FormalPara Key Points

Based on pharmacokinetic data from 5 phase 1/2 clinical studies of lucitanib in patients with advanced cancer, we show that a 2-compartment model with zero-order release into the dosing compartment adequately describes lucitanib pharmacokinetics.

A difference in release duration between lucitanib tablets and capsules was not considered clinically meaningful, and other covariates had no effect.

We observed high between-subject pharmacokinetic variability, suggesting that safety-based dose titration may optimize lucitanib exposure and maximize potential benefit for patients with advanced cancer.

1 Introduction

Development of therapies targeting angiogenesis was initially proposed as a means to starve tumors of oxygen and nutrients, thereby inhibiting tumor progression [1]. However, more recent preclinical and clinical studies of anti-angiogenic agents have revealed that these therapies may exert their anticancer effects not only by blocking tumor-driven recruitment of blood vessels but also by promoting immune activation and normalizing existing vessels to allow for enhanced drug delivery and immune cell infiltration. These proposed mechanisms provide a rationale for combinations of existing anti-angiogenic agents with other targeted therapies, requiring a detailed understanding of the pharmacokinetic parameters associated with dosing and timing [2].

Lucitanib is a potent, oral, antiangiogenic tyrosine kinase inhibitor (TKI) that selectively inhibits vascular endothelial growth factor receptors 1‒3 (VEGFR1‒3), fibroblast growth factor receptors 1‒3 (FGFR1‒3), and platelet-derived growth factor receptors alpha and beta (PDGFRα/β). Preclinical investigation suggests that lucitanib has a favorable pharmacokinetic profile with high tumor accumulation and demonstrated antitumor activity as a single agent, and in combination with anti-programmed cell death receptor-1 (PD-1) therapy in tumor xenograft models [3, 4].

The safety, pharmacokinetics, and antitumor efficacy of lucitanib as a single agent or in combination with fulvestrant have been evaluated in patients with various tumor types (E-3810-I-01, CO-3810-025, FINESSE, E3810-II-02, INES; Table 1) [5,6,7,8,9,10,11]. Study E-3810-I-01 (NCT01283945) was a phase 1/2a dose-escalation and dose-expansion study evaluating the safety, efficacy, and pharmacokinetics of lucitanib monotherapy [6]. CO-3810-025 (NCT02202746) and FINESSE (CL2-80881-001/NCT02053636) were both phase 2 studies of lucitanib monotherapy in patients with metastatic breast cancer with or without FGFR1/FGF amplification [7, 9, 11]. Study E3810-II-02 (NCT02109016) was a phase 2, open-label study of lucitanib monotherapy in patients with advanced/metastatic lung cancer with FGF-, VEGF-, or PDGF-related genetic alterations [10]. INES (CL1-80881-001/ISRCTN23201971) was a phase 1b, multicenter, dose-allocation study of lucitanib and fulvestrant in patients with estrogen receptor-positive (ER+)/human epidermal growth factor receptor 2-negative (HER2-) metastatic breast cancer who had relapsed during or after treatment with fulvestrant [8].

Table 1 Summary of clinical studies included in lucitanib PopPK analysis
Full size table

Consistent with the pharmacokinetics of many other TKIs [12], high between-subject variability (BSV) is observed with lucitanib exposure [6]. High BSV in absorption and/or disposition kinetics for therapeutic agents can be attributed to a variety of intrinsic and extrinsic factors [e.g., body weight, transporter efflux, and metabolism, organ function, and interactions with food or other medications, such as proton pump inhibitors (PPIs)] [12, 13]. Pharmacokinetic variability can result in adverse effects or suboptimal therapeutic efficacy and may require precision dosing based on individual characteristics, such as body weight or organ function [13]. Safety-based or pharmacodynamically guided dose titration has been explored for several VEGFR or multikinase inhibitors, including axitinib, erdafitinib, lenvatinib, and regorafenib [14,15,16,17].

Here, we present a population pharmacokinetics (PopPK) model developed using pooled pharmacokinetic data from the 5 clinical trials described above. The objectives of this analysis were to characterize the pharmacokinetics of lucitanib in patients with advanced cancers and to evaluate the effect of covariates on the variability of lucitanib pharmacokinetics, in order to better understand the high BSV observed in clinical studies.

2 Methods

2.1 Clinical Studies

Five clinical studies of lucitanib in patients with advanced cancers were included in the PopPK analysis (Table 1). All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee, and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. All studies were approved by an Institutional Review Board or Independent Ethics Committee. Written informed consent was obtained from each patient in all studies prior to participation.

Lucitanib (Clovis Oncology, Inc., Boulder, CO, USA) was administered orally at doses ranging from 5 to 30 mg once daily (QD) on a continuous schedule, or at 15 mg QD on 2 different intermittent schedules [5 days on treatment and 2 days off (5/2) or 21 days on treatment and 7 days off (21/7)] in 28-day cycles. Two immediate-release oral formulations, a film-coated tablet and a hard gelatin capsule, were used (see Table 1 for regimens and formulations used in each study). Intensive pharmacokinetic sampling was conducted for selected studies during the first dose cycle, and sparse sampling was conducted for up to 5 cycles. For evaluation of the effects of formulation on lucitanib pharmacokinetics, post hoc comparisons were limited to patients who received both formulations.

2.2 Assumptions

The allometric exponent was fixed to 1.0 for the relationship between body weight and volume terms and was fixed to 0.75 for the relationship between body weight and clearance terms.

2.3 Bioanalysis

Blood samples were collected in lithium heparin tubes, and processed to obtain plasma within approximately 30 min of collection. Lucitanib was extracted from plasma by protein precipitation and quantified using validated liquid chromatography‒tandem mass spectrometry methods [18]. The calibration range was 2.00‒500 ng/mL [lower limit of quantification (LLOQ), 2.00 ng/mL].

2.4 Pharmacometric Modeling

Modeling was performed using Nonlinear Mixed-Effects Model (NONMEM) v.7.4.3 (ICON Development Solutions, Ellicott City, MD, USA) with the First Order Conditional Estimation with Interaction (FOCE-I). These methods are commonly known by their abbreviations estimation method. Graphical analyses were conducted with RStudio v.1.2.5033 and R version 3.5. Goodness-of-fit (GoF) plots were generated with Xpose v.0.4.11, and visual predictive checks (VPCs), and bootstrap analyses were conducted using Perl-speaks-NONMEM (PsN) v.4.8.1. Hypotheses were generated based on biological plausibility and diagnostic graphics.

2.4.1 Base Model

Based on visual inspection of observed concentration–time datasets, a 2-compartment model with first-order absorption and first-order elimination was used as a starting point for the structural model. 1, 2-, and 3-compartment models were fit to lucitanib concentration‒time data, with the absorption process modeled as either traditional first-order absorption or zero-order release to the dosing compartment, followed by first-order absorption. BSV was assessed for duration of zero-order release (D1), first-order absorption rate constant (Ka), intercompartmental clearance/bioavailability (Q/F), apparent central volume (Vc/F), apparent peripheral volume (Vp/F), and apparent clearance (CL/F). Covariance between the random effects on Vc/F and CL/F of the central compartment was also assessed. Residual error was modeled as either a combined additive and proportional error model or a proportional error model only. Interoccasion variability was assessed for D1 and CL/F for each cycle.

To account for variability in patient weight for dosing, fixed allometric scaling exponents for body weight were assumed for all clearance and volume terms without statistical testing. This assumption was later confirmed to be appropriate based on a sensitivity analysis conducted on the final model with and without the allometric scaling components. The use of weight as opposed to body surface area for allometric scaling was also tested to ensure that the assumption was appropriate.

2.4.2 Covariate Model

To identify potential covariates for statistical testing, an exploratory analysis was first carried out by inspection of plots of post hoc estimates of CL/F and Vc/F versus covariates. All covariates were included in the dataset as time-varying. For plots that suggested a relationship, statistical testing was conducted for the specified covariate using a stepwise forward-addition/backward-elimination process. Individual covariates were retained in the model during forward addition based on a decrease in objective function value (OFV) of at least 6.635 (P ≤ 0.01, 1 degree of freedom) or an improvement in diagnostic plots. During backward elimination, covariates resulting in a minimum OFV increase of 10.828 (P ≤ 0.001, 1 degree of freedom) were retained. Exploratory hypotheses that were examined are described in Table 2.

Table 2 Exploratory covariate analysis for the lucitanib PopPK model
Full size table

2.4.3 Model Diagnostics

Base, covariate, and final models were selected based on OFV, parameter precision, and diagnostic plots (including both GoF plots and VPCs). For VPCs, 1000 simulations were performed with the selected model, and the results were graphically compared to the observed data based on the median and 95% prediction interval. VPC output was stratified by study to evaluate model fit for individual studies. Robustness of the model was assessed by bootstrapping (1000 replicates, with replacement).

3 Results

3.1 Data Summary

A total of 3540 pharmacokinetic records from 403 patients with advanced cancers enrolled in 5 clinical studies were included in the dataset for analysis. Two samples below the LLOQ were excluded. Baseline demographics and summary of covariates by study are presented in Table 3.

Table 3 Summary of covariates and patient demographics
Full size table

3.2 Pharmacokinetic Model Development

The pharmacokinetics of lucitanib were best described by a 2-compartment model with a zero-order release into the dosing compartment, followed by first-order absorption and first-order elimination. BSV was included for central CL/F and Vc/F, as well as D1 and Vp/F. Residual error was described by a proportional error model.

Covariates identified for statistical testing based on visual assessment are listed in Table 2. Addition of formulation and effects of concomitant PPIs on bioavailability (F) or cytochrome P450 (CYP) 2C8 inhibitors and renal function group effects on CL/F did not significantly improve model fit (ΔOFV < 6.636). Effects of formulation on D1, P-glycoprotein (P-gp) inhibitors on CL/F, and serum albumin on Vc/F were statistically significant, and were included during forward addition. The latter 2 covariates were rejected in backward elimination (ΔOFV < 10.828; Supplementary Table 1). Therefore, the final model included only 1 covariate: effect of formulation on D1, for which capsule release was relatively slow compared to tablet release. Given that elimination of lucitanib has been shown to be slow (half-life ~ 31‒40 h) [6], the effect of different D1 times on maximum plasma drug concentration (Cmax) is unlikely to be clinically significant. Weight was also included in the final model for scaling of CL/F and Q/F (fixed exponent, 0.75) and Vc/F and Vp/F (fixed exponent, 1.0). We did not identify any other covariates that significantly improved the model fit. Plots of post hoc random effects values for pharmacokinetic parameters (ETA in NONMEM modeling language) versus dose (Supplementary Fig. S1) were consistent with dose proportionality from 5 to 30 mg lucitanib based on visual inspection (Supplementary Fig. S1). With regard to any effect of concomitant PPI use, 112 of 403 patients (28%) had at least 1 observation or dose with concomitant PPI. Of these 112 individuals, 97 had concomitant PPI for all reported events (representing 83% of all doses and observations). Among the patients who had at least 1 dose or concentration with concomitant PPI, the mean fraction of events with concomitant PPI was 91%, and the median was 100%. A VPC of plasma lucitanib concentration during the early post-dose period (0–10 h post dose) with or without concomitant PPI is shown in Supplementary Fig. S4. A diagram of the final model structure is shown in Fig. 1.

Fig. 1
figure 1

PopPK model structure. The pharmacokinetics of lucitanib were described by a 2-compartment linear model with a zero-order release into the dosing compartment, followed by first-order absorption and first-order elimination. Volume (Vc, Vp) and clearance (Q, CL) terms were proportional to Weightexponent (a power model of weight with exponents fixed to 1 for volume terms and 0.75 for clearance terms). Duration of release from the tablet formulation (0.243 h) differed from the capsule (0.814 h). CL clearance, Ka first-order absorption rate constant, PopPK population pharmacokinetics, Q intercompartmental clearance, Vc central volume, Vp peripheral volume

Full size image

Lucitanib pharmacokinetic parameter estimates obtained using the final PopPK model were consistent with bootstrap results (Table 4). Diagnostic plots (Figs. 2, 3; Supplementary Fig. S2) also indicated that the final model described the observed lucitanib concentration–time data. Parameters were estimated with good precision, with relative standard error (RSE) from NONMEM output < 23% for all parameters. The RSE estimates were consistent with the confidence intervals (CI) from the bootstrap estimates (Table 4).

Table 4 NONMEM parameter estimates and estimates from a nonparametric bootstrap for the PopPK model of lucitanib
Full size table
Fig. 2
figure 2

Goodness-of-fit plots for the final lucitanib PopPK model. a Population prediction (ETA = 0) versus observed; b individual prediction (ETA ≠ 0) versus observed; c prediction versus conditional weighted residual; and d time versus conditional weighted residual. Blue lines indicate smoothed means (locally estimated scatterplot smoothing, LOESS [34]). Black trend lines indicate lines of unity (intercept = 0, slope = 1). Observations (concentrations or residuals, black dots) from the same individual are connected by black lines. ETA random effects values for pharmacokinetic parameters, PopPK population pharmacokinetics

Full size image
Fig. 3
figure 3

Visual predictive check for the final lucitanib PopPK model: time since first dose versus lucitanib concentration. PopPK population pharmacokinetics

Full size image

To further examine the effect of formulation (tablet vs. capsule) on the rate and extent of absorption, we plotted the conditional weighted residuals by formulation (Fig. 4) and post hoc estimates of Ka and CL/F BSV by formulation (Figs. 5, 6). An additional model was run with a between-occasion variation term to generate post hoc ETA values for Ka and CL/F by formulation, in which the occasions were defined by the formulation. All plots examining the effect of formulation on rate and extent of absorption suggested that the difference is well described by the effect of formulation on duration of the zero-order release (D1 parameter in NONMEM), which showed a difference of < 0.6 h in duration of release for the 2 formulations (D1, 0.243 h vs. 0.814 h for the tablet and capsule, respectively; Fig. 1). The model development table is provided in Table 4.

Fig. 4
figure 4

Conditional weighted residuals by formulation. Black horizontal lines represent median values, and boxes correspond to the ranges of the first and third quartiles. Upper and lower whiskers extend from the box to the largest or smallest value, respectively, within 1.5 times the interquartile range. Observations outside the whisker range are represented as dots

Full size image
Fig. 5
figure 5

Post hoc estimates of BSV of Ka by formulation. Black horizontal lines represent median values, and boxes correspond to the ranges of the first and third quartiles. Upper and lower whiskers extend from the box to the largest or smallest value, respectively, within 1.5 times the interquartile range. Observations outside the whisker range are represented as dots. BSV between-subject variability, ETA random effects values for pharmacokinetic parameters, Ka first-order absorption rate constant

Full size image
Fig. 6
figure 6

Post hoc estimates of BSV of CL/F by formulation. Black horizontal lines represent median values, and boxes correspond to the ranges of the first and third quartiles. Upper and lower whiskers extend from the box to the largest or smallest value, respectively, within 1.5 times the interquartile range. Observations outside the whisker range are represented as dots. BSV between-subject variability, CL/F apparent clearance, ETA random effects values for pharmacokinetic parameters

Full size image

4 Discussion

In this study, we have developed a PopPK model for lucitanib based on data from 5 clinical trials in patients with advanced cancer (Table 1). Lucitanib pharmacokinetics were adequately described by a 2-compartment model with zero-order release into the dosing compartment, followed by first-order absorption and first-order elimination. The only statistically significant covariate was formulation, which had an effect on duration of release. D1 was 70% shorter for tablets versus capsules. While the effect was precisely estimated (RSE, 19.4%) and statistically significant, this estimate should be interpreted with caution, as limited data were available in this zero-order release phase. Because both formulations were developed for immediate release with very short D1 values (< 1 h) relative to the slow elimination of lucitanib (half-life, ~ 31‒40 h) [6], the formulation difference did not translate into a clinically meaningful difference in the pharmacokinetic profile. As area under the curve (AUC) would not be influenced by the duration of zero-order release, and given the relatively rapid zero-order release and first-order absorption compared to elimination, the effect on Cmax was minimal and not clinically significant. An effect of body weight on volume and clearance terms was also assumed in the final model and was later confirmed to be appropriate using a sensitivity analysis. To accommodate the effects of weight, CL/F, Q/F, Vc/F, and Vp/F were defined as a power model of weight with exponents fixed to 0.75 (for clearance terms) and 1.0 (for volume terms). This step resolved the bias observed in plots of post hoc ETA for CL/F, Vc/F, D1, and Vp/F (Supplementary Fig. S3), and resulted in a 49-point reduction in the OFV (Supplementary Table 1).

We did not observe any effects of age, sex, race, ethnicity, or tumor type (by study) on lucitanib pharmacokinetics in this analysis. While effects of P-gp inhibitors on CL/F and serum albumin on Vc/F were suggested by visual analysis, further analysis showed that these effects were not statistically significant. As a result, these covariates were not included in the final PopPK model. Although limited data on concomitant use of CYP3A4 inhibitors and inducers precluded development of statistical conclusions, the results from this analysis did not show an effect of CYP3A4 and CYP2C8 inhibitors on lucitanib pharmacokinetics. In the case of CYP3A4, 46 observations from 11 patients with known concomitant inhibitors and 4 observations from 1 patient with a known concomitant inducer were included in the dataset. For CYP2D8, 118 observations from 14 patients with a known concomitant inhibitor were included in the dataset. Due to limited data, the P-gp, CYP2D8, and CYP3A4 results must be viewed as exploratory.

A number of TKIs have pH-dependent solubility, and co-administration with PPIs has been demonstrated to decrease TKI exposure in some cases [19,20,21]. Thus, understanding the effect of concomitant PPI administration on lucitanib pharmacokinetics is important to evaluate clinical impact on efficacy and safety. More patients in the study population were receiving concomitant PPIs (n = 152) compared to CYP3A4 (n = 12) or CYP2C8 (n = 14) inhibitors, yielding a more robust result showing an effect (ratio of F with concomitant PPI/without PPI) that was estimated with good precision (RSE 6.53%), though not significant (95% CI 0.83–1.08). Furthermore, the estimated effect size (6.5% reduction in F) was not clinically significant. The exploratory analysis suggested an effect of PPI on Ka (Supplementary Table 1). While the effect was statistically significant, it should be noted that the absorption profile is driven largely by the zero-order duration of release (D1 0.814 and 0.243 h for capsule and tablet, respectively), not by the relatively rapid first-order absorption process (Ka 4.86; absorption half-life 0.143 h). Consistently, diagnostic plots (VPCs) for models with and without PPI effect on Ka showed similar fits and suggested that concomitant PPI had no apparent effect on the overall pharmacokinetic profile (Supplementary Fig. S5). Thus, PPI was not included as a parameter in the final model. The effect of PPIs on lucitanib pharmacokinetics will be further assessed with emerging clinical data.

We also assessed lucitanib pharmacokinetic parameters in patients with varying degrees of renal and hepatic function. In comparison to patients with normal renal and hepatic function, no statistically significant or clinically meaningful pharmacokinetic differences (≤ 20%) were observed for patients with mild renal impairment [n = 141; creatinine clearance (CLCR) 60–89 mL/min], moderate renal impairment (n = 40; CLCR 30–59 mL/min), or mild hepatic impairment [n = 80; total bilirubin ≤ upper limit of normal (ULN) and aspartate aminotransferase (AST) > ULN, or total bilirubin 1–1.5× baseline and any AST level] [22, 23]. Patients with severe renal impairment or moderate or severe hepatic impairment were not included in this analysis (Table 3).

The current PopPK model is based on clinical studies of lucitanib monotherapy, as well as combination therapy with fulvestrant. In vitro studies have shown that lucitanib is equally metabolized by CYP2C8 and CYP3A4 (data on file. Clovis Oncology, Inc.). Fulvestrant did not significantly inhibit the major CYP isoenzymes, including CYP3A4 in vitro, nor did it change the exposure to a CYP3A substrate (i.e., midazolam in humans) [24]. No known drug–drug interactions have been reported for fulvestrant [24, 25]. Consistent with these data, the current PopPK model based on 5 clinical studies suggests that co-administration of fulvestrant has no effect on lucitanib pharmacokinetics, although the number of patients with concomitant fulvestrant was small (17 of 403).

Data from clinical trials have revealed high BSV with lucitanib pharmacokinetics [6, 26, 27]. Although the PopPK model developed in this study fits concentration–time data and estimated pharmacokinetic parameters with good precision, the covariates included in the model only explained a small part of the observed pharmacokinetic variability. In the present dataset, body weight explained 9.5% of the variance in CL/F and 11% of the variance in Vc/F.

As large variability in pharmacokinetic and pharmacodynamic responses can lead to suboptimal clinical efficacy and significant adverse effects [13], individualized dose titration is a rational approach for managing tolerability and maximizing efficacy of kinase inhibitors with high pharmacokinetic variability. For example, a retrospective analysis based on pooled data from 3 phase 2 studies of axitinib, a VEGFR inhibitor with high pharmacokinetic variability [28], showed that individualized treatment through a safety-based dose titration strategy improved efficacy and resulted in more consistent exposure across patients with metastatic renal cell carcinoma (mRCC) [14]. This dose titration strategy has been applied in subsequent clinical trials, including a phase 2 dose titration study of axitinib in patients with mRCC [29]. Dose titration of axitinib is also recommended in combinations with pembrolizumab or avelumab approved for treatment of patients with advanced RCC [30, 31]. Safety-based or pharmacodynamically guided dose titration strategies have also been explored for other multikinase inhibitors, such as regorafenib, lenvatinib, and erdafitinib [15–17].

The results from this PopPK model suggest that individualization of the dosing regimen based on safety could optimize the benefit:risk ratio for lucitanib. The phase 1b/2 LIO-1 study (NCT04042116) is exploring whether lucitanib-mediated inhibition of angiogenesis can enhance the antitumor activity of the anti-PD-1 inhibitor nivolumab in patients with advanced solid tumors [27, 32]. Initial pharmacokinetic analyses in 14 patients enrolled in LIO-1 have revealed a pharmacokinetic profile consistent with lucitanib monotherapy, including relatively high BSV [27]. Based on this observation, and no evidence of differential safety across dose levels in phase 1b [27, 32], a safety-based dose titration strategy has been recommended for the individualization of lucitanib dosing in patients with endometrial cancer, ovarian cancer, cervical cancer, or endometrial/ovarian clear-cell cancers in the phase 2 expansion cohorts. Over the course of treatment, patients who tolerate lucitanib for ≥ 28 days with no grade > 2 treatment-related adverse events, no proteinuria > 1 + (or urinary protein > 1.0 g/24 h), no grade > 1 treatment-related diarrhea, and blood pressure ≤ 150/90 mmHg not requiring any antihypertensive agents (or changes to pre-existing antihypertensive regimens if blood pressure is stable and well-controlled at baseline) may undergo dose escalation from 6 mg to 8 mg to 10 mg at subsequent cycles, provided that ≤ 2 cycles have elapsed between escalations [32, 33]. Initial phase 2 data from LIO-1 suggest that this safety-based dose-titration strategy is feasible with lucitanib [32].

5 Conclusions

The results from this study showed that a 2-compartment PopPK model with zero-order release into the dosing compartment adequately describes the pharmacokinetics of lucitanib in patients with advanced cancer. No clinically significant covariates were identified in this analysis. High BSV in pharmacokinetics supported a safety-based dose titration strategy that is being investigated in an ongoing clinical study of lucitanib to optimize lucitanib exposure and potential benefit.