Human hepatocytes and cytochrome P450‐selective inhibitors predict variability in human drug exposure more accurately than human recombinant P450s

Abstract

Background and Purpose

Drugs metabolically eliminated by several enzymes are less vulnerable to variable compound exposure in patients due to drug–drug interactions (DDI) or if a polymorphic enzyme is involved in their elimination. Therefore, it is vital in drug discovery to accurately and efficiently estimate and optimize the metabolic elimination profile.

Experimental Approach

CYP3A and/or CYP2D6 substrates with well described variability in vivo in humans due to CYP3A DDI and CYP2D6 polymorphism were selected for assessment of fraction metabolized by each enzyme (fm_CYP) in two in vitro systems: (i) human recombinant P450s (hrP450s) and (ii) human hepatocytes combined with selective P450 inhibitors. Increases in compound exposure in poor versus extensive CYP2D6 metabolizers and by the strong CYP3A inhibitor ketoconazole were mathematically modelled and predicted changes in exposure were compared with in vivo data.

Key Results

Predicted changes in exposure were within twofold of reported in vivo values using fm_CYP estimated in human hepatocytes and there was a strong linear correlation between predicted and observed changes in exposure (r ² = 0.83 for CYP3A, r ² = 0.82 for CYP2D6). Predictions using fm_CYP in hrP450s were not as accurate (r ² = 0.55 for CYP3A, r ² = 0.20 for CYP2D6).

Conclusions and Implications

The results suggest that variability in human drug exposure due to DDI and enzyme polymorphism can be accurately predicted using fm_CYP from human hepatocytes and CYP‐selective inhibitors. This approach can be efficiently applied in drug discovery to aid optimization of candidate drugs with a favourable metabolic elimination profile and limited variability in patients.

Abbreviations

ADR

adverse drug reaction

AZ1

AZD1305

CL_int

clearance intrinsic

CYP

cytochrome P450

DDI

drug–drug interaction

EM

extensive metabolizer

ESI

electrospray ionization

F

oral bioavailability

f_a

fraction absorbed

f_g

fraction escaping gut metabolism

f_g,i

fraction escaping gut metabolism in presence of inhibitor

f_m

fraction of total elimination due to hepatic metabolism

fm_CYP

fraction of total hepatic clearance due to a specific P450 isoform

HLM

human liver microsomes

hrP450

human recombinant cytochrome P450

HSM

hepatocyte suspension medium

ISEF

inter system extrapolation factor

MRM

multiple reaction monitoring

NCE

new chemical entity

P450

cytochrome P450

PBPK

physiologically based pharmacokinetics

PK

pharmacokinetic

PM

poor metabolizer

RT

room temperature

Introduction

A drug being eliminated by multiple clearance pathways has a lower risk of large variations in drug exposure in the patient population, in comparison with a drug metabolically eliminated by a single enzyme. One source of variability is drug–drug interactions (DDI) as co‐administered drugs may inhibit or facilitate the metabolic clearance, which may result in higher or lower drug concentration, respectively, in patients than intended and an increased risk of adverse drug reactions (Lynch and Price, 2007) or lack of efficacy. Drugs cleared mainly via one enzyme are highly sensitive to inhibition or induction of this only clearance route. Today, polypharmacy is more a norm than an exception, especially in the area of cardiovascular and metabolic diseases in an ageing population where patients are commonly prescribed more than 10 medications (Rifkin et al., 2010), increasing the potential of undesirable DDI. Another source of variability during drug exposure is the dependency of clearance on a polymorphic enzyme, which may result in large variations in exposure due to the different phenotypes in the patient population (Dayer et al., 1985). The risk of adverse drug reactions is higher in poor metabolizers (PM) if the safety margins are narrow (Ingelman‐Sundberg, 2004). In ultra‐rapid metabolizers (UM), there is a risk of exposure below therapeutic levels and as a consequence a limited pharmacological effect. Prodrugs dependent on metabolism by a polymorphic enzyme for conversion to the active drug carry a risk of adverse drug reactions in UM due to elevated exposure of the active drug and lack of effect in PM due to low exposure of the active drug.

http://www.guidetopharmacology.org/GRAC/FamilyDisplayForward?familyId=242&familyType=ENZYME (CYP, P450) (Harding et al., 2018) is involved in the metabolic clearance of more than 90% of all drugs (Rendic and Guengerich, 2015), which makes P450 enzymes a major contributor to DDI, as described above. Therefore, quantifying the fraction of total hepatic clearance due to a specific P450 isoform (fm_CYP) of new chemical entities (NCE) in drug discovery is of critical importance and there are several in vitro methods available (Bohnert et al., 2016). One common method is to use a panel of human recombinant P450s (hrP450). To predict fm_CYP values, intersystem extrapolation factors (ISEF) combined with abundance of the individual P450s in the liver are used to scale intrinsic clearance (CL_int) values in hrP450s to CL_int in human liver microsomes (HLM) (Proctor et al., 2004). An alternative method is to incubate NCE with human liver microsomes or hepatocytes with and without selective P450 inhibitors and determine CL_int both in the presence and absence of inhibitor (Rodrigues and Wong, 1997; Li et al., 1999; 2007; Lu et al., 2008; Desbans et al., 2014).

http://www.guidetopharmacology.org/GRAC/ObjectDisplayForward?objectId=1337&familyId=263&familyType=ENZYME and the polymorphic http://www.guidetopharmacology.org/GRAC/ObjectDisplayForward?objectId=1329&familyId=262&familyType=ENZYME enzyme metabolize more than 50% of clinically used drugs (Zanger and Schwab, 2013; Rendic and Guengerich, 2015). For this reason, we selected a representative set of substrates metabolized to different extents by these two enzymes, for which robust in vivo pharmacokinetic (PK) data from DDI studies with http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=2568 as well as studies on extensive (EM) and poor (PM) metabolizers for CYP2D6 in humans are available (Figure 1). The aim of the present study was to assess the accuracy of a predicted increase in drug exposure when co‐administered with the potent CYP3A inhibitor ketoconazole and in CYP2D6 PM versus EM. Two in vitro models, primary human hepatocytes and recombinant P450 enzymes were applied, and the predicted results from in silico tools were compared to clinical in vivo data demonstrating variation in exposure due to CYP3A victim DDI or CYP2D6 polymorphism.

Figure 1.

Names, abbreviations and chemical structures of compounds used for assessment of AUC‐fold increases by co‐administration of ketoconazole and in CYP2D6 PMs.

Methods

Group size, randomization and blinding of experimental data

The assays evaluated, both hepatocytes and hrP450, have been validated in‐house previous to this work, and demonstrate low inter‐day variability in terms of estimated CL_int and fm_CYP. Therefore, due to our confidence in the robustness of these assays, it was decided to generate only n = 3 for this study. For a few of the compounds, data had been generated before this study was performed. These data were included in the study and therefore n = 4 or n = 5 for these compounds (Tables 1 and 2).

Table 1.

CL_int in hrP450 and human hepatocytes and inhibition of metabolite formation data in human hepatocytes

	CLint (μL·min−1·pmol−1)		CLint (μL·min−1·106 per cells)			% metabolites of total MS area
Compound	hrCYP3A	hrCYP2D6	In absence of inhibitor	In presence of ketoconazole	In presence of quinidine	In absence of inhibitor	In presence of ketoconazole	In presence of quinidine
ARI	7.9	5.4	NT	NT	NT	17	5.6	16
AZ1	1.8	0.067	NA	NA	NA	22	1.3	19
AZ2	0.47	0.71	NA	NA	NA	15	6.4	NT
AZ3	0.63	1.1	4.2b	1.8b	NT	NA	NA	NA
AZ4	0.72	1.2	6.5b	3.4b	NT	NA	NA	NA
BUF	0.099	13	6.8a	6.0a	1.4a	NA	NA	NA
LOR	9.7	8.4	11	8.5	5.4	NA	NA	NA
MET	0.010	7.7	NA	NA	NA	21	15	6.3
MDZ	7.9	0.023	18	2.4	18	NA	NA	NA
TAM	1.9	3.8	NA	NA	NA	7.3	2.6	4.3
TOL	2.3	15	18	16	4.6	NA	NA	NA

All data, n = 3 if not noted otherwise, are given as mean values. NT, not tested; NA, not applicable.

^a n = 5.

^b n = 4.

Abbreviations: ARI, aripiprazole; AZ1, AZD1305; BUF, bufuralol; LOR, loratadine; MET, metoprolol; MDZ, midazolam; TAM, tamsulosin; TOL, tolterodine.

Table 2.

Average % fm_CYP values estimated in hrP450s and human hepatocytes

	hrP450		Human hepatocytes
Compound	fmCYP3A	fmCYP2D6	fmCYP3A	fmCYP2D6
ARI	97	2.3	67	7.8
AZ1	>99	0.13	94	15
AZ2	93	2.0	58	NT
AZ3	95	5.5	56b	NT
AZ4	92	5.0	48b	NT
BUF	17	65	9.3a	78a
LOR	89	2.7	21	48
MET	4.3	94	28	70
MDZ	99	0.034	87	3.8
TAM	93	6.6	65	42
TOL	80	18	8.1	73

All data, n = 3 if not noted otherwise, are given as mean values. NT, not tested.

^a n = 5.

^b n = 4.

Abbreviations: ARI, aripiprazole; AZ1, AZD1305; BUF, bufuralol; LOR, loratadine; MET, metoprolol; MDZ, midazolam; TAM, tamsulosin; TOL, tolterodine.

No systematic bias was observed to be associated with the compound position in the incubation plate or in the order of sample analysis on the LCMS instrument. Therefore, randomization of samples was deemed unnecessary.

Blinding of samples was not undertaken since prior knowledge of sample content was not expected to introduce any bias of the results. The procedure for compound incubation was performed according to a standardized methodology. Furthermore, LCMS analysis of samples and processing of data were controlled by the software used and interference by the experimentalist could be excluded.

Thawing of cryopreserved hepatocytes

A 10‐donor pool of cryopreserved hepatocytes was used for all experiments. The pool consisted of five female and five male donors, age between 19 and 80 years. The pool was characterized with respect to drug metabolizing enzymes (specific P450 activities and general activities of conjugating enzymes), and activities are at average or above compared with individual donors available at the purchase. Cryopreserved cells were moved from −150°C freezer and immediately immersed in a pre‐heated water bath kept at 37°C. When only a small ice crystal was left, the content was emptied into 50 mL hepatocyte suspension medium (HSM; William's medium E supplemented with 25 mM HEPES and 2 mM L‐glutamine, adjusted to pH 7.4) and centrifuged for 6 min at 80× g at room temperature (RT). The medium was discarded, and cells were re‐suspended in 50 mL Percoll solution and centrifuged at 100× g for 15 min at RT. The resulting pellet was resuspended in HSM; viability was determined by the trypan blue exclusion method, and cells were diluted to 2 ×10⁶ cells^. mL^‐1 in HSM. Minimum accepted viability was 75%.

CL_int experiments in human hepatocytes

For the test compounds and inhibitors, 10 mM stock solutions were prepared in DMSO. In triplicate, 10 mM DMSO test compound solutions were diluted with 50% acetonitrile to a final concentration of 100 μM. In one of the triplicates, ketoconazole was added to a concentration of 300 μM; in one triplicate, http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=2342 was added to a concentration of 100 μM, and to the third triplicate, only DMSO was added. These solutions were further diluted with HSM giving a test compound concentration of 2 μM and ketoconazole and quinidine concentrations of 6 and 2 μM respectively.

Hepatocyte suspension (25 μL) was added to each well in 96‐well plates. Each plate represented a discrete incubation time point. The plates were pre‐incubated for 10 min in an incubator at 37°C, 5% CO₂, and reactions were started by addition of 25 μL of substrate/inhibitor solution to the cell‐containing wells. The final test compound concentration in the incubation was 1 μM, and ketoconazole and quinidine concentrations were 3 and 1 μM respectively. The solvent concentration in the incubations was 0.04% DMSO and 0.48% acetonitrile.

Following brief, gentle shaking, plates were placed back in the incubator. The reactions were stopped after 2, 15, 30, 45, 60, 75, 90, 120, 150 and 180 min by addition of three volumes (150 μL) of ice‐cold stop solution (acetonitrile containing 0.8% formic acid and 5,5‐diethyl‐1,3‐diphenyl‐2‐iminobarbituric acid as internal standard). Plates were centrifuged at 3200 × g at 4°C for 20 min. The supernatant (80 μL) was transferred to a new plate and diluted with an equal volume of water and was analysed for parent compound using LC‐MSMS.

The analysis was performed on a Waters Quattro Ultima mass spectrometer with electrospray ionization (ESI) interface (Waters, Milford, MA, USA), an Agilent 1100 Series binary pump (Hewlett Packard GmbH, Waldbronn, Germany) configured with a CTC HTS PAL auto sampler (CTC Analytics AG, Zwingen, Germany). The analytical column used for chromatographic separation was an Atlantis T3 column (3 μm, 2.1 × 30 mm, Waters, Milford, MA, USA), and the mobile phases consisted of water (A) and acetonitrile (B) both containing 0.2% formic acid. The gradient used was 4% B for 0.1 min, followed by a linear increase to 95% B in 1.2 min at a flow rate of 0.7 mL·min⁻¹ and a column temperature of 55°C. The injection volume was 5 μL. The samples were analysed with multiple reaction monitoring (MRM) in positive ionization mode. Data were acquired using MassLynx 4.1 acquisition software with Waters QuanLynx data processing.

Metabolite formation experiments in human hepatocytes

The test compound solutions were prepared according to the description for the CL_int experiment with slight modification. The final test compound concentration was 4 μM, giving an intermediate test compound concentration of 400 μM. The organic solvent concentrations in the incubations were 0.07% DMSO and 0.47% acetonitrile.

The experiments were performed in the same manner as the CL_int experiments although only 0 and 240 min samples were taken, and samples were analysed using a high resolution QToF MS instrument for formation of metabolites.

Metabolite formation analysis was performed on an Acquity UPLC system interfaced with a Synapt G2 mass spectrometer (Waters, Milford, MA, USA). Chromatographic separation of metabolites and parent was carried out on an Acquity UPLC BEH C18 column (1.7 μm, 2.1 × 100 mm). The mobile phases consisted of 0.1% formic acid in water (A) and acetonitrile (B). The LC gradient was as follows: 0–6 min 10–70% B, 6–6.7 min 90% B at a flow rate of 0.5 mL·min⁻¹, and the column temperature was set to 45°C. The Synapt G2 was operated under positive ESI conditions with a capillary voltage of 0.5 kV and cone voltage of 20 V. The data acquisition was performed using MS^E with a mass range of 80–1000 Da. For the low energy MS^E acquisition, the trap and transfer energy were set to 4 and 3 V, respectively, while for the high energy MS^E acquisition, the trap was ramped from 15 to 45 V and the transfer energy was held fixed at 12 V. Data were collected in centroid mode. Leucine‐enkephaline was used as a lock mass (m/z 556.2771) for internal calibration at a concentration of 250 pmol·μL⁻¹ and a flow rate of 40 μL·min⁻¹. MassLynx 4.1 (Waters, Milford, MA, USA) was used for the data acquisition.

CL_int experiments in hrP450s

Each compound was incubated individually in nine P450 isoforms: CYP1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1 and 3A4. Test compound solutions were prepared by adding 4 μL of 10 mM stock solution in DMSO to 196 μL of acetonitrile giving a concentration of 200 μM. The hrP450 enzymes were thawed on ice and then diluted in 0.1 M, pH 7.4 phosphate buffer. Test compound was added to the enzyme solution, and reaction mixtures were pre‐incubated for 15 min at 37°C. Reactions were started by addition of NADPH (final concentration of 1 mM) and carried out at 37°C. Final concentrations of test compounds and CYP enzymes were 2 μM and 100 nM respectively. The final organic solvent concentrations in the incubations were 0.02% DMSO and 0.98% acetonitrile. After 0, 5, 10, 15 and 25 min aliquots were taken from the reaction mixture to a new plate and quenched with 120 μL of stop solution (acetonitrile containing internal standard). After quenching of the reaction, plates were centrifuged at 3200 × g for 20 min, placed at 4°C for 30 min, and then re‐centrifuged at 3200 × g for 20 min to precipitate proteins. After the centrifugation, the supernatant was transfered to a new plate, diluted with an equal volume of water and analysed with LC‐MSMS for the parent compound remaining.

The analysis was performed on a Waters XEVO TQD mass spectrometer (Waters, Milford, MA, USA) equipped with an Acquity I‐class UPLC (Waters, Milford, MA, USA) configured with an Acquity Sample Management FTN auto sampler (Waters, Milford, MA, USA). The analytical column used for chromatographic separation was an Aquity UPLC BEH C18 column (1.7 μm, 2.1 × 50 mm, Waters, Milford, MA, USA), and the mobile phases consisted of water (A) and acetonitrile (B) both containing 0.1% formic acid. The LC gradient was as follows: 0–0.2 min 5% B, 0.2–1.0 min 5–100% B, 1.0–1.4 min 100% B, 1.4–1.6 min 100–5% B, 1.6–2.0 min 5% B at a flow rate of 0.5 mL·min⁻¹ and the column temperature was set to 40°C. The injection volume was 1 μL. The samples were analysed with MRM in positive or negative ionization mode. Data were acquired using MassLynx 4.1 acquisition software and data were processed with Waters QuanLynx Software.

Calculations of CL_int values

For experiments with both hepatocytes and hrP450s, the slope value, k, was determined by linear regression of the natural logarithm of % parent remaining versus incubation time curve. The t _½ was determined from the slope value according to Equation (1).

$$t_{\mathrm{½}}\left(min\right)=\frac{\mathit{ln}2}{k}$$ (1)

The t _½ value is converted into the CL_int using Equation (2a) or (2b).

$${\mathrm{CL}}_{\mathrm{int}},\text{hepatocytes}\left(\mu \mathrm{L}{min}^{‐1}\mathrm{per}\text{million}{\text{cells}}^{‐1}\right)=\frac{ln2\times \text{volume of}\text{incubation}\left(\mu \mathrm{L}\right)}{{\mathrm{t}}_{\mathrm{½}}\left(min\right)\times \text{amount}\text{of}\text{cells}\left(\text{millions}\right)}$$ (2a)

$${\mathrm{CL}}_{\mathrm{int}},\text{hrCYPi}\left(\mu \mathrm{L}{min}^{‐1}{\left(\text{pmol}\mathrm{CYP}\right)}^{‐1}\right)=\frac{ln2\times \text{volume of}\text{incubation}\left(\mu \mathrm{L}\right)}{{\mathrm{t}}_{\mathrm{½}}\left(min\right)\times \text{amount}\text{of}\text{CYPi}\left(\text{pmol}\right)}$$ (2b)

Calculations of fm_CYP values

From CL_int values determined from hepatocytes, fm_CYP2D6 and fm_CYP3A were calculated according to Equation (3) and expressed as % reflecting the contribution of each enzyme to the overall hepatic metabolism of the compound of interest. These fm_CYP values were used as entry parameters for in vitro to in vivo predictions of the DDI risk as well as to allow comparison across different in vitro systems.

$${\mathrm{f}}_{\mathrm{m}},{\mathrm{CYP}}_{\mathrm{i}}=\frac{{\mathrm{CL}}_{\mathrm{int}}\text{without inhibitor}–{\mathrm{CL}}_{\mathrm{int}}\text{with inhibitor}}{{\mathrm{CL}}_{\mathrm{int}}\text{without inhibitor}}$$ (3)

For metabolite formation experiments, the area of all metabolites identified as compound related and the area of the parent compound was summarized with the aid of the MetaboLynx Software (Waters, Milford, MA, USA) and the area % for each entity was calculated. The sum of the area % of the metabolites was then used for the calculations according to Equation (4). It is assumed that the MS response is similar for all metabolites.

$$\%\text{inhibition}=\frac{\%\text{metabolites without inhibitor}-\%\text{metabolites with inhibitor}}{\%\text{metabolites without inhibitor}}\times 100$$ (4)

For hrP450s, the % contribution from an individual CYP isoform was calculated according to Equation (5).

$$\%\text{contribution}=\frac{{\mathrm{CL}}_{\mathrm{int}},{\text{hrCYP}}_{\mathrm{i}}\times \text{ISEF},{\text{hrCYP}}_{\mathrm{i}}\times \text{median abundance}}{{\sum }_{\mathrm{i}=0}^{\mathrm{N}}{\mathrm{CL}}_{\mathrm{int}},{\text{hrCYP}}_{\mathrm{i}}\times \text{ISEF},{\text{hrCYP}}_{\mathrm{i}}\times \text{median abundance}}\times 100$$ (5)

ISEFs were estimated in‐house for CYP3A4 and CYP2D6 (compounds in the study are predominantly metabolized by these isoforms to different extents) based on CL_int estimated by substrate depletion of probe substrates in a pool of HLM (150 donors) as well as in hrP450s and median hepatic abundance of the respective P450 given by Simcyp (Table S1) according to the published procedure (Chen et al., 2011) (Equation (6)). http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=3342 and bufuralol were selected as probe substrates for CYP3A4 and CYP2D6 respectively and ISEFs for both CYPs were in agreement with ISEFs from Simcyp (Table S1) that were used in the calculations.

$$\text{ISEF}=\frac{{\mathrm{CL}}_{int,\mathrm{HLM}}}{\text{Abundance}\mathrm{x}{\mathrm{CL}}_{int,\text{hrCYP}}}$$ (6)

In vitro to in vivo predictions

DDI predictions and comparison between EM and PM were performed using two different methods: static (with fixed inhibitor concentrations at a steady state level) and dynamic (with inhibitor concentrations that vary over time). For both methods, the predicted output was expressed as fold AUC change to allow comparison with clinical data regardless of the absolute concentration of the drugs in circulation. AUC‐fold change is a parameter required by regulatory agencies as a measure of DDI extent or inter individual variability. For both approaches, the unbound hepatic inlet concentration of ketoconazole as inhibitor was used. The dynamic and physiologically based PK (PBPK) modelling were performed with the Simcyp software (version 16, Certara, Sheffield, UK). A subset of drugs (http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=360, http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=553, http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=7216) and an AstraZeneca drug candidate (AZD1305) (Johnson et al., 2012) were selected for the PBPK modelling and all drug‐specific parameters for these drugs are given in Table S2. Physicochemical and in vitro data were generated for the compounds when literature data were not available. The fm_CYP3A and fm_CYP2D6 values estimated from human hepatocyte experiments were used for the simulations, and the retrograde method within the software was used to incorporate these values. The clearance values that were used as input can be found in Table S2. The default healthy population (n = 100) was used as a starting point for the simulations and separated into two populations with 100% EM and PM respectively. The default gender (50% male and female) and age settings (20–50 years) were used for the simulations. In each run, n = 10 trials with n = 10 individuals were simulated. Minor changes were made to the default software compound files for tolterodine and metoprolol. Two separate compound files were used for tolterodine and metoprolol, respectively, using different clearance values for the EM and PM populations. For loratadine, a new model was built with a combination of a ‘bottom‐up’ and ‘top‐down’ methodology (Tsamandouras et al., 2015), see Table S2 for input data. In order to get an estimate of clearance for loratadine, in vitro data were used (total hepatocyte CL_int) and scaled to a total clearance. The default software ketoconazole compound file was used for the 200 mg b.i.d. and the 400 mg once daily simulations. The model for AZD1305 has been published previously (Johansson et al., 2016), and only minor changes were made to incorporate the fm_CYP3A data.

The equation used for the static modelling is shown below (Equation (7)) (Obach, 2009).

$$\frac{{\mathrm{AUC}}_{\mathrm{i}}}{\mathrm{AUC}}=\frac{{\mathrm{f}}_{\mathrm{g},\mathrm{i},}}{{\mathrm{f}}_{\mathrm{g}}}\times \frac{1}{\frac{\sum {\mathrm{f}}_{\mathrm{m}}\times {\mathrm{f}}_{\mathrm{m},\mathrm{CYP}}}{1+\sum \frac{{\mathrm{I}}_{\mathrm{u}}}{{\mathrm{K}}_{\mathrm{i},\mathrm{u}}}}+1-\sum {\mathrm{f}}_{\mathrm{m}}\times {\mathrm{f}}_{\mathrm{m},\mathrm{CYP}}}$$ (7)

AUC_i

Area under the concentration time‐curve for drug in presence of inhibitor

AUC

Area under the concentration time‐curve for drug in absence of inhibitor

f_g,i

Fraction escaping gut metabolism in presence of inhibitor

f_g

Fraction escaping gut metabolism in absence of inhibitor

f_m

Fraction of total elimination due to hepatic metabolism

f_m,CYP

Fraction of total hepatic clearance due to a specific P450 isoform

I_u

Unbound inhibitor concentration

K_i,u

Unbound inhibitor constant

The values for fraction escaping gut metabolism (f_g) and fraction of total elimination due to hepatic metabolism (f_m) are given in Table S3, and the fraction escaping gut metabolism in presence of inhibitor (f_g,i) was assumed to be 1 for all compounds when ketoconazole is used as inhibitor. The f_g‐values were estimated according to Equation (8). Details are given in Table S3.

$${\mathrm{f}}_{\mathrm{g}}=\frac{\mathrm{F}}{{\mathrm{f}}_{\mathrm{h}}\mathrm{x}{\mathrm{f}}_{\mathrm{a}}}$$ (8)

f_g

Fraction escaping gut metabolism

F

Oral bioavailability

f_h

Fraction escaping hepatic metabolism

f_a

Fraction absorbed

The f_h and f_m‐values were calculated according to Equations (9) and (10) respectively assuming the compounds are cleared mainly via hepatic metabolism in addition to renal clearance. Values for total clearance, renal clearance and fraction absorbed (f_a) were retrieved from the literature (Table S3), but no numbers for f_a have been reported previously. For all the compounds studied, the f_a was stated to be good and fast and was therefore assumed to be 0.95–1 for all compounds. This assumption was supported by the intrinsic Caco‐2 cell apparent permeability values (Fredlund et al., 2017).

$${\mathrm{f}}_{\mathrm{h}}=1-\frac{{\mathrm{CL}}_{\mathrm{tot}}-{\mathrm{CL}}_{\mathrm{r}}}{{\mathrm{Q}}_{\mathrm{h}}}$$ (9)

$${\mathrm{f}}_{\mathrm{m}}=1-\frac{{\mathrm{CL}}_{\mathrm{r}}}{{\mathrm{CL}}_{\mathrm{tot}}}=1-{\mathrm{f}}_{\mathrm{e}}$$ (10)

f_m

Fraction of total elimination due to hepatic metabolism

CL_r

Renal clearance

f_e

Fraction renally excreted

CL_tot

Total clearance

Q_h

Hepatic blood flow set to 20 mL·min⁻¹·kg⁻¹

For the predictions of the AUC‐ratios in the presence and absence of ketoconazole, the unbound concentration of ketoconazole, I_u, was calculated according to Equation (11) (Einolf, 2007).

The dose, D, and dosing interval, Ƭ, of ketoconazole in the different interaction studies are given in Table S4. The unbound inhibitor constant, K_i,u, was set to 0.015 μM for ketoconazole. For the predictions of AUC increase in CYP2D6 PMs, ‘dummy values’ for I_u (400 000) and K_i,u (0.004) were used to achieve 100% inhibition of CYP2D6 to mimic the CYP2D6 PM phenotype.

$${\mathrm{I}}_{\mathrm{u}}={\mathrm{I}}_{\mathrm{in}}{,}_{\mathrm{u}}={\mathrm{I}}_{\mathrm{ss}}\times {\mathrm{f}}_{\mathrm{u}}+\frac{{\mathrm{k}}_{\mathrm{a}}\times {\mathrm{f}}_{\mathrm{a}}\times {\mathrm{f}}_{\mathrm{g}}\times \mathrm{D}\times {\mathrm{f}}_{\mathrm{u}}}{{\mathrm{Q}}_{\mathrm{pv}}}$$ (11)

I_u = I_in,_u

Maximum unbound hepatic inlet concentration of ketoconazole

I_ss

Average systemic plasma concentration of ketoconazole after repeated oral administration

f_u

Human plasma protein binding (3% for ketoconazole, from Simcyp)

k_a

Absorption rate constant (0.032 h⁻¹ for ketoconazole, from Simcyp)

f_a

Fraction absorbed (1 for ketoconazole, from Simcyp)

f_g

Fraction escaping gut metabolism (0.5 for ketoconazole, from Simcyp)

D

Administered dose of ketoconazole

Q_pv

Portal vein blood flow assumed to be 1148 mL·min⁻¹ (Davies and Morris, 1993)

Data and statistical analysis

For CL_int experiments, the quality of the data were judged by r ² or standard error of the slope. If these were not accepted due to low metabolism (r ² < 0.8 or P > 0.05 for t‐test of standard error of the slope), the compounds were tested in the inhibition of metabolite experiments instead. All calculations were carried out by using Microsoft Excel and XLfit add‐in version 5.2.2.

The data and statistical analysis comply with the recommendations on experimental design and analysis in pharmacology (Curtis et al., 2015). All experiments were performed according to validated procedures, and n = 3 was considered sufficient for each experiment. Positive controls were included to qualify each experiment. For some compounds, data already existed and therefore n > 3 for these.

Root mean square error (RMSE) and mean error (ME) of the predicted versus observed AUC‐ratios were calculated according to Equations (12) and (13) respectively. RMSE describes precision and ME accuracy of the predicted AUC‐ratios. High precision is reflected by a low RMSE value and high accuracy by an ME value close to zero.

$$\text{RMSE}=\sqrt{\frac{1}{N}\sum _{i=1}^N{\left(y_i-x_i\right)}^2}$$ (12)

$$\mathrm{ME}=\frac{1}{N}\sum _{i=1}^N\left(y_i-x_i\right)$$ (13)

N

Total number of observations in the analysis

y_i

Predicted AUC‐ratio

x_i

Observed AUC‐ratio

Materials

Human cryopreserved hepatocytes were purchased from CelsisIVT (Baltimore, MD, USA). Recombinant human CYP1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1 and 3A4 enzymes were from CYPEX (Dundee, UK). Ketoconazole, loratadine, metoprolol, midazolam, quinidine, L‐glutamine, and William's medium E without phenol red were purchased from Sigma Aldrich (Gillingham, Dorset, UK). Bufuralol was from Toronto Research Chemicals Inc. (Ontario, Canada). HBSS and HEPES were from Gibco Life Technologies (Paisley, UK). Percoll separating solution was purchased from Biochrom AG (Berlin, Germany). NADPH was from Roche Diagnostics GmbH (Mannheim, Germany). DMSO was purchased from Sigma Aldrich (Gillingham, Dorset, UK) or Solarbio S&T Co. LTD (Beijing, PR China). Acetonitrile 100% was from Rathburn Chemicals (Walkerburn, Scotland) or Merck KgaA (Darmstadt, Germany). Formic acid 98–100% was from Merck KgaA (Darmstadt, Germany) or ROE Scientific INC (Delaware, USA). http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=34, http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=488, tolterodine, 5,5‐diethyl 1,3‐diphenyl‐2‐iminobarbituric acid (used as internal standard) and AstraZeneca proprietary compounds (AZD1305, 2, 3 and 4) were provided by AstraZeneca Compound Management Department.

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 (Harding et al., 2018), and are permanently archived in the Concise Guide to PHARMACOLOGY 2017/18 (Alexander et al., 2017).

Results

Estimation of fm_CYP3A and fm_CYP2D6 by hrP450 and human hepatocytes

Eight well described (Figure 1) and three AstraZeneca proprietary compounds, all of them CYP3A and/or CYP2D6 substrates metabolized by CYP3A and CYP2D6 to different extent were selected for this study. The fm_CYP of these compounds were estimated in human hepatocytes incubated in the presence and absence of the selective CYP3A inhibitor ketoconazole or the selective CYP2D6 inhibitor quinidine (Li et al., 1999) (Table 2). For six of the selected compounds, CL_int was determined by the well‐established substrate depletion method (Table 1). The change in rate of elimination is depicted for the CYP3A substrate midazolam when co‐incubated with ketoconazole (compared with midazolam alone) (Figure S1A) and for the CYP2D6 substrate bufuralol in the presence and absence of quinidine (Figure S1B). For five of the 11 compounds, the CL_int values were below the limit of quantification. The fraction metabolized by CYP3A and CYP2D6 for these substances was determined instead by measuring the inhibition of metabolite formation following co‐incubation with ketoconazole and quinidine respectively (Leandersson et al., poster presented at the 19^th ISSX meeting in San Francisco 2014) (Table 1).

The fm_CYP3A and fm_CYP2D6 values were also estimated from CL_int using hrP450s (substrate depletion method, CL_int values given in Table 1) combined with scaling factors (Table 2). The results from hrP450s clearly predicted a larger fraction metabolized by CYP3A and a smaller fraction by CYP2D6 as compared with the results obtained in human hepatocytes (Figures 2 and 3). Metoprolol was an outlier showing a higher fm_CYP3A when estimated in hepatocytes as compared with hrP450s and vice versa when calculating the fm_CYP2D6. For all compounds, except for bufuralol and metoprolol, fm_CYP3A was predicted to be 80% or more of the total hepatic metabolism (less than 20% by CYP2D6). In contrast to hrP450s, human hepatocytes and selective inhibitors captured a wide range of estimated fm_CYP3A (8–94%) and fm_CYP2D6 (4–78%) suggesting a wider dynamic range of that method.

Figure 2.

Correlation between the average fm_CYP3A estimated in hrP450s versus human hepatocytes for 11 CYP3A and/or CYP2D6 substrates. The fm_CYP3A was estimated in hepatocytes by comparing the metabolic turnover, following incubation of the compounds in the absence and presence of the selective CYP3A inhibitor ketoconazole. In hrP450s, CL_int was estimated in 10 individual hrP450 isoforms and scaled up to an in vivo CL_int by applying ISEFs and abundance for each individual P450 enzyme. The fm_CYP3A was determined from the fraction of scaled in vivo CL_int,CYP3A over Σscaled in vivo CL_int,CYPi. Data represent mean values.

Figure 3.

Correlation between the average fm_CYP2D6 estimated in hrP450s versus human cryopreserved hepatocytes for eight CYP2D6 and/or CYP3A substrates. The fm_CYP2D6 was estimated in hepatocytes by comparing metabolic turnover in absence and presence of the selective CYP2D6 inhibitor quinidine. In hrP450s, CL_int was estimated in 10 individual hrP450 isoforms and scaled up to an in vivo CL_int by applying ISEFs and abundance for each individual P450 enzyme. The fm_CYP2D6 was determined from the fraction of scaled in vivo CL_int,CYP2D6 over Σscaled in vivo CL_int,CYPi. Data represent mean values.

Prediction of clinical CYP3A victim DDI risk using a static equation

For six of the studied compounds (aripiprazole, AZD1305, loratadine, midazolam, tamsulosin and tolterodine), information on the interaction potential by ketoconazole in vivo in humans is available (Table 3). The change in AUC by ketoconazole was compared with the predicted change in AUC from the two in vitro models applying Equation (7) (see Methods for details). Overall, there was an excellent agreement between observed and predicted AUC‐fold increases when using fm_CYP3A values estimated in human hepatocytes. The predicted AUC‐ratios were found to deviate less than twofold from what was observed in vivo, and there was no obvious systematic error in the predictions (RMSE = 0.98, ME = −0.4) (Figure 4A, Table 3). Consequently, there was a strong correlation where r ² of the regression line for predicted versus observed AUC‐ratios was high (0.83). Utilization of fm_CYP3A values estimated from hrP450 resulted in poor predictions where AUC‐fold increases were considerably overpredicted (RMSE = 9.8, ME = 6.5, r ² = 0.55) (Figure 4B, Table 3).

Table 3.

Observed and predicted AUC‐fold increase by co‐administration of ketoconazole and in CYP2D6 PM versus EM in humans

	Observed		Predicted
			Hepatocytes PBPK		Hepatocytes static		hrP450 static
Compound	ketoconazole	PM	ketoconazole	PM	ketoconazole	PM	ketoconazole	PM
ARI	1.6a	1.8a, 1.7b	ND	ND	2.3	1.1	4.8	1.0
AZ1	7.7c	NA	4.8 (4.3–5.4)n	NA	6.4	NA	8.4	NA
BUF	NA	6.3d	NA	ND	NA	4.4	NA	2.8
LOR	4.5e, 3.1f	NA	1.6 (1.6–1.7)n , o	NA	2.3	NA	11	NA
MET	NA	6.0g	NA	2.2 (1.9–2.4)n	NA	3.1	NA	11
MDZ	9.5h, 9.4i	NA	ND	NA	10	NA	32	NA
TAM	2.8j	1.6k	ND	ND	2.3	1.6	5.5	1.1
TOL	2.1l	10m	1.1 (1.1–1.1)n	10.5 (8.9–12.2)n	1.5	3.5	9.5	1.2

NA = not applicable, ND = not determined.

Abbreviations: ARI, aripiprazole; AZ1, AZD1305; BUF, bufuralol; LOR, loratadine; MET, metoprolol; MDZ, midazolam; TAM, tamsulosin; TOL, tolterodine.

^aVieira et al. (2014).

^bHendset et al. (2007).

^cJohansson et al. (2016).

^dDayer et al. (1985).

^eChaikin et al. (2005).

^fKosoglou et al. (2000).

^gLennard et al. (1982).

^hChung et al. (2006).

ⁱChen et al. (2006).

^jTroost et al. (2011).

^kChoi et al. (2012).

^lBrynne et al. (1999).

^mMalhotra et al. (2011).

ⁿAUC‐ratio presented as geometric mean ratio and 90% CI.

^oSimulated with the study design of Chaikin et al., 2005.

Figure 4.

Correlation between predicted and observed AUC‐fold increase in humans following oral co‐administration of the selective CYP3A inhibitor ketoconazole. Line of unity (solid), twofold deviation (dashed line), regression line (dotted). The fm_CYP3A values estimated in human hepatocytes (A) and hrP450s (B) were applied for the predictions using a static equation.

Prediction of variation in compound exposure in CYP2D6 PM versus EM phenotypes using a static equation

The fm_CYP2D6 values estimated by human hepatocytes were modelled using the static equation in order to predict the difference in AUC between CYP2D6 PM and EM for five compounds (aripiprazole, bufuralol, metoprolol, tamsulosin and tolterodine) metabolized by CYP2D6 to different extents. The predicted AUC‐ratio in PM versus EM was in general somewhat underestimated but, except for tolterodine (2.9‐fold underpredicted), in agreement (within twofold) with clinical observations (RMSE = 3.1, ME = −1.5). Also, there was a good correlation between the predicted versus observed AUC‐ratios (r ² = 0.82) (Figure 5A, Table 3). When fm_CYP2D6 values estimated by hrP450s were used, the AUC‐ratio was predicted within twofold, except for tolterodine (8.3‐fold underpredicted) and bufuralol (2.3‐fold underpredicted). The overall precision was poor, and there was no correlation between predicted versus observed AUC‐ratio (RMSE = 4.6, ME = −0.85, r ² = 0.20) (Figure 5B, Table 3).

Figure 5.

Correlation between the predicted and the observed AUC‐fold increase in CYP2D6 poor metabolizers compared with extensive metabolizers. Line of unity (solid), twofold deviation (dashed line), regression line (dotted). The fm_CYP2D6 values estimated in human hepatocytes (A) and hrP450s (B) were applied for the predictions using a static equation.

Prediction of clinical victim DDI risk using PBPK modelling and simulation

For the three compounds AZD1305, loratadine and tolterodine, fm_CYP3A values estimated from human hepatocytes were assessed by PBPK modelling approach in order to predict the AUC‐fold increase following oral co‐administration with ketoconazole. The predicted AUC‐ratios were similar to what was predicted by the static equation with a high degree of correlation between predicted and observed ratios (Figure 6A, Table 3).

Figure 6.

Correlation between predicted and observed AUC‐fold increase in humans for AZD1305 (AZ1), loratadine (LOR) and tolterodine (TOL) following oral co‐administration of the selective CYP3A inhibitor ketoconazole (A). Correlation between predicted and observed AUC‐fold increase in CYP2D6 PMs compared with EMs for metoprolol (MET) and tolterodine (6). Line of unity (solid), twofold deviation (dashed line). The fm_CYP3A and fm_CYP2D6 values respectively estimated in human hepatocytes were applied for the predictions using a static equation (triangles) and Simcyp (squares).

For the two CYP2D6 substrates metoprolol and tolterodine, fm_CYP2D6 values estimated by human hepatocytes were modelled in order to predict the AUC‐ratio between CYP2D6 PM and EM. The AUC‐ratio for metoprolol was about threefold underpredicted by the PBPK model while twofold underpredicted by the static equation. For tolterodine, the AUC‐ratio was well predicted by the PBPK model while threefold underpredicted by the static equation (Figure 6B, Table 3).

Discussion

Two major goals in drug discovery and development programs are to ensure human safety and efficacy of candidate drugs. Therefore, it is of outmost importance to have an accurate prediction of variability in human exposure and potential DDI risks in patients. In this study, two alternative in vitro approaches, human hepatocytes combined with P450‐selective inhibitors and hrP450s, were used to assess the risk of CYP3A victim DDI and differences in exposures between CYP2D6 EM and PM for a set of drugs with previously reported clinical victim DDI and PK profiles in humans. The compounds investigated (aripiprazole, bufuralol, loratadine, metoprolol, midazolam, tamsulosin, tolterodine, AZD1305 and three AstraZeneca proprietary compounds) are all metabolized by CYP3A and CYP2D6 to different extents. There is no significant non‐P450 mediated metabolism reported for any of these compounds (Kamimura et al., 1998; Patki et al., 2003; Zhou et al., 2009).

Using fm_CYP values from human hepatocytes, modelled by a static equation, the prediction of AUC increase due to CYP3A victim DDI and the difference in AUC between CYP2D6 PM versus EM was within twofold of reported in vivo data except for one substance (tolterodine). We expect that utilizing the fm_CYP data generated using human hepatocytes will allow us to distinguish between and to rank order drug candidates with respect to potential variations in exposure due to P450 victim DDIs and/or P450 polymorphism. This is of great value for drug design efforts in the drug discovery phase to identify and optimize compounds for a balanced metabolic profile, increasing the safety margins and decreasing the likelihood of adverse drug reactions ultimately leading to development of safer drugs for patients.

In human hepatocytes, the metabolic turnover of the well‐known CYP2D6 substrate bufuralol was, as expected, found to be inhibited significantly by quinidine. Bufuralol was also inhibited to some extent by ketoconazole (Table 1) suggesting metabolism by CYP3A, which to our knowledge, has not been reported previously. Instead, in addition to CYP2D6, bufuralol has been reported to be metabolized to some extent by CYP1A2 and CYP2C9 (Rendic and Di Carlo, 1997). Bufuralol has also been reported to be conjugated by glucuronic acid (Francis et al., 1982), which may partly explain the inhibition of its metabolism since certain UDP glucuronosyltranferases are known to be inhibited by ketoconazole (Yong et al., 2005). It was confirmed in our study that glucuronidation of bufuralol is a minor pathway of the metabolism of bufuralol which was partly inhibited by ketoconazole (data not shown).

Prediction of DDI potential and differences in PK profiles in CYP2D6 PM versus EM was less accurate when using fm_CYP values estimated in hrP450s compared with human hepatocytes. A potential caveat with the approach using hrP450s and scaling factors was recently demonstrated (Siu and Lai, 2017). Depending on the substrates selected for estimation of scaling factors, the fm_CYP values could vary significantly. This was exemplified by losartan where fm_CYP3A4 was predicted between 12–25% (using testosterone as CYP3A substrate) and 78–91% (using midazolam or nifedipine as CYP3A substrates). In a study by Chen and colleagues (2011), the accuracy of fm_CYP predicted by hrP450 was demonstrated to be in acceptable agreement with the observed fm_CYP using HLM and inhibitors. One outlier in this study was the only CYP3A and CYP2D6 mixed substrate for which a more than 10‐fold discrepancy was observed between fm_CYP2D6 when estimated by the two separate in vitro methods. This demonstrates the unsuitability of hrP450 approach for this mixed CYP3A/CYP2D6‐substrate, which is well in line with our findings for the substrates in the present study, selected for their mixed CYP3A/2D6 properties. One explanation for the issue related to studying individual P450s in isolation is the complexity of P450 function described in a recent review, such as interactions with redox partners, allosteric mediators and also with other P450 isoforms that will affect the P450 activity (Kandel and Lampe, 2014). Disrupting any of these interactions might lead to altered metabolic pathways. By using hepatocytes, a more complete ‘in vivo‐like’ in vitro system compared with hrP450s, the risk of incorrect estimation of fm_CYP is alleviated. In addition, hepatocytes represent a more complete in vitro system with regard to drug metabolizing enzymes expressing non‐P450 phase I as well as phase II enzymes. Ignoring these additional metabolic pathways could lead to erroneous prediction of victim DDIs.

The concept behind the static equation (details given in Methods) applied for assessment of AUC‐ratios due to CYP3A victim DDI and CYP2D6 polymorphism is well established, originally described by Rowland and Matin (1973). The approach with a static equation has proved to be simple to use and give accurate predictions. One key input parameter in the equation is the inhibitor concentration. The most relevant inhibitor concentration to be used in the equation has been debated in the literature. It has been shown that the use of hepatic inlet concentration is preferred rather than systemic plasma concentrations for prediction of reversible DDI (Ito et al., 2004; Obach, 2009). Therefore, predicted free hepatic inlet concentration was applied as the inhibitor concentration in our study.

Advanced PBPK based models that integrate system‐specific components as P450 abundance, differences within a population and drug specific components such as CL_int, fraction unbound in plasma and physicochemical properties can be built using commercial software (Rostami‐Hodjegan and Tucker, 2007). Several studies have shown the usefulness of PBPK modelling in predicting CYP3A DDI risk as well as capturing AUC variability for drugs that are predominantly metabolized by CYP2D6 (Vieira et al., 2014; Yoshida et al., 2016). In this work, we demonstrated the usefulness of accurately determined fm_CYP data for the build‐up of PBPK models. Simulations applying the compound file can then be used to assess DDI risk and variability within a population. These compound files are aimed for model continuum and to continuously be developed as more data become available for a drug candidate. CYP3A victim DDI risks as well as AUC‐ratio of CYP2D6 PM over EM, simulated by PBPK modeling, were well predicted for the selected compounds and very close to the predictions using static equations.

In conclusion, in vitro as well as in silico tools with a rapid turn‐around time and high accuracy in prediction of variability in drug exposure in patients are essential for risk assessment of NCE in the discovery phase to aid drug design and optimization of this potential liability. We suggest a static equation should be applied to model fm_CYP values estimated in human hepatocytes combined with P450‐selective inhibitors to assess the risk of variable exposures due to victim DDI and enzyme polymorphism in drug discovery. In our study, the concept was successfully demonstrated for the two major P450s, CYP3A and CYP2D6, but is also expected to result in reliable predictions for other P450s (or in principle for other drug metabolizing enzymes as well) provided that selective and potent inhibitors can be identified. The static equation is easy to use, and less input parameters are required compared with a dynamic PBPK model. Therefore, a static model is considered sufficient in the drug discovery phase, and the use of dynamic PBPK modelling should be used as a complementary tool for more developed compounds in late discovery phase and for clinical candidates to support and potentially refine the predictions by the static model and maintain modelling continuum. We believe that the approach described will enhance the likelihood of selecting a candidate drug for clinical development with reduced risk of adverse drug reactions and/or lack of efficacy due to large variations in drug exposure.

Author contributions

All authors substantially contributed to the conception and design of the work; analysis and interpretation of data for the work; drafting the work; revising the work critically for important intellectual content; final approval of the version to be published; and agree to be accountable for all aspects of the work, ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Conflict of interest

The authors declare no conflicts of interest.

Declaration of transparency and scientific rigour

This http://onlinelibrary.wiley.com/doi/10.1111/bph.13405/abstract acknowledges that this paper adheres to the principles for transparent reporting and scientific rigour of preclinical research recommended by funding agencies, publishers and other organisations engaged with supporting research.

Supporting information

Table S1 Median abundance and ISEF values for CYP isoforms.

Table S2 Summary of drug‐specific parameters used for PBPK modelling.

Table S3 Estimated fg and fm values.

Table S4 Summary of clinical data.

Figure S1 Metabolic turnover of midazolam (1A) and bufuralol (1B) in human 59 hepatocytes when incubated in absence (circles) and presence (triangles) of selective 60 CYP3A (ketoconazole) and CYP2D6 (quinidine) inhibitors respectively.

Lindmark, B. , Lundahl, A. , Kanebratt, K. P. , Andersson, T. B. , and Isin, E. M. (2018) Human hepatocytes and cytochrome P450‐selective inhibitors predict variability in human drug exposure more accurately than human recombinant P450s . British Journal of Pharmacology, 175: 2116–2129. doi: 10.1111/bph.14203.