The cyclooxygenase (COX) enzymes catalyze a key step in the conversion of arachidonate to PGH2, the immediate substrate for a series of cell specific prostaglandin and thromboxane synthases. Prostaglandins play critical roles in numerous biologic processes, including the regulation of immune function, kidney development, reproductive biology, and gastrointestinal integrity. There are two COX isoforms, which differ mainly in their pattern of expression. COX-1 is expressed in most tissues, whereas COX-2 usually is absent, but is induced by numerous physiologic stimuli. Surprisingly, disruption of Cox1 (Ptgs1) in the mouse did not result in gastrointestinal abnormalities. cox-2 (Ptgs2) null mice show reproductive anomalies and defects in kidney development. Epidemiologic, animal, and human data indicate that NSAIDs, inhibitors of cyclooxygenase, are chemopreventive for colon cancer. COX-2 is overexpressed in 50% of benign polyps and 80 – 85% of adenocarcinomas. Offspring from cox-2 null by ApcΔ716 matings exhibit an 86% reduction in polyp number when compared to offspring from control animals, thus providing genetic evidence that COX-2 contributes to tumor formation or growth. The in vivo mechanism by which COX-2 affects tumor growth has not been determined. It is possible that both tumor and stromally derived COX-2 could influence tumor angiogenesis and/or immune function.
Metabolites of arachidonic acid are critical for numerous biologic processes, including inflammation, ovulation, implantation, angiogenesis, platelet aggregation, and immunologic function. Eicosanoids are the products of arachidonic acid metabolism, and the cyclooxygenase (COX, prostaglandin H synthase, PGHS) enzymes play a key role in the production of eicosanoids. Significant advances have been made in understanding the role of these enzymes in certain biologic processes. The focus of this review will be limited to the role of cyclooxygenase in development information and cancer.
Arachidonic acid is a 20 carbon unsaturated fatty acid distributed throughout the lipid bilayer of the cell, and is usually esterified at the SN-2 position of phospholipids. Phospholipase enzymes cleave membrane bound arachidonate, thus making it available for conversion to bioactive lipids. Once liberated, the arachidonic acid can be metabolized through one of three major pathways: (1) the cyclooxygenase pathway; (2) the lipoxygenase pathway; or (3) the cytochrome P-450 monooxygenase pathway. Additionally, free radical peroxidation can convert arachidonic acid non-enzymatically to yield isoprostanes.
The cyclooxygenase pathway, which is the focus of this study, is the most extensively studied of the major pathways. Cyclooxygenase-1 (COX-1, prostaglandin G/H synthase-1, 8,11,14-icosatrienoate, hydrogen donor : oxygen oxidoreductase, EC 18.104.22.168) catalyzes the conversion of arachidonic acid to PGH2, the immediate substrate for a number of cell specific prostaglandin and thromboxane synthases. This occurs via a two-step process, in which the first step introduces two molecules of oxygen to arachidonate, forming the bicyclic peroxide intermediate, prostaglandin G2 (PGG2). The second step occurs in a distinct reactive site located on the other side of the molecule, and requires the diffusion of PGG2 to this site. Here peroxidation results in the reduction of PGG2 to the freely diffusible PGH2 (Figure 1). Though these enzymes are membrane bound, they do not contain transmembrane domains; rather, they possess four amphipathic helices juxtaposed such that they form a localized region of hydrophobicity. The hydrophobic region serves to anchor the lower portion of the enzyme in the membrane. The cyclooxygenase active site is located in an area of hydrophobicity near the amphipathic helices. Access to this site occurs via a channel buried in the lipid bilayer. Both substrate and inhibitors use this channel to reach the active site.
Cycloxygenase-1 was first purified from bovine vesicular glands in 1976 (Miyamoto et al., 1976). COX-1 is constitutively expressed in many tissues including kidney, lung, stomach, duodenum, jejunum, ileum, colon, and cecum of rat, dog, Rhesus monkey, and human (Kargman et al., 1996). COX-1 activity is believed to be responsible for producing cytoprotective prostaglandins, such as prostacyclin and PGE2, which are thought to be critical to maintain integrity of gastric mucosa (Allison et al., 1992; Miller, 1983; Soll et al., 1991).
In 1989, Simmons et al. identified a novel cyclooxygenase cDNA, of which the corresponding mRNA was induced by v-src transformation of chicken embryo fibroblasts (Simmons et al., 1989). In 1991, Kujubu et al. independently isolated a cDNA encoding this isoform by differential screening of 3T3 fibroblasts treated with phorbol ester (Kujubu et al., 1991). This second isoform, now known as cyclooxygenase-2 (COX-2), shares significant sequence homology and catalytic activity with COX-1. However, its expression pattern is markedly different. Most tissues, with the exception of the placenta, the macula densa of the kidney and brain, do not constitutively express COX-2 (Harris et al., 1994; Hirst et al., 1995). However, a variety of extracellular and intracellular stimuli will rapidly induce COX-2. These stimuli include lipopolysaccharide (LPS) (Fu et al., 1990; Lee et al., 1992; O'Sullivan et al., 1992), forskolin (Kujubu and Herschman, 1992), interleukin-1 (IL-1), tumor necrosis factor (TNF) (Coyne et al., 1992; Geng et al., 1995; Jones et al., 1993), serum (DeWitt and Meade, 1993; Ryseck et al., 1992), epidermal growth factor (EGF) (Hamasaki et al., 1993), transforming growth factor alpha (TGFα) (DuBois et al., 1994), interferon-γ (Riese et al., 1994), retinoic acid, platelet activating factor (PAF) (Bazan et al., 1994), endothelin (Kester et al., 1994), and arachidonic acid. Induction of COX-2 is transient, with a return to baseline within 24 – 48 h following treatment.
Cyclooxygenases and inflammation
In 1971, Vane and colleagues first demonstrated that aspirin and indomethacin inhibited prostaglandin production by blocking cyclooxygenase enzymatic activity (Vane, 1971). Since that report, it has been found that nonsteroidal anti-inflammatory drugs (NSAIDs) directly affect cyclooxygenase activity, either by covalently modifying the enzyme (as in the case of aspirin and the selective COX-2 inhibitor APHS), or by competing with the substrate for the active site (as with virtually all other NSAIDs).
Because prostaglandins participate in a number of normal physiologic functions, it is predictable that chronic blockade of cyclooxygenase leads to some undesirable side effects. Potentially life-threatening side-effects of NSAID use include gastrointestinal ulceration, bleeding and perforation. Some studies have estimated that regular users of NSAIDs have a threefold greater relative risk of developing serious gastrointestinal complications when compared to non-users (Gabriel et al., 1991). It is believed that the loss of COX-1 generated cytoprotective PGI2 or PGE2 in the stomach contributes to gastric ulceration (Miller, 1983). Therefore, inhibitors which could distinguish between the two cyclooxygenase isoforms might achieve analgesic and anti-inflammatory benefits without the accompanying undesirable gastrointestinal side effects.
The prototypical cyclooxygenase inhibitor, aspirin, blocks cyclooxygenase enzymatic activity covalently through the acetylation of Ser-530 in COX-1, and Ser-516 in COX-2 (Wennogle et al., 1995). This modification does not affect the peroxidase activity of the enzyme, and aspirin-acetylated COX-2, but not COX-1, can generate 15-HETE, which is normally produced via the lipoxygenase pathway (Capdevila et al., 1995; Xiao et al., 1997). In contrast, most other NSAIDs compete with arachidonate for active site binding. In purified enzyme assays, both non-preferential inhibitors such as indomethacin and flurbiprofen (Laneuville et al., 1994), and selective COX-2 inhibitors such as NS-398, DuP-697, and DFU (5,5-dimethyl-3-(3-fluorophenyl)-4-(4-methylsulphonyl)phenyl-2(5H)-furanone) (Copeland et al., 1994; Riendeau et al., 1997), efficiently inhibit both isoforms. COX-2 selectivity with these inhibitors is achieved only when COX-2 is pre-incubated with inhibitor prior to substrate addition. The binding of the inhibitor occurs as a two step mechanism. Initially, the inhibitor binds with a relatively high dissociation constant (Ki 140 μM for DFU), followed by a slow isomerization to a tightly bound complex with a change in the dissociation constant (Ki 0.21 μM for DFU). In contrast, inhibition of COX-1 by this class of drugs is much weaker, rapidly reversible, and detectable only at low arachidonate concentrations. The difference in the mechanism of inhibition of COX-1 and COX-2 makes it difficult to compare the selective ratios of various COX-2 inhibitors, since the arachidonic acid concentration can drastically affect the IC50 for COX-1.
Recently, a new COX-2 inhibitor has been developed (Kalgutkar et al., 1998), APHS (o-(acetoxyphenyl)hept-2-ynyl sulfide), which is twenty-one times more selective for COX-2 than COX-1. While this level of selectivity is not as great as that of some other COX-2 inhibitors, this inhibitor offers a unique advantage in that it covalently modifies the enzyme, thus insuring complete, permanent activation of COX-2 while sparing COX-1 enzymatic function.
The pro-inflammatory role of COX-2 recently has been questioned. Using the carrageenin-induced pleurisy model, investigators have demonstrated that later-stage COX-2 generated cyclopentanone prostaglandins may actually enhance resolution of inflammation (Gilroy et al., 1999). Consistent with the currently accepted model, it was found that COX-2 generated prostaglandins of the E series were pro-inflammatory during the first 24 h. At 48 h, a shift from polymorphonuclear leukocytes to mononuclear leukocytes, and a shift from PGE2 to 15-deoxy 12 – 14PGJ2 was observed in the exudate. If NS-398, a selective COX-2 inhibitor, was administered 1 h prior to carrageenin injection, then inflammation was reduced. However, if treatment was delayed for 24 h, then the inflammatory state was exacerbated. The late stage effects were reversible by adding back PGD2 or 15dPGJ2. Furthermore, investigators have found that inhibition of COX-2 in animal models for ulcer formation decreases the rate of ulcer healing (Mizuno et al., 1997; Schmassmann et al., 1998). These observations have clear clinical implications if they translate accurately from animal models of inflammation to inflammatory conditions in humans.
Phenotypic analysis of mice in which either Cox1 (Ptgs1) or cox-2 (Ptgs2) has been inactivated via homologous recombination yields information about the role of cyclooxygenases in development. Such animals were first produced and studied in 1995 (Dinchuk et al., 1995b; Langenbach et al., 1995; Morham et al., 1995). Phenotypic abnormalities, some of which were unexpected, were observed in both Cox1 and cox-2 null mice and are described below and summarized in Table 1.
Murine COX-1 null phenotype
In 1995, Langenbach et al. utilizing homologous recombination, disrupted the murine Ptgs1 gene encoding COX-1 (Langenbach et al., 1995). COX-1 had long been considered the `housekeeping' cyclooxygenase, and therefore it was surprising that these null mice were generally healthy, displaying no immediately obvious pathology. However, detailed study of the animals revealed certain phenotypic anomalies, many of which were unexpected. One expected finding in the null mice was decreased platelet aggregation in response to arachidonic acid treatment. Platelet localized COX-1 enzyme produces precursors for the synthesis of thromboxane, a potent inducer of platelet aggregation (Schafer, 1995). COX-1 deficient platelets presumably cannot produce these precursors, and therefore exhibit delayed aggregation. Also, because platelets lack nuclei, they are unable to produce COX-2 via a compensatory mechanism. Despite the decrease in platelet aggregation, the Cox1 null mice did not have any significant hemorrhagic complications.
As mentioned above, COX-1 generated prostaglandins appear to be cytoprotective to the gastric mucosa. Thus, it was hypothesized that null mice might exhibit gastric pathology. Interestingly, the Cox1 null mice generally had no identifiable gastrointestinal abnormality, despite the fact that gastric PGE2 levels in null mice were only 1% of the levels observed in wild type mice (Langenbach et al., 1995). Additionally, when null mice were treated with indomethacin, an NSAID known to induce ulcer formation, there was reduced ulceration when compared to wild type mice. Western blotting revealed that COX-2 protein levels were equivalent in wild type and Cox1 null mice. Therefore, compensation by induction of COX-2 was not occurring in these tissues. The authors speculated that compensation by non-prostaglandin pathways may have been occurring. Furthermore, they suggested that NSAID-induced gastric ulceration may develop through mechanisms other than COX-1 inhibition.
Another unexpected finding in the Cox1 null mice was a lack of significant renal pathology. COX-1 is widely expressed in the kidney, and is known to produce vasodilatory prostaglandins which are considered important for the maintenance of renal blood flow and glomerular filtration rate. NSAID-induced nephropathy is thought to result from decreased production of these renoprotective prostaglandins via COX-1 inhibition (Zambraski, 1995). In the study by Langenbach et al. three of six kidneys examined from null mice showed scant foci of immature tubules, while none of the six age-matched wild type kidneys showed these lesions. However, the lesions were of minimal severity, and did not progress, or cause appreciable alterations in kidney function (Langenbach et al., 1995).
Prostaglandins play important regulatory roles in the reproductive processes of ovulation, implantation, and parturition (Chakraborty et al., 1996; Lim et al., 1997). Matings of Cox1 null females with Cox1 null males resulted in normal size litters with few viable offspring (90% perinatal mortality). However, fertility, defined as the capacity to conceive, was normal in both male and female null mice. Furthermore, matings in which either parent is Cox1 heterozygous resulted in normal litter size and perinatal pup survival. These matings resulted in equal numbers of homozygous and heterozygous pups, suggesting that COX-1 production in either maternal or fetal tissue could restore perinatal survival. From these data, the authors speculated that prostaglandins generated by the COX-1 enzyme in placenta and/or fetal tissue are critical for parturition, i.e. the successful initiation and completion of labor (Langenbach et al., 1995).
Murine cox-2 null phenotype
cox-2 null mice develop severe nephropathology within the first 6 weeks of life (Dinchuk et al., 1995a; Morham et al., 1995). Unlike human kidney development, mouse kidney development continues postnatally. Three-day-old cox-2 null mice exhibit no kidney abnormalities. However, by 6 weeks of age, striking abnormalities appear in the subcapsular nephrogenic zone, and include glomerular sclerosis and associated tubulointerstitial injury. Compromised renal function as a result of damage to overworked immature nephrons results in increased blood urea nitrogen and serum creatinine levels. Additionally, in one study, the mice had an increased risk for development of secondary pyelonephritis (Dinchuk et al., 1995a). Interestingly, careful evaluation of later generations of the cox-2 null mice have revealed renal phenotypic abnormalities which are much less severe, possibly due to some sort of adaptation process (Lim et al., 1999).
In contrast to Cox1 null animals, in which parturition appears to be the primary reproductive abnormality, cox-2 null mice display abnormalities in every phase of the reproductive process. The phases of reproduction are sequential, complex, and very precisely timed. Studies undertaken prior to the development of cox-2 null animals, in which gonadotropin treatment resulted in rapid induction of cox-2 in granulosa cells, suggested that COX-2 was important in ovulation. Early studies in cox-2 null mice noted that most females were infertile. Analysis of the reproductive tract in the animals revealed small ovaries with absent corpora lutea. Follicular development, however, appeared normal (Dinchuk et al., 1995a). A later study in cox-2 null mice, specifically examining the reproductive process, revealed abnormalities in ovulation, fertilization, implantation, and decidualization (Lim et al., 1997). cox-2 null animals in this study displayed reduced ovulation in response to gonadotropin stimulation (nine eggs per mouse in cox-2 null mice versus 36 eggs per mouse in controls). Furthermore, the cox-2 null animals were largely infertile, likely as a result of abnormal appearing, defective oocytes. Implantation also was severely impaired in cox-2 null mice. Less than 2% of transferred wild type blastocysts implanted in cox-2 null mice, while 50% implanted in wild type mice. Decidualization, the uterine angiogenic and proliferative changes which occur during and subsequent to embryo implantation, was compromised in the cox-2 null animals. Using intra-uterine oil infusion to induce artificial decidualization, the investigators demonstrated that cox-2 null mice were significantly less likely to form decidual processes when compared with wild type mice. These findings are not unexpected, considering that Cox-2 is expressed in the mesenchymal stroma of the uterus surrounding the implantation site, and that PGE2 is considered to be important in implantation signaling. Surprisingly, though, treatment with exogenous PGE2 did not rescue decidualization. In contrast, PGI2, a prostanoid markedly increased at day 5 of pregnant wild type mice, was able to partially restore decidualization. Decidualization is the result of complex processes involving both the embryo and uterine tissue. If wild type embryos are transferred to pseudo-pregnant Cox-2 null animals, the embryos fail to implant. Thus, uterine COX-2, as opposed to COX-2 expressed in the embryo, appears to be an absolute requirement for embryo implantation (Lim et al., 1997).
Recently, both COX isoforms have been found to contribute to T-cell development (Rocca et al., 1999). It was found that PGE2 produced via COX-1 was required for thymocyte transition from the double negative (CD4−, CD8−) to double positive lineage (CD4+, CD8+). COX-2-generated prostaglandins were shown to positively affect the double negative population, and COX-2-generated PGE2 was required for CD4 single positive thymocyte development.
COX-2 in neoplasia
Colorectal cancer is the third most common cancer in the US (excluding non-melanoma skin cancer), and the second most common cause of cancer related fatalities. In 1998, there were over 130 000 new cases of colorectal cancer in the United States, and roughly 55 000 deaths (Landis et al., 1998). Colon cancer is infrequent before the age of 40; however, after 50 the incidence of colon cancer doubles with each decade of life, and 90% of all cases are diagnosed in individuals over the age of 50. Despite the development of new screening strategies, aggressive surgical and adjuvant therapy, and intensive research efforts, little progress has been made in the successful management of advanced disease. However, research efforts have yielded some promising leads, one of which is the apparent association between NSAID use and decreased cancer risk.
Early studies using animal models of colon cancer indicated that NSAIDs were chemopreventive (Kudo et al., 1980; Pollard and Luckert, 1981). Two surgeons, Waddell and Loughry, observed that in patients with Familial Adenomatous Polyposis (FAP), a familial form of colorectal cancer, sulindac decreased intestinal polyp burden. Another study investigating the relationship between aspirin use and colon cancer involved following colon cancer fatality rates prospectively among 600 000 individuals (Thun et al., 1991). The relative risk of colon cancer for individuals reporting use of 16 or more aspirin tablets a month was 0.60. There was no protective effect found with acetaminophen, an analgesic which does not affect cyclooxygenase activity. Follow-up studies indicated that aspirin use also reduced deaths from esophageal, gastric and rectal cancer (Thun et al., 1993).
In 1991, Labayle et al. again evaluated patients with the hereditary colon cancer syndrome referred to as familial adenomatous polyposis coli (FAP) (Labayle et al., 1991). Treatment of these patients with sulindac led to significant repression of polyps. In 1993, in a carefully designed, randomized, double-blind, placebo-controlled study of 40 patients with FAP, Giardiello et al. (1993) noted that patients who took sulindac had a significant reduction in the size and number of colonic polyps compared with matched controls. That same year, Nugent et al. (1993) reported that sulindac treatment caused a statistically significant reduction in rectal polyp count among 24 FAP patients. These studies added to the mounting evidence that NSAIDs had antineoplastic effects, and could affect the development of colorectal cancer, even at the earliest stages.
If NSAIDs do reduce the risk of developing colon cancer, by what mechanism do they achieve this effect? Certainly, the cyclooxygenase enzymes are known targets of NSAIDs. If cyclooxygenase contributed to tumor growth, then one might expect that levels of downstream metabolites (i.e. prostaglandins) would be increased in tumors, and that there would be abnormal cyclooxygenase expression or catalytic activity. Indeed, it has been reported that PGE2 and 6-keto PGF1α levels were elevated in colorectal cancers (Oka et al., 1994; Pugh and Thomas, 1994). The increase in prostaglandins may be explained by increased cyclooxygenase expression or increased cyclooxygenase catalytic activity. In 1994, Eberhart and DuBois reported that COX-2, but not COX-1, was elevated in colorectal cancers. They found that approximately 50% of adenomas and 80 – 85% of adenocarcinomas had increased expression of COX-2, suggesting that COX-2 could be involved in colorectal carcinogenesis (Eberhart et al., 1994). Other laboratories have since confirmed this observation (Kargman et al., 1995; Kutchera et al., 1996; Sano et al., 1995) (Figure 2).
Further evidence implicating COX-2 in colorectal carcinogenesis can be found in studies of animal models of colorectal cancer. One animal model for colorectal cancer is the genetically predisposed Multiple Intestinal Neoplasia (Min) mouse. Min mice usually develop 20 – 30 polyps throughout their lifetime. The polyps arise almost exclusively in the small intestine, and rarely, if ever, progress to carcinoma, probably because the animals die due to intestinal obstruction or anemia within the first year of life. Another commonly used animal model of colorectal cancer is the azoxymethane (AOM) treated rat. The lesions arising in the AOM rat seem to progress from aberrrant crypt foci (ACF), to polyp, and finally to carcinoma. Just as in colorectal cancer in humans, COX-2 levels are increased in both AOM rat tumors (DuBois et al., 1996; Singh et al., 1997) and Min mouse tumors (Boolbol et al., 1996; Williams et al., 1996). Sulindac sulfide effectively prevents polyp formation and causes rapid regression of existing polyps in the Min mouse (Boolbol et al., 1996; Chiu et al., 1997). In the AOM rat model, sulindac reduces ACF formation, polyp number and carcinogenesis (Pereira et al., 1994; Samaha et al., 1997). These animal studies, paralleling epidemiological and clinical data in humans, further support the role of cyclooxygenase in colorectal carcinogenesis.
To directly test the role COX-2 plays in tumorigenesis, researchers turned to a genetic model. Mice heterozygous for an adenomatous polyposis coli (APC) mutant allele develop hundreds of intestinal polyps. When cox-2 null mice were bred with ApcΔ716 mutant mice, tumor burden in the offspring decreased in a gene dose-dependent fashion (Oshima et al., 1996). This experiment provided direct genetic evidence that COX-2 is important for polyp promotion. Interestingly, residual tumors were still present in the cox-2 null background. A possible explanation for continued tumorigenesis is that compensation by COX-1 may have been occurring. This study underscores the crucial role COX-2 plays in tumor formation and growth in this animal model.
COX-2 localization in tumors
There are conflicting data regarding whether COX-2 is increased in the epithelial or the stromal component of tumors. COX-2 has been found to be increased in the epithelium of adenomas in the multiple intestinal neoplasia mouse (Min) (Williams et al., 1996), azoxymethane (AOM)-treated rat cancers (Shao et al., 1990), replication error repair (RER) positive human tumors (Karnes et al., 1998), and sporadic human colorectal cancers (Kutchera et al., 1996; Sano et al., 1995). More recently, it has been found that COX-2 expression is increased in the stromal component of adenomas from the Min mouse (Hull et al., 1999; Shattuck-Brandt et al., 1999) and the carcinogen-induced tumors in the AOM mouse (Shattuck-Brandt et al., 1999). It is possible that some of these differences in localization may be due to non-specific binding of the antibodies which were used for immunostaining or in tissue procurement or preparation.
Oshima et al. (1996) localized COX-2 expression in the polyps in ApcΔ716 by crossing these mice with mice in which the bacterial LacZ gene interrupted Ptgs2, thus placing it under the control of the Ptgs2 promoter. They observed LacZ expression in areas surrounding the adenoma, but no significant expression in the epithelial cells of the adenomas. Therefore, stromally derived COX-2 may promote tumor growth by producing bioactive prostaglandins which affect carcinoma cells in a paracrine fashion. COX-2 may thus be acting as a `landscaping tumor promoter' according to the landscaping model proposed by Kinzler and Vogelstein (1998). This model supports the notion that COX-2 expression in the stromal component of the adenoma influences tumor growth. Therefore, the mechanism(s) for the inhibition of tumor formation and growth by NSAIDs may in part involve inhibition of stromal COX-2.
This complex stromal/epithelial interaction may explain some of the difficulties in attributing the anti-neoplastic effects of NSAIDs solely to inhibition of COX-2 in cell culture studies. In cultured cancer cells, the concentration of NSAID required to inhibit the growth of the cells or to induce apoptosis is typically 3 – 4 orders of magnitude greater than the amount required to inhibit prostaglandin production. Furthermore, colorectal cancer cell lines that do not express either cyclooxygenase isoform show similar sensitivity to high dose NSAID treatment. And lastly, there is variability in the effects that different NSAIDs have on cancer cells. Some NSAIDs, such as sulindac sulfide, potently induce apoptosis, while other NSAIDs, such as the selective COX-2 inhibitor SC-58125, primarily induce cell cycle arrest, thus decreasing cellular proliferation rates. If cyclooxygenase were the only relevant target, then NSAIDs might be expected to have similar phenotypic effects. Sulindac and aspirin have been shown to inhibit NF-κB signaling by directly blocking the activity of IκB kinase β (Grilli et al., 1996; Yamamoto et al., 1999). Additionally, indomethacin, a non-selective COX inhibitor, can act as a direct ligand for peroxisome proliferator activating receptor (PPAR) α and γ, and thus may directly regulate gene transcription (Lehmann et al., 1997). It is possible that in addition to these targets, NSAIDs may interact with other, as yet unidentified, cellular targets. Perhaps one of the reasons that the effects of NSAIDs on carcinoma cells are so enigmatic may be that these in vitro assays do not accurately reflect the stromal/epithelial interactions arising in vivo.
As illustrated above, most of the confusion regarding the molecular target for NSAIDs has been generated in cell culture assays. In vivo, there is much stronger evidence that COX-2 does play a role in tumor formation. In vivo, the effective anti-neoplastic NSAID dose is comparable to the amount of drug required to inhibit prostaglandin production. Several investigators have shown that treatment of mice with low doses of selective and non-selective COX-2 inhibitors will attenuate growth of cancer cells grown as xenografts in mice (Goldman et al., 1998; Hial et al., 1976; Sheng et al., 1997; Tsujii et al., 1998; Williams et al., 1999).
COX-2 parallels between cancer and development
Some of the same processes occur in both development and cancer. Indeed, the process of uterine decidualization sometimes is referred to as a pseudomalignant state. It is possible that understanding the role COX-2 plays in development will clarify its role in cancer and vice versa (Figure 3).
During blastocyst implantation and decidualization, a vascular network must be established to support the nutritional needs of the developing embryo. In much the same way, the expansion of a tumor mass requires the genesis of a vascular network to support its metabolic demands. There is some evidence that COX-2 generated prostaglandins participate in angiogenesis, a process common to both development and cancer.
Maximal tumor expansion occurs when the tumor is able to escape immunologic surveillance, thus avoiding the anti-tumoral onslaught of activated macrophages and cytotoxic T-cells. The `semi-allogenic' blastocyst also is considered foreign material, and likewise must escape immunologic surveillance in order to survive. COX-2-generated prostaglandins have been demonstrated to be immunosuppressive. In tumor models, macrophage-secreted PGE2 suppresses both macrophage mediated and natural killer cell-mediated cytotoxicity. It has also been demonstrated that PGI2 is essential for embryo implantation. It is possible that during implantation, PGI2 is immunosuppressive, and helps to establish an immune-privileged environment in the decidualization zone.
Eicosanoids contribute to normal physiologic processes such as inflammation, development and immune function. It has been suggested that COX-1 is the `housekeeping' isoform of cyclooxygenase, and that COX-2 is rapidly inducible in response to numerous intracellular and extracellular stimuli, and acts in a pro-inflammatory fashion. Phenotypic analyses of Cox1 and cox-2 null mice have made it clear that the role of these genes in a developing organism is much more complicated than was once thought. For example, the fact that Cox1 null mice do not spontaneously develop ulcers raises the possibility that NSAID-induced gastric ulceration could occur through mechanisms other than, or in addition to, COX-1 inhibition. Unexpected characteristics are not limited to COX-1. Recently, COX-2 has been demonstrated to play a role in the resolution of inflammation and in ulcer healing. Thus, an expanded role for COX-generated eicosanoids must be considered in the maintenance of mucosal integrity.
There is ample genetic and pharmacologic evidence to implicate COX-2 in neoplasia. The precise contribution of COX-2 to neoplastic growth has not been elucidated. However, there is some evidence which suggests that COX-2 may blunt the apoptotic response in tumor cells. In this way, COX-2 may play a direct role in tumor cell growth. Additionally, there is evidence that COX-2 may indirectly modulate tumor expansion. It has been demonstrated that COX-2 induces angiogenesis in vitro, and can downregulate natural killer T-cell function; thus, COX-2 in the stromal compartment may influence tumor growth. Hopefully, future experiments will elucidate the specific roles of the cyclooxygenases in neoplasia and inflammation, and will thereby contribute to the development of safe, effective therapeutic interventions.
This work was supported in part from United States Public Health Services Grants (to RN DuBois) DK 47297, P030 ES-00267-29, PO1CA-77839. RN DuBois is a recipient of a VA Research Merit Grant and the Mina C Wallace Professor of Gastroenterology and Cancer Prevention. We also thank the TJ Martell Foundation for generous support. We thank Brigid Hogan for her insightful suggestions.