Population pharmacokinetics and site of action exposures of veliparib with topotecan plus carboplatin in patients with haematological malignancies

Abstract

Aims

Veliparib is a potent inhibitor of poly(ADP‐ribose) polymerase (PARP) enzyme. The objectives of the analysis were to evaluate the effect of baseline covariates and co‐administration of topotecan plus carboplatin (T + C) on pharmacokinetics of veliparib in patients with refractory acute leukaemia, and compare veliparib concentration in various biological matrices.

Methods

A population pharmacokinetic model was developed and effect of age, body size indices, sex, creatinine clearance (CrCL) and co‐administration of T + C on the pharmacokinetics of veliparib were evaluated. The final model was qualified using bootstrap and quantitative predictive check. Linear regression was conducted to correlate concentrations of veliparib in various biological matrices.

Results

A two compartment model with first‐order absorption with T_lag described veliparib pharmacokinetics. The apparent clearance (CL/F) and volume (V_c/F) were 16.5 l h⁻¹ and 122.7 l, respectively. The concomitant administration of T + C was not found to affect veliparib CL/F. CrCL and lean body mass (LBM) were significant covariates on CL/F and Vc/F, respectively. While a strong positive relationship was observed between veliparib concentrations in plasma and bone marrow supernatant, no correlation was observed between plasma and peripheral blood or bone marrow blasts.

Conclusions

Consistent with veliparib's physiochemical properties and its elimination mechanism, LBM and CrCL were found to affect pharmacokinetics of veliparib while concomitant administration of T + C did not affect veliparib's CL/F. Plasma concentrations were found to be a reasonable surrogate for veliparib concentrations in peripheral blood and bone marrow supernatant but not blasts. The current model will be utilized to conduct exposure‐response analysis to support dosing recommendations.

What is Already Known about this Subject

• Veliparib is an orally available, small molecule PARP inhibitor.

• The pharmacokinetics of veliparib has been characterized in patients with solid tumours.

• Topotecan or carboplatin/paclitaxel did not affect the pharmacokinetics of veliparib when administered with veliparib in patients with solid tumours.

What this Study Adds

• Veliparib pharmacokinetics is consistent in haematological malignancies and solid tumours.

• Creatinine clearance and lean body mass affect the CL/F and V/F of veliparib, respectively. T + C co‐administration does not influence CL/F of veliparib.

• Veliparib concentrations in plasma are correlated to peripheral blood and bone marrow supernatant but not blasts.

Tables of Links

TARGETS
Enzymes 2	Transporters 3
PARP1	OCT1
PARP2	OCT2
CYP1A2	OAT3
CYP2D6	BCRP
CYP3A
CYP4

LIGANDS
Topotecan
Carboplatin
Veliparib
Olaparib
Paclitaxel

These Tables list key protein targets and ligands in this article that are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY 1, and are permanently archived in the Concise Guide to PHARMACOLOGY 2015/16 2, 3.

Introduction

Poly(ADP‐ribose), or PAR, has been implicated in many cellular processes including replication, transcription, differentiation, gene regulation, protein degradation and spindle maintenance. The polymerization and recruitment of PAR to sites of DNA damage is catalysed by the poly(ADP‐ribose) polymerase (PARP) family of nuclear enzymes, in particular PARP1 and PARP2, which are activated by both single‐ and double‐strand DNA breaks and are essential to repair damage via base excision repair pathways 4, 5. Overexpression of PARP1 has been detected in a broad spectrum of human haematopoietic and epithelial malignancies 6, 7, 8. Inhibition of this repair process could serve to sensitize cells to DNA‐damaging chemotherapy and radiation therapy 9, 10. PARP inhibitors may have a special role in the treatment of BRCA‐deficient malignancies, as single agents and in combination with cytotoxic agents 10, 11.

Veliparib is an orally available, small molecule PARP inhibitor that has been found to enhance the cytotoxicity of diverse classes of DNA‐damaging agents, including topotecan, cisplatin, carboplatin, cyclophosphamide, irinotecan, radiation and temozolomide 11, 12, 13, 14. Efficacy and safety data from Phase I and II trials demonstrated that PARP inhibitors such as olaparib and veliparib in combination with cytotoxic agents were well tolerated and showed antitumour activity 15.

A previous clinical trial of T + C yielded promising results in adults with relapsed and refractory acute leukaemia 16. Building on the results of this trial, a Phase I dose escalation trial of veliparib administered twice daily (BID) on days 1–8 with T + C from day 3–7 was conducted in patients with refractory acute leukaemia 17. The data from this Phase I dose escalation trial were utilized for current population pharmacokinetic analysis. The pharmacokinetics of veliparib have been characterized in patients with solid tumours 18, 19. The aim of the current analysis is to: (1) utilize population pharmacokinetic analysis to characterize and estimate the variability in the pharmacokinetics of veliparib in patients with haematological malignancies, (2) identify the baseline covariates that affect the pharmacokinetics of veliparib, (3) evaluate the effect of co‐administration of T + C on pharmacokinetics of veliparib, and (4) correlate veliparib concentrations in various biological matrices.

Methods

Data

The pharmacokinetic data for the current analysis was obtained from a Phase I dose escalation trial of veliparib conducted in patients with relapsed or refractory acute myelogenous leukaemia. The main objective of this trial was to establish the maximum tolerated dose (MTD) of veliparib in combination with T + C 17. Veliparib was administered orally as a BID regimen on days 1–8. Topotecan (1.2 mg m⁻² day⁻¹) and carboplatin (150 mg m⁻² day⁻¹) were administered as 120 h intravenous continuous infusion from days 3–7. The doses of veliparib studied in the trial ranged from 10 to 100 mg BID. The plasma samples for measuring veliparib concentrations were collected on days 1 and 4 at pre‐dose, 0.25, 0.5, 1, 2, 4, 6, and 8 h after veliparib administration. In addition to the plasma, veliparib concentrations were also measured in the following biological matrices: peripheral blood (PB) blast, PB blast supernatant, bone marrow (BM) blast and BM supernatant. These samples were collected 4–7 h post‐dose on days 1 and 4. The veliparib concentrations in plasma, bone marrow cells and supernatant were quantified using liquid chromatography‐tandem mass spectrometry (LC/MS/MS) bioanalytical method 20.

Population pharmacokinetic modelling

Population pharmacokinetic modelling of day 1 and day 4 veliparib data was simultaneously performed using Phoenix NLME 1.4 (Pharsight, a Certara™ Company, Cary, NC) and first‐order conditional estimation‐extended least squares. Graphical exploration of the data was performed using R (version 3.1.1).

Base model

One and two compartment models with varied absorption models and first‐order elimination were explored. The pharmacokinetic parameters were assumed to be log normally distributed and between‐subject variability (BSV) was modelled using an exponential error model. Proportional and proportional plus additive residual error models were assessed. The inclusion of full covariance matrix between the random effects was based on the graphical exploration of the relationship between the random effects on the pharmacokinetic parameters. The adequacy of the base model was assessed using likelihood ratio test (LRT), precision of the parameter estimates, diagnostic plots and individual predictions.

Covariate model

Once the base model was identified, covariate evaluation was performed by plotting the random effects of pharmacokinetic parameters from the base model against each covariate and trends were observed. The covariates evaluated include: total body weight (TBW), LBM, height, CrCL, co‐administration of T + C, age, sex, body surface area (BSA) and body mass index (BMI). CrCL was calculated using the Cockcroft–Gault equation 21. LBM was computed using Boer's formula 22. The effect of continuous covariates on pharmacokinetic parameters was explored using the following structure:

$$P_i=\mathit{tvP}\bullet {\left(\frac{\mathit{COV}}{\mathit{\text{Median}}\mathit{COV}}\right)}^{\theta }$$ (1)

where P _i is the pharmacokinetic parameter in an individual, tvP is the typical value of the pharmacokinetic parameter at median value of the covariate (Median COV), COV is the covariate value in each subject, and θ is the power exponent for the covariate effect.

The effect of categorical covariates was explored using the following relationship:

$$P_i=\mathit{tvP}·\left(1+\theta ·\left(\mathit{\text{CATCOV}}\right)\right)$$ (2)

where tvP is the typical value of the pharmacokinetic parameter when CATCOV (binary categorical covariate) = 0 and θ is the proportional change in tvP when CATCOV = 1.

Covariate modelling was performed by forward addition [objective function value (OFV) decreased by at least 3.84 units (α = 0.05 , df = 1)] followed by backward elimination [OFV increased by at least 6.64 units (α = 0.01 , df = 1)] using LRT. Apart from statistical criteria, physiological relevance was considered as an important criterion for covariate selection.

Once the final covariate model was developed, the effect of between occasion variability (BOV) on the pharmacokinetics of veliparib between day 1 and day 4 was explored. BOV was modelled as an exponential term added to the BSV using the following structure:

$$P_i=\mathit{tvP}·{\left(\frac{\mathit{COV}}{\mathit{\text{Median}}\mathit{COV}}\right)}^{\theta }·\mathit{exp}\left(\mathit{\eta P}+\mathit{\eta Px}·\left(\mathit{OCC}=0\right)+\mathit{\eta Px}·\left(\mathit{OCC}=1\right)\right)$$ (3)

where ηP is the random variation of individual i from the typical value of parameter P and ηPx is the random variability in an individual when occasion (OCC) is either 0 or 1. The ηP and ηPx were assumed to be normally distributed with a mean of 0 and a variance of ω² and γ², respectively.

Final model qualification

The final model qualification was performed considering the goodness of fit plots, precision of the parameter estimates, visual predictive check (VPC) and quantitative predictive check (QPC). The precision of the final model parameters was obtained using the asymptotic standard errors and bootstrap with N = 1000 runs. VPC was conducted as follows: 200 replicates with the same design as the original trial were simulated using the mean and the variability estimates from the final model. The 5th, 50th and 95th percentiles of the observed and the simulated concentration vs. time stratified by dose were then compared.

The final model was also qualified using QPC with C _max and AUC _0‐last as the exposure metrics 23. Since AUC and C _max will be further used to elucidate exposure‐response relationships for efficacy and safety, QPC was performed for these metrics to qualify the model for its ability to predict AUC _0‐last and C _max. Non‐compartmental analysis (NCA) was conducted for each of the 200 replicates generated for the VPC to obtain C _max and AUC _0‐last at days 1 and 4. The 50th percentile of C _max and AUC _0‐last was calculated for each replicate stratified by dose group. The histogram of the 50th percentile of the simulated data was plotted and overlaid with the median of the observed data for each dose level for days 1 and 4. An ideal QPC plot is expected to have minimal bias between the observed and the predicted exposures. The prediction error (%PE) was calculated as follows:

$$\%\mathrm{PE}=\frac{\text{PRED}-\mathrm{OBS}}{\mathrm{OBS}}·100$$ (4)

where PRED is the predicted median C _max or AUC _0‐last in each of the 200 simulated datasets and OBS is the observed median C _max or AUC _0‐last. The 5th, 50th and 95th percentile of %PE were then calculated for the 10 and 80 mg dose groups.

Comparison of veliparib concentration in various biological matrices

Plasma veliparib concentrations were compared to concentrations in various biological matrices by graphical analysis and linear regression. If the time of veliparib sample collection was different between plasma and biological matrices, a population PK model was utilized to predict the plasma concentrations at the time of sample collection in the biological matrices.

Results

Table 1 shows the demographics of the patient population studied in the trial. A total of 1249 plasma concentrations from 90 patients administered doses of 10–100 mg BID were available. Figure 1 shows the mean concentration–time profile of veliparib on day 1 (without T + C) and day 4 (with T + C) at different doses. Based on the NCA results reported by Pratz et al. 17, ~30% [AUC _0‐tau (multiple dose)/AUC_0‐inf (single dose)] increase in exposure at steady state for 10 mg and 80 mg dose groups was observed after accounting for accumulation due to multiple dose administration. However, there was a 1.8‐fold increase in C _max and two‐fold increase in AUC after multiple doses between 80 mg and 100 mg dose 17. On closer examination of the data for three subjects on the 100 mg dose included in NCA, there was one outlier subject (ID 64) who had AUC _0‐inf of 1372 ng h ml⁻¹ after a single dose (similar to 40 mg dose) while AUC _0‐tau at steady state was 13 943 ng h ml⁻¹ (higher than the mean AUC at 100 mg dose) (Figure S1). Excluding this subject resulted in mean AUC _0‐inf (single dose) and AUC _0‐tau (multiple dose) of 6268 ng h ml⁻¹ (n = 3) and 6715 ng h ml⁻¹ (n = 2) after 100 mg dose. This resulted in a 1.31‐fold increase in AUC with a 1.25‐fold increase in dose (80–100 mg).

Table 1.

Baseline demographics of the study population

Demographic characteristics	Dose levels						Total
	10 mg	20 mg	40 mg	80 mg	90 mg	100 mg
n	27	13	5	37	4	4	90
Age (years) a	55 (40, 73)	64 (43, 75)	44 (33, 68)	58 (20, 73)	54 (40, 67)	61 (46, 76)	58.5 (20, 76)
Weight (kg)	78.8 ± 19.6 (49.7, 142.1)	82.0 ± 16.9 (58.5, 115.0)	88.2 ± 26.9 (56.1, 120.3)	76.2 ± 16.2 (47.5, 127.6)	88.6 ± 10.17 (78.3, 102)	76.5 ± 26.6 (58.9, 115.7)	79.0 ± 18.2 (47.5, 142.1)
Height (cm)	169.8 ± 10.3 (150, 186)	169.7 ± 8.8 (158, 184)	171.4 ± 10.7 (160, 188)	168.9 ± 11.4 (139, 191)	173.3 ± 9.3 (163, 185)	164.3 ± 18.8 (145, 190)	169.4 ± 10.8 (139, 191)
Sex (Male)	63%	53.8%	60%	51.4%	50%	25%	54.4%
Creatinine clearance (ml min −1 ) b	91.8 ± 20.9 (55.4, 120)	90.4 ± 22.9 (55.0, 120)	103.4 ± 15.7 (86.9, 120)	90.5 ± 23.9 (32.1, 120)	111.5 ± 15.6 (88.2, 120)	80.5 ± 11.9 (66.1, 93.4)	92.1 ± 22.1 (32.1, 120)
BSA (m 2 )	1.92 ± 0.27 (1.47, 2.74)	1.96 ± 0.24 (1.68, 2.41)	2.03 ± 0.33 (1.65, 2.39)	1.88 ± 0.23 (1.4, 2.48)	2.07 ± 0.09 (1.99, 2.18)	1.86 ± 0.41 (1.57, 2.47)	1.92 ± 0.26 (1.4, 2.74)
BMI (kg m −2 )	27.2 ± 5.3 (19.9, 41.1)	28.4 ± 4.7 (19.5, 36.3)	30.2 ± 9.5 (18.5, 41.4)	26.8 ± 5.4 (18.9, 43.4)	29.9 ± 6.4 (22.9, 38.4)	27. 8 ± 4.1 (22.2, 32.1)	27.5 ± 5.5 (18.5, 43.4)
LBM (kg)	55.9 ± 11.9 (38.0, 88.3)	56.5 ± 11.0 (43.7, 75.1)	59.8 ± 12.1 (48.2, 75.3)	53.6 ± 10.5 (33.67, 78.5)	58.6 ± 3.3 (54.5, 62.1)	50.6 ± 19.1 (35.7, 78.6)	55.1 ± 11.2 (33.67, 88.3)

The numbers represent mean ± standard deviation (range).

^aMedian (range).

^bCreatinine clearance was capped at 120 ml min⁻¹.

Figure 1.

Geometric mean concentration–time profile of veliparib on day 1 (without T + C) and day 4 (with T + C). Solid black lines represent concentrations at day 4 and dashed lines represent concentrations at day 1, solid black circles and error bars denote geometric mean and standard error, respectively

Population pharmacokinetic model

A two compartment model with first‐order absorption and T_lag adequately described the concentration–time profiles of veliparib. The OFV decreased by 90 units with two additional parameters for the two compartment model, compared to a one compartment model. The two compartment model was parameterized in terms of apparent volume of the central compartment (V_c/F) and peripheral compartment (V_p/F), apparent inter‐compartmental clearance (Q/F) and central clearance (CL/F), absorption rate constant (K_a) and lag time (T_lag). BSV on CL/F, V_c/F, K_a and T_lag was estimated. The covariance was estimated between random effects on CL/F and V_c/F; T_lag and K_a. The proportional error model best described the residual variability in the data. Table S1 shows the comparison of different base models.

CrCL was capped to 120 ml min⁻¹, which is the physiologically relevant maximum value of CrCL (23% of patients had CrCL > 120 ml min⁻¹). CrCL was evaluated as covariate on CL as described by Holford et al. 24. Veliparib is excreted 70% by renal elimination and 30% by hepatic metabolism 25, 26. It has been reported that CYP2D6 metabolism, additional liver metabolism, glomerular filtration and active tubular secretion account for 18%, 10%, 27% and 45% of total veliparib clearance, respectively 19. Since CrCL is the surrogate for the renal clearance through glomerular filtration, the total apparent CL in each subject (CL_i) was estimated as: (1) renal CL through filtration (tvCL_GF = tvCL_{glomerular filtration}) and (2) other pathways (tvCL_other) that included tubular secretion and hepatic metabolism.

$${\mathrm{CL}}_{\mathrm{i}}={\text{tvCL}}_{\text{other}}+{\text{tvCL}}_{\mathrm{GF}}·\left(\frac{\text{CrCL}}{90}\right)$$ (5)

where tvCL_GF represents the typical value of CL through glomerular filtration in subjects with CrCL of 90 ml min⁻¹.

Addition of CrCL on CL/F, decreased the OFV by 10 units, BSV on CL/F decreased from 40.0% (base model) to 36.8% and the trend between random effects on CL/F and CrCL disappeared (Figure S2). Based on the established covariate relationship between CrCL and CL/F of veliparib, a mild renal impaired subject (CrCL = 75 ml min⁻¹) will have 28% higher AUC as compared to a patient with normal renal function (CrCL = 120 ml min⁻¹). The effect of moderate renal impairment (30–60 ml min⁻¹) on CL/F of veliparib cannot be estimated since the majority (n = 7 out of 8) of subjects with moderate renal impairment had CrCL between 55 and 60 ml min⁻¹. After adjusting for CrCL, weight was not a significant covariate on apparent clearance and thus administration of fixed dosing is justified. As there appears to be a link between exposures and dose‐limiting toxicities, the exposure‐safety relationship will be further explored, including assessment of dose alteration based on renal function 17. Figure S3(A) shows the typical concentration–time profile of veliparib in a typical patient with mild renal impairment and normal renal function.

Different body size metrics such as TBW, LBM, BSA and BMI were explored as a covariate on V_c/F. Inclusion of LBM reduced the OFV by 17 units as compared to the base model, decreased BSV on V_c/F from 36.0% (base model) to 28.7%, and the trend between random effects on V_c/F and LBM disappeared (Figure S2). The final model included CrCL on CL/F and LBM on V_c/F as covariates. Figure S3(B) shows the effect of LBM on PK profile of veliparib when included on V_c/F. Table S1 shows the change in OFV for the selection of covariate model. Co‐administration of T + C was not found to be a significant covariate on CL/F and V_c/F (Figure S4), which would imply absence of pharmacokinetic interaction of T + C with veliparib.

In order to explain the 30% increase in exposures observed on day 4 with T + C co‐administration after accounting for accumulation due to multiple dosing, the effect of change in relative bioavailability between the two occasions (with and without T + C) was investigated. However, due to lack of biological plausibility, this was not considered further. None of the other covariates were found to be significant after inclusion of LBM on V/F and CrCL on CL/F (Figure S5).

After the development of the final covariate model, the effect of BOV on K_a, V_c/F and CL/F was explored. The absorption of veliparib was found to be highly variable. The BSV on K_a was 120% while the residual variability was 33% before inclusion of BOV on K_a. The inclusion of BOV on K_a significantly reduced the OFV by 357 units. In addition, the BSV in K_a reduced to 22% while residual variability decreased to 26%. Thus, BOV on K_a was included in the final model. A large BOV of 131.5% was estimated on K_a. There was no significant change in OFV when BOV was included on CL/F and V_c/F.

Table 2 and Figure 2 show the final model parameter estimates and goodness of fit plots, respectively. The goodness of fit plots suggests that the population pharmacokinetic model adequately described the concentration–time profiles of veliparib and the associated pharmacokinetic variability. The population parameters were estimated with %RSE < 20%. The shrinkage for the final model was less than 20% on BSV for CL/F and V_c/F, while it was 48% and 38% for BSV on K_a and T_lag, respectively. The final equations for typical CL/F and V_c/F are as follows:

Table 2.

Pharmacokinetic parameters for the final model and 1000 bootstrap samples

Parameter	Estimate (%RSE)	BSV (% CV)	Bootstrap estimate (median (95% CI))	Bootstrap BSV (% CV (mean (95% CI))
CL/Fother (l h−1)	7.97 (11.8)	34.7	7.72 (3.19, 10.85)	34.8 (28.88, 41.44)
CL/Ffitration (l h−1)	8.52 (10.1)		8.72 (5.65, 12.91)
Vc/F (l)	122.7 (4.0)	29.0	122.6 (113.7, 131.85)	28.6 (21.83, 34.99)
Q/F (l h−1)	10.1 (11.8)	Not estimated	10.2 (8.34, 12.91)	Not estimated
Vp/F (l)	97.91 (19.2)	Not estimated	100.97 (70.5, 214.39)	Not estimated
Ka (h−1)	2.15 (13.3)	21.9	2.15 (1.72, 2.70)	23.3 (7.14,39.66)
Tlag (h)	0.22 (1.3)	7.47	0.22 (0.21, 0.23)	7.68 (5.07, 10.67)
LBM exponent	0.92 (13.3)	Not estimated	0.94 (0.64, 1.21)	Not estimated
Between occasion variability on Ka (%CV)		131.53		131.15 (114.53, 147.28)
Correlation between random effects for CL/F and V/F	0.83		0.84 (0.67, 0.97)
Correlation between random effects for Tlag and Ka	‐0.79		−0.80 (−0.98, −0.52)
Residual error (Proportional) (%CV)	26% (4.7)		26% (24%, 28%)

Figure 2.

Goodness of fit plots for the final model

$$\frac{\mathrm{CL}}{\mathrm{F}}=8.0+8.5·\left(\frac{\text{CrCL}\left(\mathrm{ml}{\min }^{-1}\right)}{90}\right)$$ (6)

$$\frac{{\mathrm{V}}_{\mathrm{c}}}{\mathrm{F}}=122.7·{\left(\frac{\mathrm{LBM}\left(\mathrm{kg}\right)}{55}\right)}^{0.92}$$ (7)

Model qualification

A non‐parametric bootstrap of 1000 datasets showed preci3se estimation (%RSE < 20%) of all parameters indicating overall model stability (Table 2). Figure 3 shows the individual predicted concentration–time profiles of representative subjects. Figure 4 shows the VPC of 10 mg and 80 mg dose groups at days 1 and 4. The model qualification using VPC and QPC is shown at 10 and 80 mg because the largest number of subjects were enrolled at these dose levels. Based on the VPC plots, the model slightly underpredicts the median concentrations. However, the variability was captured adequately on both days 1 and 4. Furthermore, the model was qualified using a QPC shown in Figure 5 for days 1 and 4. The model shows slight underprediction of median C _max and AUC _0‐last at days 1 and 4.

Figure 3.

Individual predicted concentration–time profile overlaid with the observed concentrations for representative subjects. Solid black circles represent the observed data; solid blue and black lines represent the individual predicted and population predicted concentrations

Figure 4.

Visual predictive check for 10 mg and 80 mg dose groups on days 1 and 4. Solid blue and black lines depict 5th, 50th and 95th percentile of observed and simulated data, respectively; open circles denote the observed data. The grey bands represent 95% prediction intervals of the simulated data

Figure 5.

Quantitative predictive check for 10 mg and 80 mg dose groups on days 1 and 4. The histogram shows the distribution of median C _max (ng ml⁻¹) and AUC _0‐last (ng h ml⁻¹) of 200 simulated datasets; black dotted lines represent the 5th, 50th and 95th percentile of median C _max and AUC _0‐last obtained from 200 simulations; solid blue lines depict the observed median of C _max and AUC _0‐last. The median %PE for C _max and AUC _0‐last for 10 mg dose were −9.9% (90% CI: −23.6%, 9.2%) and −5.7% (90% CI: −16.7%, 8.2%) on day 1, respectively, and −8.4% (90%CI: −22.5%, 9.1%) and −12.6% (90% CI: −24.9%, 0.49%) on day 4, respectively. The median PE for C _max and AUC _0‐last for 80 mg dose were −3.8% (90% CI: −17.2%, 10.3%) and −6.7% (90% CI: −17.4%, 1.69%) on day 1, respectively, and −12.8% (90% CI: −24.4%, −2.04%) and −4.6% (90% CI: −15.3%, 6.0%) on day 4, respectively

Comparison of veliparib concentration in various biological matrices

Figure 6 shows the comparison of exposures of veliparib in plasma and various biological matrices. A total of 102 (50 on day 1 and 52 on day 4) PB blast supernatant and 33 (19 on day 1 and 14 on day 4) BM blast supernatant samples were available. A strong positive correlation was observed between plasma veliparib concentrations and PB [Pearson correlation coefficient (r) = 0.82; Figure 6(A)] and BM supernatant [r = 0.86; Figure 6(C)], respectively.

Figure 6.

Comparison of veliparib concentrations in various biological matrices. Solid red and blue circles represent day 1 and day 4 data, respectively. (A) and (B): Peripheral blood mononuclear samples were collected at the same time (4 h after dose) as the plasma samples, thus observed plasma concentration 4 h post‐dose on day 1 and day 4 were used for comparison. (C) and (D): Bone marrow supernatant samples were collected ~5 h post‐dose on day 1 and day 4, thus model predicted concentrations at the time of bone marrow supernatant samples were used for comparison. Two subjects who had veliparib concentrations of 8 and 4667 pmol cell⁻¹ × 10⁶ in peripheral blood blast were not included in the plot. Similarly, two subjects who had veliparib concentrations of 6 and 71 in bone marrow blast were not included in the plot. One subject with observed veliparib concentration in peripheral blood pellets and bone marrow pellet of 0.0170 and 6.58 pmol cell⁻¹ × 10⁶, respectively on day 4 was excluded from plot (E)

A total of 48 (22 on day 1 and 26 on day 4) PB blast and 20 (10 on day 1 and 10 on day 4) BM blast samples were available. A weak negative correlation was observed between veliparib concentrations in plasma and PB [Figure 6(B), r = 0.39] or BM blast [Figure 6(D), r = 0.21]. A strong positive correlation [Figure 6(E), r = 0.96] was observed between veliparib concentrations in PB blast and BM blast. However, due to the small number of samples (n = 6), results should be interpreted with caution.

Discussion

The main objective of this research was to characterize the pharmacokinetics of veliparib, evaluate the effect of covariates and co‐administration of T + C on pharmacokinetics of veliparib in patients with haematological malignancies. One of the important contributions of our population pharmacokinetics model was the ability to predict the concentrations in plasma at the same time as when the bone marrow supernatant or blast exposures were collected. This was essential to perform correlation analysis between plasma with bone marrow exposures. Veliparib concentrations in different biological matrices were compared to evaluate whether the drug reaches the bone marrow (i.e., the site of action) and whether plasma concentrations are representative of exposures observed in bone marrow (supernatant and blasts).

The pharmacokinetics of veliparib in the current analysis was characterized using a two compartment model with first‐order absorption and a lag time. Previously, the pharmacokinetics of veliparib has been characterized in patients with solid tumours using a one compartment model with first‐order absorption 18, 19. However, for the current data, the one compartment model with first‐order absorption did not provide adequate fit. Furthermore, another veliparib concentration–time dataset collected in breast cancer patients (data not shown; J.H. Beumer, personal communication) where plasma samples were collected until 24 h after dosing also indicated that a two compartment model fit the data better than a one compartment model. Even though the structure of the pharmacokinetic model was different, the parameter estimates of CL/F and Vc/F were consistent between patients with haematological malignancies and solid tumours 18.

The estimated CL/F was 16.5 l h⁻¹ for a subject with CrCL of 90 ml min⁻¹. Since 70% of the veliparib dose is excreted unchanged in the urine in humans through glomerular filtration and tubular secretion 25, 26, CrCL was used as a surrogate for glomerular filtration. The active tubular secretion is not expected to change with CrCL, thus the CL_other component of the total clearance comprised hepatic clearance and tubular secretion. The estimate of CL_GF was found to be consistent with the observed data. As shown in Figure S6, with a two‐fold increase in CrCL (from 60 ml min⁻¹ to 120 ml min⁻¹), a 58% increase in CL/F was observed, indicating that CrCL explained 58% of the total clearance. The estimate for CL_GF was 8.5 l h⁻¹ (i.e., 52% of the total clearance).

The estimated mean V_c/F was 122.7 l for a typical subject with LBM of 55 kg. The exponent of LBM was estimated to be 0.92. LBM has good correlation with volume of distribution and has been proposed as a better predictor of drug dosage in obese populations 27. In addition, drugs which are hydrophilic do not distribute to fatty tissues and thus volume of distribution of hydrophilic drugs like veliparib should be better correlated to LBM in obese subjects. As 64% of subjects were obese or overweight (i.e., BMI > 25), LBM was found to be a better covariate than actual body weight, as also reported by Salem et al. 18.

The estimates for Q/F, V_p/F, K_a and T_lag were 10.1 l h⁻¹, 97.9 l, 2.15 h⁻¹ and 0.22 h, respectively. The estimate of K_a was similar to that reported by Salem et al. 18. The inclusion of T_lag is empirical and may not be supported mechanistically. Inclusion of T_lag resulted in a 219 unit drop in OFV and also improved the diagnostic plots and thus was included in the model. Additionally, inclusion of BOV on K_a caused a significant drop in OFV and improved diagnostic plots and individual predictions and thus was included in the final model.

Despite a 1.3‐fold increase in veliparib exposure on day 4, concomitant administration of T + C was not found to affect pharmacokinetics of veliparib based on covariate analysis. This is consistent with the observation that the terminal slopes and thus the half‐life with single or multiple dose administration were similar 17. A change in clearance with co‐administration of T + C with veliparib would have likely manifested in change in terminal slope. Furthermore, there is no plausible physiological basis for any pharmacokinetic interaction between veliparib and T + C. Veliparib is primarily excreted by kidneys as unchanged drug (70%) 25, 26. The active tubular secretion mediated through organic cation transporters (OCT1/OCT2) plays an important role in veliparib clearance along with glomerular filtration 25. Veliparib is a weak P‐glycoprotein (P‐gp) substrate. Taken together, there is a low potential for clinically significant P‐gp or CYP‐450 mediated drug–drug interactions 28. Topotecan is also primarily excreted through urine in humans with approximately 49% of the dose recovered as total topotecan (topotecan lactone plus topotecan hydroxyl acid form) 29. The renal elimination of topotecan involves tubular secretion mediated by OAT3 (organic anionic transporters) in addition to glomerular filtration Furthermore, in vitro inhibition studies have demonstrated that human CYP2D6, CYP1A2, CYP3A and CYP4A enzyme activity is not altered by topotecan 30. In addition, Kummar et al. have also reported that there is no evidence of a significant pharmacokinetic interaction between veliparib and topotecan in patients with refractory solid tumours and lymphomas 31. Carboplatin is excreted almost exclusively through renal excretion. The total body clearance of ultra‐filterable platinum and that of the parent carboplatin molecule are roughly equivalent and correlate linearly with glomerular filtration rate 32, 33. Recently a Phase 1 study conducted in Japanese subjects with non‐small cell lung cancer also reported no impact of carboplatin/paclitaxel on the pharmacokinetics of veliparib 34. Taken together, this supports that T + C co‐administration should not be a covariate of veliparib exposure.

In addition, any change in relative bioavailability between day 1 and day 4 due to food may not be expected as the drug was administered under fasted conditions. Furthermore, food had no clinically meaningful effect on extent of veliparib absorption 35. Modest accumulation is expected with BID administration of veliparib due to short half‐life. After accounting for accumulation due to multiple dosing, a 1.3‐fold higher AUC with T + C was observed from the NCA analysis for 10 mg and 80 mg doses (highest number of patients). It is important to note that the increase of 1.3‐fold exposures is within the estimate of BSV on CL/F (35%) and thus may not be clinically relevant. In addition, a magnitude of 1.3‐fold increase in exposure is typically considered modest and may not have any clinical implications unless the drug in question has a narrow therapeutic index. Nevertheless, the clinical relevance of this exposure increase will be further explored after quantitating the exposure‐response for safety.

Veliparib concentrations were detected in BM blast and supernatant indicating the drug reaches the site of action. In addition, veliparib concentrations in plasma were correlated to concentrations in BM blast supernatant, indicating that plasma concentrations can be used as a surrogate for exposures in BM. However, no correlation was observed between exposures in plasma and PB or BM blast. It is plausible that the lack of a stronger correlation in blasts compared to the supernatant was due to P‐gp. Veliparib is a weak P‐gp substrate which, along with breast cancer resistance protein (BCRP), is highly expressed in blasts from acute myeloid leukaemia 36. We would like to acknowledge that the study was not designed to assess correlation between plasma and blasts exposures. It is possible that there is a delayed distribution in the blast which cannot be captured given the single sample per patient collected in this study. Therefore, correlation between veliparib concentration in PB blast and BM blast needs to be further characterized as this may provide an easily accessible surrogate marker of BM blast exposure.

Conclusion

The pharmacokinetics of veliparib in patients with relapsed or refractory acute myelogenous leukaemia or aggressive myeloproliferative neoplasms was characterized using a two compartment model with first‐order absorption and a T_lag. LBM and CrCL were found to be significant covariates that explained variability in V_c/F and CL/F, respectively. A modest 22% decrease in CL/F in mild renal‐impaired subjects may not warrant dose adjustment, but will be explored in a future exposure safety assessment. Co‐administration of T + C with veliparib did not affect the clearance of veliparib, which is also supported by mechanism of elimination of the veliparib, topotecan and carboplatin. Plasma concentration of veliparib can be used as a surrogate for veliparib concentrations in BM supernatant but not blasts. The current model will be utilized to conduct exposure‐response analysis to support dosing recommendations.

Competing Interest

There are no competing interests to delcare.

The project described was supported in part by NCI Cooperative Agreement U01CA070095, and UM1CA186691 (J.E.K. and M.A.R.) and by the Analytical Pharmacology Core of the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins (NIH grants P30CA006973 and UL1TR001079). Grant Number UL1TR001079 is from the National Center for Advancing Translational Sciences (NCATS), a component of the NIH, and NIH Roadmap for Medical Research. Its contents are solely the responsibility of the authors and do not necessarily represent the official view of the Johns Hopkins ICTR, NCATS or NIH.

Supporting information

Table S1 Comparison of the objective function values between different models

Figure S1 Individual concentration time profiles at day 1 and day 4 for four subjects on 100 mg dose. Solid black line represent concentrations at day 4 and dashed line represents concentrations at day 1, respectively

Figure S2 Comparison of random effects on pharmacokinetic parameters and covariates for the base and the final model

Figure S3 Effect of creatinine clearance and lean body mass on CL/F (A) and Vc/F (B) of veliparib, respectively after 80 mg dose at steady state

Figure S4 Comparison of apparent clearance and apparent volume of veliparib with and without topotecan plus carboplatin

Figure S5 Relationship between random effects of PK parameters and covariates (not included) for the final model

Figure S6 Relationship between clearance and creatinine clearance. Equation represents the linear regression fit

Mehrotra, S. , Gopalakrishnan, M. , Gobburu, J. , Greer, J. M. , Piekarz, R. , Karp, J. E. , Pratz, K. , and Rudek, M. A. (2017) Population pharmacokinetics and site of action exposures of veliparib with topotecan plus carboplatin in patients with haematological malignancies. Br J Clin Pharmacol, 83: 1688–1700. doi: 10.1111/bcp.13253.