FormalPara Key Points

Population pharmacokinetic analysis of REC-2282 (AR-42) demonstrates high pharmacokinetic variability, which may be a function of variable oral absorption

The present analysis suggested FFM does explain a very small portion of IIV (< 3%), and this supports choice of flat dosing, as the minimal decrease in pharmacokinetic variability utilizing body size-normalized dosing would not justify this dosing strategy

1 Introduction

Histone deacetylases (HDACs) are overexpressed in a variety of cancers, and HDAC inhibitors are an expanding class of anti-cancer therapeutics that induce growth arrest, differentiation and apoptosis of malignant cells [1, 2]. REC-2282 (also known as AR-42 and referred to as such from this point forward) is a novel HDAC inhibitor that exhibits antitumor activity in in vitro and in vivo models of malignancy, and similar to other pan HDAC inhibitors, suppresses tumor cell growth via a variety of mechanisms. Additionally, REC-2282 has shown promising activity in preclinical models of neurofibromatosis type 2 (NF2) [3]. Interestingly, from the first-in-human (FIH) study, the median progression-free survival for solid tumors was 3.6 months compared to 9.1 months in patients with NF2 and meningiomas [4, 5]. In agreement with these clinical findings, an informatics-based artificial intelligence (AI) approach independently suggested REC-2282 would be effective in NF2 [6]. REC-2282 is currently under clinical investigation for vestibular schwannoma and meningiomas (NCT02282917).

In preclinical studies, oral bioavailability was estimated at 26% and 100% in mice and rats, respectively (unpublished data). REC-2282 was highly bound to plasma proteins in mouse plasma (96%). Renal clearance was a minor pathway of elimination, as 1.4% and 5% of the dose was recovered in mouse and rat urine, respectively [7]. Preliminary phase 1 data from the first-in-human study with orally administered, flat-dosed REC-2282 revealed a coefficient of variation (CV) ranging from 36 to 76% among dose levels in apparent oral clearance and 24–66% in peak concentrations [2]. The maximum tolerated dose (MTD) was found to be 40 mg and 60 mg for patients with hematologic and solid tumor malignancies, respectively, both dosed three times weekly for 3 weeks followed by 1 week off in a 28-day cycle. Observed dose-limiting toxicities included grade 4 thrombocytopenia, febrile neutropenia and grade 3 neutropenia with infection and grade 4 psychosis [2, 4]. Among the 27 patients enrolled, clinical response varied from minimal response to progressive disease, and dose-dependent cytopenias were the most common adverse event. The observed wide inter-individual pharmacokinetic variability (IIV) [8] and variability in clinical response is commonly observed with many anticancer therapies, especially with oral drugs, often requiring dose reductions or termination of therapy [9, 10]. Typical strategies for individualizing doses include body-size normalization, though the benefits of this approach have been challenged repeatedly [11,12,13,14,15,16]. Alternatively, flat dosing is the most convenient and most easily managed dosing approach, especially with orally administered tablet or capsule formulations in fixed dosage strengths. However, the choice of flat vs. body size-normalized dosing, especially within a first-in-human study, should be evaluated to ensure the choice provides acceptable variability in exposures within the targeted patient population.

The primary objectives of the current study were to describe the population pharmacokinetics of REC-2282 using pooled data from NCT01129193, the first-in-human study in solid tumors and hematologic malignancies, and NCT01798901, a phase 1 study in acute myeloid leukemia, and identify covariates that explain portions of IIV in plasma pharmacokinetics. In particular, we explored various body size descriptors as potential sources of IIV. As part of this analysis, we also wanted to evaluate whether or not the use of body size-normalized dosing, as opposed to flat dosing, could further reduce observed pharmacokinetic variability.

2 Methods

2.1 Study Design, Drug Treatment and Patient Enrollment

The details of trial design concerning the clinical trials included in our analysis have been previously described [2, 4, 5, 17]. These studies were approved by The Ohio State University Institutional Review Board and conducted in accordance with the Helsinki Declaration of 1975, as revised in 1983. Patients enrolled in both trials were administered REC-2282 orally three (or four) times weekly, Monday, Wednesday, (Thursday) and Friday in 28-day cycles with 3 weeks of dosing followed by a 7-day off-treatment period (Table 1). Blood samples were collected from a total of 57 subjects enrolled on OSU09102 (NCT01129193) and OSU11130 (NCT01798901) including times points of pre-dose, (0.25), 0.5, 1, 1.5, 2, 4, 8, 10 and 24 h post-dose on either day 1, day 5 and/or day 19.

Table 1 Dosing regimens of REC-2282 summarized by dose level and clinical trial
Full size table

2.2 Pharmacokinetic Sample Analysis

Plasma concentrations of REC-2282 were determined using previously described bioanalytical methodology [2] and are summarized in Supplemental Table 1.

2.3 Pharmacokinetic Modeling and Covariate Analysis

Pharmacokinetic data were available from 55 subjects, from day 1, day 5 and/or day 19. REC-2282 plasma concentrations were available. Population pharmacokinetic models were developed using NONMEM, version 7.3, implementing the first-order conditional estimation method with interaction (FOCE-I) [18]. R (version 3.3.1; http://www.r-project.org) and the Xpose package (version 4.5.3; http://xpose.sourceforge.net) were used for visual diagnostics.

REC-2282 concentration data were evaluated with various compartmental models using lag time (ALAG) and transit models to describe the observed apparent delay in REC-2282 systemic absorption [19]. IIV was estimated using an exponential error model. Residual variability (ε) was described with an additive error model for log-transformed data. Tested covariates included age, sex, tumor type, trial, lean body weight (LBW), fat-free mass (FFM), body mass index (BMI), total body weight (TBW), body surface area (BSA) and measured lean body weight (MLBW) by CT-derived calculation [20]. Tumor type and trial were evaluated as dichotomous variables for solid vs. heme tumor and OSU09102 vs. OSU11130, respectively. LBW, FFM, BMI and CT based body composition factors were calculated as noted below. Continuous covariates were normalized using population median values, and a power model was used during the covariate model construction:

$${\uptheta }_{i }={\uptheta }_{P }\times {(\frac{\mathrm{COV}}{\mathrm{Median}})}^{{\uptheta }_{COV}}\times {e}^{{\eta }_{i}}$$
(1)

where θi is the estimated value of a parameter for individual i, θP is the estimated typical value of the parameter in the population, COV is the individual value of the particular covariate under investigation, Median is the median value of the covariate in the study population, θCOV is the estimated covariate coefficient determining its effect, and eηi is the estimated individual variation between θi and θp not accounted for by the covariate, whereby the ηis in the population are approximately log-normally distributed with a variance of ω2 and mean of zero. For categorical covariates, the following function was used:

$${\uptheta }_{i }={\uptheta }_{P }\times {(1+{\uptheta }_{\mathrm{COV}}\times\mathrm{COV})}\times {e}^{{\eta }_{i}}$$
(2)

Covariates having a significant influence (P < 0.05) were added in a forward stepwise manner until no further reduction in objective function was observed. Backward elimination was then performed using a P value of 0.01.

The final population pharmacokinetic model was evaluated for accuracy and stability via bootstrap resampling. Model parameters were estimated, and 95% confidence intervals of the bootstrap replicates were compared with parameter estimates from the final pharmacokinetic model. Model-based simulation was then performed to evaluate predictive performance of the final model using visual predictive checks (VPCs) from 1000 simulations [21]. Simulation was also performed to explore how body size-normalized dosing might have impacted IIV.

2.4 Body Size and Body Composition Assessments

The patient characteristics that were collected and evaluated as potential covariates were: age, body weight, BSA, BMI, sex, trial, tumor type, LBW and MLBW by CT. LBW was calculated using TBW and height [22]:

$$\mathrm{LBW} \left(\mathrm{kg}\right)= 1.10 \times \mathrm{TBW} \left(\mathrm{kg}\right)-120 \times {\left[\frac{\mathrm{TBW}}{\mathrm{height}\left(\mathrm{cm}\right)}\right]}^{2}\left(\mathrm{males}\right)$$
(3)
$$\mathrm{LBW} \left(\mathrm{kg}\right)= 1.07 \times \mathrm{TBW} \left(\mathrm{kg}\right)-148 \times {\left[\frac{\mathrm{TBW}}{\mathrm{height}\left(\mathrm{cm}\right)}\right]}^{2}\left(\mathrm{females}\right)$$
(4)

FFM was calculated using TBW and BMI [23]:

$$\mathrm{FFM} \left(\mathrm{kg}\right)= \frac{9.27 \times 103 \times \mathrm{TBW}\left(\mathrm{kg}\right)}{6.68 \times 103+216 \times \mathrm{BMI}} \left(\mathrm{males}\right)$$
(5)
$$\mathrm{FFM} \left(\mathrm{kg}\right)= \frac{9.27 \times 103 \times \mathrm{TBW}\left(\mathrm{kg}\right)}{8.78 \times 103+244 \times \mathrm{BMI}} \left(\mathrm{females}\right)$$
(6)

BMI [24] and BSA [25] were calculated as previously described, using the following formulas:

$$\mathrm{BSA} \left({\mathrm{m}}^{2}\right)=\sqrt{\frac{\mathrm{height}\left(\mathrm{cm}\right) \times \mathrm{weight} (\mathrm{kg})}{3600}}$$
(7)
$$\mathrm{BMI}= \frac{\mathrm{weight} (\mathrm{kg})}{\mathrm{height} ({\mathrm{m})}^{2}}$$
(8)

CT images at specific lumbar landmarks correlate to whole-body skeletal muscle in healthy adults [26, 27]. Patients who received a CT scan within 30 days of their first cycle, and therefore pharmacokinetic assessments, were evaluated for skeletal muscle area at the third lumbar vertebra (L3) [20]. Images were analyzed for cross-sectional area (CSA) (cm2) using Slice-O-Matic software V4.3 (Tomovision, Montreal, Quebec, Canada), and density was quantified in Hounsfield units (HU) assigned to each image pixel relative to reference values for air (− 1000), water (0) and skeletal muscle (− 29 to 150) as previously described [28, 29]. Muscle attenuation (MA) was determined as the average HU for each scan. Total CSA of skeletal muscle was normalized by dividing by squared height (m2) and expressed as a skeletal muscle index (SMI) (cm2/m2). MLBW was calculated using the total muscle area (cm2) from each CT scan [20]:

$$\mathrm{MLBW} \left(\mathrm{kg}\right)=\left(L3 \left({\mathrm{cm}}^{2}\right) \times 0.3\right)+6.06$$
(9)

3 Results

3.1 Patient Demographics

This study was conducted using datasets from two clinical trials with REC-2282 at The Ohio State University Comprehensive Cancer Center. Combined, OSU09102 (NCT01129193), the first-in-human trial, and OSU11130 (NCT01798901) included a total of 57 patients with relapsed or refractory malignancies. Summaries of the patient demographics and disease characteristics were previously published, and a subset of patient features are also presented in Supplemental Table 2 [2, 4, 5, 17]. Anthropometric measurements were not available for one patient. Demographic and body size factors for all patients are summarized in Table 2. Doses of REC-2282 are summarized in Table 1.

Table 2 Summary of patient characteristics from clinical trials
Full size table

3.2 Population Pharmacokinetic Model for REC-2282

A total of 882 REC-2282 plasma concentrations were obtained from 55 subjects. Forty-four (44) subjects had day 1 and either day 5 or day 19 plasma concentrations. The observed plasma concentration-time data for all 55 subjects can be seen in Fig. 1. Several models were evaluated, including one- and two-compartment models without and with lag time, and also without and with transit compartments to accommodate observed variability in apparent lag time. A two-compartment base model with one transit compartment for absorption, lag time, first-order elimination and a proportional error model best described the data (see Fig. 2). The model was parameterized in terms of clearance (CL), volume of distribution of the central compartment (Vc), inter-compartmental clearance (Q), volume of distribution of the peripheral compartment (Vp), absorption rate constant (ka) and ALAG. IIV was estimated for CL and ka with sufficiently low shrinkage (< 15%). IIV was estimated for ALAG but with high shrinkage (62.5%). Base model parameters are displayed in Table 3, and relevant diagnostic plots can be found in Supplemental Figure 1.

Fig. 1
figure 1

Concentration vs. time plots of REC-2282 (AR-42) at A day 1; B day 5; C day 19. AMT Dose levels of AR-42 (mg). Dots represent observed concentrations while lines represent the mean concentrations of the dose group

Full size image
Fig. 2
figure 2

A two-compartmental structural model describing the pharmacokinetics of REC-2282 (AR-42). ALAG absorption lag time, ka absorption rate constant, Vc volume of central compartment, CL apparent clearance, Q inter-compartmental clearance, Vp volume of peripheral compartment

Full size image
Table 3 Population parameter estimates from the base and final covariate model
Full size table

For covariate analysis, each covariate was evaluated individually on each model parameter, and those that achieved significance (P < 0.05) were evaluated within the model in multivariate analysis using a step-wise selection procedure with forward addition and backward elimination. FFM, LBW, BSA, height, weight, tumor type and sex were significant covariates on CL, and tumor type was significant on ALAG in univariate analysis (see Table 4). Multivariate analysis was challenging because of instability of the model and inability to obtain covariance estimates for most models attempted. We ultimately chose to remove IIV on the ALAG parameter, which improved model stability. Interestingly, we observed improved model fit when the categorical covariates, tumor type and formulation were included on ALAG without IIV. In the final model FFM was retained as a single covariate on CL, and tumor type and formulation were retained as covariates on ALAG. The final model presented acceptable estimation errors of the parameters (< 40%) except for tumor type (60.6%) and formulation (50%); shrinkages were sufficiently low (< 16%), while random error was high (54.5%). The detailed parameter estimates in the structural and final covariate model are listed in Table 3.

Table 4 Results of univariate covariate analysis
Full size table

Bias and stability of the final model were assessed by evaluating 95% confidence intervals of predicted parameters utilizing the bootstrap method (Table 3). One thousand resampled datasets were simulated to evaluate prediction performance of the final model by VPC, in which observed data were compared with the 95% confidence intervals of the predicted values (Fig. 3). Overall, the final model sufficiently described the observed concentrations; however, the model may fail to adequately capture the Cmax in some patients.

Fig. 3
figure 3

Visual predictive check (VPC) evaluating the final model of REC-2282 (AR-42) pharmacokinetic on Days 1 (A), 5 (B) and 19 (C)

Full size image

3.3 Analysis of Patient Body Composition

Of the 56 patients, 21 had a CT image during cycle 1 evaluable for skeletal muscle area (SMA), SMI and MA at the L3 landmark. Anthropometric measurements and body composition parameters for patients with evaluable CT scans are summarized in Table 2. The mean ± SD SMI (cm2/m2) and MA (HU) for L3 were 44.71 ± 11.1 and 31.5 ± 10, respectively. Patients presented with a variable range of BSA, weight, BMI, SMI and MA.

As an alternative to traditional anthropometric-derived body descriptors (FFM, LBW, BSA, etc.), MLBW and MA were evaluated as continuous variables in a subset covariate analysis with the 21 patients characterized above. When MLBWs were calculated and compared to anthropomorphic-derived LBW and FFM, MLBWs were consistently lower (Supplemental Figure 2). A subset univariate analysis with patients with CT scans revealed anthropomorphic-derived LBW as a significant covariate in this subset of patients, but there were no further observable trends between the individual parameter estimates and any of the other covariates evaluated.

3.4 Simulation of Flat vs. Body Size-Normalized Dosing

To further evaluate whether body size normalization of the REC-2282 dose could decrease inter-subject variability, simulations were performed to compare the difference between flat dose-based simulation and body size-normalized simulation. AUCs were simulated from a subset of patients who were administered with a 40-mg flat dose three times weekly, the same subset of patients but with a LBW-based dose, a LBW dose based on the available capsule strengths or a BSA-based dose. The 40-mg dose level was selected since this dose level contained the greatest number of patients with evaluable pharmacokinetic profiles. Simulations showed no differences in patient AUCs between these simulated dosing regimens, though differences in standard deviations were observed with the flat dose having the smallest inter-subject variability for simulated AUCs (Supplemental Figure 3). Given the lack of difference observed in simulated AUC and the higher AUC variability among the simulated body size-normalized dose regimens compared to the flat dose regimen, we concluded the flat dose method was in fact the most appropriate choice based on the current dataset.

4 Discussion

REC-2282 is in clinical development for treatment of cancer and neurofibromatosis. Thus far, five clinical trials have been initiated in solid tumor and hematologic malignancies and in vestibular schwanommas. Two of these trials were completed, and three are either on hold (2) or were terminated (1) as the new drug owner, Recursion Pharmaceuticals, establishes new clinical drug supply. Overall, REC-2282 has demonstrated acceptable safety in these early phase studies, and promising clinical activity has been observed in some patients, especially in those with schwanomas [3, 30]. Notably, REC-2282 was also discovered through novel in silico screening to have high activity in NF2-driven tumors [5], which has motivated investigators to further explore this area clinically.

In this report, we combined data from two of the completed trials, including the first-in-human study, to characterize pharmacokinetics of REC-2282 across a population of 55 patients with both solid and hematologic cancers and across a dose range from 20 mg to 80 mg. Our analysis revealed high variability in REC-2282 exposure, with dose-normalized AUC and Cmax of 24.1% and 29.6%, respectively. The data also demonstrated REC-2282 pharmacokinetics is dose-proportional across this range of flat doses. Using nonlinear mixed effects modeling, we explored a variety of structural models and performed covariate analysis to identify factors that could explain the significant observed variability. Our final model included a single transit compartment for oral absorption with ka, ALAG, a central plasma compartment and a peripheral tissue compartment. FFM was the only significant covariate identified on CL, and it described only a small portion of the overall observed pharmacokinetic variability (2.6% of IIV on apparent clearance).

Despite a large portion of the variability unexplained by the available covariate data, the model performed well in describing the data overall, though one obvious limitation was its inability to capture the time of peak concentration (Tmax). Our model consistently over-predicted Tmax, which we sought to address by incorporating first-order, zero-order and parallel zero- and first-order oral absorption with or without lag time or transit compartments. However, the model became over-parameterized when adding multiple parameters to describe the absorption process. Additionally, although IIV on ka or other absorption parameters significantly improved the Tmax predictions, this strategy resulted in high shrinkage of the IIV terms. These results suggested a significant portion of the observed IIV was due to variability in absorption. Notably, patients in these two trials were fasted prior to drug administration, which allowed us to rule out variability due to differences in food intake among enrolled patients. As we were not able to achieve acceptable parameter precision or model stability when we attempted to model various types of absorption processes, we ultimately settled on the single transit compartment model with first-order absorption, where a majority of IIV remained. In this final model, we added a lag time parameter into the transit compartment and identified tumor type and formulation as significant covariates on this parameter. Given the difference in sampling schemes, tumor type may be attributed to a study effect between these two trials. Further consideration of alternate oral absorption models, adjusted sampling schemes to better describe the absorption process and the assessment of REC-2282 bioavailability with pharmacokinetic data obtained after intravenous dosing in future clinical trials may be warranted.

During our search for covariates to describe the observed pharmacokinetic variability, and because we wanted to determine whether flat dosing vs. body size-normalized dosing was appropriate for clinical development, we interrogated various body size descriptors including anthropomorphic-derived LBW, FFM, BMI and CT-derived lean mass (MLBW) using SMI and MA to determine whether these improved our ability to describe REC-2282 pharmacokinetic variability. Notably, in the subset of patients with CT scans, we observed lower estimates of lean mass by CT compared to LBW or FFM (Supplemental Figure 3). This finding is consistent with other reports directly comparing imaging-based assessments of lean body mass with anthropomorphic-derived parameters [31] and suggests that cancer patients’ lean body mass may be consistently over estimated by traditional equations relating anthropomorphic features to lean body mass. Multiple reports suggest reduced LBM in cancer patients is associated with increased therapeutic toxicity [32, 33]. To this end, in the subset of patients with evaluable CT images, we observed patients with higher SMI completed greater than one cycle of REC-2282 compared to those completing one or less cycles (Supplemental Figure 4). However, in a subset covariate analysis we did not see a major impact from SMI or MA on IIV. Ultimately none of the other body size descriptors we evaluated provided an improved understanding of variability in our model. Of note, we observed slightly higher (15%) AUCs in females vs. males. With this, we chose to evaluate the relationship between observed dose-normalized AUC vs. FFM and observed a weak association (R2 = 0.18). Although this relationship was significant (P < 0.01), the portion of observed variability described by FFM was very low (2.6%). This exhaustive analysis of body size factors as potential covariates and the ability of only one of these factors to explain only a very small (2.6%) portion of observed pharmacokinetic variability support the decision for flat dosing instead of body size-normalized dosing as the appropriate choice for clinical development.

5 Conclusion

REC-2282 is a promising new therapeutic with the potential to treat neurofibromatosis type 2, a devastating disease currently lacking effective treatment options [34]. This first population pharmacokinetic analysis demonstrates high pharmacokinetic variability, which may be a function of variable oral absorption. Future modeling of REC-2282 absorption is warranted, though it will likely require additional data to understand the contributions of formulation, food effects, drug transport, etc. Our analysis suggested FFM does explain portions of IIV; however, like many oral drugs, the choice of flat dosing is appropriate, and little or nothing would be gained by utilizing body size-normalized dosing.