Pharmacokinetic Appraisal of Carprofen Delivery from Intra-Articular Nanoparticles: A Population Modeling Approach in Rabbits †

: Osteoarthritis is frequently treated in veterinary settings with non-steroidal anti-inflam-matory drugs (NSAID) such as carprofen (CP). Its action over the articular cartilage can be enhanced by increasing drug uptake into the cartilage, alongside its site of action, and anticipating its rapid distribution towards the bloodstream. A pharmacokinetic study to evaluate carprofen nanoparticles (NP) after intraarticular administration (IA) in rabbits was performed through a modeling allometric approach. The pharmacokinetic analysis of plasma profiles showed a rapid CP distribution out-wards the synovial chamber but mainly remaining in plasma (Vc = 0.14 L/5 Kg), according to its high protein-binding. The absorption data modeling showed the occurrence of two different re-lease–absorption rate processes after nanoparticle administration in the synovial space, i.e., a fast rate process causing a burst effect and involving the 59.5% of the total CP absorbed amount and a slow rate process, involving 40.5%. Interestingly, the CP burst effect inside the joint space enhances its diffusion towards cartilage resulting in CP accumulation in about three times higher concentrations than in plasma. In line with these results, the normalized-by-dose area under the concentration vs. time curve (AUC) values after IA were 80% lower than those observed after the intravenous. Moreover, the slower slope of the concentration–time terminal phase after IA administration vs. intravenous (IV) suggested a flip-flop phenomenon (0.35 h-1 vs. 0.19 h-1). Notably, CP clearances are predictive of the pharmacokinetic (PK) profile of CP in healthy humans (0.14 L/h/5 kg vs. 2.92 L/h/70 kg) although an over-estimation has been detected for cats or dogs (10 times and 4 times, respectively). This fact could probably be attributed to inter-species metabolic differences. In sum-mary, despite the limited number of animals used, this study shows that the rabbit model could be suitable for a predictive evaluation of the release enhancement of CP-NP towards the biophase in arthritic diseases not due to sterical retention of the formulation.


Introduction
Osteoarthritis is managed with COX or fosfolipase A2 inhibitors to decrease prostaglandin mediators. Carprofen (CP) is an anti-inflammatory, analgesic and antipyretic propionic acid-derivative, used in veterinary medicine [1] as an alternative to corticosteroidic management of local inflammations [2]. Intraarticular administration (IA), improves local action and reduces systemic effects. Drug delivery to synovial fluid lining with the biophase, i.e., articular cartilage surface [3] improves the quality of life of animals with inflammatory arthropaties [4]. Drug uptake to the cartilage requires high synovial fluid concentrations, anticipating its rapid distribution towards bloodstream [5] due to the thin layer of specialized cells of the synovial cavity that facilitates the drug diffusion through. Indeed, a rapid equilibrium is achieved between synovial fluid and plasma [5]. Nanoparticle formulations, using biocompatible [6] Poly lactic-co-glycolic acid (PLGA) polymers, are promising [7][8][9][10] to extend the drug residence times, bioavailability and duration of effects [4]. The aim of the current study is the in vivo evaluation of CP nanoparticles for IA administration through a modelling approach, in rabbits.

Animals and Drug Administration
Study protocol was approved (Ref.2015_089) by the animal welfare committee (Department of Agriculture, Livestock and Fisheries, 1997). Three 7-month-old white New-Zealand healthy male rabbits (Harlan, Barcelona, Spain) weighing 3.76-3.94 kg housed under standard conditions receiving food and water ad libitum were used. After anesthesia with xylazine (Rompun ® 20 mg/mL, BayerHispania, Spain) and ketamine (Imalgene ® 100 mg/mL, Merial, Spain), a dilution of CP Rimadyl ® 50 mg/mL (Zoetis, Spain) in physiologic saline (1:1 v/v) was administered through the right ear vein at 4 mg/kg. Secondly, CP nanoparticles 4.5 mg/mL were administered intraarticularly at 0.5 mg/kg through the right knee joint according to a cross-over design with 7 days wash-out. The left ear vein was catheterized (22G × 1.00in., Henry Schein, Hong Kong, China) and Vacutainer ® tube K2E was used for sampling (Beckton&Dickinson, Madrid, Spain).

Analysis of CP in Plasma and Knee Samples
CP Solid-phase extraction in plasma samples was performed with Discovery ® DSC-18 cartridges and Visiprep DL ® vacuum manifold (Supelco) followed by HPLC-UV as described by Parton et al. [12]. Knee samples were analyzed as described by [13] and expressed in µg/g. The HPLC method was acceptably linear within the calibration range (0.51-103.50 µg/mL). The lower limit of quantification (LLOQ) was 0.51 µg/mL.

Data Analysis. Non-Compartmental/Compartmental Pharmacokinetic/Deconvolution Analysis
Individual pharmacokinetic parameters [14] were estimated using non-compartmental analysis. The in vivo CP input rate from the nanoformulation I(t) was calculated by numerical deconvolution [15] with Phoenix-WinNonlin ® 64.8.2 Certara Inc. All CP plasma concentrations (IV and IA) were simultaneously analyzed with a population approach using NONMEM ver 7.4 [16] and Xpose R package v4.2.1, as diagnostic tool. The first-order-conditional estimation method (FOCEI) with interaction was used for parameter estimation [17]. Inter-individual variability (IIV) exponentially modeled was evaluated for each pharmacokinetic parameter. Additive and combined error models were tested for residual variability. Model selection was based on: decrease in the minimum objective function (MOFV; −2xlog likelihood); parameter precision and visual inspection of goodness-of-fit plots (Gofs). A decrease in MOFV of 7.879 between nested models was statistically significant (p < 0.005). For non-hierarchical models, the model with lowest Akaike criterion (AIC) was selected [18]. One-and two-open-compartment models with linear elimination parameterized as distributional clearance (CLD), apparent distribution volumes (V), and elimination clearance (CL) were fitted to the data. Allometric weight scaling was a priori added to all disposition parameters standardized to 70 kg body weight [19][20][21]. The power parameter was 0.75 for clearances and 1.0 for distribution volumes [22]. Table 1 and Figure 1 summarize tentatives of description of CP release/absorption from nanoparticles. The descriptive/predictive capability of the final model was evaluated through Gofs and a visual predictive check (VPC) [23]. Table 1. Tentative models for description of CP release/absorption from nanoparticles. Models are depicted in Figure 1A (models 1-2) and Figure 1B (models 3-6).

Model
Kinetics of the Release/Absorption Process Sequential vs. Parallel Number of Depots 1 1st order absorption (ka) -one 2 Zero order absorption (ka) -one 3 Two 1st order absorption processes (kfr, ksr) Parallel two 4 1st order (kfr) and zero order absorption processes (ksr) Parallel two 5 1st order (kfr) and zero order absorption processes (ksr) Sequential two 6 Two 1st order absorption processes (kfr, ksr) Sequential two

Carprofen Concentrations
Mean ± SD plasma concentrations are displayed in Figure 2. Table 2 summarizes the comparative CP concentrations in knee tissues at 9 h post-IA-administration.   Table 3 summarizes the main non-compartmental disposition/absorption parameters. Caution should be taken because the high extrapolation percentage after IA administration (>20%) suggests that release/absorption process and bioavailability (51.4%) could not be accurately characterized. A total of 47 CP plasma concentrations were analyzed by the population approach. Concentrations below the LLOQ (23.4%) were replaced by LOQ/2. CP disposition was best described by a two-compartment model. IIV was only associated with CL. The CP release/absorption from nanoparticles consisted of two first-order processes (KfR: fast and KsR: slow). Thus, the IA dose was partitioned into fast (FfR·F1·DoseIA) and slow (FsR·F1·DoseIA) absorption compartments, where FfR refers to the fraction of the administered dose for the fast absorption and FsR to the slow absorption (FsR = 1 − FfR). F1 and DoseIA the absolute bioavailabilty and the input from the IA CP dose at time of admin-istration, respectively. The VPC (Figure 3) confirmed the descriptive/predictive model capability. The circles represent the observed data. Dashed lines depict the 2.5th and 97.5th percentiles of the simulated concentrations. The solid line corresponds to the 50th percentiles of the simulated concentrations. VPC showed that most of the data fell within the 90% prediction interval and were symmetrically distributed around the median both after iv and IA administrations. The final population pharmacokinetic parameters are shown in Table 4. The visual predictive checks (VPCs) were constructed from the fixed and random estimates obtained from the final selected model. One thousand concentration-time profiles were simulated using Monte Carlo simulations after each administration route and their non -parametric 95% confidence intervals (the 2.5th and 97.5th percentiles) were calculated and represented together with the observed data for visual inspection. CL = plasma clearance; VC and VP = volumes of distribution for central and peripheral compartments; CLD = intercompartmental clearance between central and peripheral compartments; IIV and residual variability given as coefficient of variation (%). KaFR: initial rapid release rate constant; KaSR: slow release rate constant. F: Bioavailability; FFR: fraction of drug released during the initial faster phase; FSR: fraction of drug released during the slow phase; TlagSR: lag-time of the slow phase release. All final parameter estimates are shown with the relative standard error (RSE) indicated by italic numbers in parentheses, demonstrating an acceptable precision.

Pharmacokinetic Analysis
Individual in vivo cumulate release/absorption and input rate I(t) profiles from the assayed IA nanoformulation are shown in Figure 4. An initial first-order kinetics burst effect was followed by a more sustained release. . Individual in vivo cumulate release/absorption (up) and input rate I(t) (down) profiles of CP from the assayed IA nanoformulation. These profiles were in agreement with in vivo release/absorption pattern described by the final model. This profile resulted to be different than that observed in the in vivo evaluation assay designed for this nanoformulation, which resulted very useful to optimize the conditions of in vitro evaluation.

Discussion
Anatomical similarities of rabbits with other species (dog, cat) suggest the extrapolation of in vivo CP release/absorption to these species. As previously reported [12,24,25], CP disposition was best described by a two-compartment model. The steady-state distribution volume (Vss = 0.1126 L/kg mean live bodyweight) was in line with other results in rabbits [26], dogs 0.1192 L/kg [24] or cats (0.1506 L/kg) [27]. Considering a total body-water of 0.61 L/kg in live rabbits [28], the low Vss values suggest that carprofen is mainly confined to plasma according to its high protein-binding [29] as other NSAIDs [2].
Predicted CP clearance (1.99 L/h/70 kg) was in agreement with results in healthy volunteers [29] (100 mg IV: 2.916 L/h). However, somewhat higher CL was predicted for 7.1-15.8 kg dogs (0.0447 vs. 0.01487 L/h·kg) and for 1.9-6.0 kg cats (0.058 vs. 0.006 L/h/kg). Prediction failure of the allometric CP model is probably due to large interspecies differences in metabolic patterns [22]. The metabolic pathway is only known in dogs, cat and humans [30] but not in rabbits.
The CP bioavailability from the nanoparticles (51.4%) was lower than expected, since the IA route obviates first-pass effects. Analytical limitations preventing the complete characterization of the release/absorption profile (extrapolated areas >20%), or a higher iv variability than anticipated probably due to the low number of animals can be the cause. The flip-flop phenomenon of IA nanoparticles support a longer residence time than conventional formulations. The CP release/absorption profile was best described by an initial burst effect (fast release component) due to the rapid release/absorption of the unencapsulated fraction into central compartment with a rate constant of 7.38 h −1 (KfR), resulting in a high concentration gradient between synovial fluid and plasma at initial times. Afterwards, a sequential second process with a slower first order release rate constant (KsR = 0.0667 h −1 ) of microencapsulated CP occurred. The fast process contributed to the 60% of total CP reaching the systemic circulation while the slow release prevented CP from leaving the joint space rapidly. Poorly-optimized inclusion of the non-ionic surfactant beneath the PLGA matrix [31] could have contributed to such an undesirable initial burst effect, so improvement of the formulation would be required to extend the CP residence time in the synovial cavity. Additionally, the reduced intraarticular volume in rabbits has demanded the injection of a minimal volume (0.4 mL) that, although being highly concentrated to migrate towards the cartilage, does not allow the build-up of a relevant drug reservoir for a significant modified release. In any case, CP plasma levels resulted to be higher than the half maximal inhibitory concentration (IC50) for Cox2 [32] both after IV and IA administrations. The lower albumin concentrations in synovial fluid (60%) in rabbits compared to plasma resulted in higher total plasma CP concentrations than in synovial cavity once the distribution equilibrium reached, as also reported in sheep [25] and in horse [2].
The cumulative input profile predicted by deconvolution was useful to optimize previous in vitro evaluation tests. Indeed, inadequate in vitro release profiles [33] have been due to, probably, an inappropriate membrane type.

Conclusions
In vivo characterization of a new CP nanoformulation for IA administration has been performed in rabbits. The pharmacokinetic profile was scalable to other species. The CP burst effect inside the joint space enhances its diffusion towards cartilage and plasma. Although a limited number of animals, the rabbit model seems suitable for a predictive evaluation of the release enhancement of CP (not due to sterical retention of the formulation) towards the biophase of arthritic diseases.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.