Nonsteroidal anti-inflammatory drugs (NSAIDs) clearly reduce the risk of human colorectal neoplasia in epidemiological and prospective randomized clinical studies of aspirin and nonaspirin NSAIDs, including selective cyclooxygenase-2 (COX-2) inhibitors, or coxibs ( 1–3 ). In contrast, the epidemiological and clinical data on NSAIDs in reducing breast cancer risk are not consistent. This inconsistency is likely attributable to contrasting expression patterns of COX-2, a key target of NSAIDs, in breast and colon neoplasia ( 4 , 5 ), and to differing activities of individual NSAIDs (which have varying selectivity for COX-2 vs COX-1), including a potentially selective impact of certain NSAIDs on hormone receptor–positive breast tumors.
In this issue of the Journal, Takkouche et al. ( 6 ) report an extensive meta-analysis (involving 38 studies) supporting an inverse association between NSAID use and risk of breast cancer. They found a statistically significant reduction in breast cancer risk associated with use of any NSAID (relative risk [RR] = 0.88, 95% confidence interval [CI] = 0.84 to 0.93) and similar associations for aspirin (RR = 0.87, 95% CI = 0.82 to 0.92) and ibuprofen (RR = 0.79, 95% CI = 0.64 to 0.97). They found no evidence of a dose–response relationship, and some studies indicated that coxibs were also associated with a lower risk of breast cancer ( 7 , 8 ).
This large-scale meta-analysis is consistent with several smaller meta-analyses ( 9–12 ). Furthermore, NSAIDs can prevent experimental breast cancer in numerous rodent models ( 4 , 5 ). Why then do individual observational and clinical studies vary substantially in their findings on NSAID use and breast cancer risk? The answer lies in an appreciation of the likely mechanisms of NSAID-mediated breast cancer suppression.
COX enzymes, the primary molecular targets of NSAIDs, synthesize prostaglandins (PGs) from arachidonic acid ( 13 , 14 ). COX-1 is normally expressed constitutively, whereas COX-2 is an early-response gene whose expression is increased in response to growth factors, oncogenes, and cytokines and is a key component of the inflammatory response ( 5 , 14 , 15 ). Transgenic overexpression of COX-2 induces mammary tumor formation in mice ( 16 ); activation of COX/PG signaling has multiple procarcinogenic consequences, including regulation of proliferation, apoptosis, angiogenesis, invasion, and immune responsiveness ( 5 , 14 ). Patterns of COX-2 overexpression differ substantially between breast and colorectal neoplasia. The frequency of overexpression is approximately 40% in invasive breast cancer vs approximately 80% in colorectal cancer ( 4 , 17 ) and approximately 80% in breast precancers vs approximately 50% in colorectal adenomas ( 4 , 5 , 18 ). The magnitude of COX-2 overexpression also is much higher in colorectal than breast cancer. In breast cancer, COX-2 overexpression is associated with overexpression of human epidermal growth factor receptor 2 (HER2) ( 17 , 19–21 ), also called c-erbB2 and neu, which (like COX-2) is expressed at a higher frequency in ductal carcinoma in situ than in invasive breast cancer. HER2 signaling can induce COX-2 transcription in vitro ( 22 ). Therefore, it is conceivable that coxibs may selectively target breast tumors that overexpress HER2 (and thus COX-2). HER2 overexpression is associated with hormone receptor–negative breast cancer, which can be prevented in mouse models by targeting HER2 ( 23 , 24 ). Coxibs can inhibit the development of hormone receptor–positive and –negative breast cancer in rodent models ( 4 , 5 ); either pharmacologic inhibition (coxib mediated) or genetic ablation of COX-2 can suppress HER2-driven, hormone receptor–negative mammary tumorigenesis in transgenic mouse strains ( 25 , 26 ). Perhaps surprisingly in light of the COX-2/HER2 relationship, to our knowledge no epidemiological studies to date have stratified NSAID responsiveness according to HER2 expression status. However, a selective effect of NSAIDs on HER2-overexpressing cancers (due to the coupled expression of HER2 and COX-2) may be obscured by procarcinogenic contributions of constitutive COX-1.
COX-1–derived PGs can contribute to tumorigenesis, as demonstrated most clearly by decreased rodent skin and intestinal neoplasia after genetic deletion of COX-1 ( 27–29 ). Therefore, the sum of COX-1 and COX-2 activity may be a key determinant in breast carcinogenesis. Consistent with this notion, tissue levels of the protumorigenic eicosanoid PGE 2 are only halved by knocking out COX-2 in HER2-overexpressing mouse mammary glands, with a parallel 50% decrease in mammary tumor multiplicity ( 25 ). In aggregate, these preclinical data suggest that inhibiting COX-1 could be important for the anticancer activity of NSAIDs in the setting of human breast cancer, in which COX-2 overexpression is not particularly prevalent ( 4 , 17 ). Furthermore, all NSAIDs are not equal in relative potency toward COX-1 and COX-2, ranging from greater than 150-fold selectivity of aspirin for COX-1 (vs COX-2) to greater than 100-fold selectivity of coxibs (such as celecoxib) for COX-2 (vs COX-1). Between these extremes are ibuprofen and naproxen, which are relatively nonselective. Perhaps the association between NSAID use and breast cancer risk could be clarified by evaluating individual NSAIDs or NSAID classes in the subset of breast cancers that are most likely to be sensitive to each one. This paradigm is well illustrated by the available datasets concerning aspirin.
Data on the association between aspirin use and breast cancer risk are mixed—an inverse association in some studies ( 7 , 8 , 30–39 ) but no apparent benefit in others, including several large cohort studies and one randomized clinical trial ( 40–48 ). However, subset analysis with stratification according to hormone receptor status revealed a selective inverse association of aspirin use with risk of hormone receptor–positive breast cancer in some but not all studies ( 35 , 43–46 , 49 , 50 ). The apparent relationship between hormone receptor expression and aspirin sensitivity may reflect the link between COX/PG signaling and estrogen biosynthesis ( 51 ). COX-derived PGs can activate a signaling cascade leading to increased transcription of the CYP19 gene that encodes the estrogen synthetase aromatase, and thus to increased estrogen biosynthesis ( 52–55 ). (Estradiol can then increase PG levels in a positive feedback loop.) This PG-dependent pathway is thought to contribute predominantly to regulation of peripheral, rather than ovarian, estrogen synthesis and thus assumes greater importance in the postmenopausal setting. Elucidation of this molecular pathway provides a potential mechanistic explanation for the selective suppression of hormone receptor–positive breast tumors by aspirin and other NSAIDs ( 35 , 43 , 44 , 49 ) and may help to explain why the receptor-positive–NSAID sensitivity relationship has been identified only in certain studies. An aspirin-mediated reduction in peripheral estrogen synthesis is likely to be more evident in postmenopausal than premenopausal women, in whom ovarian estrogen production far outweighs peripheral production. Furthermore, a model of NSAID suppression of estrogen synthesis and breast tumorigenesis suggests that hormone replacement therapy (HRT) could abrogate any protective effects of aspirin or other NSAIDs.
COX-independent effects also may contribute to potentially protective NSAID effects ( 56 ), particularly apoptosis. These effects may be mediated in part through 15-lipoxygenase-1 (15-LOX-1), decreased expression of which in human breast cancer is associated with a poor prognosis ( 57 ). NSAIDs can restore 15-LOX-1 expression, leading to apoptosis in other systems ( 58 ).
The findings of Takkouche et al. ( 6 ) support targeting the COX/PG signaling axis to prevent breast cancer. The cardiovascular side effects associated with coxibs (well described elsewhere) and potentially other NSAIDs necessitate refocusing COX-directed cancer prevention strategies to avoid this danger ( 3 , 4 ). Current directions include agents that target other components of the COX/PG signaling pathway and combination approaches that should allow increased efficacy and diminished toxicity with lower doses of the individual agents ( 59 ). For example, low-dose celecoxib plus the rexinoid bexarotene has synergistic preventive efficacy in an animal model of breast cancer ( 60 ).
Moving forward, analyses of associations between NSAID use and breast cancer risk will likely gain clarity from stratification based on the biology of NSAID action. Responsiveness to an individual NSAID may be predicated on hormone receptor expression, menopausal status and HRT use, or HER2 status (a routinely examined clinical marker that may be a surrogate for COX-2 overexpression in the breast). Risk modeling, eg, for hormone receptor–positive or –negative cancer, is also likely to be important to future NSAID chemoprevention of breast cancer ( 61 ). Adding to the complexity, genetic polymorphisms in COX-2 (and likely other components of the pathway) are also associated with breast cancer risk, potentially through modulation of NSAID responsiveness ( 62–64 ). Elucidating the contribution of such variant alleles will advance us toward personalized use of NSAIDs for reducing breast cancer risk.
National Institutes of Health (CA119273); Breast Cancer Research Foundation to LRH.