Introduction
In recent years, several relatively large studies have examined the association between nonsteroidal anti-inflammatory drugs (NSAIDs) and breast cancer. A meta-analysis of 14 studies (six prospective studies and eight case-control studies) [
1] demonstrated a decrease of 20% in the risk for breast cancer among NSAID users. Harris and coworkers [
2] observed an inverse relationship between incidence of breast cancer and intake of NSAIDs in a prospective cohort study, with an approximately 40% decreased risk for breast cancer in regular aspirin users. More recent studies [
3,
4], including our previous analyses from the Long Island Breast Cancer Study Project [
5], demonstrated that use of aspirin (the most commonly used NSAID) is associated with a significant reduction in risk for breast cancer, especially for hormone receptor positive tumors. However, data from several recent cohort studies [
6,
7] have been less consistent.
The anticancer activity of NSAIDs is believed to be due to inhibition of cyclooxygenase (COX)-2, which is over-expressed in many types of cancer, including breast cancer, and plays a major role in tumorigenesis [
8,
9].
COX-2 encodes one of two COX enzymes that catalyze the synthesis of prostaglandins (PGs) from the dietary fatty acid arachidonic acid [
10]. Increased COX-2 levels are associated with increased angiogenesis, increased estrogen synthesis, and reduced apoptosis, all of which may stimulate tumor growth [
7]. In breast cancer PGs, particularly PGE
2, have an organ-site-specific effect of increasing the levels of aromatase, thereby increasing estrogen and progesterone synthesis [
10‐
12]. COX-2 expression has been reported in a significant proportion of preinvasive and invasive breast cancers [
10,
13], and different frequencies of COX-2 over-expression have been observed in subgroups of breast cancer patients by hormone receptor status (estrogen receptor [ER] and progesterone receptor [PR]). More frequent elevation in COX-2 expression was noted in ER/PR-negative breast cancer [
14]. COX-2 expression also had a different impact on prognosis in subgroups of tumors. Elevated expression of COX-2 was associated with poor survival in ER-positive or PR-positive tumors (
P < 0.002) but not in ER/PR-negative ones [
14]. Therefore, it is biologically plausible that the protective effect of NSAIDs comes, at least in part, from limiting either the amount or activity of COX-2 present in the cell, both of which are partly determined by specific polymorphisms [
15‐
18].
The previous inconsistent findings on the association between NSAIDs and breast cancer risk might be explained by interindividual differences in
COX-2 gene expression or be limited to certain subtypes of breast cancer. Thus, we hypothesized that potential functional genetic polymorphisms (variant alleles or haplotypes/diplotypes) in
COX-2 that result in altered expression and/or activity of the protein may modulate the inflammatory response, modifying overall breast cancer risk or risk for subtypes of breast cancer. Single nucleotide polymorphisms (SNPs) in the promoter and 3'-untranslated region (UTR) of
COX-2 were previously reported to modulate the risks for prostate, colorectal, esophageal, gastric, bladder, biliary tract, and non-small-cell lung cancer [
17,
19‐
31]. A few previous reports showed that the
COX-2 .926 G→C polymorphism examined here can result in higher COX-2 expression and lead to an approximate 30% reduction in promoter activity
in vitro [
15,
18]. No functional studies of other
COX-2 polymorphisms potentially associated with cancer risks have been reported. Little is known regarding
COX-2 polymorphism and breast cancer risk and its potential interactions with aspirin and NSAID use [
32‐
35].
Results
The distributions of COX-2 .5209 and COX-2 .8473 genotypes were compatible with those expected from Hardy-Weinberg equilibrium, and the frequencies of the variant G and C alleles were respectively 20.0% and 34.1%. The genotype distribution for COX-2 .926 significantly deviated from Hardy-Weinberg equilibrium in both controls and combined cases and controls (both P values below 0.003).
No main effects on reduced breast cancer risk were found for the three
COX-2 SNPs, even when women were categorized by menopausal status (Table
1). The frequency of carrying at least one variant allele (
C allele in
COX-2 .926,
G allele in
COX-2 .5209, or
C allele in
COX-2 .8473) did not differ significantly from the controls, even when breast cancer groups were stratified based on hormone receptor status (data not shown).
Table 1
COX-2 genotypes and risk for breast cancer by menopausal status: Long Island Breast Cancer Study Project, 1996 to 1997
All women | | | | |
COX-2 .926 (rs20417) | | | | |
GG | 670 | 691 | 1.0 (ref.) | 1.0 (ref.) |
GC + CC | 387 | 414 | 1.0 (0.8–1.1) | 1.0 (0.8–1.2) |
COX-2 .5209 (rs20432) | | | | |
TT | 685 | 694 | 1.0 (ref.) | 1.0 (ref.) |
TG + GG | 372 | 399 | 0.9 (0.8–1.1) | 0.9 (0.8–1.1) |
COX-2 .8473 (rs5275) | | | | |
TT | 475 | 467 | 1.0 (ref.) | 1.0 (ref.) |
TC + CC | 585 | 635 | 0.9 (0.8–1.1) | 0.9 (0.8–1.1) |
Premenopausal women | | | | |
COX-2 .926 (rs20417) | | | | |
GG | 217 | 234 | 1.0 (ref.) | 1.0 (ref.) |
GC + CC | 124 | 142 | 0.9 (0.7–1.3) | 1.0 (0.7–1.4) |
COX-2 .5209 (rs20432) | | | | |
TT | 225 | 229 | 1.0 (ref.) | 1.0 (ref.) |
TG + GG | 116 | 142 | 0.8 (0.6–1.1) | 0.9 (0.6–1.2) |
COX-2 .8473 (rs5275) | | | | |
TT | 148 | 153 | 1.0 (ref.) | 1.0 (ref.) |
TC + CC | 193 | 223 | 0.9 (0.7–1.2) | 0.9 (0.7–1.2) |
Postmenopausal women | | | | |
COX-2 .926 (rs20417) | | | | |
GG | 438 | 428 | 1.0 (ref.) | 1.0 (ref.) |
GC + CC | 253 | 255 | 1.0 (0.8–1.2) | 1.0 (0.8–1.2) |
COX-2 .5209 (rs20432) | | | | |
TT | 445 | 438 | 1.0 (ref.) | 1.0 (ref.) |
TG + GG | 246 | 238 | 1.0 (0.8–1.3) | 1.0 (0.8–1.3) |
COX-2 .8473 (rs5275) | | | | |
TT | 318 | 295 | 1.0 (ref.) | 1.0 (ref.) |
TC + CC | 376 | 385 | 0.9 (0.7–1.1) | 0.9 (0.7–1.1) |
Haplotypes and diplotypes for the three
COX-2 SNPs reconstructed using PHASE software [
46] resulted in eight distinct haplotypes and 18 diplotypes. Five haplotypes and eight diplotypes were sufficiently frequent to calculate ORs (Table
2). The three common haplotypes (GTT, CGC, and GTC) accounted for more than 94.7% of the alleles. None of the haplotypes exhibited an association with breast cancer risk in this population. The three most frequent diplotypes were GTT/GTT, GTT/CGC, and GTT/GTC. Compared with the most prevalent reference diplotype (GTT/GTT), an uncommon diplotype (GTT/GGC) was associated with a 50% decreased risk for breast cancer (adjusted OR = 0.5, 95% CI = 0.2 to 0.9). The OR remained unchanged when the diplotype was compared with all other diplotype combinations. However, this result is based on very small numbers of individuals. No strong associations were observed between other diplotypes and breast cancer risk.
Table 2
COX-2 haplotype and risk for breast cancer: Long Island Breast Cancer Study Project, 1996 to 1997
Haplotypeb
| | | | |
GTT | 1,359 | 1,405 | 1.0 (ref.) | |
CGC | 363 | 388 | 1.0 (0.8–1.2) | 0.85 |
GTC | 299 | 306 | 1.0 (0.8–1.2) | 0.95 |
GGC | 38 | 60 | 0.7 (0.5–1.0) | 0.06 |
CTT | 27 | 33 | 0.8 (0.5–1.4) | 0.48 |
Diplotype (haplotype pairs) | | | | |
GTT GTT | 434 | 432 | 1.0 (ref.) | |
GTT CGC | 239 | 262 | 0.9 (0.7–1.1) | 0.45 |
GTT GTC | 186 | 194 | 1.0 (0.7–1.2) | 0.69 |
GTC CGC | 49 | 51 | 0.9 (0.6–1.4) | 0.79 |
GTC GTC | 26 | 25 | 1.0 (0.6–1.8) | 0.94 |
GTT GGC | 12 | 27 | 0.5 (0.2–0.9)c
| 0.03 |
GTT CTT | 19 | 24 | 0.7 (0.4–1.4) | 0.36 |
CGC CGC | 25 | 23 | 1.1 (0.6–2.1) | 0.68 |
Others | 47 | 36 | 1.3 (0.8–2.0) | 0.31 |
An inverse association between ever use of aspirin and breast cancer risk was previously reported in this population [
5], and a similar result was observed when the analyses were restricted to women who donated blood (OR = 0.80, 95% CI = 0.65 to 0.98). When we evaluated the associations between ever use of aspirin and breast cancer risk stratified by
COX-2 genotype, the decreased OR observed among ever users of aspirin who carried at least one variant C allele of
COX-2 .8473 (OR = 0.7, 95% CI = 0.5 to 0.9) compared with nonusers carrying the wild-type TT genotype (Table
3) was similar to the overall OR obtained from ever users of aspirin. No significant interaction was observed (
P for interaction = 0.77). When restricted to hormone receptor positive cases, the reduction in breast cancer risk was slightly stronger for carriers of at least one variant C allele of
COX-2 .8473 who ever used aspirin (OR = 0.6, 95% CI = 0.4 to 0.9; Table
4), although the modest heterogeneity in the ORs was not statistically significant. However, there was significant heterogeneity in the ORs for the interaction between ever NSAID use (including both aspirin and nonaspirin NSAID use) and carrying variant C allele of
COX-2 .8473 (OR = 0.7, 95% CI = 0.5 to 1.0;
P for interaction = 0.02). No significant interaction between ever use of aspirin, ever use of NSAIDs, and variant C allele of
COX-2 .8473 was observed in ER/PR-negative breast cancer (Table
4).
Table 3
COX-2 genotypes, NSAIDs use, and risk for breast cancer: Long Island Breast Cancer Study Project, 1996 to 1997
Aspirin |
COX-2 .926 (rs20417) |
GG | 511/484 | 1.0 (ref.) | 138/156 | 0.8 (0.6–1.0) |
GC + CC | 285/291 | 0.9 (0.8–1.1) | 84/93 | 0.8 (0.6–1.1) |
COX-2 .5209 (rs20432) |
TT | 521/485 | 1.0 (ref.) | 145/160 | 0.8 (0.6–1.0) |
TG + GG | 275/282 | 0.9 (0.7–1.1) | 78/86 | 0.8 (0.6–1.1) |
COX-2 .8473 (rs5275) |
TT | 363/332 | 1.0 (ref.) | 98/101 | 0.8 (0.6–1.1) |
TC + CC | 436/439 | 0.9 (0.8–1.1) | 124/148 | 0.7 (0.5–0.9)b
|
Any NSAID |
COX-2 .926 (rs20417) |
GG | 418/412 | 1.0 (ref.) | 235/240 | 0.9 (0.7–1.1) |
GC + CC | 245/249 | 0.9 (0.8–1.2) | 126/137 | 0.8 (0.6–1.1) |
COX-2 .5209 (rs20432) |
TT | 427/417 | 1.0 (ref.) | 243/240 | 0.9 (0.7–1.1) |
TG + GG | 237/237 | 1.0 (0.8–1.2) | 118/133 | 0.8 (0.6–1.1) |
COX-2 .8473 (rs5275) |
TT | 290/285 | 1.0 (ref.) | 175/155 | 1.1 (0.8–1.4) |
TC + CC | 376/371 | 1.0 (0.8–1.3) | 186/223 | 0.8 (0.6–1.0) |
Table 4
COX-2 .8473 genotype, NSAIDs use, and risk for breast cancer by hormone receptor status, Long Island Breast Cancer Study Project, 1996–1997
ER or PR positive |
TT | Aspirin | 191/332 | 1.0 (ref.) | 46/101 | 0.7 (0.5–1.0) |
TC + CC | | 237/439 | 1.0 (0.7–1.2) | 61/148 | 0.6 (0.4–0.9)b
|
TT | Any NSAID | 147/285 | 1.0 (ref.) | 92/155 | 1.1 (0.8–1.5) |
TC + CC | | 210/371 | 1.1 (0.9–1.5) | 89/223 | 0.7 (0.5–1.0)c
|
ER and PR negative |
TT | Aspirin | 44/332 | 1.0 (ref.) | 13/101 | 1.0 (0.5–1.8) |
TC + CC | | 53/439 | 0.9 (0.6–1.4) | 22/148 | 1.1 (0.6–1.9)d
|
TT | Any NSAID | 40/285 | 1.0 (ref.) | 18/155 | 0.8 (0.5–1.5) |
TC + CC | | 39/371 | 0.8 (0.5–1.2) | 37/223 | 1.2 (0.7–1.9)e
|
When analyzed at the diplotype level, the ORs of ever use of any NSAID alone and carrying the GTT/GGC diplotype alone were, respectively, 0.9 (95% CI = 0.8 to 1.1) and 0.6 (95% CI = 0.2 to 1.4) compared with nonusers carrying all other diplotypes (data not shown). Ever use of any NSAID was associated with a pronounced reduction in OR among women carrying the diplotype of GTT/GGC (OR = 0.3, 95% CI = 0.1 to 0.9) compared with nonusers with all other diplotypes, but numbers were small in these groups and no significant interaction was observed (data not shown).
Discussion
The COX-2 pathway is now recognized to be important in human cancer development and progression [
49]. Numerous studies have suggested a role for
COX-2 in the initiation, promotion, and progression of cancers in different organs [
50‐
52]. However, little is known about the role of sequence variation within
COX-2 in breast cancer, and modification with NSAID use [
32‐
35]. In the present study, no overall associations between the three studied
COX-2 variant alleles and breast cancer risk were found. This finding is consistent with an observation from a nested case-control study conducted in Danish women (361 breast cancer cases and 361 matched controls) [
35], but it is not consistent with a report that indicated that the homozygous
COX-2 . 8473-CC genotype was associated with breast cancer risk (OR = 2.1, 95% CI = 1.3 to 3.3) in an Austrian population (500 cases and 500 controls) [
33]. The frequencies of variant CT and CC genotypes in our cases were similar to those in the prior studies [
33,
35]. We found little evidence to suggest that intake of aspirin or ever NSAID use interacted with
COX-2 genotypes to affect overall breast cancer risk. However, among women with hormone receptor positive breast cancer, there was some evidence to suggest that the reduction in risk associated with ever NSAID use (including both aspirin and nonaspirin NSAID use) was limited to women who carried the variant C allele for
COX-2 .8473; however, there was no corresponding interaction between this polymorphism and ever aspirin use alone in this subgroup.
Several previous studies, including the SNP500 database [
53], have reported the frequency of
COX-2 .8473 variant C allele in the 3'-UTR in Caucasian populations to range between 31.4% and 46.5% [
19‐
21], which is consistent with our finding (34.1%). Our larger sample size results in a more stable genotyping frequency compared with prior small studies. Although a small previous study [
20] found that carriers of the
COX-2 .8473 variant C allele had a significantly increased risk for lung cancer (ORs 2.12 for CT genotype and 4.28 for CC genotype), an expanded study in the same population [
28] failed to reproduce the association (ORs 0.96 for CT genotype and 0.97 for CC genotype). Furthermore, the significantly reduced OR observed for the GTT/GGC diplotype (OR = 0.5, 95% CI = 0.2 to 0.9) in the present study is probably a false-positive finding caused by chance because of the small number of observations (34 carriers of the diplotype with NSAID data).
Our findings on the roles of
COX-2 .8473 polymorphism and NSAIDs among women with hormone receptor positive breast cancer are biologically plausible. Polymorphisms present in the regulatory 3'-UTR of
COX-2 might hinder the binding of RNA-binding proteins and modulate mRNA stability and degradation, and finally decrease the production of protein [
54‐
56]. Therefore, it is reasonable to speculate that the
COX-2 .8473 polymorphism, located downstream of the stop codon in the adenylate/uridylate-rich 3'-UTR region [
20], could partly decrease mRNA stability and expression either by modifying the efficiency of polyadenylation signals or by affecting the binding affinity of regulatory elements. This could result in decreased cellular COX-2 activity and reduced inflammatory response, angiogenesis, and tumor growth. Continuous use of any NSAIDs appears to result, via a local blockade of the COX-2 enzyme, in a reduction in PGE
2 and aromatase synthesis. The consequent reduction in the local production of estrogen, in turn, is associated with the reduced breast cancer risk [
10‐
13,
57,
58]. However, the fact that there is no direct laboratory evidence on the function of the
COX-2 .8473 polymorphism may increase the likelihood of a spurious association or indicate the presence of some other functional SNPs in strong linkage disequilibrium with this SNP.
The more pronounced interaction between the
COX-2 .8473 polymorphism and ever NSAID use in ER-positive or PR-positive tumors than that in ER/PR-negative tumors is of particular interest. This might be an indication that different mechanisms are involved in subtypes of breast cancer. The lower frequency of COX-2 over-expression in hormone receptor positive tumors may make it easier to observe a significant protective effect of NSAID use among carriers of the variant C allele of
COX-2 .8473. This is consistent with the observation that elevated expression of COX-2 was statistically significantly associated with poor survival in ER-positive or PR-positive tumors but not in ER/PR-negative ones [
14]. The same dose of NSAIDs might have a protective effect in ER-positive or PR-positive tumors by acting on an inflammation-related pathway through decreasing COX-2 activity, reducing PGE
2 and aromatase synthesis, and decreasing estrogen production. However, there are also several discrepant reports showing that long-term daily use of aspirin significantly increased risk for ER/PR-negative breast cancer [
7], and ER-negative tumors are more sensitive to chemotherapy than are ER-positive ones [
59]. We did not observe a significant interaction between ever aspirin use, ever NSAID use, and the variant C allele of
COX-2 .8473 in ER/PR-negative breast cancer, perhaps because of limited power (Table
4). The nonsignificant interaction observed in ever aspirin use might be a reflection of either inadequate power or lack of a true association. Considering the biologic mechanism of action of aspirin and nonaspirin NSAIDs and the results obtained in the present study (Table
4), we believe that this inconsistency is due to limited power resulting from the small sample size in some subgroups of aspirin users. The observed significant interaction between ever NSAID use and the variant C allele of
COX-2 .8473 for decreasing breast cancer risk also included, to a certain extent, a contribution of ever aspirin use. The ORs for interactions between the genotype and either ever aspirin use or ever NSAID use were essentially similar (OR 0.6, 95% CI = 0.4 to 0.9 versus 0.7, 95% CI = 0.5 to 1.0). This is consistent with the results from our analysis of the interactions between nonaspirin NSAID use alone, aspirin use alone, and
COX-2 .8473 polymorphism (data not shown). Thus, the difference in interaction between ever aspirin use, ever NSAID use, and the
COX-2 .8473 polymorphism might well be the result of limited power for subgroup analyses rather than lack of a true association.
The relatively large sample size, population-based study design, and availability of both genotyping and NSAID information are the strengths of the study. The NSAID data were obtained from retrospective reporting of medication use, which might lead to recall bias. However, in order to explain our findings, cases should under-report more than controls or the accuracy of reporting should be related to genotype status or ER/PR status. Thus, it is unlikely that recall bias played a major role in explaining the present findings. Previous sensitivity analysis also indicated that the missing data would not have altered the overall conclusion of a protective effect between aspirin use and breast cancer [
5].
Another consideration was that our NSAIDs information did not include the use of prescription drugs containing NSAIDs (such as naproxen, etodolac, ketoprofen, and sulindac). The presumed 'unexposed' women who used prescription NSAIDs would be included in the nonuser category and lead to an underestimation of inverse associations. Such a misclassification of users as nonusers would tend to bias estimates of NSAID effects toward the null. The extensive information collected in the study for evaluating confounding factors and effect modifiers allowed us to assess confounding in the data analysis. ORs changed less than 10% when considering potential confounding factors individually or in multivariate models, indicating that the effects of confounding are limited. Inclusion of age at reference in logistic models minimized any possible residual confounding effect of age. Although our questionnaire also included variables on duration and frequency of aspirin and other NSAID use, providing some information on dose, the small sample size in some subgroups when categorized by COX-2 genotype limited power to evaluate dose-response effects.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
JS was responsible for genotyping polymorphisms, data analyses and interpretation, and drafting of the manuscript. MG contributed to study design, sample collection and manuscript preparation, and provided expertise in data analyses. MBT, SL, and AN contributed to the study design, sample collection, and manuscript preparation. RS was responsible for supervising laboratory work, and contributed to study design, biospecimen processing and manuscript preparation. All authors read and approved the final manuscript.