### Animals

#### Study 1

Twelve healthy Beagle dogs (6 females and 6 males) were selected after clinical examination and biochemical analysis to exclude any underlying pathologies. The bodyweights and ages were 10 ± 2.5 kg and 1 ± 0.25 years, respectively. The dogs were usually housed in large boxes (2 per box). During the different periods of the trial the animals were kept in individual stainless steel cages in a controlled environment. Dogs were fed once a day with 250 g of commercial dry food (Medium, Royal Canin SA – Aimargues – France). On the day of the trial, they were fed on the evening after the last measurements. As the inflammatory model is totally reversible without sequelae, all the dogs that participated in this trial were rehomed as companion animals at the end of the study.

#### Study 2

Four different sized-breeds from four different private breeders were tested (ten individuals were used from each breed): Anglo-French Hounds, Pointers, Cavalier King Charles Spaniels and Bernese Mountain Dogs. The bodyweights of the dogs ranged between [27.2, 37.4], [16.8, 22.3], [7, 9.8], [44.5, 62.5] kg; the ages were [0.6, 2.5], [1.9, 11.8], [1.7, 6.7], [1.4, 6.4] years for the Anglo-French Hounds, Pointers, Cavalier King Charles Spaniels and Bernese Mountain Dogs breeds, respectively. The sex ratios (male/female) were 5/5, 6/4, 4/6 and 4/6 for these same breeds. Prior to the experiment, all the animals underwent a clinical examination by a veterinarian, along with a complete biochemistry profile. No clinically important abnormality was noted during these examinations.

Animal care and conduct of the study were performed according to the Guide for the Care and Use of Laboratory Animals. The protocol was approved by the animal experimentation ethics committee of Midi-Pyrenees (MP/01/41/09/08). The study was performed in compliance with the Principles of Good Clinical Practice (CVMP/VICH/595/98) and according to the Guideline for the Conduct of Efficacy Studies for NSAIDs (EMEA/CVMP/237/01).

### Experimental design

#### Study 1

The 12 dogs were regularly trained to get accustomed to experimental and measurements conditions for at least one month before the beginning of the experiment. Furthermore, all the investigators involved received a special training for endpoints measurements. The paw inflammation model used was the kaolin model (1.55 g of Kaolin aseptically injected in the skin under general anesthesia, [15]). The measured endpoints used to evaluate the antipyretic, analgesic and anti-inflammatory effects of the NSAID were the body temperature, the paw circumference, the time to perform a creeping test under a tunnel, the lameness score, the vertical force of the paw on force plates (normalized to the dog’s body weight), the thermal pain threshold and the plantar skin temperature [15]. For body temperature a single measurement was obtained at each time using an electronic thermometer. Paw circumference duplicate measures per time point was done just above the pad using a measuring tape (DMC, Colmar, France). Skin plantar temperature was measured (in degrees Celsius) using an infrared thermometer (Raynger M. Raytek, Fisher Bioblock Scientific - Illkirch – France) at a reference site on the plantar face of the paw, at six different times for each day and measures were performed in triplicate. Data used was the average of the three measurements performed at each time point. Creeping time was measured (in seconds) in a tunnel 6.38 m long, using a stopwatch (digital stopwatch) in triplicate for each time point and creeping speed (m/s) was used as the final measurement for this endpoint. Lameness score was assessed using a numerical rating scale as developed by Giraudel et al. [35]. Vertical normalized force i.e. the maximal vertical force applied on the ground by a hind limb was measured with force plates (SATEL Veto; Patrick Savet, Blagnac, France); for each measurement time, three measures were recorded, the mean values were normalized to the dog’s body weight (*F*max/b.wt., kg/kg) and used in the data analyses. Pain was assessed using a hind paw thermal escape model. The model consisted of exposing the hind limb to a light beam delivered by a Hargreaves apparatus (model 390; IITC Inc., Woodlands Hills, CA) [13, 36]. A heat intensity corresponding to 15% of the peak heating value of 150 W was selected. The paw withdrawal time (in seconds) was measured. For each trial, three measures were recorded per measurement time, and the mean of the three measures was used in the data analysis.

The study was conducted according to a classical 2×2 crossover design. Briefly, in period one, half of the dog (3 males and 3 females), received a single oral dose of cimicoxib (approximately 2 mg/kg, one 20 mg capsules – Vetoquinol Lure france) 26.5 h after the induction of kaolin inflammation on the left paw; the other half dogs received a placebo treatment at the same time. In period 2, treatments were crossed over for each dog and inflammation was induced on the right paw. A wash-out delay of at least 5 weeks was applied between the two periods. The treatment (cimicoxib or placebo) was blinded to the investigators.

Blood samples (5 mL, heparinized tubes) were obtained from the jugular vein by direct puncture just before inflammation to perform a biochemical checkup of the animals at the beginning of the experiment. Time 0 was defined as the time of treatment administration. Blood samples (5 mL, heparinized tubes) were collected at time 0 (just before cimicoxib or placebo administration) and then at times 30, 60, 90 min, and, 4, 6, 9, 15, 30 and 48 hours post-treatment. Tubes were centrifuged at 3000 G and plasma samples were stored at -20°C.

At least three days before the induction of kaolin inflammation, the PD endpoints were measured twice on the same day to generate negative control values. These endpoints were also measured at 23.5 and 25.5 h after the induction of the inflammation (i.e. before the treatment) to generate positive control values. Subsequently, the PD endpoints were measured at 2, 4, 9, 15 hours after administration (measures on all dogs) and at 22 h and 27 h after administration depending on the presence of further cimicoxib activity.

The results of this first study suggested that there might be two subpopulations of dogs diverging in their capacities to eliminate cimicoxib. To verify this hypothesis, a PK investigation of cimicoxib following IV administration (bolus, 2 mg/kg) was performed on the 12 dogs at least 2 months after the completion of the PK/PD study. Blood samples were obtained from the jugular vein by direct puncture at time 0 (control just before the IV administration) and then 2, 15, 30, 60, 90 min. and 2, 3, 4, 6, 9, 12, 24, 30 and 48 hours after the test article administration.

#### Study 2

The half-life of elimination of cimicoxib was studied in four different sized breeds (40 dogs). Two mg/kg of cimicoxib was administered orally, in the form of 10 and 20 mg (hard gelatin capsules, Vetoquinol). For each dog, four blood samples (5 mL, heparinized tubes) were taken, at 6 h, 8 h, 10 h and 24 h after administration of cimicoxib, centrifuged (3000 G) and stored (- 20°C).

### Assay of cimicoxib in plasma

Plasma samples were analyzed by a High Performance Liquid Chromatography (HPLC) method using ultraviolet (UV) detection. Briefly, an internal standard (UR-8877) and cimicoxib (UR-8880) were extracted from plasma by a solid liquid (methanol) extraction process using HLB Oasis cartridges (Waters). The extraction yield was about 90%. The HPLC apparatus consists of a pump system equipped with an automatic injector and an UV detector (242 nm). Separation was achieved by reverse phase column with an octadecylsilane stationary phase (Merck Lichrospher 100 RP18e (125×4) mm, 5 μm) using a guard column (Merck Lichrospher 100 RP18e (4×4) mm, 5 μm). The mobile phase was a mixture of 0.65 L of ultra-pure water and 0.35 L of acetonitrile which was delivered at a flow rate of 1 mL per min. Under these conditions, cimicoxib (UR-8880) and the internal standard (UR-8877) were eluted at retention times of 6 to 7 min and 10 to 11 min, respectively. The method was linear over the calibration range of 0.01 to 5 μg/mL using a linear model weighted by 1/X^{2}. Within-day and day-to-day coefficients of variation were less than 9% and the accuracy ranged from 94 to 103%, indicating an appropriate precision and accuracy for the analytical method. The lack of interference from endogenous compounds was verified on blank plasma from untreated dogs, establishing the specificity of the method. The validated limit of quantification was 0.01 μg/mL.

### Data analysis and modeling

#### Study 1

Pharmacokinetic and PK/PD modeling were performed by least-squares regression analysis using WinNonlin Professional software (WinNonlin® software, version 4.0.1, Pharsight Corporation, Mountain View, Ca, USA). For PK analysis, individual plasma cimicoxib concentrations were fitted to polyexponential equations. The data points were weighted by the inverse of the squared-fitted value. The number of exponential (2 or 3) terms needed to obtain the best fit for each data set was determined by the Akaike’s information criteria [37] and by inspecting the plot of residuals. A biexponential equation corresponding to a mono-compartmental model for extravascular administration with a lag-time was selected (Equation 1).

\mathrm{C}\left(t\right)=\frac{\mathit{\text{FD}}{k}_{01}}{V\left({k}_{01}-{k}_{10}\right)}\left[exp\left(-{k}_{10}\times \left(t-{t}_{\mathit{\text{lag}}}\right)\right)-exp(-{k}_{01}\times \left(t-{t}_{\mathit{\text{lag}}}\right)\right]

(1)

where C(t) is the cimicoxib plasma concentration (μg/L) at time t (h), V/F (L/kg) is the apparent volume of distribution, k01 (1/h) is the rate constant of the initial ascending phase, k10 (1/h) the rate constant of the terminal phase, tlag is the lag time and D is the cimicoxib dose (mg/kg). The parameters (V/F, k01 and k10 and tlag) were estimated.

For the IV study, a non-compartmental analysis was used to determined PK parameters.

For PD analysis, a percentage of improvement was calculated for all endpoints (except for the lameness score) following Equation 2:

\mathrm{\%}\phantom{\rule{0.6em}{0ex}}\mathit{\text{improvement}}\phantom{\rule{0.5em}{0ex}}=\left(\frac{{T}^{+}-M}{{T}^{+}-{T}^{-}}\right)*100

(2)

Where M is the measurement of the endpoint at different times after administration of cimicoxib, T^{-} is the mean negative control value (before inflammation) and T^{+} is the mean positive control value (after inflammation but before any treatment).

The maximum effect was defined as the mean % of improvement for a duration corresponding to a plateau of effect as identified by the visual inspection of the graph representing the % of improvement vs. time.

As the effect cimicoxib on the local temperature and paw oedema after inflammation were present but lower than the other effects, these endpoints were not used for the PK/PD analysis. Similarly the analgesic effect was discarded for modelling because it appeared to follow a complicate pattern suggesting a dual mechanism of action, namely an anti-nociceptive phenomenon and anti-hyperalgesic one (see Discussion), requiring a more advanced modelling with appropriate data that were not available in this experiment. Finally, four endpoints were considered as suitable for the PK/PD analysis aiming at modelling the relationship between cimicoxib plasma concentrations and its antipyretic and anti-inflammatory effects, namely: body temperature (antipyretic effect), creeping speed, vertical force of the paw and clinical lameness score (anti-inflammatory effects).

The selected PK/PD models were of the class of the indirect response models as proposed by Dayneka [38]. In these models, the measured response (R) is assumed to result from factors controlling either the development (Kin) or the dissipation (Kout) of the response according to Equation 3.

\raisebox{1ex}{$\mathrm{dR}$}\!\left/ \!\raisebox{-1ex}{$\mathrm{dt}$}\right.=\mathit{\text{Kin}}-\mathit{\text{Kout}}\times \mathrm{R}

(3)

where dR/dt is the rate of change of the response over time, Kin represents the zero-order rate constant for production of the response and Kout the first-order rate constant for loss of the response. The time development of measured response was modeled by a sigmoid-Emax (maximum stimulatory effect) or sigmoid-Imax (maximum inhibition effect) function (Hill models) that can be exerted either on Kin or Kout. The parameters generated by such analyses establishing the relationship between C(t), the plasma drug concentration at time *t*, and the effect, are the “Emax” or “Imax” for the drug effect (estimated maximum indirect effect), the “EC_{50}” or “IC_{50}” for the potency (estimated plasma concentration which causes 50% of Emax or Imax), and the Hill’s coefficient (n) for the sensitivity of the different concentration-effect curves. The effect of cimicoxib on the central temperature was based on a model assuming that during the inflammation phase, thermogenesis is increased (increase in Kin) while thermolysis is reduced (reduction in Kout) until a new balance for central temperature is reached. The antipyretic effect of an NSAID consisted of an increase in thermolysis (increase of Kout) enabling the central temperature to return to its control values, according to Equation 4:

\frac{\mathit{dR}}{\mathit{dT}}=\mathit{\text{Kin}}-\mathit{\text{Kout}}\times \left(1+\left(\frac{Emax\times {C}^{n}}{E{{C}_{50}}^{n}+{C}^{n}}\right)\right)\times R

(4)

For the creeping test consisting of crossing the tunnel more or less rapidly, an empirical model postulating that the dog’s speed (m.s^{-1}) is the result of a balance between its motivation to move forward (Kin) and a braking effect (Kout) associated with the severity of inflammation with a lower Kout in control conditions than during inflammation. During inflammation, pain increases the braking effect (i.e. increase Kout) and decreases the speed to cross the tunnel. The anti-inflammatory effect of cimicoxib consisted in reducing the pain (by decreasing Kout), which in turn increased the speed to cross the tunnel as described by Equation 5:

\frac{\mathit{dR}}{\mathit{dT}}=\mathit{Kin}-\mathit{Kout}\times \left(1-\left(\frac{\mathrm{Im}\mathit{ax}\times {C}^{n}}{I{{C}_{50}}^{n}+{C}^{n}}\right)\right)\times R

(5)

For the vertical force measured with the force plate apparatus normalized by the weight of the dog (without units), Kin represents the ground reaction forces in control resting conditions and Kout represents the factors prompting the withdrawal of limb from the ground in the case of inflammation due to pain *i.e.* that Kout is increased in the case of inflammation. Thus, the anti-inflammatory effect of cimicoxib consisted in a reduction of pain (decrease of Kout) and the same model as for the creeping speed under the tunnel was used (see Equation 5).

For the clinical lameness score, an empirical model was used postulating that the clinical score, from 0 (dog without lameness) to 5 (dog does not put limb on the ground), was the result of a balance between factors promoting lameness (Kin) and factors mitigating lameness (Kout). During inflammation, Kin is increased and the effect of cimicoxib was modelled using a concentration-dependent inhibition of Kin, according to Equation 6:

\frac{\mathit{dR}}{\mathit{dT}}=\mathit{\text{Kin}}\times \left(1-\left(\frac{{C}^{n}}{I{{C}_{50}}^{n}+{C}^{n}}\right)\right)-\mathit{\text{Kout}}\times R

(6)

#### Study 2

Half-life was estimated by log-linear regression (estimating the slope) using the non-compartmental approach implemented in Winnonlin.

### Statistical analysis

Statistical analysis was carried out using Systat (Version 8.0, SPS Onc., Chicago, IL). For study 1, the comparison between the two identified dog subpopulations for cimicoxib was done with an unpaired t-test. For study 2, the effect of the breed on cimicoxib plasma half-life was analyzed with a one-way ANOVA with the breed as fixed effect factor.