Abstract
Physiologically based pharmacokinetic modeling and simulation can be used to predict the pharmacokinetics of drugs in human populations and to explore the effects of varying physiologic parameters that result from aging, ethnicity, or disease. In addition, the effects of concomitant medications on drug exposure can be investigated; prediction of the magnitude of drug interactions can impact regulatory communications or internal decision-making regarding the requirement for a clinical drug interaction study. Modeling and simulation can also help to inform the design and timings of clinical drug interaction studies, resulting in more efficient use of limited resources and improved planning in addition to promoting mechanistic understanding of observed drug interactions. These approaches have been used in GlaxoSmithKline from drug discovery to registration and have been applied to 41 drugs from a number of therapeutic areas. This report highlights the variety of questions that can be addressed by prospective or retrospective application of modeling and simulation and the impact this can have on clinical drug development (from candidate selection through clinical development to regulatory submissions).
Introduction
Although the concepts of physiologically based pharmacokinetics (PBPK) date back many years (Teorell, 1937; Benowitz et al., 1974; Harrison and Gibaldi, 1977), the use of PBPK modeling as a tool to aid drug discovery and development has increased in recent years, as illustrated by many reports and reviews in scientific literature (Poulin and Theil, 2002; Theil et al., 2003; Nestorov, 2007; Edginton et al., 2008; Rowland et al., 2011; Jones et al., 2012). This may be attributed to the evolution of the simpler PBPK models (e.g., those involving blood flow through perfusion-limited tissues) to include in vitro–in vivo extrapolation and active uptake and efflux into permeability-limited tissues (Rostami-Hodjegan and Tucker, 2007; Watanabe et al., 2009; Rostami-Hodjegan, 2012). The utility of such modeling is the translation of in vitro and/or preclinical data for a new chemical entity (NCE) into clinical pharmacokinetic predictions (De Buck and Mackie, 2007; Espie et al., 2009), thereby providing early indications of factors that may affect clinical development such as susceptibility to drug-drug interactions (DDI) and potential effects of genetic polymorphisms or disease-states. There are multiple reports that testify to the accuracy of simple arithmetic (static) approaches for predicting the magnitude of DDIs, particularly those mediated by cytochrome P450 (Obach et al., 2006, 2007; Fahmi et al., 2009; Shardlow et al., 2011); however, there are additional advantages to using dynamic PBPK modeling, allowing the prediction of drug exposure profiles (in blood and organ compartments) and the time-based effects of concomitant medications on those profiles (Einolf, 2007; Vossen et al., 2007; Kato et al., 2008; Fahmi et al., 2009; Fenneteau et al., 2010).
The commercially available PBPK software packages, such as simCYP (SimCYP Ltd., Sheffield, United Kingdom) and Gastroplus (Simulations Plus, Inc., Lancaster, CA), feature complex clinical trial design options, including multiple routes of drug administration; some also incorporate population variability, physiologic changes associated with various disease states, and the ability to link tissue concentrations to pharmacodynamic effects (Jamei et al., 2009). These capabilities are extremely useful for evaluating the potential clinical impact before undertaking extensive clinical trials and can inform both the necessity and the design of such trials.
Importantly, the use of PBPK modeling has gained increasing recognition by the regulatory agencies, as highlighted by Huang and others, illustrating the application of modeling and simulation during regulatory review (Zhao et al., 2011, 2012a,b; Huang et al., 2012; Leong et al., 2012). Discussion of modeling and simulation approaches in the recent drug interaction guidance updates from both the European Medicines Agency (EMA) (CHMP, 2012) and the U.S. Food and Drug Administration (FDA) (CDER, 2012) also demonstrates regulatory acceptance of the use of PBPK modeling to support drug registration.
In addition to supporting drug registration, we have applied these modeling approaches to impact the decision making in clinical drug development and to provide mechanistic understanding of clinical observations. In this report, we have compiled specific examples of the application of PBPK from a variety of therapeutic areas within GlaxoSmithKline (GSK).
Materials and Methods
Examples of PBPK modeling within GSK over the past 5 years were collated and interrogated. One of the most widely used commercially available software packages, as seen in these collated examples, was simCYP, a population-based pharmacokinetic simulator. This software was often selected in preference to other packages because of its capability to simulate clinical drug interaction trial designs and its population-specific information. Consequently, the examples presented and discussed in this article are those specifically involving the use of simCYP.
Although all the simCYP modeling examples are briefly described (Tables 2, 3, and 4), five of these were selected for more detailed discussion as case studies. The case studies describe various applications of simCYP versions 6 to 11 (inclusive). Unless stated otherwise, the simulations used the default simCYP system-specific properties (e.g., organ mass, blood flow) for the appropriate human population and the default drug-specific parameters for the simCYP library compounds (where used). These default values were as defined in the software version used, and the simulations have not been repeated using updated parameters that may have been incorporated into later versions. The physicochemical properties of the GSK case study compounds in combination with several key parameters measured in vitro or in vivo were input by the user and are presented in Table 1. Further details of individual simulations are presented with each case history (see the Results section).
For the in vitro parameters shown in Table 1, reaction phenotyping experiments were performed with human liver microsomes in the presence and absence of selective cytochrome P450 inhibitors and with single cDNA expressed human cytochrome P450 (CYP) enzymes to determine the relative contribution of each enzyme to the total oxidative metabolism using appropriate scaling factors (de Waziers et al., 1990; Rodrigues, 1999). CYP inhibition experiments (IC50 and KI/kinact) were performed as described previously elsewhere (Shardlow et al., 2011; Polli et al., 2013). CYP induction experiments were conducted in primary human hepatocyte culture as described by Bowen et al. (2000). The CYP3A4 induction potential was determined by measuring the changes in mRNA expression levels with EC50 and maximal induction (Emax) calculated by nonlinear regression with a modified Hill equation. In vitro permeability was determined at pH 7.4 using Madin-Darby canine kidney epithelial cells transfected with the human MDR1 gene (MDCKII-MDR1) cells. The test compounds were incubated in the presence and absence of the potent P-glycoprotein inhibitor GF120918 [N-4-[2-(1,2,3,4-tetrahydro-6,7-dimethoxy-2-isoquinolinyl)ethyl]-phenyl)-9,10-dihydro-5-methoxy-9-oxo-4-acridine carboxamide hydrochloride], and the passive membrane permeability was measured from the apical to the basolateral side (Mahar Doan et al., 2002).
The clinical data included in this paper was obtained from GSK-sponsored studies; all protocols were reviewed and approved by the appropriate ethics review committees, and written informed consent was obtained from all subjects before any protocol-specific procedures.
Results
We identified 41 examples of key uses of PBPK modeling (using simCYP) at GSK over the past 5 years. Upon detailed examination, these examples appeared to fall naturally into three major categories: those that have been used to inform regulatory communications (Table 2), those that have impacted clinical development decisions (Table 3), and those where PBPK modeling has been used to the aid the mechanistic understanding of clinical observations (Table 4). Selected examples from each category are presented in more detail in the following case studies, which represent both retrospective and prospective analyses with corroborative clinical data available in case studies 2, 4, and 5.
Case Study 1
Question.
Does PBPK modeling predict changes in the pharmacokinetics of GSK1 in patients with renal or hepatic impairment? The purpose of this simulation was to inform any dose-adjustment required and the initial clinical study designs in these patient populations.
Background.
The elimination of GSK1 in humans was primarily by hepatic metabolism, with only 5% of an oral dose being excreted as unchanged drug. In vitro data showed CYP3A4-mediated metabolism accounted for approximately 70% of total elimination with minor contributions from CYP1A2 (∼12%), CYP2C8 (∼4%), CYP2D6 (∼6%), and CYP2C9 (∼3%).
Simulations.
SimCYP (version 11) was used initially to check that simulations with a healthy volunteer virtual population corresponded with the observed pharmacokinetic profiles in a phase I clinical study. Input parameters for GSK1 are shown in Table 1. After successful reproduction of the healthy volunteer study (data not shown), further simulations were performed prospectively to predict the effect of mild, moderate, and severe hepatic impairment (Child-Pugh classification) and moderate and severe renal impairment on the exposure of GSK1. Use of simCYP to simulate the potential exposure of individuals with hepatic and renal impairment has been reported previously (Rowland Yeo et al., 2011a; Grillo et al., 2012; Zhao et al., 2012b).
Results from the simulations (Fig. 1) with subjects with mild, moderate, and severe hepatic impairment showed increases in Cmax of 1.9- to 4.4-fold compared with healthy subjects. Increases in the area under the curve (AUC) ranged from 1.9- to 4.6-fold. Based on these results and in accordance with the EMA (CHMP, 2005) and FDA (CDER, 2003) guidance for assessing the pharmacokinetics in patients with impaired hepatic function, a clinical study comparing the GSK1 plasma levels after administration of a single dose of GSK1 with those in subjects with normal hepatic function and with moderate hepatic function was proposed as an initial study design to investigate any changes in pharmacokinetics.
Simulations were also performed in virtual subjects from populations suffering from renal impairment—moderate (glomerular filtration rate [GFR] between 30 and 60 ml/min) and severe (GFR <30 ml/min). Simulations for subjects with severe renal impairment showed a maximum 1.5-fold (median data) higher exposure of GSK1 compared with healthy subjects (Fig. 1). Minimal impact on exposure was predicted in patients suffering from moderate renal impairment (data not shown). A clinical study to determine the effect of renal impairment on the pharmacokinetics of GSK1 was therefore recommended using a reduced study design in accordance with EMA (CHMP, 2004) and FDA (CDER, 2010) guidance on evaluating pharmacokinetics in renally impaired subjects. The pharmacokinetics of GSK1 would be assessed after a single dose of GSK1 administered to subjects with severe renal impairment and subjects with normal renal function. Subjects with mild and moderate renal impairment would only be studied if the data from the severe impairment groups indicated a clinically significant change in pharmacokinetics.
Impact.
The abbreviated study designs were presented to the FDA and EMA, and the reviewers agreed with the proposed clinical study designs to investigate the impact of renal and hepatic impairment on the exposure of GSK1. Without this PBPK modeling data, full clinical study designs would be required.
Case Study 2
Question.
Can PBPK modeling inform the optimal ketoconazole drug interaction study design for a drug with an extended half-life? The purpose of this case study was to explain clinical observations from a prior DDI study, to optimize the trial design for a second clinical study, and to extrapolate from the latter study the maximal DDI magnitude.
Background.
GSK5 was a confirmed CYP3A4 substrate in vitro with little contribution to metabolism from any other enzyme tested. A human radiolabeled metabolism study showed a significant contribution to drug elimination via oxidative metabolites; therefore, there was an acknowledged clinical risk of DDIs on coadministration with CYP3A4 inhibitors. A clinical DDI study with ketoconazole had been performed (before routine application of static or PBPK modeling approaches) using a standard protocol design. In the first phase, healthy volunteers received a single oral 20-mg dose of GSK5 followed by a 21-day washout period. In the second phase, the same volunteers received 400 mg of ketoconazole daily for 8 days with a single 20-mg dose of GSK5 coadministered on day 5 only. The pharmacokinetics of GSK5 was measured for 14 days postdose for both phases of the study. The change in GSK5 exposure observed after coadministration of ketoconazole [area under the curve/area under the curve (inhibited) (AUCi/AUC) = 3.3] was lower than expected given the extent of metabolism via CYP3A4, and this was hypothesized to be due to the clinical study design. GSK5 has low clearance and a long half-life (∼50 hours), and the clinical DDI study had potentially underestimated the maximal effect of ketoconazole; therefore, simulations were performed to explore and optimize the trial design for a second study.
Simulations.
In the first instance, simCYP version 7 was used retrospectively to simulate the original study design; key input parameters for the simulations are shown in Table 1. It was clear from the simulation results that the ketoconazole dosing regimen employed in the first clinical study would underpredict the maximal extent of inhibition (Fig. 2). Based on the successful retrospective prediction (simulated AUCi/AUC = 3.5 versus observed AUCi/AUC = 3.3), we then went on to simulate alternative dosing regimens, such as increasing the number of days of codosing with ketoconazole to 10 and 70 days. These simulations predicted AUCi/AUC ratios of 5 and 10.7 for 10 and 70 days, respectively, with a predicted steady-state AUC ratio of >20.
Thus, a second clinical study was performed in which ketoconazole (400 mg every day) was administered together with a single dose of GSK5 and for 9 consecutive days afterward. This second study achieved the predicted magnitude of interaction for 10 days dosing of ketoconazole (observed AUCi/AUC ratio 5) and consequently supported the simulations that had suggested that the maximal extent of inhibition would not be achieved without prolonged dosing of ketoconazole, potentially beyond 70 days—a trial design that would be impractical to perform in the clinic.
Impact.
PBPK simulations enabled an understanding of observed clinical drug interaction data, informed an improved study design, and provided an estimation of the maximal drug interaction. As GSK5 was identified as having an increased risk as a victim of DDIs, exclusion of CYP3A4 inhibitors was recommended in future clinical trials.
Case Study 3
Question.
Can ritonavir-boosting allow once-a-day dosing for a compound that exhibited a short half-life in the first time in human (FTIH) clinical study? The aim of this simulation study was to predict whether ritonavir boosting would be a viable strategy for increasing the half-life of GSK6 in future clinical studies.
Background.
GSK6 exhibited a mean half-life in the FTIH study of 5.3 hours, which resulted in plasma concentrations remaining above the minimum effective concentration (MEC) for only 5–7 hours after a single oral dose. Because GSK6 pharmacokinetics did not allow for the desired product profile of once-a-day oral dosing, PBPK modeling was used to investigate the feasibility of maintaining GSK6 plasma concentrations above the MEC for the entire dosing interval after every-day dosing in the presence of 100 mg of ritonavir.
Simulation.
The pharmacokinetics of GSK6 were simulated using simCYP version 10, with the assumptions that the overall systemic clearance of GSK6 was 100% dependent on CYP enzymes and the percentage contribution to systemic clearance by CYP3A4 and CYP2D6 was approximately 99% and 1%, respectively, based on in vitro reaction phenotyping data. Input parameters for GSK6 are shown in Table 1. The coefficient of variation on CYP3A4 abundance in the simCYP Healthy Volunteers population library file was decreased from the default value of 95% to 24% to approximate the variability in pharmacokinetics observed after a single oral dose of 70 mg of GSK6 in humans. Absorption and elimination processes were assumed to be linear. SimCYP was then used to simulate a drug interaction trial involving 10 trials of 10 subjects per trial, with GSK6 coadministered with 100 mg of ritonavir once daily for 3 days.
The model for ritonavir used the default parameters provided by simCYP, with the exception of the CYP3A4 induction parameters of Emax and EC50, which were added in accordance with the values published by McGinnity et al. (2009). The modified simCYP ritonavir model was validated against clinical results from another GSK compound for which ritonavir boosting has been explored clinically (data not shown).
The predicted outcome of this clinical trial design for GSK6 is illustrated in Fig. 3. Although ritonavir was shown to boost the exposure of GSK6, the boosted exposures exceeded the MEC in only approximately 10%–50% of the population over a simulated dose range of 70–420 mg of GSK6.
Impact.
The predicted ritonavir boosting effect was not significant enough to allow for once-a-day dosing without a significant escalation in the dose of GSK6. The project team opted to investigate twice-a-day dosing in a subsequent clinical trial.
Case Study 4
Question.
Can changing the route of administration of GSK19 mitigate the DDI risk observed with oral GSK19 and midazolam? The aim of this simulation was to inform comedication exclusion criteria.
Background.
In a clinical DDI study, oral dosing of GSK19 with oral midazolam increased midazolam AUC by approximately 3-fold (2.73–3.7, 90% confidence limits) on day 1. When GSK19 was reevaluated for a new indication and new route of administration, we used modeling and simulation to investigate whether changing the route of administration of GSK19 could mitigate the DDI risk observed with oral dosing. It has been shown that the route of administration of a DDI perpetrator can have a significant effect on the magnitude of DDI, as seen with fluconazole (Ahonen at al., 1997); however, unlike fluconazole, which is a moderate and direct inhibitor of CYP3A4, GSK19 showed metabolism-dependent inhibition of CYP3A4 in vitro and thus had the potential to be an inactivator of CYP3A4 in vivo.
Simulations.
We used simCYP version 10 retrospectively to simulate the observed oral and intravenous PK profiles of GSK19 and observed DDI with midazolam, and to prospectively evaluate the perpetrator DDI risk with a single intravenous administration of GSK19 (Fig. 4). The model input parameters are shown in Table 1. The model performance was tested against clinically observed midazolam and GSK19 exposure after both oral and intravenous dosing. The DDI trials were simulated using the midazolam file in simCYP, with minor modification to tissue partition coefficients based on Fenneteau et al. (2010) as part of the model calibration to optimize midazolam pharmacokinetics. A mean midazolam drug interaction of 2.3-fold (range: 1.2–5.2) increase in AUC was retrospectively predicted with oral GSK19 (Fig. 4A).
Simulations to investigate the effect of single intravenous doses of GSK19 on the AUC of 5 mg of oral midazolam showed a dose-dependent increase in midazolam AUC (Fig. 4A), with a maximum individual increase of about 4.8-fold for the highest simulated dose (24 mg). This modeling result is consistent with previous findings (Chen et al., 2013) where the reduced risk for a highly permeable perpetrator (such as GSK19) administered IV is a result of bypassing the impact on gut metabolism of an orally administered victim. Simulation of the recovery of enzyme activity indicated that at a dose range of 12–24 mg, the hepatic CYP3A4 enzyme level would reach ∼40%–60% of the normal level 3 days after a single intravenous infusion of GSK19 (Fig. 4B).
Impact.
The drug interaction predictions for single intravenous administration of GSK19 indicated the risk was reduced after intravenous dosing but not abolished. The predicted DDI magnitude and time profile of enzyme recovery informed clinical management of CYP3A4 substrate comedications after a single intravenous dose of GSK19.
Case Study 5
Question.
Can PBPK modeling be used to provide a mechanistic understanding of observed time-dependent changes (reduced exposures) in pharmacokinetics as well as clinical drug-drug interactions with a CYP3A4 inducer?
Background.
In vitro data have shown that GSK38 is both an inducer and substrate of CYP3A4. A clinical pharmacokinetic study with GSK38 exhibited time-dependent nonlinear (subproportional) exposure upon repeat dosing at doses greater than 40 mg in the 1–160 mg dose range. The observed less than predicted accumulation of GSK38 after repeat-dose to steady state was hypothesized to be due to autoinduction of its own major oxidative metabolizing enzyme CYP3A4. Additionally, a drug interaction study with midazolam indicated that the victim drug midazolam exposure was decreased after repeated dosing with GSK38.
Simulations.
Previous reported applications of simCYP to model and predict the clinical pharmacokinetics and DDIs of CYP3A4 inducers (rifampin, carbamazepine, and phenobarbital) with CYP3A4 substrates (midazolam, nifedipine, simvastatin, and zolpidem) using in vitro hepatocyte data had demonstrated the capability of simCYP for predicting induction-based clinical DDIs (Xu et al., 2011). Therefore, simCYP v.11 was used to retrospectively gain a mechanistic understanding of the observed midazolam drug interaction and reduced exposure of GSK38 after repeated dosing. Model input parameters are summarized in Table 1.
Simulations predicted the time-dependent and dose-dependent exposure change due to CYP3A4 induction for GSK38 by incorporating EC50/Emax values of 0.25/5.46 based on an in vitro human hepatocyte induction study (n = 3 donors) and scaled by the simCYP calculator based on the positive control data for rifampicin. In the simulation, system-specific properties such as organ mass and blood flow were the default values using the simCYP virtual population for women at aged 20–50 years to mimic the clinical trial; the systemic clearance of GSK38 was assumed to be via hepatic elimination and systemic hydrolysis. The contribution of CYP3A4 to hepatic clearance was set as 0.75 (fraction metabolized) based on information obtained from in vitro experiments.
Simulations were performed for repeat dosing of GSK38 at 160 mg—the predicted pharmacokinetic profiles and reduction in exposure with time were comparative with those clinically observed, providing further model validation as well as confirming the autoinduction due to CYP3A4 induction and metabolism (Fig. 5). DDI predictions for midazolam while codosing with GSK38 at 160 mg for 10 days were further performed to help understand its induction effect on midazolam. An AUCi/AUC ratio of 0.33 was predicted and was consistent with the clinical observation of 0.28 (see the inset chart in Fig. 5).
Impact.
PBPK modeling confirmed the hypothesis of autoinduction of CYP3A4 being the major factor for GSK38 exposure reduction on repeated dosing, and verified the mechanism of CYP3A4 induction by GSK38 in the in vivo DDI study with midazolam. In addition, this modeling exercise provided internal validation on the simCYP capability to predict pharmacokinetics and DDI effects due to CYP induction for NCEs in development.
Discussion
Over the past 5 to 6 years (since the implementation of version 6 of the software) simCYP has been used in 41 instances in GSK where we consider the impact to have been significant.
Case study 1 is an example of a simulation that impacted regulatory communications (for other examples, see Table 2). This case study describes how reduced clinical study packages for both hepatic and renally impaired subjects were proposed and accepted based on simulations in virtual populations with mild, moderate, and severe functional impairment. For this example, the choice of simCYP was dictated by its inclusion of the hepatically and renally impaired population files, which are not available in the other modeling software packages routinely used in GSK. Available data indicate that, as would be expected, hepatic enzymes are compromised in hepatic failure, in addition to other physiologic changes including a reduction in liver size, protein binding, and liver blood flow. Also, there are a number of reports suggesting renal impairment not only affects elimination of the drug in the kidney but also the nonrenal route of drugs that are extensively metabolized in the liver (Pichette and Lebond, 2003; Vilay et al., 2008). Renal failure may influence hepatic drug metabolism either by inducing or suppressing hepatic enzymes or by its effects on other variables such as protein binding, hepatic blood flow, and accumulation of metabolites. This information is incorporated in the hepatic and renal population demographics in simCYP; as GSK1 is extensively metabolized by CYP3A4, the impact of potential changes in enzyme activity were simulated to inform subsequent clinical study designs. This case history highlights the benefit of PBPK approaches to incorporate physiologic changes in patient populations and simulate any impact on drug pharmacokinetics that will ultimately inform the need for dose adjustments.
Case studies 2, 3, and 4 are examples of simulations that have directly impacted clinical development decisions (see Table 3 for other examples). These particular examples were all selected as they represent instances where dynamic (time-variant) simulations were required to provide answers that could not have been provided by static models.
Case study 2 describes a situation where a clinical DDI study was initially conducted using a suboptimal study design because the initial dosing regimen failed to take account of the victim drug’s extended half-life. This is an example where prior use of DDI simulations could have saved the time and expense of repeating the study with a better design. This is the earliest of the case studies presented; if we were faced with a similar example of a potential DDI “victim” with a prolonged half-life today, our simulations with extended dosing of inhibitor would be performed well in advance of designing a clinical DDI study. Also, since this case study, there have been reports that, for such compounds, a ketoconazole dosing regimen of 200 mg twice a day could be more effective in eliciting a maximal inhibitory response than the standard 400 mg every day (Zhao et al., 2009), and this, too, would be simulated.
Case study 3 describes the investigation of the potential for using ritonavir boosting for a CYP3A4 substrate with a very short half-life. Ritonavir is a known, potent inhibitor of CYP3A4, and it has been used particularly for human immunodeficiency virus therapies (Hill et al., 2009) to increase exposure to drugs that are sensitive CYP3A4 substrates. This case study highlights the utility of using a dynamic PBPK model that incorporates variability on model parameters, particularly for infectious disease treatments where the potential to develop antimicrobial or antiviral resistance is a concern. The ability to further analyze the simulation results and gain an understanding of what proportion of patients would be expected to fall beneath the MEC can add significant value over static models and dynamic PBPK models focused on mean values, and can lead to improved decision making within clinical development.
Case study 4 describes the use of DDI simulations to predict the extent of change in exposure of a CYP3A4 substrate when coadministered with a metabolism-dependent inhibitor (CYP3A4 inactivator), intravenously versus orally, and to estimate the length of time after the last dose of inhibitor that would be required for full recovery of enzyme activity. The case study simulations are deliberately presented as they were originally performed—that is, using previous versions of the simCYP software—but this is one example that could potentially give a different result if reevaluated using a later version because it specifically addressed the question of CYP3A4 inactivation and recovery after coadministration with a mechanism-based inhibitor. For a mechanism-based inhibitor, the simulated magnitude of the DDI and recovery time of enzyme activity depends not only on the inactivation parameters of the inhibitor but also on the value used for enzyme turnover (e.g., the turnover rate constant [kdeg] or half-life). For hepatic CYP3A4, reported half-life values vary widely with a range of 10 to ≥100 hours (Yang et al., 2008). Before version 11, the default hepatic kdeg value for CYP3A4 used in simCYP was 0.0077 hour−1 (i.e., a half-life of 90 hours). With version 11, this kdeg value was updated (Rowland Yeo et al., 2011b) to 0.019 hour−1 (i.e., a half-life of 36 hours), so it is likely that if the simulations were repeated using the current version of the software, both the extent of DDI and the recovery time of CYP3A4 activity would be less than the original predictions. Future simulations involving estimates of enzyme recovery time would include a sensitivity analysis around the kdeg value. Additional simulations would be extended to predict the time to full enzyme recovery, or to predict the potential for DDIs at the time when patients are likely to be released from inpatient care.
Case study 5 is an example of a simulation that increases the mechanistic understanding of clinical observations (see Table 4 for other examples). GSK5’s reduced systemic exposure on repeat dosing (and the reduced exposure of coadministered midazolam) could be explained by induction of CYP3A4. Unlike CYP inhibition, which is an almost immediate response, CYP induction is a slow regulatory process. Therefore, the ability to simulate the time-course of DDIs mediated by CYP induction is of considerable benefit compared with the use of static models, which focus on predicting the mean exposure changes of victim drugs in the presence of inducers at steady-state concentrations (Almond et al., 2009; Chu et al., 2009). As few examples of applications of induction modeling had been previously reported (Xu et al., 2011), this exercise enabled an internal validation for NCEs in development and promoted confidence in using simulation tools such as simCYP for the prediction of pharmacokinetic profiles as well as DDIs resulting from enzyme induction.
Of the 41 examples presented here, only 3 to date in the first category have impacted regulatory communications; all of these are fairly recent and represent the growing acceptance of PBPK modeling by regulatory bodies together with the increasing confidence of pharmaceutic companies in relying on the models. The second category of examples is currently the largest, and it includes simulations that have directly impacted clinical development decisions. Particularly notable here are examples relating to clinical DDI studies (necessity and design), comedication exclusions, and therapeutic dose regimens. The fact that the majority of our examples currently fall into this category reflects that our early use of simCYP was particularly focused on the prediction of clinical DDIs and, before the inclusion of transporters into the software, these DDIs were further limited to those mediated by the inhibition or induction of cytochrome P450 enzymes. Future applications in this category will potentially include more drug transporter predictions and enzyme/transporter interactions, in line with expanding scientific understanding in this area (Giacomini et al., 2010). The final category includes examples for which the use of simCYP has increased the mechanistic understanding of clinical observations. These cases can be described as informing the learn-and-confirm cycle and will therefore impact future clinical study designs as well as directly or indirectly impacting clinical decisions.
This report shows that, in line with reported trends, modeling and simulation within GSK, as exemplified here by our use of simCYP, is increasing in all areas of drug development. Of the key examples presented as case studies, only one was completed before 2008 whereas the other four were conducted in 2010 or later. It is also evident that, with increasing experience and awareness of new publications in the field, the modeling can be refined; for example, two of our five historic case studies would have been performed with at least some additional scenario testing if the same questions were to arise today. Looking forward, it is expected that the number of modeling examples used in regulatory communications will grow, as will examples of modeling used for pharmacokinetic-pharmacodynamic assessment and simulating pharmacokinetics of biologics. However, the prediction of the magnitude of clinical DDIs (enzyme and transporter mediated) will continue to be important as this, combined with interrogation of potential comedications and assessment of clinical risk, will obviate the conduct of unnecessary or untimely clinical studies.
Acknowledgments
The authors thank Nicoletta Pons for contributions to case study 2.
Authorship Contributions
Participated in research design: Shardlow, Generaux, Patel, Tai, Tran, Bloomer.
Conducted experiments: Generaux, Patel, Tai, Tran.
Performed data analysis: Shardlow, Generaux, Patel, Tai, Tran, Bloomer.
Wrote or contributed to the writing of the manuscript: Shardlow, Generaux, Patel, Tai, Tran, Bloomer.
Footnotes
- Received May 15, 2013.
- Accepted September 5, 2013.
Abbreviations
- AUC
- area under the curve
- AUCi
- area under the curve (inhibited)
- CYP
- cytochrome P450
- DDI
- drug-drug interaction
- EMA
- European Medicines Agency
- FDA
- U.S. Food and Drug Administration
- FTIH
- first time in human
- GF120918
- N-4-[2-(1,2,3,4-tetrahydro-6,7-dimethoxy-2-isoquinolinyl)ethyl]-phenyl)-9,10-dihydro-5-methoxy-9-oxo-4-acridine carboxamide hydrochloride
- GSK
- GlaxoSmithKline
- kdeg
- turnover rate constant
- NCE
- new chemical entity
- MEC
- minimum effective concentration
- PBPK
- physiologically based pharmacokinetics
- Copyright © 2013 by The American Society for Pharmacology and Experimental Therapeutics