The strong clinical interaction between bupropion and CYP2D6 is primarily mediated through bupropion metabolites and their stereoisomers: A paradigm for evaluating metabolites in drug-drug interaction risk

Abstract

Bupropion (BUP) is a potent clinical inhibitor of CYP2D6, although the specific mechanisms underlying this interaction are not fully understood. We comprehensively evaluated the inhibition potencies of racemic BUP and its stereoisomers, as well as its major metabolites (4-hydroxybupropion [OHBUP], threohydrobupropion [THBUP], and erythrohydrobupropion [EHBUP]), on CYP2D6-mediated dextromethorphan O-demethylation in pooled human liver microsomes. The K_i value for racemic EHBUP was 5.5-, 11.4-, and 13-fold lower than those for THBUP, OHBUP, and BUP, respectively. THBUP demonstrated over 2-fold greater potency than OHBUP and BUP. Additionally, RR-THBUP had a 2.1-fold lower K_i value than SS-THBUP, while S-BUP and RR-OHBUP exhibited 3.0-fold and 1.5-fold lower K_i values than R-BUP and SS-OHBUP, respectively, indicating modest stereoselective inhibition. The K_i values of stereoisomers EHBUP were comparable. Using a mechanistic static interaction model that incorporated in vitro K_i value and unbound steady-state concentration in plasma (C_max) or estimated liver concentrations of each inhibitor, we found significant underprediction of the observed clinical BUP-CYP2D6 interaction, indicating that no single inhibitor can predict observed in vivo BUP-CYP2D6 interaction. Accurate predictions of observed clinical interactions were achieved using all racemic (within 11.6%) or all stereoisomeric forms (within 5.4%) of BUP and its metabolites, along with their liver concentrations (but not plasma concentrations). Our findings highlight the crucial role of circulating BUP metabolites, particularly the summation of EHBUP and THBUP or their stereoisomers, in the in vivo inhibition of CYP2D6 by BUP. These data provide mechanistic and quantitative insight into the partially understood clinical CYP2D6-dependent interactions associated with BUP.

Significance Statement

This article describes comprehensively the inhibition of CYP2D6 by racemic and stereoisomers of bupropion (BUP) and its main metabolites in pooled human liver microsomes. The strong interaction between BUP and CYP2D6 observed clinically is mainly due to BUP metabolites and inhibition is stereospecific. Accounting for inhibition constants and steady-state unbound estimated liver (but not plasma) concentrations of all racemic or all stereoisomers accurately predicted the clinically observed BUP-CYP2D6 interactions.

1. Introduction

Bupropion (BUP) is a dual dopamine-norepinephrine reuptake inhibitor and nicotinic acetylcholine receptor antagonist widely used for treating major depression and seasonal affective disorder, and as an aid for smoking cessation (Dwoskin et al, 2006), in combination with naltrexone, for managing weight in obese patients (Onakpoya et al, 2020), and in combination with dextromethorphan (DEX) for treating major depression (Keam, 2023).

However, BUP causes clinically significant drug-drug interactions (DDIs) via inhibition of CYP2D6 inhibition, with a reported increase of 5.1- to 7.2-fold for drugs such as DEX (Kotlyar et al, 2005; Duong et al, 2020), desipramine (Reese et al, 2008), nebivolol (Gheldiu et al, 2016), and atomoxetine (Todor et al, 2016). Clinical case reports and some studies indicate that coadministration of BUP with CYP2D6 substrates increases the risk for toxicity (Weintraub, 2001; McCollum et al, 2004; Paslakis et al, 2010) or lack of efficacy (Fulton et al, 2019). However, despite several in vitro investigations, a disconnect remains between predicting the observed in vivo DDI magnitude from in vitro findings. This may be in part due to incomplete understanding of the mechanisms by which BUP decreases CYP2D6 activity in vivo.

Inhibition studies in human liver microsomes (HLMs) and hepatocytes consistently demonstrate that BUP is a weak inhibitor of CYP2D6, with reported IC₅₀ values ranging from 18 μM to 79 μM and a K_i value of approximately 21 μM (Hesse et al, 2000; Reese et al, 2008; Parkinson et al, 2010). These values are over 70-fold higher than the steady-state BUP plasma concentration (C_max, ∼ 0.25 μM) (Benowitz et al, 2013). This significant discrepancy between in vitro and in vivo inhibition of CYP2D6 by BUP suggests that other factors may play a major role.

In humans, BUP undergoes extensive metabolism and is primarily excreted in urine as metabolites (84%) with minimal parent drug (<1%) (Welch et al, 1987). Among the metabolites, CYP2B6-mediated oxidation to 4-hydroxybupropion (OHBUP) (Hesse et al, 2000) and reduction to 2 amino-alcohol isomers—threohydrobupropion (THBUP) and erythrohydrobupropion (EHBUP)—by mainly hepatic microsomal 11β-hydroxysteroid dehydrogenase type 1 (Meyer et al, 2013; Bamfo et al, 2023) and to a small extent by aldo-keto reductases (Skarydova et al, 2014) are pharmacologically active, contributing to both the beneficial effects and potential toxicity of BUP (Damaj et al, 2004; Silverstone et al, 2008; Costa et al, 2019). Steady-state plasma exposures of these metabolites following BUP therapy significantly exceed those of BUP itself (Daviss et al, 2005; Benowitz et al, 2013; Kharasch et al, 2019, 2020). In a previous in vitro study (Reese et al, 2008), racemic EHBUP and THBUP exhibited 12- and 4-fold higher inhibitory potency against CYP2D6, respectively, compared with BUP, linking circulating BUP metabolites to the in vivo interaction of BUP with CYP2D6 substrates. Subsequent in vitro (Sager et al, 2017) and retrospective analysis (Yeung et al, 2011; Steinbronn et al, 2021) suggested that circulating metabolites of BUP may be responsible for the observed in vivo BUP-CYP2D6 interactions. Indeed, incorporating all in vitro inhibition constants of BUP and its metabolites and their steady-state exposure in liver via static models (Reese et al, 2008) significantly improved prediction of in vivo DDIs, but underprediction still exists (predicted, ∼3.4-fold, vs observed 5.2-fold increase of desipramine area under the plasma concentration-time curve [AUC] ratio) (Reese et al, 2008).

BUP is clinically administered as a racemic mixture of its R and S forms, and additional chiral center is introduced during biotransformation, generating 6 diastereomers (SS- and RR-OHBUP, SS- and RR-THBUP, and RS- and SR-EHBUP) (Sager et al, 2016; Bamfo et al, 2023) (Fig. 1), each with unique stereoselective disposition and pharmacological profiles (Masters et al, 2016a; Costa et al, 2019; Kharasch et al, 2020). Evidence for stereoselective inhibition was suggested by the finding that EHBUP was approximately 3- to 4-fold more potent inhibitor of CYP2D6 than its isomer THBUP (Reese et al, 2008; Sager et al, 2017). In addition, the IC₅₀ values for the inhibition of CYP2D6 by S-BUP was shown to be 13.7-fold lower than that of R-BUP, while RR-OHBUP was more potent (2.8-fold) than SS-OHBUP (Sager et al, 2017). We have chromatographically separated and identified diastereomers of THBUP and EHBUP (Masters et al, 2016a,b). Yet, the stereoselective inhibition of CYP2D6 by the diastereomers of THBUP and EHBUP remain unexplored.

Fig. 1.

Human stereoselective metabolism of BUP to its main active metabolites. Enzymes involved include CYP2B6, 11β-hydroxysteroid dehydrogenase 1 (11β-HSD1), and aldo-keto reductases (AKRs).

In this study, we conducted a comprehensive in vitro evaluation to test the hypothesis that BUP metabolites and their stereoisomers exhibit potent inhibition of CYP2D6. We further hypothesized that integrating their inhibitory potencies with steady-state systemic exposure (either plasma or liver) accurately predicts the clinical BUP-CYP2D6 interaction. We determined the in vitro inhibitory potencies (IC50 and K_i values) of racemic BUP, OHBUP, THBUP, and EHBUP, along with their stereoisomers (R- and S-BUP, RR- and SS-OHBUP, RR- and SS-THBUP, and SR- and RS-EHBUP), on CYP2D6 activity using DEX O-demethylation as the probe reaction in pooled HLMs.

2. Materials and methods

2.1. Chemicals

BUP, R-BUP, S-BUP, OHBUP, RR-OHBUP, SS-OHBUP, THBUP, RR-THBUP, SS-THBUP, racemic EHBUP, RS-EHBUP, and SR-EHBUP were purchased from Toronto Research Chemicals (Toronto, Canada). The internal standard, codeine, was obtained from Sigma-Aldrich. Paroxetine was supplied by Clinical Pharmacology Analytical Core, Indiana University School of Medicine. Nevirapine (internal standard) was bought from MedChemExpress. Pooled HLMs from 50 donors with mixed biological sex (average age, 47 years; range, 5–83 years; 20 mg/mL) was purchased from XenoTech LLC and stored at −80 °C until use. DEX hydrobromide and dextrorphan (DXR) were purchased from Research Biochemicals International. NADPH was purchased from Sigma. Na₂HPO₄ was purchased from Mallinckrodt, and NaH₂PO₄ was purchased from EM Science. Dulbecco’s phosphate buffer saline was from Gibco. All chemicals and solvents were of high-performance liquid chromatography (HPLC) grade or higher. Acetonitrile, water, methanol, acetic acid, and ethyl acetate were purchased from Fisher Scientific Company LLC and were of high HPLC grade.

2.2. Liquid chromatography with tandem mass spectrometry analytical method development

The analytical method aimed at separating and quantifying the formation of DXR from DEX in incubations of pooled HLMs was initially developed using an API 3200 triple quadrupole mass spectrometer (Applied Biosystems). This was coupled to a HPLC system consisting of 2 LC-20AD pumps, a SIL-20AHT UFLC autosampler, a DGU-20A3 degasser, and a CBM-20A system controller (Shimadzu). Data acquisition and processing were conducted using Analyst software (version 1.5.1; AB SCIEX). Global parameters were optimized across all analytes: curtain gas, 25 psig; ion spray voltage, 4500 V; source temperature, 600 °C; gas 1, 60 psig; and gas 2, 50 psig. Chromatographic separation used a Phenomenex Luna C18(2) analytical column (150 × 4.6 mm i.d.; 5-μm particle size) and a mobile phase consisting of acetonitrile and 5 mM ammonium acetate at pH 3.5, with the following gradient: starting with 100% mobile phase A, followed by a linear gradient to 100% mobile phase B between 0.01 and 8 minutes. Mobile phase B was maintained at 100% until 10 minutes, followed by re-equilibration to initial conditions from 10.1 to 13 minutes, using a flow rate of 0.8 mL/min. This API 3200 method was used for initial method development, kinetic studies, and IC₅₀ determination.

Subsequent assays for determination of K_i values were performed using a QTRAP 6500+ mass spectrometer (Applied Biosystems), coupled to a ultrahigh-performance liquid chromatography (UHPLC) system (Shimadzu) equipped with 2 AD pumps, an AD autosampler, and an AD column oven. Data acquisition was managed using Analyst software version 1.7.0 with MultiQuant 3.0.2 quantification software. Parameters for the QTRAP 6500+ included the following: curtain gas, 25 psig; ion spray voltage, 5500 V; source temperature, 500 °C; gas 1, 15 psig; and gas 2, 20 psig. A Restek C8 column (250 × 4.6 mm i.d.; 5-μm particle size) was used, and the mobile phase composition was identical to that used with the API 3200 system.

Both (API 300 and QTRAP 6500+) mass spectrometers were operated with an electrospray ionization source in the positive mode with the following multiple reaction monitoring (MRM) transitions: the mass-to-charge ratio (m/z) of DEX, DXR, and the internal standard (codeine) was 258.3/157, 272.4/215.2, and 300.2/165, respectively.

2.3. Determination of kinetic of CYP2D6-mediated DEX O-demethylation

Kinetic analysis was conducted under linear conditions of microsomal protein and incubation time before initiating inhibition experiments with BUP and its metabolites. DEX concentrations ranging from 0 to 100 μM were incubated in duplicate with pooled HLMs at a protein concentration of 0.1 mg/mL. The incubation buffer consisted of phosphate buffer (0.2 M Na₂HPO₄ and NaH₂PO₄, pH 7.4). Reaction mixtures (final volume 150 μL) were preincubated for 5 minutes at 37 °C. Reactions were initiated by adding 100 μL of a NADPH (final concentration 1 mM). After incubating for 20 minutes at 37 °C, reactions were terminated by adding 500 μL of acetonitrile containing 1000 ng/mL of codeine as an internal standard. The samples were vortexed vigorously for 20 seconds and left on ice for 5 minutes, followed by centrifugation at 14,000 rpm for 5 minutes. The supernatant was transferred to cell culture tubes and extracted with 6 mL of ethyl acetate and 300 μL of glycine-sodium hydroxide buffer (1 M, pH 11.3) by shaking for 15 minutes on a shaker (Eberbach). Samples were evaporated for 2 hours, and residues were reconstituted with 60 μL of methanol from which 10 μL was injected into liquid chromatography with tandem mass spectrometry (LC-MS/MS) systems for DXR analysis.

2.4. Determination of half-maximum inhibitory concentrations

Pilot experiments were conducted to estimate IC₅₀ values, providing initial insights into the relative potency of inhibition. A single-isoform–specific substrate (DXM, 5 μM), corresponding to its respective K_m value, was preincubated for 5 minutes with HLMs (0.5 mg/mL) in phosphate buffer (0.2 M Na₂HPO₄ and NaH₂PO₄, pH 7.4). Each test inhibitor (BUP, R-BUP, S-BUP, OHBUP, RR-OHBUP, SS-OHBUP, THBUP, RR-THBUP, SS-THBUP, EHBUP, SR-EHBUP, and RS-EHBUP) was added at final concentrations ranging from 0 to 200 μM. Reactions were started by adding 100 μL of NADPH (final concentration 1 mM). After incubating for 20 minutes at 37 °C, reactions were stopped by adding acetonitrile containing 1000 ng/mL codeine as an internal standard. The samples were vortexed vigorously for 20 seconds and left on ice for 5 minutes, followed by centrifugation at 14,000 rpm for 5 minutes to separate the supernatant. The supernatant layer was transferred into cell culture tubes and extracted with 6 mL of ethyl acetate and 300 μL of glycine-sodium hydroxide buffer (1 M, pH 11.3) by shaking for 15 minutes in a shaker (Eberbach Co). Supernatants were evaporated for 2 hours and residues were reconstituted with 60 μL of methanol from which 10 μL was injected into LC-MS/MS systems for DXR analysis.

For direct, competitive, reversible inhibition, IC₅₀ values are useful for a quick comparison of enzyme inhibitors and can provide initial qualitative insights into the inhibitory potential. However, predicting DDI risks in vivo based solely on these values may be generally limited as they are highly dependent on the experimental conditions and substrate concentration selected. Therefore, additional experiments were conducted to estimate a more quantitative measure, the dissociation (inhibition) constants (K_i values). While IC₅₀ values can be used to estimate K_i values for direct, reversible inhibition (IC₅₀/2) under certain circumstances (Cheng and Prusoff, 1973; Haupt et al, 2015), the experimental conditions to determine IC₅₀ values need to be carefully designed, and the mechanisms of inhibition must be understood. The initial IC₅₀ values obtained from the pilot study and the kinetic data generated were simulated to inform the selection of optimal concentration ranges of DEX and each inhibitor for constructing Dixon plots and accurately estimating K_i values for the inhibition of CYP2D6 by BUP and its metabolites in HLMs.

2.5. Determination of K_i values

Six concentrations of DEX corresponding to 0.25×, 0.5×, 1×, 2×, 4×, and 8× the K_m value were incubated with pooled HLMs (0.1 mg/mL), along with cofactors, in duplicate, without and with multiple concentrations (0×, 0.3125×, 0.625×, 1.25×, 2.5×, and 5×) of each racemic or stereoisomers of BUP and its metabolites (BUP, R-BUP, S-BUP, OHBUP, RR-OHBUP, SS-OHBUP, THBUP, RR-THBUP, SS-THBUP, EHBUP, RS-EHBUP, or SR-EHBUP). We acknowledge that triplicate incubations could enhance data robustness but chose duplicates due to time and resource constraints, as the experiments require substantial amounts of HLMs and reagents. Dixon plots for 12 compounds (4 racemic mixtures and 8 stereoisomers) were generated using 6 concentrations of both substrate and inhibitor, each in duplicate. A matrix-matched standard curve was included, and rigorous quality control measures were applied to ensure reproducibility and reliability of the duplicate data.

Each incubation reaction was conducted in a 96-well plate format. A total of 130 μL of mixture, consisting of 0.1 mg/mL HLMs, substrate, and/or inhibitor suspended in sodium phosphate buffer (0.2 M Na₂HPO₄/NaH₂PO₄, pH 7.4), was preincubated for 5 minutes in an incubator set at 37 °C. After preincubation, the reactions were initiated by adding 20 μL of NADPH solution (1 mM final concentration). The reactions proceeded for 5 minutes under the same incubation conditions. To terminate the reactions, 300 μL of ice-cold methanol containing 1000 ng/mL codeine as an internal standard was added to each well. The samples were then vigorously vortexed for 20 seconds to ensure proper mixing, followed by centrifugation at 3000 rpm for 20 minutes and 5 μL of the supernatant was injected into an LC-MS/MS system for DXR analysis.

2.6. Microsomal protein-binding assays

We performed in vitro experiments to assess microsomal binding of each test inhibitor using equilibrium dialysis. Stock solutions (1 mg/mL) of BUP, metabolites, and paroxetine were dissolved in methanol and diluted in PBS buffer to final concentrations of 10 μM for BUP and metabolites and 1 μM for paroxetine as a positive control. HLM was diluted in PBS buffer to 0.1 mg/mL. We added 120 μL of each drug solution and 30 μL of diluted HLM to the protein chamber and 150 μL of PBS buffer to the buffer chamber. The dialysis plate was covered with adhesive film and incubated at 37 °C with shaking at 50 rpm for various times to equilibrate. Incubation was performed on a high-throughput dialysis 96B well plate (Fisher Scientific). The process ended by transferring 100 μL of incubation solution to a 96-well plate (0.65 mL tubes) containing 10 ng/mL nevirapine (internal standard) in 300 μL of ice-cold methanol. The mixture was shaken for 2 minutes at 2500 rpm and then centrifuged for 20 minutes at 4 °C. Next, 140 μL of supernatant was transferred to a new plate, and 5 μL was injected into the UHPLC/tandem mass spectrometry (MS/MS) system. Preliminary assays were performed at 10 μM of each inhibitor and 0.1 mg/mL pooled HLMs to determine stability, nonspecific binding, and optimal equilibration time. Compounds were stable (>80%) with minimal nonspecific binding after a 5-hour incubation. Specific microsomal protein-binding assays were then conducted with 10 μM of each inhibitor and pooled HLMs (0.1 mg/mL) for 5 hours in triplicate. Paroxetine (1 μM) served as a positive control.

Chromatographic separation of stereoisomers of BUP, OHBUP, EHBUP, and THBUP was achieved using a Phenomenex AMP LC column (150 × 4.6 mm; 3.0 μm). Paroxetine was eluted with a Sigma Discovery C18 column (150 × 4.6 mm; 5.0 μm). A mobile phase consisted of A (5 mM ammonium bicarbonate, pH 11 for BUP, and metabolites or 5 mM ammonium acetate for paroxetine) and B (methanol) delivered at a gradient flow rate of 0.45 mL/min. The isocratic elution was 20% A and 80% B for 15 minutes. The injection volume was 5 μL. After each injection, the needle was washed with 25% acetonitrile, 25% 2-propanol, and 50% water with 0.1% formic acid.

Quantification of stereoisomers of BUP and metabolites was performed using a UHPLC-MS/MS method (Bamfo et al, 2023; Gufford et al, 2022) with slight modifications. MS/MS analysis was performed on a QTRAP 6500+ mass spectrometer (Applied Biosystem/MDS Sciex) equipped with a Turbo V ion spray source and coupled with an AB Sciex UHPLC system. Data were acquired using Analyst software (version 1.6.3), and quantification was done via MultiQuant software (version 3.0.2). Mass spectrometry (MS) optimization was achieved by adjusting compound- and instrument-dependent parameters for MRM in positive mode. The source temperature was 550 °C for MRM, with high resolution for quadrupoles 1 and 3 and unit for third stage of tandem mass spectrometry (MS/MS/MS or MS³). Optimal gas pressures were set for all analytes and internal standards: collision gas (medium), curtain gas (25 psi), ion source gas 1 (35 psi), ion source gas 2 (20 psi), ion spray voltage (5500 V), and pause between mass ranges (5.007 ms).

Binding extent in microsome was reported as the fraction unbound (f,u) value, calculated as follows: f,u_mic = 1 − (PC − PF) / PC), where PC is the test compound concentration in the protein-containing compartment, and PF is the test compound concentration in the protein-free compartment.

2.7. Data analysis

Apparent kinetic constants (V_max and K_m) were estimated by fitting appropriate kinetic equations to the formation rate of DEX versus DEX concentration using nonlinear regression analysis in GraphPad Prism version 7.04 for Windows (www.graphpad.com). IC₅₀ values were determined by analyzing the logarithm of inhibitor concentration versus the percentage of activity remaining after inhibition, also using GraphPad software.

K_i values were calculated by fitting the inhibition data to various models of enzyme inhibition (competitive, noncompetitive, and uncompetitive) through nonlinear least-squares regression analysis using GraphPad software. The final model for each dataset was selected based on visual inspection of the Dixon plots, consideration of the sum of squares of residuals, Akaike information criterion, and Schwartz criterion values. Data are presented as mean ± SD or as averages of duplicate experiments.

2.8. Prediction of in vivo drug interactions from in vitro

In this study, we used mechanistic static interaction models as a routine, simple, and rapid prediction approach to describe BUP’s impact on CYP2D6 activity in vivo. These models require only enzyme inhibition constants (K_i or IC₅₀ values) and inhibitor concentration data, making them more accessible, but less precise in capturing all factors influencing drug interactions. While static models are effective for predicting DDIs driven by competitive enzyme inhibition of a single inhibitor, they may lack accuracy in more complex interactions, such as the current one, or in nonlinear scenarios, such as time-dependent inhibition (TDI). Physiologically based pharmacokinetic (PBPK) models may offer more accurate predictions of BUP-CYP2D6 interactions, especially given the involvement of several metabolites and their stereoisomers. Both PBPK and static models have distinct advantages and limitations in predicting DDIs. PBPK models are more comprehensive and accurate but are computationally intensive and require more detailed data. It is our intention to use our rich data set to support future detailed PBPK modeling approaches and compare PBPK based predictions with those obtained using the static models outlined in the current work.

2.8.1. Individual inhibition constants and unbound plasma or estimated liver steady-state concentrations of racemic or stereoisomers of BUP and metabolites

The in vivo ratio of the area under the plasma concentration–time curve (AUC) of CYP2D6 substrates in the presence (AUC_i) and absence (AUC_un) of each test inhibitor was predicted using steady-state unbound concentrations (C_max) in plasma ([I]_u,P) of each inhibitor, their respective in vitro inhibition constants (K_i) against CYP2D6, and the fraction of substrate metabolized by CYP2D6 (fm_CYP2D6). Steady-state plasma C_max value [I] for BUP, R-BUP, S-BUP, OHBUP, RR-OHBUP, SS-OHBUP, THBUP, RR-THBUP, SS-THBUP, EHBUP, RS-EHBUP, and SR-EHBUP were obtained from published literature (Kharasch et al, 2019, 2020). Plasma protein binding of racemic and stereoisomeric forms of BUP and its metabolites were determined from which the fraction unbound (f_u) plasma C_max was calculated [I]_u,P.

The maximum plasma concentration (C_max) of an inhibitor may not always accurately reflect the in vivo concentration available to the enzyme in the liver and, thus, may not reliably predict the extent of clinical DDIs. Direct measurement of the inhibitor concentration at the enzyme site is impractical. Unbound hepatic inlet concentrations have been proposed as alternative to predict clinical DDIs (Obach et al, 2006); this approach cannot be applied to circulating metabolites of BUP because knowledge of doses, absorption rate constants, and the fraction passing through the intestine unchanged is required. An alternative approach is to combine unbound clinical plasma concentrations, typically C_max, of BUP and its metabolites, with estimated liver-to-plasma (L:P) ratio of 5.5–9.4 obtained from quantitative whole-body autoradiography (qWBA) study in rats dosed with ¹⁴C-OHBUP (Reese et al, 2008). Thus, liver inhibitor concentrations ([I]_u,L) were estimated by multiplying the clinical unbound plasma C_max ([I]_u,P) by a factor of 5.5 and 9.4. While the qWBA data provide useful insights, we understand that applying rat data to humans involves assumptions and complexities. For instance, the qWBA data from ¹⁴C-OHBUP was used to estimate the L:P ratios of BUP and its metabolites and their stereoisomers because qWBA data were not available for these inhibitors. The possibility that the L:P ratios for BUP and its metabolites could differ if each compound had been individually dosed in separate qWBA studies cannot be ruled out. Species differences in drug metabolism, transport, and distribution may limit direct extrapolation to humans. qWBA, which measures total radioactivity, is nonselective as it does not distinguish between the liver distribution of individual metabolites. Additionally, because the qWBA data were generated from a single dose, the time-dependent changes in liver metabolism, transport, or plasma protein binding after multiple dosing may further impact hepatic concentrations over time. We acknowledge all these dynamics and uncertainties in estimating the L:P ratios were not modeled in our analysis. We also recognize that we attempted to estimate liver concentrations using limited data available to us. Therefore, the L:P scaling approach assumes similar distribution properties between rats and humans and between single and chronic dosing, which may not fully capture differential distribution. Despite these limitations, our approaches and assumptions of the scaling factor seem reassuring because our in vitro data provide both a mechanistic explanation and quantitative prediction of the clinical BUP-CYP2D6 interaction. Of note, this approach was successfully applied in previous studies (Reese et al, 2008). To strengthen the justification for using this scaling factor derived from rat qWBA data, we provide additional supporting data regarding unbound plasma concentrations (fu,p values) of BUP and its metabolites (and stereoisomers) in both rats and humans (Supplemental Table 1). These data demonstrate comparable fu,p values in rats and humans, suggesting that extrapolation from rat to human may be feasible with regard to fu,p.

The fraction of victim drug clearance due to metabolism by CYP2D6 (fm_CYP2D6) values used were 0.9 for desipramine (Reese et al, 2008) and 0.96 for DEX (Nakashima et al, 2007). The fm_CYP2D6 for nebivolol (0.93) (Briciu et al, 2014), atomoxetine (0.91) (Brown et al, 2016), and vortioxetine (0.65) (Frederiksen et al, 2022) were estimated from pharmacogenetic studies as proposed elsewhere (Tod et al, 2011). The prediction of area under the plasma concentration–time curve (AUC) changes brought about by each test inhibitor was performed using the following equation (Ito et al, 2005; Obach et al, 2006):

$$\frac{{AUC}_i}{AUCun}=\frac{1}{\begin{array}{c}\left(\frac{{fm}_{CYP2D6}}{1+\left(\frac{\left[I\right]u,PorL}{Ki}\right)}\right)+\left(1-{fm}_{CYP2D6}\right)\end{array}}$$ (1)

where AUC_i is the AUC of the substrate in the presence of the inhibitor, AUC_un is the AUC of the substrate in the absence of the inhibitor, fm_CYP2D6 is the fraction of substrate (victim drug) clearance due to metabolism by CYP2D6, 1-fm_CYP2D6 represents clearance of the substrate via nonCYP2D6 (ie, through other P450 or other enzymes and/or renal clearance), [I]_u,P and [I]_u,L are the unbound plasma (P) C_max and liver (L) concentration of the inhibitor (I), respectively, and K_i is the in vitro inhibition constant (the dissociation constant describing the binding affinity between the inhibitor and the CYP2D6 enzyme).

2.8.2. Summation of inhibition constants and unbound plasma or estimated liver steady-state concentrations of racemic or stereoisomers of BUP and metabolites

Mode of inhibition of CYP2D6 by BUP and its metabolites or their stereoisomers appear to be explained by competitive reversible model (Reese et al, 2008). The K_i values derived were generated in the same study and using the same probe substrate. In the event that multiple inhibitors, metabolites, or stereoisomers contribute to DDI risk and the mechanism of inhibition is the same, eq. 2 can be used to determine the cumulative impact on AUC_i/AUC_un ratios (Ito et al, 2005; Hinton et al, 2008), and thus this approach was used in this study. Racemic and stereoisomers of EHBUP and THBUP were found to be the most potent inhibitors compared with racemic and stereoisomers of BUP and OHBUP (see Results section). Therefore, predictions were first made by considering racemic EHBUP and THBUP together or their stereoisomers together. Then, the summation of all inhibition constants and unbound plasma or estimated liver concentrations of racemic or stereoisomers were tested in the prediction model.

Equation 2 integrates the summation of [I]_u,P/K_i or [I]_u,L/K_i, contrasting with eq. 1, which considers individual inhibitors I/K_i values:

$$\frac{{AUC}_i}{AUCun}=\frac{1}{\begin{array}{c}\left(\frac{{\text{fm}}_{CYP2D6}}{1+\left({\sum }_j^n\frac{\left[I\right]u,PjorL,uj}{Kij}\right)}\right)+\left(1-{\text{fm}}_{\text{CYP}2\mathrm{D}6}\right)\end{array}}$$ (2)

$\left({\sum }_{j=i}^n\frac{\left[I\right]u,PjorL,u,j}{Kij}\right)$ is the summation of each inhibitor’s unbound plasma or liver concentration divided by its respective inhibition constant K_ij, [I]_u,Pj is the unbound plasma C_max of the j-th inhibitor and [I]_u,Lj the estimated liver concentration of the j-th inhibitor, and K_ij is the in vitro inhibition constant of the j-th inhibitor against CYP2D6.

3. Results

3.1. Kinetic parameters of DEX O-demethylation

The formation rates of DXR in pooled HLMs were best described by the Michaelis-Menten equation (data not shown). The apparent K_m and maximum rate of metabolism (V_max) were 4.1 μM and 647 pmol/min per milligram of protein, respectively (data not shown). Consequently, a fixed DEX concentration of 5 μM, which approximates the K_m value, was selected for determining IC₅₀ values and for establishing the substrate concentration range needed to assess the K_i of BUP and its metabolites (described further).

3.2. IC₅₀ values

The inhibitory effects of various concentrations (0.1–100 μM) of racemic BUP and its metabolites on CYP2D6 activity were assessed using a single concentration of the isoform-specific probe DEX (5 μM) in pooled HLMs (0.5 mg/mL). EHBUP demonstrated potent inhibition of CYP2D6 with an IC₅₀ of 7.9 μM, while OHBUP was the least potent inhibitor, with an IC₅₀ of 48.9 μM. BUP and THBUP exhibited intermediate potency, with IC₅₀ values approximately 20 μM Fig. 2A; Table 1).

Fig. 2.

Determination of IC₅₀ values for the inhibition of CYP2D6 by racemic and stereoisomers of BUP and its metabolites. DXM (5 μM) was incubated with HLMs (0.5 mg/mL) and NADPH for 20 minutes in the absence (control) and presence of increasing concentrations of each inhibitor. IC₅₀ values were calculated from percent activity remaining vs log₁₀ of inhibitor concentrations. Inhibition profiles of CYP2D6 by racemic BUP and its metabolites (A) and stereoisomers of BUP and metabolites (B) are shown. Each data point represents the average of duplicate incubations.

Table 1.

Estimated K_i value for CYP2D6 inhibition by BUP and metabolites and fraction unbound (f_u) clinical plasma C_max

Inhibitor	IC50	Ki	Steady-State Plasma Cmaxa [I]	fu	[I]u Plasma	[I]u Liverb	[I]u/Ki Plasma	[I]u/Ki Liver
			μM		μM	μM
Stereoisomers
R-BUP	22.39	16.27	0.3	0.68	0.204	1.12–1.92	0.012	0.07–0.12
S-BUP	14.5	5.58	0.044	0.71	0.031	0.17–0.29	0.006	0.03–0.05
RR-OHBUP	14.77	8.17	4.72	0.53	2.502	13.76–23.52	0.31	1.68–2.88
SS-OHBUP	12.87	12.26	0.13	0.64	0.083	0.46–0.78	0.006	0.04–0.06
RR-THBUP	9.99	1.89	1.08	0.76	0.821	4.51–7.72	0.45	2.39–4.08
SS-THBUP	4.49	3.9	0.96	0.66	0.634	3.49–5.96	0.17	0.89–1.53
RS-EHBUP	2.45	3.03	0.11	0.77	0.085	0.47–0.80	0.027	0.15–0.26
SR-EHBUP	1.17	3.35	0.32	0.77	0.246	1.36–2.32	0.078	0.41–0.69
Racemic
BUP	19.71	12.65	0.34	0.70	0.238	1.31–2.24	0.019	0.10–0.18
EHBUP	7.88	0.97	0.42	0.77	0.323	1.78–3.04	0.33	1.83–3.13
THBUP	20.56	5.3	2.05	0.71	1.456	8.01–13.68	0.27	1.51–2.58
OHBUP	48.85	11.08	4.84	0.59	2.86	15.71–26.84	0.26	1.42–2.42

^aSteady-state C_max values were obtained from the literature (Kharasch et al, 2020).

^bCalculated based on a liver-to-plasma ratio of 5.5–9.4 (determined from qWBA studies in rats dosed with 14C-OHBUP) (Reese et al, 2008).

Similarly, IC₅₀ values for the stereoisomers of BUP and its metabolites were determined in pooled HLMs (Fig. 2B; Table 1). The diastereomers of EHBUP were identified as strong inhibitors, with IC₅₀ values of less than 2.5 μM, followed by SS-THBUP and RR-THBUP. The IC₅₀ values for SS-OHBUP, S-BUP, RR-OHBUP, and R-BUP ranged from approximately 12.9–22.4 μM.

3.3. K_i values

Figure 3 illustrates the inhibition kinetics of CYP2D6-catalyzed DEX O-demethylation by racemic BUP and its metabolites, accompanied by corresponding Dixon plots. Based on the final model selection criteria (see Data analysis section), the inhibition data fit best to competitive model of enzyme inhibition than the other models tested, indicating that these inhibitors compete directly with the substrate for binding to the active site of CYP2D6. Visual inspection of the Dixon plots further confirms the competitive nature of the inhibition for each inhibitor. The derived K_i values (Table 1) reveal that racemic EHBUP and THBUP are strong inhibitors of CYP2D6, with K_i values of 0.97 μM and 5.3 μM, respectively, whereas OHBUP and BUP exhibit weaker inhibition, with K_i values of 11.08 μM and 12.65 μM, respectively (Table 1). EHBUP is 5.5-, 11.4, and 13.0-fold more potent than THBUP, OHBUP, and BUP, respectively, while THBUP is over 2-fold more potent than OHBUP and BUP. The rank order of potency from highest to lowest was EHBUP > THBUP > OHBUP ≈ BUP.

Fig. 3.

Determination of K_i values for the inhibition of CYP2D6 by racemic BUP and its metabolites. A range of concentrations of DXM was incubated with HLMs (0.1 mg/mL) and cofactors for 5 minutes in the absence and presence of multiple inhibitor concentrations. Inhibition kinetic data were fitted to a competitive inhibition model. Left panel: Michaelis-Menten plots showing the effect of racemic BUP, OHBUP, THBUP, and EHBUP on CYP2D6 activity. Right panel: Dixon plots illustrating the competitive inhibition of CYP2D6 by the inhibitors. Data are presented as the average of duplicate incubations.

As observed with racemic BUP and its metabolites, the inhibition kinetics of CYP2D6 by the stereoisomers of BUP and its metabolites best fit a competitive inhibition model. These findings (competitive inhibition) were further supported by visual inspection of the corresponding Dixon plots (Fig. 4). For clarity, the inhibition profiles of R-BUP and its derived metabolites (RR-OHBUP, RR-THBUP, and SR-EHBUP) are presented separately from those of S-BUP and its metabolites (SS-OHBUP, SS-THBUP, and RS-EHBUP) in Fig. 4, A and B, respectively. The corresponding K_i values are listed in Table 1.

Fig. 4.

Determination of K_i values for the inhibition of CYP2D6 by R- BUP and metabolites derived from R-BUP (A) and by S-BUP and metabolites derived from S-BUP (B). A range of concentrations of DXM was incubated with HLMs (0.1 mg/mL) and cofactors for 5 minutes in the absence and presence of multiple inhibitor concentrations of stereoisomers. Inhibition kinetic data were fitted to a competitive inhibition model. (A) Above panel, Michaelis-Menten plots illustrating the inhibition of CYP2D6 by R-BUP, RR-OHBUP, RR-THBUP, and SR-EHBUP; Lower panel, Dixon plots showing the competitive inhibition kinetics of CYP2D6 by the inhibitors; (B) Above panel, Michaelis-Menten plots illustrating the inhibition of CYP2D6 by S-BUP, SS-OHBUP, SS-THBUP, and RS-EHBUP; Lower panel, Dixon plots showing the competitive inhibition kinetics of CYP2D6 by the inhibitors. Data represent the average of duplicate incubations.

Among the tested compounds, RR-THBUP exhibited the most potent inhibition of CYP2D6 activity, with a K_i value of 1.9 μM, which was 2.1-fold lower than its counterpart diastereomer, SS-THBUP (K_i = 3.9 μM). The K_i value for S-BUP (5.6 μM) was 2.9-fold lower than that for R-BUP (16.3 μM). RR-OHBUP demonstrated slightly higher potency (1.5-fold) compared with SS-OHBUP. Notably, no stereoselective inhibition was observed between the diastereomers of EHBUP, with K_i values of 3.03 μM for RS-EHBUP and 3.4 μM for SR-EHBUP (Table 1). Thus, the rank order of potency from highest to lowest was RR-THBUP > RS-EHBUP ≈ SR-EHBUP ≈ SS-THBUP > S-BUP > RR-OHBUP > SS-OHBUP > R-BUP.

Numerous studies suggest that irreversible or TDI can lead to underprediction of inhibition, but we believe this is unlikely for the BUP-CYP2D6 interaction. First, Reese et al (2008) found that neither BUP nor its metabolites act as time-dependent inhibitors of CYP2D. Another study also suggested reversible inhibition (Sager et al, 2017). Second, we estimated K_i values using various enzyme inhibition models and found that competitive inhibition best described the data. Visual inspection of the Dixon plots supports this. Third, our accurate prediction of in vivo AUC changes assuming competitive inhibition further supports this conclusion. Thus, irreversible inhibition does not appear to significantly contribute to the underprediction of the observed in vivo DDI.

3.4. Prediction of observed AUC_i/AUC_un ratios from in vitro K_i values

Predictions of AUC ratios were made for individual inhibitors (eq. 1) or for multiple inhibitors together (eq. 2) using in vitro apparent K_i values and the corresponding unbound steady-state plasma concentrations (C_max) or estimated liver concentrations of racemic BUP and its metabolites (BUP, OHBUP, THBUP, and EHBUP) or their stereoisomers (S-BUP, R-BUP, SS-OHBUP, RR-OHBUP, SS-THBUP, RR-THBUP, RS-EHBUP, and SR-EHBUP).

All K_i values derived from our experiments (Table 1) and used in our predictions were apparent values (K_i,app) without corrections for microsomal binding, assuming that binding to microsomal proteins was minimal. However, we recognize that incorporating microsomal binding data may be crucial for accurately relating in vitro K_i values to in vivo DDI predictions (Obach, 1997). Therefore, it is important to confirm the initial assumption that binding to microsomal proteins is minimal and the impact of microsomal protein binding on K_i,app is negligible. The data presented in Supplemental Table 2 indicate that the fraction unbound (f,u) of BUP and its metabolites is close to 1 for all inhibitors. The f,u value for paroxetine, which was used as a positive control, was 0.500 ± 0.072. This value aligns with the commercial microsomal binding data for paroxetine, which reported a f,u of 0.456 ± 3.9 (https://www.xenotech.com/wp-content/uploads/2022/02/XenoTech_Microsomal-Protein-Binding-of-Drugs.pdf). Based on the data presented in Supplemental Table 2, the K_i,u (unbound inhibition constant) and K_i,app values for each inhibitor are comparable, indicating that the effect of microsomal protein binding is marginal.

In Table 1 and Supplemental Table 1, the fraction (%) unbound to human plasma (fu,p) for racemic and stereoisomers of BUP and its metabolites determined using an equilibrium dialysis procedure were used for predictions. For the racemic compounds, the unbound fractions were 69.5% for BUP, 58.6% for OHBUP, 71.0% for THBUP, and 76.9% for EHBUP (Table 1 and Supplemental Table 1). For the stereoisomers, the unbound fractions were 68.4% for R-BUP, 70.6% for S-BUP, 53.2% for RR-OHBUP, 63.9% for SS-OHBUP, and 76.3% for RR-THBUP. Because RS-EHBUP and SR-EHBUP were not commercially available during the plasma protein assays, the value for racemic EHBUP (76.9%) was used for the stereoisomers. The unbound fractions in rat plasma for the racemic and stereoisomer compounds are provided in Supplemental Table 1 for comparison.

The predicted AUC ratios for each inhibitor or multiple inhibitors together were then compared with clinically observed DDI outcomes available in the literature (Table 2 and Supplemental Table 3). These outcomes were based on the pharmacokinetics of CYP2D6 substrates (desipramine, nebivolol, DEX, vortioxetine, and atomoxetine) determined both before (baseline) and after treatment with chronic doses of BUP (300 mg/day or 150 mg twice daily). Clinically observed AUC_i/AUC_un ratios for CYP2D6 substrates with an fm_CYP2D6 of ≥0.9 (desipramine, nebivolol, DEX, and atomoxetine) ranged from a 5.1- to 7.2-fold increase and for vortioxetine, which has an fm_CYP2D6 of 0.65, was a 2.28-fold increase in AUC ratio.

Table 2.

Predicted and observed AUC changes (ratios) in vivo for BUP interaction with CYP2D6 substrates

Substrate (Fm CYP2D6)a	Study Design	Racemic and stereoisomers of BUP and its metabolites	AUC Ratio Predicted		Observed AUC Ratios	Reference
			Plasmab	Liver (Average)c
Desipramine (0.9)	Open-label, 2-period. PK of 50 mg desipramine was determined in 15 healthy NMs of CYP2D6 before (day 1, control) and with 150 mg BUP (day 22) after pretreatment with BUP (150 mg/day on days 8–10 and 300 mg/day on days 11–21)	Racemic	1.73	3.95–5.09 (4.52)	5.2	Reese et al, 2008
		Stereoisomers	1.84	4.25–5.43 (4.84)
Nebivolol (0.93)	Open-label, 2-period. PK of a single 5 mg nebivolol was determined in 18 healthy volunteers: before (day 1, control) and with 300 mg BUP (day 8) following 7-day pretreatment with BUP (150 mg/day for 3 days and 300 mg/day for 4 days)	Racemic	1.77	4.38–5.89 (5.13)	7.2	Gheldiu et al, 2016
		Stereoisomers	1.89	4.77–6.36 (5.57)
Atomoxetine (0.91)	Open-label, 2-period: PK of a single 25 mg atomoxetine was determined in 18 volunteers (CYP2D6 NMs): before (day 1, control) and with 300 mg BUP (day 8) following 7-day pretreatment with BUP (150 mg/day for 3 days and 300 mg/day for 4 days)	Racemic	1.75	4.08–5.33 (4.70)	5.1	Todor et al, 2016
		Stereoisomers	1.86	4.41–5.71 (5.06)
DXM (0.96)	Open-label, 2-period: PK of a single 30 mg DXM was determined in 24 healthy volunteers on 2 occasions: before (day 1, control) and with 150 mg BUP (day 22) following pretreatment with BUP (150 mg/day for 3 days and 150 mg twice daily for 16 days)	Racemic	1.82	4.91–6.95 (5.95)	6.5	Duong et al, 2020
		Stereoisomers	1.95	5.43–7.70 (6.56)
Vortioxetine (0.65)	Steady-state PK of vortioxetine was determined at steady state (10 mg/day vortioxetine, days 1–28) in 24 healthy volunteers on 2 occasions: before (day 14, control) and with 150 mg BUP (day 28) after pretreatment with BUP (75 mg twice daily for 3 days, and 150 mg twice daily on days 18–28)	Racemic	1.44	2.17–2.38 (2.28)	2.28	Chen et al, 2013
		Stereoisomers	1.49	2.23–2.43 (2.33)

NM, normal metabolizer; PK, pharmacokinetic.

^aThe fmCYP2D6 values were obtained from published literature or estimated from pharmacogenetic studies.

^bPrediction calculated based on unbound steady-state clinical plasma C_max values of the inhibitors (Kharasch et al, 2019, 2020).

^cPrediction calculated based on the ratio of liver to the unbound plasma C_max, which was 5.5–9.4 as determined from qWBA studies in rats dosed with 14C-OHBUP (Reese et al, 2008).

3.5. Predictions using individual racemic or stereoisomers of BUP and its metabolites

Using in vitro K_i values and the unbound steady-state plasma concentrations (C_max) of racemic BUP, OHBUP, THBUP, and EHBUP, predictions were made using eq. 1. The findings are summarized in Fig. 5 (green bars), Table 2, and Supplemental Table 3. EHBUP was predicted to cause the greatest increase in AUC_i/AUC_un ratios (∼1.3-fold increase), followed by THBUP (∼1.25-fold increase) and OHBUP (∼1.24-fold increase) for CYP2D6 substrates with fm_CYP2D6 of ≥0.9. BUP was predicted to have a minimal effect on AUC_i/AUC_un ratios (∼1.02-fold increase). For vortioxetine, which has an fm_CYP2D6 value of 0.65, the predicted increases in AUC ratios were accordingly modest: 1.19-fold for EHBUP, 1.16-fold for THBUP, and 1.15-fold for OHBUP, with racemic BUP predicted to cause no change in AUC ratios. Overall, the predictions based on individual K_i values and unbound steady-state plasma concentrations resulted in a significant and marked underestimation of the observed clinical interactions (Fig. 5, green bars; Table 2 and Supplemental Table 3).

Fig. 5.

Predicted AUC_i/AUC_un changes from in vitro inhibition constants (K_i values) and steady-state plasma C_max or estimated C_max liver concentrations of racemic BUP and its metabolites. Predicted AUC_i/AUC_un changes of 5 CYP2D6 substrates (desipramine, nebivolol, vortioxetine, atomoxetine, and DXM) caused by the individual inhibitors—BUP, OHBUP, THBUP, and EHBUP—were calculated using eq. 1, incorporating the in vitro inhibition constant (K_i value) and steady-state plasma C_max or estimated C_max liver concentration of each inhibitor (values are presented in Table 1). The prediction was repeated using eq. 2 where the summation of I/K_i values of all racemic BUP and its metabolites (BUP + OHBUP + THBUP + EHBUP) was considered. These predicted AUC changes were compared with the observed AUC ratio in healthy volunteers who received a single oral dose of a CYP2D6 substrate on 2 occasions: at baseline (uninhibited) and after dosing with BUP (300 mg/day or 150 mg/day twice daily) to steady state (Table 2).

We repeated the predictions using the inhibition constants and estimated liver concentrations of each inhibitor (racemic BUP and its metabolites) in eq. 1 (Fig. 5—blue bars; Table 2 and Supplemental Table 2). EHBUP was predicted to cause the largest fold increase in AUC_i/AUC_un ratios, ranging from 2.77- to 3.16-fold, for substrates with fm_CYP2D6 ≥0.9, followed by THBUP (2.51- to 2.81-fold increase) and OHBUP (2.44- to 2.70-fold increase). BUP alone was predicted to cause a minimal change in AUC (≤1.13-fold increase). The same trend was predicted for vortioxetine as a substrate (fm_CYP2D6 = 0.65), with the predicted AUC ratios of 1.85-, 1.76-, 1.73-, and 1.09-fold increase for EHBUP, THBUP, OHBUP, and BUP, respectively. These predictions based on liver concentrations (Fig. 5, blue bars; Table 2 and Supplemental Table 3) were significantly higher compared with those made using plasma exposure (Fig. 5, green bar; Table 2 and Supplemental Table 3). Despite the liver concentration adjustment, however, the predictions using individual inhibitor still underestimated the clinically observed AUC ratios.

To test whether consideration of the individual stereoisomers could improve prediction, we used K_i values and the unbound steady-state plasma concentrations (C_max) or estimated liver concentrations of each stereoisomer of BUP and its metabolites in eq. 1 to predict in vivo AUC changes for CYP2D6 substrates. The predicted AUC ratios using plasma exposure of the individual stereoisomers are summarized in Fig. 6 (green bar), Table 2, and Supplemental Table 3. Among the stereoisomers, the largest predicted AUC increase was associated with the most potent in vitro inhibitor, RR-THBUP, which showed an average 1.39-fold increase (range, 1.37- to 1.41-fold increase) for substrates with fm_CYP2D6 of ≥0.9. In contrast, RR-OHBUP and SS-THBUP were predicted to cause approximately 1.28-fold and 1.15-fold increases in AUC ratios, respectively, for these substrates, while RS- and SR-EHBUP were predicted to cause increases of ≤1.07-fold. S-BUP, R-BUP, and SS-OHBUP were predicted to cause no change in AUC ratios (Fig. 6, green bars; Table 2 and Supplemental Table 3). For vortioxetine, the predicted fold changes in AUC ratios were smaller: 1.25-fold with RR-THBUP, 1.18-fold with RR-OHBUP, and 1.10-fold with SS-THBUP. S-BUP, R-BUP, RS-EHBUP, SR-EHBUP, and SS-OHBUP predicted no or marginal changes in vortioxetine AUC ratio. However, consistent with the findings for the racemic inhibitors, prediction of the in vivo DDI outcomes cannot be made solely on the [I]_u,P/K_i values of any individual stereoisomer.

Fig. 6.

Predicted AUC_i/AUC_un changes from in vitro inhibition constants (K_i values) and steady-state plasma C_max or estimated C_max liver concentrations of stereoisomers of BUP and its metabolites. Predicted AUC_i/AUC_un changes of 5 CYP2D6 substrates (desipramine, nebivolol, vortioxetine, atomoxetine, and DXM) caused by the individual stereoisomers of BUP, OHBUP, THBUP, and EHBUP were calculated using eq. 1, incorporating the in vitro inhibition constant (K_i value) and steady-state plasma C_max or estimated C_max liver concentration of each inhibitor (values are presented in Table 1). The prediction was repeated using eq. 2 where the summation of I/K_i values of all stereoisomers of BUP and its metabolites (R-BUP + S-BUP + RR-OHBUP + SS-OHBUP + RR-THBUP + SS-THBUP + SR-EHBUP + RS-EHBUP) was considered. These predicted AUC changes were compared with the observed AUC ratio in healthy volunteers who received a single oral dose of a CYP2D6 substrate on 2 occasions: at baseline (uninhibited) and after dosing with BUP (300 mg/day or 150 mg/day twice daily) to steady state (Table 2).

Use of the estimated liver concentrations and in vitro K_i values of the individual stereoisomers (eq. 1) significantly improved the prediction of in vivo AUC ratios compared with predictions based on plasma exposure and K_i values (Fig. 6, blue bars; Table 2 and Supplemental Table 3). RR-THBUP alone exhibited the largest fold increase, with a 3.17- to 3.73-fold change (average 3.4-fold increase) for the substrates with fm_CYP2D6 of ≥0.9, followed by RR-OHBUP with a 2.65- to 3.0-fold increase (average 2.79-fold increase) and SS-THBUP with a 1.97- to 2.10-fold increase. SR-BUP and RS-BUP predicted approximately 1.5-fold and 1.19-fold increases in AUC ratios, respectively. Predictions from R-BUP, S-BUP, and SS-OHBUP showed marginal increases of ≤1.09-fold. For vortioxetine, the predictions followed a similar pattern but with lower fold changes: 1.97-fold with RR-THBUP, 1.81-fold with RR-OHBUP, and 1.54-fold with SS-THBUP. SR- and RS-EHBUP predicted 1.3-fold and 1.13-fold increases, respectively, while S-BUP, R-BUP, and SS-OHBUP predicted increases of ≤1.06-fold. Although adjusting for liver concentrations significantly improved the accuracy of predictions for each stereoisomer of BUP and its metabolites compared with those predictions using plasma exposure (Fig. 6, green bar; Table 2 and Supplemental Table 3), none of the individual stereoisomers accurately predicted the exposure changes (AUC_i/AUC_un) observed for CYP2D6 substrates in vivo (Fig. 6, blue bars; Table 2 and Supplemental Table 3).

3.6. Predictions using the summation of racemic or stereoisomers of BUP and its metabolites

3.6.1. Summation of racemic or stereoisomers of EHBUP and THBUP

The relative potency of CYP2D6 inhibition by the reductive metabolites (EHBUP and THBUP) and their stereoisomers (SR- and RS-EHBUP and SS- and RR-THBUP) was found to be greater than that of the racemic and stereoisomeric forms of BUP and OHBUP (Table 1). The steady-state plasma exposure of these metabolites exceeds that of BUP (Daviss et al, 2006; Benowitz et al, 2013; Kharasch et al, 2019, 2020). Therefore, we first tested whether incorporating the summation of the K_i values and unbound steady-state concentrations of the reductive metabolites (EHBUP + THBUP) in plasma ([I]_u,P/K_i) or liver ([I]_u,L/K_i) into eq. 2 could account for the clinically observed AUC ratios. Using the combined in vitro K_i values and plasma concentration of racemic EHBUP and THBUP, the predicted AUC ratio was a 1.54-fold increase for substrates with fm_CYP2D6 of ≥0.09 and 1.33-fold increase for vortioxetine as a substrate (Fig. 5, green bars; Table 2 and Supplemental Table 3). When considering the summation of K_i values and the corresponding plasma steady-state concentrations of the stereoisomers of EHBUP and THBUP, the predicted AUC ratios were 1.62-fold increase; for vortioxetine, this was 1.36-fold increase (Fig. 6, green bars; Table 2 and Supplemental Table 3).

Repeating the predictions using liver concentrations of racemic EHBUP and THBUP reveal significantly higher predicted AUC ratios (a 4.11-fold increase for substrates with fm_CYP2D6 of ≥0.09 and a 2.12-fold increase for vortioxetine) (Fig. 5, green bars; Table 2 and Supplemental Table 3). The summation approach considering liver concentrations of stereoisomers of EHBUP and THBUP predicted a 4.45-fold increase for substrates with fm_CYP2D6 of ≥0.09 and 2.18-fold increases for vortioxetine. Thus, significantly improved prediction was noted using ([I]_u,L/K_i) than the the predictions using ([I]_u,P/K_i) for both recemic and stereoisomers of EHBUP and THBUP.

3.6.2. Summation of all racemic or all stereoisomers of BUP and metabolites

Although these data using the summation of racemic or stereoisomers of EHBUP and THBUP approach highlight the major role of these reductive metabolites in the overall observed DDIs, accurate prediction was not fully achieved with the use of the summation of racemic EHBUP and THBUP or all their stereoisomers together (Figs. 5 and 6; Table 2 and Supplemental Table 3). Therefore, we tested whether more accurate predictions could be achieved by taking simultaneously all K_i values of racemic or all K_i values of stereoisomers of BUP and metabolites and steady-state concentrations (plasma or liver) in eq. 2.

As shown in Fig. 5 (green bar), Table 2, and Supplemental Table 3, inclusion of all [I]_u,_P/K_i values for all racemic BUP and its metabolites into the prediction model (eq. 2) slightly improved the predicted AUC ratio increased, on average 1.77-fold (ranging from 1.73- to 1.82-fold) for substrates with fm_CYP2D6 of ≥0.9 and 1.44-fold for vortioxetine. Of the total AUC ratios predicted with all racemic forms in plasma, approximately 87% (for substrates with fm_CYP2D6 of ≥0.9) and 92% (for vortioxetine) was accounted by racemic EHBUP and THBUP. Similarly, when summing all [I]_u,P/K_i values for all stereoisomers of BUP and its metabolites is taken in to account, the predicted AUC ratio was, on average, a 1.88-fold increase (ranging from 1.84- to 1.95-fold) for substrates with fm_CYP2D6 of ≥0.9 and 1.49-fold for vortioxetine (Fig. 6, blue bars; Table 2 and Supplemental Table 3). Of the total AUC ratios predicted using all stereoisomers in plasma, approximately 86% for substrates with fm_CYP2D6 of ≥0.9 and 91% for vortioxetine were accounted for by the summation of all stereoisomers of EHBUP and THBUP.

As described earlier, use of steady-state plasma exposure of all racemic or all stereoisomers of BUP and metabolites significantly underpredicted observed clinical DDI between BUP and CYP2D6 interactions. We then next repeated the prediction using the summing of all [I]_u,L/K_i values for racemic or stereoisomers of BUP and its metabolites. The predicted AUC ratios using all racemic forms achieved accurate prediction (Fig. 5, blue bar; Table 2 and Supplemental Table 3), with predictions falling on average within 11.6% for racemic (range, 0%–29%) of the observed in vivo AUC ratios. Of the total AUC ratios predicted using all racemic forms in liver, approximately 81% for substrates with fm_CYP2D6 ≥0.9 and 93% for vortioxetine were accounted for by the summation of EHBUP and THBUP. When the summing all [I]_u,L/K_i values of stereoisomers of BUP and its metabolites is considered, the predictions of the in vivo AUC ratios were more accurate for substrates with fm_CYP2D6 of ≥0.9 (4.84- to 6.56-fold increase) and 2.33-fold increase for vortioxetine, falling on average within 5.4% for stereoisomers (range, −2.2% to 23%) of the observed in vivo AUC ratios (Fig. 6, blue bar; Table 2 and Supplemental Table 3). Of the total AUC ratios predicted using all stereoisomeric forms, approximately 81% for substrates with fm_CYP2D6 ≥0.9 and 93% for vortioxetine were accounted for by the summation of EHBUP and THBUP (Supplemental Table 3). Use of the stereoisomers improved prediction compared with using the racemic forms (on average 5.4% vs 11.6%, respectively).

We compared our model prediction performance with predictions using PBPK models. Only 1 PBPK model has been published to predict BUP’s interaction with CYP2D6 substrates (Xue et al, 2017), with some limitations. For example, while predictions were made for several CYP2D6 substrates, observed AUC values were only available for desipramine and to some extent for venlafaxine based on plasma concentrations (not AUC changes). To provide context for our static model predictions, we compared our results with the PBPK prediction of desipramine AUC change (Xue et al, 2017). The PBPK model predicted a 5.05-fold increase in desipramine AUC, which closely matched the observed 5.2-fold change (Xue et al, 2017). Our static model predicted a similar range (3.95-fold to 5.09-fold increase) (Table 2 and Supplemental Table 3), supporting the rationale for using the static models in this specific case. Both models assumed competitive inhibition.

4. Discussion

To our knowledge, this study is the first comprehensive in vitro study to test the inhibition potency of racemic and stereoisomers of BUP and its metabolites in HLMs. By incorporating the inhibition constants for all forms of racemic or all forms of stereoisomeric BUP and its metabolites, along with their estimated steady-state liver concentrations into eq. 2, we achieved highly accurate predictions of the observed clinical DDIs. The average prediction accuracy was within 5.4% for stereoisomers and 11.6% for racemic forms of the observed clinical DDIs between BUP and CYP2D6 substrates. Notably, over 80% of the predicted AUC changes were attributed to the combined effects of either racemic or stereoisomeric forms of EHBUP and THBUP. In contrast, predictions based solely on individual inhibitors using plasma or liver concentrations or summing all forms (racemic or stereoisomers) based on plasma concentrations led to significant underestimations. These findings provide valuable mechanistic insights and quantitative data regarding the partially understood CYP2D6-dependent DDIs associated with BUP.

Our findings indicate that the reductive metabolites of BUP, specifically the racemic forms of EHBUP and THBUP, are more potent inhibitors of CYP2D6 than BUP and OHBUP. This aligns with previous in vitro studies (Hesse et al, 2000; Reese et al, 2008; Parkinson et al, 2010), which demonstrated that racemic BUP and OHBUP are relatively weak inhibitors of CYP2D6 and the reductive metabolites, THBUP and EHBUP, are approximately 4- and 12-fold more potent than BUP (Reese et al, 2008). To our knowledge, ours is the first to test stereospecific inhibition of CYP2D6 by the stereoisomers of EHBUP and THBUP. We demonstrated greater inhibitory potency (K_i values ranging from 1.89 to 3.35 μM) of the diastereomers of EHBUP and THBUP compared with the stereoisomers of BUP and OHBUP, which ranged from 5.6 to 16.3 μM. RR-THBUP was approximately 2.1-fold more potent than SS-THBUP, but no stereoselective inhibition was noted for the diastereomers of EHBUP, with K_i values between 3.03 μM and 3.35 μM. We also found that S-BUP and RR-OHBUP were 3.0- and 1.5-fold more potent inhibitors of CYP2D6 than their counterparts R-BUP and SS-OHBUP, respectively. These findings are consistent with prior publications indicating that S-BUP and RR-OHBUP exhibit 13.7- and 2.8-fold greater inhibition (based on IC₅₀ values) compared with R-BUP and SS-OHBUP, respectively (Sager et al, 2017). The slight differences in stereoselectivity between our study and that study (Sager et al, 2017) may be attributed to the differing inhibition parameters used (K_i values in our study vs IC₅₀ values in theirs) and the potential racemization of BUP under various experimental conditions (Bamfo et al, 2023). Overall, our data demonstrate that BUP reductive metabolites (racemic or stereoisomers of EHBUP and THBUP) exhibit higher binding affinity for the CYP2D6 active site compared with racemic and stereoisomeric BUP and OHBUP and that these metabolites may be the main drivers of CYP2D6 inhibition in vivo, underscoring the role circulating metabolites on the observed clinical BUP-CYP2D6 inhibition. Our data also suggest modest stereoselective inhibition of CYP2D6 by BUP and its metabolites.

CYP2D6 inhibition by BUP persists for over 7 days after the last dose (Wellbutrin, 2017). Of all stereoisomers, RR-THBUP was the most potent inhibitor of CYP2D6. It has also the longest elimination half-life (∼45 hours) after the administration immediate release single 100 mg oral dose of BUP (Masters et al, 2016a). We simulated steady-state exposure of stereoisomers of BUP and its metabolites from the single-dose study and predicted that RR-THBUP had the greatest accumulation index and persists in the circulation for at least 12 days (Masters et al, 2016a). Whether this metabolite helps to explain the extended clinical inhibition of CYP2D6 by BUP after the last dose remains to be determined.

Integrating the K_i values of each inhibitor derived in vitro with the corresponding steady-state plasma C_max or estimated liver concentrations of racemic or stereoisomers of BUP and its metabolites into eq. 1 significantly underestimated the observed clinical DDI risks between BUP and CYP2D6 substrates. For EHBUP and RR-THBUP, the most potent inhibitors, the maximum fold change predicted using plasma exposure was less than 1.5-fold increase. In contrast, using liver concentrations yielded maximum increases of 2.93-fold and 3.4-fold in the AUC ratio for substrates with f_mCYP2D6 of ≥0.9. These results indicate that estimated liver concentrations improved predictions compared with plasma exposure. However, no single individual inhibitor—whether racemic or stereoisomer—fully accounted for the more than 5-fold increase in AUC ratio observed for substrates with fm_CYP2D6 of ≥0.9, regardless of whether plasma or liver concentrations were used in the modeling (Table 2 and Supplemental Table 3).

To our knowledge, ours is the first to accurately predict observed AUC ratios when the inhibition constants for all forms of racemic or all forms of stereoisomers of BUP and its metabolites, along with their estimated liver concentrations, were incorporated into eq. 2. On average, these predictions were within 11.6% for racemic BUP (range, 0%–29%) and within 5.4% for stereoisomers (range, −2.2% to 23%) of the in vivo observations for all substrates following coadministration of BUP. Our data also indicate that the summation of the reductive metabolites of BUP (racemic or stereoisomers EHBUP and THBUP) accounts for over 80% of the in vivo CYP2D6 inhibition by BUP. The overall contribution of the combined racemic or combined stereoisomeric forms of BUP and OHBUP to CYP2D6 inhibition in vivo seems marginal (<20%). A previous in vitro-to-in vivo prediction of BUP-desipramine interaction based on all K_i values and steady-state concentrations of racemic BUP and metabolites revealed prediction within 26% of the observed AUC ratio (Reese et al, 2008). In our analysis, using either all racemic forms or all stereoisomers of BUP and its metabolites in the liver, we predicted AUC ratios within 13% and 7%, respectively, of the observed 5.2-fold increase in desipramine exposure following coadministration of BUP (Table 2 and Supplemental Table 3). This improved prediction in our study could reflect to our use of a higher plasma fraction unbound, based on well characterized in vitro plasma protein binding data (Table 1). Predictions of the observed clinical AUC ratios using plasma concentrations as a surrogate for inhibitors were considerably less accurate (within >60%) compared with predictions based on liver concentrations, indicating that plasma concentrations of BUP and its metabolites are poor indicators of CYP2D6 inhibition at the enzyme’s active site in the liver.

A previous in vitro study suggested that transcriptional downregulation of CYP2D6 by BUP and its metabolites, along with reversible inhibition, could contribute to BUP-DDIs in vivo (Sager et al, 2017), but this mechanism seems unlikely. First, a follow-up study questioned its relevance for in vivo observations (Stevison et al, 2019). Second, our clinical data (Duong et al, 2020) showing increased DEX exposure with a single 150-mg dose of BUP suggests no significant role for downregulation, as transcriptional effects typically take longer to manifest. The hypothesis that BUP metabolites downregulate CYP2D6 was based on decreased CYP2D6 mRNA (Sager et al, 2017). Although unlikely, a downregulation effect may not be fully ruled out during the marked DEX exposure change on day 22 after multiple doses of BUP (Duong et al, 2020). Third, our in vitro data accurately predicted interactions using reversible inhibition alone. While TDI by BUP metabolites could contribute to the reported CYP2D6 downregulation, our data, along with prior studies (Reese et al, 2008; Sager et al, 2017), do not support the involvement of TDI.

This study highlights key findings. First, circulating metabolites of BUP, especially the reductive forms (EHBUP and THBUP), are major drivers of strong in vivo CYP2D6 inhibition by BUP. These metabolites have higher plasma exposure than BUP itself after 300 mg/day oral BUP at steady state (Daviss et al, 2005; Kharasch et al, 2019, 2020). Second, plasma concentrations alone significantly underpredicted DDI outcomes, indicating that plasma exposure may not reflect liver concentrations, where BUP and its metabolites likely accumulate. In addition, we found that no single inhibitor, based on plasma or liver concentrations, fully explains the observed BUP-CYP2D6 clinical DDIs. Therefore, accurate predictions of CYP2D6 inhibition in vivo required including the inhibition constants for all racemic or stereoisomeric BUP forms and metabolites, along with estimated steady-state liver concentrations. Fourth, inhibition constants and in vivo predictions suggest modest stereospecific inhibition of CYP2D6 by BUP and its metabolites. This, along with the marked stereoselective disposition of BUP metabolites (Masters et al, 2016a; Kharasch et al, 2020; Gufford et al, 2022), may differentially affect the strong and sustained in vivo DDIs observed between BUP and CYP2D6 substrates. Together, these findings provide valuable mechanistic insights and a quantitative understanding of BUP’s CYP2D6-dependent DDIs. Several drugs, including BUP, have metabolites that contribute to DDI risks, and incorporating inhibition constants for both the parent drug and its metabolites is essential for accurate predictions (Isoherranen et al, 2009; Yeung et al, 2011). Examples include norfluoxetine, the main metabolite of fluoxetine, enhances CYP2D6 inhibition, leading to stronger, longer-lasting DDIs (Sager et al, 2014), itraconazole metabolites boost its CYP3A inhibition, improving in vitro-to-in vivo extrapolation (Isoherranen et al, 2004; Templeton et al, 2008), and considering both amiodarone and its metabolites improves predictions of CYP2D6 and CYP2C9 inhibition (McDonald et al, 2015). Therefore, the approaches used in this study may offer a paradigm in vitro approach for evaluating the role of circulating metabolites and their stereoisomers as key factors in DDI risks.

Conflict of interest

The authors declare no conflicts of interest.