Cyclooxygenases (COXs), also known as prostaglandin H synthases or prostaglandin endoperoxide synthases (E.C. 1.14.99.1), are fatty-acid oxygenases of the myeloperoxidase superfamily that are most closely related to the pathogen-inducible oxidases and linoleate diol synthases of plants and fungi 1. The purification of COX-1 (then called simply COX) from sheep 2 and bovine 3 seminal vesicles in 1976 led to the cloning of the COX-1 gene in 1988 456. For many years, it was thought that the constitutively active COX-1 protein was the only cyclooxygenase in eukaryotic cells, but in 1991 a second, inducible enzyme was identified through studies of cell division; this second enzyme is now called COX-2 78. The structure of the human COX-1 and COX-2 genes are shown in Figure 1a, and the properties of COX-1 and COX-2 and the genes encoding them are compared in Table 1.

All vertebrates investigated, including cartilaginous fishes, bony fishes, birds, and mammals, have two COX genes: one encoding the constitutive COX-1 and another the inducible COX-2. COX-1 and COX-2 share approximately 60-65% amino-acid identity with each other; COX-1 orthologs (without the signal peptide) share approximately 70-95% amino-acid identity across vertebrate species and COX-2 orthologs share 70-90%. Additionally, coral (of the phylum Cnidaria) and sea squirt (ascidian) each have two COX genes, which may have arisen from gene-duplication events independent from those that produced vertebrate COX-1 and COX-2 9. It is clear that the vertebrate, coral and ascidian COX genes all descend from a common ancestor.

Intron-exon junctions are highly conserved in all species, with a few notable exceptions: the vertebrate COX-1 genes contain an extra intron (intron 1), and the ascidian and coral genes have extra introns or lack some exons in the regions that encode exons 6, 7 or 11 in vertebrate COX-1 9. The exon structures of COX genes largely reflect the domains encoded by the proteins (Figure 1b). The structure of human cyclooxygenase genes and their expression and regulation have been reviewed elsewhere 10.

COX genes have not been found in insects, unicellular organisms, or plants, although prostaglandins, their products, have been found in some of these organisms 11. Recently, an enzyme that catalyzes the synthesis of prostaglandin E₂ from arachidonic acid (the substrate of COXs) was cloned from the protozoan Entamoeba histolytica. This enzyme shows no clear structural similarity to COXs, suggesting that alternative evolutionary paths to prostaglandin synthesis have evolved in some organisms 12.

COXs are all close to 600 amino acids in size and have a similar primary structure 1314 (Figure 1b). The crystal structure of sheep COX-1 (minus the post-translationally cleaved signal peptide), was obtained in 1994 15; human and mouse COXs have since been crystallized and show strikingly similar features 1617. After the signal peptide, the amino terminus of the protein contains a single epidermal growth factor (EGF) module with conserved disulfide bonds that functions as a dimerization domain. This is followed by a series of four amphipathic helices that anchors the protein to one leaflet of the membrane. This 'monotopic' type of insertion into a membrane has been found only in this enzyme and a few other proteins such as squalene cyclase and S-mandelate dehydrogenase 1819. The remainder of the protein consists of the catalytic domain, which has two distinct cyclooxygenase and peroxidase active sites.

COXs are highly conserved, and few significant differences are seen in the dimerization, membrane-binding and catalytic domains between COXs from different species. The amino-terminal hydrophobic signal peptides differ significantly in length between species. In the case of two splice variants of canine COX-1, the signal peptide has been found not to be cleaved from the enzyme when expressed in insect cells 20.

The catalytic domain contains conserved alpha-helical structures and a heme-binding motif shared with other peroxidases 15. COXs are glycosylated on asparagine in all organisms. One N-glycosylation site (Asn143, using the numbering of human COX-1) is absolutely conserved, and other sites are shifted only slightly in different homologs. For example, Asn410 in sheep COX-1 (orthologous to Asn4O9 in human COX-1), found to be essential for folding 21, is shifted to Asn394 in coral COXs. The cyclooxygenase active site is a narrow tunnel, approximately 8 Å wide and 25 Å long, that opens in the membrane-binding domain (Figure 2). This site accepts the arachidonic acid that is liberated from the membrane by cellular phospholipases. Amino acids lining this site are largely hydrophobic and serve to 'solvate' the hydrophobic substrate into the site 22. Exceptions to this hydrophobicity are Arg119, Tyr384 and Ser529 (numbered according to the human COX-1 enzyme sequence). Arg119 coordinates the carboxyl group of arachidonic acid by a salt bridge or hydrogen bond; Ser529 coordinates the geometry of attack in the complex bis-dioxygenation reaction performed; and Tyr384 forms a critical tyrosyl radical that initiates the cyclooxygenase reaction by abstraction of hydrogen from arachidonic acid (see Mechanism section). Non-steroidal anti-inflammatory drugs (NSAIDs) competitively inhibit the cyclooxygenase active site; an exception is the NSAID aspirin, which covalently modifies the enzyme by acetylating Ser529. In contrast to the cyclooxygenase active site, the peroxidase site is a solvent accessible cleft located on the surface of the enzyme furthest from the membrane (Figure 2).

COX-1 is ubiquitously and constitutively expressed in mammalian tissues and cells, whereas COX-2 is highly inducible and is generally present in mammalian tissues at very low levels, unless increased by one of many types of stimuli such as cytokines and growth factors.

Both COXs are largely located on the lumenal side of the endoplasmic reticulum (ER) membrane and the nuclear envelope, although they have also been detected in some situations in lipid bodies, mitochondria, filamentous structures, vesicles and in the nucleus 23242526. The lumen of the ER is important for both the structure and function of COXs: its oxidative potential allows formation of the disulfide bonds of the enzymes, and N-linked glycosylation - which occurs in the ER - appears to be necessary for proper protein folding 21. Moreover, the final product of COXs, prostaglandin H₂, is sufficiently non-polar to diffuse through the membrane of the ER to isomerases located on the cytosolic surface of the ER or in the cytosol (Figures 2, 3). Lipid bodies may provide a similar environment, but the role of COXs in the nucleus is unknown.

Both classes of COX are bifunctional enzymes with two distinct catalytic activities: cyclooxygenase (or bis-dioxygenase) activity and peroxidase activity (Figure 3a). The primary products of COXs were first detected in human seminal fluid by clinicians studying uterine contraction 27. Thought to be the product of the prostate gland, these highly potent bioactive compounds were given the name prostaglandins. They are synthesized in virtually all tissues in vertebrates, however, and some organisms that lack prostate glands, such as corals, also synthesize prostaglandins. Thus, in many respects the term prostaglandin is a misnomer. Initially, the enzyme activity that synthesized prostaglandins was frequently called prostaglandin synthetase, but because it does not require ATP it is now called prostaglandin G/H synthase to fit the nomenclature convention. It is more popularly known as cyclooxygenase, a name that only partially describes the enzyme since it refers to only one of its two enzymatic activities.

Prostaglandin isomers - including thromboxane and prostaglandins D₂, E₂, F_2α, and I₂ (prostacyclin; Figure 3b) - function in numerous physiological and pathophysiological processes, such as pyresis (fever), algesia (sensitivity to pain), inflammation, thrombosis, parturition, mitogenesis, vasodilation and vasoconstriction, ovulation, and renal function. Prostaglandin isomers act upon G-protein-coupled receptors 28, and there are multiple receptors for some isoforms (such as prostaglandin E₂). Prostaglandins are short-lived in vivo (with half-lives of seconds to minutes), and act in an autocrine or a paracrine rather than an endocrine fashion. COX-1 was first studied in tissue and cell homogenates, and in this context was shown by Vane 29 to be the inhibitory target of NSAIDs.