Garret A. FitzGerald, 153 Johnson Pavilion, School of Medicine, University of Pennsylvania, Philadelphia, PA 19104-6084, USA. Tel.: +12158981184; fax: +12155730135. E-mail: email@example.com
Summary. Prostaglandins, products of the cyclo-oxygenase (COX) enzymes, can both promote and restrain atherothrombosis. While non-steroidal anti-inflammatory drugs (NSAIDs) selective for inhibition of COX-2 predispose to myocardial infarction, heart failure, hypertension and stroke, suppression of products of COX-1, such as thromboxane (Tx) A2, underlie cardioprotection from low-dose aspirin. Data from clinical pharmacology, rodent models, human genetics, observational studies and randomized trials provide insight into the implications of inhibiting COX product synthesis or function. Many lines of evidence afford a mechanistic explanation for the cardiovascular (CV) hazard from NSAIDs. Elucidation of the biology of this pathway using diversified approaches is also relevant to understanding the implications of substrate rediversion following inhibition of enzymes downstream of COXs, such as the microsomal PGE synthase (mPGES)-1 and of combining D prostanoid antagonism with niacin to attenuate facial flushing.
Prostaglandins (PGs) are formed from arachidonic acid by the sequential action of PG G/H synthases, commonly known as cyclo-oxygenases (COXs) and a series of product-specific synthases that are expressed in a fashion such that most cells make one or two major PGs (Fig. 1). These act in turn on G-protein-coupled receptors. Their deletion has revealed a remarkably diverse and contrasting range of biological effects attributable to these lipids .
The importance of PGs became widely appreciated with the evidence that low-dose aspirin reduces the secondary incidence of myocardial infarction and stroke by about a quarter. Although a myriad of effects have been attributable to aspirin, one needs to invoke no more than suppression of formation of thromboxane (Tx) A2 by platelet COX-1 to explain these clinical results. Aspirin inhibits COX irreversibly by chemically modifying its catalytic center, rendering the anucleate platelet a particular target. Low-dose (<100 mg day−1) aspirin is relatively (but not exclusively) selective for COX-1. Placebo-controlled trials of low-dose aspirin also reveal that it increases the incidence of serious gastrointestinal (GI) bleeds, reflecting suppression not just of platelet TxA2, but also of gastro-epithelial prostaglandin (PG)E2 and PGI2– the latter also known as prostacyclin. While benefit from aspirin outweighs this risk in the case of secondary prevention of cardiovascular (CV) disease, the issue is much more nuanced in patients who have never had a serious atherothrombotic event; here prevention of myocardial infarction by aspirin is numerically balanced by the serious GI bleeds it precipitates .
More recently, the development of non-steroidal anti-inflammatory drugs (NSAIDs) targeted specifically at inhibition of COX-2 has re-emphasized the importance of this system. Here, while efficacy is broadly conserved, the incidence of serious GI events is reduced when compared with traditional (t)NSAIDs that inhibit both COXs reversibly together. However, seven placebo-controlled trials demonstrate that three structurally distinct members of this group of drugs increase the incidence of CV events .
Methods and Results
The first evidence that predicted the possibility that COX-2 inhibition would elevate CV risk derived from clinical pharmacology. Studies in healthy volunteers revealed in 1997 (before the first drug was launched in 1999) that COX-2 was unexpectedly the major source of PGI2 biosynthesis, as reflected by urinary excretion of its major urinary metabolite (PGIM). It was suggested that this reflected hemodynamic induction of COX-2 in endothelium and that the studies that would lead to approval of these drugs were too small and too short to address the possibility of a CV hazard. These studies were dismissed by some on the grounds that (i) other cardioprotective systems, particularly the ability of endothelium to elaborate nitric oxide (NO), would compensate for suppression of PGI2 and (ii) that the origin of PGIM was unknown and deemed highly unlikely to reflect endothelial biosynthesis of PGI2.
Evidence that depression of PGI2 was important in itself was afforded by experiments in mice lacking the PGI2 receptor (the IP). Deletion of the IP augmented the vascular proliferative response to injury and the attendant platelet activation. This phenotype was rescued by co-incident deletion of the TxA2 receptor (the TP). This evidence, that PGI2 restrained cardiovascular effects of TxA2in vivo, was criticized by some as irrelevant to drug action in humans as: (i) it reflected complete loss of PGI2 action, while PGIM was suppressed by 60–80% by the drugs in humans and (ii) was only relevant to mice.
Further experiments addressed the first of these concerns . A biochemically characterized drug regimen specific for inhibition of COX-2 accelerated thrombogenesis in a second model system in mice, an effect intermediate between the impact of deleting one or both copies of the IP. Similar results were obtained with (i) a mouse expressing a COX-2 mutant that removes COX, but retains the hydroperoxidase action of PG G/H synthase (a genetic mimic of the effects of a selective pharmacologic inhibitor); (ii) a structurally distinct inhibitor; and (iii) mice in which COX-2 has been either knocked down or knocked out.
At about this time, the first data from the randomized comparison of rofecoxib and naproxen – the VIGOR (Vioxx Gastrointestinal Outcomes Research) study  was published. This revealed that although serious GI events were less frequent in patients randomized to rofecoxib, serious cardiovascular events were more common. The subsequent publication of the full data revealed an almost numerical balance between serious GI vs. serious CV events (Table 1). This result appeared consistent with the suggestion that COX-2 inhibition might predispose to CV events. However, this was dismissed by some who proposed that while rofecoxib was neutral in this respect, naproxen was cardioprotective. Certainly, the pharmacokinetics of naproxen is quite variable and, in some individuals, drug exposure is sufficient to sustain platelet inhibition over the typical 12 h dosing interval. Indeed, some subsequent epidemiologic data and an overview analysis of randomized trials are consistent with the possibility that naproxen may be the only commonly used, reversible tNSAID that is cardioprotective in some patients; a weak ‘aspirin effect’. However, if the estimate of the difference between the two treatment groups in the VIGOR trial was accurate, even a complete ‘aspirin effect’ of naproxen would account only for half of the difference; such a possibility did not exclude a CV hazard from rofecoxib.
Table 1. The concluding data from the VIGOR study
Source: Curfman GD, Morrissey S, Drazen JM. Expression of concern reaffirmed. N Engl J Med 2006; 354: 1193.
Number of events
Complicated GI events
Serious thromboembolic events
Observational studies prior to the withdrawal of rofecoxib – the traditional approach to monitoring drug safety – delivered somewhat mixed messages. Some, but not all, suggested a CV hazard at a high dose of rofecoxib, 50 mg day−1, but missed a signal at 25 mg day−1, the more commonly prescribed dose. All missed a signal from celecoxib, a shorter-lived and somewhat less selective drug and from valdecoxib, a highly selective inhibitor with which there was less experience.
The results of a placebo-controlled chemoprevention study of 25 mg day−1 rofecoxib, the APPROVe (Adenomatous Polyp Prevention on Vioxx) study [5,6], provided the first unequivocal evidence of a CV hazard from a COX-2 inhibitor and prompted the withdrawal of rofecoxib from the market. Interestingly, some claimed that this was unforeseeable and mechanistically inexplicable. Others claimed that this would not involve other drugs, as it was attributable to some unknown ‘off-target’ effect of rofecoxib.
As evidence accrued from subsequent placebo-controlled trials that both celecoxib and valdecoxib also conferred cardiovascular risk, it was suggested that the ‘mechanism’ that underlay this was a treatable condition, hypertension. It had long been known that hypertension variably complicated treatment with NSAIDs and we had shown that COX-2 inhibition, mutation or deletion or deletion of the IP would elevate blood pressure in mice. Thus, the suggestion that this was a ‘mechanism’ distinct from suppression of COX-2-derived PGs was somewhat eccentric.
Subsequent events also suggested that while hypertension was no doubt an important player, thrombosis seemed to be the dominant element of the phenotype. Thus, overview analysis of the clinical trials – albeit in the absence of individual patient data – revealed a much clearer signal for myocardial infarction than stroke . Furthermore, the apparent similarity of risk conferred by celecoxib and rofecoxib would also be consistent with this theory. A highly non-linear relationship between inhibition of platelet TxA2 and TxA2-dependent platelet-activation mean that these two drugs are functionally equivalent with respect to platelet inhibition (neither causes it) despite modest differences in biochemical selectivity for inhibition of COX-2. Thus, both are similarly ‘functionally selective’ in the platelet albeit not in other organs, such as the vasculature and the heart .
A further line of evidence implicating disruption of the COX-2/PGI2 pathway emerged from human genetics. Thus, polymorphisms in COX-2, PGI2 synthase and the IP have been variously associated with hypertension and adverse CV outcome after angioplasty (e.g. Ref. 8,9).
Meantime, an outstanding mechanistic concern was driven by the failure of some investigators to detect COX-2 in endothelium ex vivo. This was hardly surprising – experiments in vitro suggested that it would be induced by the shear forces exerted by flowing blood and the time to examination ex vivo was highly variable. In any event, human COX-2 was cloned initially from endothelial cells and other investigators were able to detect expression of endothelial COX-2 mRNA and protein ex vivo . Nonetheless, it was proposed that all the PGI2 in endothelial cells came from COX-1. This led to the curious suggestion that a CV hazard from drugs selective for inhibition of COX-2 actually reflected greater inhibition of COX-1 in endothelial cells than in platelets  and was unrelated to inhibition of COX-2 – itself absent from platelets and a minor player at the most in endothelial cells.
Generation of mice in which the COX-2 gene was tagged for cell selective deletion permitted. Selective deletion of COX-2 in endothelial cells or in vascular smooth muscle cells accelerates thrombogenesis and coincident deletion in both cell types exacerbates the phenotype. Interestingly, there is a similar impact on blood pressure, integrating the ‘hypertension’ theory into a single mechanism. Finally, while it is impossible to ascribe a tissue of origin to a urinary metabolite in humans, both endothelial and vascular smooth muscle cell deletion of COX-2 depresses urinary PGIM and the degree of depression relates inversely to the elevation of blood pressure in mice.
Placebo-controlled trials of COX-2 inhibitors also reveal an increased incidence of heart failure. Each of these trials has too few cases to permit analysis of covariates. However, hypertension may predispose to heart failure. Interestingly, selective deletion of COX-2 in cardiomyocytes results in mild heart failure and a predisposition to arrhythmogenesis. Further studies will determine the relevance of these results to clinical observation including an elevated incidence of sudden cardiac death ascribed to an unspecified ’off-target’ effect of rofecoxib.
Our original proposal was that NSAIDs selective for inhibition of COX-2 would predispose to CV events by suppressing cardioprotective products of the enzyme, especially PGI2. This would reflect the constraining properties of PGI2 on all endogenous factors that promoted platelet activation, atherogenesis and hypertension. It was never suggested, although often characterized, as a simple ‘balance theory’ between just two mediators, PGI2 and TxA2. Experiments in mice, using TP deletion or knock down of COX-1 suggested that COX-1 inhibition might mitigate, but not eliminate the risk of hypertension as well as thrombosis.
We also proposed that the consequences of this fundamental mechanism, although dependent on the selectivity for inhibition of COX-2, would also be conditioned by factors relevant to drug exposure – dosage and frequency of dosing, pharmacokinetics and drug potency – as well as the underlying cardiovascular risk of the patient. Deletion of the IP enhances the response to thrombogenic stimuli but does not result in spontaneous thrombosis. Indeed, we also suggested that risk transformation might occur during chronic treatment of patients at low risk as deletion of the IP predisposed to initiation and early development of atherosclerosis in mice. Finally, heterogeneity was noted with respect to selectivity for inhibition of COX-2 amongst the earlier generation NSAIDs; specifically diclofenac has selectivity indistinguishable from celecoxib .
Subsequent evidence supported these predictions. The time to risk detection in the placebo-controlled trials was brief – a couple of weeks – in the studies conducted in coronary artery bypass patients – and delayed beyond a year in chemoprevention and Alzheimer studies where the patients were selected to be at low CV risk [4,5,11–15]. Overview analysis of the celecoxib trials related risk both to dosage and antecedent cardiovascular risk . Overview of the NSAID trials with integration of data on human pharmacology related risk across the class to drug potency and inhibition of COX-2 derived PGI2 . Epidemiologic studies increasingly detected a CV hazard for diclofenac (e.g. Ref. 17,18). The incremental risk of cardiovascular events persisted 1 year after drug cessation in the VIGOR patients, consistent with a slowly resolving (if at all) drug-induced process, such as atherosclerosis . Although no studies were designed to determine whether low-dose aspirin might attenuate the risk of COX-2 inhibitors, post hoc analysis was consistent with this hypothesis in the Therapeutic Arthritis Research and Gastrointestinal Event Trial (TARGET) study, where the difference in CV events between patients receiving lumiracoxib and naproxen was attenuated by low-dose aspirin . Finally, the strength of the respective signals of GI and CV risk across randomized comparisons of the NSAIDs relates to the degree of selectivity spanned by the comparators .
In the 12 years since a mechanism for a CV hazard from drugs that inhibit COX-2 was proposed, conclusive evidence for such a hazard has emerged from seven placebo-controlled trials of structurally distinct NSAIDs. Evidence consistent with the proposed mechanism has emerged from clinical pharmacology, proof of concept studies in rodents and other species, observational studies, human genetics and randomized comparisons amongst NSAIDs. This is the most expansive and diversified data set that exists for any mechanism of drug action to date. By contrast, evidence – that remains unaddressed – rejecting the original hypothesis has yet to emerge.
Many challenges and opportunities remain within the field of PGs and their cardiovascular biology. There is considerable inter-individual differences amongst even healthy volunteers dosed under standardized conditions with NSAIDs . Can we exploit this heterogeneity to determine whether individuals differ in their efficacious response to NSAIDs and if so, why? Can we identify markers – genomic variation, biased and unbiased biomarkers of drug response, and physiologic responses like a rise in blood pressure – that identify individuals at risk or at emerging risk during chronic dosing of CV effects of NSAIDs?
Attention has shifted downstream particularly to mPGES-1 . A major source of PGE2, deletion of this enzyme seemed as effective as NSAIDs in some, but not all models of pain in mice. Interestingly, mPGES-1 knockouts (KOs), unlike COX-2 or IP KOs were not predisposed to thrombogenesis and seemed less likely to develop hypertension. However, this might reflect substrate rediversion to augment production of PGI2, the converse of what happens with inhibition of COX-2. Unfortunately, we now know that PGI2, just like PGE2 can mediate pain; thus, what affords a bland cardiovascular profile to drugs that target this pathway may undermine their primary efficacy as analgesics. Meantime, we have found that mPGES-1 deletion actually restrains atherogenesis and formation of angiotensin II induced aortic abdominal aneurysm formation in mice. Perhaps syndromes of cardiovascular inflammation will turn out to be the primary opportunity for investigating the efficacy of these drugs .
Finally, there is much yet to be learned about the role of PGs in cardiovascular function. For example, we have recently found that deletion of the F prostanoid receptor – the FP – for PGF2α reduces blood pressure and atherogenesis by disrupting renin release in the kidney . PGD2 is a particular focus of interest as activation of one of its receptors – the DP1 – mediates facial flushing induced by niacin; recently, a drug that combines niacin and a DP1 antagonist has been shown to attenuate this phenomenon . PGD2 is also a rediversion product enhanced in mPGES-1 KO mice. However, little is known about the cardiovascular biology of the DP1, other than its existence on human platelets where, just like the IP, it is coupled to adenylate cyclase activation.
In summary, the integration of previously segregated fields of basic and clinical scientific endeavor predicted and explained the CV hazard from NSAIDs selective for inhibition of COX-2: a nice example of translational therapeutics. A similar approach will be deployed increasingly to realize new opportunities and manage risk of all emerging therapies, not just those targeting the PG pathway.
Supported by grants (PO1-HL62250, P50-HL-83799, HL 54500 and 1U54-RR-023567) from the National Institutes of Health and (Jon Holden DeHaan Foundation-0730314N, 0430148N and 0430148N) from the American Heart Association. Dr FitzGerald is the McNeil Professor of Translational Medicine and Therapeutics.
Disclosure of Conflict of Interests
Dr Fitzgerald receives support for investigator initiated studies from Bayer, Crystal Genomics and Boehringer Ingelheim. During the past year he has received compensation for consultation on drugs relevant to this manuscript from Logical Therapeutics, Synosia, Astra Zeneca, Merck, Novartis and Nicox.