Prostaglandin H Synthase 2 Variant (Val511Ala) in African Americans May Reduce the Risk for Colorectal Neoplasia¹

PTGS³ enzymes convert arachidonic acid to prostaglandin H₂, a precursor to all of the other prostanoids. Two forms of human PTGS are known, PTGS1 and PTGS2. They are similar in crystal structure and inhibited by NSAIDs (1, 2), but the enzymes are encoded by different genes (3, 4). PTGS1 was first prepared from sheep and bull seminal vesicles (5, 6, 7), and may be a housekeeping enzyme involved in cell signaling. PTGS2, on the other hand, was discovered in chicken embryo fibroblasts induced by Rous sarcoma virus (8), in Swiss 3T3 mouse cells treated with tetradecanoylphorbol acetate (9), and in C127 mouse fibroblasts stimulated by serum or glucocorticoids (10). Both PTGS1 and PTGS2 localize to the nuclear envelope and endoplasmic reticulum of cells (11), but PTGS2 is absent from many cell types unless induced by tumor promoters, growth factors, or cytokines.

PTGS2 occurs at high levels in colon cancer (12, 13, 14, 15, 16). Support for the hypothesis that PTGS2 is involved in development of colon cancer comes from evidence that fewer polyps occurred after disruption of the PTGS2 gene in mice prone to intestinal polyposis (17). Also, celecoxib, a selective PTGS2 inhibitor, blocked colon tumors in mice (18, 19) and in patients with familial adenomatous polyposis (20, 21). In a case-control study of colon cancer, Kune et al. (22) unexpectedly found a 0.53 relative risk for cancer with regular aspirin use among 715 cases and 727 controls. Some 15 other studies have found similar results (reviewed in Ref. 23).

We hypothesized that naturally occurring PTGS2 variants might mimic long-term NSAID use (24) and give additional insight into biochemical mechanisms of colon cancer prevention. Here we describe: (a) identification of a PTGS2 polymorphism among African-Americans; (b) cell transfection experiments to check function of the polymorphism; (c) modeling of the crystal structure of the variant enzyme; and (d) epidemiological studies of the polymorphism in relation to colorectal adenomas and cancer.

We used blood specimens from the University of California Los Angeles Tissue Typing Laboratory to search for common PTGS2 variants. Specimens were from blood drives aimed at finding bone marrow donors from specific ethnic groups. Random batches of specimens in acid-citrate-dextrose tubes without identifiers were obtained between February 1992 and November 1993. Most blood specimens were from the Los Angeles area, but the laboratory also received specimens from around the world. Volunteers were generally unrelated, between the ages of 18 and 65, and without a history of chronic disease, including cancer. There was no gender preference and no duplicate sampling. Ethnic groups surveyed were: African-American, Chinese (Hong Kong), Filipino, Hispanic, Indian (Asian), Japanese, Korean, Samoan, and Caucasian.

We used 1 × 1 mm squares of dried blood on blotter paper (No. 903 paper; Schleicher and Schuell, Keene, NH) for PCR templates. Each PCR contained 40 μm of each dNTP, 1 μm of each primer, 20 μg ml⁻¹ BSA, buffers recommended by Taq polymerase suppliers, and a total volume of 10–40 μl. MgCl₂ was optimized for each pair of primers. Heating was done in a Perkin-Elmer 9600, Applied Biosystems 9700, or MJ Research PTC-100 thermal cycler (95°C for 15 min; then the temperature was lowered to 85°C for 10 min while 0.25 units of Taq polymerase was added; then 26–32 cycles of 94°C for 30 s, the annealing temperature for 40 s, and 72°C for 60 s; then 72°C for 5 min). PCR primers were selected by use of a computer program (Table 1), with theoretical annealing temperatures matched to within 3–5°C.

We used DNA heteroduplex analysis to screen for variants in exons and flanking intron regions in 47 African-American and 47 Caucasian subjects (25, 26). PCR conditions were as above, with total volumes of 10 μl and 0.2 μCi of [α³²P]dCTP or 1 μCi of [α³³P]dATP (3000 Ci mmol⁻¹; ICN Pharmaceuticals, Irvine, CA). Primer pairs, annealing temperatures, and sizes of PCR products were: EX1L1 and EX1R1 (54°C; 329 bp); EX23L2 and EX23R2 (50°C; 627 bp); EX4L3 and EX4R3 (50–51°C; 364 bp); EX5L4 and EX5R4 (54°C; 319 bp); EX67L5 and EX67R5 (50°C; 659 bp); EX8L6 and EX8R6 (48°C; 413 bp); EX9L7 and EX9R7 (50°C; 296 bp); and EX10L and UTR-R (48°C; 603 bp). After amplification, products were heated to 98°C for 5 min, incubated at 68°C for 1 h to form heteroduplexes, and electrophoresed on 10% polyacrylamide gels to detect variants (27).

Variants were identified by Sanger dideoxy DNA sequencing (United States Biochemical, Cleveland, OH) through either direct sequencing of PCR products or sequencing of PCR products that were cloned into plasmids. For direct sequencing, PCR products were treated with exonuclease I and shrimp alkaline phosphatase (United States Biochemical) or purified from agarose gels (GENECLEAN; Bio101, Vista, CA). For plasmid sequencing, PCR products were cloned into a plasmid vector (TA cloning vector; Invitrogen Corp., San Diego, CA). Forty to 60 clones containing inserts were grown individually, pooled, and used for preparing plasmid DNA.

We also sequenced an AU-rich region 3′ of the PTGS2 stop codon in six subjects with the V511A polymorphism to look for DNA variation that might be associated with V511A. The AU-rich region is thought to play a role in degradation of mRNA (see “Results”). Specifically, we amplified exon 10 and the adjoining 3′ untranslated region (primers EX10L and UTR-R; 48°C; 603 bp) and used UTR-L and UTR-R as sequencing primers.

A plasmid expression vector containing the human PTGS2 coding sequence was provided by W. Smith and D. DeWitt. We subcloned into plasmid pUC19 a SalI restriction fragment containing wild-type coding and flanking regions (1.9-kb total). The 5′ flanking region was: GTC GAC(SalI site)-CGA ATT C(EcoRI/ApoI site)-GCG GCC GCG TGA GAA CCG TTT ACC(24 bp junction piece)-ATG(start of PTGS2 coding sequence). The 3′ flanking region was: TAG(stop codon)-AAG TCT AAT GAT CA(PTGS2 flanking sequence)-ACC CTT CTC ACC TCG GCC GAT AAG CTC TAG AGC GGT CGA C(40 bp junction piece)-GTC GAC(SalI site). We prepared a coding unit that produced the same amino acid sequence as described by Kosaka et al. (4).

We introduced the V511A variant into the coding sequence by site-directed mutagenesis [Quantum Biotechnologies, Inc., Laval, Quebec, Canada; mutagenic oligonucleotide: 5′-pGA AAC CAT GGT AGA AGC TGG AGC ACC ATT CTC-3′ (PHS2–1629); closing oligonucleotide: PUC-1222B, 5′-pCC ACT GGC AGC AGC CAC TGG TAA CAG GAT TA-3′]. Presence of the variant and absence of PCR artifacts were checked by sequencing. Other PTGS2 variants in our survey were not built into expression vectors, because they did not change the amino acid sequence.

Expression vectors were prepared by subcloning the SalI restriction fragment above into vector pOSML-1 (28, 29), choosing recombinants with the desired orientation, double banding on cesium chloride gradients, and sequencing again to verify presence or absence of the variant.

COS-1 cells were grown until near confluency in DMEM with 8% calf serum and 2% FCS. The cultures were split 1:2 roughly 16 h before transfections. Cells were transfected by use of DEAE-dextran and chloroquine (30) with either an expression vector carrying wild-type PTGS2, the V511A variant, or an “empty vector” containing no PTGS2 coding sequence. After 48 h, transfected cells were removed with a rubber policeman, collected by centrifugation, resuspended, and lysed by sonication. Microsomal membranes were prepared as described, and cyclooxygenase activity was measured by monitoring the initial rate of O₂ uptake by use of an O₂ electrode (28, 29). Each assay reaction contained 250–500 μg of microsomal protein, 1 mm phenol, 85 μg of hemoglobin, 0.1 m Tris-Cl (pH 8.0), and 1–100 μm arachidonic acid in a volume of 50 μl. To assess thermal stability of PTGS2, catalytic activities of variant and wild-type proteins in microsomal membrane preparations were measured after incubation at 37°C for 0, 2, 4, 6, 8, 12, and 24 h.

The crystal structures of ovine PTGS1 (3.1 Å resolution and Rfactor = 18.5%; PDB entry 1CQE) and murine PTGS2 (3.1 Å resolution and Rfactor = 28%; PDB entry 3PGH), both complexed with the NSAID flurbiprofen, were superimposed by least squares superposition of equivalent Cα atoms (root square mean = 0.51Å) and were used to generate images of enzyme structure. For clarity in comparing crystallographic structures of PTGS isozymes, we use a consensus numbering system that starts with methionine 1 of the ovine PTGS1 enzyme. The homologous amino acids corresponding to ovine PTGS1 and human PTGS2 that are mentioned in the text are (PTGS1:PTGS2): Met525:Val511; Leu384:Leu370; and Tyr385:Tyr371. Thus, position 511 in PTGS2 is called position 525 whenever crystal structures are discussed below.

As the coordinates of human PTGS2 were not yet publicly available, the L525V (leucine to valine) and L525A (leucine to alanine) substitutions were modeled in the mouse PTGS2 structure, using Setor 4.14.7 (31), to evaluate the V525A substitution in human PTGS2. This procedure was straightforward as the sequences of PTGS2 from humans and mice are highly homologous (>95% identity) in this region, and because human PTGS2, mouse PTGS2, and ovine PTGS1 are all structurally homologous to each other (32, 33). Cavity volume measurements were made using the program Grasp (34).

Subjects were from two study populations. Blood specimens from 240 African-Americans (119 cases and 121 controls) came from the USC/Kaiser study, which consisted of roughly 1700 subjects enrolled after examination by flexible sigmoidoscopy at either of two Southern California Kaiser Permanente Medical Centers (Bellflower and Sunset; done during the period from January 1991 through August 1993 or from October 1995 through March 1999). Eligible subjects were 50–74 years old, spoke English, gave informed consent, and lived in the metropolitan Los Angeles area. There was no history of invasive cancer, inflammatory bowel disease, familial polyposis, previous bowel surgery, symptoms suggestive of gastrointestinal disease, or disability precluding an interview. Cases had a first-time diagnosis of an adenoma, confirmed by histology. Controls had no current or past polyp. Controls were randomly selected and individually matched to cases by gender, age, sigmoidoscopy date, and center. Indications for sigmoidoscopy were “routine” for 44% of the control subjects and 45% of the cases, “specific minor symptoms” for 13% of the control subjects and 16% of the cases, and were not given for 43% of the control subjects and 39% of the cases. The average depth of the sigmoidoscope was 59 cm for control subjects and 55 cm for cases. Fifteen case subjects also had carcinoma in situ. During a 45-min in-person interview, subjects provided information on smoking, therapeutic drug use, physical activity, height, weight, family history of cancer, and other factors. The interviewer was unaware of case or control status for 70% of case subjects and 87% of controls. Overall participation rates in the study were 84% for cases and 82% for controls (number interviewed/number eligible; Ref. 35).

Blood specimens from 140 African-Americans (42 cases and 98 controls) came from the UNC study, which consisted of roughly 800 subjects examined by colonoscopy at the UNC Hospitals during the period from August 1998 through March 2000. Eligible subjects were ≥30 years of age and gave informed consent. They agreed to have biopsy specimens taken during the clinically indicated colonoscopy, a blood sample drawn, and a telephone interview asking about lifestyle and diet (by use of a quantitative food frequency questionnaire developed at the National Cancer Institute, Bethesda, MD). Subjects were excluded if there was a previous adenoma, a history of cancer, colitis of any type (e.g., radiation, infectious, and idiopathic), polyposis (defined as ≥100 polyps), a previous colon resection, unsatisfactory colon preparation as judged by the endoscopist, or an incomplete examination (cecum not reached). Cases had a first-time diagnosis of an adenoma, confirmed by histology. Controls had no adenoma and no disease of the colon identified. The most common indication for the colonoscopy was bleeding (i.e., anemia or a positive fecal occult blood test), and the proportion of individuals with each of the indications was the same among cases and controls. The interviewers were not aware of case-control status, and the overall response rate was 83% (number biopsied/number eligible).

Dried blood spots on blotter paper were prepared from blood samples that had been collected by the principal investigators of these studies (R. W. H. and R. S. S.). The blotter papers were labeled with a code number, but no other identifier. One-mm squares of dried blood were tested in duplicate for the V511A polymorphism by use of allele-specific PCR, without knowledge of case or control status (see Fig. 1,d). To double-check for homozygous V511A, samples found to carry the polymorphism were amplified again and tested for the polymorphism by use of restriction enzyme AluI (site present in variant allele; see Fig. 1 c).

DNA specimens came from 396 African-Americans (138 cases and 258 controls) from the Multiethnic Cohort Study, a cohort assembled to investigate diet and cancer in Hawaii and Los Angeles (36). African-American participants (16.3% of a total cohort of 215,251) were recruited from mainly Los Angeles County between 1993 and 1996. Subjects were 45–75 years old and completed a baseline questionnaire on diet, smoking, sun exposure, physical activity, prior medical conditions, medications, reproductive history, family history of cancer, and demographic factors. The medications section asked about NSAIDs and NSAID-containing drugs, including duration of use. All of the incident colorectal cancer cases occurring in the cohort were identified through the rapid reporting systems of the Los Angeles County Cancer Surveillance Program and the Hawaii Tumor Registry, two population-based cancer registries that are members of the Surveillance, Epidemiology, and End Results Program of the National Cancer Institute. Control subjects were a random sample of the entire cohort. Cases and controls were recontacted to obtain a blood sample. Among African-Americans, 69% of eligible cases and 70% of eligible controls donated a blood sample.

Genotyping and retesting for homozygous V511A were done as described above, except DNA extracted from lymphocytes was used for PCR (as little as 10 ng/PCR). DNA samples were labeled with a study number but no other identifier.

ORs and 95% CIs for colorectal adenomas or cancer were calculated by use of unconditional logistic regression. Use of unconditional logistic regression on the USC/Kaiser study allowed analyses to include genotype information on unmatched subjects. Unmatched controls occurred when, for example, their matched cases did not speak English or were found to have invasive cancer at the follow-up colonoscopy. Unmatched cases occurred when we were unable to interview an eligible control. ORs were adjusted for age (within 5-year intervals) and gender. Conditional and unconditional analyses of the reduced data set gave similar results. Results were also stratified by NSAID use, because such drugs inhibit PTGS2 and prevent some cases of colorectal cancer. Adjustment for factors besides age and gender was not possible, because other variables were not available in all three of the data sets. Conceivably, other factors may be confounders in such analyses, but there have been no reports of strong confounders of NSAID effects.

Variants identified by heteroduplex analysis and DNA sequencing are shown in Table 2. Only the V511A polymorphism changed the amino acid sequence (Fig. 1). The V511A heterozygote frequency among African-American blood specimens from the University of California Los Angeles Tissue Typing Laboratory was 8 of 107 (allele frequency, 0.04). The result was consistent with the frequency in our epidemiological studies (0.045; see below). Among Indians, the V511A heterozygote frequency was 5 of 75 (allele frequency, 0.03). V511A was not detected among Chinese, Hispanic, or Caucasian individuals screened by heteroduplex analysis, or among Filipino, Japanese, Korean, or Samoan individuals screened by allele-specific PCR.

Our screening for variants in exon 1 also covered 193-bp directly 5′ of the ATG start codon, which contains a cyclic AMP response element (TTCGTCA; −193 to −187 relative to the ATG codon), the TATA box (TATAAAA; −165 to −159), and the mRNA cap site (at −134; Ref. 4). No variation in this promoter region was detected.

At the other end of the gene, the 2.9-kb region 3′ of the stop codon is AU-rich and contains at least 21 copies of an AUUUA element (4). AUUUA elements are believed to enhance degradation of mRNA (37). The 116-bp, AU-rich region directly after the PTGS2 stop codon contains six AUUUA elements and has been shown to inhibit protein translation (38). We sequenced the 116-bp region in six V511A carriers to look for DNA variants that might be associated with V511A. No variation was found (specimen numbers 3014, 3040, 3049, 3061, 3062, and 3065).

The coding sequence with V511A was built into a plasmid vector by use of site-directed mutagenesis and expressed in COS-1 monkey kidney cells. Wild-type and variant PTGS2 were similar in V_max, K_m, and thermal stability at 37°C (Table 3).

In the native PTGS2 enzyme, residue 511 (referred to below as residue 525, in accordance with the consensus PTGS1/PTGS2 numbering used for crystal structure comparisons; see “Materials and Methods”), lies inside a tightly packed hydrophobic pocket adjacent to the cyclooxygenase active site (Fig. 2,A; Refs. 32, 33, 39). Val 525 in human PTGS2 abuts against Leu 384, which forms part of the active site wall and is a close neighbor of the radical donor/acceptor, Tyr 385. The residues that form the pocket are highly conserved among different species and PTGS isozymes: mouse PTGS2 has a leucine at position 525 instead of a valine, and ovine PTGS1 contains Met 525 and two other conservative substitutions. These differences cause little or no change in the packing of the pocket (Fig. 2, B–D). However, the replacement of Val 525 with Ala is predicted to open a 53 ų (moderate-sized) cavity within the pocket and a smaller, adjacent 20 ų hole (Fig. 2 E).

The presence of a large cavity next to Leu 384 will alter its conformation and could directly perturb the structure of the active site. For example, movement of Leu 384 may allow the Tyr 385 catalytic residue to assume different conformations. Movement of Leu 384 into the cavity may also increase the space within the active site and allow nonproductive conformations of arachidonate (40, 41). Finally, the segment 384–388 forms an unusually stretched helix that contains the two key catalytic residues, Tyr 385 and His 388 (the proximal heme ligand). Repacking around Leu 384 could relax the stretched helix, allowing Tyr 385 to move away from its ideal position. These alterations would be expected to affect cyclooxygenase function.

The V511A heterozygote frequency among African-American controls from the USC/ Kaiser (12 of 121), UNC (10 of 98), and Multiethnic Cohort (21 of 258) studies was 43 of 477, corresponding to an allele frequency of 0.045 (95% CI, 0.033–0.060). No homozygote was found among all 776 of the cases and controls (1 in 492 expected with an allele frequency of 0.045).

In the USC/Kaiser and UNC studies, the OR for colorectal adenomas with the V511A polymorphism was in the direction of a protective effect but not statistically significant, possibly because of small sample sizes (OR, 0.56; 95% CI, 0.25–1.27; Table 4B). Among African-Americans in the Multiethnic Cohort Study, the OR for colorectal cancer with the V511A polymorphism was 0.67 (95% CI, 0.28–1.56; Table 5B). Thus, there was a suggestion of a protective effect for V511A in all three of the case-control populations.

We analyzed V511A separate from NSAID use, because NSAIDs alone may prevent some cases of colorectal neoplasia. Data on NSAID use were available for 180 subjects (75%) in the USC/Kaiser study and 111 subjects (79%) in the UNC study. NSAID users were defined as subjects who took such medication at least once per week for some period of time during the 1-year or the 5-year period before the interview, for the USC/Kaiser or UNC studies, respectively. Excluding NSAID users, the OR for adenomas with V511A was 0.29 (95% CI, 0.08–1.08; Table 4B).

Data on NSAID use were available for 368 subjects (93%) from the Multiethnic Cohort Study. NSAID users were defined as subjects who ever took such medications at least twice per week for at least 2 years. Excluding NSAID users, the OR for cancer among subjects with V511A was 1.19 (95% CI, 0.39–3.61; Table 5B).

We also analyzed NSAIDs separate from V511A, to check effects of these drugs alone. Comparing NSAID users to nonusers in the combined USC/Kaiser and UNC sample, the OR for adenomas when V511A carriers were excluded was 0.56 (95% CI, 0.32–0.96; P = 0.035; Table 4C). The result is compatible with the recognized protective effect of NSAIDs.

Among NSAID users in the Multiethnic Cohort Study, the OR for colorectal cancer was 0.82 when V511A carriers were excluded (95% CI, 0.50–1.33; Table 5C). Again, none of these ORs was statistically significant.

Finally, we compared subjects who were V511A carriers and/or NSAID users to noncarriers who were nonusers. The analysis may represent low versus high PTGS2 activity. For adenomas, ORs for this comparison were 0.52 in both the USC/Kaiser study (95% CI, 0.28–0.95; P = 0.034) and the combined adenoma study (95% CI, 0.31–0.86; P = 0.011; Table 4D). For colorectal cancer, comparison of V511A carriers and/or NSAID users to noncarriers who were nonusers gave an OR of 0.78 (95% CI, 0.49–1.23; Table 5D).

We found seven variants in PTGS2 coding or flanking intron regions. These DNA variants are among the 46 variants seen previously in or around the PTGS2 gene.⁴ Reported variants that change the coding sequence include I1M (Ile1Met), R228H (Arg228His), E488G (Glu488Gly), and V511A. Of these, only V511A was observed in our survey, possibly because heteroduplex analysis may be only 90% sensitive (25).

Our allele frequency (0.045; 95% CI, 0.033–0.060) for V511A in African-Americans was close to a reported frequency of 0.083 (95% CI, 0.023–0.20; Ref. 42). The polymorphism has been reported in Asians at an allele frequency of 0.06 (42). However, we found V511A only in Indians, among the six Asian populations screened.

We looked for association of V511A with variation in PTGS2 gene control regions. No variant was found in a 59-bp PTGS2 promoter region immediately 5′ of the mRNA cap site, although a minimal PTGS2 promoter may include 100-bp or more (43). Also, the 116-bp, AU-rich region directly after the PTGS2 stop codon has been shown to inhibit protein translation (38). We did not find any variation in this region in 6 (of 6) subjects with V511A. Thus, there is no sign of association between V511A and variation in a gene control region in our study.

Work on PTGS2 variants focused on V511A, because the polymorphism changes an amino acid that interacts with residues at the active site of the enzyme. Despite the potential for causing significant structural changes in the active site, the polymorphism did not produce detectable differences in enzyme kinetic parameters (V_max and K_m) or stability for the utilization of arachidonate, when the enzyme was expressed in COS-1 cells (Table 3). Observed K_m values were comparable with the published K_m for human PTGS2 (5.6 μm; Ref. 44). Recently, Fritsche et al. (42) also reported that no functional differences were observed between the V511A variant and wild-type PTGS2. They tested three PTGS2 substrates (arachidonic acid, 2-arachidonyl glycerol, and linoleic acid) and measured effects of four different PTGS2 inhibitors.

Such lack of functional changes may reflect assay conditions in vitro, where arachidonate concentrations were 1–100 μm. However, in the cell, PTGS2 is believed to use arachidonate released from the nuclear envelope by cytosolic phospholipase A₂ (45, 46). Physiological arachidonate levels may be fairly low and may be associated with different enzyme properties. For example, Swinney et al. (47) found that PTGS2 binds an inhibitor (SC-58125) 26-fold better at 50 nm arachidonate than at 20 μm arachidonate. An allosteric conformation transition at low arachidonate levels was proposed as an explanation.

Moreover, the in vitro assays monitored PTGS2 activity by following oxygen consumption, a method that does not detect altered products (48). PTGS enzymes oxygenate several different polyunsaturated fatty acids other than arachidonate (49, 50) to form bioactive products. Subtle changes in enzyme structure, because of amino acid substitution, can affect the turnover of alternate substrates, whereas only marginally affecting arachidonate utilization (49, 50). Hence, it may be premature to conclude that there is no functional difference between the V511A variant and wild-type PTGS2, until a more thorough enzymological analysis is done.

Case-control analysis of African-Americans in two studies showed a ∼0.5 OR for colorectal adenomas among V511A carriers (Table 4B). Similarly, a third study suggested protection against colorectal cancer (OR ∼0.7; Table 5B). The consistency among the three studies tends to support potential validity of a protective effect. Larger sample sizes are needed to confirm our interpretation.

To additionally analyze our small numbers of V511A carriers, we combined V511A carriers with NSAID users and compared them with noncarriers who were nonusers. NSAID users are presumed to have low PTGS2 activity because of inhibition of the enzyme by such drugs. Therefore, the comparison corresponds to low versus high PTGS2 activity, if V511A lowers enzyme activity in vivo. Results were compatible with a protective effect against colorectal adenomas for V511A and/or NSAID use (OR, 0.52; 95% CI, 0.31–0.86; P = 0.011; Table 4D). Additional studies are needed to confirm the result.

Possible effects of V511A, if any, were smaller in the Multiethnic Cohort Study than in the adenoma studies. For example, the OR for colorectal cancer was 1.19 when NSAID users were excluded. However, results are not inconsistent, because the CIs were wide and overlapped with those for the adenoma studies. These apparent differences may reflect differences between cancer and adenomas or different definitions of NSAID use. Alternatively, the results may indicate lack of a V511A effect.

Wiesner et al. (51) tested the PTGS2 locus for linkage to colon neoplasia by use of sib pairs. No evidence for linkage was found among 46 families, but 44 of the families were Caucasian. Absence of functional PTGS2 variants among Caucasians in our survey is compatible with lack of linkage in the sib pair study.

Mechanisms by which PTGS2 and NSAIDs may affect colon cancer are being studied. Prescott and White (52) concluded that PTGS2 overexpression occurs after loss of both APC alleles and after formation of the early polyp in Min mice. They suggested that PTGS2 promotes tumors through prostaglandin receptor signaling. The prostaglandin receptor EP2 appears to be one of the mediators (53). Peroxisome proliferator-activated receptor γ (a nuclear receptor and tumor suppressor) may also be involved (54, 55, 56, 57, 58). Lower numbers of intestinal adenomas in Apc^Min/+, cPLA₂^−/− mice, which lack the enzyme for the first step in prostanoid synthesis, also support a prostaglandin mechanism for tumorigenesis (59). Angiogenic factors (60) or c-MYB (61) may also play a role.

Each of the three studies used here to analyze the V511A polymorphism have specific strengths. For example, advantages of the sigmoidoscopic case-control study were screening asymptomatic individuals for a first-time diagnosis of adenomas and participation of >80% of cases and control individuals. Awareness by interviewers of case (30%) or control (13%) status was not expected to introduce bias, because interviewers were well trained, and NSAIDs were not a major topic of the interview. A weakness was that adenomas proximal to the left colon cannot be seen with sigmoidoscopy. Roughly 50% of colon cancers occur in the proximal segment (62). Studies show that up to 15–17% of control individuals who have no polyp within view of the sigmoidoscope may have a polyp in the right colon (63). However, our analyses included 140 patients who underwent full colonoscopy, in addition to the 240 subjects screened by sigmoidoscopy. Findings in the colonoscopy group were similar to those in the sigmoidoscopy group (Table 4). Furthermore, misclassification of controls because of incomplete colon screening by sigmoidoscopy would bias the study toward the null. Our results would actually underestimate the true effect of the PTGS2 variant under these circumstances. Finally, the 396 colorectal cancer cases and controls from the Multiethnic Cohort Study should not be subject to selection biases that may be associated with colonoscopy referrals.

In summary, in three large, ethnically diverse studies, we found the suggestion of an inverse association between a variant form of PTGS2 and colorectal neoplasia. Larger sample sizes are needed to assess our initial report of possible protection against colon neoplasia by the V511A polymorphism. The results are consistent with epidemiological studies and may suggest the importance of PTGS2 inhibitors in reducing the risk of colorectal cancer.