Population Pharmacokinetics of Crenolanib in Children and Young Adults with Brain Tumors

Abstract

Purpose:

Crenolanib, an oral inhibitor of platelet-derived growth factor receptor, was evaluated to treat children and young adults with brain tumors. Crenolanib population pharmacokinetics and covariate influence were characterized in this patient population.

Methods:

Patients enrolled on this phase I study (NCT01393912) received oral crenolanib once daily. Serial single-dose and steady state serum pharmacokinetic samples were collected and analyzed using a validated LC-ESI-MS/MS method. Population modeling and covariate analysis evaluating demographics, laboratory values, and co-medications were performed. The impact of significant covariates on crenolanib exposure was further explored using model simulations.

Results:

Crenolanib serum concentrations were analyzed for 55 patients (2.1–19.2 years-old) and best fitted with a linear two-compartment model, with delayed absorption modeled with a lag time. A typical patient (8-year-old, body surface area [BSA] 1 m²) had an apparent central clearance, volume, and absorption rate of 41 L/h, 54.3 L, and 0.19 /h, respectively. Patients taking acid reducers (histamine H₂ antagonists or proton pump inhibitors) concomitantly exhibited about 2- and 1.7-fold lower clearance and volume (p<0.0001 and p=0.018, respectively). Crenolanib clearance increased with BSA (p<0.0001), and absorption rate decreased with age (p<0.0001). Model simulations showed co-treatment with an acid reducer was the only covariate significantly altering crenolanib exposure and supported the use of BSA-based crenolanib dosages vs flat-dosages for this population.

Conclusions:

Crenolanib pharmacokinetics were adequately characterized in children and young adults with brain tumors. Despite marked increased drug exposure with acid reducer co-treatment, crenolanib therapy was well-tolerated. No dosing adjustments are recommended for this population.

Introduction:

Pediatric high-grade glioma (pHGG) are highly aggressive heterogenous tumors with a two-year survival outcome of less than 20% (1, 2). As much as half of pediatric HGG tumors are diffuse intrinsic pontine glioma (DIPG), a tumor of the pons in the brain stem controlling vital body functions such as vision, speech, and heart rate (1–3). DIPG is uncommon in adults, but is usually diagnosed in young children, and remains the most common cause of death in children with brain tumors with a median two-year survival of less than 10% (1, 4).

Standard of care therapy for pHGG includes maximal safe surgical resection, when feasible, and focal irradiation. Chemotherapy is frequently employed in lieu of radiation therapy in very young children, and investigational systemic therapy is often administered after irradiation in older children and young adults (2). However, currently available chemotherapeutic agents have shown limited consistent survival benefit.

Whole-genome profiling analyses has revealed platelet derived growth factor receptor α (PDGFRA) as a predominant target of amplification and activating somatic mutation in both pHGG and DIPG (3–6). PDGFR is a family of cell-surface receptor tyrosine kinases that contribute to cell proliferation, embryonic development, and promote the maturation of mesenchymal cell types in different organs (7, 8). Dysregulated PDGFR signaling can lead to tumorigenesis through the dysregulation of multiple downstream pathways such as PI3K/AKT/mTOR pathway, also frequently altered in pediatric glioma (9, 10). As such, PDGFR has emerged as a promising therapeutic target for pHGG and DIPG.

Crenolanib is an orally bioavailable benzimidazole that potently and selectively inhibits PDGFR α/β, as well as the FMS-related tyrosine kinase 3 (FLT3) internal tandem duplication (11–13). Preclinical studies have demonstrated the anti-angiogenic and antitumor properties of crenolanib in different disease models: gastrointestinal stromal tumors, non-small-cell lung cancer, systemic sclerosis, and acute myeloid leukemia (14–17). Clinical efficacy of crenolanib was evaluated in adults with solid tumors in phase I and II trials (11, 15). When crenolanib was dosed at between 100–340 mg once daily, the drug was well tolerated, exhibited a slow absorption profile, and a terminal half-life of 14±4.2 hours which was similar across all dose levels.

This report presents the population pharmacokinetic data from the first pediatric phase I trial of crenolanib, developed in children and young adults with newly diagnosed DIPG or recurrent HGG (SJPDGF; NCT01393912). Population modeling and covariate analysis were performed to characterize crenolanib disposition, quantify the inter- and intra-patient variability, and identify patient covariates with significant impact on drug exposure.

Material and Methods

Study design

The patients included in this study were treated on a phase I clinical trial (SJPDGF; NCT01393912) conducted at St. Jude Children’s Research Hospital, evaluating oral crenolanib therapy in children and young adults with newly diagnosed DIPG and recurrent HGG, including DIPG. The details of the clinical trial have been reported elsewhere (18). The Institutional Review Board approved this trial and written informed consent was obtained from patients, parents, or legal guardians before the onset of any study procedures. Patients with newly diagnosed DIPG were enrolled on Stratum A and received oral crenolanib once daily with concomitant radiation therapy during the first 6 weeks of therapy, followed by continuous once daily crenolanib for up to two years. Patients with recurrent, progressive, or refractory HGG were enrolled on Stratum B and received continuous once daily crenolanib for up to two years. In both strata, the following dose escalations were evaluated: 100, 130, 170, and 220 mg/m².

Blood sampling and bioanalysis

Serial pharmacokinetic studies were conducted on all patients enrolled on this protocol during the first crenolanib course on days 1 and 28 (±3 days) to characterize single-dose and steady state drug disposition. On day 1, serial crenolanib samples were collected at pre-dose and at 1, 2, 4, 8 (±2), 24 (±6), and 48 (±6) hours after dose. The crenolanib dose on day 2 was held during the pharmacokinetic studies. On day 28, serial crenolanib samples were collected pre-dose, and at 1, 2, 4, 8 (±2), and 24 (±6) hours post-dose. Unpublished in vitro data showed that crenolanib was primarily metabolized by human cytochrome CYP3A enzyme family. Most patients were expected to receive dexamethasone, a potent CYP3A4 inducer, as concomitant treatment; thus, additional pharmacokinetic studies were performed. In consenting patients taking dexamethasone during course 1, serial crenolanib samples were collected pre-dose, and at 1, 2, 4, 8 (±2), and 24 (±6) hours post-dose, approximately two weeks after discontinuation of dexamethasone, in a subsequent course.

At each time point, whole blood (2 mL) was collected in BD vacutainer serum collection tubes, allowed to sit at room temperature for 30 minutes, then centrifuged for 10 minutes at 4°C at 1,500 ×g. The extracted serum was stored at −80°C. Serum crenolanib concentrations were determined using a validated LC-ESI-MS/MS method with a lower limit of quantitation (LLOQ) of 5 ng/mL (11 nM) (19).

Population pharmacokinetic modeling

Course 1 day 1, day 28, and off-dexamethasone crenolanib serum concentration-time data were analyzed using nonlinear mixed-effects modeling, which permitted characterizing the typical crenolanib pharmacokinetics and quantifying the variability among patients (20). The model characterizing crenolanib disposition was based on empiric compartments described using differential equations including pharmacokinetic parameters of interest. Each parameter was assumed to take a typical population value with variances describing inter-individual and inter-occasion variabilities. The model parameters were assumed to follow a log-normal distribution. The Stochastic Approximation Expectation Maximization algorithm in Monolix (v2019R2. Antony, France: Lixoft SAS, 2019. http://lixoft.com/products/monolix/) was used to estimate the parameters. Inter-individual and inter-occasion variabilities were assumed to be normally distributed and implemented on the model-parameters using an exponential model. Additive, proportional, and combined residual error models were tested to quantify the unexplained variability of the model. Error terms were assumed to be normally distributed.

Different model structures were tested to describe the crenolanib pharmacokinetics: zero or first order absorption, one, two or three compartmental distribution, and linear or non-linear elimination. Lag time or transit compartments were also explored to describe a possible delay in absorption. In the absence of intravenous data, volume and clearance parameter estimates were apparent. Both inter-patient and inter-occasion variabilities were tested on each model-parameter with a forward/backward stepwise approach, as well as correlations between the pharmacokinetic parameters. Dose amount (nmol) was used as model input. Data below the LLOQ were censored according to the Beal method M3 (21). Model selection was based upon changes in the minimum objective function value (OFV), precision of parameter estimates (relative standard error), and visual inspection of goodness-of-fit plots.

Covariate analysis

Once the population pharmacokinetic model was selected, a covariate analysis was performed to determine patient characteristics that influence the pharmacokinetic parameters. Tested covariates included: age, total actual bodyweight, height, body surface area (BSA) (22), gender, stratum (A/B), concomitant medications, serum albumin, alkaline phosphatase, alanine-aminotransferase, aspartate aminotransferase, total bilirubin, serum creatinine, and hematocrit. Concomitant medications encompassed dexamethasone (Yes/No) and acid reducers (Yes/No). Histamine H-2 receptor antagonists (i.e., ranitidine and famotidine), and/or proton pump inhibitors (i.e., omeprazole, lansoprazole, and pantoprazole) were used as acid reducers in this study and were tested as two individual covariates and one combined covariate. The values of each covariate reported on days 1 and 28 were incorporated into the dataset. The continuous covariates were modeled according to an exponential model, or a power model scaled to the population median covariate value. The best model was selected based upon the objective function value, the extent of parameter variability explained by the addition of the covariate, and the precision of the parameter estimate. Categorical covariates were modeled according to an exponential shift model.

The individual model-derived parameters were plotted against the covariate values to assess potential relationships to further evaluate, using the stepwise forward inclusion and backward exclusion method (23). Each covariate was added univariately to the parameter of interest and selected based upon a decrease in the OFV by at least 3.84 units (corresponding to p < 0.05 based on the χ² test), a decrease in the inter-individual and/or inter-occasion variability, and improved model fits. The covariate displaying the highest change in OFV was retained in the model and the process was repeated until no covariates were statistically significant. Then, backward elimination was performed, where each covariate was removed individually and retained in the model only if the OFV increased by at least 6.63 (X²-test, p-value < 0.01). This process was repeated until all remaining covariates were significant.

Model evaluation

The precision of the parameters estimated from the final population model was evaluated using a nonparametric bootstrap resampling method (24). A total of 500 replicates of the original dataset were generated using random sampling with replacement. The model was applied to each replicate dataset. Summary statistics for each parameter were calculated based upon the model-runs (i.e., mean and 90% bootstrap confidence interval). Non-parametric bootstrapping was performed with R-based Rsmlx 2.0.4 package (R speaks ‘Monolix’, http://rsmlx.webpopix.org).

The predictive performance of the final model was evaluated using prediction-corrected visual predictive checks (25). A total of 1000 dataset replicates simulated conditionally on the final model parameters. The observed data were overlaid on the 5^th, 50^th, and 95^th percentiles of the model-simulations to visually assess concordance between the observed and model-based simulated data. All the data were corrected using the median of the population prediction within the associated time bin.

Model simulations

Model simulations were performed to further explore the extent of the effects of significant identified covariates on crenolanib exposure, defined by area under the concentration-time curve (AUC), and maximum concentration (C_MAX). The final population parameter estimates were used to simulate crenolanib exposures after single and repeated dosages at the maximum tolerated dosage defined at 170 mg/m²/d (18), for a virtual population (N=500) with the same characteristics as the studied population and for one typical individual (8 years old, BSA 1 m², with no concomitant treatment). Specific virtual patients were created by assigning combinations of extreme and nominal values of the significant covariates included in the model. Crenolanib exposures were simulated for these specific virtual patients using the model (without any variability) after single and repeated dosages (170 mg/m²/d) to explore the change in drug exposure from the typical individual within the range of covariate values. Similar simulations were also performed after a single flat dose of 170 mg crenolanib to specifically assess the impact of BSA on crenolanib exposures, and the need for BSA-based crenolanib dosing for the pediatric population.

Results

Data summary

A total of 55 patients were enrolled in this study and participated in the Course 1 Day 1 pharmacokinetic studies. Steady state data on Course 1 Day 28 were missing from 9 patients due to tolerance issues or disease progression leading to treatment interruption. Most patients were given dexamethasone concomitantly with crenolanib therapy (i.e., 30/55 on Day 1 and 32/46 on Day 28). However, off-dexamethasone pharmacokinetic data were only collected from 12 patients in a subsequent course. The patient population age ranged from 2.1 to 19.2 years old, and 76.4% of patients were Caucasian. All the patient characteristics, including the clinical covariates and concomitant medications of interest are summarized in Table 1.

Table 1.

Characteristics of patients included in the pharmacokinetic studies

Patient characteristics (N = 55)
Patients enrolled on PK studies n
Day 1	55
Day 28	46
Post-dexamethasone	12
Female / male n (%)	29 (53) / 26 (47)
†Ethnicity n (%)
White	42 (76.4)
Black	7 (12.7)
Other	6 (10.9)
Stratum A / B n (%)	32 (58) / 23 (42)
Crenolanib dosage n (%)
100 mg/m2	15 (27.3)
130 mg/m2	12 (21.8)
170 mg/m2	13 (23.6)
220 mg/m2	15 (27.3)
Age (y)	8.13 (2.1–19.2)
Height (cm)	126 (85.8–174)
Weight (kg)	28.2 (11.2–98.2)
Body surface area (m2)	1.0 (0.54–2.2)
Albumin (g/dL)	4.2 (3.3–5.1)
Alkaline phosphatase (U/L)	113 (42–413)
Aspartate aminotransferase (U/L)	22 (12–38)
Alanine aminotransferase (U/L)	19 (5–84)
Bilirubin (mg/dL)	0.2 (0.1–1)
Creatinine (mg/dL)	0.4 (0.2–0.8)
Concomitant medication on day 1 / day 28 n (%)
Dexamethasone	30 (54.5) / 32 (69.6)
Acid reducer – H2-antagonists	23 (42) / 25 (54.3)
Acid reducer – Proton pump inhibitors	5 (9) / 9 (19.6)

Data are reported as frequency or median (range).

^†Self-declared ethnicity. Other include Asian (n=1) and unknown (n=5)

A total of 718 crenolanib serum concentration-time points were collected. Wide interpatient variability was observed in the concentration-time data after both single dose and at steady state drug administration. About 7% of the concentrations were found below LLOQ, among which 67% were samples collected after 8 hours post-dose.

Population pharmacokinetic and covariate analysis

Crenolanib serum concentration-time data were best described by a two-compartment model with first-order absorption delayed with a lag time parameter, and first-order elimination (Fig. 1). The model was described with the following parameters: lag time (T_lag), absorption rate constant (k_a), apparent central clearance (CL/F) and volume (V1/F), apparent peripheral clearance (Q/F) and volume (V2/F) volume. The unexplained residual variability was described by a proportional error model. No correlation between parameters was significant. Before including any covariates, the final population estimates for T_lag, k_a, CL/F, V1/F, Q/F and V2/F were 0.59 hr, 0.13 /h, 20.4 L/h, 34.6 L, 4.47 L/h, and 316 L, respectively. Inter-individual variability was estimated on all parameters except V2/F, and inter-occasion variability was associated with T_lag, CL/F, and V1/F. Associated shrinkages calculated based on a ratio of variances were all below 55%, except for Q and V2/F (79% and 75%, respectively); thus, covariates were not tested on peripheral parameters.

Fig. 1.

Crenolanib pharmacokinetic model structure

Crenolanib serum concentrations were described using a two-compartment with delayed first-order absorption and first-order elimination with apparent central clearance (CL/F) and volume (V1/F), apparent peripheral clearance (Q/F) and volume (V2/F), absorption rate constant (k_a) and absorption lag time (T_lag).

An exploratory covariate analysis showed potential associations between individual pharmacokinetic parameters and all evaluated covariates except patient stratum, gender, hematocrit, and serum creatinine. The stepwise covariate analysis further identified three covariates with significant impact on crenolanib pharmacokinetics: BSA, age, and acid reducer concomitant treatment. Crenolanib CL/F and V1/F were significantly lower in patients receiving concomitant acid reducers (2-fold lower p<0.0001 and 1.7-fold lower p=0.018, respectively), as shown in Fig. 2a and 2b. Acid reducer co-treatment was included in the model as one categorical covariate encompassing the use of histamine H-2 receptor antagonists and/or proton pump inhibitors. Crenolanib CL/F increased with BSA according to a power model centered to the median BSA value with a coefficient of 1.03 (p<0.0001), suggesting a linear correlation (Fig. 2c). Last, the absorption rate k_a decreased with patient age (p<0.0001), and that was best described with an exponential model (Fig. 2d). With the inclusion of significant covariates, the interpatient variability (standard deviation) of crenolanib CL/F, V1/F, and k_a decreased by 29.7% (from 0.516 to 0.363), 46% (from 0.407 to 0.246), and 39.8% (from 0.186 to 0.112), respectively. Additionally, the inter-occasion variability associated with crenolanib CL/F and V1/F decreased by 11.8% (from 0.568 to 0.510) and 13.7% (from 0.956 to 0.825), respectively. The final equations describing crenolanib CL/F, and V1/F, and k_a are reported below:

$$CL/F_i=CL/F_{\mathit{\text{pop}}}\cdot {\frac{BSA}{\mathit{\text{median}}}}^{{\beta }_{BSA}}\cdot e^{AR\cdot {\beta }_{AR-CL}}$$

$$V1/F_i=V1/F_{\mathit{\text{pop}}}\cdot e^{AR\cdot {\beta }_{AR-V}}$$

$$k_{a,i}=k_{a,\mathit{\text{pop}}}\cdot e^{Age\cdot {\beta }_{Age}}$$

P_i and P_pop terms represent the individual and population (typical) parameters, and β terms are the estimated covariate coefficients. AR was assigned a value of zero or 1 if acid reducer cotreatment was used. The final population and random effect parameter estimates are reported in Table 2. The relative standard errors associated to the parameter estimates were below 50%, except for the population estimate of V2/F, and the inter-individual variability of V1/F. Removal of the inter-individual terms associated to those parameters was tested, but significantly impaired the estimation of other parameters and increased the residual error thus, no change was implemented. Despite the inclusion of significant covariates, large unexplained inter-individual and inter-occasion variabilities remained in the model.

Fig. 2.

Pharmacokinetic parameter and covariate associations. Boxplots of crenolanib apparent clearance (a) and volume (b) vs acid reducer concomitant treatment (Yes/No). Scatterplot of crenolanib apparent clearance and body surface area (c). Scatterplot of absorption rate constant ka and patient age (d). Green color indicates data from patients taking concomitant acid reducers.

Table 2.

Final pharmacokinetic population parameter estimates

Parameter	Final model	Non-parametric bootstraps
	Estimate (RSE%)	Mean	90% CI
Population estimates
Absorption lag time Tlag (h)	0.525 (14.4)	0.547	0.404; 0.668
Absorption rate constant ka (/h)	0.191 (10.4)	0.192	0.165; 0.230
Age coefficient on ka βAge	−0.0404 (20.5)	−0.0403	−0.0567; −0.0260
Central clearance CL/F (L/h)	41 (12.2)	42.0	35.2; 52.6
AR coefficient on CL/F βAR-CL	−0.701 (18.8)	−0.687	−0.917; −0.480
BSA coefficient on CL/F βBSA	1.03 (24.6)	1.09	0.57; 1.60
Central volume V1/F (L)	54.3 (18.4)	54.0	35.7; 78.6
AR coefficient on V1/F βAR-V	−0.516 (42.3)	−0.477	−0.866; −0.061
Peripheral clearance Q/F (L/h)	3.84 (32.5)	3.09	1.75; 5.12
Peripheral volume V2/F (L)	412 (86.7)	194	66.3; 694
Inter-patient variabilities (IIV)
Absorption lag time Tlag	0.524 (27.7)	0.496	0.119 – 0.761
Absorption rate constant ka	0.112 (37.7)	0.091	0.026 – 0.181
Central clearance CL/F	0.363 (25.4)	0.367	0.117 – 0.521
Central volume V1/F	0.246 (124)	0.182	0.062 – 0.539
Peripheral clearance Q/F	0.536 (20.2)	0.506	0.271 – 0.788
Inter-occasion variabilities (IOV)
Absorption lag time Tlag	0.645 (18.3)	0.591	0.379 – 0.802
Central clearance CL/F	0.510 (13.9)	0.532	0.404 – 0.647
Central volume V1/F	0.825 (14.2)	0.860	0.615 – 1.08
Residual proportional error	0.304 (4.44)	0.306	0.28 – 0.33

RSE% relative standard errors, AR acid reducer concomitant treatment, BSA body surface area

IIV and IOV are reported as standard deviation.

The diagnostic plots of the final pharmacokinetic model including covariates are reported in Fig. 3. The goodness-of-fit plots based on the model observations and predictions did not show any significant bias or model misfits (Fig. 3a–d). The prediction-corrected visual predictive checks based on model simulations showed that the central tendency and the variability of the crenolanib data after both single dose and at steady-state were accurately described by the final population pharmacokinetic model (Fig. 3e–f). The precision of the parameter estimates was further assessed using non-parametric bootstraps. The estimated parameters were close to the mean of the bootstrap estimates and none of the 90% confidence intervals included a value of zero (Table 2).

Fig. 3.

Diagnostic plots for the population pharmacokinetic model. Observed concentration-time data vs population (a) and individual model predictions (b). Individual weighted residuals (IWRES) vs time (c) and model predictions (d). In panels (a) and (b), solid lines represent the unity line. Red crosses are data below limit of quantification. Prediction-corrected visual predictive checks for single-dose (e) and steady state data (f). Circles are observed data, crosses are data below limit of quantification, the solid and dashed lines depict the model-predicted and empirical (observed) 5^th, 50^th, and 95^th percentiles, respectively, and the shaded areas represent the 90^th confidence interval around the model-predicted percentiles. Green color indicates data from patients taking concomitant acid reducers.

Identified covariates and crenolanib exposure

Model-based simulations were performed to assess the impact of the identified covariates on crenolanib overall exposure. The influence of covariates on crenolanib AUC_0−∞ and C_MAX after a single dose of 170 mg/m² crenolanib is shown in Fig. 4. Co-treatment with acid reducers resulted in ~2-fold increased AUC_0−∞ and 1.8-fold increased C_MAX relative to the nominal patient not taking this co-medication. The effects of age and BSA were negligible on crenolanib AUC_0−∞, and remained minimal on C_MAX, within the range of covariate values observed in the population, after a BSA-adjusted dose. Model simulations with a flat dose (170 mg) showed that in patients with the extreme values of BSA observed in the population (i.e., 0.54 and 2.21 m²), the range of crenolanib AUC_0−∞ and C_MAX were 45 to 185%, and 53.5 to 151%, relative to the nominal patient with a median BSA of 1 m². These results highlighted the need for BSA-adjusted crenolanib dosages for children and young adults.

Fig. 4.

Forest plots of crenolanib area under the concentration curve AUC_0−∞ (a) and maximum concentration C_MAX (b) using the final model for covariate evaluation. Vertical lines depict crenolanib AUC_0−∞ and C_MAX for a typical patient with nominal covariate values. Horizontal bars show the effect of single covariate in the range of the observed covariate after 170 mg/m² and flat 170 mg dosing.

Model simulations were also performed after repeated crenolanib doses (170 mg/m²/d and 170 mg/d) and showed similar results (data not shown). Between single-dose and steady state dosing, assuming no changes in co-medications, small accumulation was observed in crenolanib exposure. Mean crenolanib AUC_0–24h after single dose and at steady state were determined at 8947 and 9668 h·nM in patients not taking co-medication, and at 16,190 and 18,605 h·nM in patients taking acid reducers.

Discussion

The pharmacokinetics of crenolanib, a selective PDGFR α/β inhibitor, was characterized for the first time in children and young adults with high-grade glioma. Crenolanib disposition was described using a linear two-compartment model with delayed absorption. The crenolanib absorption rate and apparent oral clearance were influenced by patient age and BSA, respectively. Co-treatment with acid reducers significantly increased crenolanib exposure by decreasing both drug clearance and volume. Model simulations support the use of BSA-adjusted dosages versus flat dose in pediatrics. No other dosing adjustment was considered relevant in this population.

Only a few pharmacokinetic studies in adults for crenolanib (previously named CP865,696) have been published. In the first-in-human study, adult patients with advanced cancers showed linear but highly variable crenolanib pharmacokinetics between 60 and 340 mg once or twice daily (11). The mean terminal half-life ranged from 12.3 to 18.5 hours. After 280 mg once daily (which approximately corresponds to 168 mg/m²), the mean (CV%) steady state crenolanib AUC_0–72h was 25,928 (92%) h·nM. In adult patients with newly diagnosed FLT3 mutant acute myeloid leukemia, crenolanib exhibited a 6.8 hour terminal half-life, minimal drug accumulation after repeated doses, and a mean clearance of ~60 L/h (range 10.7 to 592 L/h) similar between day 1 and day 15 of crenolanib therapy (26). Overall, crenolanib pharmacokinetics characterized in our pediatric population was similar compared to the adult population. In our study, mean crenolanib clearance was 38.5 L/h (58.8 L/h without acid reducers and 26.1 L/h with acid reducers), minimal accumulation was observed at steady state, and mean steady state AUC_0–24h after 170 mg/m² ranged from 9,476 to 68,508 h·nM with a mean of 20,725 h·nM.

Our covariate analysis identified three covariates with a significant influence on crenolanib pharmacokinetic parameters. Crenolanib absorption rate decreased with increasing age within the range 2.1–19.2 years old. This may be related to CYP3A4 ontogeny or result from different developmental factors related to the gastrointestinal tract which are altered as children age (27). For instance, in children compared to adults, intestinal transit time is shorter, gastric pH is lower, and intestinal bile concentration is lower. These are all factors than can impact drug absorption rate (28). It is important to note that if patient age was significantly correlated with ka, overall, it did not have a significant impact on crenolanib AUC_0−∞, and only a moderate impact on C_MAX. Therefore, further age-based dosing adjustments would not be relevant.

Co-treatment with acid reducers had considerable impact on crenolanib exposure (AUC_0−∞ and C_MAX), by significantly decreasing both apparent clearance and volume (i.e., by 2-fold and 1.7-fold), respectively. Here, acid reducers encompassed both histamine H-2 receptor antagonists (ranitidine and famotidine), and proton pump inhibitors (omeprazole, lansoprazole, and pantoprazole); however, ranitidine was by far the main acid reducer used in this population. By suppressing gastric acid secretion and increasing stomach pH from pH ~1 to pH ~4, acid reducing agents can significantly alter the dissolution, absorption, and bioavailability of oral drugs that have a pH-dependent solubility (29). For weak base drugs, with an acid dissociation constant (pKa) less than 4, coadministration with acid reducers typically decreases the drug dissolution and absorption, and thus, decreases drug systemic exposure. This has been observed for other tyrosine kinase inhibitors, such as dasatinib and crizotinib (29, 30). However, weak acid drugs, such as crenolanib which has a pKa ~9.8, are more soluble in neutral pH than in acidic pH fluids. Thus, co-administration of acid reducers increasing the stomach pH, may increase weak acid drug solubility and absorption, and leads to higher bioavailability and drug exposure (29). Higher crenolanib bioavailability and exposure in the presence of acid reducers would be consistent with lower observed apparent crenolanib clearance and volume.

To our knowledge, no other studies have explored and identified the influence of acid reducers cotreatment on crenolanib pharmacokinetics. Interestingly, our initial investigations regarding crenolanib exposure and cotreatment relationships were focused on the concomitant use of dexamethasone. This corticosteroid drug was used to manage neurological symptoms and to reduce nausea and vomiting, which were expected to occur during crenolanib therapy. A potential drug-drug interaction was anticipated between dexamethasone, an inducer of the CYP3A enzyme family, and crenolanib, metabolized by CYP3A enzymes; thus, optional off-dexamethasone pharmacokinetic studies were performed. During our first covariate analysis, significantly lower crenolanib clearances were observed in patients taking dexamethasone. A decreased clearance leading to higher drug exposure was the opposite effect that we expected from an induction of crenolanib metabolism by dexamethasone. Additionally, the analysis of the off-dexamethasone pharmacokinetic data did not show any clear trend supporting an impact of dexamethasone. These findings suggested that another factor was responsible for the low crenolanib clearance in patients taking dexamethasone and led us to explore other administered comedications.

The most frequent comedications given to our patient population included antiemetic agents, antibiotics, and acid reducers as previously described. As expected, we found that more than 75% of patients taking dexamethasone during the pharmacokinetic studies were also taking ranitidine or another acid reducer, as acid reducers are frequently co-administered with corticosteroids to reduce the risk of gastrointestinal complications. The correlation observed between acid reducer cotreatment and crenolanib clearance was stronger than that observed with dexamethasone comedication and was pharmacologically meaningful. Therefore, we believed that the lower clearances observed in this study were caused by using acid reducers. We can’t conclude that dexamethasone does not have any impact on crenolanib metabolism and exposure, as this effect may have been masked by the one of acid reducers. It is also possible that the used dexamethasone dosages did not lead to concentrations high enough to significantly induce crenolanib metabolism. Lastly, it is important to note that several of the acid reducers used in the population (i.e., ranitidine, omeprazole, and pantoprazole) have been reported as weak CYP3A4 inhibitors (31). Therefore, the impact on crenolanib exposure by acid reducers could be explained by an alteration of drug dissolution and potential inhibition of hepatic metabolism.

The last covariate that significantly influenced crenolanib exposure was patient BSA. During the stepwise covariate analysis, both BSA and bodyweight showed an effect on crenolanib apparent clearance. However, the addition of BSA explained more interindividual variability than that of bodyweight. Thus, the impact of BSA was retained in the final model. Crenolanib apparent clearance was found to linearly increase with BSA. This result highlighted the need for BSA-adjusted crenolanib dosages compared to flat doses in our pediatric and young adult population, which was also demonstrated by our model simulations. Indeed, the use of flat dose in our population considerably increased the variability in drug exposure within the range of patient BSA. Overall, despite the large pharmacokinetic inter-individual variability and the marked differences in crenolanib exposure caused by acid reducer cotreatment, no dosing adjustments for a particular group of patients are suggested. Higher crenolanib AUC were not associated with more frequent or severe toxicities (18). Moreover, a targeted crenolanib exposure has not yet been determined for children with brain tumors to maximize treatment efficacy or minimize toxicities. If a targeted exposure is defined, the developed pharmacokinetic model will be helpful to define the crenolanib dosages necessary to reach this targeted exposure.

One main limitation of this study is the lack of external model validation. Despite a reasonable number of pediatric patients enrolled in the pharmacokinetic analysis, the sample size was still not large enough to constitute an external group to further validate the model. The significant influence of concomitant use of acid reducing agents needs further investigations in other patient population. Last, the evaluation of pharmacokinetic-pharmacogenetic relationships would be of interest and may explain aspects of the remaining pharmacokinetic variability.

Conclusion

A population pharmacokinetic model of crenolanib in children and young adults with newly diagnosed DIPG or recurrent HGG was developed for the first time. Our results support the safe use of crenolanib in the pediatric population with BSA-adjusted dosages. Concomitant use of acid reducers was found to significantly increase crenolanib exposure, although no dosing alterations were considered warranted. Patient age and BSA also contributed to the large pharmacokinetic variability observed in the population. Our validated pharmacokinetic model showed good predictive performance and can be used to guide the design of future clinical trials with crenolanib in pediatric populations.