Physiologically based pharmacokinetic (PBPK) models are mathematical representations of pharmacokinetic processes. In a bottom-up approach all determining physico-chemical and biological interactions between a drug and the body are implemented into virtual compartments, mapping the anatomical architecture and the physiological and biochemical properties of the body. Integration over time then allows for calculation of concentration-time curves of a modeled drug in specific body-compartments in silico (Kuepfer et al., 2016). While PBPK modeling has become an established methodical approach in drug development and latterly in regulatory decision making (Zhao et al., 2012), there is now an emerging use of PBPK modeling to simulate pathological conditions (Radke et al., 2016). Still, the predictive performance of PBPK models in specific clinical settings with heterogeneous and chronically ill patients characterized by numerous unknown individual and clinical factors needs to be further tested.

Cyclosporine is a drug with a high inter- and intraindividual pharmacokinetic variability and a narrow therapeutic window. As a cyclic oligopeptide (Rüegger et al., 1976) cyclosporine shows a poor water-solubility but high intestinal permeability (Amidon et al., 1995). It is predominantly a substrate of cytochrome P450 (CYP) 3A4 (Kronbach et al., 1988) and P-glycoprotein (P-GP) (Saeki et al., 1993). Due to a high protein binding predominantly to lipoproteins (Lemaire and Tillement, 1982), the fraction unbound in blood (f_U) is low (1–17%) and depends on measurement methods and examined individuals (Akhlaghi and Trull, 2002). Cyclosporine shows saturable accumulation in blood cells and peripheral tissue resulting in a partially non-linear pharmacokinetic behavior (Foxwell et al., 1988; Tanaka et al., 2000). Moreover it shows a high affinity for fatty and lymphatic tissue (Kahan et al., 1983). Cyclosporine is eliminated almost exclusively in form of its metabolites via bile (Venkataramanan et al., 1985).

Since the early 1980’s cyclosporine has been used for immunosuppression after renal transplantation and can impressively reduce acute rejection episodes early after transplantation. Yet long-term allograft survival is still inadequate with a progressively decreasing allograft function occurring in most patients within 10 years after transplantation (Nankivell et al., 2003). Causes of so called chronic-allograft-injury include, among others, cyclosporine nephrotoxicity and immunological rejection of the transplant which should both be prevented by an optimal immunosuppressive regime. Although dosing and target levels are poorly researched (KDIGO Clinical Practice Guideline for the Kidney Disease Improving Global Outcomes, 2009) and must be adjusted in consideration of individual patient factors and (immunosuppressive) co-medication, a high variability of cyclosporine exposure might correlate with a worse long-term outcome (Kahan et al., 2000). In clinical practice cyclosporine blood levels are measured regularly and invasively via therapeutic drug monitoring (TDM) to avoid side effects, particularly nephrotoxicity, as well as rejection episodes and to adapt dosing. However, details on procedure and the actual benefit of TDM for efficacy and safety of cyclosporine therapy have been studied insufficiently (KDIGO Clinical Practice Guideline for the Kidney Disease Improving Global Outcomes, 2009).

PBPK modeling can serve to further investigate the complex pharmacokinetic of cyclosporine. In the 1990’s there have been several approaches to investigate the non-linear pharmacokinetic behavior of cyclosporine after intravenous application by PBPK modeling (Kawai et al., 1998; Tanaka et al., 1999). One model has been applied to analyze the intravenous pharmacokinetic of cyclosporine in children (Gérard et al., 2010). Attempts to describe the oral absorption of cyclosporine by whole-body PBPK modeling either publish incomplete data (Darwich et al., 2013) or rely on semi-mechanistic approaches only (Gertz et al., 2013). Drug levels from TDM allow to retrospectively evaluate the predictive ability of a PBPK model in clinical care (Polasek et al., 2018).

The aim of this study was to implement a PBPK model for the oral application of cyclosporine and to assess its predictive performance in a clinical setting by comparing model based predictions of trough levels with observed trough levels from TDM of renal transplant outpatients.

The aim of this study was to test the predictive performance of a PBPK model in real-life clinical care by comparing predicted trough levels of a PBPK model developed for orally administered cyclosporine with observed trough levels measured by TDM in renal transplant outpatients. Up to now, PBPK models are hardly applied in clinical routine. Our work has shown, that taking into account patient's biometric parameter a PBPK model shows a satisfying accuracy with a small systematic bias and a reasonable precision for a heterogenous and chronically ill group of patients.

While in most patients predicted trough levels were in good accordance with observed drug levels, a relevant deviation (more than threefold) was found in some patients at least for a few cyclosporine trough levels. Prediction errors were more pronounced in patients receiving higher cyclosporine doses showing an overprediction in most of these cases. In the literature, factors influencing the absorption process are discussed as being substantial for the pharmacokinetic variability of cyclosporine. These factors include diet, intestinal motility and notably factors with an impact on the intestinal cytochrome P450 system.

Comedications inducing or inhibiting CYP3A4 have a strong impact on cyclosporine’s pharmacokinetics by lowering plasma levels up to one third or doubling of oral bioavailability, respectively (Hebert et al., 1992; Gomez et al., 1995). In our study, documented comedication data were limited to a few concomitantly taken immunosuppressive compounds (azathioprine, mycophenolate mofetil, and corticosteroids) which might be considered as a potential source for bias. Whereas conflicting results have been reported for corticosteroids, no clinically relevant influence on cyclosporine pharmacokinetics have been reported for azathioprine and mycofenolate mofetil (Kuypers, 2008; Lam et al., 2008). Hence, a limited impact of the immunosuppressive agents taken by the patients on the cyclosporine pharmacokinetics cannot be fully excluded. However, we did not found relevant discrepancies regarding the predictive accuracy of the developed cyclosporine PBPK model after stratification by corticosteroid intake (data not shown). Due to the limited documentation of comedications, we cannot fully exclude an intake of other comedications influencing the pharmacokinetics of cyclosporine to a relevant extent. Inclusion of those interacting (non-) immunosuppressive comedications may further improve the predictive accuracy of cyclosporine PBPK models.

It is known that expression of CYP3A4 can change in chronic kidney disease (Rowland Yeo et al., 2011; Zhao et al., 2012) which mostly develops in the clinical course after kidney transplantation and was present in our clinical population too. We stratified the model’s prediction by patient’s GFR (data not shown) and found no correlation between GFR and accuracy of model prediction. This is in accordance with the available evidence. Correspondingly no dose adjustments are recommended for patients with impaired kidney function (Novartis Pharma, 2011) implicating that changes of CYP3A4 expression in chronic kidney disease are of minor pharmacokinetic relevance for cyclosporine. Still, modeling of cyclosporine metabolism in chronic kidney disease seems to be an interesting spot for further research. The representation of intestinal cyclosporine metabolism within the model is based on literature based local CYP3A4 expression and on an intrinsic clearance fitted to match a literature based value for F_I of 0.47. Other attempts to derive the intestinal extraction rate for cyclosporine from in vitro-data by PBPK modeling resulted in a 5-fold underestimation of the supposed value for F_I (Gertz et al., 2011). Considerable variations in local expression influencing clearance in vivo are well known for CYP3A4 (Paine et al., 1997). Moreover, for cyclosporine its binding not only to plasmatic but also to cytosolic lipoproteins could be of particular relevance for intestinal and systemic metabolism. Lipoproteins are involved in many highly regulated physiological and pathophysiological processes which might also concern lipoprotein-bound cyclosporine (Gupta and Benet, 1990). Distribution of cyclosporine into lipoproteins shows an extensive inter- and intraindividual variation (Sgoutas et al., 1986) and lipoproteins vary greatly according to the current metabolic state in general, during cyclosporine therapy and with concurrent chronic kidney disease in particular. Depending on the method for measurement, pre-analytics and examined individuals values for f_u vary therefore from 1 to 17% (Akhlaghi and Trull, 2002) while the fraction unbound was a very sensitive parameter in the current modeling with a distinct influence on simulated concentration-time curves. Concerning this aspect the developed PBPK model might not cover all physiological and pathophysiological conditions relevant for a mechanistic representation and mapping of cyclosporine pharmacokinetic variability but might offer prospects for further research.

Intestinal P-GP is discussed as another potential cause for the high variability of cyclosporine absorption (Fricker et al., 1996; Lown et al., 1997). The quantification of the intestinal P-GP efflux transport kinetics in the current model is based on the efflux transport of cyclosporine at the blood-brain-barrier of rats in vivo (Tanaka et al., 1999) which is attributed to P-GP (Goralski et al., 2006). Using numerical values for the P-GP efflux transport kinetics from in vitro studies (Saeki et al., 1993; Fricker et al., 1996) resulted in a fraction absorbed of ∼0.01, i.e. almost no absorption. Difficulties in translating P-GP in vitro data to in vivo observations are well known and might be at least partially caused by the complex interplay of intestinal solubility, intestinal permeability and intestinal metabolism. Modeling and simulation in the current work confirmed the assumption that a P-GP-efflux-transport in the small intestinum might be of minor importance for the pharmacokinetic profile of cyclosporine since a large amount of drug must be absorped to enable the extensive intestinal metabolism known for cyclosporine. P-GP might be of particular relevance to prevent absorption from the large intestine thereby forming an absorption window for cyclosporine in jejunum and ileum (Fricker et al., 1996). Moreover, a relevance of P-GP-efflux for the systemic distribution of cyclosporine was discussed before (Schinkel et al., 1995), but not considered in the current model.

The intravenous pharmacokinetics of cyclosporine within the model are marked by nine specific binding processes in blood cells and in parenchymatous organs leading to the high volume of distribution and a long half-life. While the accumulation of cyclosporine in blood cells due to specific and high-affinity-binding to cyclophilins is well studied, binding of cyclosporine in peripheral tissue and its quantification was hypothesized based on tissue concentrations in rates and estimations made by PBPK-modeling (Tanaka et al., 1999; Tanaka et al., 2000). An actual physiological correlate is unknown or at least presumably nonspecific (Ryffel, 1993). The validity of an interspecies translation between rats and humans was not studied.

Rather than specific binding processes, biotransformation and specific transports, physico-chemical parameters, that is solubility and intestinal permeability, appeared to be very sensitive parameters having a strong impact on predicted concentration-time curves and trough levels in the developed PBPK model and the conducted simulations by influencing absorption. Depending on the solvent, values for cyclosporine solubility range from 7.3 μg/ml in water (Ismailos et al., 1991) up to 250 μg/ml in postprandial human intestinal fluid (Persson et al., 2005). A solubility of 190 μg/ml as used in the current work seems to be a reasonable parameter value resulting in a solubility limit and a decline of the fraction absorbed at doses above 300 mg per os. This is in accordance with observed data and the reduction of dose has been discussed as a possible reason for the increase of oral bioavailability that can be observed in the course of cyclosporine immunosuppressive therapy early after transplantation (Ptachcinski et al., 1986). With respect to the fact that the absorption of cyclosporine seems to be food-dependent (Gupta and Benet, 1990), enhanced mechanisms of digestion prior to absorption might be involved in the absorption process. The prediction of drug levels while considering food intake and adapting solubility in the PBPK model might help to map very low and very high drug levels.

To the best of our knowledge, two other whole-body PBPK models depicting the absorption process of cyclosporine can be found in the literature. The first model (Darwich et al., 2013) relies on the Simcyp® compound file for cyclosporine, available in the Simcyp® Simulator compound library. Only few model parameter values are reported within the publication itself. F_I is supposed to be much higher and F_A to be considerably smaller than in the current work but no references for these assumptions are published. In the second model (Gertz et al., 2013) a drug liberation process is implied and the intestinal permeability is set higher than in our model. No P-GP kinetics are considered and the intestinal metabolism is represented semi-mechanistically only.

Considering solubility, intestinal metabolism and P-GP-efflux transport mechanistically our model can be used to examine the influence of clinical aspects on cyclosporine dosing (such as non-immunosuppressive co-medication, nutrition, dyslipidemia and chronic kidney disease). This might help to further understand inter- and intraindividual variability and to improve clinical outcome. While drug levels from TDM are a valuable starting point for testing the predictive performance of a PBPK model in a clinical context, further research should evaluate the PBPK model prospectively using full pharmacokinetic profiles.

In summary, the current study has shown that PBPK modeling offers valuable contributions for pharmacokinetic research in clinical populations.