Differential inhibition of cyclooxygenase isoenzymes in the cat by the NSAID robenacoxib


  • Present address of J. M. Giraudel: Novartis Centre de Recherche Santé Animale SA, CH-1566 Saint-Aubin FR, Switzerland.

Prof. Peter Lees, Department of Veterinary Basic Sciences, Royal Veterinary College, Hawkshead Campus, North Mymms, Hatfield, Hertfordshire AL 9 7TA, UK. E-mail: plees@rvc.ac.uk


Robenacoxib is a new nonsteroidal anti-inflammatory drug (NSAID) developed for use in companion animal medicine. The objectives of this study were: to quantify the inhibitory actions of robenacoxib on cyclooxygenase (COX) isoenzymes in feline whole blood assays; to establish blood concentration–time profiles of robenacoxib after intravenous and subcutaneous dosing in the cat and; to predict the time courses of inhibition of COX isoforms by robenacoxib. COX-1 and COX-2 activities in heparinized feline whole blood samples were induced with calcium ionophore and lipopolysaccharide, respectively. Inhibition of thromboxane B2 provided a marker of both COX-1 and COX-2 activities and a nonlinear parametric mixed effects modelling approach was used to establish the pharmacodynamic parameters describing this inhibition. Mean values (and prediction intervals) of IC50 were 28.9 (16.4–51.1) μm (COX-1) and 0.058 (0.010–0.340) μm (COX-2). These parameters were used to compute several selectivity indices. Selectivity IC ratios (COX-1:COX-2) were 502.3 (IC50/IC50), 451.6 (IC95/IC95) and 17.05 (IC20/IC80). Based on a clinically recommended dosage regimen of 2 mg/kg, it was predicted that the corresponding mean robenacoxib blood concentration over the first 12 h after drug administration corresponded to 5% inhibition of COX-1 and 90% inhibition of COX-2.


Nonsteroidal anti-inflammatory drugs (NSAIDs) are extensively used in human and veterinary medicine. The principal mechanism of action of NSAIDs is inhibition of cyclooxygenase (COX), the rate limiting enzyme responsible for the conversion of arachidonic acid to prostaglandins and thromboxane A2. The identification of a second inducible COX gene (Kujubu et al., 1991) led to the hypothesis that the therapeutic effects of NSAIDs are due to inhibition of the corresponding encoded prostaglandin endoperoxide synthase, COX-2, whereas the toxic effects on the gastrointestinal tract and the kidney, and bleeding complications, are primarily or solely due to the inhibition of COX-1. This hypothesis provided the rationale for development of selective COX-2 inhibitors, the coxibs, with improved gastrointestinal tolerability and without effect on platelet aggregation. The original hypothesis has been modified. COX-2 is now known to be constitutive in several organs, including the kidney, so that toxic effects on that organ cannot be excluded with coxibs. Moreover, several authors have proposed roles for COX-1 (as well as COX-2) in generating mediators of acute inflammation (Wallace, 1999; Giuliano & Warner, 2002). Finally, roles for COX-2 derived mediators in mucosal defense and repair have been proposed (Straus et al., 2000; Gilroy et al., 2004).

Most COX-2 selective drugs are slow time-dependent reversible inhibitors of COX-2 but exert a simple competitive inhibitory action on COX-1 at higher concentrations (Lora et al., 1998; Walker et al., 2001). The time-dependent inhibition of COX-2 has been ascribed to an inhibitor-induced conformational change to a very high affinity complex, which, when it is only slowly reversible, accounts for pharmacological and therapeutic effects which persist after plasma concentrations have decreased below the IC50 (Lora et al., 1998).

In vitro isolated enzyme assays do not have sufficient accuracy to establish selectivity and potency for time-dependent inhibitors of COX-2 (Laneuville et al., 1994; Griswold & Adams, 1996; Blain et al., 2002). Cellular and whole animal estimates of potency and selectivity are required to confirm the results of in vitro isolated enzyme assays. However, the whole blood assays allow for time-dependent inhibition by NSAIDs, because the blood cells are preincubated with the drug being tested. These assays have additional advantages when the aim of the study is to investigate the clinical relevance of selective COX-2 inhibition (Pairet & Van Ryn, 1998). Indeed, the blood cells activated are target cells for the anti-inflammatory effects (monocytes) and side-effects (platelets) of NSAIDs. Whole blood used in COX-1 and COX-2 assays is taken from the same animal at the same time, prostaglandin or thromboxane synthesis is measured from arachidonic acid released from endogenous stores and the binding of the drug to plasma proteins which occurs in vivo is taken into account. Results obtained with whole blood assays have therefore been proposed to correlate reasonably well with biological activity of the test compounds in vivo (Patrignani et al., 1994; Pairet & Van Ryn, 1998).

Whilst some authors still categorise NSAIDs on the basis of chemical structure, classification according to pharmacological profile is increasingly used. NSAIDs are now usually described as selective COX-1 inhibitors, nonselective COX inhibitors (most classical NSAIDs), preferential COX-2 inhibitors (such as etodolac, nimesulide, meloxicam and diclofenac) and selective inhibitors of COX-2, when there is only very limited inhibition of COX-1 activity at therapeutic dosages. COX-2 selective inhibitors include etoricoxib, firocoxib, lumiracoxib, rofecoxib and valdecoxib. Robenacoxib is a new coxib similar in structure to lumiracoxib that is being developed for use in companion animal medicine. The pharmacological properties of robenacoxib in the rat were reported by King et al. (2008).

The objectives of this study were: 1) to use recently developed (and optimized) whole blood assays (Giraudel et al., 2005) to describe the inhibitory actions of robenacoxib on COX isoenzymes; 2) to optimize the analysis of data generated with these assays; 3) to establish new criteria of selectivity providing improved interpretation of in vitro data; and 4) to integrate the data with the results of pharmacokinetic studies to predict time courses of differential inhibition of COX-1 and COX-2 in vivo and to predict a possible dosage regimen.

Materials and methods

The studies were approved by the Royal Veterinary College Ethics and Welfare Committee.

In vitro study

Blood samples were taken from 10 healthy domestic short-haired cats of both sexes. The cats were maintained in a temperature controlled environment (20 ± 2 °C) and loose-housed in a colony. Weights and ages of cats ranged from 3.5 to 5.1 kg and 1 to 1.5 years, respectively. The maximum blood volume collected on any one occasion was 7 mL/kg b.w. Samples were collected from the jugular veins of fasted, sedated animals (intramuscular administration of 0.2 mg/kg midazolam obtained from Roche Products Ltd., Welwyn Garden City, Herts., UK) and 10 mg/kg ketamine (Fort Dodge Animal Health, Southampton, Hants., UK). Blood was withdrawn with 10-mL prefilled heparinized syringes (20 units heparin per mL of blood obtained from C P Pharmaceuticals Ltd., Wrexham, UK) connected to a 21G butterfly catheter (Terumo Europe N.V., 3001 Leuven, Belgium) inserted in the jugular vein for the duration of sampling. For each sample, blood from a single donor was mixed and then divided into 500 μL aliquots in loosely capped polypropylene tubes (Becton Dickinson Labware Europe, 38800 Le Pont De Claix, France). Half the tubes were used for the COX-1 assay and half for the COX-2 assay.

Assay conditions for evaluation of NSAID inhibition of COX-1 and COX-2 are summarized in Fig. 1. Optimal stimulus concentrations and incubation times were determined in a previous study (Giraudel et al., 2005). Briefly, COX-1 and COX-2 assays were standardized, such that time courses of incubation with the test compounds, endogenous supply of substrate and conditions of COX expression were as similar as possible in the two assays. COX-1 was stimulated using calcium ionophore (A23187, Sigma Aldrich Co. Ltd., Dorset, UK) and COX-2 was stimulated by lipopolysaccharide derived from Escherichia coli serotype 026:B6 (Sigma Aldrich Co. Ltd). At the end of incubation, eicosanoid production was stopped by placing the tubes on ice. Plasma was collected following centrifugation (10 min at 2000 g and 4 °C) and frozen at −20 °C prior to use. Inhibition of thromboxane B2 (TxB2) production was used as a measure of inhibition of both COX-1 and COX-2 activity. Thromboxane B2 and TxB2 antiserum were obtained from Sigma Aldrich Co. Ltd. and Tracer [3H] TxB2 was obtained from Amersham Pharmacia Biotech Ltd., Amersham, Herts., UK.

Figure 1.

 Time courses of incubation of whole blood with stimuli and test compound (robenacoxib) in dimethylsulphoxide (DMSO) in COX-1 and COX-2 assays. Feline blood was stimulated (a) with calcium ionophore (A23187) which leads to platelet activation with subsequent formation of TxB2 by COX-1 or (b) with bacterial lipopolysaccharide (LPS), after which TxB2 production is mainly due to COX-2 induction in monocytes.

Inhibition of COX-1 and COX-2 was assessed by adding a range of concentrations of the test compound (robenacoxib, Novartis Animal Health Inc., Basel, Switzerland) dissolved in dimethyl sulphoxide (DMSO obtained from Sigma Aldrich Co. Ltd) to blood samples. The final concentrations of robenacoxib used were 0.05; 0.2; 0.5; 2; 5; 10; 20; 50; 100; 200; 500; 1000 μm for the COX-1 assay and 0.005; 0.01; 0.02; 0.05; 0.1; 0.2; 0.5; 2; 5; 20; 50; 200 μm for the COX-2 assay. Assay validation procedures and data are as described by Giraudel et al. (2005).

Plasma concentrations of TxB2 were determined in duplicate as previously described (Higgins & Lees, 1984) by radioimmunoassay without extraction after dilution of samples in assay buffer (1:10, 1:30, 1:100 or 1:500 depending on the concentration of TxB2). For each blood sampling, four standard curves were prepared using known concentrations of TxB2 (20, 50, 100, 200, 500, 1000, 2000, 5000 and 10000 pg/mL), each standard curve incorporating 1:10, 1:30, 1:100 or 1:500 TxB2-free plasma. For each sample, two dilutions were prepared to ensure at least one reading on the linear part of the standard curve. The detection limit was 20 pg/mL (i.e., 0.2 ng/mL for the 1:10 dilution).

Enzyme inhibition was expressed as a percentage of the control value (robenacoxib concentration = 0 μm). Percentage inhibition in relation to test compound blood concentration was fitted with the software Scientist (MicroMath Research, St Louis, MO, USA) using the following Hill equation:


where %Inhibition is the test compound inhibition expressed as percentage of the control value, C is the test compound concentration (as independent variable), IC50 (expressing potency) is the test compound concentration giving 50% of Imax, I0 is the baseline inhibition, Imax + I0 (expressing efficacy) is the maximal response achieved by the test compound, and n is the Hill coefficient (expressing sensitivity) giving the slope of the concentration-effect relationship.

Several approaches were used to describe inhibition of COX isoforms by robenacoxib. The two-stage method (individual analyses followed by calculation of mean parameters) permitted determination of a set of mean parameters describing inhibition profiles of COX-1 and COX-2 by robenacoxib. For each COX isoenzyme, values for IC50, Imax, I0 and n were then used as initial estimates for the simultaneous analysis of all data sets using a nonlinear parametric mixed effects model. The estimation method was a First Order Maximum Likelihood Method using the software WinNonMix (Pharsight Corporation, Mountain View, CA, USA). The following Hill equation was used as the pharmacodynamic model:


where % Inhibij and Cij represent the inhibition percentage and the concentration for the ith concentration level and the jth cat, I0j, Imaxj and nj the baseline inhibition, the efficacy and the parameter describing selectivity in the jth cat and εij is the residual error term for the ith concentration level and the jth cat.

The between subject variability of PD parameters was described by the following statistical model:


where IC50j represents the potency in the jth cat (j = 1–10), IC50avg the geometric mean for the potency and η(IC50j) represents the deviation from the average value for the jth cat; the η(IC50j) were assumed to be normally distributed.

A similar model was used for the parameter describing selectivity (n). For the two other fixed variables (Imax and I0), another model described the between subject variability (example given for I0):


where I0j represents the baseline inhibition in the jth cat (j = 1–10), I0avg the arithmetic mean for the baseline inhibition and η(I0j) represents the deviation from the average value for the jth cat; the η(I0j) were assumed to be normally distributed.

The inclusion of a positive control was considered to be unnecessary, as all assay procedures have previously been developed and appropriately validated (Giraudel et al., 2005).

Thromboxane B2 concentrations of control samples in both assays are presented in Table 1. In the COX-2 assay TxB2 was present at a concentration of 0.82 ng/mL when the test compound was added (AT0) (Table 1). As this TxB2 production cannot be suppressed by the test compound, TxB2 concentrations present 1.5 h after commencing the incubation were subtracted from the final TxB2 concentrations before calculation of inhibition percentages in the COX-2 assay.

Table 1.   TxB2 plasma concentration (ng/mL) of controls used in whole blood assays (results are expressed as mean ± SD for 10 assays)
COX-2 assayCOX-1 assay
ControlsTxB2 conc. (ng/mL)ControlsTxB2 conc. (ng/mL)
  1. AT0: pretreatment with aspirin but incubation stopped just before adding test compound (t + 1.5 h); S: 8-h incubation with aspirin, DMSO and saline; A0: blood pretreated with aspirin (10 μg/mL) and 6-h incubation with LPS (100 μg/mL in saline); LL: 8-h incubation without aspirin but with DMSO and LPS; TT: no incubation (blood samples centrifuged just after sampling); CT0: incubation stopped immediately before adding robenacoxib (t + 1.5 h); CT: 8-h incubation with DMSO; C0: 4-h incubation with DMSO (test compound, 0 μm) and 4-h incubation with A23187 (50 μm in DMSO).

AT00.82 ± 0.68TT0.71 ± 0.62
S2.93 ± 1.23CT01.25 ± 1.51
A05.00 ± 0.96CT12.54 ± 6.53
LL17.07 ± 9.61C0217.02 ± 167.14

Pharmacokinetic study

For the pharmacokinetic study eight healthy domestic short-haired cats of both sexes were maintained in individual stainless steel cages in a temperature controlled environment (20 ± 2 °C). Weights and ages of cats ranged from 4.1 to 5.3 kg and 1.5 to 2 years. The cats were randomly allocated to two groups. Three days prior to the first robenacoxib administration a jugular catheter (4.0 French, 8.0 cm long with a 17 cm long extension tube, V-PUM-401J-V8-UQ, Cook Veterinary Products Inc., Bloomington, IN, USA) was inserted in each cat under sedation (subcutaneous administration of 1 mg/kg xylazine (Bayer plc Animal Health, Newbury, Berks., UK) followed 20 min later by intravenous administration of 5 mg/kg ketamine). On the day preceding the intravenous administration of robenacoxib, a cephalic venous catheter was placed under sedation in each of the four cats.

A 2*2 cross-over design (0.1 mg/kg robenacoxib intravenously vs. 1 mg/kg robenacoxib subcutaneously with a wash-out period of 4 days) was used. For intravenous administration a 1 mg/mL solution of robenacoxib was prepared in tris(hydroxymethyl)aminoethane, propylene glycol and Water for Injection and administered through the cephalic catheter. For the subcutaneous administration a 10 mg/mL solution (Novartis Animal Health Inc.) was injected on the dorsal mid-line in the region of the last thoracic vertebrae.

Blood samples (1 mL) were withdrawn from the jugular catheter into EDTA tubes, before dosing and 2, 5, 10, 30 and 60 min and 2, 4, 6, 8, 12 and 22 h post intravenous administration or 5, 20, 40 and 60 min and 2, 4, 6, 8, 12 and 22 h post subcutaneous administration. The blood samples were stored at –20 °C until assayed.

A sensitive GLP validated analytical method using high-pressure liquid chromatography with ultraviolet detector (HPLC-UV) and liquid chromatography-mass spectrometry (LC-MS) following solid phase extraction was used (Jung et al., 2008). Briefly, the method involved an initial analysis of an unknown specimen by HPLC-UV, covering the concentration range of approximately 500–20 000 ng/mL and, if required, a subsequent analysis by LC-MS, covering the range of approximately 3–100 ng/mL. Depending on the results obtained with the HPLC-UV method, some samples were diluted, in order not to exceed a concentration of 100 ng/mL in the LC-MS method. The limit of quantification (LOQ) was 3 ng/mL.

Pharmacokinetic parameters and variables were calculated by noncompartmental analysis using WinNonlin Professional software (WinNonlin®, version 4.0.1; Pharsight Corporation). All values were expressed as arithmetic mean ± SD except for the terminal half-life for which the harmonic mean and the corresponding 95% prediction interval (95% PI) was calculated.

Prediction of dosage for efficacy and safety

Prediction of in vivo doses for efficacy and safety can be undertaken using in vitro determined concentrations, albeit with caution and with qualifications and provided they have been obtained in whole blood assays, using the following equation:


where Dose/Dosing Interval is the dose to be administered in vivo to give, in steady-state conditions, an average concentration over the dosing interval equal to the Target Concentration. Total Clearance is the total blood clearance of the test compound, F is the bioavailability and Target Concentration is the blood concentration of interest (in terms of beneficial or side-effects). The Target Concentration is usually an in vivo concentration that has been demonstrated to be therapeutically both efficacious and safe.


Inhibition profiles of robenacoxib for COX-1 and COX-2

To determine the magnitude of COX inhibition by robenacoxib, individual animal data for each of 10 cats were analysed, followed by calculation of mean parameters (the two-stage method). Distributions of individual parameters for IC50 and n were assumed to be log-normal. Potency and selectivity were therefore best described by the geometric mean and corresponding prediction interval (Table 2). A second approach to data processing comprised averaging the inhibition percentages for each concentration of robenacoxib and fitting it, as if it was the curve of a single cat (Fig. 2). This is the naïve averaging of data approach. In a third approach, simultaneous fitting of the 10 inhibition curves for each COX isoenzyme, also known as naïve pooled data analysis, allowed calculation of a further set of mean parameters but none of these two last sets of figures are presented because these approaches have potential drawbacks that may lead to a distortion of computed mean parameters and do not provide adequate determination of data variability. Therefore, a fourth approach utilising nonlinear parametric mixed effects modelling, as classically used in population analysis, was undertaken. This approach allowed an accurate estimation of the average parameters describing COX inhibition and determination of the dispersion of individual parameters in the population, as described by prediction intervals (Table 3). These mean values were then used to compute several selectivity/safety indices that are likely to be informative of the potential clinical value of in vitro determined COX-2 selectivity (Table 4). The less than maximal inhibition of COX-2 (Imax + I0 of 87.6%) is believed to be model-dependent, as a similar finding has been reported for two other NSAIDs (Giraudel et al., 2005). Therefore, it should not be inferred that maximal inhibition of COX-2 cannot be achieved with robenacoxib. Inhibition percentages for both isoenzymes were consequently re-scaled on 0–100% scales to compute selectivity indices and to predict inhibition percentages of COX-1 for different inhibition percentages of COX-2.

Table 2.   Mean parameters of the Hill equation describing COX-1 and COX-2 inhibition by robenacoxib in whole blood assays
Parameters*COX-1 assayCOX-2 assay
  1. *Results are expressed as arithmetic mean (I0 and Imax + I0) or geometric mean (IC50 and n) and corresponding 95% prediction interval (PI) for 10 cats. Mean parameters for both isoenzymes were determined by averaging parameters of the individual Hill equations (2-stage method). I0 is the baseline inhibition, Imax + I0 (maximal response) corresponds to the true efficacy of robenacoxib for both isoenzymes (see Methods section), IC50 is the concentration producing 50% of Imax and n represents the Hill coefficient.

I0 (%)−17.9−24.8
Lower 95% PI−60.7−75.0
Upper 95% PI24.925.4
Imax + I0 (%)101.488.5
Lower 95% PI84.768.3
Upper 95% PI118.2108.6
Lower 95% PI4.90.011
Upper 95% PI102.20.390
Lower 95% PI0.370.37
Upper 95% PI1.702.13
Figure 2.

 Mean concentration response curves for COX-1 and COX-2 inhibition by robenacoxib in whole blood assays. Inhibition percentages (mean + SEM) were calculated by averaging for each concentration individual inhibition percentages of 10 cats. For each cat inhibition was expressed as percentage of the individual control value (robenacoxib concentration = 0 μm). The average percentages were then fitted using a Hill equation (naïve averaging of data approach).

Table 3.   Average values (geometric mean for IC50 and n and arithmetic mean for I0 and Imax + I0) for the parameters of the Hill equation describing COX-1 and COX-2 inhibition by different concentrations of robenacoxib in 10 cats
Parameters*COX-1 assayCOX-2 assay
  1. *Parameters for both isoenzymes were determined using a nonlinear parametric mixed effects model (see Methods section). For each parameter a 95% prediction interval (PI) comprising 95% of individual parameters were derived.

I0 (%)−13.8−24.8
Lower 95% PI−53.2−54.6
Upper 95% PI25.65.0
Imax + I0 (%)103.187.6
Lower 95% PI102.873.4
Upper 95% PI103.4101.8
IC50 (μM)28.90.058
Lower 95% PI16.40.010
Upper 95% PI51.10.340
Lower 95% PI0.550.37
Upper 95% PI1.261.79
Table 4.   Three categories of indices describing the selectivity of robenacoxib in feline whole blood assays
  1. *All indices were computed with the parameters of the Hill equations for COX-1 and COX-2 and determined using a nonlinear parametric mixed effects model (see Methods section). Corresponding mean percentages, initially ranging from I0 to Imax + I0, were re-scaled on 0–100% scales. Classical selectivity ratios are presented. Selectivity is also expressed as suggested safety factors, which are indicated as a ratio of a cut-off concentration corresponding to a level of COX-1 inhibition above which unacceptable side-effects might occur divided by a concentration producing a level of inhibition of COX-2 which might be required for therapeutic efficacy. The third category of indices gives the percentage by which COX-1 is inhibited for a given inhibition percentage of COX-2.

Classical selectivity ratios (ICX COX-1/ICX COX-2)
Other selectivity ratios (ICX COX-1/ICY COX-2)
% Inhibition of COX-1 for a fixed % inhibition of COX-2 (on 0–100% scales)
 % Inhibition of COX-1 for IC50 COX-20.56
 % Inhibition of COX-1 for IC80 COX-22.31
 % Inhibition of COX-1 for IC90 COX-25.17
 % Inhibition of COX-1 for IC95 COX-210.52
 % Inhibition of COX-1 for IC99 COX-239.17

There was clear separation of COX inhibition profiles (Fig. 2), an IC50 COX-1/IC50 COX-2 ratio of 502, an IC80 COX-1/IC80 COX-2 ratio of 478, an IC20 COX-1/IC80 COX-2 ratio of 17 and only 10.5% inhibition of COX-1 for 95% inhibition of COX-2 (Table 4).

Pharmacokinetic profile of robenacoxib and predicted COX inhibitions

Blood concentration–time profiles in eight cats are presented in Fig. 3. After intravenous administration the total blood clearance was relatively high (0.63 ± 0.06 L/kg/h) and the steady-state volume of distribution relatively small (0.20 ± 0.02 L/kg). The bioavailability from the subcutaneous route of administration was high (93.7 ± 10.0%) and peak blood concentration (Cmax 960 ± 241 μg/L) was achieved rapidly (Tmax 0.7 ± 0.2 h) after administration of 1 mg/kg robenacoxib. The terminal half-lives were short for both routes of administration: 0.34 h (95% PI: 0.28–0.43) and 0.97 h (95% PI: 0.72–1.48) for the intravenous and subcutaneous routes, respectively.

Figure 3.

 Semi-logarithmic plot of mean robenacoxib blood concentration (μg/L) vs. time (h) after intravenous administration of a dose of 0.1 mg/kg and subcutaneous administration of 1 mg/kg in eight cats in a cross-over design. Vertical lines represent SD.

Prediction of dosage for efficacy and safety

Taking IC50 for COX-2 inhibition as the lowest target concentration, the predicted dose of robenacoxib is 0.3 mg/kg/24 h (or 1.5 mg/kg/24 h if the target concentrations is the IC80). All dose predictions with target concentrations corresponding to different inhibition levels of COX-1 and COX-2 are presented in Table 5.

Table 5.   Predicted robenacoxib doses corresponding to several levels of COX-1 and COX-2 inhibition and predicted COX-1 and COX-2 inhibition levels for a proposed clinically effective dose of robenacoxib of 2 mg/kg subcutaneously
  1. *Predictions are based on in vitro results obtained using a nonlinear parametric mixed effects model (10 cats) and in vivo pharmacokinetic data obtained after a subcutaneous administration of 2 mg/kg robenacoxib (10 cats) (Giraudel et al., 2008).

  2. Average blood concentration over the first 12 h following robenacoxib administration, which corresponds approximately to the time period during which the robenacoxib blood concentrations remained above 3 ng/mL.

Predicted in vivo dose for 50% inhibition of COX-2 (mg/kg/24 h)0.27
Predicted in vivo dose for 80% inhibition of COX-2 (mg/kg/24 h)1.53
Predicted in vivo dose for 95% inhibition of COX-2 (mg/kg/24 h)10.50
Predicted in vivo dose for 10% inhibition of COX-1 (mg/kg/24 h)9.81
Predicted in vivo dose for 20% inhibition of COX-1 (mg/kg/24 h)26.01
In vivo maximum blood concentration (μm) obtained with a dose of 2 mg/kg administered s.c.5.30
% inhibition of COX-1 corresponding to Cmax19.6
% inhibition of COX-2 corresponding to Cmax97.5
In vivo average blood concentration (μm) obtained with a dose of 2 mg/kg administered s.c.0.85
% inhibition of COX-1 corresponding to the average concentration5.1
% inhibition of COX-2 corresponding to the average concentration89.8


Assay design and data analysis

Patrignani and co-workers introduced a whole blood assay that has subsequently been used extensively to assess the biochemical selectivity of NSAIDs as inhibitors of COX isoforms (Patrignani et al., 1994). However, a disadvantage of this test system is the time discrepancy between the COX-1 and the COX-2 assays (Giuliano & Warner, 1999; Giuliano et al., 2001). Therefore, in the present study the incubation times for the two assays were identical and TxB2 was used as a marker of both COX-1 and COX-2 activities. As discussed previously (Giraudel et al., 2005) Thromboxane B2 is preferred to prostaglandin E2 (PGE2) for the COX-2 as well as the COX-1 assay, as lipopolysaccharide induces a strong expression of PGE2 synthase (Matsumoto et al., 1997; Dieter et al., 2000). This may potentially bias the measurement of COX-2 activity when it is determined by PGE2 concentration. For the COX-2 assay it has already been shown that 10 μg/mL aspirin is the optimal concentration to avoid a contribution of COX-1 to TxB2 production, without causing significant inhibition of COX-2 (Giraudel et al., 2005). The marked difference in TxB2 concentrations between A0 and LL in the present study reinforces the opinion that blood should be pretreated with aspirin when TxB2 is used as a marker of COX-2 activity (Demasi et al., 2000).

With very low concentrations of robenacoxib, there was an apparent small stimulation of TxB2 production compared to control samples; this was taken into account in our mathematical model by a negative inhibition (Fig. 2). The induction of COX-1 and COX-2 by low concentrations of inhibitors has been described previously with acetaminophen and for some NSAIDs (Swinney et al., 1997; Toutain et al., 2001; Giraudel et al., 2005). On the other hand, with very high concentrations of both robenacoxib (greater than 100 μm) and other test compounds (Giraudel et al., 2005), TxB2 production increased again, whereas for concentrations in the range 10–100 μm TxB2 production was almost completely inhibited. The clinical significance of these findings is uncertain because these high concentrations cannot be achieved in vivo with clinically recommended dosages. Some authors have described an NSAID-induced COX-2 expression that could account for this increase in TxB2 production (Simmons et al., 1999; Botting, 2000). Moreover, it is interesting to note that diclofenac was particularly effective in inducing an acetaminophen-sensitive COX-2 activity and robenacoxib and diclofenac have similar structures. Diclofenac is a COX-2 preferential inhibitor that induces a time-dependent, pseudoirreversible inhibition of COX-2 (Brideau et al., 1996; Patrignani et al., 1997; Cryer & Feldman, 1998; Giuliano & Warner, 1999;Simmons et al., 1999; Wearner et al., 1999). The higher selectivity of robenacoxib for COX-2 is probably attributable to the acetyl and fluorine substituents on the phenyl rings, which increase the steric bulk.

Inhibition profiles of COX-1 and COX-2 by robenacoxib demonstrate that it is indeed one of the most COX-2 selective drugs, possibly even more selective than valdecoxib, rofecoxib and etoricoxib, for which IC50 COX-1/IC50 COX-2 ratios in a human whole blood assay (Brideau et al., 1996) were between 30 and 106 (Riendeau et al., 2001). These results require confirmation using the same assay conditions in the same species. Lumiracoxib, a drug whose structure is very similar to robenacoxib, has approximately the same selectivity in human whole blood assays (IC50 COX-1/IC50 COX-2 ratio = 400) (Capone et al., 2003).

Parameters presented in Table 2 were computed using the so-called 2-stage method, the most common approach when processing data obtained in whole blood assays. Many authors do not indicate the approach used and naïve averaging of data (Cryer & Feldman, 1998) or naïve pooled data analyses (Panara et al., 1998) have probably also been extensively performed. A major advantage of using a nonlinear parametric mixed effects model to analyse the data is that it provides unbiased estimates of the parameters of interest. From these parameters were derived 95% PI comprising 95% of individual parameters (Table 3). Another advantage is the possibility of incorporating sparse data in the overall analysis. In this study the data were rich enough to derive individual parameters for all cats, but when only few NSAID concentrations can be tested because of smaller blood volumes taken this is not possible and only a so-called population approach can be used to analyse such sparse data sets.

Indices of COX selectivity

As the inhibition profiles for COX-1 and COX-2 may not be parallel, selectivity ratios should be based on clinically relevant inhibition percentages (Wearner et al., 1999; Giraudel et al., 2005), for example 80–95% inhibition of COX-2 and 0–20% inhibition of COX-1. For NSAIDs there is uncertainty concerning the precise levels of COX inhibition associated with an adequate therapeutic response on the one hand and the occurrence of side-effects on the other. Nevertheless, it has been suggested that most NSAIDs probably inhibit COX-2 by 80% or more in vivo when used at therapeutic doses (Wearner et al., 1999; Lees, 2003). It is indeed probable that a drug for which more than 80–90% COX-2 inhibition and less than 10–20% COX-1 inhibition is achieved over most of the dosing interval is likely to be relatively safe in terms of gastrointestinal and platelet side-effects.

When considering both IC50 and IC80 COX-1/COX-2 ratios, the selectivity for COX-2 of robenacoxib exceeded 450, as inhibition profiles for the two isoenzymes were almost parallel. When the Hill coefficients for the curves are dissimilar, however, it becomes particularly relevant to calculate other selectivity indices. For example, an in vitro IC10 COX-1/IC90 COX-2 ratio greater than unity or less than 20% inhibition of COX-1 for 95% inhibition of COX-2, as obtained with robenacoxib, demonstrates high COX-2 selectivity for clinically relevant inhibition levels, whether the curves for COX-1 and COX-2 inhibition are parallel or not.

The potential clinical relevance of differential inhibition of COX-1 and COX-2 is further indicated by the plot of inhibition percentage of COX-2 against corresponding inhibition percentage of COX-1 (Fig. 4). By adding to the graph cut-off values for each COX isoform, it becomes clear that even for high levels of robenacoxib mediated COX-2 inhibition (80–95%) corresponding COX-1 inhibition remains low (less than 20%). For comparative purposes, data has also been plotted for a reference drug, meloxicam, in the cat using the same assay conditions (Giraudel et al., 2005). There was virtually no meloxicam concentration for which COX-1 was not inhibited by at least 20% and, in the predicted range of therapeutic concentrations (80% or greater inhibition of COX-2), COX-1 was inhibited by more than 40%. The similarity of these meloxicam data with results obtained in humans is striking (Panara et al., 1999). For both species, the preferential COX-2 inhibition obtained in vitro is not sufficient to clearly separate the effects of meloxicam on the two isoenzymes when clinically relevant concentrations are considered.

Figure 4.

 Inhibition percentages of COX-2 and corresponding inhibition percentages of COX-1 for a range of concentrations of robenacoxib. Mean inhibition percentages for both isoenzymes were computed with the average values determined using the population approach. Percentages, initially ranging from 0 to Imax, were re-scaled on 0–100% scales. For given mean inhibition percentages of COX-2 corresponding robenacoxib concentrations were calculated using the average Hill equation for COX-2 and these concentrations were then incorporated in the average Hill equation for COX-1 to calculate corresponding COX-1 inhibitions. Dotted lines indicate cut-off values for inhibition of COX-1 (20% inhibition COX-1 = percentage above which a risk of side-effects is assumed) and COX-2 (80% inhibition for COX-2 = percentage above which a good therapeutic effect is expected). Data for meloxicam generated in a previous investigation are included for comparison.

Ex vivo whole blood assays, in which the drug is administered directly to the animal or patient, are a further development in determination of the likely in vivo COX-2 selectivity of test compounds. Ex vivo assays have the advantage of taking into account activity due to both parent drug and active metabolites. Nevertheless, when the contribution of active metabolites is limited, it has been shown that predictions based on in vitro data can correlate well with ex vivo determined COX inhibitions at therapeutic concentrations. This has been shown for carprofen (Taylor et al., 1996; Giraudel et al., 2005), meloxicam (Panara et al., 1999) and etoricoxib (Dallob et al., 2003). In this circumstance the in vitro inhibition profile of the drug can serve as a surrogate for the effect of a NSAID on COX-1 and COX-2 in vivo (Cryer & Feldman, 1998; Panara et al., 1999).

Prediction of dosage for efficacy and safety

For robenacoxib administered intravenously at a dosage of 0.1 mg/kg, blood clearance was relatively high, the steady-state volume of distribution was low and, in consequence, the terminal half-life was short. However, in a pharmacokinetic study in the cat using a robenacoxib dosage of 2 mg/kg administered intravenously, noncompartmental analysis provided a smaller body clearance (0.44 vs. 0.63 L/kg/h) and a longer terminal half-life (1.43 vs. 0.34 h) than the values obtained with the much lower intravenous dose (0.1 mg/kg) in this study (Jung et al., unpublished). This latter finding is attributable to the fact that the dose was too small in the current study to characterize adequately the terminal phase of elimination.

Using the in vitro IC80 for COX-2 inhibition as a target blood concentration, a daily dose of 1.5 mg/kg was computed.

Figure 5 illustrates the mean blood concentration–time profile of robenacoxib in 10 cats receiving a 2 mg/kg dose subcutaneously in a previous investigation (Giraudel et al., 2008). Peak concentrations were achieved rapidly (Cmax of 1736 ± 409 μg/L at an observed Tmax of 0.9 ± 0.1 h) and the terminal half-life was short (1.09 h, 95% PI: 0.82–1.63, noncompartmental analysis) (Giraudel et al., 2008).

Figure 5.

 Arithmetic plot of mean robenacoxib blood concentration (μg/L) vs. time (h) after subcutaneous administration of 2 mg/kg in 10 cats. Vertical lines indicate SD (data from Giraudel et al., 2008). In vitro determined cut-off values for COX-1 (10% and 20%) and COX-2 (80% and 95%) inhibition are indicated by horizontal lines.

Our predictions indicate that the mean robenacoxib blood concentration over the first 12 h following drug administration corresponds to 5% inhibition of COX-1 and 90% inhibition of COX-2, with COX-1 inhibition never exceeding 20% (Table 5).

For a NSAID such as robenacoxib with a short terminal half-life, predictions of in vivo efficacy based on in vitro enzyme inhibition do not fully allow for the slowly reversible dissociation process that is classically described for COX-2 selective drugs, nor is the likelihood of ready accumulation in and slow clearance from inflammatory sites taken into account. This latter feature has been extensively studied using a tissue cage model of carrageenan-induced inflammation to measure NSAID concentrations and corresponding molecular effects (inhibition of PGE2 production, etc.) in inflammatory exudate (Landoni et al., 1995, 1999). A similar finding has been reported for robenacoxib in the rat (King et al., 2008) and cat (Pelligand et al., unpublished data). The chemical structure of robenacoxib closely resembles that of lumiracoxib and the lipophilic character of the latter drug enables rapid distribution to sites of inflammation to achieve good anti-inflammatory activity (Mukherjee et al., 1996; Torres-Lopez et al., 1997; Waver et al., 2003). If similar considerations hold for robenacoxib, it would be expected to possess a rapid onset of anti-inflammatory response and to maintain activity after plasma concentrations have decreased below detectable levels.

Although the potency for inhibition of gastric PGE2 synthesis correlates well with potency for COX-1 inhibition in whole blood for a large number of NSAIDs (Cryer & Feldman, 1998), the preferred approach remains assessment of inhibition of COX-1 activity in the gastric mucosa (Dallob et al., 2003). Consequently, definitive proof of a desirable degree of COX-2 selectivity can be obtained only in vivo when, in the relevant tissues such as gastric mucosa and synovial tissue (Dallob et al., 2003; Duffy et al., 2003) and across the entire therapeutic concentration range, there is significant suppression of COX-2 activity with slight or no effect on COX-1.

Finally the therapeutic response to NSAIDs may not be related solely to COX-2 inhibition. Some studies indicate that COX-1 may contribute to the generation of pro-inflammatory prostanoids and hence to clinical signs, and some NSAIDs have been shown to possess actions other than COX inhibition [decrease in cytokine (TNF-α, IL-6) or COX-2 production, increase in inducible nitric oxide synthase production, inhibition of NF-kappa B activation] (Armstrong & Lees, 2002; Chen et al., 2004; Coruzzi et al., 2004). These findings, together with the existence of side-effects unrelated to COX inhibition, emphasize that the ultimate demonstration of both safety and efficacy must be made in clinical studies. It is nevertheless necessary to make initial predictions on the basis of in vitro COX inhibition data when selecting new drugs and doses for subsequent preclinical and clinical testing.


The work was supported by Novartis Animal Health Inc., Switzerland. The authors would also like to thank Katey Gardner and Mike Andrews from the Royal Veterinary College for skilled technical assistance.