Cyclooxygenase-2 and its regulation in inflammation

There is no doubt that the identification of an inducible isoform of cyclooxygenase (here referred to as cyclooxygenase-2 or COX-2) has brought about a renaissance in prostanoid biochemistry, pharmacology and therapeutics. This area is now as vigorous as it was 20 years ago when thromboxane A2 (TXA2) and prostacyclin (PGI2) were discovered1’2 and indeed is almost as active as that of nitric oxide (NO), with which it shares many features and correlations. One of the most important features of COX-2 is its close association with a variety of inflammatory mediators and its consequent description as the COX isoform involved in and responsible for many of the signs of inflammation. It is also generally accepted that COX-1 is the constitutive isoform involved in the physiological actions of prostaglandins (PGs) in the stomach and kidney, the inhibition of which leads to gastric ulceration and nephropathy as side effects of anti-inflammatory therapy with nonsteroidal anti-inflammatory drugs (NSAIDs). The initial findings and many of the subsequent developments are based on the techniques of molecular biology3’4 and are often expressed in terms unfamiliar to many researchers already established in inflammation. Our purpose in this review is to summarize the progress made so far in characterizing the regulation of COX-2, to evaluate its role in inflammation and, as a consequence, to assess the utility of the selective inhibitors of COX-2. In order to establish the appropriate context for the analysis of regulatory mechanisms, we shall


Introduction
There is no doubt that the identification of an inducible isoform of cyclooxygenase (here referred to as cyclooxygenase-2 or COX-2) has brought about a renaissance in prostanoid biochemistry, pharmacology and therapeutics. This area is now as vigorous as it was 20 years ago when thromboxane A2 (TXA2) and prostacyclin (PGI2) were discovered 1'2 and indeed is almost as active as that of nitric oxide (NO), with which it shares many features and correlations. One of the most important features of COX-2 is its close association with a variety of inflammatory mediators and its consequent description as the COX isoform involved in and responsible for many of the signs of inflammation. It is also generally accepted that COX-1 is the constitutive isoform involved in the physiological actions of prostaglandins (PGs) in the stomach and kidney, the inhibition of which leads to gastric ulceration and nephropathy as side effects of anti-inflammatory therapy with nonsteroidal anti-inflammatory drugs (NSAIDs).
The initial findings and many of the subsequent developments are based on the techniques of molecular biology 3'4 and are often expressed in terms unfamiliar to many researchers already established in inflammation. Our purpose in this review is to summarize the progress made so far in characterizing the regulation of COX-2, to evaluate its role in inflammation and, as a consequence, to assess the utility of the selective inhibitors of COX-2. In order to establish the appropriate context for the analysis of regulatory mechanisms, we shall (C) 1996 Rapid Science Publishers first consider the molecular biochemistry of COX-2 and its possible place in physiology. We shall also refer to work on COX-1 where necessary.
Molecular Biochemistry of COX-2 Although the early work on COX-2 utilized animal sources, information relating to the human form of this enzyme is steadily accumulating. Since the practical outcome of COX-2 research would be the more efficient alleviation of human inflammatory conditions, emphasis will be placed on results obtained with the human protein, along with data from animal sources wherever relevant. Here, for clarity and simplicity, the molecular biochemistry of COX-2 will be considered at three separate levelsmits DNA, its RNA and the enzyme protein; further details of the molecular biolog.y 4 of COX-2 are available in two recent reviews. 3' DNA The gene for COX-2 is located on chromosome 1 in both human and mouse cells. [5][6][7][8] The small size of the COX-2 gene (7.5-9kbp6'7'9) is compatible with its inclusion in the group of inducible, immediate early lg0enes, few of which are over 10 kbp in length. It is relevant here to note that the human gene for TXA2 synthase is larger (75 kbp 1) and, like that for human PGI2 synthase, is only weakly (two-fold) inducible. 12 '3 The COX-2 gene has ten exons, one less than that for COX-1. 6,v Overall, the descriptions from different groups, of human genomic DNA for COX-2 6'7'9 are in agreement and show many similarities between the human 4 gene and the corresponding murine gene, underlining the close relationship between species. The eDNA for human COX-2 was first derived from HUVEC cells. 5 More recently eDNA prepared from a human line of erythroleukaemia cells (HEL cells7) exhibited virtually an identical sequence with only two nucleotide differences.
In contrast to the similarity shown so far between species in the protein coding and 3'flanking regions of COX-2 DNA, there are important species-related differences in the 5'flanking region of the COX-2 gene, where the promoters and transcription factors bind. Whereas in the human gene there are putative binding sites for a variety of transcription factors including AP-2, SP-1, NF-IL6, NFkB and a cAMP responsive element (CRE) along with a TATA box and a TPA-response element in the first intron, 6'7'9'6'7 the corresponding region of the mouse gene appears to lack a CRE, 4 NFkB or NF-IL6 site, 7 although the others are present.
The rat gene which has over 80% identity with the mouse gene in this region also lacks CRE, NFkB or AP-2 sites and a TATA box but includes a site for NF-IL6. &19 However more recent analysis of the mouse gene for COX-2 in an osteoblastic cell line has found an action of and sites for, NFkB; the same authors have also 20 identified a NFkB binding site in the rat gene. mRNA Reflecting the similarity in cDNA for COX-2 across species, there is also considerable similarity in mRNA for COX-2, which at 4 kb is almost twice the size of that for COX-1 (2.8 kb) in all 21 species so far examined. Since the enzyme proteins (COX-1 and COX-2) are very similar in size, just over 600 amino acid residues, most of the difference in mRNA for COX-1 and COX-2 is taken up by the extensive 3'-untranslated region in COX-2 mRNA. This region includes several copies of the Shaw-Kamen 'instability' sequence, the actual number varying 2 between species from 14 to 18 in animals. 2 In two examples of human mRNA, 17 and 22 copies were found.7'9 These sequences are characteristic of rapidly degraded RNA 7'9 and have been found in the mRNA for other immediate early proteins. 2 However such sequences do not occur in the mRNA for COX-1 in any species.
Estimates of the half-life of COX-2 mRNA vary with the cells studied and with the stimuli used for induction of the protein. In an endothelium- 24 derived cell line (ECV304) with IL-1 as the 306 Mediators of Inflammation Vol 5 1996 inducing agent, COX-2 mRNA had a half-life of about 1 h. In the same system with transcription blocked, IL-1 was able to prolong the halflife of existing COX-2 mRNA to about 90 min, thus contributing also at a post-transcriptional stage to the overall induction of the enzyme. In an epithelial cell line (EGV6), COX-2 mRNA induced by the phorbol ester, TPA, had a halflife of 30 min. 25-It appears that the 'built-in' instability of the COX-2 mRNA is an essential component of the regulation of this protein and hence of its activity.
Protein structure and function The COX proteins are very similar, both between species and between isoforms, as they both carry out the same two separate catalytic functions, oxidation of arachidonate to PGG2 and reduction of peroxide, specifically that of 4 26,-27 PGG2 to PGH2.
The differences in protein structure are small and chiefly outside what is considered to be the catalytically active site. 4 '26 The two isoforms are almost identical in size, COX-1 is about 602 residues whereas COX-2 comprises 604 residues. The major differences in sequence are at the N terminal where COX-2 has 17 less amino acids in the signal peptide and at the C-terminal where COX-2 has 18 more residues than COX-1. 4 '26 The central parts of the proteins where the catalytic and substrate binding sites are located, are almost identical. The tyrosyl groups crucial for the oxidation and the histidines interacting with the haem group are all highly conserved as is the serine acetylated by aspirin.
(i) Substrate binding sites There are important functional differences between the isoforms which suggest that the active site in COX-2 is larger or has a looser fit than that in COX-1. This has been deduced from various mutations at the serine residue, which is acetylated by aspirin in either enzyme, Ser 28.29 530 in COX-1 or Ser 516 in COX-2.
(The different numbers for similarly placed residues in the two isoforms is due to the longer N terminal sequence in COX-l; the numbering for COX-1 is thus about 14 in advance of that for COX-2.) Mutation of serine to alanine in either isoform altered neither Km nor PG production but did confer protection against the irreversible inhibition caused by aspirin, since alanine cannot be acetylated. 29 However, mutation of serine to asparagine (isosteric with acetylated serine) has a strikingly differential effect; the COX-1 mutant lost cyclo-oxygenase activity whereas the COX-2 mutant retained full activity and an unchanged Km. Substitution with a larger not important in binding of substrate as marked amino acid, glutamine, abolished cyclo-oxygemutations at this site (Val to Lys or Glu) did not nase activity in both isoforms. 28'29 materially alter Km for AA of the COX-2 Another indication of the larger active site in protein. 33 COX-2 may be drawn from the effects of aspirin An attempt to exploit the selectivity sugon catalytic activity. This compound irreversibly gested by the larger substrate binding site in inhibits the production of PGs by COX-1 or COX-2 had an unexpected outcome. It was COX-2 through the acetylation of Ser 530 or Ser argued that, as the acetylation of COX-2 still 516. Nevertheless, acetylated COX-2 but not allows binding of AA to give 15-HETE, acylation acetylated COX-l, still oxidized arachidonic acid with a larger acyl group should prevent any (AA) to an alternative product, 15-HETE.  oxidation of AA by encroaching further into the This finding would suggest that there is space binding area. In the event, the most potent for AA to bind to acetylated COX-2 close analogue of aspirin was valeryl salicylic acid but enough to the active site for oxidation to occur it was a selective inhibitor of COX-1 with no even though the orientation is not adequate for inhibition of COX-2. 34 The explanation for this the full cyclo-oxygenase reaction to take place, result is still to be put forward. Further support comes from the effect of It is important to note that in the COX mutant another substitution of this serine in COX-2, proteins and in the acetylated native COX, the with methionine; this leads to a 'pseudo-acetyperoxidase activity catalysed by an active site lated' form in that the mutant protein shows on the other side of the haem group from that increased production of 15 were compatible with the model of COX-1 Because COX-1 and COX-2 are catalytically and structure derived from X-ray crystallographic structurally almost identical, it is likely that if analysis. 1'32 The Ser 530 lies halfway along a selective inhibitors of COX-2 are to be found tunnel leading up to the active site and it is then these would not bind to any catalytically relatively easy to imagine how the acetylation of relevant site which would be the same for both Ser 530 would block access of substrates to the isoforms, but to some other region possibly active site at the head of the tunnel. On the unique to COX-2. However access to substrate basis of the biochemical results for COX-2, one must still be denied in order to inhibit COX-2 would assume that Ser 516 either lies further activity. This line of reasoning would explain within the wall of the tunnel or that in COX-2 why most of the selective COX-2 inhibitors so there is enough room for AA to 'squeeze past' far disclosed are not carboxylic acids (as are the acetyl group and to bind close enough to most COX-1 inhibitors)but interestingly contain the tyrosine-haem complex to allow oxidation a different common grouping, the sulphonto 15-HETE, but not to tare up the configura-amide or sulphone group. tion which leads to the cyclic endoperoxide Support for this suggestion for different bind-(PGG2).
ing sites for the two types of inhibitors comes Another possible factor is the residue on the from another mutant of COX-1 in which Arg opposite side of the tunnel, Ile523 in COX-1 120 was replaced with Glu. 37'38 This positively which in COX-2 is substituted by a1509, one charged residue (Arg 120) is located at the methylene group smaller than lie. This location opening of the active site tunnel 1 and was also provides the only difference between the assumed to be the binding site for the carbactive sites of the two isoforms. It is therefore oxylic acid group in the substrate fatty acids. It possible that the extra methylene group of lie was also assumed to provide a binding site for 523 creates enough of a narrowing of the the 'old', carboxylic acid, non-selective, COX tunnel so that in combination with the acetyl inhibitors. These assumptions were fully congroup on Ser 530, access to the active site in firmed by the characteristics of the (Arg 120-COX-1 is essentially prevented. In COX-2, the Glu) mutant COX-l, which showed a 100-fold presence of Yal 509 would not only allow a higher Km for AA and a much reduced susceptwider range of fatty acid configurations to gain ibility to the carboxylic acid inhibitors including access as substrates in the normal enzyme but indomethacin, flurbiprofen and diclofenac all of also provide less of a 'choke point' in the which did not inhibit the mutant enzyme at acetylated enzyme. Nevertheless, Val 509 was concentrations between 200-and 8000-fold higher than their IC50 values for the wild-type 37 However two COX-2 selective corn-enzyme. pounds, the sulphone DuP 697 and a sulphonamide analogue, were more potent as inhibitors of the mutant enzyme. Although this increased potency is probably due to the decreased binding of substrate AA (DuP 697 is a competitive reversible inhibitor of the wild-type COX-1 enzyme), there was clearly no decrease in the efficacy of the sulphonamide compounds and hence no loss of their binding to the mutant protein. In the other report, 38 the Arg 120-Glu mutant of ovine COX-1 showed no COX activity at all. Flurbiprofen binding was thus assessed with the Arg 120-Gln mutant which had 5% of the wild-type activity. This mutant did not show time-dependent inhibition and flurbiprofen's IC50 value had increased from 5 bM (wild type) to 1 mM in the mutant. These results are probably adequate evidence for the importance of the Arg 120 residue in the binding of the 'old' non-selective inhibitors but the binding sites for the COX-2 selective inhibitors still remain to be determined, although two recent reports 33'39 provide some clues to its location.
In both, mutants of COX-2 have been generated with Val 509 being changed to lie, as in COX-l, and, in the mutants, selective COX-2 inhibitors were much less potent and less capable of causing the time-dependent inhibition of COX-2, characteristic of wild-type COX-2. However in one report, 33 the COX-2 selective inhibitors (nimesulide, DuP697, NS398, SC58125) were still able to bind and inhibit presumably on a 'reversible' basis. The 'old' NSAIDs were unaffected by this mutation. 39 Here the Val/Ile substitution appears to be crucial in determining the activity and possibly selectivity of inhibitors of COX-2. The possible effects of this substitution have already been discussed above in terms of substrate access but it is less easy to visualize the crucial influence of a methylene group in the binding of strongly polar sulphone/sulphonamide compounds such as the selective COX-2 inhibitors. The crystal structure of human recombinant COX-2 was described at a recent meeting 4 and although the full report is not yet avai{able some details relevant to this point have emerged. As expected the crystal structure of COX-2 is almost identical to that of COX-1. However for cox-2, two conformations appear to be possible for 4o the binding of inhibitors, one in which inhibitor binds to both Arg 120 and Tyr 355 (see Ref. 38) as polar sites (the closed conformer) and the other (open conformer) in which inhibitor binding excludes Arg 120. Since mutants of Arg 120 do not change binding of COX-2 selective inhibitors, the open conformer would appear to be the most liRely form of COX-2 bound to a selective inhibitor. However these changes are at the mouth of the substrate tunnel and involve polar residues not the nonpolar Val 509 which has such striking effects on inhibitor efficacy. It may thus be necessary to re-assess the role of another polar residue, Glu 524 (in COX-l). This negatively charged amino acid is close enough in the crystal 3'4 to form a salt bridge with the positively charged Arg 120. Although Glu 524 was not important for en-. zymic activity in COX-l, 38 it, along with Tyr 335 could provide polar binding sites in COX-2, alternative to Arg 120. This residue Glu 524/510 is also immediately adjacent to the lie 523/Val 509 and it may be that the crucial effects of the Ile/Val substitution on inhibitor binding are actually to alter the configuration of the next residue, Glu 524/510. Clearly we need more information before the binding site for COX-2 selective inhibitors and the nature of the conformational change associated with their action are fully elucidated. (See Note Added in Proof). The largest difference between the two proteins is the C-terminal extension (18 amino acids) and the lack of the 17 residues at the Nterminus in COX-2. Although the C-terminus in COX-13 and probably in COX-2, is also distant from the active site, it clearly exerts a considerable influence on the function of the enzyme as site-directed mutations at the extreme C-terminal of recombinant COX-1 had unexpectedly strong effects on activity. 4 Frustratingly, in COX-1 crystals, the X-ray analysis appears to extend reliably only to residue 586 of the 600 total residues in the ovine protein. 3 The threedimensional structure of this C-terminal extension in COX-2 has also not been reported 4 although in the full report more information on this region may be presented. Mutants of COX-2 with alterations in this region will be invaluable in this analysis. Another feature of inhibitor interaction with the COX proteins is the time-dependent irreversibility of some compounds (apart from aspirin). This varies between inhibitors and isoforms and is discussed further below but one aspect is relevant here. An inference from time-dependent and irreversible inhibition is that there may be chemical reaction between inhibitor and enzyme, for instance the formation of a 42 Schiff base.
These interactions and subsequent conformational changes in protein 43 re different structure a from acetylation by aspirin as all oxidation of AA is prevented, i.e. there is no formation of 15-HETE as in acety-lated COX-2. [28][29][30] The selective, time-dependent receptor family of proteins whose endogenous inhibition of COX-2 42-45 would imply that conligands are still undefined, although their activformational changes in COX-1 derive from interity as transcription factors is well established. 53 actions different from those that inactivate PGD2, PGJ2 and other related cyclopentenone COX-2. It is important to note that neither PGs bind to PPAR, and cause the protein to act reversible nor irreversible inhibitors affect the as a transcription factor for reporter genes 54-56 peroxidase activity of the proteins which conand also to stimulate the differentiation of tinue to function normally in the presence of fibroblasts into adipocytes. 54'55 These PGs may the inhibitors. 29'35'36 Another important conse-be the endogenous ligands for the PPARs, quence of the conformational rearrangement of which might also explain the effects on cell the enzyme protein with inhibitor would be growth already described for these prostthat the structure of COX-1 crystallized with anoids. [57][58][59] An intriguing finding is the stimulainhibitor 31 could be different from that of tion of transcription of the haem oxygenase enzyme crystallized with substrate, as already gene in rat cells by PGJ2 derivatives 6'61 in recognized. 46'47 relation to the possible role of haem oxygenase 62 in the inflammatory response. (iii) Intracellular location and function The extent to which COX-2 accumulates in The C-terminal region may also be especially this location rather than in the ER could provide important in securing the protein to the endoa control mechanism for this growth regulatory plasmic membranes. For both COX-1 and COXfunction, additional to any control of catalytic 2, the most commonly suggested location is in activity by inhibitors. The additional amino-acid the membrane of the endoplasmic reticulum residues in the C-terminus of COX-2 could be (ER) with an additional locus for COX-2 on the critical in the location of the enzyme in the nuclear membrane.  Results from the crysnuclear membrane and spontaneous mutations tallographic analysis suggested that the enzyme in this region of the protein might affect the is bound to the ER by a sequence of short distribution between the ER and nuclear sites. helical stretches of the molecule which prob-More COX-2 in the nuclear membrane could, ably only interact with one-half of the lipid for instance, be associated with decreased apopbilayer, i.e. there is no transmembrane tosis in epithelial cell lines. 63 structure. 3 This description may now need to be modified. Ren et al. 5

using antibodies
Physiological and specific for particular sequences in COX- 1 Pathophysiological Roles for protein (including the C-terminus, the active COX-2 site and glycosylation sites), together with selective lysis of cellular membranes have sug-The accepted role for COX-2 is to provide PGs gested that the C-terminus may cross the ER in a range of inflammatory and host defence membrane to the cytoplasm, while agreeing conditions, which could be termed a 'pathologiwith a luminal location for the rest of the cal' role. Other functions such as involvement molecule. Such a model would be compatible in mitogenesis and reproduction inferred from 38 1957 with the crystallographic analysis where the Cexperimental results' could be considterminus beyond Arg 586 was not resolved and ered as physiological or pathophysiological.
other results attributing importance to amino These other roles have been supported by the acid residues distant from the active site. 4]'52 results from 'knock-out' animals in which a Although analogous mutations in COX-2 have gene coding for a particular protein is selecnot yet been assessed, these findings re-emphatively inactivated. The logical purpose of knocksize the importance of the structural analysis of out strains is to expose deficiencies in the the C-terminal region of the proteins, knock-out animal and from these defects to The intracellular location of the COX protein deduce the roles played by the missing protein. is important because only COX-2 is found on In the present context the relevant proteins are the nuclear membrane where it would be COX-1 and COX-2. ideally positioned to participate in mitogenesis, normal or neoplastic. Inflammation A possible mechanism for this participation is provided by recent findings, in three different Two separate reports 64'65 of COX-2 knock-out systems, of the activation by PGJ2 derivatives of mouse strains (null mice) have appeared and in the peroxisome proliferator activated receptor each a model of acute inflammation (oedema type gamma (PPARy). The PPARs (cz, and y) following AA applied to the ear) was used to are members of the steroid hormone nuclear test responses in the null mice. Both agreed that the null mice had normal ear oedema in response to AA and the interpretation of this result was that COX-l, still present, was able to generate the PGs required. Two other models of acute inflammation (PMA-ear oedema and carrageenin paw oedema) also gave normal results in null mice. The only model tested which failed in COX-2 null mice was one of LPS-induced hepatotoxicity which depends on induction of COX-2 in macrophages and/or Kupffer cells. 4 The implications of these experiments in mice lacking COX-2 are that where COX-1 is normally present (ear skin, paws) this isoform will substitute for the missing COX-2; where COX-1 is absent or at very low levels, as in macrophages, the inducing agent fails to generate the usual response.

Reproduction and development
An essential role for COX-2 in embryonic development would be deduced from the severe morphological defects and consequent functional failures in the kidneys of COX-2 null mice, 4'5 which lead to their early death (some at about 8 weeks and most by 6 months). A similarly absolute requirement for COX-2 in female fertility must be deduced from the failure of null female mice to ovulate. 4 It is important to realize that these failures in null mice must represent a highly localized generation of COX products as the nature of the products from COX-2 is the same as those derived from COX-1 and COX-1 is still fully active in the COX-2 null mice.
A strong linkage between isoform and physiological function was further supported by the 66 results from COX-1 knock-out mice. Here the null mice are healthy, without developmental defects and with no ovulatory changes. However null pups carried by a null mother were mostly (> 90%) born dead; either a heterozygous mother or some heterozygous pups restored viability of the whole litter to normal. Clearly the COX-2 prese.nt in the COX-1 null mice cannot provide the PGs needed for foetal survival but COX-1 present in some littermates will ensure survival of all foetuses. Again the separate events in reproduction appear to have differing absolute requirements for COX isoforms.
Implications of results from knockout mice The possibility of compensatory changes suggested for inflammation in COX-2 knock-out mice could also explain why in COX-1 null mice there was no gastric ulceration or NSAID-type nephropathy. Both these effects would have been predicted from the known effects of aspirin and other NSAIDs in normal animals. However in COX-1 null mice the compensation was not due to increased amounts of COX-2 activity as gastric tissue from the null mice synthesized less than 1% of the normal amount of PGs. 66 The most likely alternative compensation would be from NO; this mediator is, like PGI2, a vasodilator and ulceration is associated with vasconstriction in the gastric microcirculation. 67 Like COX, both constitutive and induced NOS can be expressed in endothelium. One predictable consequence of this alternative pathway would be that NOS inhibitors would be ulcerogenic in the COX-1 null mice, whereas they are not noticeably so in normal animals. 67 If the knock-out mice show that COX-1 and COX-2 have separate and important physiological roles in reproduction and COX-2, unlike COX-I, plays an essential part in foetal development, what do the knock-out mice tell us about their pathological importance in inflammation? One conclusion is that there is a clear need for COX-2 in certain forms of inflammation, perhaps all those related to the actions of LPS. 64 However the logical deduction from the other results is that COX-2 is not relevant to the inflammatory models used for many years to screen for anti-inflammatory compounds. On the basis of the knock-out results alone, several major pharmaceutical companies are wasting time, effort and money in searching for selective COX-2 inhibitors; these compounds will not decrease inflammation nor will they affect the incidence of NSAID-induced gastric ulcers as that effect is not connected with the presence or absence of COX-1 activity. In contrast to these logical deductions, there is a considerable body of empirical experimental evidence clearly demonstrating both the anti-inflammatory efficacy of selective COX-2 inhibitors and their decreased ulcerogenicity when compared with the more COX-1 selective inhibitors. This paradox is not without hope of resolution; there are still many relevant measurements to be made in the null strainsmlevels of COX-1 or COX-2 activity, of PEA2 activity, the effect of selective inhibitors and many othersmand when these results are gathered, another synthesis of the apparently opposing views may be possible.

Regulation of COX-2 Activity
It is now clear that the activity of COX-2, as expressed by the synthesis of PGs, is normally controlled through the synthesis of the protein.
Several of the transcription factors effective on the COX-2 gene are known to be stimulated by inflammatory cytokines. Other cytokines and corticosteroids can alter the half-life of the inherently unstable mRNA, either increasing or decreasing translation into protein. For the protein itself, one option is to control the provision of substrate arachidonic acid (AA), although the most obvious regulator of activity would be a selective COX-1 or COX-2 enzyme inhibitor. All these possibilities, considered in more detail below, are summarized in Fig. 1.

Regulation by inflammatory factors
A great variety of agents, mostly derived from inflammatory situations, have been used to induce COX-2 activity. The most frequently used inducing stimuli are IL-1 and lipopolysaccharide (LPS; used here as a synonym for bacterial endotoxin) and not as might have been expected the inflammatory stimuli often used for in vivo models, carrageenin64'68-71 or 72 These and other stimuli used with zymosan.
human cells are listed in Table 1. in this scheme for completeness but are not considered any further in the text as a means of regulating COX-2 levels. Note that factors affecting phospholipase induction or action could independently or co-operatively influence PG synthesis by altering the amounts of endogenous AA available to COX-2. The levels of PG production in vivo under physiological or pathological conditions will reflect the sum of changes in COX-2 and phospholipase activities. Intervention with COX-2 inhibitors allows exogenous control, overriding the endogenous mechanisms. The references given are restricted to cells and tissues of human origin and to papers published in 1994, 1995 and early 1996. These are given as a guide to the range of systems (stimuli, cells, tissues or in vivo) used and not as a complete list of all the work on COX-2.
Differences between the effects of LPS and IL-1 in cultured cells will clearly depend on the ability of the cell line to release IL-1 in response to LPS; such differences are less likely in vivo or ex vivo where a wide range of cell types have been exposed to the inducing agent. However as at least two cytokines, IL-10 and IL-4, decrease induction (see below), their synthesis in vivo following IL-1 or LPS treatment could modify the final level of COX-2 activity attained. In endothelial cells COX-2 is induced readily by LPS, in some cases through the release of 73 76 TNF, PDGF and other cytokines.
However the initial stage of this process, the binding of LPS to the cell membrane, is still unclear as endothelial cells do not express the particular LPS-binding membrane proteins (CD 14) that may be used as receptors on leukocytes,77-v9 a major cell type responding to LPS with induction of COX-2. There are some indications that soluble forms of CD 14 are involved in the mediation of responses to LPS in endothelial ceils. 77,80 Whereas most cytokines so far studied increase induction of COX-2, there are examples of inhibition by cytokines. Two interleukins, IL-10 and IL-4, already known to antagonize other effects of 'pro-inflammatory' cytokines, 81'82 decreased COX-2 levels in monocytes stimulated by LPS or Con A, 79'83'84 but in mast cells, 85'86 IL-IO potentiated, whereas IL-4 still inhibited, induction of COX-2 by c-kit ligand and IL-1. It is possible that this discrepancy in the effects of IL-10 is related to the cell types involved; more studies would be needed to define such a selectivity.
The TGF proteins, TGF and TGF, present conflicting results for analysis. As might be expected from its mitogenic activity, TGFx was able to stimulate COX-2 production in epithelia 87 and to increase PGE2 output in human amnion cells and in osteoblasts. 88'89 TGF synergized with IL-1 or TNFc 9'91 to increase PG output, perhaps because TGFc can also induce IL-1 receptors. 91 Although TGF potentiated the induction of COX-2 caused by phorbol ester in fibroblasts, 92'93 it had no effect when given alone. In macrophages, 94 the same cytokine inhibited the induction of COX-2 by LPS, more in keeping with its general anti-inflammatory profile. The apparent divergence of the effects of TGF[ may be more readily rationalized on the basis of cell types; for an overall anti-inflammatory and wound-healing effect, it would be reasonable to de-activate leukocytes and to stimulate fibroblasts at the same time. TGF also affects NOS induction in a range of cells 95  PGE2 also stimulated COX-2 expression in mouse macrophages treated with LPS. 1 In these cells and in two other cell systems, rat epithelial cells stimulated with TGF and phorbol ester and in human PMNs stimulated with LPS, COX inhibitors (indomethacin or sulindac sulphide) decreased COX-2 induction. 87 '99 The positive feedback implied in these systems contrasts sharply with the negative feedback loops more usually found in inflammatory conditions where PGE2 and other agonists that raise cAMP, actually decrease cytokine secretion from 102 104 macrophages and lymphocytes.
Indeed the breaking of this loop with COX inhibitors (NSAIDs) in chronic inflammation is believed to increase cytokine production and subsequent degradation of joint cartilage while relieving pain and oedema; thus symptomatic benefit is undermined by a continuing or even acceler- 105,106 ated disease process.
(iii) Free radicals and nitric oxide Another totally different class of compounds associated with inflammatory situations are the reactive oxygen intermediates or oxygen derived free radicals (ODFR), such as superoxide anion (O) and the hydroxyl radical (OH-). A role for these species in the induction of COX-2 has also been proposed. 17 '18 These highly reactive chemical species are present at sites of inflammation, leukocytes generate ODFR in the phagocytic process (the oxidative burst) and NO is another free radical, present during inflammation, synthesized by iNOS and capable of interacting with the ODFR to increase free radical actions. Thus, there are many opportunities for interactions between these radicals and COX-2. In rat mesangial cells following LPS, IL-1 or TNFct, radical scavengers (thioureas) or other antioxidants inhibited the usual increase in COX-2 mRNA, COX-2 protein and PGE2 synthesis. 17 The effects were selective in that induction of other proteins (chemokines) by LPS or of COX-2 by other stimuli (phorbol esters, serum) were not similarly inhibited. At present, the actual mechanism of the interaction between the ODFR and the transcription of the COX-2 gene is not fully known, although ODFR are known to activate NFkB, 19'11 one of the transcription factors active on the COX-2 gene. However, in the rat the COX-2 gene may not exhibit a binding site for this transcription factor. 19,21 Another free radical associated with inflammation is nitric oxide (NO) and its effects on COX activity remain difficult to summarize briefly. Both enzymes, COX and NO synthase (NOS), have inducible isoforms, which are induced in inflammatory situations by the same cytokines and both genes belong to the family of immediate early genes. Thus interaction between these two enzyme systems is highly likely; the difficulty arises from the outcome of that interaction. There was potentiation of PG output in the presence of NO 11-115 apparently by direct interaction of NO with the haem in COX protein,116 although this interpretation has been questioned.liT,118 The opposite effect, inhibition of PG output .b. NO, has also been observed in macrophages. Action of PGs on NO output is less frequently encountered 12'121 although such effects might be deduced from the interactions of raised cAMP with NO output. 122-124 In one case a truly reciprocal relationship, with PGs inhibiting NO output and NO inhibiting PG output, by which a constant level of vasodilator tone may be maintained has been described in human saphenous vein. 125 This last example is unlikely to involve induced forms of either enzymes and is perhaps less relevant to the present discussion but it does illustrate the range of interactions possible between these enzymes. The most valid conclusion would seem to be that there is no generalization and that each situation with its particular combination of species, cell type and stimulus has to be evaluated individually.
(iv) Contribution of phospholipases A potential source of confusion in the interpretation of experiments involving COX-2 induction is the simultaneous induction of PEA2 activit most frequently with IL-1 as inducing agent. 6,126-132 Increased PEA2 activity implies increased provision of free AA which could be as easily available to COX-1 as to COX-2. Thus the final effect, increased synthesis of PGs, is not necessarily due solely to increased COX-2. One prediction from this co-induction is that the effectiveness of inhibitors of COX-2 could depend on the level of PEA2 induced at the same time and that level could vary with the nature of the major inducing agent (IL-1, PAF or TNF) and with the cell type involved. The localization of the human gene for cytosolic PEA2 to the same chromosomal region as the gene for COX-2 (lq25), raises the possibili_ty of coordinate regulation of these enzymes. 133'13 (v) Effects of corticosteroids The major inhibitory mechanism affecting COX-2 induction, both in terms of experimental results and of pathophysiological relevance is the action of corticosteroids, most frequently demonstrated with dexamethasone (Dex). In many conditions, susceptibility of the production of PGs, enzyme protein or mRNA to inhibition by Dex is used as clear evidence for COX-2 induction. This simple conclusion is however clouded by other actions of Dex including the inhibition of the induction of PEA2132 and inhibition of PEA2 activity itself via lipocortin. 135'136 For instance over 20 years ago it was shown that the output of PGs from freshly isolated lungs from untreated guineapigs, conditions in which COX-2 induction should be minimal, was decreased by infusions of Dex. 17 There is nevertheless clear evidence for decreased induction of both COX-2 protein and mRNA in the presence of Dex. 25' 93'13g-141 There are also a few examples of corticosteroids increasing PG production. The mRNA for COX-2 but not that for COX-1 was increased 20-fold in human amnion cell cultures exposed to Dex for 16 h. 142 Interestingly, oestradiol and progesterone did not increase COX-2 mRNA but cortisol did. More typically, in human decidual tissue, Dex and progesterone were inhibitory.143 As with other examples of corticosteroid inhibition of mRNA, the mechanism of action of Dex on COX-2 induction is poorly elucidated; either increased degradation of an already short-lived mRNA8, 44 or decreased transcription are possi-Regulation by substrate ble contributors to the overall affect.
One of the continuing controversies in the analysis of COX action centres on the influence Regulation via intracellular signalling of different sources of substrate on activity.
Tyrosine kinases play important roles in the There is a clear difference between the utilizaintracellular signalling pathways for COX-2 in-tion of endogenous and exogenous AA in that duction in a range of cells; in endothelial cells, the supply of endogenous substrate is conepithelial cells and macrophages this was trolled by enzymes outside the COX cascade, demonstrated with tyrosine kinase in-chiefly by the balance between phospholipid hibitors 7'74 '45-48 and in mesangial cells by hydrolysis and reor trans-acylation. 161 '162 The measurement of protein tyrosine phos-levels of endogenous substrate can be increased phorylation. 72,[149][150][151] The oncogene v-src en-and PG output consequently stimulated in a coding a tyrosine kinase is by itself enough to number of cells and tissues (without induction cause COX-2 induction in T cells. 44 Whereas of COX-2) by agents such as thrombin, hista- 163 the induction by EGF 25 probably utilized the mine, brad kinin, PAF or the crosslinking of Y 164 receptor tyrosine kinase, the kinases used by IgE receptors. Ths stimulation s charactensother inducing agents have not yet been clearly tically of short duration, less than 30 min and identified. However the inhibition by tyrosine there is no doubt that endogenous substrate is kinase inhibitors such as erbstatin and genistein utilized by COX-1 to increase PG synthesis. The 73 145 146 152 of both COX-2 and iNOS induction experiments in which COX-2 utilizes endogenoffers a new mode of anti-inflammatory action, ous substrate are characteristically of longer which would have the advantage of not acting duration (6-24 h) and entail incubation of cells on constitutive enzyme activities. Such an effect with an inducing agent such as LPS, TPA or may contribute to the anti-inflammatory proper-PDGF which is present throughout the incubaties of leflunomide 5 which was shown to tion. The PGs accumulated over these longer inhibit the EGF-stimulated tyrosine kinase, 154 periods are increased many-fold in the presence the Src tyrosine kinases 55 and the synthesis of of the inducer molecule, relative to those in PGE2 induced by LPS in human leukocytes. 99 control incubations. Most of the inducing agents Another signalling pathway, via protein kinase will increase phospholipase action as well as C (PKC), also appears to be involved but here inducing COX-2 protein. 126'128-'165 Again the net effects of stimulating PKC activity are there is no doubt that COX-2 can utilize more variable. In most cases, increased PKC endogenous substrate to form PGs. activity was associated with induction of COX- The problems appear with the use of exogen- 87 147 59 Another confusing factor is increased output of PGs. However in some the well-established down-regulation of PKC on preparations where both COX-1 and COX-2 are continued stimulation by phorbol esters. In the present, the induced enzyme appears to conpresent context this was illustrated by the tribute no additional amount of PGs over that stimulation of COX-2 mRNA by 5HT (mediated seen with COX-1 alone, leading to the proposivia 5HT2 receptors and PKC activation)or by tion that COX-2 does not utilize exogenous short exposure to phorbol ester. However pre-aa. 92 49 In human skin fibroblasts for short times, typically 10 min, compared with after IL-1 stimulation, PKC appeared to be the 6 h incubations to show COX-2 induction. 92 major protein kinase with only minor contribu-Within 10 min over 100 ng PGE2/ml was genertions from PKA or tyrosine kinases. 16 However ated, whereas over 6 h there was accumulation in ovarian tissue both PKA and PKC were of only about 20ng/ml. An equally plausible activated during LH and GnRH stimulation of explanation would be that although COX-2 was COX-2. 56 Other intracellular second messen-induced several-fold, the actual amounts of gers identified in osteoblasts include cAMP and COX-2 protein were still low relative to the PLC, stimulated by PGE2 or iloprost (a stable amount of COX-1 present. It is thus not possible analogue ofPGI2) and PGF2a respectively. 96 to come to a definite conclusion on the selective accessibility of exogenous substrate to either isoform on the basis of these experiments.
A variation of this hypothesis was the conclusion drawn from work with only endogenous substrate in mast cells. 164 Here release of PGD2 and COX-2 protein were induced by incubating the mast cells with a mixture of cytokines. Incubation in this medium for 5-10 h increased PGD2 output many-fold to about 5-10 ng/ml. 'Acute' stimulation of PGD2 biosynthesis by the crosslinking of IgE receptors in these cells was not increased until after 24 h incubation with the cytokine mixture, by which time the COX-2 protein had fallen again to near normal levels. There was certainly no increase in the 'acute' release of PGD2 at the time of peak COX-2 induction. From this the authors concluded that the induced COX-2 did not have access to the increased AA released during IgE stimulated PGD2 synthesis and that only COX-1 was utilized, even when COX-2 had been induced, to form PGs subsequent to IgE stimulation. A comparison of the times over which these experiments were performed shows that the IgE stimulation provided about 5 ng PGD2/IO 6 cells in 10 min whereas the cytokine stimulation took 10 h to provide 10 ng/10 6 cells. On this basis the induced COX-2 would provide about 0.2 ng in 10 min; this contribution would be rather difficult to detect against the total of 5 ng produced. The authors may be correct in postulating different coupling of stimuli to the isoforms but the experimental results are not an adequate test of their hypothesis. Indeed in a more recent paper from this group, the 'stimulus selective' linkage of COX-1 or COX-2 action has been replaced by a 'time selective' hypothesis. 168 If distinct pools of substrate for COX-1 and COX-2 do exist they are more likely to be defined on a spatial basis than on a simple endogenous/exogenous substrate criterion. Indeed if COX-2 shares the general three-dimensional structure proposed for COX-l, 31 with the membrane anchors defining the entrance to the active site tunnel which guides AA cleaved from the adjacent membrane up to the oxidative site of the enzyme, it is not immediately obvious how one isoform could favour exogenous AA over the freshly hydrolysed product of the underlying ER membrane. However this model of COX action would assume a phospholipase in close proximity to the COX protein and a more likely basis of selectivity is in the phospholipase activated to supply endogenous AA. Although most emphasis has been placed on PEA2 in this context, the action of either PLC or PLD can also give rise to free AA. Different ligands for cell membrane receptors will activate these phospholipases differentially. Furthermore, each phospholipase has its own substrate selectivities and the distribution of the phospholipids is not uniform throughout all membranes. 162 There is also evidence for a selective locus of COX-2 in the nuclear membrane, apart from the location in the ER demonstrable for both isoforms. 5 Thus a combination of which phospholipase is activated by which stimulus and which phospholipid is closest to a particular COX protein might appear to give a degree of selectivity between isoforms in terms of the substrate.
Selective Inhibition of COX-2 One disadvantage of regulating COX-2 by interference with the processes of transcription, translation or intracellular signalling pathways is that, at present, selectivity of effect is low. This is well recognized for the corticosteroids which will prevent induction of many proteins apart from COX-2 and could be equally true for antagonists of or interference with, transcription factors such as NFkB or NF-IL6169 or for the inhibitors of tyrosine kinase. Logically the most selective effect would be attained by inhibition of the enzyme protein and this consideration coupled with the effectiveness of COX inhibitors already known has led to an extensive search for new inhibitors with a selective action on COX-2. Particularly, such selective agents should be free of the most significant side effects associated with COX-1 inhibition, gastric ulcers and nephropathy.

Assessment of selectivity
The initial stages of this search were concerned with assessment of the selectivity of the known NSAIDs or COX inhibitors and very soon established one major difficulty in the analysis, a marked variation in the selectivity ratio (IC50 for COX-I:IC50 for COX-2) for any given compound; a high ratio representing selectivity for COX-2 inhibition. For instance, ratios for indomethacin ranged from 20170 to 0.1.171 This variability is due to variation, at every level, in the experimental conditions of the assays. Different types of cell are used, derived from different species, as whole cells, homogenates, purified extracts or recombinant proteins expressed in bacterial, insect or animal cells. Further variation is introduced in the time of pre-incubation with inhibitor, the concentration of exogenous substrate or the use of endogenous substrate. The last three factors contribute significantly to the trations of AA could thus be higher, providing marked differences in selectivity. For instance more competition for the binding of inhibitor. with 10 min pre-incubation, indomethacin at Even the time dependent irreversible inhibitors 1.6 bM completely abolished COX-2 activity will show some reversal of inhibition when whereas COX-1 activity suffered only 50% in-exposed to high AA concentrations. 42 hibition. However, with no pre-incubation, indo- The value of a model system must lie in its methacin had an IC50 of 13.5 bM for COX-1 ability to predict compounds with selective whereas for COX-2 the IC50 was over COX-2 inhibition in vivo and one test of that 1 000 l,m. 172 Clearly two very different selectiv-value is to assess the compounds already known ity ratios would be calculated from IC50 values to exhibit such activity alongside those NSAIDs obtained under these two experimental condi-with the worst side effect profiles. 75 Thus tions. Time-dependent inhibition of COX-2 but compounds such as NS398, SC58125 and not of COX-1 by NS 398 was the major reason CGP28238 must be clearly separated from for its selectivity with human recombinant NSAIDs such as piroxicam, azapropazone or enzymes. 4' 44 A similar differential time-depen-ketoprofen. Most model systems achieve this dency was reported for CGP28238, another separation and will probably be efficient COX-2 selective inhibitor. 42 This feature is not screens for selective COX-2 inhibitors. However, the sole determinant of selectivity as several it must be remembered that over the last 25 COX-1 inhibitors also show time-dependent years, reliance on in vitro screening with COX inhibition of either isoform. 42'44'45 purified from ram seminal vesicle (now known Another source of variability with important to be almost entirely COX-l) must have led to practical consequences is the nature of the the rejection of many COX-2 selective inhibitors enzyme system used, i.e. in whole cells, homo-before they could be tested in vivo and a genates or purified enzymes. Many groups have similar absolute reliance on a single in vitro used human platelets as a source of COX-1 and screen for COX-2 could lead to similar mistakes. a variety of cells (renal mesangial cells, macrophage cell lines, peripheral blood monocytes) Progress in develomn o selective stimulated with IL-1 or LPS to provide COX-2.

inhiBitors
The IC50 values for indomethacin for whole cell preparations vary but are always lower than There were at least two examples of possible those reported for cell free preparations. 7 This selective COX-2 inhibitors in the literature; both is true also for selective inhibitors; the IC50 for exhibited a good anti-inflammatory effect in CGP28238 against COX-2 was 15 nM in whole vivo with little or no inhibition of the standard cells but 750 nM even after prolonged preincu-COX preparation from ram seminal vesicle bation with purified enzyme. 42 Similarly, in-together with low ulcerogenic activity. The first creased potencies (lower IC50) have been noted of these compounds, nimesulide, was patented for aspirin, ibuprofen and even salicylate in over 20 years ago and has been available latterly whole cells compared with values obtained in in several European countries as an out-ofbroken cell or purified enzymes. 74 There are patent, non-prescriltion analgesic and anti-inno obvious explanations for this phenomenon, flammatory agent. The other, CGP28238, is Preferential concentration of the inhibitor in closely related in structure (see Fig. 2) and was lipid of the ER membrane to give a locally first reported in 1989. 77 A third close relative, higher concentration than in the bulk solution NS398, was described many years later as a would not explain why inhibitory potency is selective inhibitor of COX-2.178 lost in broken cells as crude homogenates after Subsequent development has led to DuP centrifugation would contain enzyme still at-697, 79 SC 58125 8 and L-745,337, 8 with tached to fragments of ER. One possibility is many other similar compounds less extensively that in whole cells the concentration of free AA studied. 42'82-85 The most striking feature of is kept low by restraining phospholipase activ-this new generation of NSAIDs is that none of ity (through low intracellular calcium, for in-them are carboxylic acids, like the 'old' NSAIDs, stance) and increasing re-or trans-acylation into and all have the sulphone or sulphonamide lipid. '2 Thus, the initial binding of the grouping. The simplest (and perhaps simplistic) inhibitor to the enzyme in whole cells takes inference from this is that the selective COX-2 place with little or no competition from the inhibitors bind to a site that is different from substrate AA, allowing a maximal inhibitory that used by the 'old' NSAIDs 7 and, as effect. In a broken cell preparation the calcium proposed earlier, would suggest a structure concentration is much higher than the normal unique to the COX-2 protein which, at first intracellular level, lipase activities and concensight, is most likely to be the 18 Table 2). The sulphonamide grouping is present in all except meloxicam in which it forms part of a cyclic structure. None of these compounds has a carboxylic acid grouping characteristic of the 'older' NSAIDs.
insert at the C-terminus. It is important to note that neither the 'old' nor the new selective COX inhibitors affect the peroxidase activity of the proteins, implying that the binding of COX-2 inhibitors, like that of COX-1 inhibitors, does not disturb the three-dimensional structure of the protein on the 'other' side of the haem ring. 29,35,36 In spite of all the reservations outlined above, the search for selective COX-2 inhibitors has been remarkably successful. To some extent this reflects the efficacy of the exemplar compound, nimesulide, but all the subsequent developments exhibit the predicted properties. They all have good selectivity in vitro with pure enzymes or in cell systems with ratios of IC50 favouring COX-2 in all assays and some values are shown in Table 2. Furthermore they exert anti-inflammatory activity in a range of models, acute and chronic, as well as anti-pyretic and analgesic activities and at these doses there is little or no gastric ulceration. Clearly, it is possible to achieve selective inhibition of COX-2 and now it seems only a matter of refining the effective structures to combine the highest selectivity with the best pharmacokinetics and provide compounds for clinical evaluation. One result of the selectivity of COX inhibition coupled with a better understanding of the conditions in which COX-2 can be induced is that it may now be reasonable to use COX-2 inhibitors in therapeutic areas such as endotoxin shock or asthma where the 'old' NSAIDs were ineffective. The greater potency against COX-2 together with lack of toxicity on stomach and kidney could allow a reduction in PG output via the induced enzyme while allowing the 'beneficial' output from constitutive COX-1. Another potential therapeutic area of considerable promise for COX-2 inhibitors is suggested by the negative correlation between colon cancer and NSAIDs; 186 recently aspirin was shown to reduce the risk of colorectal cancer by almost half. 187 The crucial observations were that the COX-2 isoform was present only in malignant tissue 188'189 and conferred resistance to apoptosis, 63'19 implying an important role for COX-2 in neoplastic growth. The side effects or toxicity of COX-2 inhibitors are not easy to predict; certainly those of the 'old' NSAIDs should be absent, by definition. From the evidence of the knock-out mice, 64'65 the major toxicities will be on the

Summary
Elucidation of the regulation of COX-2 provides an instructive example of the interaction between molecular biology and applied pharmacology. The basic science of the identification of the isoforms and the stimuli for induction was rapidly transformed into a new and powerful therapeutic concept, NSAIDs without the usual side effects. We now know a great deal about COX-2 from the gene to the crystal structure of the protein, its substrate sites and its intracellular location, much more than about many other enzymes of pharmacological importance. However in one significant aspect this encouraging utilization of molecular biology has failed; for all our knowledge, the design of selective inhibitors has not been based on a careful study of the structure of the protein and its interactions with substrate but, as in the past, on chemical variations of a molecular structure, found empirically to be effective. Moreover, deductions based on the knock-out mice would deny much of the equally empirical evidence correlating inflammation with COX activity. Nevertheless it was molecular biology that disclosed the important place of COX-2 in reproduction, that raised new possibilities for Compound alC50 for COX-2 bRatio IC50 the physiological role of COX-1 and that elucidated the correlation between COX-2 and neoplastic growth. In this last context, there is an intriguing possibility for which there is no direct evidence yet but which is entirely assessable with molecular biological techniques, namely that COX-2, like the products of other immediate-early genes, has effects on gene transcription and/or translation that do not entail the oxidation of AA to PGs. In all these roles and in the new ones to come, the analysis of regulatory mechanisms, physiological, pathophysiological and pharmacological, will remain central to scientific and clinical progress.