Physiologically‐based pharmacokinetic modelling to predict oprozomib CYP3A drug–drug interaction potential in patients with advanced malignancies

Abstract

Aims

Oprozomib is an oral, second‐generation, irreversible proteasome inhibitor currently in clinical development for haematologic malignancies, including multiple myeloma and other malignancies. Oprozomib is a rare example of a small molecule drug that demonstrates cytochrome P450 (CYP) mRNA suppression. This unusual property elicits uncertainty regarding the optimal approach for predicting its drug–drug interaction (DDI) risk. The current study aims to understand DDI potential during early clinical development of oprozomib.

Methods

To support early development of oprozomib (e.g. inclusion/exclusion criteria, combination study design), we used human hepatocyte data and physiologically‐based pharmacokinetic (PBPK) modelling to predict its CYP3A4‐mediated DDI potential. Subsequently, a clinical DDI study using midazolam as the substrate was conducted in patients with advanced malignancies.

Results

The clinical DDI study enrolled a total of 21 patients, 18 with advanced solid tumours. No patient discontinued oprozomib due to a treatment‐related adverse event. The PBPK model prospectively predicted oprozomib 300 mg would not cause a clinically relevant change in exposure to CYP3A4 substrates (≤30%), which was confirmed by the results of this clinical DDI study.

Conclusions

These results indicate oprozomib has a low potential to inhibit the metabolism of CYP3A4 substrates in humans. The study shows that cultured human hepatocytes are a more reliable system for DDI prediction than human liver microsomes for studying this class of compounds. Developing a PBPK model prior to a clinical DDI study has been valuable in supporting clinical development of oprozomib.

What is Already Known about this Subject

• Oprozomib, an oral, second‐generation, irreversible proteasome inhibitor, has demonstrated anti‐tumour activity in preclinical models and clinical trials of patients with haematologic malignancies.

• Oprozomib has a half‐life of ~1 h; cytochrome P450 (CYP) enzymes play a minor role in its metabolism.

• An in vitro study showed oprozomib causes time‐dependent inhibition of CYP3A4/5 in human liver microsomes.

What this Study Adds

• Oprozomib suppresses CYP3A4 mRNA expression and enzymatic activity in human hepatocytes.

• A physiologically‐based pharmacokinetic (PBPK) model predicted low potential for oprozomib to inhibit CYP3A4 substrate metabolism in humans, confirmed by the clinical drug–drug interaction study.

• This study demonstrates the utility of PBPK modelling in drug interaction risk assessment and clinical programme development.

Introduction

Proteasome inhibitors are established treatments for multiple myeloma and other malignancies. The first‐in‐class reversible proteasome inhibitor bortezomib and the more recently approved irreversible proteasome inhibitor carfilzomib 1 are given subcutaneously or intravenously for treatment of newly diagnosed and relapsed and/or refractory multiple myeloma.

Oprozomib is an oral, irreversible proteasome inhibitor undergoing clinical evaluation for the treatment of haematological malignancies, including multiple myeloma. As a tripeptide epoxyketone and an analogue of carfilzomib that covalently bonds with the active site N‐terminal threonine of the 20S proteasome, oprozomib was designed to provide improved dosing flexibility and patient convenience 2. Preclinical studies have shown that oprozomib has potent anti‐multiple myeloma activity in tumour‐derived cell lines and animal models 3. Oprozomib has shown clinical activity in patients with haematologic malignancies 4, 5, 6, 7. In the phase Ib/II study in patients with haematologic malignancies, the maximum tolerated dose (MTD) of single‐agent oprozomib was 300 mg day⁻¹ when administered the first 2 days every 7 days of a 14‐day cycle (days 1, 2, 8 and 9; 2/7 schedule) or 240 mg day⁻¹ when administered on the first 5 days of a 14‐day cycle (days 1–5; 5/14 schedule) 7. On the 2/7 step‐up dosing schedule (240/300 mg day⁻¹), single‐agent oprozomib appears to have promising activity with higher overall response rate (mean of 34%, n = 38) in patients with relapsed multiple myeloma compared with those of 5/14 schedule 8.

Oprozomib has a short half‐life of approximately 1 h, with peak concentration between 0.6 and 2 h in patients with advanced solid tumours following oral administration 9. There is no evidence of accumulation after repeated dosing. Area under the plasma concentration curve (AUC) and peak plasma concentration (C _max) generally increase with increasing dose from 30 to 180 mg 9, and from 150 to 300 mg 7. High between‐subject variability in patients with haematologic malignancies was observed with inter‐subject percent coefficient of variation (%CV) ranging from 70% to 90% 7.

Oprozomib is primarily metabolized by epoxide hydrolases and peptidases; cytochrome P450 (CYP) enzymes play a minor role in its metabolism and elimination 10. In an in vitro study using human liver microsomes (HLMs), oprozomib showed time‐dependent inhibition (TDI) of human CYP3A4 10, and suppressed CYP3A mRNA expression, which is not commonly seen for small molecule drugs. The question remains regarding the optimal approach to predict the potential for drug–drug interaction (DDI) between oprozomib and CYP3A4 substrates.

Early understanding of clinically relevant DDI is critical in optimal development of antineoplastic agents, particularly those given orally, as a number of concomitant medications are routinely used in cancer patients 11. Physiologically‐based pharmacokinetic (PBPK) modelling has been used successfully to facilitate design of DDI studies and extrapolation to unstudied drug interactions, and to inform drug labelling 12, 13, 14. The objectives of the present study were several‐fold: firstly, to assess the utility of in vitro primary human hepatocyte data for prediction of oprozomib and CYP3A clinical DDI risk in light of oprozomib suppressing CYP3A mRNA; second, incorporating human hepatocyte data into a PBPK model to prospectively predict the clinical effect of oprozomib on CYP3A4 activity; and third, to conduct a clinical DDI study to confirm the model's predictive capabilities in patients with advanced malignancies. This study highlights an example of the model‐based learning and confirming paradigm to enhance the efficiency of clinical development 15.

Methods

In vitro assay using cultured human hepatocytes

Primary cultures of cryopreserved hepatocytes from three separate donors (denoted as lots 295, 312 and 318; obtained from Corning Life Sciences – Discovery Labware [Woburn, MA, USA]) were used. Hepatocytes on collagen I‐coated 96‐well plates were exposed to the test article for a total of 3 days, with media change approximately every 24 h. Oprozomib was incubated at concentrations of 0.003, 0.01, 0.03, 0.1, 0.3, 1, 3 and 10 μM. CYP3A4 suppression was measured by mRNA expression assays selective for CYP3A4 by real‐time polymerase chain reaction analysis. CYP3A4 in situ catalytic activity suppression was determined using the specific probe substrate testosterone. Positive control inducers and solvent vehicle controls were included. The 6β‐hydroxytestosterone metabolite formations in the test samples were analysed by liquid chromatography–tandem mass spectrometry (LC–MS/MS) using an AB Sciex 4000 or AB Sciex 4000 Qtrap™ mass spectrometer, with a standard metabolite range of 0.016–10 μM. All LC–MS/MS runs met the acceptance criteria as described, as follows: the accuracy of predicted concentrations ≥75% of both standard and quality control samples were within ±20% of their nominal values (25% for lower limit of quantification or lowest concentration quality control sample). In addition, the toxicity potential of the test article to human hepatocytes was tested using the 3‐[4,5‐dimethyl‐2‐thiazolyl]‐2,5‐diphenyl‐2H‐tetrazolium bromide (MTT) assay. The absorbance of formazan was measured on a plate reader at a wavelength of 570 nm.

PBPK model

The PBPK model was developed with the Simcyp population‐based simulator (version 13.2; Simcyp, Sheffield, UK) using hepatocyte CYP3A4 activity, oprozomib pharmacokinetics (PK) and physicochemical properties of oprozomib. Simcyp models complex systems using in vitro experimental data and demographic and physiological data from a specific patient population. The algorithm, physiological basis and differential equations used by the Simcyp software have been described previously 16, 17. Simulation parameters were based on clinical trial data and in vitro human hepatocyte studies (details described below). Oprozomib PK parameters have been reported previously 9. Oral plasma clearance and volume of distribution of oprozomib were derived from a population PK analysis (unpublished data on file, Amgen Inc., Thousand Oaks, CA, USA).

The PBPK model used downregulation data from hepatocyte incubation, rather than TDI data from microsome incubation, because the hepatocyte system is more relevant to epoxyketone metabolism 18. The suppression/inhibition parameters minimum CYP enzyme activity expressed as a fraction of vehicle control (E _min) and concentration resulting in 50% of maximum suppression (EC₅₀) were derived from the in vitro effect of oprozomib on CYP3A4‐catalysed testosterone 6β‐hydroxylase activity following treatment of cultured human hepatocytes for 3 days.

The clinical profile of oprozomib was modelled with a minimal PBPK model (i.e. with tissues with similar kinetic behaviour combined and a minimum number of physiological parameters) included. Specifically, the minimal PBPK model in Simcyp is a lumped PBPK model with only three compartments, predicting only the systemic, portal vein and liver concentrations. Default values from the Simcyp library were adopted for some parameters: the mean degradation constant (k _deg) values for CYP3A4 in the liver and intestine were 0.0193 and 0.03 h⁻¹, respectively 19.

It was assumed that the magnitude of oprozomib suppression for CYP3A4 in the intestine was the same as in the liver 20, 21. In addition, within the Simcyp simulator, it was assumed that oprozomib suppresses the rate of CYP3A4 synthesis in the liver according to the following equation 20, 22, 23:

$$\frac{d{\left[\mathit{\text{Enzyme}}\right]}_t}{\mathit{dt}}=\underset{\text{Rate of synthesis}}{\underbrace{{\mathrm{k}}_{\mathrm{deg}}•{\left[\mathit{\text{Enzyme}}\right]}_0}}•\underset{\text{Suppression factor}}{\underbrace{\left(1+\frac{\left({\mathrm{E}}_{min}-1\right)\mathrm{x}{\left[\mathrm{I}\right]}_t}{{\mathrm{EC}}_{50}+{\left[\mathrm{I}\right]}_t}\right)}}-\underset{\text{Rate of degradation}}{\underbrace{{\mathrm{k}}_{\mathrm{deg}}•{\left[\mathit{\text{Enzyme}}\right]}_t}}$$

where [Enzyme]_t is the amount of active CYP3A4 enzyme at any given time in the liver; [Enzyme]₀ is the basal amount of CYP3A4 enzyme in the liver; [Enzyme]_t = [Enzyme]₀ at t = 0; E _min is the minimum CYP3A4 enzyme activity (i.e. the maximum suppression) expressed as a fraction of vehicle control; EC₅₀ is the concentration that results in 50% of E _min (i.e. half of the maximum suppressive effect); [I]_t is the perpetrator (oprozomib) concentration at time t; and k _deg is the degradation rate of CYP3A4 enzyme in the liver.

The simulated oncology population size was 100 (10 trials with 10 virtual patients each), and was built using data from 2597 cancer patients 24. In addition, because a 28% reduction in CYP3A4 abundance in the liver and gut has been noted in the cancer population 25, 26, we adjusted our simulated oncology population in Simcyp accordingly. Virtual patients were 18–50 years of age, with a 1:1 male:female ratio. The key inputs to the PBPK model are shown in Supplementary Table S1. The suppressive effect was calculated as a ratio of the AUC or C _max of the victim drug in the presence vs. absence of oprozomib. Mean values for the trials and 90% confidence intervals (CIs) are reported.

Sensitivity analyses were conducted to assess the effect of a range of CYP3A4 EC₅₀ and oprozomib E _min (each varying by up to 10‐fold of the observed in vitro result) on the AUC or C _max ratios of midazolam. Three‐dimensional plots of AUC or C _max ratios vs. the ranges of EC₅₀ and E _min values were generated.

Clinical DDI study design

The evaluation of oprozomib and midazolam DDI described in this study is part of a larger, phase I, open‐label, multicentre study of oprozomib that assessed PK in patients with advanced haematologic and solid tumour malignancies (NCT02244112). The study was composed of a food effect/calculated QT interval assessment and the DDI part. Here we report only the oprozomib and midazolam DDI results from the study. The primary objective of the DDI study was to evaluate the PK parameters of midazolam, including C _max, AUC from time zero to time of last quantifiable concentration (AUC_0–last) and AUC to time infinity (AUC_0–inf) in the presence and absence of oprozomib. Additional midazolam PK parameters assessed included time to C _max (t _max) and half‐life (t _1/2); hydroxymidazolam PK parameters were also assessed. The secondary objective was to evaluate oprozomib safety and tolerability.

Clinical DDI study patients

Adult patients with a histologically confirmed diagnosis of advanced malignancies and no defined standard therapy were eligible. Other inclusion criteria included Eastern Cooperative Oncology Group performance status of 0–2 and adequate hepatic (bilirubin ≤1.5 times the upper limit of normal [ULN], alanine aminotransferase and aspartate aminotransferase ≤3 times ULN), renal (calculated or measured creatinine clearance ≥30 ml min⁻¹ by Cockroft‐Gault) and bone marrow (absolute neutrophil count ≥1000 mm⁻³, haemoglobin >7 g dl⁻¹, platelet count >30 000 mm⁻³) function.

Patients were excluded if they received strong P‐glycoprotein inhibitors ≤14 days or potent CYP3A4 inhibitors or inducers ≤7 or 14 days, respectively, prior to the first dose of oprozomib. In addition, concomitant medications, including moderate CYP3A inhibitors and inducers and strong P‐glycoprotein inhibitors and inducers, were prohibited during the PK portion of the study. Other exclusion criteria included myocardial infarction ≤6 months of enrolment, conduction abnormalities or active heart failure (New York Heart Association class III or IV); other uncontrolled intercurrent illness (e.g. diabetes); clinically significant gastrointestinal abnormality such as gastric reduction surgery, dysphagia or inability to swallow tablets, or pancreatic insufficiency; primary malignancies of the central nervous system; known human immunodeficiency virus positivity or hepatitis B surface antigen positivity, or suspected hepatitis C infection; or active infection requiring treatment ≤2 weeks of the first oprozomib dose. See Supplementary Methods for full inclusion and exclusion criteria.

The institutional review board of the investigational site reviewed and approved the clinical study protocol. The study was conducted in accordance with the Declaration of Helsinki. All patients provided written, informed consent.

Clinical DDI study dosing and PK assessments

The oprozomib and midazolam dosing schedules are shown in Figure 1. During period 1, patients received a single oral midazolam dose (2 mg) followed by a 3–6‐day washout. During period 2, patients received oprozomib 300 mg orally (150 mg ×2 extended‐release tablets) on days 1, 2, 8 and 9 of two consecutive 14‐day cycles. The 300‐mg dose of oprozomib was previously determined to be the MTD for patients with haematologic malignancies 7; this dose was chosen to maximize the possibility of demonstrating an interaction between the substrate and interacting drug, consistent with US Food and Drug Administration guidance 22. Oral dexamethasone (4 mg) and an oral 5‐hydroxytryptamine type‐3 (5‐HT3) antagonist were given 30–60 min prior to each oprozomib dose, to reduce the incidence and severity of gastrointestinal toxicity. Additional doses of the 5‐HT3 antagonist could be administered at the investigator's discretion.

Figure 1.

DDI study design: ^aThe first day of oprozomib administration was day 1 (there was no day 0). ^bThe washout is defined as a period to allow midazolam to be eliminated from the body, and when no drug is administered. ^cOn PK analysis days, patients fasted for 2 h before and after midazolam administration. C1D1, cycle 1 day 1; C2D2, cycle 2 day 2; DDI, drug–drug interaction; MDZ, midazolam; OPZ, oprozomib; PK, pharmacokinetic

Midazolam PK was assessed in period 1 and both oprozomib and midazolam PK were assessed in period 2; sampling times are shown in Figure 1. Patients must have received the planned dose of midazolam in period 1, and at least one midazolam dose and the first planned oprozomib dose in period 2 to be included for PK assessments. Plasma concentrations of midazolam, hydroxymidazolam and oprozomib were measured using a validated LC–MS/MS method. The intra‐ and inter‐day %CV for these analytes were ≤10%. The lower limit of quantification was 1.00 ng ml⁻¹ for oprozomib and 0.100 ng ml⁻¹ for midazolam and hydroxymidazolam, respectively. PK parameters were analysed by noncompartmental methods using Phoenix WinNonlin v.6.4 (Pharsight^®, St. Louis, MO, USA).

Statistical analysis

A sample size of 18 evaluable patients was estimated to allow for an approximately 65% probability that the 90% CI of the geometric mean ratio (GMR) for the midazolam PK parameters with vs. without oprozomib would lie between 0.80 and 1.25 if the true GMR = 1 (i.e. no DDI). The sample size was based on the assumption of a %CV of 30% for AUC_0–last, AUC_0–inf and C _max of midazolam.

The effects of oprozomib on midazolam PK parameters, AUC and C _max, were evaluated using mixed‐effects models on log (ln)‐transformed data with treatment (i.e. period) as a fixed effect and patients as a random effect. The 90% CIs for the GMRs of PK parameters for test (midazolam in the presence of oprozomib) vs. reference (midazolam in the absence of oprozomib) were calculated from the antilog of the corresponding CIs for the differences between the means on the log scale.

All patients who received at least one dose of either oprozomib or midazolam were included in the safety analysis.

Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY 27, and are permanently archived in the Concise Guide to PHARMACOLOGY 2017/18 28.

Results

In vitro data in human hepatocytes

The effect of oprozomib on CYP3A4 mRNA expression and enzymatic activity was evaluated following a standard 3‐day treatment of cryopreserved human hepatocytes (lots 295, 312 and 318). Rifampicin (RIF) was run in parallel and used as a positive control. RIF (10 μM) resulted in a 6.7‐fold increase in CYP3A4 mRNA expression and 4.7‐fold increase in enzyme activity, demonstrating suitability of the test system to assess the effect of test articles on CYP3A4 transcription and activity. When treated with oprozomib 0.003–10 μM, a concentration‐dependent decrease in CYP3A4 mRNA expression and enzyme activity was observed in hepatocytes (Figure 2). The half maximal inhibitory concentration (IC₅₀) ranged from 0.27 to 0.30 μM for CYP3A4 mRNA and from 0.20 to 0.41 μM for CYP3A4 activity for the three donors. At 1 μM, there was a >90% decrease in CYP3A4 mRNA expression and a >40% decrease in enzyme activity. Relative to solvent control, oprozomib did not affect the viability of hepatocytes following a 3‐day treatment at concentrations of ≤1 μM. At higher concentrations (≥3 μM for lot 318 and 10 μM for lot 295), hepatocyte viability was reduced to <70%, with morphological changes indicating cytotoxicity.

Figure 2.

In vitro effects of oprozomib: (A) Effect of oprozomib on CYP3A4‐catalysed testosterone 6b‐hydroxylase activity in normal human hepatocytes; (B) Effect of oprozomib on CYP3A4 mRNA in normal human hepatocytes. Results are expressed as the mean (n = 3) of fold ± SD. Bottom figures present data without positive controls. RIF‐10, rifampicin 10 μM

PBPK model predictions

The PBPK model of oprozomib was first optimized using oprozomib clinical PK data (unpublished data on file, Amgen Inc., Thousand Oaks, CA, USA), and a virtual oncology population was used to simulate the DDI effect of oprozomib. The model and approach were initially validated by comparing the simulation results to published carfilzomib‐midazolam DDI data 18, given that carfilzomib and oprozomib were expected to have similar DDI potential (unpublished data on file, Amgen Inc., Thousand Oaks, CA, USA). Oprozomib PK at the 300 mg dose (given once‐daily for two consecutive days [QDx2] every 7 days) was well captured in the simulation with regard to its absorption profile, peak concentration and elimination profiles (Figure 3). The PBPK model predicted maximum suppression of liver and gut CYP3A4 activities of 6.9% and 13.8%, respectively, following administration of oprozomib 300 mg (Figure S1). The suppression effect peaked at approximately 7–9 h after the second oprozomib dose, and returned to baseline prior to the initiation of the next treatment round of QDx2 every 7 days. The model also predicted oprozomib 300 mg should not cause a clinically relevant increase in the exposure to CYP3A4 substrates (predicted mean AUC ratio [90% CI] on cycle 2 day 2 after repeated dosing: midazolam, 1.11 [1.10–1.12]; simvastatin, 1.19 [1.17–1.20]; triazolam, 1.11 [1.10–1.12]; predicted mean C _max ratio [90% CI] on cycle 2 day 2 after repeated dosing: midazolam, 1.07 [1.06–1.07]; simvastatin, 1.16 [1.14–1.17]; triazolam, 1.06 [1.05–1.07]). The predicted in vivo oprozomib effect on midazolam AUC was <1.25‐fold.

Figure 3.

Observed vs. PBPK model predicted mean values of oprozomib (300 mg QDx2) systemic concentration in plasma over time. QDx2, once‐daily for two consecutive days. Observed: squares. Predicted: solid line

Sensitivity analyses were conducted to assess the effect of a range of EC₅₀ (0.03–3.20 μM) and the E _min (0.003–1.0) on the AUC or C _max ratios of midazolam (Figure S2). These sensitivity analyses predicted low DDI potential. In a scenario in which both EC₅₀ and E _min were lowered by 10‐fold vs. the in vitro results, predicted midazolam AUC and C _max ratios were 1.34‐fold and 1.18‐fold, respectively.

Clinical DDI study results

Patient demographics

A total of 21 patients were enrolled. Baseline demographic and disease characteristics are shown in Table 1.

Table 1.

Patient and disease characteristics at baseline

Characteristic	n = 21
Age, years
Median (range)	61.0 (23–85)
Male sex, n (%)	13 (61.9)
Race, n (%)
Caucasian	16 (76.2)
African‐American	1 (4.8)
Other	4 (19.0)
Years since diagnosis
Median (range)	3.20 (0.7–9.8)
ECOG PS, n (%)
0	5 (23.8)
1	15 (71.4)
2	1 (4.8)
Type of malignancy, n (%)
Advanced solid tumour	18 (85.7)
Multiple myeloma	1 (4.8)
Non‐Hodgkin's lymphoma	1 (4.8)
Other	1 (4.8)

ECOG PS, Eastern Cooperative Oncology Group performance status

Pharmacokinetics of midazolam with or without oprozomib

The observed data from the DDI study showed minimal changes in exposure (AUC or C _max) to midazolam in the presence of oprozomib (Table 2 and Figure 4), as predicted by the PBPK model. The mean (%CV) for midazolam AUC_0–last was 57.8 (67.7%), 62.1 (54.0%) and 46.0 (43.9%) h ng⁻¹ ml⁻¹ at baseline (midazolam alone), cycle 1 day 1 (single dose oprozomib) and cycle 2 day 2 (repeated dose oprozomib), respectively. The mean (%CV) for midazolam C _max was 15.4 (63.9%), 20.1 (59.6%) and 15.9 (56.2%) ng ml⁻¹ at baseline, cycle 1 day 1 and cycle 2 day 2, respectively. Similar results were observed for the major metabolite of midazolam, hydroxymidazolam; no marked differences in exposure to hydroxymidazolam were observed with or without oprozomib coadministration.

Table 2.

PK parameters of midazolam after administration of midazolam with or without oprozomib in the DDI study

PK Parameters	Period 1: Midazolam alone		Cycle 1 Day 1: Midazolam + oprozomib		Cycle 2 Day 2: Midazolam + oprozomib		Mixed effects model: GMR b (90% geometric CI c ) Midazolam + oprozomib vs. midazolam alone
	n	Mean (% CV)	n	Mean (% CV)	n	Mean (% CV)	Single dose	Repeated dose
AUC 0–last (ng h ml −1 )	21	57.8 (67.7)	20	62.1 (54.0)	15	46.0 (43.9)	1.17 (1.02–1.35), n = 20	0.93 (0.78–1.10), n = 15
AUC 0–inf (ng h ml −1 )	20	59.4 (68.9)	16	61.5 (58.6)	14	48.6 (43.5)	1.28 (1.15–1.42), n = 16	0.98 (0.85–1.14), n = 14
C max (ng ml −1 )	21	15.4 (63.9)	20	20.1 (59.6)	15	15.9 (56.2)	1.33 (1.16–1.53), n = 20	1.14 (0.93–1.40), n = 15
t max (h)	21	0.52 (0.25–1.0)a	20	0.50 (0.23–4.0)a	15	0.27 (0.25–2.0)a	–	–
t 1/2 (h)	20	5.57 (33.2)	16	5.35 (37.3)	14	4.77 (43.0)	–	–

AUC, area under the plasma concentration–time curve; AUC_0–inf, AUC from time zero to infinity; AUC_0–last, AUC from time zero to time of last quantifiable concentration; C _max, maximum observed drug concentration; CV, coefficient of variation; GMR: geometric mean ratio; PK, pharmacokinetic; t _1/2, half‐life; t _max, time to maximum concentration

^aData presented as median (minimum–maximum)

^bCalculated using least‐squares means (ln‐transformed data) according to the formula e^{(Treatment a – Treatment b)} × 100

^c90% geometric confidence interval (CI) using ln‐transformed data

Single dose data were assessed on cycle 1 day 1 and repeated dose data were assessed on cycle 2 day 2

Figure 4.

Plasma midazolam concentration–time profiles after oral administration of midazolam with and without oprozomib.^a N‐values varied by time point. MDZ, midazolam; OPZ, oprozomib

The mixed‐effects modelling results are shown in Table 2. Consistent with the high PK variability (with %CV ranging from 43.9% to 68.9%), the 90% CIs of the GMRs were relatively wide. On average, the mean midazolam ratio for AUC and C _max (oprozomib/without oprozomib) was no greater than 1.33 after single or repeated oprozomib dosing. There was generally a good concordance between the PBPK model prediction and the observed data from the DDI clinical study (Table 3), consistently showing a maximal increase in midazolam exposure (AUC or C _max) of no more than 33% after oprozomib treatment. The wide CIs of the GMRs were not accurately reflected by the PBPK model using simulations for 100 oncology patients. Nevertheless, the results from the clinical DDI study were similar to those predicted by the PBPK model, in that oprozomib did not cause a relevant change in CYP3A4 activity following single or repeated oprozomib dosing.

Table 3.

Comparison of mean ratios (midazolam + oprozomib vs. midazolam alone) by the PBPK model to the observed data in the oprozomib DDI study

		Single dose	Repeated dosing
AUC 0–inf	PBPK prediction	1.04 (1.03–1.04)	1.11 (1.10–1.12)
	Observed data	1.28 (1.15–1.42)	0.98 (0.85–1.14)
C max	PBPK prediction	1.02 (1.01–1.02)	1.07 (1.06–1.07)
	Observed data	1.33 (1.16–1.53)	1.14 (0.93–1.40)

PBPK predicted values are the mean ratio (90% CI) of simulation for 100 oncology patients after single dose on cycle 1 day 1 (C1D1) or repeated dosing on cycle 2 day 2 (C2D2). For the observed data in the DDI study, expressed as GMR (90% CI), single dose data were assessed on cycle 1 day 1 and repeated dose data were assessed on cycle 2 day 2

AUC, area under the plasma concentration–time curve; AUC_0–inf, AUC from time zero to infinity; CI, confidence interval; C _max, maximum observed drug concentration; DDI, drug–drug interaction; GMR, geometric mean ratio; PBPK, physiologically‐based pharmacokinetic

Safety

The mean (standard deviation) number of oprozomib doses was 16.0 (20.0) and the median number of cycles was 2 (range: 1–20 cycles). Adverse events (AEs) are shown in Supplementary Table S2. All patients were reported to have ≥1 AE during treatment. Grade 3 or greater AEs were reported in 13 (62%) patients and serious AEs were reported in 10 (48%) patients.

Sixteen (76%) patients reported oprozomib treatment‐related AEs (grade ≥3, 4 [19%]); serious, 2 [10%]). The most frequently reported treatment‐related AEs (defined as any AEs not designated by the investigator to be unrelated to treatment) were nausea (71%), diarrhoea (43%), vomiting (38%) and fatigue (33%). The following were reported in >1 patient: abdominal pain, decreased appetite and increased alanine aminotransferase. No patient discontinued oprozomib due to a treatment‐related AE. Three deaths occurred within 30 days of the last dose of investigational product. All were considered related to cancer progression.

Discussion

Similar to carfilzomib, oprozomib is a rare example of a small molecule drug that suppresses CYP mRNA. This unusual property elicits uncertainty regarding the optimal approach for predicting its DDI risk and related questions regarding clinical study design elements during early clinical development (e.g. inclusion/exclusion criteria, considerations for combining with other study drugs). To address these questions, we first conducted in vitro studies in human hepatocytes, and incorporated the results into PBPK models to predict the CYP3A4‐mediated DDI potential of oprozomib. We report results of an oprozomib clinical DDI study, which confirmed the PBPK model prediction that oprozomib 300 mg would not cause a clinically relevant change in exposure to CYP3A4 substrates. The results demonstrate the utility of primary hepatocyte data in conjunction with PBPK modelling to predict clinical CYP3A4 DDI prospectively for small molecule drugs that affect CYP3A mRNA expression.

Because very few small molecule drugs on the market or in clinical development cause CYP mRNA suppression, the field for prospectively predicting the DDI potential of such drugs is still evolving. A PBPK approach has been used successfully to predict the effect of mRNA suppression of CYP enzymes by cytokines, in particular interleukin 6, on the PK of other drugs 16, 20. However, the utility of a PBPK model incorporating primary hepatocytes data for DDI prediction has not been established for small molecule drugs that cause CYP3A mRNA suppression.

The potential of an investigational drug to inhibit CYP enzymes is commonly studied by an initial in vitro screening using HLMs to determine its inhibition mechanism (e.g. reversible or TDI) and inhibition potency (e.g. inhibitory constant [K_i]) 22. With HLMs, oprozomib was shown to be a weak time‐dependent inhibitor of CYP3A4 (IC₅₀ > 30 μM for reversible inhibition; IC₅₀ reduced to 5.3 and 12 μM, respectively, with testosterone and midazolam as the CYP3A4 substrate and a 30‐min oprozomib pre‐incubation with HLMs in the presence of nicotinamide adenine dinucleotide phosphate [NADPH]) 10. The TDI of oprozomib on CYP3A4/5 observed in HLMs was NADPH‐dependent, suggesting it is likely due to CYP‐mediated metabolite(s). However, in vitro and in vivo studies on oprozomib have shown that in humans it is primarily metabolized by epoxide hydrolases rather than CYP‐mediated metabolism 10. Incubation in HLMs may overestimate CYP‐mediated oprozomib metabolism and results may not translate to humans in vivo. PBPK modelling using oprozomib TDI kinetic parameters (K_i and rate of enzyme inactivation derived from in vitro studies in HLMs) suggested an approximate 1.5 to 1.9‐fold increase in exposure for midazolam and simvastatin (data not shown). This approach would have overestimated the observed clinical DDI risk of oprozomib. In contrast, PBPK modelling of clinical DDI risk using data from in vitro hepatocyte studies showed close concordance with the clinical DDI study results, indicating that cultured human hepatocytes may be a more reliable system for DDI prediction than HLMs for this class of compounds. These findings are consistent with what has been observed with carfilzomib, another tetrapeptide epoxyketone proteasome inhibitor that also suppresses CYP3A mRNA. Despite a TDI effect observed in vitro in HLMs, carfilzomib did not affect CYP3A4 activity clinically 22, and human hepatocytes better predicted this lack of DDI in vivo 18. HLM‐based prediction has been observed to substantially overestimate clinical DDI for other classes of drugs as well, and prediction using hepatocyte data could provide an alternative approach 16.

Although the mechanism of mRNA suppression by oprozomib is unknown, oprozomib downregulated mRNA expression and reduced the enzymatic activities of CYP1A2 and 2B6 in cultured human hepatocytes, in addition to its effect on CYP3A4. These effects are similar to those previously observed for carfilzomib 18. The decrease in CYP3A mRNA expression is not due to effects on cell viability, which was unchanged in the presence of oprozomib at the testing concentrations. Many CYP enzymes are transcriptionally suppressed by inflammatory stimuli, and downregulation by cytokines can also occur by post‐transcriptional mechanisms via nitric oxide (NO) and proteasomal degradation‐dependent pathways 29. While we did not measure CYP3A protein levels in our study, proteasome inhibitors can reduce CYP3A protein levels 30. The relationship between inducible NO synthase, nuclear factor kappa‐light‐chain‐enhancer of activated B cells activation, and proteasomal activity 31 and the interplay between NO synthase and pregnane X receptor 32 could potentially account for the mRNA modulation observed. Additional studies are needed to elucidate the molecular mechanism of CYP mRNA suppression by oprozomib.

It should be noted that due to the clastogenicity potential of oprozomib, the clinical DDI study was conducted in oncology patients instead of healthy volunteers. Unlike studies in healthy volunteers, high PK variability was observed in this study, which is not well captured by the PBPK model. While this study excluded those who used strong P‐glycoprotein inhibitors or strong/moderate CYP3A inhibitors or inducers, patients were allowed to use a low dose of dexamethasone (4 mg), which is a weak CYP3A4 inducer 33, and an oral 5‐HT3 antagonist to improve tolerability. In addition, proton pump inhibitors such as lansoprazole, and the antidiarrhoeals loperamide or diphonoxylate/atropine, were allowed in the study. It has been documented that PK variability is generally higher in cancer patients than those of healthy volunteers 24; PK variability in the cancer patients of this study could be attributed to concomitant medications, or to concurrent diseases possibly leading to impaired hepatic function, which could impact the absorption, distribution and metabolism of oprozomib. The observed data indicated a trend towards decreased midazolam exposure (AUC and C _max) upon repeated oprozomib dosing vs. observations after a single dose. The underlying mechanism for this observed trend is unclear, but is likely due to the high PK variability and small sample size in this study; thus, this trend is not captured by the PBPK model.

The PBPK approach illustrated here demonstrates an example of the learning and confirming paradigm to enhance the efficiency of clinical development as proposed by Sheiner et al. 15. Investment in developing a PBPK model early (prior to a dedicated clinical DDI study) has been valuable in supporting clinical development of oprozomib. Preliminary understanding of the CYP3A4 DDI risk provided by the PBPK model was used to inform the inclusion/exclusion criteria of multiple oprozomib trials, and to support the clinical study design of a combination study with other agents that are sensitive substrates of CYP3A4. This validated PBPK model can now be used to support future clinical programmes, including studies to understand the DDI potential in different clinical scenarios, including various dose regimens or formulations.

Competing Interests

Y.O. is a former employee of Amgen, Inc., a current employee of BeiGene, and is a stockholder in Amgen, Inc. and BeiGene. Y.X. is a current employee of and stockholder in Amgen, Inc. L.G. has received consulting fees for advisory board service to Amgen, Inc., Novartis and Roche, and for data safety and monitoring board membership/service to Celgene and Novartis; and holds stock valued at less than $20 000 in Amgen, Inc., Celgene, Clovis Oncology and Sanofi Pasteur, and stock valued at greater than $20 000 in Incyte. R.D.H. has received research grants from Amgen, Inc. A.M. has nothing to disclose. K.P.P. reports funding provided to South Texas Accelerated Research Therapeutics San Antonio for the conduct of clinical trials from 3D Medicines, Abbvie, ADC Therapeutics, Amgen, Inc., ArQule Inc., Calithera Biosciences, Curagenix, Daichi Sankyo, EMD Serono, Incyte, Medimmune, Merck, Peloton Therapeutics and Regeneron. Z.W. is a former employee of and current stockholder in Amgen, Inc. R.E.C. is a former employee of Onyx Pharmaceuticals, Inc. D.E.P. is a current employee of Aileron Therapeutics, Inc., and a stockholder in Aileron Therapeutics, Inc., Amgen, Inc. and Loxo Oncology. A.M.T. has received research grants from Baxalta, Bayer, Boston Biomedical, EMD Serono, Foundation Medicine, Karus Therapeutics, Immatics, Onyx Pharmaceuticals, Inc. and Placon Therapeutics; and has served on the speakers' bureau for Roche Europe.

The authors wish to thank all of the patients, families, caregivers, investigators, research nurses, study coordinators and support staff who contributed to this study.

This study was supported by Onyx Pharmaceuticals, Inc., an Amgen subsidiary, South San Francisco, CA. Medical writing and editorial assistance was provided by BlueMomentum, an Ashfield company, part of UDG Healthcare PLC, and funded by Amgen, Inc.

Contributors

Y.O., Y.X., L.G., R.D.H., A.M., K.P.P., Z.W., R.E.C., D.E.P. and A.M.T. wrote the manuscript. Y.O., Y.X., L.G., R.D.H., R.E.C., D.E.P. and A.M.T. designed the research. Y.O., Y.X., L.G., R.D.H., A.M., K.P.P., Z.W., R.E.C., D.E.P. and A.M.T. performed the research. Y.O., Y.X., L.G., R.D.H., A.M., K.P.P., Z.W., R.E.C., D.E.P. and A.M.T. analysed the data.

Supporting information

Table S1 Key oprozomib parameters for the PBPK model

Table S2 Incidence of adverse events

Figure S1 Predicted effect of 2 once‐daily administrations of 300 mg oprozomib (QDx2 every 7 days) on CYP3A4 intrinsic clearance in liver and small intestine

Figure S2 Automated parameter sensitivity analysis

Ou, Y. , Xu, Y. , Gore, L. , Harvey, R. D. , Mita, A. , Papadopoulos, K. P. , Wang, Z. , Cutler, R. E. Jr , Pinchasik, D. E. , and Tsimberidou, A. M. (2019) Physiologically‐based pharmacokinetic modelling to predict oprozomib CYP3A drug–drug interaction potential in patients with advanced malignancies. Br J Clin Pharmacol, 85: 530–539. 10.1111/bcp.13817.