Genetic Variation in Prostaglandin E2 Synthesis and Signaling, Prostaglandin Dehydrogenase, and the Risk of Colorectal Adenoma

Aspirin and other nonsteroidal anti-inflammatory drugs (NSAID) are effective chemopreventive agents for colorectal adenoma and cancer (1-4). The main targets of NSAIDs are the cyclooxygenase-1 (COX-1) and COX-2 proteins, which catalyze the first step in the prostaglandin synthesis pathway. This pathway metabolizes ω-6 (n-6) and n-3 poly-unsaturated fatty acids into prostaglandins, many of which are potent inflammatory mediators. Prostaglandin E₂ (PGE₂), one of the main downstream products of COX metabolism, is upregulated in colon cancer (5) and is associated with increased cell proliferation, survival, and motility (6-8). Apc Min mice treated with PGE₂ have increased intestinal tumor number and size, compared with untreated mice (9). Conversely, PGE₂ receptor knockout mice show reduced intestinal tumorigenesis (10-12).

PGE₂ is produced by PGE₂ synthase, of which there are three forms (13). One of these, microsomal PGE₂ synthase-1 (PGES), is induced by proinflammatory stimuli and is preferentially coupled to COX-2 over COX-1 in several tissues (14). PGES is overexpressed in colorectal adenoma and cancer (13, 15), and transfection of PGES in concert with COX-2 resulted in faster growth, increased cell aggregation, and abnormal cell morphology (16), indicating that induction of PGES may be driving prostaglandin-related inflammatory processes.

PGE₂ signals through four G protein–coupled cell surface receptors called EP1-4. The four receptor types have different downstream mechanisms of signal transduction, resulting in a wide variety of functions, although EP2 and EP4 have similar modes of action (17). Of the four EP receptors, EP2 and EP4 have been the most consistently linked to colorectal carcinogenesis (10, 12, 17, 18). APC^Δ716/EP2 receptor-null mice had fewer intestinal polyps, reduced tumor growth, and reduced angiogenesis (10, 11, 19), indicating that PGE₂ signaling through this receptor is an important contributor to colorectal neoplasia. Higher EP2 expression was associated with poorer colorectal cancer survival in a study of 99 Swedish cancer patients (18).

In mouse models, knocking out EP4, either through genetic deletion or through treatment with an EP4-receptor antagonist reduced the formation of aberrant crypt foci and decreased polyp formation (12). In an in vitro study, treatment of cells with an EP4 agonist countered the chemopreventive effects of NSAID treatment (20), indicating that PGE₂ signaling through the EP4 receptor also plays an important role in the development of colorectal neoplasia. However, in mouse models of colitis, a condition predisposing to colorectal cancer, treatment with EP4 agonists ameliorates symptoms (21, 22), indicating that the full extent of the relationship between EP4 and colon carcinogenesis is not yet well understood.

Prostaglandins can be inactivated by 15-hydroxyprostaglandin dehydrogenase (PGDH; ref. 23). APC Min ^−/− mice exhibit lower PGDH expression than APC Min ^+/− mice, and Pgdh knockout mice have greatly increased tumor development (24, 25). Human colon cancer cell lines exhibit low levels of PGDH expression compared with normal colon mucosa (24, 26), and expression of PGDH mRNA in human colon cancer tissues is greatly reduced compared with normal colon tissue (24-26). PGDH expression is also reduced in inflammatory bowel disease (27), a condition associated with markedly increased colorectal cancer risk.

Use of NSAIDs substantially reduces PGE₂ levels (28-30), induces PGDH expression (24, 31, 32), and may decrease expression of EP2 (33), indicating that reduction of PGE₂ production or signaling may be important mechanisms by which NSAIDs prevent colorectal neoplasia. Dietary factors can also influence prostaglandin levels. n-3 polyunsaturated fatty acids (PUFAs), such as eicosapentaenoic acid, compete with arachidonic acid for metabolism by the prostaglandin pathway, resulting in production of 3-series prostaglandins that have reduced inflammatory potential (34-36). Thus, higher intakes of n-3 fatty acids, which are abundant in fatty fish, may also result in reduced risk of colorectal adenoma or cancer.

Given the known relevance of this pathway to colorectal carcinogenesis, we evaluated the genetic variability in key proteins related to PGE₂ synthesis, inactivation, and signaling in relation to colorectal polyp risk. We have previously shown that genetic variability in the prostaglandin synthesis pathway may alter the chemopreventive associations between adenoma risk and NSAID use or fish intake (a proxy for n-3 polyunsaturated fatty acid intake; refs. 37-40). We and others have screened the coding regions of PGES, EP2, and EP4 for polymorphisms (41, 42). We investigated potential associations between tagSNPs in PGES, EP2, EP4, and PGDH, as well as two rare candidate nonsynonymous single nucleotide polymorphisms in EP4, and risk of colorectal adenoma. We further explored the potential interactions between these polymorphisms and use of aspirin or other NSAIDs and fish intake.

Participant recruitment has been previously described (4, 43, 44). Briefly, adenoma cases and colonoscopy-screened polyp-free controls were recruited through a large multiclinic gastroenterological practice in the Twin Cities area of Minnesota from April 1991 to April 1994. Eligibility criteria have been described elsewhere (43), participants were recruited before colonoscopy and were ages 30 to 74 y; English-speaking residents of the Twin Cities metropolitan area with no known genetic syndrome associated with increased risk of colon neoplasia; and with no individual history of cancer (except nonmelanoma skin cancer), prior colorectal polyps, or inflammatory bowel disease. The participation rate for all patients who underwent colonoscopy was 68%.

Information on use of aspirin or other NSAIDs, diet, physical activity, anthropometrics, demographics, and medical history was obtained through questionnaire. Participants reported the average weekly consumption of aspirin and other NSAIDs, and the duration of use. Study participants were provided with a list of 14 common aspirin brands and 24 common nonaspirin NSAIDs. Non-NSAIDs, which were to be excluded from responses (e.g., acetaminophen), were also listed for guidance.

Information about regular weekly fish intake over the year before polyp diagnosis or reference date was obtained using a food frequency questionnaire. The food frequency questionnaire was an adaptation of the Willett semiquantitative food frequency questionnaire, which has been studied previously for validity and repeatability within the Nurses' Health Study cohort (45), the Iowa Women's Health Study cohort (46), and the Health Professionals Follow-up Study cohort (47). Participants were asked four questions about their usual intakes of various types of fish, including canned tuna, fatty fish, and white fish. These responses were combined to form a total fish intake variable.

Polymorphisms that may affect protein levels or function are called candidate polymorphisms. In our recent screening of PGES, EP2, and EP4 (41), only two nonsynonymous polymorphisms in EP4, Thr176Ile [minor allele frequency (MAF), 1%) and Val294Ile (MAF, 3%), were found in Caucasians. Of these, Thr176Ile was predicted to affect protein function using the SIFT and PolyPhen algorithms (41). We genotyped both of these nonsynonymous SNPs in the present study.

The coding regions, as well as 2KB 5′ and 3′ of PGES, EP2, EP4, and PGDH, were resequenced in 23 individuals of European descent and 24 individuals of African-American descent by the University of Washington-Fred Hutchinson Cancer Research Center Variation Discovery Resource (42). TagSNPs were selected from the European-descent population using the LD Select algorithm developed by Carlson and colleagues (48), with a cutoff MAF of 4% (i.e., any variant that occurred twice among the European population) and an r² value of 0.90. This resulted in the selection of 10 tagSNPs in PGES, 11 in EP2, 11 in EP4, and 32 in PGDH, estimated by the Genome Variation Server (49) to cover 85% of the common (MAF, ≥4%) variation in these loci (see Supplementary Table S1). A total of 51 SNPs successfully converted to the Illumina GoldenGate genotyping platform. Most of the polymorphisms that failed to convert (10 of 13) were “singleton” SNPs, not tagging for any other polymorphisms; the three failed SNPs that tagged “bins” of linkage disequilibrium (LD), as well as two candidate polymorphisms, were subsequently genotyped individually at Fred Hutchinson Cancer Research Center (see below).

The majority (51 of 56) of SNPs were genotyped using the Illumina GoldenGate bead-based genotyping technology at the Translational Genomics Institute (TGen). For polymorphisms that were not amenable to high-throughput methods, genotyping was conducted at Fred Hutchinson Cancer Research Center. These polymorphisms (EP4 Thr176Ile, EP4 Val294Ile, PGES −1254T>C, PGDH 12184A/−, and PGDH 19979A>T) were detected by allelic discrimination using the 5′ nuclease assay on a 7900HT sequence detection system (Applied Biosystems). The 5′ nuclease genotyping assays were validated by genotyping 92 individuals by both 5′ nuclease assay and restriction fraction length polymorphism (RFLP). There were no discrepancies between the two assays. The 5-μL genotyping reactions contained the Taqman universal master mix, primers, probes, and 2 ng of genomic DNA. Positive controls for all the genotypes as well as two negative controls were included on each plate.

Intraplate and interplate replicates at a rate of ∼5% were included on all plates and in all batches. Blinded duplicates were also included on all plates as another quality control measure. Genotype data from 30 CEPH trios (Coriell Cell Repository, Camden, NJ) genotyped by the HapMap project were used to confirm the reliability and reproducibility of the genotyping. Genotypes were excluded from analyses if SNPs had <85% call rate, <85% concordance with blinded or nonblinded duplicates, or Hardy-Weinberg Equilibrium P < 0.0001 (see Supplementary Table S1). Eight tagSNPs were excluded due to quality control errors, leaving a total of 46 tagSNPs and two candidate SNPs in four genes included in the analyses presented here.

Unconditional, logistic regression was used to estimate the odds ratio (OR) and corresponding 95% confidence interval (CI) for the associations between genotypes in PGES, EP2, EP4, and PGDH and risk of adenoma. Most SNPs were analyzed using indicator variables for the heterozygous and homozygous variant genotype groups (codominant or unrestricted model). If fewer than 10 cases or controls had the homozygous variant genotype, we grouped the homozygous variant genotypes with the heterozygous genotypes for analysis (dominant model). In main-association models of genotypes, associations were adjusted for age and sex. Effect modification by NSAID use or fish intake was evaluated in two ways: (a) by the inclusion of multiplicative interaction terms in logistic regression models and (b) by testing for a difference in trends within strata of genotype or NSAID/fish, coded as a continuous variable. Because use of NSAIDs or dietary intakes of fish may be associated with other known risk factors for colorectal neoplasia, we adjusted our NSAID- and fish-gene interaction analyses for age, sex, body mass index, dietary intakes of fiber, alcohol, energy, postmenopausal hormone use, and smoking. Because of possible racial differences in genotype frequencies, our analyses were restricted to Caucasian individuals (97% of the total study population). To obtain tests for trend, the genotypes were treated as a continuous variable, with the wild-type genotypes coded as 0, the heterozygous coded as 1, and the homozygous variant coded as 2.

All statistical analyses were carried out using SAS v.9. As a secondary analysis, P values were adjusted for multiple testing taking into account correlated tagSNPs using the p(ACT) method (50). We corrected for multiple testing on the gene level, i.e., all tests within a gene were corrected for the total number of SNPs within a gene.

Haplotype analysis and principal components analysis were used to evaluate the combined association of tagSNPs on a gene level (i.e., all SNPs in a gene were analyzed together). Haplotype blocks were determined in Haploview (51) using the Gabriel method (52); the resulting blocks were individually analyzed in R version 2.8.0 using a modified version of the haplo.stats package (53), which allows adjustment for age and sex. All haplotypes with >5% frequency in the controls were included in the logistic regression models as a categorical variable; haplotypes with <5% frequency were grouped and included in the model as one variable. The most common haplotype among the controls was used as the reference group. Gene-level significance was determined using a score test of the haplotypes included in the model.

For principal components analysis (54), we determined the number of principal components that explained at least 80% of the variance in adenoma risk and performed logistic regression using those major components, adjusted for age and sex. The gene-level significance was determined using a likelihood ratio test comparing a model that contained the principal components and one that did not.

We evaluated gene-gene interactions using two methods: (a) the Tukey 1 degree of freedom interaction test proposed by Chatterjee et al. (55) and (b) logic regression (56). For the Chatterjee method, we performed the test of interaction in two ways: (a) by deriving scores for each parameter and then testing the maximum score against the normal distribution (i.e., asymptote-based testing), or (b) by permutation testing. Logic regression is a model selection method that uses Boolean logic to find combinations of variables that predict the outcome variable (56). For the purposes of these comparisons, we ran models of various sizes to compare the fit and we also performed the Monte Carlo Logic Regression (57) to obtain a list of SNPs that show up frequently in the best fitting model, suggesting that these SNPs might be important predictors of disease risk. All gene-gene interaction analyses were conducted in R (version 2.8.0).

Characteristics of the study population have been previously described (4, 43, 58) and are shown for the genotyped subset in Table 1 (n = 483 adenoma cases and 582 polyp-free controls). Briefly, the study population was mostly Caucasian (97%); adenoma cases tended to be older than controls and more likely to be male. Regular aspirin or other NSAID use was somewhat more common among controls than among adenoma cases (aspirin: OR, 0.63; 95% CI, 0.44-0.90; other NSAIDs: OR, 0.50; 95% CI, 0.31-0.82; ref. 4). The quality control results for each SNP are shown in Supplementary Table S1.

Associations between risk of adenoma and tagSNPs in PGES, EP2, EP4, and PGDH are shown in Tables 2 and 3. We observed a marginally significant association with EP2 4950G>A (OR_{GA or AA vs. GG}, 0.71; 95% CI, 0.52-0.99). There were no individuals with the variant genotype in the candidate SNP EP4 Thr176Ile in the study population, so data for that polymorphism are not shown. There was no association with the candidate SNP EP4 Val294Ile (Table 2). The heterozygous genotype for EP4 −1307G>A was associated with a 40% increase in adenoma risk compared with the most common genotype (OR_{GA vs. GG}, 1.42; 95% CI, 1.06-1.91). An increase in risk was also observed for the homozygous variant genotype group; however, this was not statistically significant (Table 2).

For PGDH, we observed two associations of marginal statistical significance (Table 3). The heterozygous genotype of 782A>G was associated with decreased adenoma risk compared with wild-type (OR_{AG vs. AA}, 0.76; 95% CI, 0.57-1.01), although the homozygous variant genotype was not associated with polyp risk. Similarly, the heterozygous genotype of 1543G>A was associated with a decreased risk of marginal statistical significance compared with wild-type (OR_{GA vs. GG}, 0.78; 95% CI, 0.58-1.07), whereas the homozygous variant genotype was not associated with altered risk. No associations were observed for tagSNPs in PGES.

We observed several interactions between fish intake and tagSNPs in PGES, EP4, and PGDH. In PGES, a statistically significant interaction with fish intake was observed for PGES −664A>T (Table 4; Fig. 1). Fish intake was not associated with adenoma risk among those with the wild-type genotype (OR_{1-2/wk v. <1/wk}, 0.93; 95% CI, 0.65-1.34; OR _{>2/wk v. <1/wk}, 1.31; 95% CI, 0.83-2.04); however, among those with the variant genotypes, higher fish intake was associated with decreased adenoma risk (OR_{>2/wk v. <1/wk}, 0.56; 95% CI, 0.28-1.13; P_{trend interaction} = 0.01).

For the candidate polymorphisms EP4 Val294Ile, among the Val/Val genotype group, increasing fish intake was associated with a possibly decreased risk of adenoma (OR_{>2/wk v. <1/wk}, 0.87; 95% CI, 0.58-1.31), whereas among those with at least one Ile allele, higher fish intake was associated with substantially increased adenoma risk (P_{trend interaction} = 0.01; Table 4). However, the EP4 Val294Ile is a rare polymorphism and the numbers of subjects with a variant allele were very small (n = 44), indicating that this interaction requires confirmation in a larger population.

For PGDH 31659T>C (3′ untranslated region), no association with fish intake was observed among those with the common genotype. Those with the heterozygous or homozygous variant genotypes had a reduced adenoma risk with increasing weekly fish intake (P_{trend interaction} = 0.02; Table 4). No interactions with fish intake were observed for EP2. None of the observed interactions remained statistically significant when adjusted for multiple comparisons.

We also evaluated NSAID interactions, but we observed no strong interactions. For two SNPs [EP2 9814C>A (intron 1) and PGDH 343C>A (intron 1)], there were marginal or statistically significant interactions (see Supplementary Table S2). We analyzed NSAID interactions using the wild-type non-NSAID users as the reference group. For the EP2 SNP, the risk reduction associated with regular NSAID use was more pronounced among those with at least one of the more common alleles. However, for PGDH 343C>A among NSAID nonusers, the variant genotypes were associated with increased risk (OR_{CA or AA vs. CC}, 1.98; 95% CI, 1.18-3.33); among regular NSAID users, decreased adenoma risk was observed for all genotypes (P_interaction = 0.03). No interactions with NSAID use were observed for SNPs in PGES or EP4. None of the interactions remained statistically significant when adjusted for multiple comparisons.

Haplotype and principal components analysis results are shown in Supplementary Tables S3 and S4, respectively. There were no statistically significant associations observed for any of the four genes.

The results of Chatterjee gene-gene interaction testing are shown in Supplementary Table S5. No statistically significant gene-gene interactions were detected. The results of logic regression modeling are shown in Supplementary Tables S6 to S8. In Supplementary Table S6, the variables that were included in models of various sizes are shown. The most commonly occurring pairs of SNPs in 100,000 iterations of the Monte Carlo Logic Regression are shown in Supplementary Table S7. None of the pairs of SNPs occur with ≥1% frequency in 100,000 iterations, indicating that important SNP-SNP interactions are unlikely in these genes. The most commonly occurring SNPs are shown in Supplementary Table S8. The most commonly occurring single SNPs in 100,000 iterations of Monte Carlo Logic Regression were EP4 −1307G>A, which occurred with a frequency of 13.3%, and EP2 4950G>A, which occurred with a frequency of 7.3%; these were the only SNPs that occurred with a frequency of >5%. Our finding that no single SNP was selected for inclusion in >15% of fitted models indicates that none of the SNPs we genotyped in these four genes is strongly related to risk of adenoma. This reflects the main SNP association findings, in which we detected very few statistically significant main associations.

We investigated genetic variability in a biochemical pathway that has been unequivocally linked to colorectal carcinogenesis and has not yet been previously studied in the context of colorectal adenoma. Our results indicate that genetic variability related to PGE₂ signaling is generally not related to risk of colorectal adenoma. However, some data from this study suggest that genetic variability in PGES, EP2, EP4, or PGDH may alter the anti-inflammatory effects of specific fatty acids or NSAID use. We hypothesized that NSAID use or high fish intake would be associated with lower risk of colorectal neoplasia, predominantly among those with genetic variants that are likely to increase PGE₂ production or signaling. We observed a statistically significant association for a tagSNP in EP2 and in EP4, but no other associations between tagSNPs in PGES or PGDH and risk of colorectal adenoma. Further, we observed several statistically significant or marginally significant interactions with NSAID use or fish intake, indicating that the consequences of genetic variability in prostaglandin synthesis may be further modified by underlying dietary or pharmacologic proinflammatory or anti-inflammatory profiles. For the tagSNPs that were associated with adenoma risk or interacted with inflammatory exposures, we investigated their LD with any potentially functional SNPs (Supplementary Table S9) by evaluating SNP locations, regulatory regions, and conservation across species. However, no SNPs with clear functional significance were identified.

To our knowledge, no previous studies have examined associations between genetic variants in PGES, EP2, EP4, and PGDH and risk of adenoma. However, although our findings require confirmation in additional studies, our results suggest similar patterns to previous reports on interactions between polymorphisms in COX-1, COX-2, and PGIS, in which the inverse associations with NSAID use or fish intake were limited to certain genotype groups (37-40). Two EP2 tagSNPs that were included in this study, −616G>C and −166G>A, were associated with aspirin-intolerant asthma in Korean individuals (59). Although that study was small and 77 polymorphisms were tested without correction for multiple testing, those results indicate that polymorphisms in EP2 may be a predictor of who is likely to benefit from (or even experience adverse reactions from) aspirin. Two EP2 SNPs included in the present study (−1722A>G and −616G>C) were not associated with hypertension in a study of 266 Japanese subjects (60) and were also not associated with adenoma risk in the present study. In a recent whole genome association study of Crohn's disease, an inflammatory bowel condition, SNPs upstream of EP4 linked to altered EP4 expression were associated with increased risk (61). However, these SNPs were not included in the present study because they are over 200 KB upstream of EP4. In a study of lymphoma risk, three polymorphisms in PGES that were not included in this study (but one of which was tagged in our study by another SNP) were not associated with risk of lymphoma (62).

This study has several limitations. First, a large number of statistical tests were done, increasing the likelihood of false positives. Although we corrected for multiple testing, we chose to report the uncorrected P values. Given both the importance of PGE₂ for colorectal carcinogenesis and the fact that no one has yet reported on potential associations with tagSNPs in these genes for colorectal neoplasia risk, we decided to report all our findings for possible replication by other groups. Conversely, due to the relatively small number of study subjects, it is possible that true associations or gene-environment interactions were missed.

Intakes of specific types of fish were not measured in this study. Because fish vary in their content of n-3 polyunsaturated fatty acids, total fish intake is not a precise measure of n-3 polyunsaturated fatty acid intake. Further, we did not have a quantitative measure of dietary n-6 polyunsaturated fatty acids nor of fish oil or other fatty acid-containing supplements. This limited our power to detect true interactions between dietary fatty acids and genetic variability related to PGE₂ signaling on risk of colorectal adenoma.

This study has several strengths. First, all controls who participated in this study underwent colonoscopy before enrollment. Because colorectal polyps are common in older adults, an unscreened population-based control group would include undiagnosed adenoma cases, resulting in disease misclassification and an attenuation of any true associations. In addition, by using tagSNPs as well as rare nonsynonymous SNPs to investigate associations between these target genes and the risk of colorectal adenoma, we achieved comprehensive coverage of these genes, increasing the likelihood of observing any existing true association. The prostaglandin synthesis pathway is an important one for colorectal carcinogenesis and our complete coverage of the genes and their 5′ and 3′ regions, as well as some life-style factors that are likely to influence prostaglandin synthesis, provides a more complete investigation of this pathway.

In summary, the results from the present study provide some limited evidence that genetic variability in genes related to PGE₂ levels and signaling may be modifiers of the relationship between NSAID use or fish intake and risk of colorectal adenoma.

No potential conflicts of interest were disclosed.

We thank Dr. Roberd Bostick and Lisa Fosdick for their contributions to the initial establishment of this study.

Grant Support: Grants R01 CA114467, R01 CA112516, and R03 CA123577.

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.