The use of IVIVCs for regulatory applications is well documented in the literature and in relevant Regulatory Agency guidance (1, 2, 3). However, application of IVIVC is not confined only in the late-stage development and regulatory space. Formulation composition and process changes are commonplace during drug product development as a drug candidate progresses from early to late stage clinical trials. For formulation and biopharmaceutics scientists, understanding the clinical impact of these changes is critical to enabling the design and development of a successful pharmaceutical product. For modified release (MR) products, the composition of the formulation is specifically intended to provide a targeted input rate that optimizes the compound pharmacokinetics. Thus the ability to use dissolution data and translate them via an IVIVC to a clinical pharmacokinetic outcome can significantly simplify the formulation development process. For the MK-0941 case study presented here, the study represented the initial introduction of a MR formulation in the clinic. The study was designed with an IVIVC in mind to guide future formulation development.
At a dose of 4 mg, MK-0941 is a BCS Class III compound based on previously reported low–moderate cell line permeability (Table I) (16). However, bioavailability in preclinical species was moderate to high (48% in rats and 100% in dogs at doses 0.3–3 mg/kg), indicating compound is reasonably well-absorbed in vivo. In the absence of conclusive in vivo data indicating high permeability and with the lower bioavailability in the rats, the designation of BCS III is maintained for the purposes of this manuscript. The compound exhibited regional dependent absorption with relative bioavailability of 40% after intra-colonic administration (relative to oral administration). There is no specific mechanism that the regional dependent absorption is attributed to and it is believed to, at least partially, be a reflection of the compound moderate lipophilicity (logDpH7.4 = 1.64) and the lower absorptive area in the colon. The compound is a P-gp substrate (BAAB ratio of 16.1 in human MDR1 transfected LLCPK1; experiment conducted at 5-μM concentration) but demonstrated moderate–high bioavailability in preclinical species and linear PK across a dose range of 1–6 mg tested in the single ascending dose study; thus it is unclear if P-gp is affecting oral absorption. Given the observed region-dependent absorption MK-0941 represented an interesting model compound to attempt a comparison of traditional convolution-based IVIVC with PBPK-based IVIVC. To our knowledge, no publication is available that has attempted such evaluation based on a dedicated clinical study with two formulation technologies (matrix and multiparticulate MR systems).
The deconvolution–convolution-based approach is the most commonly used approach for development of IVIVCs for MR products. However, the deconvolution approach works better for compounds with fully linear pharmacokinetics. In cases where the total in vivo input appears to be dependent on the in vitro release rate (i.e., relative bioavailability of formulations differs), establishment of these linear models is not straightforward. Two publications have previously reported attempts to establish IVIVC for a compound with regiodependent absorption or differential bioavailability. Balan et al. (18) demonstrated that while a traditional convolution-based IVIVC was not successful for metformin, a compound known to exhibit regiodependent absorption, a successful IVIVC could be obtained by a modified approach that can be considered a convolution-based approach that allows for the model to take into account the differential bioavailability between formulations. More recently, Kakhi et al. (15) proposed a piecewise-linear variable absorption scale model, based on deconvolution–convolution approach, to obtain an IVIVC for a compound with differential bioavailability for one of the formulations. For MK-0941, we adopted a model similar to that reported by Kakhi et al. The model resulted in significant improvement in description of the in vivo absorption over time and a much improved correlation over a traditional time scale/shift model. The model generally adequately described the pharmacokinetic profile across the six formulations, although it should be acknowledged that depending on the model setup prediction errors did not fully meet the requirements for a regulatory submission (eg. while the multiparticulate only model passes all standard internal validation criteria, the matrix model fails the Cmax criteria due to underprediction of the slow formulation). However at the early stages of formulation development that this study for MK-0941 was undertaken, the observed IVIVC would still be considered extremely useful in guiding the formulation development efforts.
Absorption/PBPK modeling is increasingly being used to guide formulation development. However there are relatively few reports specifically discussing the application of such models to MR products. Lukakova et al. (14) described the use of absorption models developed in GastroPlus to accurately predict the observed pharmacokinetics of MR formulations of metoprolol, a BCS I compound and of Adinazolam. In the latter case, the model was based on fit of the in vivo release rates to the pharmacokinetic data since in vitro release data were not available. For metoprolol, the authors were able to obtain a reasonable IVIVC by adjusting the colonic ASF values, similar to what was attempted in this manuscript. The authors concluded that even after this adjustment absorption of the slowest formulations was not fully predicted, however, overall prediction error was acceptable (14–20%). Our simulations/IVIVC for MK-0941 indicates a very similar behavior; the slowest formulation of MK-0941 results in the highest prediction error (∼20% prediction error similar to what was reported for metoprolol). In a more recent manuscript, Brown et al. (19) discussed the application of GastroPlus models to guide formulation development of a HIV-1 attachment inhibitor phosphate ester prodrug. The authors also employed a regiodependent adjustment on the ASF factors to describe the absorption of the compound across the GI tract. While no IVIVC was attempted and assessment of prediction errors was not carried out, the simulation results from the authors provided same rank order with their projections.
In our study, we attempted to expand on these previous manuscripts by looking at the application of the ACAT model across six formulations developed by two different processes (three matrix and three multiparticulates). We further compared the predictions obtained from the absorption modeling efforts to the predictions that are obtained if a more traditional IVIVC is established. In general, we found that the two models resulted in comparable overall predictability and both models allowed for a good description of the overall formulation behavior. The traditional convolution-based IVIVC model required the implementation of the piecewise-linear variable absorption scale model to obtain best model description. The ACAT model required an adjustment of the ASF factors that could be seen as an analogous approach to allow for differential rate of absorption. The major difference between the models is that while for the traditional IVIVC model a correlation function between in vitro dissolution and in vivo absorption was implemented, the absorption model assumed that the in vitro dissolution is representative of in vivo release; the reasonable prediction of the pharmacokinetic data solely by adjustment of the ASF factor could be considered as an indication of a physiological reason for the differential behavior of the formulations rather than a disconnect between in vitro and in vivo release. We acknowledge that there are multiple other options to approach the absorption/PBPK IVIVC model. Ideally, a single set of ASF values would be set across both matrix and multiparticulate formulations and a secondary correlation between in vitro and in vivo dissolution would be set separately for each one; these relationships may deviate from the 1:1 relationship used in the current model. In the absence of detailed studies mapping out the regional absorption of a compound (such data are not routinely generated during clinical development) to separately fit the ASF values, any model structure attempted on the MR formulations will result in interdependent estimates for ASF and in vitro–in vivo dissolution correlations that cannot be separately identified as unique values. Thus, we chose to present here the simplest approach where only the ASF values are fitted with the in vitro dissolution used as input. Despite not representing a unique solution, we believe our analysis indicates that absorption/PBPK modeling IVIVC could be considered as an alternative to the traditional IVIVC approach to guide formulation development for MK-0941 at early stages
When trying to evaluate the relevance of an in vitro permeability assay and establish an IVIVC, the permeability determined in vitro should be compared with some in vivo end point that is directly related to or influenced by in vivo permeability. This end point can be of a pharmacokinetic or pharmacodynamic nature. Typically, a linear (or multiple linear) regression is performed and the strength of the correlation is statistically evaluated.
5.2.1 Orally administered drugs
For orally administered drugs, the parent compound and metabolite concentrations in blood, urine, and feces can easily be monitored. Hence, there is a possibility of obtaining various measures more or less related to permeability in the gastrointestinal (GI) tract. The fraction of dose absorbed (FA), defined as the relative amount of the drug that passes from the lumen of the GI tract into the tissue of the GI tract determined by absolute bioavailability or mass balance studies after oral administration, is a possible in vivo parameter for IVIVC with in vitro permeability methods for oral characterization (Zhao et al., 2001). When correlating Caco-2 in vitro permeability (Papp) with FA, a sigmoid relationship was observed in various reports despite the weak correlation observed in several of them (Grass & Sinko, 2002; Yazdanian, Glynn, Wright, & Hawi, 1998). Similar results were obtained when comparing in vitro permeability in the Madin–Darby canine kidney (MDCK) cell with FA (Irvine et al., 1999). This can be related to the existence of other mechanisms of absorption (active transport or efflux), solubility issues, or badly predicted FA values, to name a few. In fact, if only passively absorbed drugs are considered, a good correlation was observed between Caco-2 permeability and FA (Matsson et al., 2005).
Direct human jejunal permeability (Peff) can be determined in vivo using the in situ perfusion (Loc-I-Gut) method; good correlations have been observed with jejunal rat segments in vitro using the Ussing chamber. Good correlations were also observed between in vivo human jejunal permeability and FA in humans for both passive and carrier-mediated absorbed drugs (Lennernas, Nylander, & Ungell, 1997). As such, these direct determinations of human in vivo permeability provide a better comparison for IVIVC purposes. When subjected to this comparison, a strong correlation was observed between the in vitro Caco-2 permeability of 15 passively absorbed drugs and the corresponding in vivo human permeability (Sun et al., 2002). Using a multiple linear regression, Caco-2 permeability was also correlated with human jejunal permeability in an extended number of drugs by including an additional molecular descriptor (number of rotatable bonds in the molecule) that could compensate the larger density of tight junctions presented in Caco-2 cells compared with the human intestine (Figure 5.1; Paixao, Gouveia, & Morais, 2012). Overall, it is accepted that Caco-2 cells are able to identify highly absorbed drugs; this has been frequently used, at least as supportive, under the Biopharmaceutics Classification System concept for biopharmaceutical purposes (Lindenberg, Kopp, & Dressman, 2004).
5.2.2 Pulmonary administration
Although studies have sought to characterize the systemic profiles of drug in serum/plasma after pulmonary administration, it is more difficult to establish IVIVC for inhaled compounds compared with those delivered orally because fewer medicines are marketed for inhalation (Manford et al., 2005). Within this limited data setting, cultured human airway epithelial (Calu-3) cells and the human bronchial epithelial (16HBE14o-) cell line model have demonstrated excellent IVIVC, despite their differences in transendothelial electrical resistance. In reality, diffusive absorption kinetics of intact lung for test molecules could be well correlated with the Papp values obtained from non-lung epithelial cell models such as the Caco-2 cell system (Tronde et al., 2003). Therefore, the use of lung cell models would become more crucial to assess the absorption kinetics of test molecules undergoing certain lung cell-specific mechanisms, such as active transport and metabolism (Sakagami, 2006).
5.2.3 Blood–brain barrier
For the blood–brain barrier (BBB), traditional in vivo approaches used to establish the brain penetration of central nervous system (CNS) drugs consisted of estimating the extent of the steady-state brain distribution in preclinical species, defined as the ratio of total drug concentrations in the brain versus in the blood or plasma. However, this type of data have several limitations; alternatively, correlations can be based on BBB permeability measured by in situ brain perfusion (Liu, Tu, Kelly, Chen, & Smith, 2004). Good correlations were reported for passively transported drugs between in vitro permeability measured in primary bovine brain microvessel endothelial cells (BBMEC) and permeability obtained from rat in situ brain perfusion, although in vitro permeability values were significantly higher than their in vivo equivalent (Pardridge, Triguero, Yang, & Cancilla, 1990). On the contrary, no overall correlation between Papp measured in MDR1-MDCKII and permeability determined using total concentrations measured by rat in situ brain perfusion was found for a dataset of 50 compounds. This may be related to the need to perform a correction of in vivo total concentrations by the fractions unbound (fu) in blood and brain (Summerfield et al., 2007). In another study, IVIVC were improved by correcting for binding before performing the correlation analysis (Hellinger et al., 2012). In this case, good correlations were found in a dataset of 10 passive diffusion and P-glycoprotein (P-gp) substrate drugs between mouse in vivo permeability, corrected for the fu,brain/fu,plasma ratio, and various in vitro BBB models, such as rat primary brain capillary endothelial cells co-cultured with astrocytes and pericytes, as well as non-BBB models, such as MDR1-MDCK, MDCK, Caco-2, and highly P-gp–expressing vinblastine-treated VB-Caco-2.
Another important factor for BBB permeability is to assess the prediction of active efflux at the BBB for P-gp substrates using in vitro systems. This correlation analysis can be carried out, for example, between in vitro efflux ratios (defined as the ratio between permeability measured in opposite directions across the monolayer) and AUCbrain to AUCplasma ratios in P-gp knockout mice versus wild-type mice. To this aim, MDCK, the in vitro efflux ratio of human MDR1-MDCK and mouse Mdr1a-MDCK cell lines seems to be effective in identifying P-gp from non–P-gp substrates (Feng et al., 2008). A good IVIVC for the in vitro efflux ratio across MDR1-MDCK cells for P-gp substrates was also found, but only after this efflux ratio was corrected by the fu,brain/fu,plasma ratio in a manner analogous to that described for permeability (Summerfield et al., 2006). Because accounting for binding has been shown to improve correlations, a better alternative may be to carry out correlations with permeability measured based on unbound brain extracellular fluid (ECF) concentrations obtained using microdialysis as in vivo reference data. With this type of data, a linear correlation was found between log Papp from BBMEC, Caco-2, and MDR1-MDCK cell lines and the in vivo microdialysis AUCbrain/AUCblood ratio for nine drugs (Hakkarainen et al., 2010). However, for all three cell systems, the correlation was only statistically significant when two outliers, which were both known efflux transporter substrates, were removed from the regression dataset. These two compounds had much higher in vitro permeability compared with their in vivo values, and the authors justified their removal from the correlation analysis by suggesting that the mechanism responsible for their in vivo efflux may have been saturated at the concentrations used in vitro. This highlights the importance of choosing physiologically relevant concentrations in vitro for transporter substrates, to be able to obtain meaningful correlations with in vivo data (Ball, Bouzom, Scherrmann, Walther, & Decleves, 2013).
5.2.4 Topical application
In contrast to oral administration, in which absorption aims for a systemic effect, topical application to the skin is usually used for local treatment. Therefore, the main interest lies in determining the drug level within the skin, to evaluate the dermal bioavailability of compounds or assess the bioequivalence between different formulations. To this aim, in vivo reference data for IVIVC can be obtained by using pharmacodynamic responses (e.g., the concentration-dependent vasoconstriction effect of corticosteroids), dermatopharmacokinetic approaches such as recovery studies (judged from the amount of substance missing within the recovered formulation after finishing incubation), skin segmentation studies (using the tape stripping method), and in vivo dermal microdialysis (Schaefer, Hansen, Schneider, Contreras, & Lehr, 2008).
Because of the greater variability and possible saturation of the pharmacodynamic response, in a study to assess the relative bioavailability of five clobetasol marketed products, Lehman and Franz (2014) found that the in vitro permeation test, using cryopreserved human skin, resulted in a total clobetasol absorption varying 10-fold from highest to lowest product, whereas the vasoconstrictor assay found that this same difference was less than twofold. Other examples with mometasone furoate (Narkar, 2010) and betamethasone valerate (Franz et al., 1999) also suggested that the in vitro permeation test might be more sensitive than the vasoconstrictor assay. Despite several technical difficulties, microdialysis is an in vivo method with large variability that results in coefficients of variation between 50% and 100% (Narkar, 2010). As such, IVIVC based on this type of data are rare. The tape stripping method seems to be a good predictor for the in vivo skin permeability. The rationale behind this method is that the topically applied drug must pass through the first barrier, the stratum corneum, to be available to the underlying tissues, and in some instances be absorbed into the systemic circulation. To support this, good correlations were shown between the concentration of drugs in the stratum corneum at 30 min and the total amount absorbed in the systemic circulation over 4 days for compounds with widely different physicochemical properties (Rougier, Lotte, & Maibach, 1987). Using this in vivo technique, several studies have shown good correlations between the uptake of nanoparticles (Raber et al., 2014) and the penetration of sucrose stearate emulsions (Klang et al., 2012) in an in vitro pig ear skin model compared with in vivo in the human forearm.