In this case-only analysis, there was evidence of departure from multiplicativity indicating a gene-environment interaction between EPHX1 c.337T>C and smoking, particularly in smokers with a history of greater than 20 pack-years of smoking. In addition, there was a gene-gene interaction between CYP1A1 c.1384A>G and EPHX1 c.337T>C. Therefore, our results suggest that (1) the impact of cigarette smoking on CRC risk is synergistically increased among individuals who carry a variant allele of EPHX1 c.337T>C and (2) individuals who carry variant alleles of both CYP1A1 c.1384A>G and EPHX1 c.337T>C are more susceptible to CRC. Furthermore, in an analysis stratified by smoking status, we found that the OR for gene-gene interaction effect was statistically significant for the EPHX1 homozygous variant genotype (CC) in both never-smokers and ever-smokers but that the OR was higher in ever-smokers. Similarly, the OR for gene-gene interaction was significant in males but not in females, although overall, there was no significant interaction between gender and either of the SNPs. These results were based on a limited number of patients, but they suggest that compared with nonsmokers with wild-type alleles for both SNPS, smokers with variant alleles of CYP1A1 c.1384A>G and smokers with the homozygous variant genotype for EPHX1 c.337T>C are at significantly increased risk for CRC and the gene-gene interaction effect may be gender specific, as it was evident only in males.
In our previous study on individuals with Lynch syndrome, we found evidence for multiplicative interaction between CYP1A1 c.1384A>G and EPHX1 c.337T>C (P for interaction term = 0.036; Wald χ2 P = 0.044) with a greater than multiplicative hazard ratio for the combined effect of having a variant allele of each of these SNPs (hazard ratio, 3.09, 95% CI, 1.58–6.04; P = 0.001). The purpose of the present study was to see if these findings were replicable in cases of nonsyndromic CRC. Even though the study designs used to obtain the interaction estimates were different (retrospective cohort versus case-only design), the results from both studies were statistically significant.
Most of the association studies for EPHX1
and smoking in colorectal carcinogenesis have examined colorectal adenoma as the outcome [15
]. Of three studies with CRC as the outcome one reported an increased frequency of the c.337C variant allele in CRC cases compared to controls [19
], one found a reduced CRC risk associated with the c.337C allele [20
] and one large case-control study found no association [21
]. Similarly, there have been conflicting results for the interaction effect between smoking and EPHX1
. Ulrich et al. [18
] found that the variant EPHX1
c.337C allele increased adenoma risk among smokers and the risk was highest among those with greater than 25 pack-years of smoking (similar to our findings for CRC risk), whereas other studies reported reduced risk associated with the low activity c.337C allele in the presence of smoking [15
]. Further validation of the influence of EPHX1
on risk of CRC in the presence of smoking may therefore be warranted.
Few studies have evaluated the association between CYP1A1
and CRC, though CYP1A1
has been extensively evaluated in other smoking-related cancers (reviewed in [22
]). Slattery et al. examined the association between CYP1A1
c.1384A>G and smoking in risk of CRC in 1026 cases and 1185 controls and found that the individuals at highest risk for CRC were men who were currently smoking and had any CYP1A1
variant allele. The authors concluded that the impact of smoking on CRC risk may depend on CYP1A1
]. Fan et al. used a case-only study similar to our own study to determine the interactions between certain polymorphisms in metabolic enzymes and smoking in 207 Chinese patients with CRC and found a significant gene-gene interaction between CYP1B1
1294G and SULT1A1
638A alleles (OR 2.68, 95% CI = 1.16–6.26) and gene-environment interaction between CYP1B1
1294G and smoking (OR 2.62, 95% CI = 1.01–6.72) [24
]. However, the results of their study and our study cannot be compared since the polymorphisms examined in the two studies were different.
PAHs in cigarette smoke are substrates for both CYP1A1 and EPHX1, and these two enzymes act sequentially to metabolize PAHs. Therefore, a biological interaction effect may exist between CYP1A1 and EPHX1. First, CYP1A1 converts benzo(a)pyrene to the active benzo(a)pyrene 7,8 epoxide. This is then hydrated by EPHX1 to a transhydrodiol derivative, benzo(a)pyrene 7,8 diol, a product that is less toxic [16
]. However, the diol derivative is also a primary substrate for CYP enzymes that oxidize it further to benzo(a)pyrene 7,8 dihydrodiol 9,10 epoxide (BPDE), which is highly reactive and capable of forming DNA adducts. Therefore, these genes may interact to play a more complex role in cancer susceptibility.
The case-only approach was appropriate for our study since it was used to validate a priori findings of an interaction effect between two SNPs. However, a case-only study does have the disadvantage of not allowing evaluation of the independent effect of either of the exposures, smoking alone or the CYP1A1
genotypes alone, but only allowing evaluation of their interactions. It also does not allow assessment for departures from joint additive effects (can only test departures from joint multiplicative effects) of the exposure and genotype or the genotypes with each other. However, the case-only design is efficient (smaller sample size required to assess interaction than in a case-control design) [14
] and offers less potential for misclassification of exposures.
Though both CYP1A1
c.1384A>G and EPHX1
c.337T>C are nonsynonymous SNPs and therefore likely have functional consequences, we queried two programs (PolyPhen [http://genetics.bwh.harvard.edu/pph/
] and SIFT [http://blocks.fhcrc.org/sift/SIFT.html
]) that predict the impact of an amino acid substitution on protein function to assess whether these SNPs are potentially deleterious. Polyphen classified CYP1A1
c.1384A>G as “benign,” and SIFT classified it as “tolerated.” EPHX1
c.337T>C, on the other hand was classified as “possibly damaging” by Polyphen and “intolerant” by SIFT. The EPHX1
c.337T>C SNP may therefore be important to follow up on in functional studies in CRC.
Our study was underpowered for subgroup analysis by race/ethnicity due to the low frequency of the CYP1A1 variant genotype. The OR for interaction between CYP1A1 c.1384A>G and EPXH1 c.337T>C (homozygous variant) lost significance when we analyzed the non-Hispanic Whites alone (OR(EPHX1:CC genotype) = 2.2, 95% CI = 0.87–5.56, P = 0.09). However, evidence for interaction between EPHX1 c.337T>C and smoking remained significant when the analysis was limited to non-Hispanic whites—that is, among whites, ever-smokers with one or two copies of the EPHX1 variant c.337C allele had OR = 1.4 (95% CI = 1.05–1.97, P = 0.02) compared to never-smokers with the TT genotype. Overall, including ethnicity as a covariate while testing for interaction, did not alter the main effect estimates. Analyzing interaction effects in a larger sample of each ethnic subgroup may be required to validate these findings more globally.
Our finding of evidence for a gene-gene interaction between CYP1A1
c.1384A>G and EPHX1
c.337T>C in risk for sporadic CRC is especially meaningful, since to our knowledge this interaction has not previously been described and since this finding validates a similar interaction seen in our previously reported study [9
] in a different study population. Future plans for these findings would include evaluating the gene-gene and gene-environment relationship as predictors of CRC recurrence and survival. In conclusion, while low penetrance genes like CYP1A1
may raise the cancer risk only slightly independently, in combination they may greatly increase cancer susceptibility. Therefore, individuals who have multiple genetic susceptibility alleles and are smokers may be a subgroup that could be targeted for more intensive interventions than is recommended for the general population.