Marked Interindividual Variability in the Response to SelectiveInhibitors of Cyclooxygenase-2

SUSANNE FRIES,* TILO GROSSER,* THOMAS S. PRICE,* JOHN A. LAWSON,* SHIV KAPOOR,*SUSAN DEMARCO,* MATHEW T. PLETCHER,‡ TIM WILTSHIRE,‡ and GARRET A. FITZGERALD**The Institute for Translational Medicine and Therapeutics, University of Pennsylvania, Philadelphia, Pennsylvania, and ‡The GenomicsInstitute of the Novartis Foundation, San Diego, California

Background & Aims: Variability in response to drugsmay influence both efficacy and safety. Cyclooxygenase(COX)-2 inhibitors pose a cardiovascular risk by poten-tially increasing the likelihood of thrombosis, hyperten-sion, and atherogenesis. Differences between individu-als in the response to COX-2 inhibitors would beexpected to influence their susceptibility to cardiovascu-lar complications. We examined the variability in degreeand selectivity of COX-2 inhibition in humans in re-sponse to celecoxib and rofecoxib. Methods: Fiftyhealthy volunteers received placebo, rofecoxib (25 mg),and celecoxib (200 mg), randomized by order. COX-1and COX-2 inhibition was determined using ex vivo andin vivo indices of enzymatic activity. A subset of 5 indi-viduals underwent 5 replicate studies to estimate vari-ability in drug response both within and between sub-jects. Results: Despite the higher COX-2 selectivity ofrofecoxib in vitro, the average selectivity attained by 25mg rofecoxib and 200 mg celecoxib in vivo were notdifferent. However, there was considerable variability atan individual level in the degree of COX-2 inhibition andselectivity attained by both drugs. Approximately onethird of the variability was attributable to differencesbetween individuals, suggesting the contribution of ge-netic sources of variance, such as candidate polymor-phisms detected in COX-1 and CYP2C9. Conclusions:The actual degree of selectivity for inhibition of COX-2achieved by the coxibs relates both to chemical proper-ties of the drug and to factors within an individual thatmodulate drug response. These sources of variabilitymight be exploited to identify patients uniquely suscep-tible to benefit or at developing risk of cardiovascularcomplications.

The coxibs, selective inhibitors of cyclooxygenase(COX)-2, were designed to inhibit the major enzy-

matic source of the prostaglandins (PGs), which mediatepain, inflammation, and, perhaps, tumorigenesis, whilesparing COX-1-derived PGs, which contribute domi-nantly to gastric cytoprotection.1 Two coxibs, rofecoxib2

and lumiracoxib,3 have been shown in controlled trials toreduce the incidence of serious gastrointestinal (GI) ad-

verse effects when compared with traditional nonsteroi-dal anti-inflammatory drugs (tNSAIDs). Celecoxib re-duced the number of colorectal polyps compared withplacebo in patients with familial adenomatous polyposisin a controlled clinical trial.4 However, 3 structurallydistinct coxibs, rofecoxib, valdecoxib, and celecoxib, in-creased the incidence of myocardial infarction and strokein randomized controlled trials,2,5–8 suggesting that se-lectivity for inhibition of COX-2 may confer a cardio-vascular hazard. Rofecoxib and valdecoxib have recentlybeen withdrawn from the market. Both compounds aremore selective for COX-2 than celecoxib in vitro.9 De-spite such differences, all of the coxibs depress substan-tially prostacyclin (PGI2), leaving platelet COX-1-derived thromboxane (Tx) A2 unaffected, in contrast toaspirin or tNSAIDs, which inhibit both COX-1 andCOX-2.10,11 PGI2 acts as a constraint on all agonists,which elevate blood pressure, activate platelets, andstimulate atherogenesis.12–14 Thus, drug selectivity forinhibition of COX-2 is thought to contribute to thelikelihood of hypertension, myocardial infarction, andstroke resulting from treatment with coxibs.15,16

The detection of cardiovascular complications attrib-utable to coxibs relates in patients to their underlyingrisk of cardiovascular disease.17 Thus, it seems rationalthat patients with identifiable cardiovascular risk factorswill be excluded from treatment with selective COX-2inhibitors. However, cardiovascular adverse events at-tributable to the coxibs also occurred in patients whowere apparently at low initial risk of cardiovascular dis-ease,6,7 suggesting that both rofecoxib and celecoxibcaused a risk transformation during extended treatmentof these individuals. The emerging hazard would be

Abbreviations used in this paper: COX, cyclooxygenase; NSAIDs,nonsteroidal anti-inflammatory drugs; PGs, prostaglandins; PGI2, pros-tacyclin; PGI-M, 2,3-dinor-6 keto prostaglandin F1!; TxB2, thrombox-ane; Tx-M, 11-dehydrothromboxane B2.

© 2006 by the American Gastroenterological Association0016-5085/06/$32.00

doi:10.1053/j.gastro.2005.10.002

GASTROENTEROLOGY 2006;130:55–64

expected to relate to multiple factors, including thedegree of selectivity for inhibition of COX-2 actuallyattained within an individual, plasma concentrations,and half-life and duration of drug action.15,16 Such phar-macokinetic and pharmacodynamic characteristics havebeen increasingly associated with both drug safety andefficacy and may be successfully exploited to individual-ize therapy18 when heritable mechanisms or predictivehost factors such as gender, race, age, or weight can beidentified.19 However, environmental sources of variabil-ity such as nutritional status and physical activity canmask clinically relevant associations. The relative impor-tance of stable interindividual sources of variability indrug response—such as genetic variants—comparedwith less predictable, environmental sources of variabil-ity in the response to coxibs is unknown.

We examined the variability, both within and be-tween subjects, in response to celecoxib and rofecoxib, ina placebo-controlled crossover study. Our studies suggestthat even healthy individuals respond quite differently tocoxibs. This variance might confound the detection ofcardiovascular events attributable to drug exposure. Onthe other hand, interindividual variability in drug re-sponse might be exploited to identify patients in whomthe drug exhibits unique efficacy or in whom the evolu-tion of cardiovascular risk is accelerated. This may per-mit definition of the parameters within which thesedrugs2,3 might be administered chronically to patientsinitially at low risk of cardiovascular disease.

Materials and MethodsStudy Design

This randomized, double-blind, placebo-controlledprotocol was approved by the Institutional Review Board ofthe University of Pennsylvania Health System and by theAdvisory Council of the General Clinical Research Center(University of Pennsylvania, Philadelphia). Screening, enroll-ment, and follow-up of healthy study volunteers were per-formed at the General Clinical Research Center from January2002 to January 2004. Written informed consent was obtainedfrom all volunteers. All had an unremarkable medical history,physical examination, and routine hematologic and biochem-ical screen and were within 30% of ideal body weight. Subjects

were nonsmokers and abstained from the use of aspirin andtNSAIDs, as assessed by history and platelet aggregometry, forat least 2 weeks before enrollment.20 Routine hematology,biochemistry, and urinalysis were assessed at time of screeningand at 24 hours after administration of the drugs on comple-tion of the study.

Treatments and Assessment

Fifty healthy volunteers (Table 1), aged between 21and 43 years, received, in random order, a single dose ofplacebo, celecoxib (200 mg), and rofecoxib (25 mg) underdouble-blind conditions, separated by washout periods of atleast 2 weeks. The randomization sequence was determined bythe General Clinical Research Center pharmacist by allocationof the drugs to numbered containers. Inhibition of plateletCOX-1 was assessed by measurement of serum TxB2.21 COX-2inhibition was assessed ex vivo by measurement of lipopolysac-charide-stimulated PGE2 in plasma.22 The coefficients of vari-ation for repeatability and reproducibility were 4.3% and4.1%, respectively, for COX-1 inhibition and 1.9% and 4.3%,respectively, for COX-2 inhibition. Measurements were per-formed immediately before the administration of drug (0 hrs)and 4 hours thereafter. Urinary 2,3-dinor-6keto PGF1! (PGI-M), an index of PGI2 biosynthesis,23 and 11-dehydro TxB2

(Tx-M),24 reflective of TxA2 formation in vivo, were assessed at0 and 4 hours in spot urine samples that were collected 30minutes after voiding. The coefficients of variation for bothassays were "10%. A subset of the volunteers (n # 5) pro-gressed through the entire protocol on 5 occasions, each sep-arated by 2 weeks, to assess intraindividual variability in drugresponse in 5 replicates.

Plasma Drug Concentrations

Plasma samples were analyzed by liquid chromatogra-phy tandem mass spectrometry (Quattro Ultima, Micromass,Beverly, MA) using L755100 (Merck Frosst, Point Claire,Canada) and SC58125 (Cayman, Ann Arbor, MI) as internalstandards for celecoxib and rofecoxib, respectively. Sampleswere chromatographed on a Luna 3$ C18 (150 % 2.0 mm)column (Phenomenex, Torrance, CA) and quantitated usingnegative-atmospheric-pressure ionization and selected reactionmonitoring of m/z 380.2¡316.1 (celecoxib), 384.2¡305.1(SC58125), 314.2¡215.1 (rofecoxib), and 328.2¡313.1(L755100). The coefficients of variation for these assays were5.4% for celecoxib and 8.6% for rofecoxib measurements.

Table 1. Characteristics of the Subjects

Female (n # 22) Male (n # 28) All (n # 50)

Age (y) (25%, median, 75%) 23.3, 26.0, 31.8 23.0, 29.5, 35.3 23.0, 27.5, 34.5Ethnicity, No. (%)

White 12 (54.6) 17 (60.7) 29 (58.0)Black 7 (31.8) 5 (17.9) 12 (24.0)Asian 3 (13.6) 6 (21.4) 9 (18.0)

Body mass index (25%, median, 75%) 21.0, 23.5, 28.2 23.3, 25.3, 27.8 21.8, 24.8, 28.0

56 FRIES ET AL GASTROENTEROLOGY Vol. 130, No. 1

Single Nucleotide PolymorphismGenotyping

Twenty-six single nucleotide polymorphism (SNP) as-says (Table 2) were performed, using the Sequenom MassAR-RAY system (Sequenom, San Diego, CA).25 These assays ro-bustly detected 1083 out of 1482 genotypes. The additional399 genotypes were determined by resequencing (ABI 3730xl;Applied Biosystems, Foster City, CA).

Statistical Analysis

The primary response variables were ex vivo COX-2inhibition, expressed as the ratio of postdrug to predrugplasma PGE2; ex vivo COX-1 inhibition, expressed as the ratioof postdrug to predrug serum TxB2; urinary PGI-M inhibi-tion, expressed as the ratio of concentration postdrug to post-placebo administration; urinary Tx-M inhibition, expressed asthe ratio of concentration postdrug to postplacebo administra-tion; and plasma drug concentrations 4 hours postdrug admin-istration. Derived variables were ratios of primary measures asestimates of COX-2 selectivity. All variables were log trans-formed to approximate normality. Four data points (3 mea-

surements of Tx-M and 1 plasma concentration of celecoxib)were excluded because of technical errors in the assays.

Each subject in the study had either 1 or 5 replicate sets ofmeasurements. Hence, the study followed an unbalanced hi-erarchical design with the repeated factor DRUG nestedwithin measurement REPLICATE nested within study SUB-JECT. However, the effects of REPLICATE grouped withinSUBJECT were minimal, and a regression model was adoptedin which the multivariate outcomes were grouped only bySUBJECT. These semiparametric regression models were es-timated by generalized estimating equations using a robust“sandwich” variance estimator and identity link functions forthe mean and scale models (geepack v0.2-10 in R 2.0-1;http://www.cran.R-project.org26). The multiple measuresgrouped by SUBJECT were assumed to have an exchangeablecorrelation structure: This is equivalent to a hierarchical modelwith random intercepts grouped within SUBJECT. Variancecomponents estimates were calculated using Bayesian hierar-chical linear models of this type, with Markov Chain MonteCarlo estimation in winBUGS 1.4.1.27 Comparisons of theprimary response variables between the treatment groups had

Table 2. Single Nucleotide Polymorphisms

Gene DomainAccessiona/reference

numbers PositionBase/aminoacid change

Allele frequency

Majorallele

Minorallele

COX-1 Promoter (Halushka et al45) &842 G/A 1 0Signal rs1236913 8 Trp/Arg 0.96 0.04

rs3842787 17 Pro/Leu 0.95 0.05EGF rs3842789 53 Arg/His 1 0Catalytic rs3842792 185 Lys/Thr 0.97 0.03

(Ulrich et al43) 230 Gly/Ser 1 0rs5789 237 Leu/Met 0.98 0.02(Ulrich et al43) 341 Lys/Arg 1 0rs5791 359 Lys/Arg 1 0rs5792 443 Ile/Val 1 0rs5793 467 Lys/Glu 1 0rs5794 481 Val/Ile 0.99 0.01

COX-2 Promoter (Papatili et al44) &765 G/C 1 0Signal rs20426 1 Met/Ile 1 0Membrane rs4987011 61 Thr/Arg 1 0Catalytic rs3218622 228 Arg/His 1 0

rs4648279 428 Pro/Ala 1 0rs5272 488 Glu/Gly 1 0rs5273 511 Val/Ala 0.97 0.03

C-terminus rs3218625 587 Gly/Arg 1 0CYP2C9 Catalytic rs1799853b 144 Arg/Cys 0.92 0.08

rs2256871c 251 His/Arg 0.99 0.01rs1057909 358 Tyr/Cys 1 0rs1057910d 359 Ile/Leu 0.95 0.05VAR_013515e 359 Ile/Thr 1 0VAR_013516f 360 Asp/Glu 0.99 0.01

aAccession numbers refer to the NCBI (rs . . . ) or the EXPASY (VAR_ . . . ) SNP database.bCYP2C9!2.cCYP2C9!9.dCYP2C9!3.eCYP2C9!4.fCYP2C9!5.

January 2006 VARIABILITY IN THE RESPONSE TO COXIBS 57

a power in excess of 90% to detect a moderate effect size of0.50 for 2-sided Student t test of paired observations with typeI error rate of 0.05 in a study of 50 subjects. Results aregraphed on logarithmic scales where error estimations wereperformed on log transformed data.

Exploratory tests of genetic association on multiple outcomemeasures were conducted using multivariate analysis of vari-ance (MANOVA), with the dependent variables averaged overreplicate measurements for each subject. Analyses were con-ducted in 2 ways: with data points unweighted and with datapoints weighted by the number of replicate sets of measure-ments.

ResultsFifty volunteers received, in random order, pla-

cebo and single therapeutic doses of rofecoxib and cele-coxib to compare directly the responses to the drugswithin the same subjects. Baseline levels of the drugresponse parameters (medians) were as follows: SerumTxB2 408 ng/mL (25%: 280 ng/mL, 75%: 557 ng/mL);LPS induced PGE2 28 ng/mL (25%: 16 ng/mL, 75%: 47ng/mL); urine Tx-M 293 pg/mg creatinine (25%: 185pg/mg creatinine, 75%: 439 pg/mg creatinine); urinePGI-M 81 pg/mg creatinine (25%: 55 pg/mg creatinine,75%: 117 pg/mg creatinine). Both rofecoxib and cele-coxib inhibited COX-2-dependent PGE2 formation exvivo significantly when compared with placebo (Figure1A). Small placebo effects may be attributable to diurnalvariation because predrug and postdrug measurementswere performed consistently at 0800 and 1200 hours.There was no difference in the degree of COX-2 inhibi-tion attained by the 2 drugs (rofecoxib 44.8% vs cele-coxib 47.5%). Celecoxib also depressed COX-1-derivedserum TxB2, but the effect was small in comparison withthe inhibition of PGE2 (Figure 1B). Both drugs de-pressed the formation of PGI2 in vivo as assessed byurinary PGI-M excretion but not the formation of TxA2

as assessed by excretion of urinary Tx-M (Figure 1C and1D). There was no difference between the drugs in theireffects on these measures of COX-2 and COX-1 activityin vivo (Figures 1C and 1D).

The degree of selectivity for COX-2 in vitro—a chem-ical property of a COX inhibitor—is commonly ex-pressed as the ratio of the drug concentrations required toinhibit the enzymatic activities of COX-2 and COX-1 by50% (Figure 2A). Based on these criteria, rofecoxib ismore selective than celecoxib.1 However, pharmacoki-netic and pharmacodynamic variations between individ-uals would be expected to affect the degree of COX-2selectivity actually attained in humans, which can bedescribed by the ratio of COX-2 inhibition vs COX-1inhibition at any given plasma concentration (Figure

2B). When the degree of COX-2 selectivity attained insubjects was estimated using this measure, both drugsdiffered significantly from placebo but not from eachother at this time point after dosing (Figure 2C).

There was considerable variability in both ex vivo andin vivo indices of inhibition of both COX-2 and COX-1(Figures 1E–H). However, drug response correlatedpoorly with plasma drug concentrations of rofecoxib(median, 215 ng/mL; range, 115–658 ng/mL) and cele-coxib (median, 575 ng/mL; range, 179–1372 ng/mL) at4 hours after dosing (data not shown). Plasma drug levelsafter dosing showed a trend toward lower concentrationswith increasing body mass index (BMI). However, thisattained significance (P " .05) only for rofecoxib: a unitincrease in BMI was associated with a drop in plasmaconcentration of approximately 2%.

We assessed intra- and interindividual variability indrug response to determine the likely contribution ofrandom and environmental fluctuations rather than in-terindividual factors to variability. Thus, 5 replicate setsof measurements were performed in 5 volunteers (Figure1I–L). The simplified model underlying this experimen-tal design assumed that the total variability as observed inthe whole population (Figure 1E-H) was composed ofinterindividual variability, intraindividual variability, andsome portion of random variability (eg, technical variabil-ity). The replicate measurements served to determine thecombination of intraindividual variability and randomvariability (Figure 1I–L). Thus, assessment of the totalvariability (Figure 1E–H) and its 2 components intrain-dividual and random variability (Figure 1I–L) allowedestimation of the third component, interindividual vari-ability, using a variance component analysis. The averageresponses for both drugs in the replicate measurementsclosely approximate the observations made within thewhole population. Variability within these individuals(intraindividual and random variability) was less than thevariability in the population as a whole (total variability).This can be visualized by expressing for each individualthe variances of the replicate measurements and compar-ing the average variance of measurements within thesame subject with the total variance (Figure 3A–D). Thereplicate measurements in 3 out of 4 response variablesexhibited a lower variation within individuals than in thewhole population, indicating that interindividual variabil-ity, beyond that attributable to random and environmen-tal influences, is detectable within the population. In-deed, variance component analyses suggested that up toapproximately 30% of the total variability in the mea-sures of COX-1 and COX-2 inhibition may be attributedto interindividual variability (data not shown).

58 FRIES ET AL GASTROENTEROLOGY Vol. 130, No. 1

Genetic variability is a likely contributor to suchdifferences in drug response between individuals.Thus, we genotyped the study population for poly-morphisms in COX-1, COX-2, and cytochrome P450(CYP) 2C9, the principal metabolizing enzyme ofcelecoxib, to examine their potential role in drugresponse (Table 2). We targeted variants potentiallyaffecting either baseline expression or activity of the

enzymes or their interaction with the coxibs (Table 2).Fourteen volunteers (28%) were heterozygous for avariant of COX-1 located in the signal peptide or itscatalytic site; 3 (6%) were heterozygous for a variantin the catalytic site of COX-2; and 14 (28%) wereheterozygous for a variant of CYP2C9. Thus, thepopulation was genetically quite heterogeneous re-garding the enzymes most likely to affect drug re-

Figure 1. Comparison of 4 primary drug responses to placebo, rofecoxib, and celecoxib. The 4 columns show the drug responses, COX-2 andCOX-1 inhibition, and urinary excretion of the prostacyclin metabolite (PGI-M) and the thromboxane metabolite (Tx-M). The 3 rows show theaverage effects of the 3 treatments; the variation of the responses in the whole population, by plotting all individual observations from which theaverage effects are derived; and the variability of replicate measurements within 5 individuals. All data are plotted on log scales to visualize thefull range of variability. First row (A–D): The average effects are shown as fitted means and standard error of the mean. There was inhibition ofCOX-2 by both drugs (A), a minor degree of inhibition of COX-1 by celecoxib (B), similar depression by both drugs of PGI-M (C), and no effect ofeither on Tx-M (D). Second row (E–H): Variation of the responses in the study population. Whereas the main effects, such as inhibition of COX-2(E) are evident, there is a striking variability within the population. Third row (I–L): Variability of replicate measurements within 5 individuals(designated A to E on the abcissae) as repeatability-reproducibility plots. Data points of replicate measurements are connected by vertical graylines. A grey box indicates the maximal range of observations for each drug treatment. A horizontal grey line in each box represents the averageoverall measurements within the box. There was also a striking intraindividual variability, which appears only slightly smaller than the variabilityof the whole population (second row), and replicate measurements (third row), which are plotted on the same scales.

January 2006 VARIABILITY IN THE RESPONSE TO COXIBS 59

sponse; a total of 24 out of 50 individuals had 1 ormore genetic variant. No homozygous variants oc-curred, and they did not associate with gender. Thisstudy was not designed to address comprehensivelythe impact of SNPs on drug response, so we confinedour analyses to those SNPs that occurred in 4 ormore volunteers. Thus, CYP2C9*2 (Arg144Cys) andCYP2C9*3 (Ile359Leu) were tested for their effect onplasma concentrations of celecoxib, and the Trp8Argand Pro17Leu variants of COX-1 were tested for theireffects on COX-1 response.

CYP2C9*2 occurred in 8 individuals (5 white, 3Asian) and was associated with higher plasma concentra-tions of celecoxib, after adjusting for the linear effects of

age (Figure 3A). BMI and age were not significant co-variates in this model. Effects on the response to cele-coxib were not statistically significant. CYP2C9*3 hadno effect on plasma concentrations. Only the Pro17Leuvariant of COX-1 was associated with a statisticallysignificant (MANOVA, P # .009, unweighted; P #.011, weighted) pharmacodynamic effect—a reduction inCOX-1 inhibition both ex vivo and in vivo. Post hocanalysis revealed an association with an absence of theCOX-1 inhibition by celecoxib ex vivo (Figure 4B) andan absence of COX-1 inhibition in vivo by rofecoxib(Figure 4C). No effects of BMI or age were found in theseanalyses; males showed greater Tx-M suppression withboth rofecoxib and celecoxib than females (P " .05).

Figure 2. (A) Model of the estimation of the degree of COX-2 selectivity in test systems. Selectivity is described as the ratio of the concentrationsrequired to inhibit the activity of the isozymes by 50% (IC50 for COX-1/IC50 for COX-2). (B) Model of the estimation of COX-2 selectivity attainedin humans. Selectivity is described as the ratio of achieved isozyme inhibition at a given plasma concentration. (C) Assessment of the ratio ofex vivo COX-2 inhibition to COX-1 inhibition as an estimate of attained isoform selectivity 4 hours postdosing. The average effects are shown asfitted means and standard error of the mean. Although both drugs differed significantly from placebo, no difference in selectivity between thedrugs was evident.

Figure 3. Comparison of the intraindividual variability with total variability in the study population. Variances of replicate measurements of the4 primary drug responses (A–D) are plotted for each subject and treatment (colored dots connected by grey lines). Green horizontal lines indicatethe average variances of replicate measurements (intraindividual variability). Blue horizontal lines represent the derived variance of allmeasurements within the whole population (total variability). The variances of replicate measurements (intraindividual variability) are lower thanthe variances of the whole population (total variability) for the drug responses COX-2 inhibition ex vivo (A), depression of urinary prostacyclinmetabolite excretion as a marker of COX-2 inhibition in vivo (C), and depression of urinary thromboxane metabolite excretion as a marker of COX-1inhibition in vivo (D).

60 FRIES ET AL GASTROENTEROLOGY Vol. 130, No. 1

DiscussionVariability between individuals in their response

to drugs is well recognized.19 Nevertheless, the typicalparadigms of drug development and approval infer acommon response to a limited number of doses within atherapeutic category. Inhibitors of COX-2 are no excep-tion. Indeed, a considerable variability in the plasmaconcentration/enzyme inhibition response relationshipswas noted as the first member of the class, celecoxib,10

entered the US market. Typically, assays in whole humanblood have been utilized to compare the selectivity forinhibition of COX-2 among members of the class.9

Although the multiple factors that can distort the rela-tionship between the degree of selectivity attained invitro and in vivo have been noted,1,28 the implications ofthis observation have received little attention. Evidenceexists in animal models that suggests that the degree ofselectivity attained in vivo may relate to cardiovascularoutcomes, including hypertension, myocardial infarc-tion, and stroke.12,14 However, this hypothesis has neverbeen addressed in clinical studies.

Rofecoxib has been withdrawn from the market be-cause of an excess of myocardial infarction and stroke inpatients receiving this drug (25 mg/day) in the Adeno-matous Polyp Prevention on VIOXX (APPROVe) trial.7

This hazard evolved slowly over time, becoming firstevident after 18 months of treatment, in only 1%–2% ofpatients. A similar pattern, again consistent with a time-dependent transformation of cardiovascular risk was ob-served in the Adenoma Prevention with Celecoxib (APC)study of celecoxib.5 Thus, a small minority of the pa-tients in both studies, apparently initially at low risk ofcardiovascular disease, proceeded to increase that risk toa point that culminated in clinical events.16 The exploi-tation of variability in drug response to identify patientsat emerging risk of cardiovascular disease requires first adetermination of whether variability in response betweenindividuals exceeds that attributable to environmentalvariability within individuals.

COX-2 inhibitors depress biosynthesis of PGI2, whilenot affecting platelet COX-1-derived TxA2,10,11 unlike

aspirin21 or tNSAIDs.20 PGI2 acts as a constraint in vivoon the cardiovascular effects of TxA2,12 which includeplatelet aggregation,29 elevation of blood pressure,30 andacceleration of atherogenesis.14,31 This last effect may beparticularly pertinent to the “latent period” before theemergence of cardiovascular risk in the APPROVe andAPC studies.6,7 Deletion of the PGI2 receptor—the IP—accelerates the initiation and early development of ath-erosclerosis in 2 mouse models.14,32 In this context,COX-2-derived PGI2 acts to limit the interactions ofleucocytes and platelets with the vasculature and theattendant oxidant stress, thereby restraining atherogen-esis.14,32 Epidemiologic approaches and, indeed, clinicaltrials have relatively poor precision for detection of un-common, but serious, adverse effects, such as clinicalevents prevalent in the relevant populations. Thus, bi-omarkers of these mechanistic effects might be combinedwith measures of drug response (eg, blood pressure) andselectivity to identify individuals at accelerated risk ofcardiovascular disease, prior to the occurrence of clinicalevents. Research is necessary to determine the paradigmswithin which COX-2 inhibitors can be prescribed forextended periods safely in individuals initially at lowcardiovascular risk, including their potential use for can-cer prevention.33

We sought to explore the variability in response tocoxibs to examine the practicality of such an approachbased on biomarkers of drug response. We assessed ther-apeutic doses of celecoxib and rofecoxib, using a placebo-controlled, crossover design to investigate variability inplasma drug concentrations and pharmacodynamic re-sponses, as assessed by both ex vivo and in vivo indices ofCOX-1 and COX-2 activity. Doses of celecoxib (200mg) and rofecoxib (25 mg) were selected that have showncomparable effectiveness in clinical models of acutepain.34,35 Thus, perceptible pain relief is achieved within1 hour of administration of these doses, and their max-imum analgesic effect (2–3 hours postdosing) is approx-imately two thirds that of high doses of tNSAID com-parators.34,35 Despite the higher COX-2 selectivity ofrofecoxib in vitro, the average selectivity attained by

Figure 4. Association of polymorphismsin CYP2C9 and COX-1 with plasma cele-coxib 4 hours postdosing (A), COX-1 inhi-bition ex vivo by celecoxib (B), and depres-sion of the urinary thromboxanemetabolite by rofecoxib (C). (A) Eight indi-viduals, heterozygous for CYP2C9*2, hadhigher plasma concentrations of celecoxibthan those without this variant. (B and C)The Pro17Leu variant of COX-1 was asso-ciated with a reduction in COX-1 inhibitionboth ex vivo (B) and in vivo (C).

January 2006 VARIABILITY IN THE RESPONSE TO COXIBS 61

these doses of rofecoxib and celecoxib in vivo was notdifferent, suggesting that both efficacy and cardiovascu-lar and gastrointestinal safety profiles of the drugs afteracute dosing might be similar. This might differ underchronic dosing conditions when the more prolongedhalf-life (on average) of rofecoxib would be expected toinfluence both relative efficacy and the adverse effectprofile. However, again, interindividual differences inpharmacokinetics might modulate that expectation atthe individual level. We determined both interindi-vidual and intraindividual variability in drug response,the latter to assess the stability of the biochemical phe-notype and to get an impression of whether variationattributable to environmental and random effects wouldbe sufficient to obscure a genetic contribution. Signifi-cant intraindividual variability was evident in all param-eters as 5 individuals progressed through the entire studyon 5 occasions. Some of the variation, evident on placebo,may have derived from circadian variability because sam-ples were obtained at discrete time points before (0800hours) and after (1200 hours) dosing. Diurnal variationin the indices of COX activity has not been described,but diurnal variation in the pharmacokinetics ofNSAIDs,36,37 including celecoxib,38 attributable in partto circadian variation in drug metabolizing enzymes,39 iswell recognized. The intraindividual variability in re-sponse to both drugs is striking. This may reflect in partvariability in assays (although all coefficients of variationwere "10%), sources of variability that may be attenu-ated under steady-state conditions, and random environ-mental effects that, despite standardization of certainconditions of dosing (time, posture, environment, fast-ing/feeding), remain unrecognized. Little comparable in-formation is available with other drugs in which intra- vsinterindividual variability in response has been comparedso comprehensively.40–42

Despite substantial intraindividual variability in re-sponse, this was exceeded by the variation in the subjectpopulation as a whole. Thus, the variance of both plasmadrug concentrations (data not shown) and 3 out of 4pharmacodynamic parameters were greater in the entirepopulation than for replicate measurements within indi-viduals. We would anticipate that this divergence wouldbe even more pronounced with comparison between andwithin individuals when dosing had attained steadystate.

Thus, additional to the potential value of measure-ments of pharmacokinetic and pharmacodynamic re-sponse, genetic sources of variance might contributedetectably to differences in drug response over and abovethe background “noise” attributable to analytic variabil-ity and fluctuating features of the environment. Our

analysis showed that almost half of the study populationhad variants in the target enzymes, some with allelicfrequencies of 5% or more. Information is only begin-ning to emerge on genetic variation in the COX en-zymes,43–45 and our study was not designed or sized toafford a comprehensive analysis of genetic variants,which may modulate the response to coxibs. Notwith-standing these restrictions, 2 of the SNPs tested showedevidence of allelic association with elements of drugresponse. The CYP2C9*2 is a variant associated withreduced activity of a major metabolizing enzyme ofcelecoxib46 and was associated with elevated plasma con-centrations of the drug 4 hours after administration.However, although this and other genetic variants ofCYP2C9 are thought to affect the metabolic clearance ofperhaps 20% of all prescribed drugs,47 acquired or envi-ronmental factors may also contribute to variability indrug clearance.48

The Pro17Leu variant of the COX-1 enzyme appearedto be associated with a failure of inhibition of thrombox-ane formation with both drugs. All COX-2 inhibitors arerelatively, as opposed to absolutely, selective for COX-2and, at sufficient concentration, become inhibitors ofboth COXs in vitro (Figure 2A), just like tNSAIDs.1 Theinterindividual variability of inhibition of COX-1 evi-dent 4 hours after dosing with either coxib was substan-tial and may be attributable in part to this variant. Thegreater the inhibition of COX-1-derived TxA2 in anindividual, the more the gastrointestinal and cardiovas-cular profile of a coxib would be expected to resemblethat of a tNSAID. Thus, determination of the factors thatinfluence the actual selectivity for inhibition of COX-2attained within an individual may identify biomarkers ofrelevance to clinical outcome.

In summary, we report substantial variability bothwithin and between individuals in their response to acutedosing with both celecoxib and rofecoxib. SNPs in bothCOX-1 and CYP2C9 were apparently associated withelements of drug response, although population stratifi-cation may confound these findings. It is likely that amore comprehensive approach will shed further light ongenetic sources of variation. Validated genetic markersmight be usefully combined with biomarkers of athero-genesis, functional (eg blood pressure), kinetic, and dy-namic parameters of drug action to identify individualswho accelerate their development of cardiovascular riskduring treatment with coxibs. Such a strategy mightconserve the value of extended dosing with selectiveinhibitors of COX-2 for patients at low risk of cardio-vascular diseases who have previously exhibited gastro-intestinal intolerance of tNSAIDs.

62 FRIES ET AL GASTROENTEROLOGY Vol. 130, No. 1

References1. FitzGerald GA, Patrono C. The coxibs, selective inhibitors of cy-

clooxygenase-2. N Engl J Med 2001;345:433–442.2. Bombardier C, Laine L, Reicin A, Shapiro D, Burgos-Vargas R,

Davis B, Day R, Ferraz MB, Hawkey CJ, Hochberg MC, Kvien TK,Schnitzer TJ. Comparison of upper gastrointestinal toxicity ofrofecoxib and naproxen in patients with rheumatoid arthritis.VIGOR Study Group. N Engl J Med 2000;343:1520–1530.

3. Schnitzer TJ, Burmester GR, Mysler E, Hochberg MC, Doherty M,Ehrsam E, Gitton X, Krammer G, Mellein B, Matchaba P, GimonaA, Hawkey CJ. Comparison of lumiracoxib with naproxen andibuprofen in the Therapeutic Arthritis Research and Gastrointes-tinal Event Trial (TARGET), reduction in ulcer complications: ran-domised controlled trial. Lancet 2004;364:665–674.

4. Steinbach G, Lynch PM, Phillips RK, Wallace MH, Hawk E, GordonGB, Wakabayashi N, Saunders B, Shen Y, Fujimura T, Su LK,Levin B. The effect of celecoxib, a cyclooxygenase-2 inhibitor, infamilial adenomatous polyposis. N Engl J Med 2000;342:1946–1952.

5. Ott E, Nussmeier NA, Duke PC, Feneck RO, Alston RP, SnabesMC, Hubbard RC, Hsu PH, Saidman LJ, Mangano DT. Efficacy andsafety of the cyclooxygenase 2 inhibitors parecoxib and valde-coxib in patients undergoing coronary artery bypass surgery.J Thorac Cardiovasc Surg 2003;125:1481–1492.

6. Solomon SD, McMurray JJ, Pfeffer MA, Wittes J, Fowler R, Finn P,Anderson WF, Zauber A, Hawk E, Bertagnolli M. Cardiovascularrisk associated with celecoxib in a clinical trial for colorectaladenoma prevention. N Engl J Med 2005;352:1071–1080.

7. Bresalier RS, Sandler RS, Quan H, Bolognese JA, Oxenius B,Horgan K, Lines C, Riddell R, Morton D, Lanas A, Konstam MA,Baron JA. Cardiovascular events associated with rofecoxib in acolorectal adenoma chemoprevention trial. N Engl J Med 2005;352:1092–1102.

8. Nussmeier NA, Whelton AA, Brown MT, Langford RM, Hoeft A,Parlow JL, Boyce SW, Verburg KM. Complications of the COX-2inhibitors parecoxib and valdecoxib after cardiac surgery. N EnglJ Med 2005;352:1081–1091.

9. Tacconelli S, Capone ML, Sciulli MG, Ricciotti E, Patrignani P. Thebiochemical selectivity of novel COX-2 inhibitors in whole bloodassays of COX-isozyme activity. Curr Med Res Opin 2002;18:503–511.

10. McAdam BF, Catella-Lawson F, Mardini IA, Kapoor S, Lawson JA,FitzGerald GA. Systemic biosynthesis of prostacyclin by cycloox-ygenase (COX)-2: the human pharmacology of a selective inhibi-tor of COX-2. Proc Natl Acad Sci U S A 1999;96:272–277.

11. Catella-Lawson F, McAdam B, Morrison BW, Kapoor S, Kujubu D,Antes L, Lasseter KC, Quan H, Gertz BJ, FitzGerald GA. Effects ofspecific inhibition of cyclooxygenase-2 on sodium balance, hemo-dynamics, and vasoactive eicosanoids. J Pharmacol Exp Ther1999;289:735–741.

12. Cheng Y, Austin SC, Rocca B, Koller BH, Coffman TM, Grosser T,Lawson JA, FitzGerald GA. Role of prostacyclin in the cardiovas-cular response to thromboxane A2. Science 2002;296:539–541.

13. Qi Z, Hao CM, Langenbach RI, Breyer RM, Redha R, Morrow JD,Breyer MD. Opposite effects of cyclooxygenase-1 and -2 activityon the pressor response to angiotensin II. J Clin Invest 2002;110:61–69.

14. Kobayashi T, Tahara Y, Matsumoto M, Iguchi M, Sano H, Mu-rayama T, Arai H, Oida H, Yurugi-Kobayashi T, Yamashita JK,Katagiri H, Majima M, Yokode M, Kita T, Narumiya S. Roles ofthromboxane A2 and prostacyclin in the development of athero-sclerosis in apoE-deficient mice. J Clin Invest 2004;114:784–794.

15. FitzGerald GA. COX-2 and beyond: approaches to prostaglandininhibition in human disease. Nat Rev Drug Discov 2003;2:879–890.

16. FitzGerald GA. Coxibs and cardiovascular disease. N Engl J Med2004;351:1709–1711.

17. Furberg CD, Psaty BM, FitzGerald GA. Parecoxib, valdecoxib, andcardiovascular risk. Circulation 2005;111:249.

18. Israel E, Chinchilli VM, Ford JG, Boushey HA, Cherniack R, CraigTJ, Deykin A, Fagan JK, Fahy JV, Fish J, Kraft M, Kunselman SJ,Lazarus SC, Lemanske RF Jr, Liggett SB, Martin RJ, Mitra N,Peters SP, Silverman E, Sorkness CA, Szefler SJ, Wechsler ME,Weiss ST, Drazen JM. Use of regularly scheduled albuterol treat-ment in asthma: genotype-stratified, randomised, placebo-con-trolled cross-over trial. Lancet 2004;364:1505–1512.

19. Wood AJ. Racial differences in the response to drugs—pointersto genetic differences. N Engl J Med 2001;344:1394–1396.

20. Catella-Lawson F, Reilly MP, Kapoor SC, Cucchiara AJ, DeMarcoS, Tournier B, Vyas SN, FitzGerald GA. Cyclooxygenase inhibitorsand the antiplatelet effects of aspirin. N Engl J Med 2001;345:1809–1817.

21. Patrignani P, Filabozzi P, Patrono C. Selective cumulative inhibi-tion of platelet thromboxane production by low-dose aspirin inhealthy subjects. J Clin Invest 1982;69:1366–1372.

22. Panara MR, Renda G, Sciulli MG, Santini G, Di Giamberardino M,Rotondo MT, Tacconelli S, Seta F, Patrono C, Patrignani P. Dose-dependent inhibition of platelet cyclooxygenase-1 and monocytecyclooxygenase-2 by meloxicam in healthy subjects. J PharmacolExp Ther 1999;290:276–280.

23. FitzGerald GA, Pedersen AK, Patrono C. Analysis of prostacyclinand thromboxane biosynthesis in cardiovascular disease. Circu-lation 1983;67:1174–1177.

24. Catella F, FitzGerald GA. Paired analysis of urinary thromboxaneB2 metabolites in humans. Thromb Res 1987;47:647–656.

25. Wiltshire T, Pletcher MT, Batalov S, Barnes SW, Tarantino LM,Cooke MP, Wu H, Smylie K, Santrosyan A, Copeland NG, JenkinsNA, Kalush F, Mural RJ, Glynne RJ, Kay SA, Adams MD, FletcherCF. Genome-wide single-nucleotide polymorphism analysis de-fines haplotype patterns in mouse. Proc Natl Acad Sci U S A2003;100:3380–3385.

26. Yan J, Fine J. Estimating equations for association structures.Stat Med 2004;23:859–880.

27. Spiegelhalter DJ, Thomas A, Best NG. (1999). WinBUGS Version1.2 User Manual. Technical manual. MRC Biostatistics Unit,Cambridge, UK. Available at: http://www.mrc-bsu.cam.ac.uk/bugs/winbugs/contents.shtml. Accessed November 28, 2005.

28. Patrono C, Patrignani P, Garcia Rodriguez LA. Cyclooxygenase-selective inhibition of prostanoid formation: transducing bio-chemical selectivity into clinical read-outs. J Clin Invest 2001;108:7–13.

29. Thomas DW, Mannon RB, Mannon PJ, Latour A, Oliver JA, Hoff-man M, Smithies O, Koller BH, Coffman TM. Coagulation defectsand altered hemodynamic responses in mice lacking receptorsfor thromboxane A2. J Clin Invest 1998;102:1994–2001.

30. Francois H, Athirakul K, Mao L, Rockman H, Coffman TM. Role forthromboxane receptors in angiotensin-II-induced hypertension.Hypertension 2004;43:364–369.

31. Cayatte AJ, Du Y, Oliver-Krasinski J, Lavielle G, Verbeuren TJ,Cohen RA. The thromboxane receptor antagonist S18886 but notaspirin inhibits atherogenesis in apo E-deficient mice: evidencethat eicosanoids other than thromboxane contribute to athero-sclerosis. Arterioscler Thromb Vasc Biol 2000;20:1724–1728.

32. Egan KM, Lawson JA, Fries S, Koller B, Rader DJ, Smyth EM,FitzGerald GA. COX-2 derived prostacyclin confers atheroprotec-tion on female mice. Science 2004;360:1954–1957.

33. Wang D, Mann JR, DuBois RN. The role of prostaglandins andother eicosanoids in the gastrointestinal tract. Gastroenterology2005;128:1445–1461.

January 2006 VARIABILITY IN THE RESPONSE TO COXIBS 63

34. Malmstrom K, Daniels S, Kotey P, Seidenberg BC, Desjardins PJ.Comparison of rofecoxib and celecoxib, two cyclooxygenase-2inhibitors, in postoperative dental pain: a randomized, placebo-and active-comparator-controlled clinical trial. Clin Ther 1999;21:1653–1663.

35. Morrison BW, Fricke J, Brown J, Yuan W, Kotey P, Mehlisch D. Theoptimal analgesic dose of rofecoxib: overview of six randomizedcontrolled trials. J Am Dent Assoc 2000;131:1729–1737.

36. Markiewicz A, Semenowicz K. Time dependent changes in thepharmacokinetics of aspirin. Int J Clin Pharmacol Biopharm1979;17:409–411.

37. Clench J, Reinberg A, Dziewanowska Z, Ghata J, Smolensky M.Circadian changes in the bioavailability and effects of indometh-acin in healthy subjects. Eur J Clin Pharmacol 1981;20:359–369.

38. FDA. New Drug Application 20-998. Clinical Pharmacology andBiopharmaceutics Review Celecoxib. Bethesda, MD: Food andDrug Administration, 1998:1–60.

39. Hogenesch JB, Panda S, Kay S, Takahashi JS. Circadian tran-scriptional output in the SCN and liver of the mouse. NovartisFoundation Symposium 2003;253:171–183; discussion 52–55,102–109.

40. Drazen JM, Silverman EK, Lee TH. Heterogeneity of therapeuticresponses in asthma. Br Med Bull 2000;56:1054–1070.

41. Chladek J, Grim J, Martinkova J, Simkova M, Vaniekova J,Koudelkova V, Noiekova M. Pharmacokinetics and pharmacody-namics of low-dose methotrexate in the treatment of psoriasis.Br J Clin Pharmacol 2002;54:147–156.

42. Stahle L, Moberg L, Svensson JO, Sonnerborg A. Efavirenzplasma concentrations in HIV-infected patients: inter- and intra-individual variability and clinical effects. Ther Drug Monit 2004;26:267–270.

43. Ulrich CM, Bigler J, Sibert J, Greene EA, Sparks R, Carlson CS,Potter JD. Cyclooxygenase 1 polymorphisms in African-Americanand Caucasian populations. Hum Mutat 2002;20:409–410.

44. Papafili A, Hill MR, Brull DJ, McAnulty RJ, Marshall RP, HumphriesSE, Laurent GJ. Common promoter variant in cyclooxygenase-2represses gene expression: evidence of role in acute-phase in-flammatory response. Arterioscler Thromb Vasc Biol 2002;22:1631–1636.

45. Halushka MK, Walker LP, Halushka PV. Genetic variation in cy-clooxygenase 1: effects on response to aspirin. Clin PharmacolTher 2003;73:122–130.

46. Tang C, Shou M, Mei Q, Rushmore TH, Rodrigues AD. Major roleof human liver microsomal cytochrome P450 2C9 (CYP2C9) inthe oxidative metabolism of celecoxib, a novel cyclooxygenase-IIinhibitor. J Pharmacol Exp Ther 2000;293:453–459.

47. Lee CR, Goldstein JA, Pieper JA. Cytochrome P450 2C9 polymor-phisms: a comprehensive review of the in vitro and human data.Pharmacogenetics 2002;12:251–263.

48. Scordo MG, Pengo V, Spina E, Dahl ML, Gusella M, Padrini R.Influence of CYP2C9 and CYP2C19 genetic polymorphisms onwarfarin maintenance dose and metabolic clearance. Clin Phar-macol Ther 2002;72:702–710.

Received August 22, 2005. Accepted September 28, 2005.Address requests for reprints to: Garret A. FitzGerald, MD, Institute for

Translational Medicine and Therapeutics, University of PennsylvaniaSchool of Medicine, 153 Johnson-Pavilion, 3620 Hamilton Walk, Phila-delphia, Pennsylvania 19104-6084. e-mail: garret@spirit.gcrc.upenn.edu;fax: (215) 573-9135.

Supported by The National Heart Lung and Blood Institute (HL62250and HL54500), the National Center for Research Resources (MO1RR00040), the American Heart Association (Pennsylvania-DelawareAffiliate Postdoctoral Fellowship, to S.F.; Scientist Development Grant0430148N, to T.G.), the Wellcome Trust (London, United Kingdom)(T.S.P.: Training Fellowship in Mathematical Biology to T.S.P.), OrchidBiosciences (Princeton, NJ), and the Genomics Institute of the NovartisFoundation (genotyping).

S.F. and T.G. contributed equally to this paper.

64 FRIES ET AL GASTROENTEROLOGY Vol. 130, No. 1