|Home | About | Journals | Submit | Contact Us | Français|
We recently reported that heterocyclic amines (HCAs) are associated with prostate cancer risk in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. We now employ extensive genetic data from this resource to determine if risks associated with dietary HCAs (PhIP, MeIQx, DiMeIQx) from cooked meat are modified by single nucleotide polymorphisms (SNPs) in genes involved in HCA metabolism (CYP1A1, CYP1A2, CYP1B1, GSTA1, GSTM1, GSTM3, GSTP1, NAT1, NAT2, SULT1A1, SULT1A2, and UGT1A locus). We conducted a nested case-control study that included 1,126 prostate cancer cases and 1,127 controls selected for a genome-wide association study for prostate cancer. Unconditional logistic regression was used to estimate odds ratios (ORs), 95% confidence intervals (CIs) and p-values for the interaction between SNPs, HCA intake and risk of prostate cancer. The strongest evidence for an interaction was noted between DiMeIQx and MeIQx and the polymorphism rs11102001 downstream of the GSTM3 locus (p-interaction 0.001 for both HCAs; statistically significant after correction for multiple testing). Among men carrying the A variant, the risk of prostate cancer associated with high DiMeIQx intake was two-fold greater than those with low intake (OR=2.3, 95% CI: 1.2-4.7). The SNP, rs11102001, which encodes a nonsynonymous amino acid change P356S in EPS8L3, is a potential candidate modifier of the effect of HCAs on prostate cancer risk. The observed effect provides evidence to support the hypothesis that HCAs may act as promoters of malignant transformation by altering mitogenic signaling.
Recent studies suggest that exposure to heterocyclic amines (HCAs) derived from meats cooked at high temperatures, such as pan-frying or barbequing, may increase the risk of prostate cancer. Several animal and human experimental studies have demonstrated the carcinogenicity of three HCAs in particular: PhIP (2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine), MeIQx (2-amino-3,8-dimethylimidazo-[4,5-b]quinoxaline), and DiMeIQx (2-amino-3,4,8-trimethylimidazo-[4,5-f]quinoxaline). In animal models, PhIP increases mutation frequency (1) and tumor incidence (2) in the prostate. In vitro work in human prostate cells has shown that PhIP increases genotoxicity and DNA adduct levels (3-5); PhIPDNA adducts have been also been detected in vivo in human prostate cells (6,7). Oral administration of MeIQx induces tumors in rodents at multiple tissue sites (8). The N-hydroxy metabolite of MeIQx leads to prostate hyperplasia in rats and induces MeIQx-DNA adduct formation in human prostate epithelial cells (5,9). DiMeIQx, which is similar in chemical structure to MeIQx, is mutagenic in bacterial assays (10), but has not been extensively evaluated as an animal or human carcinogen.
Epidemiologic studies of HCA intake and prostate cancer are limited. One large prospective study found a significant increased risk of prostate cancer for individuals in the highest quintile of PhIP intake (11), while another prospective study found elevated risks associated with increased MeIQx and DiMeIQx (12). Two small case-control studies, however, found no association between these HCAs and prostate cancer (13,14).
Once HCAs enter the body, they undergo a series of chemical reactions in order to be eliminated. These reactions are highly dependent on particular xenobiotic metabolic enzymes (XME) and include both phase I and phase II enzymes. The metabolism of MeIQx, the structurally similar DiMeIQx, and PhIP has been extensively described (15). Cytochrome P450 (CYP) enzymes (phase I) including CYP1A1, CYP1A2, and CYP1B1 are involved in the bioactivation of these compounds. Phase II enzymes including sulfotransferases (SULTs), N-acetyltransferases (NATs), UDP-glucuronosyltransferases (UGTs), and glutathione S-transferases (GSTs) are responsible for further metabolism and detoxification. Single nucleotide polymorphisms (SNPs) in genes that code for these enzymes may result in differential metabolism of HCAs and their intermediates and thus may be related to prostate cancer risk. SNPs in genes related to xenobiotic metabolism have been inconsistently associated with prostate cancer risk (16-19). The conflicting findings may be a result of an oversimplification in assuming these genes act alone to alter risk. The complex interaction with environmental exposures, like those from dietary HCA intake, may offer more insight into the importance of XMEs and prostate cancer risk.
By combining data from a genome wide association study of prostate cancer and dietary HCA intake from questionnaire data, we evaluated the interaction of dietary HCA (PhIP, MeIQx, DiMeIQx) intake, polymorphisms in genes involved in HCA metabolism (CYP1A1, CYP1A2, CYP1B1, GSTA1, GSTM1, GSTM3, GSTP1, NAT1, NAT2, SULT1A1, SULT1A2, and UGT1A locus) and risk of prostate cancer.
Participants were selected from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, a randomized, controlled, multi-site trial to test the efficacy of screening methods for these four cancers. Details of this trial have been described elsewhere (20,21). Briefly, the PLCO trial participants were individuals 55-74 years old who were enrolled between 1993 and 2001 and reported no history of prostate, lung, colorectal, or ovarian cancer. Participants were randomized to either the screening or control arm of the trial. Men randomized to the screening arm were offered a prostate-specific antigen (PSA) test and digital rectal exam at baseline and annually thereafter for 3 years, followed by 2 years of screening with PSA alone. Cancer diagnoses were ascertained from mailed annual questionnaires and supplemented by trial screening results and linkage to cancer registries and the National Death Index to enhance completeness. Cases of prostate cancer were confirmed by review of medical records. The study protocol was approved by the institutional review board at each study center and the National Cancer Institute and participants provided informed consent.
A subgroup of men from the screening arm of the PLCO trial were included in the National Cancer Institute’s genome-wide association study of prostate cancer called Cancer Genetics Markers of Susceptibility (CGEMS), an initiative that has genotyped approximately 550,000 SNPs in selected prostate cancer cases and controls.* Details of the CGEMS case and control selection are described elsewhere (22). In brief, cases and controls were Caucasian, completed the baseline questionnaire, provided a blood sample, had no prior history of prostate cancer before randomization, and had at least one PSA test before October 1, 2003. Cases were oversampled for aggressive disease (Gleason score ≥7 or Stage ≥ III) and were diagnosed between October 1993 and September 2003. A total of 1,230 prostate cancer cases were selected. Controls were selected by incidence density sampling resulting in the identification of 1,204 different men and 1,230 control selections (1,179 subjects sampled once, 24 subjects sampled twice, and one subject sampled three times). Cases and controls were frequency-matched on age at randomization, fiscal year of randomization, and time since initial screen.
Additional eligibility criteria for the current nested case-control study follow those of the previous meat and meat-mutagen analysis from the PLCO cohort (11). Cases and controls where ineligible if they did not complete the PLCO food frequency questionnaire (FFQ, n=148), were extreme outliers for reported energy intake (those in the top or bottom 1% of intake, n=37), were missing information on HCAs (n=1), or missed more than seven items on their FFQ (n=21), resulting in a total sample size of 2,253 (1,126 cases and 1,127 controls).
Dietary data was collected as part of the PLCO trial from a validated 137-item FFQ at baseline. The FFQ included detailed information about meat-cooking methods (barbequing, grilling, pan frying, and broiling) and extent of doneness (rare, medium, well done, or very well done) for commonly consumed meats (steak, bacon, sausage, pork chop, and hamburgers) often cooked by different methods. Information obtained from the FFQ was used in conjunction with a mutagen database, CHARRED (23), to estimate intake of PhIP, MeIQx, and DiMeIQx.
Selection of SNPs, genotyping, and quality control procedure for the CGEMS prostate cancer study is described in detail in the Supplementary Methods section of the manuscript by Yeager et al. (24) which can be found online†. Briefly, genotyping of the CGEMS prostate cancer study was performed under contract by Illumina Corporation using the HumanHap240 and HumanHap300 platforms, which constituted a fixed panel of 561,494 tagSNPs. Common (minor allele frequency ≥ 5%) SNPs were identified using the method described by Carlson et al (25) assuming a r2 threshold > 0.8 for the HapMap CEU subjects of European ancestry. For the European population, this panel is expected to cover close to 90% of the common SNPs in HapMap phase 1. Genotype data for all SNPs in the current study were tested among controls for possible deviation from Hardy-Weinberg proportions.
We selected genes in phase I and phase II enzymes that are directly involved in the metabolism of the HCAs under study (26, 27). The following 10 genes or gene regions (combined due to proximity and overlapping regions) were included in the current study: CYP1A1 and CYP1A2 region, CYP1B1, GSTA1, GSTM1, GSTM3, GSTP1, NAT1, NAT2, SULT1A1 and SULT1A2 region, and the UGT1A locus. For each gene or gene region a 20kb upstream and 10kb downstream margin was used to select SNPs.
A total of 127 SNPs were selected for analysis; eight SNPs (rs13406898, rs17038616, rs17864673, rs17868306, rs7192559, rs8191431, rs9341252, rs9939264) showed little or no variation (minor allele frequency <1%) and thus were not included in regression modeling. All other SNPs had a minor allele frequency ≥ 1% in our population.
Differences between cases and controls with respect to descriptive characteristics were assessed using a chi-square test for categorical variables and t-test for continuous variables. Unconditional logistic regression was used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for the association between HCA intake and prostate cancer, the association between XME SNPs and prostate cancer (data not shown), and the interaction between SNPs in XME genes, HCA intake and prostate cancer. SNP-HCA interactions were examined using a multiplicative model. Interactions were assessed by including the cross-product terms for the gene (SNPs) and HCA intake as well as the main effect terms in a logistic regression model. Genotypes were categorized using ordinal values of 1 (homozygote wild-type), 2 (heterozygote), and 3 (homozygote variant) assuming an additive model in the regression. We evaluated the interaction under different genetic models (codominant, dominant, and recessive); however, these results were similar to those presented assuming an additive model and are therefore not shown. In instances where the homozygous variant genotype was rare, this was combined with the heterozygous group. HCA intake was categorized as low (0-39th percentile), medium (40-79th percentile), and high (≥ 80th percentile) intake, since the intake distribution is skewed and previous analyses have identified the top quintile as potentially the most important (12). Models were adjusted for age at selection, study center, and total energy intake. The p-value for each SNP-HCA interaction was computed by comparing nested models with and without the cross-product terms using a likelihood ratio test. Interactions less than or equal to p = 0.15 are presented.
Given the small p-values, the presence of functional variants, and the consistent indication of interaction between these genes and HCAs, associations between HCA intake and prostate cancer stratified by GSTM3 and GSTP1 genotypes were further evaluated for risk patterns. The association between joint categories of HCA intake and the same SNPs was also estimated; the referent group was defined as the combination of the homozygote wild-type genotype and the lowest HCA intake group. Statistical analyses were performed using STATA version 9.0.
Haplotype-HCA interaction analyses assuming an additive model for each haplotype were pursued for GSTM3 and GSTP1; haplotype-HCA interactions were explored for the GSTP1 locus with PhIP and MeIQx and for GSTM3 with all HCAs. Haplotypes were estimated using the expectation-maximization (EM) algorithm (28) in HaploStats version 1.2.1 for the R-programming language. Haplotypes with a frequency of less than 1% were collapsed into a single category and the most common haplotype was used as the referent. The p-value for each haplotype-HCA interaction was computed by comparing nested models with and without the cross-product terms using a likelihood ratio test.
To test the robustness of our findings, the false discovery rate (FDR) (Benjamini– Hochberg adjustment) method (29) was applied. The FDR is the expected proportion of false discoveries among the discoveries. FDR values were calculated separately for each HCA from the results of 119 tests (i.e., total number of SNPs studied minus those with no variation) evaluating the association between each SNP-HCA interaction and the risk of prostate cancer. Interactions were deemed significant at an FDR = 0.20 level; this indicates that 1/5 discoveries would be expected to be false.
A summary of the SNPs, in chromosomal order, within each gene or gene region that were evaluated in this study is given in Supplementary Appendix A. Among the control group, 14 genotypes deviated from Hardy-Weinberg proportions (p<0.05) but no significant deviations were observed among the SNPs presented in the results tables. Cases were slightly older than controls; however, this difference was not statistically significant (Table 1).
Table 2 presents the ORs and 95% CIs for the main effect of HCAs and risk of prostate cancer risk in the current analysis and in the previous analysis of the PLCO cohort. Consistent with the previous analysis, there was an elevated risk of prostate cancer in the top quintile of PhIP intake when compared with the referent interaction category (OR= 1.11, 95% CI 0.86, 1.44), although in this modified population the risk did not reach statistical significance. The current analysis has approximately 200 fewer prostate cancer cases than previously reported and is over-sampled for advanced disease, some of whom were not included in the previous cohort analysis and thus likely explains the few differences in risk estimates.
Fifteen interactions with p-values less than or equal to 0.15 were observed between HCAs and several phase I/II enzyme SNPs including those in the CYP1B1, GSTP1, GSTM3, and UGT1A gene and gene regions (Table 3). Interaction p-values between MeIQx and DiMeIQx and the GSTM3 polymorphism, rs11102001, were particularly small (p-interaction 0.001 for both HCAs) and indicated that the effect of HCA intake on the risk of prostate cancer may differ by genotype. The FDR adjusted p-values for the interactions involving MeIQx-rs11102001 and DiMeIQx-rs11102001 were both 0.12; thus, statistically significant at the FDR =0.20 level. The GSTM3 SNP, rs11102001, also appeared to modify the association between PhIP and prostate cancer risk (p-interaction = 0.04). Similarly, the GSTP1 nonsynonymous SNP, rs1695, appeared to modify the association between PhIP and MeIQx intake and prostate cancer (p-interaction = 0.03 for both HCAs). Neither these nor other interactions were deemed significant after adjustment using the FDR method (FDR>0.20).
Associations between HCA intake and prostate cancer stratified by GSTM3 and GSTP1 genotypes are presented in Table 4. Among individuals carrying the AG or AA genotype (rs11102001), the risk of prostate cancer for those with high intake of MeIQx and DiMeIQx was increased compared to those with low intake (OR=1.7, 95% CI 0.8, 3.6, p-interaction = 0.001 and OR=2.3, 95% CI 1.2, 4.7, p-interaction = 0.001, respectively). Among individuals carrying the GG genotype, risk of prostate cancer was decreased compared to those with low MeIQx and DiMeIQx intake (OR=0.6, 95% CI 0.5, 0.8 and OR=0.7, 95% CI 0.5, 0.8, respectively). Increased intake of PhIP or MeIQx was inversely associated with risk of prostate cancer among individuals with AA or AG (rs1695) genotypes, but a positive association was observed among individuals with the GG genotype (high vs. low PhIP: OR=1.5, 95% CI 0.7, 3.1 and high vs. low MeIQx: OR=1.5, 95% CI 0.7, 3.3). Similar findings were observed among homozygous variant genotypes GG for rs2274536 (GSTM3), TT for rs18887546 (GSTM3), and GG for rs6591256 (GSTP1) where high HCA intake was associated with increased risk of prostate cancer compared with low intake while wild type variants were associated with decreased risk. Thus, the effect of HCAs on prostate cancer appears to depend on genotype.
The association between joint categories of HCA intake and GSTM3 and GSTP1 genotypes and prostate cancer risk are summarized in Supplementary Appendix B. The patterns of joint effects were similar to the findings from the analyses stratified by genotype, with an increased association with high HCA intake observed among homozygous variant genotypes, however the magnitude of these effects tended to be in the OR=1.1-1.5 range.
GSTP1 and GSTM3-haplotype interactions were also evaluated given that several SNPs in these genes showed interesting interactions. PhIP and MeIQx interactions with GSTP1 haplotypes were associated with prostate cancer (p-interaction 0.06 and 0.02, respectively; data not shown). These associations were completely driven by the rs1695 polymorphism, where haplotypes that included the variant allele at rs1695 resulted in increased risk of prostate cancer. Important associations for GSTM3 haplotype-HCA interactions were also present, GSTM3-MeIQx p-interaction= 0.04 and GSTM3-DiMeIQx p-interaction= 0.08 (data not shown). Haplotypes that carried the variant alleles in rs7483, rs2274536, rs1887546, and rs11102001 were observed to increase risk, however, these haplotype frequencies were extremely rare (<0.01%) resulting in unstable estimates and are therefore not shown.
Our findings offer the first comprehensive analysis of the interaction between HCA intake, genetic polymorphisms in xenobiotic metabolizing genes that are known to metabolize these compounds and prostate cancer risk. We observed that the effect of HCA intake on the risk of prostate cancer differs by GSTM3 and GSTP1 genotypes in particular; interactions with the EPS8L3 Pro356Ser polymorphism (rs11102001) just downstream of the GSTM3 locus were statistically significant at the FDR = 0.20 level and the GSTP1 Ile105Val polymorphism (rs1695) also appeared to modify risk.
We observed suggestive HCA interactions with 3 SNPs located in the GSTM3 region (rs2274536, rs1887546, rs11102001). All three of these SNPs are located within the 10kb downstream margin of GSTM3 and are located in another gene called EPS8L3, epidermal growth factor receptor pathway substrate 8-like 3. This gene, and other member of the EPS8 family (EPS8L1-2), encodes proteins that are responsible for Ras to Rac signaling leading to actin remodeling or cytoskeletal integrity (30-32). The Ras signaling pathway regulates normal cell proliferation. Ras and Ras-related proteins are often deregulated in cancers, leading to increased invasion, metastasis, and decreased apoptosis (33). Ras activation is a component of the signaling pathways for virtually all the receptors shown to be upregulated in advanced prostate cancer (34). The polymorphic site at codon 356 in exon 12 (rs11102001), where a guanine-to-adenosine (G-A) transition occurs, causes a proline to serine substitution and had a minor allele frequency of 7% among controls in our population, which is consistent with the minor allele frequency observed in HapMap for Caucasians (7%). While no literature describes any loss of activity associated with this polymorphism or that this amino acid substitution is likely to be damaging (PolyPhen‡ prediction score = 0.286), carriers of the variant allele (A) with high MeIQx and DiMeIQx intake were at higher risk for prostate cancer compared to those with low intake. A recent study showed that PhIP stimulates cellular signaling pathways and resulted in increased growth and cell migration in human mammary epithelial cell lines (35). Thus, increased HCA exposure could similarly act as a promoter of malignant transformation by increasing mitogenic signaling.
Alternatively, the mechanism of action responsible for the observed effect might be linked to xenobiotic metabolism pathways. GSTM3 is expressed in prostate tissues (36,37) and acts to detoxify active HCA metabolites by conjugation with glutathione (27). Altered expression of the enzyme could lead to differential clearance of activated HCA metabolites resulting in an accumulation of DNA damaging species, which could increase the risk for carcinogenesis at this site. The three EPS8L3 SNPs are located just downstream of GSTM3 and show strong linkage disequilibrium with variants in the GSTM3 locus; therefore, it is possible that these SNPs are surrogate markers of other variants in the GSTM3 locus that were not genotyped. Alternatively, these downstream variants could alter GSTM3 expression. After adjustment using the FDR method, the MeIQx-rs11102001 and DiMeIQxrs11102001 interactions appear to be the most interesting findings, suggesting a real modification of prostate cancer risk. Thus, further examination of variants in this region is warranted.
HCA intake was estimated from a meat-cooking FFQ module, which included questions about meat type, cooking method, and cooking time, or ‘doneness’ level, used in conjunction with a HCA database. HCA formation in meat generally increases with temperature and doneness level. PhIP is the most abundant HCA, then MeIQx and DiMeIQx is the least abundant; IQ (2-amino-3methylimidazo[4,5-f]quinoline) and MeIQ (2-amino-3,4-dimethylimidazo[4,5-f]quinoline are not typically detected in meat samples (38). The questionnaire and procedure for estimating HCA intake has been validated and found to be an acceptable method for HCA exposure assessment (39).
We observed an effect for DiMeIQx and MeIQx in the presence of GSTM3 variants. Although DiMeIQx is consumed at a low level with respect to other HCAs in the diet, DiMeIQx and MeIQx are observed to be more potent mutagens than PhIP (40) thus their impact on carcinogenesis may be greater. There is little information in the literature with respect to the activity of GST enzymes in the detoxification of DiMeIQx and MeIQx. As the formation of PhIP is the highest, the pathway for detoxification of this compound is better described with respect to several Phase II enzymes (27). Information on DiMeIQx is lacking with respect to detoxification pathways and carcinogenicity data, although it is known to be mutagenic in bacterial assays. The continued evaluation of DiMeIQx and MeIQx is needed as more studies of HCAs and cancer risk implicate these more potent HCAs.
We also observed lesser effects for HCA interactions with one promoter polymorphism (rs6591256, position -1415) and one nonsynonymous polymorphism (rs1695, isoleucine to valine substitution) in GSTP1. GSTP1 is the major GST identified in benign prostate hyperplasia tissue samples relative to other GST enzymes (37,41). In human prostate tissue, however, expression of this enzyme is silenced via hypermethylation of the promoter region (42), occurring in approximately 90% of prostate adenocarcinomas (42,43) suggesting that alteration of GSTP1 activity is related to prostate carcinogenesis. The polymorphic site at codon 105 in exon 5, where an adenosine-to-guanine (A-G) transition causes an isoleucine to valine substitution (rs1695), has also been extensively described (44). The valine substitution results in lower enzyme activity (45,46) for certain substrates and thus may impair detoxification of carcinogens. While these interactions were not statistically significant, consideration of both genetic and environmental exposures may offer additional insight into our understanding given the known biologic impact of altered GSTP1 expression in prostate carcinogenesis.
Although we observed some modifying effects in SNPs in the UGT1A locus and in CYP1B1, the magnitude of these findings was not as large as those for GSTM3 and GSTP1. We also did not observe a strong modifying effect for SNPs in CYP1A1, CYP1A2, GSTA1, GSTM1, NAT1, NAT2, SULT1A1, and SULT1A2. The lack of findings in these genes does not mean that they do not play a role in altering risk via differential HCA metabolism. Variants in CYP1A2 and in the NATs have been well described in HCA metabolism and often implicated to potentially alter cancer risk (47,48). CYP1A2 is the principal hepatic CYP involved in the bioactivation of HCAs to their N-hydroxy metabolites and NATs are responsible for further bioactivation of N-hydroxy-HCA metabolites to genotoxic N-acetoxy-HCA esters (29). It is possible that we did not find an association for these genes because the bioactivation process is less important than the detoxification processes performed by the GSTs or that expression of these enzymes in the target tissue, the prostate, is lower. However, it is also possible that important SNPs in these genes and others were not genotyped or captured by our tagSNPs or that the power to detect these associations was low.
Strengths of our study included a relatively large sample, a detailed assessment of dietary HCA intake, and a comprehensive characterization of the genes involved in HCA metabolism. Given that this study was nested within the screening arm of the PLCO trial, selection and surveillance bias is minimal because both cases and controls had equal opportunity to be detected with prostate cancer. While this is the largest study to evaluate the interaction between dietary HCA intake and genetic variants in XME genes and risk for prostate cancer, the presence of low frequency variants limits the power to detect significant interactions. Thus, we have attempted to identify interesting associations that need to be followed up in future studies. Despite the potential limitation in power, it was important to characterize the top HCA intake category using the highest quintile of exposure based on previous positive associations with PhIP and prostate cancer only in the top quintile of intake. This study is comprised of only non-Hispanic Caucasian men, which limits the generalizability of our findings to other race/ethnic groups with different genetic compositions; however, as a result of the racial homogeneity, population stratification is unlikely to be a significant source of bias in this study (49).
Data from genome wide association studies have yielded novel and interesting insights into genetic risks for prostate cancer. These types of studies, however, do not take into consideration the complex interaction between genetic variants and environmental exposures. In conclusion, variants in two genes known to detoxify HCAs, GSTM3/EPS8L3 and GSTP1, modify the association between dietary HCA intake and prostate cancer. Despite the fact that genome wide scans have not identified these genes as important prostate cancer risk factors in main effect studies, additional studies with more power should continue to evaluate these genes in relation to environmental interactions and prostate cancer.
This research was supported by the Intramural Research Program of the National Institutes of Health (National Cancer Institute (Division of Cancer Epidemiology and Genetics) and by grant TU2 CA105666 from the National Cancer Institute.