The core role of a clinical pharmacologist is to improve the care of patients through safe and effective use of medicines and thus to recommend appropriate dosage regimens for individual patients or stratified subgroups of patients. Population pharmacokinetic (POP-PK) models, which aim to describe the covariates of variability in drug concentrations among individuals in the target patient population receiving clinically relevant doses of the drug of interest, are often developed for this purpose. Patient demographic and physiological variables, including body weight and metabolic functions, information such as concomitant medications and occasionally, but not often, genotypes of relevant enzymes and transporters, are included as covariates as it is known that these can affect dose–concentration relationships. Although POP-PK modeling (a ‘top-down’ approach) is
viewed to be an integral part of the drug develop-ment process, it is generally recognized that determining covariates is not always straight-forward and complications caused by bias and competition between multiple variables are a concern [1].
Modeling and simulation of the processes that define the plasma concentration–time course of a drug – namely, absorption, distribution, metabolism and elimination (ADME) – using a mechanistic approach can help to predict the potential exposure of a drug at a given dose in individual patients [2–4]. The key components of this approach are the ‘system’ (i.e., the human body) parameters, drug-related information (e.g., physicochemical characteristics deter-mining the permeability through membranes, partitioning to tissues, binding to plasma pro-teins or affinities toward certain enzymes and
Karen Rowland Yeo*1, Masoud Jamei1 and Amin Rostami-Hodjegan1,2
1Simcyp Limited, Blades Enterprise Centre, John Street, Sheffield S2 4SU, UK2School of Pharmacy and Pharmaceutical Sciences, Faculty of Medical and Human Sciences, University of Manchester, Stopford Building, Oxford Road, Manchester M13 9PT, UK*Author for correspondence: Tel.: +44 114 292 2332 Fax: +44 114 272 0275 [email protected]
The development of in vitro–in vivo extrapolation (IVIVE), a ‘bottom-up’ approach, to predict pharmacokinetic parameters and drug–drug interactions (DDIs) has accelerated mainly due to an increase in the understanding of the multiple mechanisms involved in these interactions and the availability of appropriate in vitro systems that act as surrogates for delineating various elements of the interactions relevant to absorption, distribution, metabolism and elimination. Recent advances in the knowledge of the population variables required for IVIVE (demographic, anatomical, genetic and physiological parameters) have also contributed to the appreciation of the sources of variability and wider use of this approach for different scenarios within the pharmaceutical industry. Initially, the authors present an overview of the integration of IVIVE into ‘static’ and ‘dynamic’ models for the quantitative prediction of DDIs. The main purpose of this review is to discuss the application of IVIVE in conjunction with physiologically based pharmacokinetic modeling under a systems biology approach to characterize the potential DDIs in individual patients, including those who cannot be investigated in formal clinical trials for ethical reasons. In addition, we address the issues related to the prediction of complex DDIs involving the inhibition of cytochrome P- and transporter-mediated activities through multiple drugs.
Predicting drug–drug interactions: application of physiologically based pharmacokinetic models under a systems biology approachExpert Rev. Clin. Pharmacol. 6(2), 143–157 (2013)
transporter proteins) and the study design (e.g., dose, route and frequency of administration, concomitant drugs and food). The development of in vitro–in vivo extrapolation (IVIVE), a ‘ bottom-up’ approach, to predict the pharmacokinetic param-eters has accelerated mainly due to the increasing availability of in vitro systems that act as surrogates for in vivo reactions relevant to ADME, and also as a result of advances in the understanding of the required population variables (demographic, anatomical, genetic and physiological parameters) [5]. Application of IVIVE in conjunction with physiologically based pharmacokinetic (PBPK) modeling under a systems biology approach can help inform the design of clinical studies in healthy volunteers or patients and to characterize the potential interactions in individual patients [3]. Over the past decade, there has been growing interest in the utility of PBPK modeling during drug development [6] and as a result of this, the number of submissions to regulatory bodies includ-ing PBPK modeling has been increasing [7,8]. Indeed, there were 25 submissions to the US FDA between June 2008 and December 2011 that included PBPK modeling. Of these, 60% were related to the prediction of drug–drug interactions (DDIs) [8].
DDIs remain an important issue in clinical practice and during the discovery and development of new drugs. The purpose of this review is to discuss the application of PBPK modeling to predict DDIs and to elucidate how a systems biology approach can be used to characterize potential interactions in individual patients, including those who for ethical reasons cannot be investigated in formal clinical trials and who may be on multiple drugs. The authors also present two case studies to demonstrate the above approach and introduce the application of top-down fitting used in conjunction with bottom-up IVIVE to provide optimal models for the prediction of DDIs a priori.
Drug–drug interactionsIn clinical practice, patients are often prescribed multiple drugs for the treatment of comorbidities. A DDI is said to occur when coadministration of drugs results in a pharmacological or clini-cal response that differs from the one observed when either drug is given alone. DDIs can result from changes in the pharma-codynamics or pharmacokinetics of a drug. Pharmacodynamic interactions occur when drugs compete for the same receptor or have similar (or opposing) downstream pharmacological actions; they can be synergistic, additive or antagonistic. Many of these interactions can be predicted from the knowledge of the pharmacology of each drug. Application of a mechanistic modeling approach to assess pharmacodynamic DDIs in vivo has been described [9,10]. There has been a growing recogni-tion of the fact that innovation can be increased dramatically by merging a systems biology approach with PKPD modeling. This has led to the emergence of systems pharmacology, which is viewed as the quantitative analysis of the dynamic interac-tions between drug(s) and a biological system to elucidate the behavior of the system as a whole, as opposed to the behavior of its individual constituents [10]. Mechanistic PKPD modeling is now advocated by both academic and industrial researchers and more recently by regulators [201,202]. Further discussion on
this topic is beyond the scope of the current review; the focus is on the application of a mechanistic approach to predict the pharmacokinetic interactions, more specifically metabolic and transporter-related DDIs.
Clinical relevance of DDIsPharmacokinetic interactions can alter the absorption, protein binding, metabolism and excretion of a drug. The amount of a drug being absorbed can be increased or decreased by modifying stomach pH, by affecting a transport system or by physically forming a complex. Decreased absorption of ketoconazole occurs following the administration of the proton-pump inhibitor ome-prazole. Suppression of gastric acid secretion by omeprazole leads to an elevated pH. However, a low gastric pH is required for an effective dissolution of ketoconazole in the stomach [11].
Some of the most clinically important DDIs are considered to be those related to alterations in metabolism by enzyme inducers or inhibitors in the gut or liver and are associated with a change in the plasma concentration of the affected drug, which may in turn lead to a severe adverse reaction. More recently, it has become increasingly evident that DDIs involving transporters are an issue during drug development and in clinical practice [12,13]. Although there are many DDI studies published in the literature, we discuss here a few key examples that illustrate the potential complexity of DDIs, especially when multiple inhibitory mechanisms and elimination organs are involved.
Many clinically relevant DDIs involving cytochrome P450 3A4 (CYP3A4), the most abundant P450 in human liver and intestine, are due to time-dependent inhibition consequent to metabolism of the inhibitor to a reactive intermediate that binds irreversibly or quasi-irreversibly to the enzyme [14,15]. These DDIs are persistent as they require synthesis of new enzyme for recovery. The with-drawal of the potent mechanism-based inhibitor mibefradil from the market a year after its introduction in 1997, as a consequence of reports of serious DDIs with CYP3A4 substrates, illustrates the importance of assessing the potential impact of mechanism-based inhibitors in drug development [16].
Although interactions in the kidney are considered to be less important, some DDIs for drugs that are mainly renally excreted are clinically relevant; indeed, a 50% increase in plasma exposure of the antidiabetic metformin following the coadministration of the histamine H2-receptor antagonist cimetidine has been reported [17]. Initially, it was thought that this interaction was mediated through the inhibition of the organic cation trans-porter 2, an active drug transporter expressed in cells of the proximal kidney tubules. However, more recently, it has been shown that MATE1, another transporter expressed in the kid-ney, is involved in the disposition of metformin and that this transporter is more likely to be responsible for the interaction with cimetidine [18]. Despite the interaction being moderate, it is of particular concern to clinicians, because metformin has the propensity for causing adverse effects, especially lactic acidosis, which may be related to high circulating concentrations of the drug. Other DDIs that involve the kidney include the cardiac glycoside digoxin, which is a substrate of the ATP-dependent
Yeo, Jamei & Rostami-Hodjegan
145www.expert-reviews.com
Review
efflux transporter P-glycoprotein (P-gp) and is mainly excreted unchanged in the kidney. In vitro and in vivo animal experiments have demonstrated the importance of both renal and intesti-nal P-gp in the disposition of digoxin [19,20]. Numerous drugs, including well-characterized P-gp inhibitors such as quinidine and verapamil, have been reported to reduce both the renal and nonrenal components of clearance of digoxin, thus causing an increase in levels above the upper limit of safety [21–23]. Again, despite the relatively small increase in the exposure of digoxin (34%), patients require close monitoring for untoward signs of digitalis-mediated toxicity because of the narrow therapeutic window of digoxin.
The contribution of the organic anion-transporting peptide (OATP) family of solute carriers to the hepatic elimination of many drugs and associated DDIs is significant [24]. The anti-tuberculosis drug rifampin, which is known for its strong induc-ing effect on drug-metabolizing enzymes, is also a relatively potent inhibitor of OATP1B1 and OATP1B3 in vitro [25]. In one study, a single 600 mg intravenous dose of rifampin caused a 600% increase in the mean plasma area under the curve (AUC) of atorvastatin, probably by inhibiting the OATP1B1- and/or OATP1B3-mediated hepatic uptake of atorvastatin [25]. Many OATP1B1 substrates, in addition to being substrates of other drug transporters, are often subject to metabolism by multiple CYP enzymes [26], which makes them susceptible to complex DDIs involving inhibition through multiple routes. For example, CYP3A4, CYP2C8 and OATP1B1 contribute to the disposition of repaglinide [27,28]. Although the plasma exposure of repaglinide is only moderately increased by CYP2C8 or CYP3A4 inhibitors (1.4- to 1.6-fold; [29,30]), a mean eightfold increase is observed fol-lowing their coadministration with the fibrate gemfibrozil, which leads to significant augmentation of the glucose-lowering effect of repaglinide [31]. Although gemfibrozil itself does not inhibit CYP3A4 and is only a moderate inhibitor of CYP2C8, its glucu-ronide conjugate is a relatively potent inhibitor of both CYP2C8 and OATP1B1. The magnitude of the observed interaction may be explained by combined inhibition of the hepatic uptake through OATP1B1 and of CYP2C8-mediated biotransformation of repaglinide. When itraconazole is given in addition to gemfi-brozil, the plasma AUC of repaglinide increases by 19-fold [31], although it ranges from 12.9- to 24.7-fold for the 12 individuals. This last example in particular demonstrates the importance of being able to predict the complex DDIs that may involve inhibi-tory metabolites, CYP- and transporter-mediated interactions, multiple inhibitors and interindividual variability.
Quantitative assessment of DDIs: static modelsAssessment of the DDI potential of drugs in development has been facilitated by advances in the understanding of the different CYP isoforms involved in drug metabolism and the availability of in vitro screening systems such as hepatic microsomes, hepato-cytes and recombinant enzymes. Quantitative assessment on the basis of the change in the area under the plasma drug concentra-tion–time curve (AUC
po) of a victim drug caused by competitive
enzyme inhibition can be estimated using Equation 1:
A inh
AUCpo UCpo
=+
1
1 [ ]
,
IKi u
(1)
where [I] is the relevant inhibitor concentration at the enzyme site and K
i,u is the inhibitor constant corrected for nonspe-
cific microsomal binding, which can be measured in vitro [32]. Estimation of the value of [I] requires some approximation since the enzyme site concentration cannot be easily measured. Indeed, there has been much debate about the most appropriate concentra-tion to use; the FDA recommends the use of mean total plasma C
max following the administration of the highest proposed clinical
dose, whereas Ito et al. have suggested the use of the maximum unbound concentration at the inlet to the liver as it appears to provide better predictions [33]. The equation should only be used in the case of a substrate whose in vivo clearance involves a single inhibited enzyme and when there is no gut metabolism [32]. This simple approach is often used during drug discovery and is also recommended in guidance documents issued by regulatory bodies [201,202]. Equation 1 has subsequently been expanded to account for the fraction of the dose of the victim drug metabolized by the inhibited enzyme (fm) [34] according to Equation 2.
AUCpoinh
AUCpo fm 1 [I]
K(1 fm)
i,u
=
+
+ −
1
(2)
However, it should be noted that the above equation only applies if the renal clearance (CL
R) of a drug is negligible com-
pared with the hepatic clearance (CLH), as discussed by Hisaka
et al. [35]. A more complete form of the equation that should be applied when CL
R contributes significantly to the clearance of a
drug is given as follows:
AUCpoinh
AUCpo =
+
++ −
+1
11
CLQ
fmIK
fm
R CLf
R
H
i u
B R
H[ ] ( )
.
,
PP H
RH
B
HP H
CLu
CLQ
Rf CLu
.
.
int,
int,
1 1+ +
(3)
where RB, Q
H, f
HP and CLu
int,H are the blood to plasma ratio of
the drug, hepatic blood flow, ratio of unbound drug concentration in the liver to the total concentration in the plasma and hepatic intrinsic clearance, respectively.
The more commonly applied Equation 2 has also been expanded to consider enzyme inhibition in the gut wall during first-pass [36,37], mechanism-based inhibition [36–38] and enzyme induc-tion [39–42]. Predictions of DDIs involving mechanism-based inhibition or induction requires knowledge of the degradation
Predicting drug–drug interactions
Expert Rev. Clin. Pharmacol. 6(2), (2013)146
Review
rate constant of the affected enzyme(s) [43] and drug-dependent data (k
inact and K
I: the rate constant defining the maximal rate of
enzyme inactivation and the inhibitor concentration associated with half maximal inactivation, respectively; E
max and EC
50 are the
maximal increase in the level of induced enzyme and the unbound inducer concentration associated with half maximal induction, respectively). Despite their simplicity, equations describing ‘static’ models appear to have been reasonably successful in recovering the observed AUC ratios for CYP-mediated competitive and mechanism-based inhibition and induction [40–42,44–52].
Current draft guidance issued by the FDA [201] provides recom-mendations on the basis of the in vitro data for the assessment of P-gp-mediated drug interaction potential for P-gp substrates and inhibitors [53,54]. For inhibitors, a clinical DDI study with digoxin is recommended if the maximum systemic inhibitor concentration [I
1] at steady state relative to inhibitory potency (K
i) (or concen-
tration of inhibitor to inhibit 50% of the P-gp activity; IC50
) ratio is >0.1. The use of the gut concentration of the inhibitor [I
2] (dose/250 ml) has been proposed as an alternative to predict
the drug interactions that occur during the absorption phase, but using [I
2]/IC
50 > 10 as the cutoff. Fenner et al. reported the results
of a study where both [I1]/IC
50 and [I
2]/IC
50, assumed to be rep-
resentative of interactions in the systemic and gut compartments, respectively, were calculated and compared with peak plasma con-centrations at steady state (C
max)
i,ss/C
max,ss and AUC
i/AUC ratios
from 26 clinical trials involving digoxin and nine P-gp inhibitors [55]. Use of the [I
1]/IC
50 > 0.1 cutoff was associated with a high
percentage of false negatives (24 and 41% on the basis of AUC and C
max, respectively), whereas use of the [I
2]/IC
50 >10 cutoff
generated a lower number of false negatives but a much higher proportion of false positives (50 vs 13%). Comparable results were obtained in the ‘P-gp IC
50 working initiative’, where 23
industrial laboratories compared their in-house cutoffs as well as overall cutoffs generated in four different in vitro systems and for 16 compounds [56]. Despite the fact that in vitro methods, model substrates and inhibitors for P-gp are well established, there is still a need for a more accurate quantitative approach for the predic-tion of DDIs involving P-gp. For other transporters, there is a lack of specific substrates and inhibitors that makes the assessment of the transporter-mediated DDI potential difficult. However, static models have been used for the prediction of DDIs involv-ing the transporter OATP1B1 and to date, the results appear to be positive [57–60].
Although the view is still held by some that static models remain an integral part of the drug development process, this empirical approach is being increasingly replaced by PBPK modeling in the pharmaceutical industry due to superior predictive power [61]. More complex DDIs involving a time- or concentration-dependent com-ponent can also be simulated and aspects of clinical study design such as the dosage regimen or formulation effects can be assessed. PBPK modeling is not a new concept, and until recently, its applica-tion has been largely related to the field of environmental toxicol-ogy rather than to drug development. However, there have been an increasing number of applications of PBPK modeling in the pharmaceutical industry over the past few years (see next section).
PBPK–IVIVE-linked modelsPBPK models consist of three major components: system-specific properties, drug-related parameters and the structural model. System-specific properties include organ mass or volume, arterial and venous blood flows through the individual organs and tissue composition. Drug-related parameters include plasma protein-binding affinity, membrane permeability, enzymatic stability, tis-sue partitioning and transporter-mediated uptake or efflux. The structural model comprises the anatomical arrangement of tissues and organs of the body as distinct entities, linked by perfusing blood and is coded as a series of linked differential equations that are solved numerically to track the time course of drug concentra-tions in blood and tissues after dosing. A whole-body PBPK model contains submodels describing the processes of ADME. In each of these submodels, drug-related data are scaled by system-specific properties to predict the pharmacokinetic parameters relevant to ADME. PBPK–IVIVE-linked models [3], the parameters defin-ing ADME processes and their extrapolation from in vitro data have already been described in detail elsewhere [2,4,62]. However, to demonstrate the approach, we give an overview of predic-tion of clearance using IVIVE as this is particularly relevant for quantitative assessment of DDI.
Prediction of clearance: a systems biology approachPrediction of drug clearance from in vitro enzyme kinetic data is performed in two steps and has been described in detail previously [63–65]. Initially, intrinsic clearance per unit enzyme is converted to a whole organ intrinsic metabolic clearance using relevant scal-ing parameters for the liver (system properties). This value is then combined with binding parameters and liver blood flow to extrapolate to a whole organ clearance [63,64]. There are numerous in vitro screening systems available for assessing metabolic intrin-sic clearance, including human liver subfractions (microsomes, cytosol and hepatocytes) and recombinantly expressed enzymes. The use of expressed enzymes allows the intrinsic clearance values to be expressed per unit of enzyme (e.g., CYP or UGT) and are scaled to a whole liver metabolic clearance (CLu
int,H) according
to Equation 4:
CLuint,H = ××
=ISEF ji
V i rhEnzi Enzi abundance
Kmi rhEnZii
n max ( )
( )1∑∑
=∑
× ×
j
n
MPPGL Liver weight
1
_
(4)
where there are i metabolic pathways for each of j enzymes; ‘rh’ indicates recombinantly expressed enzyme; V
max is the maximum
rate of metabolism by an individual enzyme; Km is the Michaelis
constant; Enz abundance is the amount (pmol P450) per milli-gram of microsomal protein; MPPGL is the amount of microso-mal protein per gram of liver [66]; and intersystem extrapolation factor (ISEF) is a scaling factor that compensates for any difference in the activity per unit of enzyme between recombinant systems and hepatic enzymes [67]. The relative success of IVIVE in using data from a variety of in vitro systems, including human liver microsomes, recombinant enzymes and hepatocytes to predict
Yeo, Jamei & Rostami-Hodjegan
147www.expert-reviews.com
Review
hepatic clearance [65,68–71] is probably due to the increasing amount of available information on appropriate scaling factors [66,72].
For accurate prediction of DDIs, it is necessary to ensure that the fraction of the victim drug metabolized by the inhibited enzyme (fm) is correct. When using recombinant systems, the fm value for each contributing enzyme is calculated intrinsically through the bottom-up approach and is obviously dependent on the metabolic clearance of the enzyme, as well as the abundance and ISEF value [2]. For human liver microsomes, chemical inhibi-tors can be used to estimate the contribution of each enzyme to the overall metabolic intrinsic clearance of a compound. Youdim et al. used both approaches for a series of drugs that were mainly metabolized by CYP3A4 and found that the former (recombinant enzymes) gave improved accuracy when comparing predicted magnitudes of DDIs against observed data [50].
Building in system properties creates the opportunity for quantitative assessment of the impact of covariates on the IVIVE of whole organ metabolic intrinsic clearance; these include genetics [73], age (ontogeny) [74], ethnicity [75], sex [76], pregnancy [77], obesity [78], comorbid diseases such as cirrhosis [79,80] and chronic kidney failure [81,82] and environmental fac-tors such as smoking [83]. For clearance predictions, understand-ing the value of the unbound fraction of the drug in plasma in an individual, based on the concentration levels of albumin, α-acid-glycoprotein and plasma lipids, is an integral part of the exercise. Although this can be measured in vitro for any target population, defining protein binding on the basis of binding affinities to various plasma proteins and partition to lipids enables automatic extrapolations to groups whose modi-fied plasma protein levels are well characterized, such as the elderly, pregnant women and neonates [84].
Another key advantage of PBPK–IVIVE-linked models is the ability to include sources of physiological and biochemical variability in the system parameters and to simulate the expected pharmacokinetics in a population of individuals rather than just for an average subject. A virtual population can be generated from values and formulae describing demographic, anatomical and physiological variables using a correlated Monte Carlo approach [5]. Equations describing the distributions of the system param-eters relevant to IVIVE and PBPK modeling are derived from distributions of data on the basis of real populations and patients [5]. This allows the prediction of variability prior to clinical stud-ies in contrast to a statistical approach (POP-PK analysis), which requires previous clinical data to characterize variability. Several studies have demonstrated how incorporation of variability into PBPK–IVIVE-linked models can recover the variability in expo-sure of some drugs in a population [71,85,86]. This aspect is particu-larly important when considering the risk associated with a DDI, as it is usually a few individuals with certain characteristics who are a greater concern than the average individual.
Prediction of clearance: a dynamic modelIt is obvious that the IVIVE scaling approach described for the prediction of clearance will not automatically consider the effects of time- or concentration-related nonlinearity. However, once
it is incorporated into the differential equations describing the PBPK models, CLu
int,H can be expressed as a concentration- or
time-variant parameter using the following Equation 5:
CLuint,H = ××
+ISEF ji
V i rhEnzi Enzi abundance
Kmi rhEnZi S
max ( )
( ) [ ]llivi
n
j
n
MPPGL Liver weight
=∑
=∑
× ×
11
_
(5)
where [S]liv
is the concentration of the drug in the liver (assumed to be the same as that at the enzyme site). In addi-tion to incorporating the (1+[I]/Ki) factor on the denominator to account for competitive inhibition by a perpetrator, submodels defining a dynamic pool of time-variant enzymes that respond to induction or suppression of enzyme synthesis and accelerated or stabilized enzyme degradation can also be introduced [39,43,87]. These dynamic models are more accurate than static models in recovering the extent of metabolic interactions, especially those involving time-dependent changes in enzyme abundance [51,61,88]. This improved accuracy is partly due to the capability of the PBPK models to consider the time-varying concentrations of substrate and inhibitor (rather than the use of a single average value of [I]). These models have also been used to assess more complex scenarios involving simultaneous dose-dependent inhi-bition and induction [88] as well as the inhibitory effects of both parent drug and metabolites [87,89,90]. In addition, they have been used to evaluate aspects of experimental design including dosage, choice of dosage form and the timing of dosage of interacting compounds [91,92].
Applications of PBPK models for the prediction of DDIsOver the past decade, the number of publications involving PBPK modeling has largely increased, demonstrating the widespread use of this approach within the scientific community [93–97]. As a result of this and increasing availability of commercial platforms that integrate this methodology, such as the Simcyp population-based simulator (Simcyp, Sheffield, UK) [203], PK-Sim (Bayer, Leverkusen, Germany) [204] and Gastroplus™ (Simulations Plus, CA, USA) [205], there has been growing interest in the applica-tion of PBPK modeling by the pharmaceutical industry [6,98]. There are now several examples of the use of PBPK–IVIVE-linked models during drug development for decision-making on such issues as candidate selection, first-in-human dose finding, assess-ment of DDI potential and definition of appropriate study designs involving DDIs, or inclusion/exclusion criteria for conducting studies of drugs metabolized by polymorphic enzymes [99–102]. Moreover, pharmaceutical companies are now including this modeling approach in dossiers submitted to regulatory agencies [7,8]. Indeed, there were 25 such cases included in submissions to the FDA between June 2008 and December 2011, and of these, 60% were related to the prediction of DDIs using commercially available software [8]. Although some cases related to the predic-tion of complex DDIs, they all involved CYP-mediated interac-tions. Despite the recent progress made with respect to incor-porating transporters into PBPK models [103–106] and predicting
Predicting drug–drug interactions
Expert Rev. Clin. Pharmacol. 6(2), (2013)148
Review
transporter-mediated DDIs [103–105,107], many challenges remain. For example, there are issues with the in vitro systems and the physiological limitations in mimicking the in vivo situation, in particular the interplay between enzymes and transporters and the possible compensatory increase in activities of one or more other transporters when the activity of a transporter is suppressed. In addition, relevant scaling factors such as abundances of individual proteins are required for IVIVE of transporter data [108]. Recent advances in using LC-MS techniques to measure the abundance and variability of each transporter in various tissues will facili-tate this, although some coordination and global validation of the results using the same set of samples might be necessary to overcome some discrepancies in the reported values [109].
Ultimately, PBPK–IVIVE-linked models incorporating both transporters and CYP enzymes as well as other intrinsic and extrinsic patient factors are required for the assessment of an individual’s risk of DDIs, especially those involving multiple inhibitors. The latter are a particular regulatory concern and it is not possible to conduct all possible combinations of in vivo studies [110]. Simulations can be used to assess the worst-case combination(s) for in vivo evaluation. This is most likely to involve the combination of drugs that inhibit different enzymes or transporters such that the effects are more than additive, whereas in the case of drugs competing for the same enzyme, the net effect tends to default to that of the more potent compound [34]. Similarly, if both transporters and enzymes are inhibited, as in the case of repaglinide when it is coadministered with gemfibrozil, the magnitude of interaction may be greater than anticipated [31]. PBPK modeling can be applied to assess these complex scenarios and help provide information on patients who are at risk, or those who for ethical reasons cannot be investigated in formal clinical trials.
This is of particular importance for a pediatric population. Rapid changes in organ maturation, organ blood flow and body composition occur in children. In addition, there is a variable ontogeny with respect to enzymes and transporters. The vari-ability in the maturation rates of various human liver enzymes suggests that the typical fm estimated for an adult cannot be assumed to be a fixed value from birth to adulthood. As the fm is the main determinant of the level of susceptibility to DDI, the significance of a given metabolic route in the biotransformation of a particular drug in neonates and young children could be much higher or much lower than in adults. Incorporation of the relevant scalars [74,111] into PBPK–IVIVE-linked models can be used to predict the magnitude of interaction in young children of differing ages [3]. Although the predictions from such mod-els cannot be validated due to a the lack of observed data, and therefore should be treated with caution, they do allow one to ask ‘what if ’ questions when clinical studies, for ethical reasons, cannot be performed.
Case studies: integration of bottom-up and top-down approachesAlthough the application of PBPK models can lead to successful prediction of DDIs, in some cases it may not be possible owing to
a lack of in vitro data or knowledge gaps in the models ,or more likely that the ADME properties of the drug of interest have not been characterized fully. In these instances, clinical data can be used in conjunction with available in vitro data to estimate the missing or ‘unknown’ parameters. When parameter estimation is applied to a PBPK model, it is important to retain the prior physiological knowledge it embodies. To facilitate and accelerate model building and covariate recognition in drug development, the top-down POP-PK analysis of clinical [112,113] data, includ-ing maximum likelihood [114,115] or Bayesian methodology [116], can be used in combination with bottom-up PBPK approaches. Bayesian methods extend the maximum likelihood approach by incorporating previous distributions on the various unknown parameters. Previous knowledge from different sources, such as in vitro data or clinical trial data, can be incorporated in the fitting algorithm. Therefore, Bayesian approaches have emerged as the best suited for PBPK models, given the large amount of previous information they incorporate [117]. Indeed, this approach was used by Idkaidek and Arafat to investigate the pharmacoki-netics of ibuprofen under microgravity and normal gravity condi-tions [118]. Plasma concentration–time profiles of ibuprofen for six healthy volunteers were incorporated into the Simcyp simu-lator and used in conjunction with a PBPK model and in vitro data for ibuprofen to estimate the absorption kinetic parameters of ibuprofen under normal gravity conditions. The model was then used to predict the pharmacokinetics under microgravity conditions.
In some cases, where clinical data for individual subjects are not available, POP-PK analysis cannot be applied, but it is still pos-sible to apply a top-down fitting approach using the mean in vivo concentration–time profile. We present two detailed examples of application of the Simcyp simulator [204] to demonstrate how this methodology can be used to develop robust models of drugs with complex kinetics (undergo autoinhibition or are taken up into the liver by transporters). The general approach is described below:
• Step 1: usually when developing a PBPK-IVIVE-linked model, the initial step involves using the available in vitro data for the drug of interest to simulate concentration–time profiles that are then compared with observed data;
• Step 2: if the predicted and observed profiles are not consistent, in vivo data can be used in conjunction with the known in vitro data to fit for a missing or unknown parameter that is required as an input to the PBPK–IVIVE-linked model;
• Step 3: the model is then validated to ensure that inclusion of the unknown parameter allows recovery of the observed data. The latter should be taken from validation set; that is, clinical data that have not been used for the development of the origi-nal model [8]. If the model is not able to recover the observed data, then model should be revised accordingly. This iterative procedure is captured in the article by Vieira et al. [119].
The refined and validated model can then be used to predict DDIs prospectively in individuals at extreme risk or in subjects who for ethical reasons cannot be investigated in formal clinical
Yeo, Jamei & Rostami-Hodjegan
149www.expert-reviews.com
Review
trials. In the following examples, the Simcyp simulator was used to generate a population of virtual individuals using a correlated Monte Carlo approach [5,71]. Each population was generated using values and formulae describing demographic, anatomical and physiological variables [5,71].
Case study 1: paroxetineParoxetine is a selective serotonin reuptake inhibitor antidepres-sant that is used to treat major depression and obsessive–com-pulsive, panic, social anxiety and generalized anxiety disorders in adult outpatients. The agent is an established perpetrator of DDIs and when coadministered with substrates that are metabolized by CYP2D6, including desipramine [119,120], metoprolol [121] and atomoxetine [122], it causes a fivefold to eightfold reduction in the clearances of the victim drugs. Paroxetine is also metabolized by CYP2D6 and exhibits nonlinear accumulation kinetics in extensive metabolizer (EM) subjects of CYP2D6 during single and multiple dosing [123,124]. Following administration of a single dose of paroxetine, there is a sevenfold difference in the median total clearance of poor metabolizer (PM) and EM subjects of CYP2D6, which is then reduced to twofold at steady state. The nonlinear kinetics of paroxetine are much more prominent in EMs than PMs, mainly due to time-dependent autoinhibition of the CYP2D6-mediated metabolism [125]. Indeed, it has been reported that paroxetine causes concentration- and time-dependent inhibi-tion of human liver microsomal CYP2D6 activity in vitro. There is also biochemical evidence of the formation of a metabolite-intermediate complex with CYP2D6 via a carbene–heme moiety, thus confirming that paroxetine is a mechanism-based inhibitor of CYP2D6. Other enzymes, including CYP1A2, CYP3A4/5 and CYP2C19, contribute to the metabolism of paroxetine but to a lesser extent [126].
Simulations of paroxetine were performed using in vitro data reported by Jornil et al. [127] and the study design described by Sindrup et al. [125]. Therefore, virtual trials of nine EM Caucasian subjects receiving 10 mg paroxetine once daily (q.d.) for 14 days were generated. The simulated plasma con-centration–time profiles were able to recover the observed data after a single dose of paroxetine (FigurE 1, StEp 1) but not follow-ing multiple dosing (data not shown). Initially, it was postu-lated that this could be due to an overestimation of either the contribution of CYP2D6 to the overall metabolism of parox-etine or of the CYP2D6 inactivation potency of paroxetine or an underestimation of the turnover rate for CYP2D6 [128]. However, when concentration–time profiles were generated for paroxetine in PM subjects, it was observed that even after a single dose, the exposure was too high (FigurE 1, StEp 1). IVIVE of enzyme kinetic data for CYP1A2, CYP3A4/5 and CYP2C19 was not able to recover the observed clearance of paroxetine in PMs, and therefore it was likely that there was an additional unknown metabolic route that had not been identified during the assessment of in vitro activities. Hence, a top-down fitting approach (in vivo concentration–time profile of paroxetine after a single oral dose in PMs [124]) was combined with bottom-up extrapolation of all prior in vitro data for paroxetine within
the parameter estimation module of the Simcyp simulator to obtain an estimate of 61.9 µl/min/mg protein for the additional metabolic component (FigurE 1, StEp 2). Inclusion of the additional component of metabolism improved the model fit to in vivo data during multiple dosing of paroxetine in Caucasian PM and EM subjects (FigurE 1, StEp 3).
The refined paroxetine model was then used to predict the DDI with terbinafine (using default parameters specified in the Simcyp simulator [203]) in Japanese EM subjects and compared with observed data reported by Yasui-Furokori et al. [129]. System parameters for IVIVE in a Japanese population, such as those describing the demographic, anatomical and physiological vari-ables (including enzyme abundance), were the same as those reported by Inoue et al. [75]. The predicted increase in plasma AUC
(0–∞) after a single oral dose of paroxetine (victim drug) dur-
ing coadministration of terbinafine (125 mg q.d.) in Japanese EM subjects was 2.6-fold (range for ten virtual trials: 1.6- to 3.0-fold), which was reasonably consistent with the observed value of 3.0-fold. The variability across trials and in the clinical study is shown in FigurE 2a. At least one of the virtual trials was similar to that of the observed study with respect to the median and variability (FigurE 2a). Of note, in the virtual Japanese population, the range of values for the predicted increase in AUC
(0–∞) was
1.9- to 14.9-fold. The individual with the highest predicted AUC ratio had a relatively low CYP2D6 turnover and a high fm
CYP2D6.
Simulated profiles of active CYP2D6 in the liver indicate that approximately 20% remains following the administration of a single oral dose of 20 mg paroxetine (FigurE 2B). Although chronic dosing of paroxetine was not performed during the clinical study, Yasui-Furokori et al. discussed the magnitude of interaction dur-ing multiple dosing of both paroxetine and terbinafine [129]. The authors suggested that this may be reduced after repeated doses due to autoinhibition of paroxetine metabolism and that further studies were required to confirm this. In the absence of such data, we have simulated the impact of terbinafine coadministra-tion with paroxetine during chronic dosing and found that the predicted AUC ratio is indeed attenuated: it was found to be 1.08-fold (range for ten virtual trials: 1.05- to 1.19-fold). The simulations confirm that there is <5% active CYP2D6 remain-ing after chronic dosing with paroxetine (FigurE 2B). Although there were no in vivo data to confirm whether the prediction was correct, the fact that the paroxetine model was able to recover observed data for other scenarios may provide some confidence in the prospective DDI assessment.
Case study 2: repaglinideRepaglinide is a short-acting meglitinide analog antidiabetic drug used in the treatment of Type 2 diabetes mellitus [127]. It lowers blood glucose concentrations by enhancing glucose-stimulated insulin release in pancreatic β cells. Repaglinide is rapidly absorbed following oral administration and undergoes first-pass metabolism, resulting in a 60% bioavailability. CYP3A4 and CYP2C8 are the main enzymes responsible for the oxidative metabolism of the compound [27,28]. The AUC of repaglinide is increased markedly in homozygous carriers of the SLC01B1
Predicting drug–drug interactions
Expert Rev. Clin. Pharmacol. 6(2), (2013)150
Review
521T>C (Val174Ala) single-nucleotide polymorphism, suggest-ing that it is a substrate of the SLCO1B1-encoded hepatic uptake transporter OATP1B1 [130].
Previous in vitro information on the metabolism and kinetics of repaglinide (using default parameters specified in the Simcyp simulator [203]) were incorporated into a PBPK model within the Simcyp simulator. While there are in vivo data to support the OATP1B1-mediated hepatic uptake of repaglinide, in vitro parameters describing this transport were not available to the authors. Hence, we combined a top-down fitting approach (in vivo concentration–time profile of repaglinide [131] with bottom-up extrapolation of all previous in vitro data within the parameter estimation module of the Simcyp simulator to obtain an estimate of 282 µl/min/million cells for hepatic uptake clear-ance of repaglinide via OATP1B1 (FigurE 3, StEp 2). Inclusion of
hepatic uptake through OATP1B1 improved the model fit to the in vivo data [29,30]. However, more importantly, on average, predicted increases in plasma AUC
(0–∞) of repaglinide during
the coadministration of the CYP2C8 inhibitor trimethoprim ( 1.3-fold; 160 mg q.d.) [29] and the CYP3A4 inhibitor clarithro-mycin ( 1.4-fold; 250 mg twice daily) [30] were consistent with observed values of 1.6- and 1.4-fold, respectively. Furthermore, the model was able to recover the transporter-mediated DDI with cyclosporin (100 mg twice daily) [132]; predicted versus observed AUC ratios were 2.0 and 2.4, respectively (FigurE 3, StEp 3). Finally, coadministration of all three inhibitors with repaglin-ide in our virtual clinical trial was associated with a 5.6-fold increase in AUC
(0–∞) on average (FigurE 4), but values ranged
from 2.3- to 17.7-fold in the virtual population. Although there were no in vivo data to confirm whether the predicted
Figure 1. The workflow for the development of a physiologically based pharmacokinetic model for paroxetine using ‘top-down’ and ‘bottom-up’ approaches. EM: Extensive metabolizer; IVIVE: In vitro–in vivo extrapolation; PBPK: Physiologically based pharmacokinetic; PM: Poor metabolizer.
PBPK-IVIVE
In vitro parametersparoxetine
PBPK-IVIVE
In vitro parametersparoxetine
+
Step 1
Step 2
PBPK-IVIVE
In vitro parametersparoxetine
Step 3
EM12
10
Sys
tem
ic c
on
cen
trat
ion
(n
g/m
l)S
yste
mic
co
nce
ntr
atio
n (
ng
/ml)
8
6
4
2
00 12 24 36 48
Time (h)
0
10
20
30
40
50
60
0 48 96 144 192 240 288 336Time (h)
Parameter X
EM
Parameter X
PM
Parameter estimation
PM
0 12 24 36 48Time (h)
0
5
10
15
20
25
30
35
Sys
tem
ic c
on
cen
trat
ion
(n
g/m
l)S
yste
mic
co
nce
ntr
atio
n (
ng
/ml)
0102030405060708090
0 48 96 144 192 240 288 336Time (h)
05
10152025
0 4 8 12 16 20 24Time (h)
Sys
tem
icco
nce
ntr
atio
n (
ng
/ml)
Yeo, Jamei & Rostami-Hodjegan
151www.expert-reviews.com
Review
Figure 2. Predicted drug–drug interaction between paroxetine and terbinafine in a Japanese population. (A) Predicted median AUC ratios (95% CIs) of paroxetine following a single oral dose of 20 mg in the absence of terbinafine and coadministered with terbinafine (125 mg once daily for 6 days) in ten different randomly selected trials of virtual Japanese subjects (n = 12) and observed (black circle) values. (B) Predicted profiles of active CYP2D6 remaining in the liver following a single oral dose of 20 mg paroxetine in the absence of terbinafine (solid line) and coadministered with terbinafine (125 mg once daily for 6 days) in a population of virtual Japanese subjects (n = 120; 10 × 12). The lower dotted line is representative of the active CYP2D6 remaining due to autoinhibition of paroxetine metabolism during chronic dosing. AUC: Area under the curve. Data taken from [128,129].
8.0
7.0
6.0
5.0
4.0
3.0
2.0
1.00
0
10
20
30
40
50
60
70
80
90
100
0 24 48 72 96 120 144 168 192 216
Time (h)
Act
ive
CY
P2D
6 (%
)
1 2 3 4 5 6 7 8 9 10 11 12
Trial groups
AU
C r
atio
A B
Figure 3. The workflow for the development of a physiologically based pharmacokinetic model for repaglinide using ‘top-down’ and ‘bottom-up’ approaches. b.i.d.: Twice daily; IVIVE: In vitro–in vivo extrapolation; PBPK: Physiologically based pharmacokinetic.
PBPK-IVIVE
In vitro parametersrepaglinide
+
Step 2
PBPK-IVIVE
In vitro parametersrepaglinide
Step 3
OATP1B1-mediatedhepatic uptake
OATP1B1-mediatedhepatic uptake
0
1
2
3
4
5
6
7
0 2 4 6 8
Co
nce
ntr
atio
n (
µg
/ml)
Time (h)
0
1
2
3
4
5
6
7
22 24 26 28 30 32
Co
nce
ntr
atio
n (
ug
/ml)
Time (h)
Control – repaglinide(0.25 mg)
+ cyclosporin(100 mg b.i.d. for 2 days)
Parameter estimation
05
10152025
0 4 8 12 16 20 24Time (h)
Sys
tem
icco
nce
ntr
atio
n (
ng
/ml)
Predicting drug–drug interactions
Expert Rev. Clin. Pharmacol. 6(2), (2013)152
Review
magnitude of interaction for the complex DDI involving inhi-bition of CYP2C8-, CYP3A4- and OATP1B1-mediated uptake was accurate, the fact that the repaglinide model was able to recover observed data for each of the clinical DDIs involving a single inhibitor indicates that it is likely to be.
Expert commentaryApplication of mechanistic PBPK models for the prediction of DDIs in adults has been widely accepted by most of the leading pharmaceutical companies and regulatory bodies. Indeed, docu-ments providing guidance on the assessment of the DDI potential of drugs in development were updated and issued earlier this year by both the FDA [201] and the EMA [202]; the use of PBPK modeling has been advocated by both agencies. Furthermore, a discussion on best practice in the use of PBPK modeling to address regulatory questions in the area of clinical pharmacology has also been published by Zhao et al. [8]. Therefore, it appears that PBPK modeling is definitely ‘here to stay’ as commented on recently in a review [133].
While it is encouraging to observe that PBPK–IVIVE-linked models are being increasingly used by the pharmaceutical indus-try, it has become apparent that there are many ‘system parameters’ that are lacking, such as abundance of enzymes and transporters and other relevant proteins and information on physiology and biology in different ethnic populations and disease groups, all of which are relevant for the prediction of DDIs. Despite this, efforts have been much focused on the development of refined in vitro systems for the accurate prediction of the drug-related ADME parameters. Economic constraints within the pharma-ceutical industry have led to growth in precompetitive research collaborations – it is envisaged that this will lead to increasing availability of the system parameters as the resource issue can be shared among interested parties.
On the basis of numerous examples published in the literature and case studies provided in this review, it is evident that sophisticated PBPK–IVIVE-linked mod-els on the basis of bottom-up approaches or a combination of both bottom-up and top-down approaches can be used to pre-dict DDIs in the ‘average human’ with reasonable accuracy. While attempts have been made to confirm that the predicted variability in a population is consistent with that observed in vivo, in most cases it is not possible to do this as clinical DDI studies are generally performed in a small number of healthy volunteer subjects. Similarly, for the prediction of complex DDIs involving multiple inhibitors, there are few in vivo studies available to validate the models. Furthermore, in pediatric or other special populations, observed DDIs are often based on individual case reports rather than controlled clinical studies.
Usually, in a clinical pharmacology setting, there are some in vivo data that can be used to refine (if necessary) and validate the PBPK models to some extent. This should provide some confidence that the model can then be applied prospectively for the prediction of DDIs in the patient population of interest or high-risk individuals with comorbid states (if system param-eters are available) or for the prediction of complex DDIs with multiple mechanisms or inhibitors. The latter are a particular regulatory concern, and it is not possible to cover the various permutations of combinations by conducting in vivo studies [109]. There will always be the ongoing debate about whether PBPK–IVIVE-linked models can replace actual clinical stud-ies. In the absence of clinical data, PBPK models can be used to assess the worst-case scenario for in vivo evaluation and to provide guidance to prescribers for labeling.
Five-year viewApplication of IVIVE in conjunction with PBPK models for the prediction of DDIs resulting in pharmacokinetic changes is now widely accepted by the pharmaceutical industry and regulatory bodies. There is a growing recognition of the fact that a mechanistic modeling approach is also required to assess the pharmacodynamic DDIs in vivo [9,10]. Combining mechanistic PD models and vari-ability in pharmacological response (including receptor genotype) with PBPK–IVIVE-linked models can perhaps be viewed as the first step toward the provision of ‘personalized medicine’. PBPK-PD models are likely to become useful in the healthcare arena as an educational tool and but more importantly to provide information to clinicians on a safe, effective, individualized dosage and potential DDIs [3,6]. However, appropriate education for potential prescribers and users will have to be considered in parallel to addressing the continuous quest for robust data relating to system parameters that are required for the PBPK and PD models.
Figure 4. Predicted drug–drug interactions of repaglinide coadministered with trimethoprim, clarithromycin and cyclosporin. Predicted AUC ratios of repaglinide during the coadministration of the CYP2C8 inhibitor trimethoprim (1.3-fold; 160 mg b.i.d.), the CYP3A4 inhibitor clarithromycin (1.4-fold; 250 mg b.i.d.) and the OATP1B1 inhibitor cyclosporin (2.0-fold; 100 mg b.i.d.) were reasonably consistent with all observed data, and therefore, the repaglinide model was used to assess the magnitude of interaction when all three inhibitors were coprescribed concomitantly. AUC: Area under the curve; b.i.d.: Twice daily. Data taken from [29,30,131,132].
Pre
dic
ted
fo
ldin
crea
se in
AU
C
1234567
+Trimethoprim(160 mg b.i.d)
+Clarithromycin(250 mg b.i.d)
+ All three
inhibitors
+Cyclosporin
(100 mg b.i.d)
1.6-fold 1.4-fold 2.4-fold
Observed data
CYP2C8 CYP3A4 OATP1B1
No observed data
Yeo, Jamei & Rostami-Hodjegan
153www.expert-reviews.com
Review
Key issues
• Drug–drug interactions (DDIs) remain an important issue in clinical practice and during the development of new drugs.
• Currently, many drugs in development are metabolically stable, and transporter-mediated disposition is becoming more of a concern. Therefore, issues related to the prediction of DDIs are becoming more complex.
• Physiologically based pharmacokinetic (PBPK) models are increasingly replacing more empirical approaches for predicting DDIs in the pharmaceutical industry due to the fact that they account for time-varying concentrations of interacting drugs.
• The use of PBPK modeling to assess the DDI potential of drugs in development is advocated in the recent DDI guidance issued by the US FDA.
• Numerous examples in the literature demonstrate that PBPK models can be used to accurately predict DDIs in the ‘average human’. More clinical data are required to determine whether the application of the systems biology approach allows accurate recovery of the variability associated with the DDI in a population and if those at extreme risk can be identified.
• Despite the recent progress made with respect to incorporating transporters in PBPK models and predicting transporter-mediated DDIs, many challenges remain, including the lack of relevant scaling factors (e.g., abundance data for individual proteins), which are required for in vitro–in vivo extrapolation of transporter data.
• Application of a ‘top-down’ fitting approach in combination with ‘bottom-up’ in vitro–in vivo extrapolation can be used to develop optimized PBPK models that can then be used prospectively to predict the potential interactions in individuals who for ethical reasons cannot be investigated in formal clinical trials.
ReferencesPapers of special note have been highlighted as:•ofinterest••ofconsiderableinterest
1 Ribbing J, Jonsson EN. Power, selection bias and predictive performance of the population pharmacokinetic covariate model. J. Pharmacokinet. Pharmacodyn. 31(2), 109–134 (2004).
2 Rostami-Hodjegan A, Tucker GT. Simulation and prediction of in vivo drug metabolism in human populations from in vitro data. Nat. Rev. Drug Discov. 6(2), 140–148 (2007).
3 Rostami-Hodjegan A. Physiologically based pharmacokinetics joined with in vitro–in vivo extrapolation of ADME: a marriage under the arch of systems pharmacology. Clin. Pharmacol. Ther. 92(1), 50–61 (2012).
4 Gibson GG, Rostami-Hodjegan A. Modelling and simulation in prediction of human xenobiotic absorption, distribution,
metabolism and excretion (ADME): in vitro–in vivo extrapolations (IVIVE). Xenobiotica 37(10–11), 1013–1014 (2007).
5 Jamei M, Dickinson GL, Rostami-Hodjegan A. A framework for assessing inter-individual variability in pharmacokinetics using virtual human populations and integrating general knowledge of physical chemistry, biology, anatomy, physiology and genetics: a tale of ‘bottom-up’ vs ‘top-down’ recognition of covariates. Drug Metab. Pharmacokinet. 24(1), 53–75 (2009).
6 Rowland M, Peck C, Tucker G. Physiologically-based pharmacokinetics in drug development and regulatory science. Annu. Rev. Pharmacol. Toxicol. 51, 45–73 (2011).
7 Zhao P, Zhang L, Grillo JA et al. Applications of physiologically based pharmacokinetic (PBPK) modeling and simulation during regulatory review. Clin. Pharmacol. Ther. 89(2), 259–267 (2011).
8 Zhao P, Rowland M, Huang SM. Best practice in the use of physiologically based pharmacokinetic modeling and simulation to address clinical pharmacology regulatory questions. Clin. Pharmacol. Ther. 92(1), 17–20 (2012).
9 Jonker DM, Visser SA, van der Graaf PH, Voskuyl RA, Danhof M. Towards a mechanism-based analysis of pharmacody-namic drug–drug interactions in vivo. Pharmacol. Ther. 106(1), 1–18 (2005).
10 van der Graaf PH, Benson N. Systems pharmacology: bridging systems biology and pharmacokinetics-pharmacodynamics (PKPD) in drug discovery and development. Pharm. Res. 28(7), 1460–1464 (2011).
11 Chin TW, Loeb M, Fong IW. Effects of an acidic beverage (Coca-Cola) on absorption of ketoconazole. Antimicrob. Agents Chemother. 39(8), 1671–1675 (1995).
12 Giacomini KM, Huang SM, Tweedie DJ et al. Membrane transporters in drug development. Nat. Rev. Drug Discov. 9(3), 215–236 (2010).
13 Huang SM, Zhang L, Giacomini KM. The International Transporter Consortium:
AcknowledgementsThe authors thank James Kay for his assistance with the preparation of this manuscript.
Financial & competing interests disclosureKR Yeo and M Jamei are employed by, and A Rostami-Hodjegan is currently seconded to on a part-time basis, Simcyp (a Certara company), which pro-vides a commercial physiologically based pharmacokinetic in vitro–in vivo
extrapolation simulator for the pharmaceutical industry. The simulator is freely available, after the completion of the relevant workshops, to centers of excellence in pharmacology and pharmacometrics within the academic and other not-for-profit institutions for research and teaching purposes. The authors have no other relevant affiliations or financial involvement with any organiza-tion or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.
No writing assistance was utilized in the production of this manuscript.
Predicting drug–drug interactions
Expert Rev. Clin. Pharmacol. 6(2), (2013)154
Review
a collaborative group of scientists from academia, industry, and the FDA. Clin. Pharmacol. Ther. 87(1), 32–36 (2010).
15 Venkatakrishnan K, Obach RS. Drug–drug interactions via mechanism-based cytochrome P450 inactivation: points to consider for risk assessment from in vitro data and clinical pharmacologic evaluation. Curr. Drug Metab. 8(5), 449–462 (2007).
16 Huang SM, Lesko LJ. Drug–drug, drug–dietary supplement, and drug–citrus fruit and other food interactions: what have we learned? J. Clin. Pharmacol. 44(6), 559–569 (2004).
17 Somogyi A, Stockley C, Keal J, Rolan P, Bochner F. Reduction of metformin renal tubular secretion by cimetidine in man. Br. J. Clin. Pharmacol. 23(5), 545–551 (1987).
18 Tsuda M, Terada T, Ueba M et al. Involvement of human multidrug and toxin extrusion 1 in the drug interaction between cimetidine and metformin in renal epithelial cells. J. Pharmacol. Exp. Ther. 329(1), 185–191 (2009).
19 Schinkel AH, Wagenaar E, van Deemter L, Mol CA, Borst P. Absence of the mdr1a P-glycoprotein in mice affects tissue distribution and pharmacokinetics of dexamethasone, digoxin, and cyclosporin A. J. Clin. Invest. 96(4), 1698–1705 (1995).
20 Ito S, Koren G, Harper PA, Silverman M. Energy-dependent transport of digoxin across renal tubular cell monolayers (LLC-PK1). Can. J. Physiol. Pharmacol. 71(1), 40–47 (1993).
21 Rameis H. Quinidine–digoxin interaction: are the pharmacokinetics of both drugs altered? Int. J. Clin. Pharmacol. Ther. Toxicol. 23(3), 145–153 (1985).
22 Pedersen KE, Christiansen BD, Klitgaard NA, Nielsen-Kudsk F. Effect of quinidine on digoxin bioavailability. Eur. J. Clin. Pharmacol. 24(1), 41–47 (1983).
23 Hedman A, Angelin B, Arvidsson A et al. Digoxin-verapamil interaction: reduction of biliary but not renal digoxin clearance in humans. Clin. Pharmacol. Ther. 49(3), 256–262 (1991).
24 Niemi M. Role of OATP transporters in the disposition of drugs. Pharmacogenomics 8(7), 787–802 (2007).
25 Lau YY, Huang Y, Frassetto L, Benet LZ. effect of OATP1B transporter inhibition on the pharmacokinetics of atorvastatin in healthy volunteers. Clin. Pharmacol. Ther. 81(2), 194–204 (2007).
26 Niemi M, Pasanen MK, Neuvonen PJ. Organic anion transporting polypeptide 1B1: a genetically polymorphic transporter of major importance for hepatic drug uptake. Pharmacol. Rev. 63(1), 157–181 (2011).
27 Bidstrup TB, Bjørnsdottir I, Sidelmann UG, Thomsen MS, Hansen KT. CYP2C8 and CYP3A4 are the principal enzymes involved in the human in vitro biotransfor-mation of the insulin secretagogue repaglinide. Br. J. Clin. Pharmacol. 56(3), 305–314 (2003).
28 Kajosaari LI, Laitila J, Neuvonen PJ, Backman JT. Metabolism of repaglinide by CYP2C8 and CYP3A4 in vitro: effect of fibrates and rifampicin. Basic Clin. Pharma-col. Toxicol. 97(4), 249–256 (2005).
29 Niemi M, Kajosaari LI, Neuvonen M, Backman JT, Neuvonen PJ. The CYP2C8 inhibitor trimethoprim increases the plasma concentrations of repaglinide in healthy subjects. Br. J. Clin. Pharmacol. 57(4), 441–447 (2004).
30 Hatorp V, Hansen KT, Thomsen MS. Influence of drugs interacting with CYP3A4 on the pharmacokinetics, pharmacodynamics, and safety of the prandial glucose regulator repaglinide. J. Clin. Pharmacol. 43(6), 649–660 (2003).
31 Niemi M, Backman JT, Neuvonen M, Neuvonen PJ. Effects of gemfibrozil, itraconazole, and their combination on the pharmacokinetics and pharmacodynamics of repaglinide: potentially hazardous interaction between gemfibrozil and repaglinide. Diabetologia 46(3), 347–351 (2003).
32 Rowland M, Matin SB. Kinetics of DDIs. J. Pharmacokinet. Pharmacodyn. 1(6), 553–567 (1973).
33 Ito K, Iwatsubo T, Kanamitsu S, Nakajima Y, Sugiyama Y. Quantitative prediction of in vivo drug clearance and drug interac-tions from in vitro data on metabolism, together with binding and transport. Annu. Rev. Pharmacol. Toxicol. 38, 461–499 (1998).
34 Rostami-Hodjegan A, Tucker GT. ‘In silico’ simulations to assess the ‘in vivo’ conse-quences of ‘in vitro’ metabolic DDIs. Drug Discov. Today Tech. 1, 441–448 (2004).
35 Hisaka A, Ohno Y, Yamamoto T, Suzuki H. Theoretical considerations on quantita-tive prediction of drug–drug interactions. Drug Metab. Pharmacokinet. 25(1), 48–61 (2010).
36 Wang YH, Jones DR, Hall SD. Prediction of cytochrome P450 3A inhibition by verapamil enantiomers and their metabo-
lites. Drug Metab. Dispos. 32(2), 259–266 (2004).
37 Galetin A, Burt H, Gibbons L, Houston JB. Prediction of time-dependent CYP3A4 drug–drug interactions: impact of enzyme degradation, parallel elimination pathways, and intestinal inhibition. Drug Metab. Dispos. 34(1), 166–175 (2006).
38 Mayhew BS, Jones DR, Hall SD. An in vitro model for predicting in vivo inhibition of cytochrome P450 3A4 by metabolic intermediate complex formation. Drug Metab. Dispos. 28(9), 1031–1037 (2000).
39 Almond LM, Yang J, Jamei M, Tucker GT, Rostami-Hodjegan A. Towards a quantita-tive framework for the prediction of DDIs arising from cytochrome P450 induction. Curr. Drug Metab. 10(4), 420–432 (2009).
40 Kato M, Chiba K, Horikawa M, Sugiyama Y. The quantitative prediction of in vivo enzyme-induction caused by drug exposure from in vitro information on human hepatocytes. Drug Metab. Pharmacokinet. 20(4), 236–243 (2005).
41 Shou M, Hayashi M, Pan Y et al. Mod-eling, prediction, and in vitro–in vivo correlation of CYP3A4 induction. Drug Metab. Dispos. 36(11), 2355–2370 (2008).
42 Ohno Y, Hisaka A, Ueno M, Suzuki H. General framework for the prediction of oral drug interactions caused by CYP3A4 induction from in vivo information. Clin. Pharmacokinet. 47(10), 669–680 (2008).
43 Yang J, Liao M, Shou M et al. Cytochrome p450 turnover: regulation of synthesis and degradation, methods for determining rates, and implications for the prediction of drug interactions. Curr. Drug Metab. 9(5), 384–394 (2008).
44 Kozawa M, Honma M, Suzuki H. Quantitative prediction of in vivo profiles of CYP3A4 induction in humans from in vitro results with a reporter gene assay. Drug Metab. Dispos. 37(6), 1234–1241 (2009).
45 Shou M. Prediction of pharmacokinetics and drug–drug interactions from in vitro metabolism data. Curr. Opin. Drug Discov. Devel. 8(1), 66–77 (2005).
46 Obach RS, Walsky RL, Venkatakrishnan K, Gaman EA, Houston JB, Tremaine LM. The utility of in vitro cytochrome P450 inhibition data in the prediction of drug–drug interactions. J. Pharmacol. Exp. Ther. 316(1), 336–348 (2006).
47 Grime K, Riley RJ. The impact of in vitro binding on in vitro–in vivo extrapolations, projections of metabolic clearance and
Yeo, Jamei & Rostami-Hodjegan
155www.expert-reviews.com
Review
clinical drug–drug interactions. Curr. Drug Metab. 7(3), 251–264 (2006).
48 Hyland R, Dickins M, Collins C, Jones H, Jones B. Maraviroc: in vitro assessment of drug–drug interaction potential. Br. J. Clin. Pharmacol. 66(4), 498–507 (2008).
49 Xu L, Chen Y, Pan Y, Skiles GL, Shou M. Prediction of human drug–drug interac-tions from time-dependent inactivation of CYP3A4 in primary hepatocytes using a population-based simulator. Drug Metab. Dispos. 37(12), 2330–2339 (2009).
50 Youdim KA, Zayed A, Dickins M et al. Application of CYP3A4 in vitro data to predict clinical drug–drug interactions; predictions of compounds as objects of interaction. Br. J. Clin. Pharmacol. 65(5), 680–692 (2008).
51 Fahmi OA, Hurst S, Plowchalk D et al. Comparison of different algorithms for predicting clinical drug–drug interactions, based on the use of CYP3A4 in vitro data: predictions of compounds as precipitants of interaction. Drug Metab. Dispos. 37(8), 1658–1666 (2009).
52 Gandelman K, Zhu T, Fahmi OA et al. Unexpected effect of rifampin on the pharmacokinetics of linezolid: in silico and in vitro approaches to explain its mecha-nism. J. Clin. Pharmacol. 51(2), 229–236 (2011).
53 Zhang L, Zhang YD, Strong JM, Reynolds KS, Huang SM. A regulatory viewpoint on transporter-based drug interactions. Xenobiotica 38(7–8), 709–724 (2008).
54 Agarwal S, Arya V, Zhang L. Review of P-gp inhibition data in recently approved new drug applications: utility of the proposed [I1]/IC50 and [I2]/IC50 criteria in the P-gp decision tree. J. Clin. Pharma-col. doi:10.1177/0091270011436344 (2012) (Epub ahead of print).
55 Fenner KS, Troutman MD, Kempshall S et al. Drug–drug interactions mediated through P-glycoprotein: clinical relevance and in vitro–in vivo correlation using digoxin as a probe drug. Clin. Pharmacol. Ther. 85(2), 173–181 (2009).
56 Lee CA. Application of receiver operating characteristics to refine the prediction of potential digoxin drug interactions. Drug Metab. Dispos. (2012) (In Press).
57 Shitara Y, Sato H, Sugiyama Y. Evaluation of drug–drug interaction in the hepatobil-iary and renal transport of drugs. Annu. Rev. Pharmacol. Toxicol. 45, 689–723 (2005).
58 Hinton LK, Galetin A, Houston JB. Multiple inhibition mechanisms and
prediction of drug–drug interactions: status of metabolism and transporter models as exemplified by gemfibrozil-drug interactions. Pharm. Res. 25(5), 1063–1074 (2008).
59 Karlgren M, Ahlin G, Bergström CA, Svensson R, Palm J, Artursson P. In vitro and in silico strategies to identify OATP1B1 inhibitors and predict clinical drug–drug interactions. Pharm. Res. 29(2), 411–426 (2012).
60 Yoshida K, Maeda K, Sugiyama Y. Transporter-mediated drug–drug interactions involving OATP substrates: predictions based on in vitro inhibition studies. Clin. Pharmacol. Ther. 91(6), 1053–1064 (2012).
61 Einolf HJ. Comparison of different approaches to predict metabolic drug–drug interactions. Xenobiotica. 37(10-11), 1257–1294 (2007).
62 Baneyx G, Fukushima Y, Parrott N. Use of physiologically based pharmacokinetic modeling for assessment of drug–drug interactions. Future Med. Chem. 4(5), 681–693 (2012).
63 Rane A, Wilkinson GR, Shand DG. Prediction of hepatic extraction ratio from in vitro measurement of intrinsic clearance. J. Pharmacol. Exp. Ther. 200(2), 420–424 (1977).
64 Wilkinson GR. Clearance approaches in pharmacology. Pharmacol. Rev. 39(1), 1–47 (1987).
65 Houston JB. Utility of in vitro drug metabolism data in predicting in vivo metabolic clearance. Biochem. Pharmacol. 47(9), 1469–1479 (1994).
66 Barter ZE, Bayliss MK, Beaune PH et al. Scaling factors for the extrapolation of in vivo metabolic drug clearance from in vitro data: reaching a consensus on values of human microsomal protein and hepatocellularity per gram of liver. Curr. Drug Metab. 8(1), 33–45 (2007).
67 Proctor NJ, Tucker GT, Rostami-Hodjegan A. Predicting drug clearance from recombinantly expressed CYPs: intersystem extrapolation factors. Xenobiotica 34(2), 151–178 (2004).
68 Iwatsubo T, Suzuki H, Sugiyama Y. Prediction of species differences (rats, dogs, humans) in the in vivo metabolic clearance of YM796 by the liver from in vitro data. J. Pharmacol. Exp. Ther. 283(2), 462–469 (1997).
69 Obach RS. Prediction of human clearance of twenty-nine drugs from hepatic microsomal intrinsic clearance data: an
examination of in vitro half-life approach and nonspecific binding to microsomes. Drug Metab. Dispos. 27(11), 1350–1359 (1999).
70 Riley RJ, McGinnity DF, Austin RP. A unified model for predicting human hepatic, metabolic clearance from in vitro intrinsic clearance data in hepatocytes and microsomes. Drug Metab. Dispos. 33(9), 1304–1311 (2005).
71 Howgate EM, Rowland Yeo K, Proctor NJ, Tucker GT, Rostami-Hodjegan A. Prediction of in vivo drug clearance from in vitro data. I: impact of inter-individual variability. Xenobiotica 36(6), 473–497 (2006).
72 Johnson TN, Tucker GT, Tanner MS, Rostami-Hodjegan A. Changes in liver volume from birth to adulthood: a meta-analysis. Liver Transpl. 11(12), 1481–1493 (2005).
73 Dickinson GL, Rezaee S, Proctor NJ, Lennard MS, Tucker GT, Rostami-Hod-jegan A. Incorporating in vitro information on drug metabolism into clinical trial simulations to assess the effect of CYP2D6 polymorphism on pharmacokinetics and pharmacodynamics: dextromethorphan as a model application. J. Clin. Pharmacol. 47(2), 175–186 (2007).
74 Johnson TN, Rostami-Hodjegan A, Tucker GT. Prediction of the clearance of eleven drugs and associated variability in neonates, infants and children. Clin. Pharmacokinet. 45(9), 931–956 (2006).
75 Inoue S, Howgate EM, Rowland-Yeo K et al. Prediction of in vivo drug clearance from in vitro data. II: potential inter-ethnic differences. Xenobiotica 36(6), 499–513 (2006).
76 Chetty M, Mattison D, Rostami-Hodjegan A. Sex differences in the clearance of CYP3A4 substrates: exploring possible reasons for the substrate dependency and lack of consensus. Curr. Drug Metab. 13(6), 778–786 (2012).
77 Abduljalil K, Furness P, Johnson TN, Rostami-Hodjegan A, Soltani H. Anatomical, physiological and metabolic changes with gestational age during normal pregnancy: a database for parameters required in physiologically based pharma-cokinetic modelling. Clin. Pharmacokinet. 51(6), 365–396 (2012).
78 Ghobadi C, Johnson TN, Aarabi M et al. Application of a systems approach to the bottom-up assessment of pharmacokinetics in obese patients: expected variations in clearance. Clin. Pharmacokinet. 50(12), 809–822 (2011).
Predicting drug–drug interactions
Expert Rev. Clin. Pharmacol. 6(2), (2013)156
Review
79 Johnson TN, Boussery K, Rowland-Yeo K, Tucker GT, Rostami-Hodjegan A. A semi- mechanistic model to predict the effects of liver cirrhosis on drug clearance. Clin. Pharmacokinet. 49(3), 189–206 (2010).
80 Edginton AN, Willmann S. Physiology-based simulations of a pathological condition: prediction of pharmacokinetics in patients with liver cirrhosis. Clin. Pharmacokinet. 47(11), 743–752 (2008).
81 Rowland Yeo K, Aarabi M, Jamei M, Rostami-Hodjegan A. Modeling and predicting drug pharmacokinetics in patients with renal impairment. Expert Rev. Clin. Pharmacol. 4(2), 261–274 (2011).
82 Zhao P, Vieira Mde L, Grillo JA et al. Evaluation of exposure change of nonrenally eliminated drugs in patients with chronic kidney disease using physiologically based pharmacokinetic modeling and simulation. J. Clin. Pharmacol. 52(Suppl. 1), 91S–108S (2012).
83 Plowchalk DR, Rowland Yeo K. Prediction of drug clearance in a smoking population: modeling the impact of variable cigarette consumption on the induction of CYP1A2. Eur. J. Clin. Pharmacol. 68(6), 951–960 (2012).
84 McNamara PJ, Alcorn J. Protein binding predictions in infants. AAPS PharmSci. 4(1), E4 (2002).
85 Willmann S, Höhn K, Edginton A et al. Development of a physiology-based whole-body population model for assessing the influence of individual variability on the pharmacokinetics of drugs. J. Pharmacokinet. Pharmacodyn. 34(3), 401–431 (2007).
86 Cubitt HE, Yeo KR, Howgate EM, Rostami-Hodjegan A, Barter ZE. Sources of interindividual variability in IVIVE of clearance: an investigation into the prediction of benzodiazepine clearance using a mechanistic population-based pharmacokinetic model. Xenobiotica 41(8), 623–638 (2011).
87 Rowland Yeo K, Jamei M, Yang J, Tucker GT, Rostami-Hodjegan A. Physiologically based mechanistic modelling to predict complex drug–drug interactions involving simultaneous competitive and time-de-pendent enzyme inhibition by parent compound and its metabolite in both liver and gut – the effect of diltiazem on the time-course of exposure to triazolam. Eur. J. Pharm. Sci. 39(5), 298–309 (2010).
88 Wang YH. Confidence assessment of the Simcyp time-based approach and a static mathematical model in predicting clinical drug–drug interactions for mechanism-
based CYP3A inhibitors. Drug Metab. Dispos. 38(7), 1094–1104 (2010).
89 Fahmi OA, Maurer TS, Kish M, Cardenas E, Boldt S, Nettleton D. A combined model for predicting CYP3A4 clinical net drug–drug interaction based on CYP3A4 inhibition, inactivation, and induction determined in vitro. Drug Metab. Dispos. 36(8), 1698–1708 (2008).
90 Vossen M, Sevestre M, Niederalt C, Jang IJ, Willmann S, Edginton AN. Dynami-cally simulating the interaction of midazolam and the CYP3A4 inhibitor itraconazole using individual coupled whole-body physiologically-based pharmacokinetic (WB-PBPK) models. Theor. Biol. Med. Model. 4, 13 (2007).
91 Zhang X, Quinney SK, Gorski JC, Jones DR, Hall SD. Semiphysiologically based pharmacokinetic models for the inhibition of midazolam clearance by diltiazem and its major metabolite. Drug Metab. Dispos. 37(8), 1587–1597 (2009).
92 Yang J, Kjellsson M, Rostami-Hodjegan A, Tucker GT. The effects of dose staggering on metabolic drug–drug interactions. Eur. J. Pharm. Sci. 20(2), 223–232 (2003).
93 Zhao P, Ragueneau-Majlessi I, Zhang L et al. Quantitative evaluation of pharma-cokinetic inhibition of CYP3A substrates by ketoconazole: a simulation study. J. Clin. Pharmacol. 49(3), 351–359 (2009).
94 Nestorov I. Whole-body physiologically based pharmacokinetic models. Expert Opin. Drug Metab. Toxicol. 3(2), 235–249 (2007).
95 De Buck SS, Mackie CE. Physiologically based approaches towards the prediction of pharmacokinetics: in vitro–in vivo extrapolation. Expert Opin. Drug Metab. Toxicol. 3(6), 865–878 (2007).
96 Bouzom F, Walther B. Pharmacokinetic predictions in children by using the physiologically based pharmacokinetic modelling. Fundam. Clin. Pharmacol. 22(6), 579–587 (2008).
97 Espié P, Tytgat D, Sargentini-Maier ML, Poggesi I, Watelet JB. Physiologically based pharmacokinetics (PBPK). Drug Metab. Rev. 41(3), 391–407 (2009).
98 Perdaems N, Blasco H, Vinson C et al. Predictions of metabolic drug–drug interactions using physiologically based modelling: two cytochrome P450 3A4 substrates coadministered with ketocona-zole or verapamil. Clin. Pharmacokinet. 49(4), 239–258 (2010).
99 Bouzom F, Ball K, Perdaems N, Walther B. Physiologically based pharmacokinetic
(PBPK) modelling tools: how to fit with our needs? Biopharm. Drug Dispos. 33(2), 55–71 (2012).
100 Grillo JA, Zhao P, Bullock J et al. Utility of a physiologically-based pharmacokinetic (PBPK) modeling approach to quantita-tively predict a complex drug–drug-disease interaction scenario for rivaroxaban during the drug review process: implications for clinical practice. Biopharm. Drug Dispos. 33(2), 99–110 (2012).
101 Shaffer CL, Scialis RJ, Rong H, Obach RS. Using Simcyp to project human oral pharmacokinetic variability in early drug research to mitigate mechanism-based adverse events. Biopharm. Drug Dispos. 33(2), 72–84 (2012).
102 Chen Y, Jin JY, Mukadam S, Malhi V, Kenny JR. Application of IVIVE and PBPK modeling in prospective prediction of clinical pharmacokinetics: strategy and approach during the drug discovery phase with four case studies. Biopharm. Drug Dispos. 33(2), 85–98 (2012).
103 Sinha VK, Snoeys J, Osselaer NV, Peer AV, Mackie C, Heald D. From preclinical to human – prediction of oral absorption and drug–drug interaction potential using physiologically based pharmacokinetic (PBPK) modeling approach in an industrial setting: a workflow by using case example. Biopharm. Drug Dispos. 33(2), 111–121 (2012).
104 Fan J, Chen S, Chow EC, Pang KS. PBPK modeling of intestinal and liver enzymes and transporters in drug absorption and sequential metabolism. Curr. Drug Metab. 11(9), 743–761 (2010).
105 Watanabe T, Kusuhara H, Sugiyama Y. Application of physiologically based pharmacokinetic modeling and clearance concept to drugs showing transporter-mediated distribution and clearance in humans. J. Pharmacokinet. Pharmacodyn. 37(6), 575–590 (2010).
106 Darwich AS, Neuhoff S, Jamei M, Rostami-Hodjegan A. Interplay of
Yeo, Jamei & Rostami-Hodjegan
157www.expert-reviews.com
Review
metabolism and transport in determining oral drug absorption and gut wall metabolism: a simulation assessment using the ‘Advanced Dissolution, Absorption, Metabolism (ADAM)’ model. Curr. Drug Metab. 11(9), 716–729 (2010).
107 Jones HM, Barton HA, Lai Y et al. Mechanistic pharmacokinetic modeling for the prediction of transporter-mediated disposition in humans from sandwich culture human hepatocyte data. Drug Metab. Dispos. 40(5), 1007–1017 (2012).
108 Harwood MD, Neuhoff S, Carlson GL, Warhurst G, Rostami-Hodjegan A. Absolute abundance and function of intestinal drug transporters: a prerequisite for fully mechanistic in vitro–in vivo extrapolation of oral drug absorption. Biopharm. Drug Dispos. doi:10.1002/bdd.1810 (2012) (Epub ahead of print).
109 Varma MV, Lai Y, Feng B, Litchfield J, Goosen TC, Bergman A. Physiologically based modeling of pravastatin transporter-mediated hepatobiliary disposition and drug–drug interactions. Pharm. Res. 29(10), 2860–2873 (2012).
110 Ohtsuki S, Schaefer O, Kawakami H et al. Simultaneous absolute protein quantifica-tion of transporters, cytochromes P450, and UDP-glucuronosyltransferases as a novel approach for the characterization of individual human liver: comparison with mRNA levels and activities. Drug Metab. Dispos. 40(1), 83–92 (2012).
114 Davidian M, Gallant AR. Smooth nonparametric maximum likelihood estimation for population pharmacokinet-ics, with application to quinidine. J. Pharmacokinet. Biopharm. 20(5), 529–556 (1992).
115 Pillai GC, Mentré F, Steimer JL. Non-linear mixed effects modeling – from methodology and software development to driving implementation in drug develop-ment science. J. Pharmacokinet. Pharmacodyn. 32(2), 161–183 (2005).
116 Wakefield J, Bennett J. The bayesian modeling of covariates for population pharmacokinetic models. J. Am.Stat. Assoc. 91, 917–927 (1996).
117 Gelman A, Rubin DB. Markov chain Monte Carlo methods in biostatistics. Stat. Methods Med. Res. 5(4), 339–355 (1996).
118 Idkaidek N, Arafat T. Effect of micrograv-ity on the pharmacokinetics of Ibuprofen in humans. J. Clin. Pharmacol. 51(12), 1685–1689 (2011).
119 Vieira ML, Zhao P, Berglund EG et al. Predicting drug interaction potential with a physiologically based pharmacokinetic model: a case study of telithromycin, a time-dependent CYP3A inhibitor. Clin. Pharmacol. Ther. 91(4), 700–708 (2012).
120 Brøsen K, Hansen JG, Nielsen KK, Sindrup SH, Gram LF. Inhibition by paroxetine of desipramine metabolism in extensive but not in poor metabolizers of sparteine. Eur. J. Clin. Pharmacol. 44(4), 349–355 (1993).
121 Alderman J, Preskorn SH, Greenblatt DJ et al. Desipramine pharmacokinetics when coadministered with paroxetine or sertraline in extensive metabolizers. J. Clin. Psychopharmacol. 17(4), 284–291 (1997).
122 Hemeryck A, Lefebvre RA, De Vriendt C, Belpaire FM. Paroxetine affects metoprolol pharmacokinetics and pharmacodynamics in healthy volunteers. Clin. Pharmacol. Ther. 67(3), 283–291 (2000).
123 Belle DJ, Ernest CS, Sauer JM, Smith BP, Thomasson HR, Witcher JW. Effect of potent CYP2D6 inhibition by paroxetine on atomoxetine pharmacokinetics. J. Clin. Pharmacol. 42(11), 1219–1227 (2002).
124 Kaye CM, Haddock RE, Langley PF et al. A review of the metabolism and pharma-cokinetics of paroxetine in man. Acta Psychiatr. Scand. Suppl. 350, 60–75 (1989).
125 Sindrup SH, Brøsen K, Gram LF et al. The relationship between paroxetine and the sparteine oxidation polymorphism. Clin. Pharmacol. Ther. 51(3), 278–287 (1992).
126 Bertelsen KM, Venkatakrishnan K, Von Moltke LL, Obach RS, Greenblatt DJ. Apparent mechanism-based inhibition of human CYP2D6 in vitro by paroxetine: comparison with fluoxetine and quinidine. Drug Metab. Dispos. 31(3), 289–293 (2003).
127 Jornil J, Jensen KG, Larsen F, Linnet K. Identification of cytochrome P450 isoforms involved in the metabolism of
paroxetine and estimation of their importance for human paroxetine metabolism using a population-based simulator. Drug Metab. Dispos. 38(3), 376–385 (2010).
128 Venkatakrishnan K, Obach RS. In vitro–in vivo extrapolation of CYP2D6 inactivation by paroxetine: prediction of nonstationary pharmacokinetics and drug interaction magnitude. Drug Metab. Dispos. 33(6), 845–852 (2005).
129 Yasui-Furukori N, Saito M, Inoue Y et al. Terbinafine increases the plasma concentra-tion of paroxetine after a single oral administration of paroxetine in healthy subjects. Eur. J. Clin. Pharmacol. 63(1), 51–56 (2007).
130 Dornhorst A. Insulinotropic meglitinide analogues. Lancet 358(9294), 1709–1716 (2001).
131 Niemi M, Backman JT, Neuvonen M, Neuvonen PJ, Kivistö KT. Rifampin decreases the plasma concentrations and effects of repaglinide. Clin. Pharmacol. Ther. 68(5), 495–500 (2000).
132 Kajosaari LI, Niemi M, Neuvonen M, Laitila J, Neuvonen PJ, Backman JT. Cyclosporine markedly raises the plasma concentrations of repaglinide. Clin. Pharmacol. Ther. 78(4), 388–399 (2005).
133 Rostami-Hodjegan A, Tamai I, Pang KS. Physiologically based pharmacokinetic (PBPK) modeling: it is here to stay! Biopharm. Drug Dispos. 33(2), 47–50 (2012).
Websites
201 US FDA. Guidance for industry drug inter-action studies: study design, data analysis and implications for dosing and labeling. www.fda.gov/downloads/Drugs/ GuidanceComplianceRegulatoryInforma-tion/ Guidances/UCM292362.pdf
202 EMA. Guideline on the Investigation of Drug Interactions. www.ema.europa.eu/docs/en_gb/document_ library/scientific_guide-line/2010/05/ WC500090112.pdf
203 Simcyp Ltd. The population-based simulator. www.simcyp.com
204 Bayer Technology Services. PK-Sim software for WB-PBPK modelling and simulation. www.systems-biology.com/products/pk-sim. html
205 Simulations Plus, Inc. GastroPlus. www.simulations-plus.com
