Vinblastine pharmacokinetics in mouse, dog, and human in the context of a physiologically based model incorporating tissue‐specific drug binding, transport, and metabolism

Abstract Vinblastine (VBL) is a vinca alkaloid‐class cytotoxic chemotherapeutic that causes microtubule disruption and is typically used to treat hematologic malignancies. VBL is characterized by a narrow therapeutic index, with key dose‐limiting toxicities being myelosuppression and neurotoxicity. Pharmacokinetics (PK) of VBL is primarily driven by ABCB1‐mediated efflux and CYP3A4 metabolism, creating potential for drug–drug interaction. To characterize sources of variability in VBL PK, we developed a physiologically based pharmacokinetic (PBPK) model in Mdr1a/b(−/−) knockout and wild‐type mice by incorporating key drivers of PK, including ABCB1 efflux, CYP3A4 metabolism, and tissue‐specific tubulin binding, and scaled this model to accurately simulate VBL PK in humans and pet dogs. To investigate the capability of the model to capture interindividual variability in clinical data, virtual populations of humans and pet dogs were generated through Monte Carlo simulation of physiologic and biochemical parameters and compared to the clinical PK data. This model provides a foundation for predictive modeling of VBL PK. The base PBPK model can be further improved with supplemental experimental data identifying drug–drug interactions, ABCB1 polymorphisms and expression, and other sources of physiologic or metabolic variability.


| INTRODUC TI ON
Vinblastine (VBL) is a vinca alkaloid-class chemotherapeutic that functions as a microtubule poison. VBL binds to β-tubulin to destabilize tubulin polymers 1,2 and prevent further microtubule assembly resulting in mitotic arrest, the induction of apoptosis, and subsequent cell death. 3,4 VBL has been used successfully to treat both hematologic and solid tumors in both human 5 and veterinary medicine. 6 Despite its efficacy, VBL is frequently dose-reduced due to a variety of toxicities that manifest in hematologic, gastrointestinal, and neurologic pathologies with myelosuppression being most common. 5 Understanding the physiologic and biochemical components that drive VBL pharmacokinetics (PK) is thus essential to help predict drug exposure, associated toxicities, and reduce related morbidities.
Tissue distribution and intracellular retention of VBL are driven by tubulin binding and interactions with tubulin have been extensively studied. 2,7 The mechanism underlying VBL-tubulin binding is a ligand-induced plus ligand-mediated isodesmic selfassociation reaction, which ultimately results in spiral protofilament structures. 8,9 Vinca affinity to binding sites present at the terminal ends of microtubules underlies its antimitotic activity as it results in microtubule depolymerization and subsequent metaphase arrest. 10,11 Tubulin concentrations have been shown to vary substantially by tissue 12 and the specific β-tubulin isoform predominantly expressed in cancer can determine vinca alkaloid sensitivity or resistance. 13 The tissue distribution and elimination of VBL are also influenced by ATP-binding cassette transporter B1 (ABCB1). ABCB1 expression in normal tissues is primarily in epithelial cells with secretory/excretory functions and endothelial cells of capillary blood vessels serving a barrier function. 14 This includes the apical surface of endothelial cells of brain capillaries, intestinal and renal tubular epithelial cells, and the canalicular surface of hepatocytes illustrating the role of this protein in drug disposition. 15 ABCB1 presents a limitation to cellular drug uptake as it functions as an efflux transporter for many lipophilic substrates including VBL. 16 Previous studies established that wild-type (WT) mice dosed with VBL showed no brain accumulation, whereas ABCB1 knock-out (KO) mice (Mdr1a/b(−/−)) accumulated VBL in the brain and gut and showed slower elimination. 17,18 Metabolism of VBL is primarily attributed to the CYP3A family of enzymes, specifically CYP3A4 and CYP3A12 in humans 19 and canines, 20 respectively. The primary metabolite of VBL identified in humans and dogs is the active 4-deacetylvinblastine and has been reported to have a LD 50 lower than that of the parent. 21 As VBL is commonly administered with concomitant medications, drug-drug interactions that potentiate competition for metabolic enzymes through enzyme induction and inhibition are of particular concern when designing VBL therapeutic regimens.
Addressing these variables present a challenge to predicting clinical response but designing physiologically based pharmacokinetic (PBPK) models permits a quantitative approach to reduce the variable complexity and improve understanding of drug-specific PK. 22 PBPK models are based on the principle of mass balance and utilize a system of mathematical equations to represent key physiological processes: absorption, distribution, metabolism, and elimination. Kinetic terms characterizing enzymatic reactions that govern metabolism, protein-facilitated transport of drug, or intracellular binding are preferentially scaled from in vitro data. The concentration-time profiles generated through computational simulation provide a predictive methodology of supplementing traditional noncompartmental analysis by providing a foundation for in vitro-in vivo (IVIVE) extrapolation, allowing for interspecies scaling, elucidating implications of physiological variability, and ultimately improving drug risk assessment. 23 Previous human PBPK models have been published for vincristine with incorporation of tubulin as a key driver of drug disposition. 24,25 These previous models, however, utilized a relative tubulin expression value for all relevant tissues. To our knowledge, the presented model is the first PBPK model of VBL, and the first vinca alkaloid model to incorporate variability in organ tubulin expression and provide an interspecies scaled model from mouse to canine and human. The utility of this model is to illustrate the role of the biophysical drivers of vinca drug disposition with an emphasis on the key tissues at risk of VBL toxicity as well as provide a predictive cross-species model to enhance dosing strategies.

| Chemicals
Vinblastine sulfate salt and vinorelbine ditartrate salt hydrate were purchased from Sigma-Aldrich. Vinblastine (for injection) and vincristine were obtained from the pharmacy at the Colorado State University Veterinary Teaching Hospital. 4-desacetylvinblastine was purchased from Santa Cruz Biotechnology. Acetonitrile and methanol used for LC-MS/MS were of ULPC/MS grade and were purchased from Fisher Scientific. All other reagents were of analytical grade and were purchased from commercial suppliers. dosed IV with 2 mg/kg VBL prepared in 0.9% benzyl alcohol. Whole blood and tissues were harvested throughout a time range of 0.08 to 6 h. Whole blood was collected with heparinized needles and transferred to a microcentrifuge tube and was spun at 1200xg for 10 min at 4°C. The resulting serum and tissues were flash-frozen in liquid nitrogen and stored at −80°C until ready for analysis.
Microsome incubation and Michaelis-Menten kinetics experiments were modified from previously published methods. 26 Microsome incubation reactions for VBL were conducted at a time range from 0 to 45 min with a starting substrate concentration of 1 μg/ml for both mouse and canine microsomes. Incubation reactions were performed in technical singlet. Michaelis-Menten kinetics experiments were performed following a discontinuous methodology using the following parameters, determined from the microsome incubations: mouse VBL (0.5 mg/ml protein, 5 min); canine VBL (0.5 mg/ml, 20 min). These conditions were chosen as they represented linear loss of parent with respect to time, satisfying the steady-state assumptions for Michaelis-Menten kinetics analysis.
Kinetics experiments were performed in technical doublet. Both microsome incubation and kinetics reactions were stopped following the same method and analyzed by LC-MS/MS using the described method. Loss of parent was quantified absolutely and converted to velocity (ng/ml/min) and used as a surrogate for product formation. Michaelis-Menten kinetics parameters were estimated from velocity versus substrate data using GraphPad Prism v7.0d (GraphPad Software Inc).  The dwell times for each ion transition were 500 ms. Q1 and Q3

| LC/MS/MS instrumentation and conditions
were both operated in unit resolution. Column retention times were 2.83 min for vinblastine, 2.36 min for vincristine, and 1.80 min for 4-desacetylvinblastine. No interfering peaks were detected at the monitored ion transitions in extracted matrix.

| Data analysis
Quantitation of vinblastine and its metabolite was based on linear standard curves in spiked matrix (serum) using the ratio of vinblastine or metabolite peak area to vincristine peak area and 1/x 2 weighting of linear regression. Analyst v1.5.1 software (Applied Biosystems) was used for peak area integration. For serum samples, the lower limit of vinblastine detection was 1 ng/ml and the curve was linear from 1 to 750 ng/ml. For 4-desacetylvinblastine, the lower limit of quantitation was 1 ng/ml and the curve was linear from 1 to 100 ng/ml. All calculated concentrations from unknown samples fell within the linear range of the respective standard curves. Accuracy of standard curves and quality control samples (low, medium, and high concentrations) were within 15% for vinblastine and 10% for 4-desacetylvinblastine and >80% of quality control samples showed an accuracy >85%.

| PBPK model development
The PBPK model for VBL, represented by the schematic in Figure 1 is characterized by a set of nine distinct tissue compartments: plasma, brain, lung, bone marrow, slowly perfused, rapidly perfused, kidney, liver, and gut. The model's objective is to provide a unique and predictive, cross-species modality of describing the absorption, distribution, metabolism, and elimination of VBL. This model thus aims to elucidate the implications of drug pharmacokinetics and mechanisms of distribution, ultimately providing a modeling base for extrapolation of simulated concentration profiles to clinical settings.
The primary drivers of vinca disposition are fraction unbound drug to plasma proteins, intercellular tubulin binding, metabolism, and ABCB1-mediated transport. Each tissue is characterized by its corresponding volume and flow rate indicated in Table 1. With exception to the brain, tissue compartments are defined by a well-stirred, flow-limited model in which the concentration of drug leaving the compartment is equal to the unbound drug concentration within the tissue. The blood-brain barrier (BBB) is distinctive as the tight junctions and prolific expression of efflux transporter proteins provide a highly restrictive barrier to many lipophilic molecules. The brain is thus defined by a permeability-limited model dictated by a modification of Fick's Law of diffusion in which the permeability-surface area product constitutes the tissue permeation coefficient, the capillary surface area permitting diffusion, and the thickness of the cellular membrane. 32 ABCB1 is highly expressed at the blood-brain barrier as well as in the liver, gut, and kidneys, facilitating drug excretion. 14,15

| Model parameterization & equations
Physiological parameters for tissue-specific percent body weight and cardiac output for mouse, canine, and human were sourced from Brown et al. 27 and presented in Table 1. All tissue densities were approximated to be 1 kg/L. Cardiac output for respected species was estimated using a previously derived allometric scaling relationship for unanesthetized mice. 33 The percentage of drug bound to plasma proteins, primarily alpha-1 acid glycoprotein, is 48-99% with as much as 75% bound to serum proteins in dog plasma. 5,30 The rate of change in the amount of drug within generic, non-eliminating compartments is governed by the following mass balance equation: where A C is the amount of drug in the tissue, V C is the volume of the tissue (assuming static volume), C C is the concentration within the tissue, Q C is the tissue perfusion rate, C A is the arterial blood drug concentration, and C VT is the venous blood drug concentration.
A key mechanistic feature of this model is the tissue-specific intracellular binding of VBL to tubulin. The variability in tubulinbinding capacity of different tissues has been shown to have a deterministic role in vinca alkaloid tissue distribution. 12 To account for intracellular VBL retention in all tissues incorporated into the PBPK model, the following equation was used to determine the venous blood drug concentration: where C T is the concentration of drug in the compartment, P C is the partition coefficient, B C is intracellular-binding capacity of drug to tubulin, and K D is the drug-specific-binding affinity to tubulin. The use of a nonlinear mathematical equation to characterize drug binding to intracellular macromolecules has been reported in describing the disposition of docetaxel 34 and methotrexate 22 as a function of binding at low drug concentrations. Tissue-specific-binding capacities, B C , were obtained from studies where tubulin concentrations were determined from tubulin-binding capacities for colchicine 12 and are reported in Table 2. A strong correlation between tissue-to-plasma partition coefficients and binding capacity for the vinca alkaloid, vincristine, has been reported for mouse, rat, dog, and monkey. 12 Relative tissue tubulin-binding capacities were thus assumed to be equivalent for the mouse, dog, and human PBPK model developed herein. The intrinsic value for VBL-binding affinity for tubulin, K D , is reported in Table 2 as 196 nmol/L. The binding affinity report was measured using a ligandmediated model where the affinity for liganded heterodimers for spiral polymers was found to be the major determinant of overall vinca drug affinity to tubulin. 35 Partition coefficients, P C , for respective tissues were determined by performing parameter optimization in MATLAB's tool, SimBiology, using a nonlinear least squares regression estimation method from the optimization toolbox. The statistical modeling  Table 1.
The rate of change in the amount of drug in tissues with ABCB1 expression is as follows: In which ABCB1-mediated efflux of drug is characterized by Michaelis-Menten saturable kinetics, where F ABCB1,T is the scaling factor for the relative expression of ABCB1 transport protein in relevant tissues, V max,ABCB1 is the maximum velocity of ABCB1 transport out of the tissue compartment, and K m is the associated Michaelis-Menten constant.
Scaling factors, F ABCB1,T , were acquired from Systems Pharmacology for ABCB1 relative tissue expression as previously reported. 36 The transfer of drugs at the BBB is unique as the mass transfer out of the vascular space surrounding the brain tissue is hindered by a permeability barrier. The highly restrictive nature of the BBB to exogenous substances results in a diminished flux of drug into the tissue, warranting the use of a permeability-limited modeling approach. 32 Coupling the ABCB1 transport equations as expressed above with permeability-limited modeling equations from Choi et al., 23  where V CNS is the volume of the brain tissue, V BB is the volume consisting of brain blood, C CNS,T is the concentration of drug in the brain tissue, PSA is the permeability surface area product, sf ABCB1,CNS is the relative ABCB1 expression in brain tissue, and C BB is the concentration Schematic representation of a physiologically based pharmacokinetic (PBPK) model including key organs involved in vinca drug ADME following a bolus IV dose. Solid lines are representative of blood flows and dashed lines represent clearance from organs by ABCB1-mediated transport, metabolism (V max,met and K m,met ), biliary excretion (V max,bil and K m,bil ), or glomerular filtration (GFR).
of drug in the brain blood. V CNA and V BB were set to 97% and 3% of total brain volume, respectively. 37 The PSA value representing the flux of VBL at the BBB was determined by optimizing the simulated  Table 1, characterizes the baseline flux of VBL at the BBB in the absence of ABCB1-mediated transport and was used in subsequent dog and human models. V max and K m associated with ABCB1-driven transport were simultaneously fit to wild-type brain tissue PK data and determined to be 928.8 nM/h and 6.41 nM as reported in Table 3.
Extrapolation of ABCB1 kinetic parameters to the mouse and human models was conducted using a K m value set to 5.76 μM as reported in literature for human ABCB1 ATPase activity for VBL. 40 The corresponding V max,ABCB1 value was optimized to in vivo canine PK data. The analyzed PK data for VBL-treated canines included the following concurrent treatments groups: prednisone, omeprazole, diphenhydramine (n = 8), prednisolone (n = 4), and no concurrent medications (n = 1) as shown in Figure 3. Omeprazole has been shown to be an inducer of CYP3A4 in human hepatocytes 41 and an inhibitor of ABCB1 in Caco-2 cell lines. 42 Canine model parameter values were thus fit to the patients without concurrent treatment of omeprazole to minimize the influence of potential drug-drug interactions.
The primary routes of vinca drug elimination are through metabolism, biliary excretion, and glomerular filtration. Clearance from the gut, kidneys, and liver is also mediated by ABCB1 as it contributes to tubular secretion and biliary excretion. 15 Previous studies have shown a decrease in fecal excretion from 20%-25% in wild-type mice to 3%-9% in Mdr1a/b (−/−) mice at doses of 1 and 6 mg/kg. 18 To account for drug elimination, additional terms were added to Equation (3). Metabolism is represented by Equation (6), biliary clearance is represented in Equation (7), and glomerular filtration is modeled using Equation (8). Metabolism and biliary transport were modeled with saturation kinetics. Metabolic kinetic parameters for the mouse model were derived experimentally using mouse microsomes of 0.5 mg/ml and extrapolated using 46 mg protein/g liver for in vitro-in vivo scaling. 38 Metabolic kinetic parameters were estimated from fitting Web-plot digitized data from canine liver microsmal studies by 20 and extrapolated using 55 mg protein/g liver for IVIVE. 39 Scaling of metabolic V max was determined using the following equation as reported previously 26 : where V max,mic is the Michaelis-Menten rate constant determined experimentally using liver microsomes and P mic is the micoromal protein per gram of liver. Values for species-specific kinetic parameters are reported in Table 3. Biliary excretion in the mouse model was developed using baseline biliary excretion parameters, V max,B and K m,B , through the previously described optimization method to Mdr1a/b (−/−) fecal data as shown in Figure 1G. The presence of VBL in fecal data of Mdr1a/b (−/−) is assumed to be attributed to Mrp2/ABCC2 present on canalicular cells of hepatocytes in rodents 43 as VBL has been identified as a substrate for human MRP2/ABCC2. 44 Biliary excretion was assumed to be primarily driven by ABCB1-mediated transport in the wild-type mouse, canine, and human models. Biliary transport kinetic values were thus equivalent to those of ABCB1 as reported in Table 3 and scaled to the liver using the tissue-specific relative expression scaling factor indicated in Table 2.

| Computer simulation and software
The PBPK model and simulations were conducted using SimBiology The sampling of values was constructed using random sampling with rank correlation matrix.

| Pharmacokinetic analysis
Pharmacokinetic parameters were calculated using Phoenix WinNonlin, version 8.3, Certara. Area under the curve (AUC) was calculated using a linear-log trapezoidal method.

| Data analysis
The predictive capability of the model was evaluated by calculating the prediction error (PE), median absolute performance error (MAPE %), the median performance error (MPE %), and the root mean squared performance error (RMSPE %) as indicated in Equations 9-12. 45 The prediction was calculated as follows: The measure of the precision of the prediction, evaluated was evaluated by the MAPE % and calculated with Equation 10 where n is the total number of samples per tissue.
The bias of the prediction was calculated by MPE% as:   Figure 2 and well predict the observed PK data. The role of ABCB1 is most prominently shown in the brain PK and simulations in Figure 2B where the wild-type  Table 4. Ratios of observed versus simulated PK parameters for corresponding tissues were calculated to provide a direct comparison. The results showed that the predicted AUC ratios were within the accepted validating criteria of twofold with the largest discrepancy of 78% for the gut. 46 The ratios for terminal half-life for serum and tissues were predominantly within the range of twofold, apart from wild-type brain and gut with values of 0.45 and 2.07, respectively.
The predictive performance of the PBPK model was evaluated by calculating MAPE%, MPE%, and RMSPE % as reported in Table 7.
The accuracy of the prediction, measured by MAPE%, ranged be-

| VBL PK and model simulations in canines
After validating the accuracy of the above PBPK model, the wildtype mouse model was scaled to canine using the physiological The PBPK model simulations and PK data are presented in Figure 3. An Akima spline fit for each treatment group in Figure 3B highlights the discrepancy between the concentration versus time profiles between the prednisolone and the POD treatment groups.

F I G U R E 3
Observed canine plasma PK with indicated patient concomitant medication treatment groups: prednisolone only (n = 4), not reported (n = 1), and prednisone/omeprazole/ diphenhydramine (n = 8); Monte Carlo PBPK model simulations (n = 100) are represented by range, SD, and mean outputs (A). Observed canine plasma PK with an Akima spline fit to concomitant medication treatment groups (B). Monte Carlo simulation of canine plasma PK following an induction in metabolism due to omeprazole-induced increase in CYP3A4 mRNA expression by twofold as determined 41 (C).

F I G U R E 4 AUC by concomitant medications. Boxplot of AUC
values for canine PK respective of concomitant treatment groups shows median ± interquartile range and whiskers represent min/ max values; two-tailed un-paired t-test with **p < .01.
AUC and terminal half-lives were determined for mean simulated and PK data based on indicated concomitant treatment groups presented in Table 5. Ratios of observed versus simulated AUC's and terminal half-lives were determined. For all patients, independent of concomitant medications, and patients concurrently treated with prednisolone, the AUC ratios were 1.7 and 0.8, respectively. For the POD patient group, the AUC ratio was determined to be 3.2, falling above a twofold discrepancy. The terminal half-life ratios were between 0.55 and 1.97 for the three patient groupings. The predictive performance of the simulation is shown in Table 7 with MAPE% values of 28.5%, 24.8%, and 60.2% for all patients, prednisolonetreated, and POD-treated, respectively. MPE% values were positive for only the prednisolone treatment group and negative for all and POD-treated patient groups, indicating an over-prediction of the model for the latter two groups.
Comparison of AUCs for the POD patient group and prednisolone/non-reported group showed substantially lower values for POD as shown in Figure 4. Omeprazole has been shown in a previous study to be an inducer of CYP3A4 in human HepG2 cells. 41 Although there are currently no published data indicating the induction of the canine ortholog, CYP3A12, by omeprazole, the potential drug-drug interaction was simulated by increasing the rate of metabolism by twofold, corresponding to the reported induction of CYP3A4 expression. 41 An analysis of the predictive performance of the escalated metabolism simulation showed improvement in the MAPE% for the POD treatment group from 60.2% to 42.1%.

| VBL PK and model simulations in humans
A human PBPK model was developed using human physiological and clearance parameters as reported in Tables 1 and 3

. A Monte
Carlo simulation was performed as described for the canine model.

| DISCUSS ION
Physiologically based pharmacokinetic models were introduced by Teorell in 1937 50 and first applied to cross-species modeling of anticancer drugs for methotrexate. 22 Subsequently, PBPK models have been described for multiple anticancer agents including the cytotoxic drugs, doxorubicin, 51 docetaxel, 34 and cisplatin 52 as well as targeted agents including lapatinib, 53 crizotinib, 54   a MAPE% is the median absolute prediction error, which is a measure of the precision of the prediction. b MPE% is the median prediction error, which is a measure of the bias of the prediction. c RMSPE% is the root mean squared performance error, which is a measure of the accuracy of the prediction. d All canine patients enrolled in study independent of concomitant medications. e Canine patients concomitantly treated with prednisolone only.
f Canine patients concomitantly treated with prednisone, omeprazole, and diphenhydramine. g Canine model for simulated omeprazole-induced increase in CYP3A4 expression and metabolic activity for prednisone + omeprazole + diphenhydramine (POD) concomitant treatment group. describe disposition, metabolism, and elimination. The mechanistic framework utilizing key drivers that govern VBL disposition presents a useful tool for providing insight to clinically relevant questions with regard to patient-specific variability such as co-morbidities and the effect of drug combinations encountered in therapeutic treatment regimens. An improved understanding of how patient variability affects VBL PK can thus be used to provide predictions in clinical situations and accordingly modify concurrent medications or doses to provide improved treatment to cancer patient populations.

ACK N OWLED G M ENTS
The

CO N FLI C T O F I NTE R E S T
No author has an actual or perceived conflict of interest with the contents of this article.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data and modeling code that support the findings of this study are available from the corresponding author upon reasonable request.