Population pharmacokinetic and pharmacogenetic analysis of 6-mercaptopurine in paediatric patients with acute lymphoblastic leukaemia

Abstract

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT

• The cytotoxic effects of 6-mercaptopurine (6-MP) were found to be due to drug-derived intracellular metabolites (mainly 6-thioguanine nucleotides and to some extent 6-methylmercaptopurine nucleotides) rather than the drug itself.

• Current empirical dosing methods for oral 6-MP result in highly variable drug and metabolite concentrations and hence variability in treatment outcome.

WHAT THIS STUDY ADDS

• The first population pharmacokinetic model has been developed for 6-MP active metabolites in paediatric patients with acute lymphoblastic leukaemia and the potential demographic and genetically controlled factors that could lead to interpatient pharmacokinetic variability among this population have been assessed.

• The model shows a large reduction in interindividual variability of pharmacokinetic parameters when body surface area and thiopurine methyltransferase polymorphism are incorporated into the model as covariates.

• The developed model offers a more rational dosing approach for 6-MP than the traditional empirical method (based on body surface area) through combining it with pharmacogenetically guided dosing based on thiopurine methyltransferase genotype.

AIMS

To investigate the population pharmacokinetics of 6-mercaptopurine (6-MP) active metabolites in paediatric patients with acute lymphoblastic leukaemia (ALL) and examine the effects of various genetic polymorphisms on the disposition of these metabolites.

METHODS

Data were collected prospectively from 19 paediatric patients with ALL (n = 75 samples, 150 concentrations) who received 6-MP maintenance chemotherapy (titrated to a target dose of 75 mg m⁻² day⁻¹). All patients were genotyped for polymorphisms in three enzymes involved in 6-MP metabolism. Population pharmacokinetic analysis was performed with the nonlinear mixed effects modelling program (nonmem) to determine the population mean parameter estimate of clearance for the active metabolites.

RESULTS

The developed model revealed considerable interindividual variability (IIV) in the clearance of 6-MP active metabolites [6-thioguanine nucleotides (6-TGNs) and 6-methylmercaptopurine nucleotides (6-mMPNs)]. Body surface area explained a significant part of 6-TGNs clearance IIV when incorporated in the model (IIV reduced from 69.9 to 29.3%). The most influential covariate examined, however, was thiopurine methyltransferase (TPMT) genotype, which resulted in the greatest reduction in the model's objective function (P < 0.005) when incorporated as a covariate affecting the fractional metabolic transformation of 6-MP into 6-TGNs. The other genetic covariates tested were not statistically significant and therefore were not included in the final model.

CONCLUSIONS

The developed pharmacokinetic model (if successful at external validation) would offer a more rational dosing approach for 6-MP than the traditional empirical method since it combines the current practice of using body surface area in 6-MP dosing with a pharmacogenetically guided dosing based on TPMT genotype.

Introduction

6-Mercaptopurine (6-MP) is a purine antimetabolite widely used in the maintenance chemotherapy of childhood acute lymphoblastic leukaemia (ALL), the most common cancer in children. 6-MP has been a key component of almost every successful therapeutic regimen for low- to mild-risk ALL. However, it was not until the early 1980s that 6-MP cytotoxic effects were found to be due to drug-derived intracellular metabolite concentrations rather than the plasma level of 6-MP itself [1].

Following oral administration, 6-MP undergoes extensive biotransformation by three enzymes, two of which are catabolic, xanthine oxidase (XO) and thiopurine S-methyl transferase (TPMT) and one anabolic, hypoxanthine phosphoribosyl transferase (HPRT). XO metabolises 6-MP to 6-thiouric acid (6-TU), whereas TPMT methylates 6-MP to 6-mMP. HPRT carries out the first anabolic step to produce 6-thioinosine monophosphate (6-TIMP) and subsequently the active 6-thioguanine nucleotides (6-TGNs). 6-TIMP can alternatively be methylated by TPMT, yielding 6-methylmercaptopurine nucleotides (6-mMPNs) [2]. Finally, it is hypothesized that 6-TIMP is converted successively into 6-thioinosine diphosphate (6-TIDP) and triphosphate (6-TITP) to form 6-TIMP once again by the action of the enzyme inosine triphosphatase (ITPA) [3].

Cytotoxic effects of 6-MP are achieved primarily through the incorporation of 6-TGNs into the DNA of leucocytes, due to their structural similarity to the endogenous purine based guanine [2, 4]. Moreover, 6-mMPNs are strong inhibitors of purine de novo synthesis, which is a well-established protocol to achieve immunosuppression [5, 6]. Despite these facts, the pharmacokinetics of the active metabolites of 6-MP remain poorly explored. A better understanding of the disposition of 6-TGNs and 6-mMPNs would therefore be immensely helpful in improving the design of dosing regimens for 6-MP.

In this study, the pharmacokinetics of 6-TGNs and 6-mMPNs in paediatric patients with ALL under 6-MP maintenance chemotherapy were examined prospectively and, for the first time, a population pharmacokinetic model for 6-MP active metabolites, 6-TGNs and 6-mMPNs, in paediatric patients with ALL was developed. In developing this model, potential factors that could lead to variability in 6-MP cytotoxic metabolites that would be particularly helpful in improving the dosing guidelines for 6-MP were assessed.

Methods

Patients and data collection

The study was approved by the National Health Service Office for Research Ethics Committees in Northern Ireland. Informed parental consent was obtained for each child before enrolment in the study. In addition, verbal assent was obtained from older children (>10 years) after provision of a verbal description of the study and what it involved.

Data were collected from 19 paediatric patients attending the Haematology and Oncology Outpatient Department at the Royal Belfast Hospital for Sick Children and who had been diagnosed to be suffering from ALL. Blood samples were taken from children who had been on continuous/maintenance 6-MP therapy for at least 1 month and who had received constant daily doses for at least 1 week. Patients who satisfied the above criteria but who had received intensification therapy or red blood cell (RBC) transfusion within the previous 2 months were excluded.

Blood samples were obtained at a phase of treatment when children had an indwelling cannula for vincristine therapy and at least 12 h after the preceding 6-MP dose. (Note, the blood sample was taken prior to the administration of vincristine.)

Maintenance chemotherapy for ALL patients consisted of daily oral 6-MP and weekly methotrexate. The 6-MP dose was titrated to the target protocol dose of 75 mg m⁻² day⁻¹, adjusted for each child according to leucocyte count and the presence of clinically relevant infections. Additionally, a monthly dose of intravenous vincristine was given to all children irrespective of blood count. Chemotherapy was administered usually for 2 or 3 years. The children had their full blood counts assessed at each clinic visit (every 2 weeks) for bone marrow toxicity.

The blood samples (1.5 ml) were collected in ethylenediamine tetraaceticacid (EDTA) tubes and kept on ice until centrifuged at 1000 g for 10 min to separate plasma from RBC. Separated plasma was frozen at −20°C in Eppendorf® tubes while RBC were washed twice with a balanced salt solution, then suspended at a density of 8 × 10⁸ RBC per 200 µl and kept frozen at −20°C until required for further processing. These latter samples for the determination of metabolite content were taken on a maximum of five occasions (one sample per occasion), at monthly intervals, over the study period.

A sample of blood (1 ml) from each patient, taken on one occasion only, was collected in an EDTA tube and kept at −20°C without centrifugation for genotyping the enzymes of interest (200 µl of whole blood was sufficient for this purpose).

In addition to information on dosing and times of sampling, the following data were collected for each child: age, weight, surface area, gender, ongoing pathology (e.g. renal and/or hepatic impairment), concomitant drug therapy, lab test results and records of any side-effects experienced. The demographic and clinical characteristics of the study participants are shown in Table 1.

Table 1.

The demographic and clinical characteristics of the ALL population

Parameter	n = 19
Gender (F : M)	6 : 13 (32% : 68%)
Age [median (range) years]	10 (3–17)
Weight [median (range) kg]	33.4 (13.2–77.5)
Body surface area [median (range) m2]	1.14 (0.59–2)
6-MP daily dose [median (range) mg]	50 (10–100)
6-MP daily dose [median (range) mg kg−1]	1.42 (0.16–3.46)
6-MP daily dose [median (range) mg m−2]	40 (5.88–76.47)
Co-medication during 6-MP chemotherapy
Methotrexate weekly dose [median (range) mg]	15 (5–25)
Cotrimoxazole (b.d. twice weekly) [median (range) mg]	360 (120–480)
Haematological parameters
Hb [median (range) g dl−1]	12.9 (10.9–16.5)
WBC [median (range) × 109 l−1]	3.3 (1.2–9.1)
PLT [median (range) × 109 l−1]	282 (66–648)
ANC [median (range) × 109 l−1]	1.68 (0.3–8.3)
XO, TPMT, and ITPA genotypes
XO A1936→G (heterozygotes/homozygotes)	1/0
XO A2107→G	1/0
TPMT*3A	1/0
TPMT*3B	–
TPMT*3C	2/0
ITPA C94→A	3/0
ITPA IVS2+21A→C	2/1

ALL, acute lymphoblastic leukaemia; 6-MP, 6-mercaptopurine; WBC, white blood cells; PLT, platelets; ANC, absolute neutrophil count; XO, xanthine oxidase; TPMT, thiopurine methyltransferase; ITPA, inosine triphosphatase.

Assay of 6-MP metabolites

RBC concentrations of 6-MP active metabolites, 6-TGNs and 6-mMPNs, were measured by a reversed high-performance liquid chromatography methodology that was developed earlier [7]. Intraday and interday coefficients of variation (CV) were 3.1 and 4.3%, and the limits of quantification were 13 and 95 pmol per 8 × 10⁸ RBC, respectively.

Genotyping of TMPT, ITPA and XO

All patients were screened for seven common polymorphisms in three enzymes involved in 6-MP metabolism (XO, TPMT and ITPA) that are potentially linked to the pharmacodynamics, toxicity and treatment outcome of 6-MP; two polymorphisms in XO (A1936→G and A2107→G) and ITPA (C94→A and IVS2+21A→C) and three polymorphisms in TPMT (TPMT*3A, TPMT*3B and TPMT*3C).

Detection of the various single nucleotide polymorphisms (SNPs) in the enzymes’ genetic loci was based on TaqMan® genotyping assays (ABI, Foster City, CA, USA) using somatic cell DNA extracted from patient blood samples (QIAmp® DNA Blood Mini kit; Qiagen, Hilden, Germany). The conditions used for polymerase chain reaction and subsequent detection of the genotypes were as described in the manufacturer's instructions.

Population pharmacokinetic modelling

The pharmacokinetics of 6-MP active metabolites, 6-TGNs and 6-mMPNs, were determined using a population approach in which concentrations from ALL patients were analysed simultaneously to produce estimates of the pharmacokinetic parameters. RBC concentration–time profiles of 6-TGNs and 6-mMPNs were used for nonlinear mixed effect modelling by extended least squares regression using nonmem (version VI, level 1.1 with double precision) [8] installed on a personal computer in conjunction with DIGITAL Visual Fortran compiler (version 5.0.A). Patients were assigned randomly to either an index group (n = 15) for the development of the pharmacokinetic model, or to the validation group (n = 4) for the purpose of assessing the predictive performance of the derived model. The first-order conditional estimation (FOCE) method with interaction was used to estimate population mean parameters, interindividual variability (IIV) in these parameters and residual variability between measured and predicted metabolite concentrations.

The concentration–time courses of 6-MP metabolites were described by using a one-compartment model with first-order absorption and elimination. The absorption rate constant (k_a) and the bioavailability factor (F) of the model were fixed at 1.3 h⁻¹ and 22%, respectively, according to the literature [1, 9]. The pharmacokinetic parameters estimated from this model (implemented using PREDPP subroutine ADVAN6) were clearance (CL) for 6-TGNs and 6-mMPNs and the fractional metabolic transformation of 6-MP into 6-TGNs.

The relationship between the parent drug and its metabolites was defined according to the following differential equations:

where k_a is the absorption rate constant of the parent drug, k₂₀ is the elimination rate constant of 6-MP (k₂₀ = 0.53 h⁻¹ based on literature [1, 9]), k_me is the metabolic transformation rate constant of the parent drug into either 6-TGNs or 6-MPNs. FMi is the fractional metabolic transformation into the metabolite i (6-TGNs are designated by the number 3 and 6-mMPNs are designated by the number 4, FM₄ = 1 – FM₃), and A, C are the amount/concentration of the drug or metabolite at the time t. k₂₀ = k_me + k_other (k_other is the elimination rate constant of bioavailable 6-MP transformed by other metabolic processes, k_other = 0.22 × k₂₀ = 0.1166 h⁻¹ based on the literature [9]).

An exponential error model was used to describe the deviations of the individual's clearance from the true (but unknown) population mean values:

where CL_i is the CL of the ith individual, TVCL is the typical population estimate for CL and η_i,CL is a random variable that distinguish the i^th individual's parameter from the population mean values as predicted by the regression model and is assumed to be normally distributed in the log domain with a geometric mean of zero and a variance of . An exponential model was chosen for the IIV, since pharmacokinetic parameters are usually log-normally distributed. In modelling this variability, the square root of the variance was interpreted as the CV.

Residual variability, which describes the residual error between the measured and the predicted metabolite concentrations, was modelled using additive, proportional and combined error structures. The additive error model, however, best described the residual variability. This variability could arise from intra-individual variability in the pharmacokinetic parameters, inaccuracy in the timing of sample collection and dosage administration, assay error, model misspecification, or non-adherence to therapy.

C_ij is the measured and C_pred,ij is the model predicted metabolite concentration of the i^th individual at the j^th sampling time and ε_ij is the residual error term, which is a random variable with zero mean and variance of σ². The magnitude of this residual error or variability was expressed as a standard deviation (SD).

Regression model

The initial analysis for the population pharmacokinetics of 6-MP metabolites, 6-TGNs and 6-mMPNs, was conducted without including any patient covariates in the model (BASE model). The conditional estimates of η_i,CL were obtained from this BASE model and then plotted against the following covariates: age, gender, weight, body surface area (BSA), TPMT, ITPA and XO genotypes in order to identify any potential relationship between CL and the covariates. The influence of the identified covariates on CL was individually assessed by incorporating them in the BASE model (univariate analysis). The regression relationship for CL was modelled in a linear or nonlinear way for continuous covariates:

• Proportional: CL = θ_CL × (1 + θ_COV × COV)

• Power: CL = θ_CL × COV^θCOV

• Exponential: θ_CL × e^{θCOV × COV}

where COV is a general continuous covariate and θ_CL and θ_COV are the regression coefficients to be estimated by nonmem.

Categorical covariates (1 or 0) were examined using a multiplicative model:

• 

where θ_CL is the population value in the absence of the covariate (COV = 0) and θ_COV is the fractional change when the covariate is present (COV = 1). Similar models were used to investigate the effect of covariates on FM₃.

For each model, the improvement in the fit obtained on addition of a fixed effect variable (covariate) into the model was assessed using a likelihood ratio test. The change in the objective function value (ΔOBJF) produced by the inclusion of a covariate represents a statistic that is proportional to minus twice the log-likelihood of the data and approximates a χ² distribution with degrees of freedom equal to the difference in the number of structural parameters (θs) between two models. A decrease in the OBJF ≥ 6.63 was considered statistically significant (P < 0.01, 1 degree of freedom) for the addition of one fixed effect.

The goodness of fit for each model was also assessed by examining the precision of parameter estimates (i.e. standard errors of the mean), the decrease in interindividual and residual variability, and graphs of residuals (RES), weighted residuals (WRES) and measured metabolite concentrations plotted separately against predicted concentrations.

As a result of the univariate analysis, each model with significant effect was ranked according to its ΔOBJF compared with the BASE model. The model with the largest ΔOBJF was designated as the INTERMEDIATE model and multiple regression analysis with forward selection was performed where covariates were incorporated into the INTERMEDIATE model one by one along the rank order established by univariate analysis until all significant covariates were included and no further statistically significant reduction in OBJF was obtained (FULL model). The FULL model was then subjected to stepwise, backward elimination to obtain the FINAL model. An increase in the OBJF ≥6.63 (P < 0.01, 1 degree of freedom) was required to retain the covariate in the FINAL model.

Model evaluation

Since external validation using a new dataset from another study is extremely difficult in paediatric studies, internal validation using data splitting or resampling techniques is an appropriate alternative [10]. In the present study, internal validation using the data-splitting technique was performed to verify the predictive value of the population model.

The predictive performance of the model was assessed in terms of bias (mean prediction error) and precision (root mean square prediction error) by comparing the measured concentrations in the validation group (n = 4) with the corresponding predicted values by the population model using post hoc Bayesian forecasting. This was achieved by fixing the structural and variance model parameters to the values estimated in the final population model.

The mean prediction error (ME) and root mean square prediction error (RMSE) were determined according to the following formulae [11]:

The prediction error (pe_i) was calculated as:

where C_obs and C_pred represent the observed and predicted concentrations, respectively.

In addition, WRES and measured metabolite concentrations were plotted separately against the predicted concentrations to assess visually the deviations of model predicted from observed metabolite concentrations in the validation group.

In order to evaluate the performance of the final model, obtained by fitting the full dataset (comprising index and validation groups combined together), a posterior visual predictive check was performed by simulating from the final estimates and comparing the distribution of the observations with the simulated distribution. The adequacy of the model was demonstrated by plotting the time course of the observations along with the prediction interval for the simulated values.

Results

On oral administration of 6-MP, large interindividual differences were observed as regards metabolite concentrations and the course of elimination as shown in Figure 1. It is apparent from the graph that patients with TPMT mutations had higher 6-TGN concentrations but compared with other patients they had relatively low 6-mMPN concentrations in erythrocytes.

Figure 1.

Individual 6-thioguanine nucleotide (6-TGN) (A) and 6-methylmercaptopurine nucleotide (6-mMPN) (B) concentration plots. Patients having any mutation are highlighted and their corresponding types are displayed. The therapeutic lower and upper limits suggested in literature are indicated by the dashed lines. TPMT mutant (▵); ITPA mutant (□); TPMT and ITPA mutant (○)

At certain time points after treatment with the drug, 6-TU could also be identified in addition to the parent drug, 6-MP. However, their low levels (due to the sampling time chosen in this study as stated above) did not qualify them for incorporation into nonmem analysis. nonmem analysis in this pharmacokinetic study, therefore, was performed using the measured erythrocyte levels of 6-TGNs and 6-mMPNs in all samples (n = 75 samples, 150 concentrations) obtained from 19 paediatric patients with ALL receiving 6-MP maintenance chemotherapy (a maximum of five samples was obtained per patient, one sample per occasion).

Pharmacokinetic modelling

In the initial model, the data were described with a one-compartment model with first-order absorption and elimination since there were no points to enable the accurate evaluation of the distribution phase. In addition, since most of the kinetic data were collected in the post-absorption phase, k_a and F (the bioavailability factor) could not be reliably estimated. Hence, their values were fixed according to the literature (1.3 h⁻¹ and 22%, respectively) throughout the analysis [1, 9]. An additive error model best described the residual variability and was used in the basic pharmacokinetic model (BASE) in order to be used for further analysis.

The population estimates from the BASE model for FM₃ (the fractional metabolic transformation of 6-MP into 6-TGNs), CL_6-TGNs (6-TGN clearance) and CL_6-mMPNs (6-mMPN clearance) were 0.028, 0.0125 l h⁻¹ and 0.0231 l h⁻¹ with an interindividual variability (%CV) of 69.9 and 38.1% for CL_6-TGNs and CL_6-mMPNs, respectively. The residual variability corresponded to a SD of 0.18 and 9.04 mg l⁻¹ of packed RBC for 6-TGN and 6-mMPN metabolite levels, respectively.

Regression models

Plotting of the conditional estimates of η_i,CL for the metabolites 6-TGNs and 6-mMPNs in paediatric patients with ALL from the BASE model vs. covariates showed some potential relationships. An example is shown in Figure 2.

Figure 2.

Plots of the conditional estimates of η_i,CL6-TGNsvs. body surface area. The solid line indicates the Lowess smooth line

Univariate analyses showing covariates with significant effects on FM₃, CL_6-TGNs and CL_6-mMPNs are presented in Table 2. The addition of several covariates resulted in a reduction of the OBJF of >6.63 (P < 0.01). These covariates were TPMT mutations affecting FM₃, CL_6-TGNs and CL_6-mMPNs parameters, along with weight (WT) and BSA, which affected CL_6-TGNs.

Table 2.

Summary of univariate analysis showing covariate models with significant effects on FM₃, CL_6-TGNs or CL_6-mMPNs of 6-TGN and 6-mMPN metabolites

Effect on	Model	ΔOBJF*	Parameter estimate	P-value
FM3		−8.69		<0.005
	TPMT = 1 in case of mutation, otherwise = 0
CL6-TGNs		−7.95		<0.005
CL6-TGNs		−7.32		<0.01
	TPMT = 1 in case of mutation, otherwise = 0
CL6-mMPNs		−7.15		<0.01
	TPMT = 1 in case of mutation, otherwise = 0
CL6-TGNs		−6.92		<0.01

^*Reduction in the OBJF in comparison with the basic model (BASE). A ΔOBJF ≥6.63 is statistically significant at P < 0.01 for 1 degree of freedom. 6-TGN, 6-thioguanine nucleotide; 6-mMPN, 6-methylmercaptopurine nucleotide; TPMT, thiopurine methyltransferase.

In the model that incorporated weight as a covariate affecting CL_6-TGNs, clearance was standardized using an allometric model that is based on theoretical and empirical evidence that CL is proportional to the ¾ power of weight [12].

The covariate that caused the largest reduction in the OBJF was the dichotomous covariate model incorporating the effect of TPMT mutations on FM₃. Therefore, it was declared as an INTERMEDIATE model, subsequent addition to which of the significant covariates was analysed. Multivariate analysis incorporating both BSA and TPMT mutations as covariates showed that in the presence of the effect of TPMT mutations on FM₃, the effect of BSA on CL_6-TGNs remained significant. The results of multivariate analysis together with the various steps in building the FULL model are presented in Table 3. Backward elimination of any covariate from the FULL model increased the OBJF by >6.63. Therefore, both factors TPMT mutations and BSA were retained in the FINAL model.

Table 3.

Summary of stepwise model building

Model	Effect	ΔOBJF*	P-value
BASE	No covariates included	–	–
INTERMEDIATE	TPMT mutations on FM3	−8.69	<0.005
FULL	TPMT mutations on FM3	−12.85	<0.0005
	BSA on CL6-TGNs

^*Change in OBJF in comparison to preceding model. A ΔOBJF ≥6.63 is statistically significant at P < 0.01 for 1 degree of freedom. TPMT, thiopurine methyltransferase; BSA, body surface area.

Adequate correlations between predicted and observed RBC concentrations of 6-TGNs and 6-mMPNs were observed in the FINAL model, as shown in Figures 3 and 4. The scatterplots show that the population predicted and individual predicted concentrations of 6-TGNs and 6-mMPNs were in reasonable agreement with the measured concentrations around the line of identity, although some underprediction could be observed, particularly at higher concentrations.

Figure 3.

Scatter plots of observed vs. population predicted (A) and individual predicted (B) red blood cell (RBC) concentrations of 6-thioguanine nucleotides (6-TGNs) in the index group of the FINAL model. The solid line represents the line of identity

Figure 4.

Scatter plots of observed vs. population predicted (A) and individual predicted (B) red blood cell (RBC) concentrations of 6-methylmercaptopurine nucleotides (6-mMPNs) in the index group of the FINAL model. The solid line represents the line of identity

Model evaluation

Bias (ME) and precision (RMSE) of the predictive performance of the FINAL model were tested in the validation group (n = 4) between the measured and predicted metabolite concentrations. ME computed values for 6-TGNs and 6-mMPNs were −0.047 and 1.67 mg l⁻¹, respectively. The corresponding RMSE values were 0.16 and 4.84 mg l⁻¹, respectively. Both were less than the residual unexplained variability of the index group, which corresponded to a SD of 0.188 and 9.04 mg l⁻¹ for 6-TGNs and 6-mMPNs, respectively.

The scatter plots of weighted residuals vs. model predicted RBC concentrations of 6-TGNs and 6-mMPNs (Figure 5) showed that both were randomly distributed and the weighted residuals lay within ±2 units of the null ordinate of perfect agreement. Specific examples of the predictive capability of the final model are shown for two representative individual patients in the validation group in Figure 6, which shows the time course of measured and post hoc predicted RBC concentrations of 6-TGNs and 6-mMPNs.

Figure 5.

Scatter plots of weighted residuals vs. predicted concentrations of 6-thioguanine nucleotides (6-TGNs) (A) and 6-methylmercaptopurine nucleotides (6-mMPNs) (B) in red blood cells (RBC)

Figure 6.

Longitudinal assessment of the predictive performance of the FINAL model in two representative patients from the validation dataset: (A) 12-year-old boy and (B) 6-year-old girl. (▪) Observed and (•) model-predicted 6-thioguanine nucleotide (6-TGN) concentrations. (▴) Observed and (▾) predicted 6-methylmercaptopurine nucleotide (6-mMPN) concentrations

Final population pharmacokinetic model

The computed population structural parameter estimates, the interpatient variability (CV%) and the residual variability obtained by fitting the full dataset to FINAL model are presented in Table 4.

Table 4.

Population pharmacokintics of 6-TGNs and 6-mMPNs, obtained by fitting the full dataset (19 children) to the BASE and FINAL models

Parameter	Symbol	Base model Estimate	ω (CV%)	Final model Estimate	ω (CV%)	SE (%)
FM3		0.0295	–	0.0191	–	64.1
CL6-TGNs (l h−1)		0.0138	65.7	0.00914	33.6	56.8
CL6-mMPNs (l h−1)		0.0227	33.9	0.0228	33.1	14.2
TPMT	θTPMT	–	–	2.56	–	35.9
BSA	θBSA	–	–	1.16	–	49.0
ωCL6-TGNs		0.431		0.113		81.2
ωCL6-mMPNs		0.115		0.11		82.6
σ6-TGNs (mg l−1)		0.171		0.177		17.5
σ6-mMPNs (mg l−1)		9.04		8.42		19.3

Structural models:

Random effects models:

where FM₃ is the fractional transformation of 6-MP into 6-TGNs; TPMT, 1 if the patient had TPMT mutation and 0 otherwise; CL_6-TGNs is 6-TGN clearance; BSA is body surface area in m²; CL_6-mMPNs is 6-mMPN clearance; ω is the interindividual variability; η is squared ω; σ_6-TGNs and σ_6-mMPNs are the residual variabilities of 6-TGNs and 6-mMPNs. 6-TGN, 6-thioguanine nucleotide; 6-mMPN, 6-methylmercaptopurine nucleotide; TPMT, thiopurine methyltransferase.

The final population model for 6-TGNs and 6-mMPNs in RBC was:

where BSA is body surface area in m² and TPMT = 1 in case of mutation, otherwise = 0.

For a hypothetical individual with population median value of BSA (i.e. 1.14 m²), the model-predicted FM₃, CL_6-TGNs and CL_6-mMPNs would be 0.019, 0.011 l h⁻¹ and 0.0288 l h⁻¹, respectively. The FM₃ would increase by 256% to 0.049 if the patient had a TPMT mutation. The model for CL_6-TGNs found that CL_6-TGNs was proportional to the 1.16 power of BSA, resulting in an estimated range of 0.0049–0.02 l h⁻¹ across the BSA range of 0.59–2.0 m² among the study group.

The median individual Bayesian estimates for FM₃, CL_6-TGNs, CL_6-mMPNs and BSA normalized CL_6-TGNs values obtained by fitting the full dataset from the study population to the FINAL model are presented in Table 5. These estimates are determined as a postprocessing step using the measured concentrations, in contrast to the population parameter estimates, which are derived from covariate information. Since the individual Bayesian estimates are drawn from a distribution where the population estimates reflect the posterior mode of the marginal likelihood distribution for that parameter, a possibility for differences in this summary parameter could be introduced. In order to evaluate the predictive performance of the final model obtained from the full dataset, a posterior predictive check was performed (Figure 7).

Table 5.

Individual Bayesian estimates obtained from the FINAL population model

Parameter	Median (P5, P95)
FM3 (no TPMT mutation)	0.0191
FM3 (with TPMT mutation)	0.0491
CL6-TGNs (l h−1)	0.0112 (0.0046, 0.0235)
CL6-mMPNs (l h−1)	0.0226 (0.0153, 0.0310)
CL6-TGNs (l h−1 m−2)*	0.0082 (0.0059, 0.0151)

P₅, 5th percentile; P₉₅, 95th percentile.

^*Calculated using the body surface area of each individual. 6-TGN, 6-thioguanine nucleotide; 6-mMPN, 6-methylmercaptopurine nucleotide; TPMT, thiopurine methyltransferase.

Figure 7.

Visual predictive check of the final model fitted to the full dataset (n = 19 patients). A plot of the time course of the observed concentrations of 6-thioguanine nucleotides (6-TGNs) (A) and 6-methylmercaptopurine nucleotides (6-mMPNs) (B) along with the median and 90% prediction intervals for the simulated values. Median prediction (------); 90% prediction interval (—); Observed concentrations (•)

The interpatient variability (CV%) for the population pharmacokinetic parameters of CL_6-TGNs and CL_6-mMPNs were 33.6 and 33.2%, respectively. The interpatient variability of CL_6-TGNs was reduced from 65.6 to 33.6% by the inclusion of BSA covariate, which explained most of the variability in CL_6-TGNs between individuals. The residual unexplained variability (SD) was 0.177 and 8.42 mg l⁻¹ for 6-TGNs and 6-mMPNs, respectively, which translates to a CV% of 24.3 and 57.6% at the mean RBC concentrations of 6-TGNs and 6-mMPNs measured in the full dataset (0.729 and 14.61 mg l⁻¹, respectively).

Discussion

Current empirical dosing methods for oral 6-MP result in highly variable drug and metabolite concentrations, as demonstrated by the present study and other investigators [13, 14]. The variability presumably arises in part from individual differences in bioavailability. Other factors, however, that could contribute to this variability were evaluated in the present study for paediatric patients with ALL using a population pharmacokinetic modelling approach. The different factors studied were age, gender, WT and BSA, along with various genetic factors such as polymorphisms in XO, TPMT and ITPA enzymes.

Quantitative evaluation of pharmacokinetic parameters of 6-MP seems especially attractive and is a potentially important prognostic factor in cancer chemotherapy [15]. In addition, the study of drug distribution is of particular relevance for paediatric patients, given the effect of maturational changes on organ function and body composition that can affect drug disposition [16]. For these reasons, we developed in this study, for the first time, a population pharmacokinetic model for 6-MP and its metabolites for paediatric patients with ALL.

The pharmacokinetic model developed revealed considerable IIV in the clearance of both 6-MP metabolites investigated, 6-TGNs and 6-mMPNs. This variability, however, coincides with the highly variable RBC concentrations of 6-TGNs and 6-mMPNs reported previously in children taking identical doses of 6-MP [13, 14]. The estimated IIV (CV%) of clearance in the BASE model, fitted to the index group, was 69.9 and 38.1% for 6-TGNs and 6-mMPNs, respectively. The model, however, showed large reduction in the IIV of 6-TGN clearance (69.9–29.3%) and a significant reduction in the OBJF (–7.95, P < 0.005) when BSA alone was incorporated into the model as a covariate affecting 6-TGN clearance. This indicated that large part of the IIV in 6-TGN clearance was explained by differences in BSA.

Both the WT and BSA had significant effects on 6-TGN clearance in this study. BSA, however, was found to be a better predictor of clearance than WT. The final model obtained by fitting the full dataset indicated that CL_6-TGNs increased by 116% per 1-m² increase in the BSA resulting in an estimated range of 0.0049–0.02 l h⁻¹ across the BSA range of 0.59–2.0 m² among the study group.

Clinical experience also confirms the effect of impaired renal function on the concentration of 6-MP metabolites and its clinical outcome. It has been shown that patients with impaired renal function have increased susceptibility to 6-MP side-effects. In the present study, none of the patients had clinically significant renal impairment and hence renal function was mainly proportional to patient's BSA.

The rationale behind the use of BSA as a criterion for dosing in anticancer chemotherapy was outlined about 50 years ago [17], giving rise to the practice of using BSA for dosing anticancer therapy. The current approach for dosing anticancer drugs is to administer a standard dose, which is normalized to BSA, and then to adjust or individualize subsequent doses based on the severity of drug toxicity. Even though this clearly has practical and economic implications, its clinical value has, in recent years, been questioned [18]. One objection to the use of BSA in measuring drug dosage was the difficulty in measuring BSA. Apart from the inaccuracies inherent in methods for BSA calculation [19], they are dependent on the accuracy of weight and height measurements used in BSA calculation. Another alternative was the use of normograms to avoid inaccurate determination of BSA. However, the reliability of these normograms tends to differ [20].

Another objection was that pharmacokinetic studies of anticancer drugs revealed substantial interpatient variability in plasma drug concentrations when the dose was based on BSA [13, 21]. The potential consequences of this variability in systemic drug exposure are life-threatening toxic effects in patients who are exposed to excessive drug concentrations and tumour progression in patients who achieve subtherapeutic drug concentrations. The identification of specific factors that account for variability in drug disposition among patients can, therefore, lead to more rational dosing methods.

In the present study, the most influential covariate examined was TPMT genotype on the fraction of metabolic transformation, FM₃. Its inclusion in the model resulted in the greatest reduction in OBJF (ΔOBJF = −8.69, P < 0.005). Hence, it explained part of the variability in FM₃ between individuals. In addition, the final model predicted an increase in FM₃ of 256% if the patient had a TPMT mutation. This would mean preferential 6-TGN production and 6-mMPN underproduction if the patient had a TPMT mutation. This is in perfect agreement with previous studies, which demonstrated an inverse correlation between TPMT activity and the levels of 6-TGNs in RBC, suggesting that an inherited decrease in the methylation step (due to the TPMT mutation) results in shunting 6-MP metabolism away from 6-mMPNs towards overproduction of 6-TGNs [22, 23]. More importantly, the optimal dose of 6-MP, which was determined in the standard empirical fashion, was also related to TPMT genotype [22, 24].

No prior studies, however, have provided specific 6-MP dosing guidelines based on TPMT genotype. In the present study, for the first time, the potential utility of TPMT genotype in prospectively defining the percentage of 6-MP converted to 6-TGNs or 6-mMPNs as predicted by FM₃ was illustrated. This would identify patients who are less tolerant of 6-MP standard doses based on their capacity to metabolize the drug. One limitation of the developed model, however, was its inability to account for variation in TPMT activity among patients having the wild-type. This could be the reason for underprediction observed in the model for 6-mMPN concentrations at higher levels. Higher concentrations of 6-mMPNs are probably due to high TPMT activity that could not be accounted for by the model. Phenotyping patients, however, by measuring TPMT enzymatic activity could resolve this limitation and allow for individualization of drug doses among the whole population.

The other genetic covariates tested in this study (ITPA, XO mutations) were not statistically significant and therefore not included in the FINAL model. It should be noted, however, that the IIV not explained by the model might have resulted, in part, from these covariates. It is also possible that the small number of patients studied (n = 19) was not enough to identify more potential sources of IIV. Therefore, the developed population model will need to be tested and refined by larger prospective studies that have the potential to encounter and quantify more covariates.

The reasonably large residual (unexplained) variability in the model might be due to large IIV in the pharmacokinetic parameters, interoccasional variability, assay errors, or model misspecification. Since the data in most patients were collected for a period of several months, this might have added to the variability within the same patient. Moreover, the significant IIV in drug absorption may explain some of the variability obtained with the orally administered 6-MP [25]. Non-adherence to therapy could also have contributed to this variability. Children with ALL who have reached remission always remain under a prolonged and complex course of chemotherapy in spite of being practically asymptomatic. These factors are likely to contribute to missing some doses given the home therapy of 6-MP and the absence of obvious consequences at the time. Therefore, non-adherence is somehow expected in this group of patients and should be considered.

Although the robustness of the pharmacokinetic model developed in this study was validated internally using the data-splitting technique, it is highly recommended that the model be subjected to external validation before using it to guide dosage adjustments. If successful, this would offer a more rational dosing approach than the traditional empirical method, since it would combine the current practice of using BSA in 6-MP dosing with a pharmacogenetically guided dosing based on TPMT genotype.

Competing interests

None declared.