Cyclooxygenase-2 Polymorphisms, Aspirin Treatment, and Risk for Colorectal Adenoma Recurrence—Data from a Randomized Clinical Trial Free

Cyclooxygenase-2 (COX-2) catalyzes the rate-limiting step in the production of prostaglandins, potent mediators of inflammation. Chronic inflammation plays an important role in the development and progression of colorectal cancer. Aspirin inhibits COX-2 activity and lowers the risk for colorectal adenomas and cancer. We investigated whether common genetic variation in COX-2 influenced risk for colorectal adenoma recurrence among 979 participants in the Aspirin/Folate Polyp Prevention Study who were randomly assigned to placebo or aspirin and followed for 3 years for the occurrence of new adenomas. Of these participants, 44.2% developed at least one new adenoma during follow-up. Adjusted relative risks and 95% confidence intervals (95% CI) were calculated to test the association between genetic variation at six COX-2 single-nucleotide polymorphisms and adenoma occurrence and interaction with aspirin treatment. Two single-nucleotide polymorphisms were significantly associated with increased adenoma recurrence: for rs5277, homozygous carriers of the minor C allele had a 51% increased risk compared with GG homozygotes (relative risk, 1.51; 95% CI, 1.01-2.25), and for rs4648310, heterozygous carriers of the minor G allele had a 37% increased risk compared with AA homozygotes (relative risk, 1.37; 95% CI, 1.05-1.79). (There were no minor allele homozygotes.) In stratified analyses, there was suggestive evidence that rs4648319 modified the effect of aspirin. These results support the hypothesis that COX-2 plays a role in the etiology of colon cancer and may be a target for aspirin chemoprevention and warrant further investigation in other colorectal adenoma and cancer populations.(Cancer Epidemiol Biomarkers Prev 2009;18(10):2726–33)

Mounting evidence from experimental and observational studies supports the role of inflammation in the development and progression of cancer, including colorectal cancer (1-4). Cyclooxygenase-2 (COX-2), also known as prostaglandin G/H synthase-2, is an inducible enzyme that catalyzes the rate-limiting step in the production of prostaglandins and plays a key role in inflammation (5). COX-2 may also have an important role in the etiology of colorectal and other cancers (6, 7). It is up-regulated in colorectal adenomas and cancer; approximately half of all colorectal adenomas and >85% of colorectal cancers have elevated levels of COX-2 (8-11). Increased COX-2 levels have been shown to correlate with later stage, larger tumor size, presence of lymphatic metastases, and risk for recurrence and poorer survival in human colorectal cancers (12-15). Expression and activity of COX-2 is thought to contribute to tumor promotion and carcinogenesis through stimulation of cell proliferation, inhibition of apoptosis, and promotion of angiogenesis and invasiveness (16, 17).

Important additional evidence supporting a role for COX-2 in the inflammation–colon carcinogenesis pathway comes from observational studies and randomized clinical trials showing that nonsteroidal anti-inflammatory drugs, including aspirin (18-21) and selective COX-2 inhibitors (22-25), reduce the risk for colorectal adenoma and cancer. Notably, aspirin use preferentially reduced the risk for colorectal cancers that overexpressed COX-2, suggesting that the anticancer benefit of aspirin is mediated, at least in part, by effects on COX-2 (26).

Twin studies indicate that heritable factors play a relatively large role in the causation of sporadic colorectal cancer, an estimated 35% (27). Although knowledge about the genetic factors that contribute to sporadic cancer is limited, several genomewide association studies have been published with replicated findings (28, 29). Genetic variation in the COX-2 gene may alter enzyme expression or activity, thereby altering the production of prostaglandins and potentially modulating an individual's inflammatory response and risk for colorectal cancer. In addition, it is possible that polymorphisms in the COX-2 gene could alter an individual's response to nonsteroidal anti-inflammatory drugs and thus modify the chemopreventive effect of nonsteroidal anti-inflammatory drugs in colorectal neoplasia (30).

Associations of polymorphisms in the COX-2 gene with colorectal adenomas and colorectal cancer have recently been investigated (31-39), including the analyses of interactions with aspirin use (33, 40-44), and the findings have been mixed. This existing literature is limited by the incomplete analysis of genetic variation in the COX-2 gene, small sample sizes and inconsistent and poorly defined measurement of nonsteroidal anti-inflammatory drug use. Further investigation into the association between COX-2 genotype and colorectal neoplasia and interaction with nonsteroidal anti-inflammatory drug use is needed to clarify the role of genetic variation in COX-2 and may identify individuals who would benefit the most from the use of nonsteroidal anti-inflammatory drugs in the prevention of colon cancer.

The present analysis was undertaken to investigate the impact of COX-2 polymorphisms on risk for colorectal adenoma recurrence and the interaction between COX-2 genotype and aspirin treatment among participants at high risk for adenoma recurrence who were randomly assigned to placebo or aspirin in a large adenoma chemoprevention trial (19). By providing a more comprehensive analysis of the genetic variation in a relatively large population with a precisely defined and randomized exposure to aspirin treatment, this investigation overcomes many of the limitations of the existing literature.

We conducted a cohort analysis of the association between COX-2 genotypes and colorectal adenoma recurrence among participants from the Aspirin/Folate Polyp Prevention Study (19, 45). Eligible participants were 21 to 80 years old with no history of colorectal cancer or any familial colorectal cancer syndrome but with a recent history of a histologically confirmed colorectal adenoma and a complete colonoscopy within 3 mo before enrollment with no remaining colorectal polyps. At enrollment, data were collected on potential risk factors for colorectal cancer, including cigarette smoking, alcohol use, and body mass index (BMI). After a run-in period during which compliance with study procedure was assessed, a total of 1,121 subjects were randomized to aspirin treatment (placebo, 81 or 325 mg daily) and folic acid treatment (placebo or 1 mg daily) in a 3 × 2 factorial design and followed until a subsequent colonoscopy was done. The study was completed by 1,084 participants (96.7% of 1,121 randomized) who had a follow-up colonoscopy at least 1 y after randomization (mean intervention period ± SD, 32.7 ± 3.8 mo). Of these participants, 979 (90.3%) had DNA available for genotyping and are included in the analyses presented here. This work was approved by the institutional review boards at all participating institutions, including the National Cancer Institute, where genotyping was done, and each of the collaborating clinical centers. All subjects provided written informed consent.

Genomic DNA was isolated from whole blood cells using proteinase K digestion and phenol-chloroform extraction. The COX-2 polymorphisms were genotyped with the 5′ nuclease Taqman allelic discrimination assay using the ABI 7900 Sequence Detection System (Applied Biosystems), as previously described (46). Briefly, oligonucleotide probes were labeled with two fluorescent dyes, FAM and VIC, to distinguish between the two alleles of each biallelic single-nucleotide polymorphism. All assays were designed and developed using Assays-by-Design (Applied Biosystems). All primers and probes were synthesized by Applied Biosystems. Each 384-well assay contained internal quality controls of homozygous wild type, heterozygous, and homozygous variant alleles for the respective polymorphisms along with no template controls. Assays were set up in 384-well plates using 2.5 μL of the 2× Taqman Universal Master Mix (no AmpErase uracil-N-glycosylase), including forward/reverse primers and FAM/VIC-labeled probes, and 2 to 5 ng of DNA in a final volume of 5 μL. The thermal cycling conditions for the ABI 7900HT Sequence Detector included an initial setting of 95°C for 5 min, followed by 40 cycles each of 92°C for 15 s and 60°C for 1 min. Data output was processed and downloaded electronically into the analysis program. The SDS 2.1 analysis software (Applied Biosystems) was used to determine the genotype calls.

Eight haplotype tagging single-nucleotide polymorphisms were chosen for genotyping based upon their minor allele frequencies of 9% and above as detailed at the SeattleSNPs Web site¹²

: rs689466 (798), rs20417 (1228), rs2745557 (2331), rs5277 (3355), rs20432 (5229), rs5275 (8494), rs2206593 (9123), and rs4648310 (11027). The numbers in parentheses refer to positions in the gene (Genbank entry AY382629). A total of 19 (2%) blinded replicates were included in the set of DNAs genotyped. Because the concordance rate for rs689466 was very low (47%), this single-nucleotide polymorphism was dropped from the analysis. For the remaining seven single-nucleotide polymorphisms, the concordance rate was 100% and the completion rates ranged from 96% to 99.6%. Observed and expected genotype counts among controls (participants with no adenoma recurrence) were evaluated for Hardy-Weinberg equilibrium by a 1 degree of freedom χ² test. One single-nucleotide polymorphism, rs2206593, deviated substantially from Hardy-Weinberg equilibrium (P < 0.0001), with fewer heterozygotes than expected. In addition, the minor allele frequency observed (0.34) was much higher than expected (0.03-0.10) in populations of European ancestry (CEU) based on information in the National Center for Biotechnology Information database (dbSNP), indicating likely genotyping error. Thus, rs2206593 was also dropped from the analysis. The remaining six single-nucleotide polymorphisms, which all exhibited Hardy-Weinberg equilibrium and allele frequencies similar to expected from available databases, were included in the analyses (see Table 1). Gene coverage was determined using the tagger application in the Haploview software program (47) by downloading CEU genotypes from the HAPMAP database and measuring the percentage of common variants (defined as single-nucleotide polymorphism minor allele frequency >0.03) tagged (r² > 0.8) by genotyping these six single-nucleotide polymorphisms.

At enrollment, subjects were asked to complete a questionnaire in which they identified their race or ethnic background by selecting one of the following: (1) White, not of Hispanic origin; (2) Black, not of Hispanic origin; (3) Hispanic; (4) American Indian or Alaskan Native; (5) Asian or Pacific Islander; (6) other; (7) uncertain. There were no responses of (7) uncertain. In Table 1, categories 4 and 6 are not shown because of the very small numbers in these two groups. For multivariable analyses, categories 2 to 6 were combined into one group to create a dichotomous variable.

Comparisons of subject characteristics by adenoma recurrence status were done by Pearson χ² tests for categorical variables and two-sample t tests for continuous variables. The principal outcome measured during the trial was the occurrence of one or more adenomas during randomized treatment; however, we also evaluated the occurrence of one or more advanced lesions (defined as tubulovillous or villous adenomas, adenomas ≥1 cm in diameter, adenomas with severe dysplasia, and invasive cancer). Risk ratios were calculated using an overdispersed generalized linear regression model for the Poisson distribution as an approximation to the binomial family (48). Relative risks and 95% confidence intervals (95% CI) were used to estimate the association between COX-2 genotype and risk for adenoma recurrence. Unadjusted, minimally adjusted (for age, gender, and race), and maximally adjusted (for age, gender, race, BMI, smoking status, alcohol use, clinical center, follow-up time, aspirin treatment, and folate treatment) relative risks were calculated. Results for crude and fully adjusted models were nearly identical to the minimally adjusted relative risks that are reported in the tables. Homozygosity for the most frequent genotype was set as the reference category, and relative risks were calculated by comparing the heterozygotes to the reference and the minor allele homozygotes to the reference using indicator variables (that is, an unrestricted genetic inheritance mode). Where appropriate (that is, rs5277), a recessive inheritance mode was also assessed by comparing minor allele homozygotes to a reference group including major allele homozygotes and heterozygotes.

Phased haplotype pairs and probabilities were estimated using Powermarker V3.25 (49) using the EM algorithm (50). Generalized linear regression was used to estimate haplotype association with risk for adenoma recurrence taking haplotype uncertainty (probabilities) into account. The haplotype was modeled as a continuous variable wherein the number of copies of each allele was multiplied by its probability to obtain a continuous variable, which was used as the predictor variable. The most common haplotype was used as the reference and omitted from the model. For each haplotype, the model provides an estimate of the risk associated with each additional copy of the specified haplotype. The most frequent haplotypes (with frequencies > 2%) were analyzed individually, and the remaining rare haplotypes were pooled.

We also evaluated whether aspirin treatment or participant characteristics interacted with COX-2 genotypes to modify associations with adenoma risk using interaction terms in the regression models and Wald tests. The following participant characteristics were considered: gender, age (above versus below median), BMI (<25, 25 to <30, ≥30 kg/m²), current alcohol use (drinker versus nondrinker), and smoking status (current smoker versus nonsmoker). To increase power for tests of effect modification, we combined the heterozygotes and the minor allele homozygotes into one category and compared it to the reference group of the major allele homozygotes. Stratified analyses were used to obtain stratum specific estimates of risk and confidence intervals. Effect modification was only assessed for the risk for any adenoma because of the small numbers of advanced lesions.

Analyses of study treatment with aspirin or folate were conducted according to the intention-to-treat principle. Two-sided P < 0.05 was considered statistically significant. We did not adjust for multiple testing as per convention in epidemiologic studies testing a priori defined hypotheses. Stata (version 9) was used for all analyses.

Demographics and other selected characteristics of the subset of participants with DNA available for COX-2 genotyping are presented in Table 2. Among the 979 participants, 433 (44.2%) had a recurrence of one or more colorectal adenoma during follow-up. Age, gender, and race were all associated with risk for adenoma recurrence: individuals with at least one colorectal adenoma recurrence during follow-up were slightly older than those who did not have a recurrence (59.1 and 56.4 years, respectively; P < 0.001), males were at greater risk compared with females (P = 0.004), and Hispanics and Asians seemed to be at reduced risk compared with Caucasians (P = 0.063 and 0.052, respectively). In addition, cigarette smoking, BMI, and alcohol use were associated with risk for adenoma recurrence: current smokers were at greater risk compared with nonsmokers (P = 0.001), individuals with a BMI between 25 and <30 kg/m² or ≥30 kg/m² were at greater risk than those with a BMI < 25 kg/m² (P = 0.018 and 0.005, respectively), and those who consumed alcoholic drinks were at greater risk than nondrinkers (P = 0.008). As observed in the full analysis of the trial (19, 45), individuals who were randomized to 81 mg/d of aspirin treatment were less likely to have a recurrence compared with those randomized to the placebo arm (P = 0.017), whereas 325 mg/d aspirin (P = 0.86) and folate had no significant effect (P = 0.40). Finally, the mean follow-up time did not differ according to adenoma recurrence status (P = 0.32).

Minor allele frequencies and gene locations for the six COX-2 single-nucleotide polymorphisms that were included in this analysis are shown in Table 1. All six single-nucleotide polymorphisms were in Hardy-Weinberg equilibrium among controls (participants with no adenoma recurrence), and genotype frequencies were similar to those expected for a population with predominantly European ancestry.¹³

Notably, the gene frequencies for several of the single-nucleotide polymorphisms (rs5277, rs20432, rs5275) varied statistically significantly by self-reported race and ethnicity (Table 1). Because of linkage disequilibrium, genotyping these six single-nucleotide polymorphisms captured ∼80% of the common genetic variants (single-nucleotide polymorphisms with minor allele frequency > 0.03) in the COX-2 gene.

We investigated whether any of the six COX-2 single-nucleotide polymorphisms were associated with the recurrence of any adenoma or with advanced lesions (Table 3). Two COX-2 single-nucleotide polymorphisms were associated with a statistically significant increased risk for adenoma recurrence: rs5277 and rs4648310. For rs5277, minor allele homozygotes (CC genotype) had a 49% increased risk for any adenoma compared with the common GG genotype (unadjusted absolute risk, 63.6% versus 46.7%; adjusted relative risk, 1.49; 95% CI, 1.00-2.23). Similar results were seen when the analysis was restricted to Caucasians only (adjusted relative risk, 1.61; 95% CI, 1.08-2.39). The pattern seen with rs5277 was indicative of a recessive inheritance mode because heterozygotes did not have an increased risk and was statistically significant when analyzed as such (adjusted relative risk, 1.48; 95% CI, 1.00-2.21; P = 0.05; not shown). For rs4648310, there were no minor allele homozygotes; however, the heterozygotes (AG genotype) had a statistically significant 35% increased risk for any adenoma compared with those with the common AA genotype (unadjusted absolute risk, 60.4% versus 43.3%; adjusted relative risk, 1.35; 95% CI, 1.03-1.77). There was also a nonsignificant 40% increase risk for an advanced lesion among these heterozygotes (adjusted relative risk, 1.40; 95% CI, 0.67-2.91). Similar results were seen when the analysis was restricted to Caucasians only for any adenoma (adjusted relative risk, 1.34; 95% CI, 1.01-1.76) or advanced lesions (adjusted relative risk, 1.42; 95% CI, 0.68-2.97). There was no evidence for heterogeneity between the results for advanced versus nonadvanced adenomas for any of the COX-2 genotypes analyzed (data not shown). In addition, in stratified analyses, there was no evidence for interaction between any COX-2 genotype and subject characteristics, including gender, age, BMI, alcohol use, or smoking status (data not shown).

The r² correlation between the two single-nucleotide polymorphisms associated with risk (rs5277 and rs4648310) was 0.16, indicating that they are not in strong linkage disequilibrium. In addition, when both single-nucleotide polymorphisms were included in the same regression model, their associations with risk (relative risks) remained substantially similar to those seen in the single single-nucleotide polymorphism models (in Table 3), although the associations were no longer statistically significant because of reduced power (data not shown).

Haplotype analysis revealed six common haplotypes with a frequency > 2% (Table 4). The five most common haplotypes were not associated with risk for adenoma recurrence. However, the sixth most common haplotype, with a frequency of 2.7%, was associated with a statistically significant 37% increased risk per copy of the haplotype relative to the most common haplotype (adjusted relative risk, 1.37; 95% CI, 1.03-1.82). Notably, this haplotype contained variant alleles at rs5277 and rs4648310, both of which were associated with deleterious effects at the genotype level (see Table 3). For advanced adenomas (data not shown), there was an increased risk for similar magnitude that was not statistically significant (adjusted relative risk, 1.36; 95% CI, 0.62-2.96).

Next, we evaluated whether there was evidence for an interaction between any COX-2 genotype and aspirin treatment on risk for adenoma recurrence (Table 5). There was suggestive evidence for an interaction between rs4648310 genotype and 81 mg of aspirin treatment. As observed in the overall population, 81 mg of aspirin treatment daily was associated with a protective effect among major allele homozygotes (AA genotype; adjusted relative risk, 0.76; 95% CI, 0.63-0.91). However, 81 mg of aspirin was not protective among heterozygotes (AG genotype): relative risk is 1.55 (95% CI, 0.98-2.46) compared with AA homozygotes given placebo (Table 5) and relative risk is 1.26 (95% CI, 0.61-2.57) compared with AG heterozygotes given placebo (not shown). This interaction did not reach statistical significance (P = 0.11) unless the placebo and 325 mg aspirin groups were combined (P = 0.037; not shown). Similar results were obtained when this analysis was restricted to Caucasians (not shown).

In this analysis of a randomized aspirin trial, we investigated the impact of common COX-2 variants on risk for colorectal adenoma recurrence and aspirin chemoprevention. We observed statistically significant increased risks for adenoma recurrence of 49% for the rs5277 CC genotype and 35% for the rs4648310 AG genotype relative to the homozygoud wild type genotypes. Notably, for each copy of a common haplotype containing both variant alleles, there was a statistically significant 37% increased risk for adenoma recurrence. In addition, there was suggestive evidence that the protective effect of treatment with 81 mg/d aspirin on risk for colorectal adenoma recurrence was modified by the rs4648310 genotype because the aspirin protective effect was not seen among AG heterozygotes.

A recent review summarized the associations of COX-2 polymorphisms with risk for colorectal adenoma and cancer (51). Although results on 17 COX-2 polymorphisms from 14 studies were reviewed, the ability to reach conclusions was hampered by small sample sizes and other significant limitations in the studies and the analyses (51). Four of the COX-2 polymorphisms that we examined in the current study have been investigated previously for associations with colorectal adenoma, including rs20417, rs20432 and rs5275 (for which we observed no association), and rs5277 (for which we observed an increased risk). For rs20417, pooled analysis showed no association with adenomas, but a trend toward increased risk for cancer that reached statistical significance only among Asians (51). For rs5275, pooled analyses showed a nonsignificant trend toward a protective effect for adenomas and no association with cancer (51). For rs20432, no associations were found for adenomas or cancer (51). For rs5277, a synonymous single-nucleotide polymorphism in exon 3, we found a statistically significant increased adenoma risk, whereas in the pooled analysis (51), there was a nonsignificant trends toward a reduced adenoma risk based on two previous studies (42, 44) and toward an increased cancer risk cancer based on two previous studies(34, 52). Thus, our data about rs5277 seem to be more consistent with the previous cancer data than the adenoma data. We also examined two COX-2 polymorphisms not previously investigated: rs2745557 and rs4648310. We did not detect an association for rs2745557 with adenoma risk. However, rs4648310, located in the 3′ untranslated region, was associated with an increased risk for any adenoma and a nonsignificant increased risk for advanced adenoma.

Another recent review examined interactions of COX-2 polymorphisms with nonsteroidal anti-inflammatory drug use in preventing colorectal neoplasia (53). A total of 17 polymorphisms have been investigated across six studies, but no statistically significant interactions have been reported, possibly because of limited sample sizes (53). A borderline significant interaction was reported for rs20417 (40) but was not confirmed in a different study (42) or in the present analysis. In the current study, we found suggestive evidence for an interaction of aspirin treatment with rs4648310, which also had a main effect on adenoma risk. However, the power to detect interactions was limited in the present analysis, and it will be important to follow-up on this finding in other populations because of the potential clinical implication for aspirin chemoprevention. Nonetheless, at the population level, the impact will be limited because of the low frequency of the variant allele. As discussed previously, aspirin is not thought to inhibit COX-2 at the doses used here; however, it may suppress the expression of COX-2 (19). Regardless, the mechanism is not clear, and this finding may be due to chance. Finally, it is worth noting that the effects of aspirin are likely to involve multiple pathways in addition to COX-2, including COX-1 and other non-COX mechanisms (54, 55).

The functional effects of the two single-nucleotide polymorphisms associated with adenoma risk, rs4648310 and rs5277, are not yet known. In addition, it cannot be determined from this investigation whether they represent causative single-nucleotide polymorphisms or whether they are simply markers in linkage disequilibrium with unknown or unmeasured causative single-nucleotide polymorphisms. However, based on their association with increased risk, one possibility is that they may either cause or be associated with an increase in COX-2 activity or expression. Notably, rs4648310 was also recently associated with an increased risk for aggressive prostate cancer, and this effect was modified by dietary ω-3 fatty acids (56).

There are several potential limitations to consider in the current analysis. The generalizability of the results may be limited because the study was conducted on predominantly Caucasian volunteers in a clinical trial with a previous history of adenoma. The endpoint of the study was occurrence of new adenomas rather than colorectal cancer. In addition, we had limited power to investigate risk for advanced adenomas, which were uncommon during follow-up (10% of the population) and interactions with two levels of aspirin treatment. Finally, the associations detected were modest, and our findings may be due to chance. Nonetheless, there are several important strengths associated with this study, including the relatively large sample size and the prospective study design, which make recall or selection bias unlikely. In addition, the outcome (adenoma) was uniformly assessed by a single study pathologist who was blinded to study treatment assignment. Importantly, exposure to aspirin was by randomized treatment with excellent compliance and follow-up. Finally, our analysis encompassed ∼80% of the common genetic variants in the gene.

In summary, we observed an increased risk for colorectal adenoma recurrence in association with two variants (rs5277 and rs4648310) in the COX-2 gene, as well as a possible interaction between aspirin use and one of these variants (rs4648310). Although these results provide support for the hypothesis that COX-2 plays a role in the etiology of colon cancer and may be a target for aspirin chemoprevention, the associations were modest and may be due to chance. Thus, it is important to interpret these results with caution and to replicate them in other colorectal adenoma and cancer populations.

J.A. Baron: Consultant/Advisory Board, Bayer; Minor Consultant, Pozen. R.S. Sandler: Consultant/Advisory Board, Pozen. Bayer provided the aspirin and placebo tablets for the clinical trial.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank all individuals who participated in the Aspirin/Folate Polyp Prevention Study and made this research possible, and Bayer for providing the aspirin and placebo tablets for the clinical trial.