in either study; thus, we pooled data across all ethnic groups for each study. For the adenoma study, results of the minimally and fully adjusted models were similar and no SNP was found to be statistically associated with disease risk (P-values ≥ 0.01) (supplementary Table 2 is available at Carcinogenesis Online). The lowest P-value observed was for rs4646425 in CYP1A2 (P = 0.010), where carrying the minor allele T was associated with an OR of 0.57 (95% CI: 0.37–0.87). After adjustment for multiple comparisons, this association was no longer statistically significant.

For the CRC study, Table III shows ORs and 95% CIs for the variants with P-value < 0.01 from the single-variant analysis. In the minimally adjusted models, none of the P-values was significant after multiple comparison adjustment (all P-values > the non-significance threshold 0.00167). After adjusting for other risk factors, the only variant with a P-value that passed the conservative threshold of 0.0013 was ARNT rs12410394 (adjusted P-value 0.0008). For this SNP, each copy of the A allele was associated with an OR of 0.73 (95% CI: 0.61–0.88) in the fully adjusted models compared with the OR of 0.93 (95% CI: 0.81–1.06) found in the adenoma study. The two other ARNT SNPs in Table III are in strong LD with rs12410394 (|D′| = 0.97 for rs2228099 and |D′| =1 for rs3215133).

Table III.

ORs (95% CIs) with the lowest P-values (P < 0.01) for genetic variants and CRC risk

Gene	Chr	Position (dbSNP b126)	SNP	MAFc	Genotype	Nd	Minimally adjusteda		Fully adjustedb
							OR (95% CI)	Ptrende	OR (95% CI)	Ptrende
ARNT	1	149068897	rs3215133	0.039	TT	464/574	1.0	0.02	1.0	0.009
					−/T or −/−	28/54	0.57 (0.35, 0.92)		0.50 (0.29, 0.84)
		149075513	rs2228099	0.398	CC	200/205	1.0	0.008	1.0	0.003
					GC	215/278	0.80 (0.62, 1.05)		0.72 (0.54, 0.95)
					GG	70/114	0.62 (0.44, 0.89)		0.58 (0.40, 0.85)
		149126810	rs12410394	0.480	GG	160/148	1.0	0.005	1.0	0.0008
					AG	227/294	0.73 (0.55, 0.97)		0.66 (0.48, 0.89)
					AA	105/159	0.62 (0.44, 0.87)		0.54 (0.37, 0.78)
CYP1A2	15	72834479	rs2470890	0.314	CC	266/293	1.0	0.006	1.0	0.003
					TC	166/216	0.78 (0.59, 1.03)		0.72 (0.53, 0.96)
					TT	60/93	0.57 (0.37, 0.87)		0.54 (0.34, 0.85)
		72849450	rs11072508	0.319	CC	264/292	1.0	0.005	1.0	0.00173
					TC	171/217	0.80 (0.60, 1.05)		0.70 (0.52, 0.94)
					TT	61/97	0.54 (0.35, 0.83)		0.52 (0.33, 0.82)
		72852697	rs4886410	0.319	CC	264/295	1.0	0.007	1.0	0.003
					GC	170/214	0.81 (0.61, 1.06)		0.72 (0.53, 0.96)
					GG	62/97	0.56 (0.36, 0.85)		0.54 (0.35, 0.85)

BMI, body mass index

^aAdjusted for the matching factors age, sex and race.

^bFurther adjusted for BMI 5 years before diagnosis, daily energy intake, pack-years of smoking, ever use of aspirin consecutively for 3 months, years of schooling, lifetime hours of physical activity, daily intake of calcium and daily intake of non-starch polysaccharides from vegetables. Sample size was reduced by 43 (23 controls/20 cases) due to missing values.

^cMAF in all CRC subjects combined.

^dNumber of cases/controls.

^eAfter multiple comparison adjustment, P-values < 0.0013 were considered statistically significant and P-values ≥ 0.00167 were non-significant. Note: none of these SNPs was significantly associated with adenoma (P-values > 0.05).

The CYP1A2 variants rs2470890, rs11072508 and rs4886410 were suggestively associated with CRC risk (0.00167 < P-values ≤ 0.003) (Table III). These three SNPs are in high LD with each other (pairwise |D′| ≥ 0.92). rs11072508 showed the strongest association with each copy of the minor allele T being associated with a reduced CRC risk (OR: 0.72, 95% CI: 0.58–0.88, P: 0.00173).

We also searched for important associations under dominant, recessive and codominant genetic models for all variants. Two AHR variants rs17137566 and rs2066853 in close LD (pairwise |D′| = 0.97) were found to have a P-value < 0.01 in the CRC study (P-values = 0.01 and 0.008, respectively, in a dominant and codominant model), although these effects were not statistically significant after multiple comparison adjustment (data not shown).

Next, we explored differences in effects by anatomical location of the tumor (number of cases: right colon, 146; left colon, 185; rectum, 135) for the SNPs in Table III, using all controls as the reference group in a multinomial logistic regression model. The tests for heterogeneity across subsites were not significant (P-values > 0.05).

Supplementary Tables 2 and 3 (available at Carcinogenesis Online) show the ORs from the single-variant association analyses for the 45 variants, for each ethnic group and overall in the adenoma and CRC study, respectively. Results from the haplotype analysis were consistent with those from the single-variant analysis and did not provide stronger evidence of associations (results not shown).

We also tested for interaction between single genetic variants (G) and environmental exposures (E) or NAT2 in both studies in the fully adjusted models. No significant interaction was detected (all P_interaction > 0.00167). Table IV shows the two G × E effects with a P_interaction < 0.005. In the adenoma study, the effect of total red meat consumption appeared to be different across genotypes of AHR rs3757824 (P_interaction = 0.003). More red meat consumption was associated with an increased adenoma risk among genotype TT and a decreased risk among genotype CC, although the effect in neither stratum was significant (P’s > 0.05). Interactions between this SNP and MeIQx and processed red meat intake also had relatively small P-values (P_interaction ≤ 0.02).

Table IV.

Gene (G) × environment (E) effects with the lowest P-values (P < 0.005) in the colorectal adenoma and colorectal cancer studies^a

G	G level	E	E level	Nb	Within G		OR (95% CI)	P (OR)c	Pinteraction
					OR (95% CI)	Ptrend
Adenoma study
AHR rs3757824	TT	Total red meat (g/day)	<42	130/168	1.0	0.09	1.0 (reference)	—	0.003
			42–89	165/248	0.78 (0.57, 1.08)		0.78 (0.57, 1.08)	0.13
			>89	233/217	1.31 (0.91, 1.87)		1.31 (0.91, 1.87)	0.14
	CT		<42	82/131	1.0	0.64	0.81 (0.56, 1.17)	0.27
			42–89	110/138	1.23 (0.83, 1.82)		1.00 (0.70, 1.43)	0.99
			>89	118/174	0.94 (0.62, 1.43)		0.76 (0.52, 1.13)	0.17
	CC		<42	27/24	1.0	0.07	1.46 (0.79, 2.71)	0.23
			42–89	18/28	0.60 (0.26, 1.39)		0.87 (0.45, 1.70)	0.69
			>89	19/33	0.46 (0.20, 1.08)		0.68 (0.35, 1.33)	0.26
CRC study
CYP1B1 rs1056837	CC	Di-MeIQx (ng/day)	<0.25	102/129	1.0	0.30	1.0 (reference)	—	0.004
			0.25–2	89/110	0.99 (0.66, 1.50)		0.99 (0.66, 1.50)	0.97
			>2	102/131	0.81 (0.54, 1.22)		0.81 (0.54, 1.22)	0.31
	CT		<0.25	50/59	1.0	0.99	1.09 (0.66, 1.81)	0.73
			0.25–2	56/81	0.77 (0.45, 1.33)		0.84 (0.53, 1.35)	0.47
			>2	54/54	1.00 (0.56, 1.80)		1.09 (0.66, 1.80)	0.73
	TT		<0.25	12/15	1.0	0.01	0.94 (0.39, 2.29)	0.89
			0.25–2	10/19	0.59 (0.19, 1.85)		0.55 (0.24, 1.28)	0.17
			>2	20/6	5.04 (1.43, 17.78)		4.73 (1.73, 12.9)	0.002

BMI, body mass index

^aAdjusted for age, sex and ethnicity, daily energy intake, lifetime hours of recreational physical activity, and additionally for recruitment site and examination procedure, BMI, pack-years of smoking, alcohol intake, folate intake in the adenoma study, and BMI 5 years before diagnosis, pack-years of smoking, ever use of aspirin, years of schooling, daily intakes of calcium and non-starch polysaccharides from vegetables in the CRC study.

^bNumber of cases/controls.

^cP-value for OR compared with the reference group. After multiple comparison adjustment, P-values < 0.0013 were considered statistically significant and P-values ≥ 0.00167 were non-significant.

For the CRC study, the interaction between CYP1B1 rs1056837 and Di-MeIQx intake (P_interaction = 0.004) is shown. No effect of Di-MeIQx was observed within rs1056837 genotypes CC or CT (P’s ≥ 0.30). However, among individuals with the TT genotype, an intake of Di-MeIQx in the upper tertile (>2 ng/day) was associated with a significantly increased CRC risk (OR: 5.04, 95% CI: 1.43–17.8, P = 0.01), compared with the lower tertile of intake (<0.25 ng/day). Nonetheless, note that the sample size was small in the subgroup with TT genotype and high intake of Di-MeIQx—only 20 cases and six controls.

Discussion

In this investigation, we found several variants in genes involved in the bioactivation of PAHs and HAAs to be associated with the risk of CRC. The strongest association and, indeed, the only one that remained statistically significant after adjusting for multiple comparisons was between ARNT rs12410394 and CRC. A marginally significant association with CRC was also observed for the CYP1A2 region containing rs2470890, rs11072508 and rs4886410. These results suggest that the genomic regions around these SNPs may contain variants that affect disease risk. After adjusting for multiple comparisons, no interaction was detected between the genetic variants studied and smoking, red meat, HAAs or NAT2. Finally, the strongest main effects were observed for CRC and not for adenoma, suggesting that these genetic effects may be more important for the later rather than the earlier stages of colorectal neoplasia.

We identified no past study between CRC and ARNT variants. ARNT (also known as hypoxia inducible factor-1β or HIF 1B) is a member of the basic helix-loop-helix Per/AHR/ARNT/Sim family of transcription factors. ARNT and its interacting molecules (HIF1a, HIF2a and AHR) form heterodimeric complexes, which mediate cellular responses to hypoxia and environmental toxins, such as dioxins and PAHs. Putative ARNT-binding sites may exceed 13  000 in the genome (25) and include genes involved in glucose metabolism, vascular function and oxygen transport. The ARNT gene is located in a region on chromosome 1q21 with replicated linkage to type-2 diabetes in various populations. SNP rs188970, a perfect proxy for the synonymous rs2228099 that we found to be marginally associated with CRC risk (Table II), has recently been associated with active insulin response (P ≤ 0.005) (26). Since type 2-diabetes is a risk factor for CRC, the role of ARNT in susceptibility to diabetes is consistent with the main effect observed here. The SNP with the strongest association, rs12410394, was not genotyped by the HapMap project (27); no functional information was available for this variant in dbSNP (http://www.ncbi.nlm.nih.gov/projects/SNP/).

Selected SNPs in CYP1A1, CYP1A2 and CYP1B1 have been examined in past studies and the results have been inconsistent. Two CYP1A1 polymorphisms in high LD, namely, the T6235C (rs4646903) and the missense A4889G or I462V (rs1048943) variants, have been associated with lung cancer and with CRC/adenoma, as a main effect and/or as interacting with smoking (28–31). However, these SNPs were not associated with CRC in a large study in Ontario (1095 cases and 1890 controls) (32). In our studies, these two SNPs were not associated with either disease (P ≥ 0.08). Two other CYP1A1 variants, rs1799814 or T461N and rs2606345, were found to be inversely associated with CRC in a Spanish case–control study of 377 cases and 326 controls (33). We analyzed rs1048943, which is 2 bp away from rs1799814, and it was not associated with either disease.

Variants in CYP1A2 have also been investigated in relation to CRC risk. In a French study of 1023 cases and 1121 controls, Küry et al. (34) reported no association for the intronic rs762551 and synonymous rs2470890; however, rs2470890 in our study was marginally associated with CRC risk after multiple comparison adjustment (P = 0.003, Table III). rs762551 was not associated with CRC in the Ontario study (32) but was associated with an increased CRC risk in the aforementioned Spanish study (33). We genotyped rs2472299, which is in complete LD (r² = 1.0) with rs762551, and found no strong effects (all P-values ≥ 0.02). Moreover, a Taiwanese case–control study of CRC reported no main effect for rs2069514 (also known as −3860G>A) but an interaction was observed between this SNP and NAT1*10 in increasing CRC risk (35). Finally, in a large UK case–control study (2561 cases and 2595 controls), rs2069522 (within 2 kb of the 5′ end of the coding region) was found to be weakly associated with CRC risk (36). In our studies, rs2069514 and rs2069522 were not associated with CRC or adenoma.

CYP1B1 rs10012 or R48G has been associated with an increased CRC risk in some data (33) but not in others (32). rs162558 (within 2 kb of the 5′ end of the coding region) was reported to be associated with CRC in a large UK case–control study (36). We genotyped rs162557 (not genotyped by HapMap), which is 1.4 and 3.1 kb from rs162558 and rs10012, respectively, and no important effect was observed. In some past studies, CYPI1B1 rs1056836 (L432V) was not associated with CRC (32,34) but, in others, this SNP was found to interact with intake of red meat cooked by high-temperature methods (32) and with smoking (37) in increasing CRC risk. We analyzed rs1056837 or D449E, which is in very high LD (r² = 1.0) with rs1056836 and found a moderate interaction effect with Di-MeIQx on CRC risk (Table III). A similar but weaker interaction was suggested with total red meat intake (P = 0.02). Thus, this apparently consistent observation across studies of an interaction between SNPs in this region and exposure warrants further study.

We found only two studies on colorectal neoplasia and AHR rs2066853 (R554K). Previously, this SNP did not appear to be associated with CRC (32) or adenoma (38). However, in our study, there was some suggestion of an increase in CRC risk for the A allele of rs2066853 using a codominant model (P = 0.008).

Thus, past results for genes encoding carcinogen metabolizing enzymes and CRC risk have been mostly inconsistent. We note that future systematic reviews or meta-analyses may provide more insights into these associations.

Strengths of our study include a systematic approach to testing the association of genetic variants in the bioactivation pathway for PAHs and HAAs with colorectal neoplasia and their interaction with relevant exposures and strict control of type I error rates. However, the lifestyle exposure assessment in both studies was conducted after diagnosis and recall bias could be a concern. Similarly, we did not adjust our analysis for potential population stratification within each ethnic group; it has been demonstrated that this type of bias tends to cancel each other and is small in a study with multiple ethnic groups (39). In addition, the majority of the controls in the adenoma study only received a flexible sigmoidoscopy and adenomas located in the proximal colon may have been missed. Finally, our sample size was relatively small in the CRC study, especially for the interaction analyses. In order to detect GxE and GxG interactions, we tried the classification tree approach (40) to select important risk factors for CRC among all environmental factors (see Methods) and genetic variants; however, no genetic effect was included in the final selected models—this is consistent with the notion that CRC is mainly caused by environmental factors and also reflects our limited sample size in certain nodes (results not shown).

In summary, the present investigation identified ARNT as a putative risk locus for CRC. Evidence was weak for the existence of interaction effects between genetic variants in the CYP1A1/A2, CYP1B1, AHR and ARNT genes and environmental risk factors on colorectal neoplasia. The association of ARNT with CRC warrants replication in larger studies.

Supplementary material

Supplementary Tables 1–3 can be found at http://carcin.oxfordjournals.org/

Funding

National Cancer Institute (R01 CA60987, CA72520).

Glossary

Abbreviations

AHR

aromatic hydrocarbon receptor

ARNT

aromatic hydrocarbon receptor nuclear translocator

CI

confidence interval

CRC

colorectal cancer

HAA

heterocyclic aromatic amine

HWE

Hardy–Weinberg equilibrium

MAF

minor allele frequency

LD

linkage disequilibrium

OR

odds ratio

PAH

polycyclic aromatic hydrocarbon

SNPs

single nucleotide polymorphisms