Imatinib has revolutionised the treatment for cancer and led to a subsequent discovery of a class of drugs known as small molecule kinase inhibitors (Rowland et al., 2017). It is approved as the first-line treatment for chronic myeloid leukaemia (CML) and gastrointestinal stromal tumours (GIST) in adult patients and for CML and Philadelphia chromosome-positive (Ph+) acute lymphoblastic leukaemia (ALL) in children and adolescents (Suttorp et al., 2018a). A phase III clinical trial highlighted that imatinib was well tolerated and effective for newly diagnosed paediatric CML, with a 5-year progression free survival of 94% (Suttorp et al., 2018b). A 5-year follow-up of imatinib (340 mg/m²/d) in combination with conventional chemotherapy drugs (e.g., cyclophosphamide, methotrexate, and cytarabine) showed a favorable outcome in children with Ph+ ALL, similar to that of bone marrow transplantation (Schultz et al., 2014).

The prevalence of childhood CML and Ph+ ALL, however, is very low, accounting for around 2% of all leukaemias and 3%–5% of ALL in children, respectively (Coebergh et al., 2006). Therefore, an optimal dose for imatinib in paediatric patients, let alone its potential drug-drug interactions, has been less widely explored. Imatinib is mainly metabolised by cytochrome P450 (CYP)3A4 and CYP2C8 (Barratt and Somogyi, 2017), and thus, has a potential for drug interactions with modulators of these CYP enzymes. A clinically significant interaction between imatinib and carbamazepine, a CYP3A and CYP2C8 inducer, was described in a 12-year old CML patient with epilepsy (Taguchi et al., 2014). However, little is known about imatinib interactions with other potential perpetrator drugs in paediatric patients. Conducting a dedicated clinical interaction study in paediatric population remains challenging owing to the ethical and logistical constraints (Barker et al., 2018). Clearly, a feasible and systematic approach to address this gap is warranted.

Physiologically based pharmacokinetic (PBPK) modeling can account for anatomical and physiological growth and organ maturation underlying age-related changes in the pharmacokinetics of a drug of interest (Yellepeddi et al., 2019). This facilitates an extrapolation across the age spectrum (Kuepfer et al., 2016). The PBPK approach has been increasingly embraced by regulatory authorities for the purposes of informing dose selection, providing simulation-based trial design and investigating potential drug interactions in paediatric populations (Cole et al., 2017; Bi et al., 2019). According to applications related to PBPK that were submitted to the US Food and Drug Administration (FDA) from 2008 to 2017, PBPK analyses are mainly intended for evaluating and predicting enzyme-based drug interactions (60% of all applications), followed by utilization in paediatric area (15%) (Grimstein et al., 2019). PBPK modeling and simulation has been an integral part of drug development for paediatric cancers (Rioux and Waters, 2016). PBPK models which can capture developmental changes in biological components are useful in describing in paediatrics the pharmacokinetics of anticancer drugs, including etoposide (Kersting et al., 2012), busulfan (Diestelhorst et al., 2014), docetaxel (Thai et al., 2015), actinomycin D (Walsh et al., 2016) and nilotinib (Heimbach et al., 2019).

A PBPK model for imatinib that incorporates maturational changes of key drug-metabolising enzymes and age-dependent organ development can help inform optimal dose selection in children. PBPK modeling and simulation also provides a greater understanding of potential drug interactions with imatinib in this vulnerable patient group which remains largely unexplored. The aim of this study was to develop and implement a paediatric PBPK model of imatinib for investigating optimal dosing regimens in children and the vulnerability to drug interactions relative to adults with a range of CYP3A modulators.

In this study, a PBPK model for imatinib was developed and verified in adults and subsequently extrapolated to children and adolescents (aged 2–18 years). The verified PBPK model was then implemented to explore optimal dosing regimens for imatinib in children and to evaluate potential drug interactions with CYP3A modulators. The workflow of this study is summarized in Figure 1 .

We developed a PBPK model for imatinib in adult populations and extrapolated its use to paediatrics. The PBPK model was able to describe imatinib pharmacokinetics in both populations and had a capability to predict drug interactions with a range of CYP3A modulators.

A paediatric PBPK model for imatinib has been reported previously in a regulatory document submitted to European Medicines Agency (2013). Unfortunately, a lack of details regarding this PBPK model’s structure and parameters limits its further use and interpretation. The PBPK model in the current study was verified to a larger set of clinically published pharmacokinetic data and its implementation was extended to predict drug interactions in paediatrics.

Scaling drug doses from adults to children is far from a straightforward process (Johnson, 2008). Both population pharmacokinetic and PBPK approaches have been used independently or in combination to guide drug dosing in paediatric patients (Johnson, 2005). A population pharmacokinetic model incorporating body weight as a primary covariate with an allometric exponent, e.g. ¾ for clearance, often does not perform well in infants and young children due to maturation of drug eliminating processes (Germovsek et al., 2017). In most cases, the predictions are improved by employing a sigmoidal ontogeny function driven by postmenstrual age (Anderson and Holford, 2011). However, the maturation half-life and Hill coefficient which parameterise the function vary across different drugs (Holford et al., 2013; Germovsek et al., 2017) and thus, sufficient number of individuals with age around the maturation half-life is necessary for precise parameter estimation. PBPK modeling and simulation offers an alternative approach to evaluate an optimal dosing regimen in the paediatric population. It integrates drug-specific inputs and system-related parameters, the latter of which encompass developmental changes in physiology and maturational rates of drug-metabolising enzymes and proteins involved in drug disposition (Maharaj and Edginton, 2014). This approach enables extrapolation from adults or between age groups within paediatric populations and increases the mechanistic understanding of potential sources of interindividual variability in systemic exposure to a drug.

The ontogeny profiles of key CYP enzymes responsible for imatinib metabolism ( Figure 2 ) are based on a meta-analysis of in vitro CYP activity in post-mortem livers of donors from different ages (Johnson et al., 2006). The maturation functions tend to underestimate the apparent clearance of CYP3A substrates in neonates and infants (Salem et al., 2014). Two in vivo-derived algorithms have been proposed to improve the prediction (Salem et al., 2014; Upreti and Wahlstrom, 2016). The Upreti and Wahlstrom model for CYP3A4 maturation has been shown to perform better with less underprediction of clearance (Johnson et al., 2019). However, as expected, the PBPK simulations that implemented this ontogeny for children older than 2 years of age yielded a similar result to that of in vitro maturation function (results not shown). Therefore, the latter, which is incorporated in the Simcyp Simulator (version 17) by default, was utilized throughout the simulations. Developmental changes in organ size, particularly liver volume were also incorporated in the PBPK model. Liver volume was most parsimoniously described by a nonlinear regression against BSA as shown in Eq. 6 (Johnson et al., 2005). Interestingly, this equation was in concordance with an allometric weight model with an exponent of ¾ in estimating liver volume from infants to adolescents (Fanta et al., 2007). The correlation between liver volume and BSA alone was superior than that with other covariates (Johnson et al., 2005), in agreement with the findings of a nonlinear mixed effect modeling approach (Small et al., 2017). All the ontogeny equations used in the current study were driven by postnatal age. Postmenstrual age is more useful if preterm neonates are included in PBPK simulations (Abduljalil et al., 2019; Germovsek et al., 2019).

It is noteworthy that the PBPK models may overestimate clinically observed peak plasma concentrations (C_max) since they report predicted concentrations at the central venous compartment rather than the peripheral vein from which blood (plasma) samples were collected. This is particularly important for intravenous (i.v.) administration routes where a substantial amount of drug is delivered to central venous compartment directly and equilibration to the peripheral venous sites may not be instantaneous (Musther et al., 2015). A PBPK model prediction of drug concentrations at a peripheral sampling site based on contributions from surrounding tissues (e.g., adipose, muscle, and skin) as proposed by Musther et al (2015) has proven to be useful to correct the PBPK predictions at initial time following i.v. administrations. As depicted in Figure S1 , implementation of this strategy within the Simcyp Simulator improved the PBPK model predictions of C_max following a 1-h infusion of imatinib (100 mg) (Peng et al., 2004) compared to that of central venous compartment (prediction differences of imatinib C_max of 0.99 vs. 1.42). Prediction differences for other pharmacokinetic parameters of imatinib were similar between the two strategies (results not shown). Conversely, the peripheral sampling site model has little to no effect on PBPK predictions of C_max of imatinib given orally (results not shown). Unlike i.v. administration over a short period of time, oral administrations of drugs are likely to give sufficient time for central venous compartment (pooled venous return) and peripheral vein in the arm to equilibrate (Musther et al., 2015).

The observed interindividual variability of imatinib concentrations in children on day 1 appeared to be higher than that at steady-state from the corresponding patient cohort ( Figures 3F–I ). The reason for this trend was not clear, but may be related to a lower between individual variability in CYP3A4 activity due to the autoinhibition by imatinib following chronic exposure (Filppula et al., 2012; Filppula et al., 2013a). PBPK simulations also highlighted a higher interindividual variability of imatinib concentrations at a fixed daily dose compared to a BSA-normalized dosing regimen ( Figures 3M, N vs. 3F–L ). A daily dose administered on a mg/m² basis in paediatric populations is usually preferred to body weight-based and flat-fixed dosing regimens owing to more favorable pharmacokinetic variability, particularly over a wide age range (Bartelink et al., 2006; Hempel and Boos, 2007).

A clear exposure-response relationship for imatinib has not been established in younger patients with CML. Thus, the proposed targets in children and adolescents were based on the concentration known to be safe and efficacious in adults (C_min ranging from 1,000 to 3,200 ng/ml (Lankheet et al., 2017)). This was further supported by the similar biological and clinical features of CML observed in adult and younger patients (Barr, 2010), with only a slight difference, particularly a higher leukocyte count presented in the latter (Millot et al., 2005). Paediatric and adult patients also had comparable response and safety profiles (e.g., occurrence of grade 3/4 haematological toxicities and musculoskeletal adverse events) to an equivalent dose of imatinib (Millot et al., 2011). This was not the case for solid tumours harboring mutations in the gene that encodes tyrosine kinase KIT (e.g. GIST). Imatinib exerted minor anticancer activity in children with GIST compared to the adult cohort, despite similar systemic concentrations (Geoerger et al., 2009).

The observed trend of a higher interindividual variability of simulated C_min in children aged between 2 and 5 years and middle-aged adults compared to other age groups ( Figure 4 ) was likely attributed to a higher variability within these age bands due to maturational changes of CYP enzymes that have not attained adult levels of expression (Johnson et al., 2006) and a reduction of total hepatic clearance related to a decrease of liver weight and scaling factor (e.g., microsomal protein per gram of liver/MPPGL) (Barter et al., 2007; Chetty et al., 2018), respectively.

PBPK simulations suggested a similar C_min following imatinib doses of 230 and 340 mg/m²/d in paediatrics and 400 and 600 mg/d in adult population, respectively (p > 0.01). This was in agreement with the finding in clinical studies in children with Ph+ leukaemias or GIST which indicated a similar systemic exposure of imatinib at daily doses of 230 and 340 mg/m² compared to those of adult patients treated with 400 and 600 mg/d of imatinib, respectively (Champagne et al., 2004; Geoerger et al., 2009). The C_min target of 1,000 ng/ml was predicted to be attainable by a 230 mg/m²/d dose in paediatric age groups (similar to an adult dose of 400 mg/d), albeit with a large subset of the population below the target. Due to this variability, a higher dose of 340 mg/m²/d (corresponds to an adult dose of 600 mg/d) might be needed. This was in line with the recommendation for the treatment of CML in children with the recommended initial doses of 260–300 mg/m²/d and 400 mg/m²/d for chronic and accelerated phases, respectively (de la Fuente et al., 2014).

There was a good agreement between PBPK model prediction and clinically observed changes in imatinib concentrations due to the coadministration of carbamazepine in adult and paediatric populations ( Figure 6 ). It should be noted that clinical pharmacokinetic data in the latter came from one Japanese paediatric patient (a case study) (Taguchi et al., 2014). PBPK simulations in paediatrics refer to European ancestry, from which the ontogeny functions for drug-metabolising enzymes and AAG were derived (Johnson et al., 2006). However, our previous simulation study suggested little to no difference in imatinib pharmacokinetic between people from Japanese and European ancestry (unpublished).

Clinical drug-drug interaction (DDI) data in adults may not be suitable for extrapolation across all paediatric age bands (Salem et al., 2013b; Salerno et al., 2019). The magnitudes of enzyme-based DDI are dictated by the level of contribution (f_m) and maturational rates of corresponding CYP enzymes (Salem et al., 2013a). In this study, a PBPK modeling approach was utilized to evaluate drug interactions with imatinib in paediatrics. The trend and extent of interactions between imatinib and CYP3A modulators (carbamazepine, rifampicin and ketoconazole) were predicted to be similar between paediatric and adult populations, despite a slight difference in the simulated means and interindividual variabilities ( Figure 7 ). Imatinib inhibits its own CYP3A4-mediated metabolism following multiple-dosing regimen (Filppula et al., 2012). Thus, the effect of CYP3A modulators on imatinib metabolism was likely to be diminished following a long-term use of imatinib, as observed in a clinical interaction study between imatinib and ritonavir (van Erp et al., 2007). The extent of modulation by CYP3A inhibitors, either direct (reversible) or mechanism-based inhibitors, e.g., ketoconazole and ritonavir, respectively was predicted to be more affected following repeated-dose administration of imatinib, compared to that observed with CYP3A inducers (e.g., rifampicin and carbamazepine). This was due to limited residual CYP3A activity which can further be inhibited in the former. Since imatinib undergoes little to no metabolism in the enterocytes (Barratt and Somogyi, 2017), inducers of CYP3A confined to intestinal enzymes (e.g., hyperforin in St John’s wort) are unlikely to affect steady-state CL/F of imatinib (Adiwidjaja et al., 2019).

The limitation of this study is a lack of specific maturation functions for children with cancer implemented in the PBPK model. The trend of developmental changes in organ size, CYP enzymes and plasma proteins observed in healthy children may not hold true for the paediatric cancer population (Thai et al., 2015). A further limitation to this study is the exclusion of children less than 2 years of age from the simulations ( Figures 4 and 7 ) due to a paucity of clinical pharmacokinetic data for this age group (CML is exceptionally rare in very young children (de la Fuente et al., 2014)). Moreover, there is a high uncertainty in the maturation pattern of CYP3A4 in this challenging age group (Johnson et al., 2019), which is further complicated by the potential presence of CYP3A7 enzyme. The latter is absent in adults, but expressed at a high level during foetal life and decreases progressively throughout the first 2 years after birth (Allegaert and van den Anker, 2019). A further study to elucidate CYP3A7 contribution to imatinib metabolism is necessary in order to perform a PBPK prediction with confidence in children less than 2 years.

In conclusion, a PBPK model for imatinib was successfully developed in adults and extrapolated to the paediatric population. The PBPK model was able to describe clinical pharmacokinetic data from published studies observed in adults, children and adolescents. PBPK simulation suggested an optimal dosing regimen range for imatinib of 230–340 mg/m²/d in paediatrics, in concordance with the recommended initial dose for treatment of childhood CML. The simulations also highlighted that children and adults being treated with imatinib have similar vulnerability to drug interactions that modulate drug metabolising enzyme activity. These findings suggest that at steady-state, imatinib is more susceptible to hepatic induction compared to inhibition of CYP3A enzymes.

All datasets generated for this study are included in the article/ Supplementary Material .

JA, AB, and AM wrote the manuscript, designed the research, and contributed to the interpretation. JA performed the simulations and analyzed the data.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.