Abstract
Cytochrome P450 3A (CYP3A) is a frequent target for time-dependent inhibition (TDI) that can give rise to drug-drug interactions (DDI). Yet many drugs that exhibit in vitro TDI for CYP3A do not result in DDI. There were 23 drugs with published clinical DDI evaluated for CYP3A TDI in human liver microsomes (HLM) and hepatocytes (HHEP), and these data were used in static and dynamic models for projecting DDI caused by inactivation of CYP3A in both liver and intestine. TDI parameters measured in HHEP, particularly the maximal rate of enzyme inactivation, were generally lower than those measured in HLM. In static models, the use of estimated average unbound organ exit concentrations offered the most accurate projections of DDI with geometric mean fold errors of 2.0 and 1.7 for HLM and HHEP, respectively. Use of maximum organ entry concentrations yielded marked overestimates of DDI. When evaluated in a binary fashion (i.e., projection of DDI of 1.25-fold or greater), data from HLM offered the greatest sensitivity (100%) and specificity (67%) and yielded no missed DDI when average unbound organ exit concentrations were used. In dynamic physiologically based pharmacokinetic modeling, accurate projections of DDI were obtained with geometric mean fold errors of 1.7 and 1.6 for HLM and HHEP, respectively. Sensitivity and specificity were 100% and 67% when using TDI data generated in HLM and Simcyp modeling. Overall, DDI caused by CYP3A-mediated TDI can be reliably projected using dynamic or static models. For static models, average organ unbound exit concentrations should be used as input values otherwise DDI will be markedly overestimated.
SIGNIFICANCE STATEMENT CYP3A time-dependent inhibitors (TDI) are important in the design and development of new drugs. The prevalence of CYP3A TDI is high among newly synthesized drug candidates, and understanding the potential need for running clinical drug-drug interaction (DDI) studies is essential during drug development. Ability to reliably predict DDI caused by CYP3A TDI has been difficult to achieve. We report a thorough evaluation of CYP3A TDI and demonstrate that DDI can be predicted when using appropriate models and input parameters generated in human liver microsomes or hepatocytes.
Introduction
Alterations in the catalytic activities of cytochrome P450 enzymes represent a common mechanism of drug-drug interactions (DDI). Administration of one drug (the precipitant or perpetrator) which inhibits, inactivates, or induces the expression of a P450 enzyme will lead to changes in exposure to a second drug (the object or victim) that is cleared by that enzyme. The scientific and medical literature describe myriad examples of DDI that arise via this mechanism, and a large public database that summarizes the knowledge of DDI has been assembled (https://didb.druginteractionsolutions.org/).
Our understanding of the human P450 enzymes and their substrate and inhibitor specificities, along with the development of methods and algorithms to translate in vitro drug metabolism data to in vivo pharmacokinetics through the use of equations and physiologically based pharmacokinetic (PBPK) modeling platforms, has led to the routine collection of in vitro P450 inhibition data. DDI are an undesired phenomenon in new drugs, and in vitro P450 inhibition data are used in drug design and the selection of candidate compounds for further development as drugs. Such data are also used in developing strategies to evaluate DDI in the clinical phase of drug development, to decide what clinical pharmacokinetic studies need to be conducted in anticipation of observing a DDI (Grimm et al., 2009; Zhang et al., 2010; FDA, 2020).
Among the human P450 enzymes, CYP3A is the most important, since it is involved in the metabolic clearance of many drugs (Guengerich, 1995; Zhou, 2008). Thus, this enzyme is of focus for in vitro evaluation of the potential for DDI. Furthermore, this enzyme is prone to time-dependent inhibition (TDI), wherein the degree of inhibition observed increases with increasing incubation time. Some of the most notorious perpetrators of DDI are TDI of CYP3A, including clarithromycin, verapamil, diltiazem, and mibefradil, among others (Backman et al., 1994; Gorski et al., 1998; Mullins et al., 1998; Jones et al., 1999). In fact, mibefradil was withdrawn from clinical use due to CYP3A TDI (Prueksaritanont et al., 1999). Among CYP3A TDI, most are mechanism-based inactivators (MBI) in that the compound must be acted upon by the enzyme for the inhibition to occur. In P450 incubations, TDI and MBI can be distinguished in that the latter require the inclusion of NADPH, the cosubstrate required for P450 catalysis. MBI of P450 enzymes can occur via several biochemical mechanisms, and the most well known are 1) the formation of a complex between a metabolic intermediate of the inactivator and the heme iron (referred to as an M-I complex); 2) adduction of the inactivator to porphyrin, or 3) adduction of the inactivator to the protein backbone of the enzyme. Irrespective of the biochemical mechanism of MBI, all three will result in permanent cessation of the activity of enzyme molecules and DDI, with DDI abating only upon stopping treatment with the perpetrator drug and resynthesis of the enzyme in vivo (Zhou et al., 2004). Thus, during the drug research and development processes, considerable attention is focused on TDI of CYP3A to avoid this unwanted property.
Observations in our laboratories arising through the routine deployment of CYP3A TDI assays (unpublished observations) and those reported by others (Zimmerlin et al., 2011) have shown the preponderance of CYP3A TDI among test compounds to be high. Yet, for several compounds that have shown CYP3A TDI and which were predicted to demonstrate DDI, the result in clinical pharmacokinetic studies with sensitive CYP3A marker substrates, such as midazolam, showed no DDI (Ring et al., 2005; Wang et al., 2013). A reason for this mismatch is currently unknown. Others have attempted to engineer the conditions of in vitro incubations in an attempt to better represent the in vivo situation by the inclusion of plasma in the in vitro incubations (Mao et al., 2011, 2012). In a previous report, it was demonstrated that several drugs that showed measurable TDI of CYP3A in human liver microsomes (HLM) and hepatocytes (HHEP) do not cause DDI in vivo (Eng et al., 2021). In that work, the focus was on the use of CYP3A TDI data in an early drug research stage when drug design teams are seeking and selecting compounds lacking potential TDI. The conclusion from that study was that a cutoff for the observed inactivation rate constant, kobs (run at a test concentration of 30 µM), is needed under which no compounds will cause in vivo DDI. Even with such a cutoff, several drugs with kobs over that boundary were still not perpetrators of DDI. Such a boundary can be considered oversimplified, as it is known that other factors besides kobs will influence whether a DDI will be observed (e.g., dose and exposure). Nevertheless, this was an important finding and defines a better understanding of how to use in vitro CYP3A TDI data in decision making.
In the present report, the objective was to build upon the observations from Eng et al., 2021 and query the fidelity of CYP3A TDI data in the projection of DDI using measurements of kinact and KI generated from both HLM and HHEP assays, along with several other necessary input values (e.g., various exposure parameters, plasma free fraction, estimated intestinal exposure, among others). Simple equations that relate in vitro TDI data to in vivo DDI (a.k.a. static methods) (Ernest et al., 2005; Obach et al., 2007; Grimm et al., 2009) as well as PBPK modeling (a.k.a. dynamic methods) (Rowland Yeo et al., 2011; Mao et al., 2013) were used, and various input parameters were evaluated. These methods were tested with 23 drugs that have reported clinical pharmacokinetic studies with sensitive CYP3A marker substrates (mostly midazolam). This represents the largest dataset for this purpose generated under the same experimental conditions. The results have shown that in vitro TDI data tend to over-project DDI, but that with application of established input parameters, success can be obtained using either HLM or HHEP.
Materials and Methods
Materials
Research was conducted on human tissue acquired from a vendor that was verified as compliant with internal policies, including institutional review board/independent ethics committee approval. Pooled HLM, consisting of 36 male and 14 female donors, were purchased from Sekisui XenoTech (Kansas City, KS). Cryopreserved pooled HHEP, consisting of 4 male and 6 female donors, pooled mixed-gender human plasma collected with K3EDTA, and male human liver homogenate were purchased from BioIVT (Westbury, NY). Monobasic and dibasic potassium phosphate buffers, magnesium chloride, NADPH, HEPES, Dulbecco’s PBS, and DMSO were purchased from Sigma (St. Louis, MO). M-PERTM buffer was purchased from Thermo Fisher Scientific (Waltham, MA). Midazolam was purchased from USP (Rockville, MD). 1'-Hydroxymidazolam and [2H4]1'-hydroxymidazolam were synthesized at Pfizer (Groton, CT). Williams' E medium was purchased from Gibco (Dublin, Ireland). Commercially obtained chemicals and solvents were of high-performance liquid chromatography or analytical grade. Tested drugs (typical purity >95%) were either synthesized internally at Pfizer (Groton, CT) or purchased from one of the following sources: Sigma-Aldrich (St. Louis, MO), Toronto Research Chemicals (North York, Ontario, Canada), MedChemExpress (Monmouth Junction, NJ), TCI (Portland, OR), or APExBio (Houston, TX).
Identification of Test Drugs
The University of Washington Drug Interaction Database (https://www.druginteractionsolutions.org) was used to compile a list of drugs for which clinical CYP3A interaction studies were conducted (Table 1). Studies in which midazolam was dosed via the oral route were preferred. In a couple of instances, studies containing midazolam dosed via the intravenous route or other CYP3A probe substrates were chosen. The magnitude of DDI (AUCR) was determined based on the ratio of the probe substrate AUC in the presence (area under the concentration versus time curve when coadministered with an inhibitor) and absence (AUC) of the test drug. For drugs where more than one interaction study was published at the same dose and route, the weighted average AUCR based on the number of subjects per study was calculated using eq. 1 below: where the weighted average AUCR (W) is the quotient between the summation of the product of each observed mean AUCR (Xi) and the number of subjects in that study (wi) divided by the total number of subjects in all studies.
Binding to Plasma, Liver Microsomes, and Liver Homogenate
Binding of the test drugs to human plasma, liver microsomes, and liver homogenate were determined based on methods previously described by Di et al., 2017. Briefly, binding experiments were performed by equilibrium dialysis. The matrix was spiked with 1 µM of test drug (donor) and allowed to dialyze against Dulbecco’s PBS (receiver) for a duration of 6 hours with a 12000–14000 molecular weight cutoff membrane in a humidified incubator supplemented with 5% CO2 at 37°C. All incubations were performed in quadruplicate. At equilibrium, matrix and buffer samples were collected and matrix matched before addition of 4 volumes of acetonitrile containing a cocktail of internal standards (50 ng/ml tolbutamide and 5 ng/ml terfenadine). Samples were vortexed and centrifuged for 5 minutes at approximately 2300 x g at room temperature. The supernatant was collected and analyzed directly by LC-MS/MS (method details are in Supplemental Table 2). Determination of the fraction unbound (fu) in human plasma, liver microsomes, and liver homogenate were determined by methods described in (Riccardi et al., 2017).
Human Hepatocyte Unbound Partition Coefficient
The unbound partition coefficient in HHEP for each of the test drugs was determined following the methods described by Riccardi et al., 2018. Briefly, HHEP (0.5 million hepatocytes/ml) suspended in WEM supplemented with l-glutamine and HEPES (50 mM) were incubated with 1 µM of test drug for 2 hours. At the end of the incubation, the HHEP were centrifuged at 40 x g for approximately 5 minutes to pellet the cells. The supernatant was reserved for analysis. The pellet was washed three times with ice-cold PBS, and the final pellet was lysed with M-PERTM buffer. The reserved supernatant and lysate were matrix matched before the addition of 4 volumes of acetonitrile containing a cocktail of internal standards (200 ng/ml diclofenac and 25 ng/ml indomethacin). Samples were vortexed and centrifuged for 5 minutes at approximately 2300 x g at room temperature. The supernatant was mixed with an equal volume of water containing 0.2% formic acid and analyzed directly by LC-MS/MS (Supplemental Table 2). In vitro Kp,uu of the test drugs in HHEP were calculated by methods described by Riccardi et al., 2018. Partitioning into hepatocytes was demonstrated to be fairly rapid, and equilibrium was generally achieved by 10 minutes (Treyer et al., 2018, 2019).
IC50 Determination in Human Liver Microsomes and Human Hepatocytes
HLM and HHEP were used to determine the IC50 of CYP3A enzyme activity. HLM incubations (0.01 mg/ml) were supplemented with MgCl2 (3.3 mM) and NADPH (1.3 mM) in potassium phosphate buffer (100 mM, pH 7.4). HHEP (0.2 million hepatocytes/ml) were suspended in WEM supplemented with l-glutamine and HEPES (50 mM). Drug stocks, at incrementing concentrations, were prepared at 100 times the final incubation concentration (up to 100 µM final) in a mixture of organic and aqueous solvents, typically acetonitrile and water (Supplemental Table 1). Midazolam, the probe substrate, was prepared at 10x the final concentration (2 to 3 µM, corresponding to KM in HLM and HHEP, respectively) in either potassium phosphate buffer or WEM, for HLM or HHEP, respectively. The final total solvent in the incubations was ≤1%. The incubation was initiated with the addition of drug stock immediately followed by the probe substrate. After a 4-minute incubation in HLM or a 10-minute incubation in HHEP, the reaction was terminated by the addition of either two or four volumes of acetonitrile containing internal standard (100 ng/ml [2H4]1'-hydroxymidazolam) for HLM and HHEP incubations, respectively. All reactions were carried out at 37°C, at a final volume of 200 µl, in duplicate. Samples were vortexed and centrifuged for 5 minutes at approximately 2300 x g at room temperature. The supernatant was mixed with an equal volume of water containing 0.2% formic acid and analyzed directly by LC-MS/MS (methods described below). A standard curve (0.250–250 nM) and inhibitor quality control (iQC, 50 nM) of 1'-hydroxymidazolam was prepared in duplicate at the final protein concentration in the assay for either HLM or HHEP. The inhibitor concentration included in the iQC was determined based on the highest concentration tested in the assay. Standards and iQC samples were processed in the same manner as the incubation samples.
Time-Dependent Inhibition in Human Liver Microsomes
TDI of CYP3A was measured in HLM (0.3 mg/ml) supplemented with MgCl2 (3.3 mM) and NADPH (1.3 mM) in potassium phosphate buffer (100 mM, pH 7.4). Drug stock solutions, at incrementing concentrations up to 300 µM (final), were prepared at 100 times the final incubation concentration in a mixture of organic and aqueous solvents, usually acetonitrile and water (Supplemental Table 1). The final total solvent in the primary incubations was ≤1%. The incubation was initiated with the addition of drug stock to the microsomal mixture. At various time points, generally up to 40 minutes, an aliquot of the mixture was transferred to an activity incubation mixture containing midazolam (20.9 µM, 10-fold KM), MgCl2 (3.3 mM), and NADPH (1.3 mM) in potassium phosphate buffer (100 mM, pH 7.4), generally resulting in a 20-fold dilution. After 6 minutes, the activity reaction was terminated by the addition of two volumes of acetonitrile containing internal standard (100 ng/ml [2H4]1'-hydroxymidazolam). All reactions were carried out at 37°C, at a final volume of 200 µl, in duplicate. Samples were vortexed and centrifuged for 5 minutes at approximately 2300 x g at room temperature. The supernatant was mixed with an equal volume of water containing 0.2% formic acid and analyzed directly by LC-MS/MS (methods described below). A standard curve (0.250–250 nM) and inhibitor quality control (iQC, 50 nM) of 1'-hydroxymidazolam was prepared in duplicate at the final protein concentration in the activity assay. The inhibitor concentration included in the iQC was determined based on the highest concentration tested in the assay divided by the fold-dilution of the primary to secondary incubation. Standards and iQC samples were processed in the same manner as the incubation samples. For several drugs, assay conditions were modified due to rapid inactivation or potent inhibition at the initial time point. Modifications to the incubation conditions can be found in Supplemental Table 1.
Time-Dependent Inhibition in Suspension Human Hepatocytes
TDI of CYP3A was measured in HHEP in a manner similar to that described by Chen et al., 2011. Drug stock solutions, at concentrations up to 300 µM (final), were prepared at 10-times the final incubation concentration in a mixture of organic and aqueous solvents, usually acetonitrile and water (Supplemental Table 1). The incubation was initiated with the addition of drug stock to HHEP (0.45 million hepatocytes/ml) suspended in WEM supplemented with l-glutamine and HEPES (50 mM), in a total volume of 50 µl. At various time points (typically up to 120 minutes unless otherwise stated), a 200 µl aliquot of the activity incubation mixture consisting of midazolam (final concentration 30 µM, approximately 10-fold KM) in media was added to the incubation wells, resulting in a 5-fold dilution of the primary incubation. After a 20-minute activity reaction, the incubation was terminated by the addition of two volumes of acetonitrile containing internal standard (100 ng/ml [2H4]1'-hydroxymidazolam). All reactions were carried out at 37°C in a humidified incubator (75% relative humidity, 5% CO2) in duplicate. The final total solvent in the primary incubations was ≤1%. Samples were vortexed and centrifuged for 5 minutes at approximately 2300 x g at room temperature. The supernatant was mixed with an equal volume of water containing 0.2% formic acid and analyzed directly by LC-MS/MS (methods described below). A standard curve (0.58–1000 nM) and inhibitor quality control (iQC, 100 nM) of 1'-hydroxymidazolam was prepared in duplicate at the final protein concentration in the activity assay. The inhibitor concentration included in the iQC was determined based on the highest concentration tested in the assay divided by the fold-dilution of the primary to secondary incubation. Standards and iQC samples were processed in the same manner as the incubation samples. For several drugs, assay conditions were modified due to rapid inactivation or solubility limitations of the test drug. Modifications to the incubation conditions can be found in Supplemental Table 1.
LC-MS/MS Methodology for the Quantitation of 1'-Hydroxymidazolam
LC-MS/MS analysis was conducted on either a Sciex 5500 or 6500 triple quadrupole mass spectrometer (Framingham, MA) fitted with an electrospray ion source operated in positive ion mode using multiple reaction monitoring. An Agilent 1290 binary pump (Santa Clara, CA) with a CTC Leap autosampler (Leap Technology, Carrboro, NC) was programmed to inject 10 µl of sample on a Halo 2.7 µm C18 2.1x30 mm column (Advanced Materials Technology, Wilmington, DE). A binary gradient was employed using 0.1% (v/v) formic acid in water (mobile phase A) and 0.1% (v/v) formic acid in acetonitrile (mobile phase B) at a flow rate of 0.5 ml/min. Mass-to-charge transitions for analytes 1'-hydroxymidazolam and [2H4]1'-hydroxymidazolam were 342.2 → 324.2 and 346.2 → 328.2, respectively. Analytes were quantified against a standard curve using Analyst software (Sciex). A linear regression with either a weighting of 1/x or 1/x2 was used. Standards and iQCs were accepted if the calculated concentrations were ±25% of their nominal concentration. Acceptance of the iQC demonstrates that the inhibitor did not interfere with 1'-hydroxymidazolam signal on the MS.
Data Analysis
Estimation of Ki.
Percent activity remaining of the CYP3A enzyme was determined by normalizing the concentration of 1'-hydroxymidazolam in the presence of varying concentrations of test drug to the concentration of 1'-hydroxymidazolam in the solvent control. The concentration of inhibitor corresponding to a 50% decrease in activity (IC50) was generated using GraphPad Prism 8 (La Jolla, CA). Since inhibition experiments were conducted at the KM of midazolam, Ki was estimated as IC50 divided by two (IC50/2) assuming competitive inhibition (Cheng and Prusoff, 1973). Free Ki was determined by correcting Ki with the free fraction determined in microsomes (fu,mic) or Kp,uu for HLM or HHEP experiments, respectively.
Estimation of KI and kinact.
Data analysis methods previously described by Yates et al., 2012 were used for the estimation of kobs, KI, and kinact. Briefly, kobs was determined by normalizing the 1'-hydroxymidazolam concentration in each sample to that of the mean solvent control concentration in the initial time point, plotting the natural log of percent remaining activity versus preincubation time, and then calculating the slope of the line (−kobs) using the initial linear portion of the curve. A statistical test was done at each concentration of test drug to determine whether kobs was statistically different from the within-experiment solvent control, i.e., a parallel lines test, shown in eq. 2.
Here, kobs[I], kobs[0µM], and S.E. represent the inactivation rate for an inhibitor at a single concentration, inactivation rate with solvent control, and standard error, respectively. A statistically significant TDI is defined when the parallel lines test yields a P value of <0.05. When possible, KI and kinact parameters were determined using nonlinear regression of the three-parameter Michaelis-Menten equation below:
[I] represents the concentrations of the test drug in the primary incubation, kinact is the maximal inactivation rate, and KI is the inactivator concentration at half kinact.
For compounds where substrate inhibition was observed at the higher concentrations, nonlinear regression of a four-parameter substrate inhibition model (eq. 4) was used to fit the curve.
In the equation above, Ks is a dissociation constant for binding, enabling a better fit of the data.
In six instances, the relationship between kobs and [I] did not yield enough of a hyperbola to reliably estimate values for KI and kinact. For these, a composite slope of the initial linear portion of the kobs vs [I] curve was determined to represent the ratio of kinact/KI. The static model equations and Simcyp require individual parameters as input values. To accomplish this, KI,u was arbitrarily set at a high value of 1 mM, and the slope was used to calculate kinact. This value for kinact along with the value of 1 mM for KI,u were then used as input values for DDI projections, under the reasonable assumption that [I]in vivo ≪ 1 mM. For compounds where KI could be determined, KI,u was determined by correcting KI with fu,mic or Kp,uu for HLM or HHEP experiments, respectively. Analyses were performed using Microsoft Excel (Redmond, WA) and GraphPad Prism 8.
Predicting Magnitude of DDIs.
The magnitude of DDI (AUCR) can be described as a ratio of the area under the concentration versus time curve when coadministered with an inhibitor divided by AUC when administered alone. Mathematical models to determine the extent of DDI while incorporating competitive inhibition and time-dependent inactivation in the liver and gut have been extensively described (Rowland and Matin, 1973; Mayhew et al., 2000; Wang et al., 2004; Obach et al., 2006, 2007; Fahmi et al., 2008) and are summarized in the equation below (eq. 5).
In the above equation, kdeg represents the hepatic (CYP3Ah, 0.00032 minute−1) and intestinal (CYP3Ag, 0.00050 minute−1) degradation rates of the CYP3A enzyme (Obach et al., 2007; Rowland Yeo et al., 2011). The intestinal kdeg is based on a half-life of 23.1 hours, which is the natural turnover rate of enterocytes in vivo (Greenblatt et al., 2003). The hepatic kdeg is based on a half-life of 36 hours, which was derived from a clinical DDI study conducted by Fromm et al., 1996 (Obach et al., 2007). The fraction of the victim drug metabolized by the CYP3A enzyme is represented as fm(CYP3A). In this analysis, fm(CYP3A) for the victim drugs midazolam, buspirone, terfenadine, triazolam, and simvastatin are 0.93, 0.94, 0.74, 0.92, and 0.92, respectively (Obach et al., 2006; Yadav et al., 2018). The fraction of the victim drug escaping intestinal metabolism is represented by Fg. For the victim drugs midazolam, buspirone, terfenadine, triazolam, and simvastatin, Fg are 0.57, 0.21, 0.40, 0.75, and 0.66, respectively (Paine et al., 1996; Brown et al., 2005; Galetin et al., 2006, 2008, 2010; Gertz et al., 2008; Shou et al., 2008). The terms [I]h and [I]g represent inhibiting concentrations of the test drug in the liver and intestine, respectively. Unbound test drug concentrations (Cu) resulting in enzyme inhibition in the liver were estimated as unbound steady state maximum hepatic inlet concentration (Cmax,hepatic inlet,u) using the equation below (eq. 6) (Kanamitsu et al., 2000) or unbound steady state maximum (Cmax,systemic,u) or average (Cavg,systemic,u) systemic concentrations.
In the equation above, ka is the oral absorption rate of the test drug (or 0.1 minute−1), Fa is the fraction of test drug absorbed after oral administration (assume 1), Fg is the fraction of test drug escaping intestinal metabolism (assume 1), fu,p is the free fraction of the test drug in plasma, BPR is the blood-to-plasma ratio (assume 1), and Qh is the liver blood flow [1617 ml/min (Yang et al., 2007a)].
Test drug concentrations resulting in enzyme inhibition in the intestine were estimated as total or free inhibiting concentrations in the enterocyte (Ig) as defined in eq. 7a by (Rostami-Hodjegan and Tucker, 2004), unbound steady state maximum portal vein concentration (Cmax,portal,u, eq. 7b) or average portal vein concentration (Cavg,portal,u, eq. 7c).
The parameters are the same as described above, with Qent representing the intestinal blood flow [300 ml/min (Yang et al., 2007b)], Qpv represents the portal vein blood flow (1213 ml/min), which is approximated as 75% of hepatic blood flow, and τ is the dosing interval for the inhibitor. Free enterocyte inhibiting concentrations are calculated as [I]g corrected for fu,plasma. Compound specific inputs can be found in Supplemental Table 4.
Simcyp Modeling.
Simcyp version 19 release 1 (19.0; Certara, Princeton, NJ) was used to simulate the time courses of victim and perpetrator concentrations in plasma. Simulations were conducted using a design of 10 trials with 10 subjects using the age range of 20–50 years and 1:1 male to female ratio. Simulations were performed in a virtual population library of healthy volunteers supplied by Simcyp (Sim-Healthy Volunteers). The hepatic and intestinal CYP3A kdeg were 0.0193 hour−1 (t1/2 = 36 hours) and 0.03 hour−1 (t1/2 = 23.1 hours), respectively, consistent with those used in static modeling. The intestinal concentrations were the enterocyte exit concentration (portal vein) with fu,gut = fu,plasma, whereas liver concentrations were the liver exit concentration. A summary of the trial designs for all simulations are listed in Table 1. In the DDI studies, the fold-increase in AUCR (e.g., AUCinfinity ratios in the single-dose studies and AUCτ ratios in the multiple-dose studies) was calculated from the ratios of the simulated values in treatment groups relative to control groups. Geometric means of pharmacokinetic parameters generated from Simcyp simulations were compared with the clinically observed geometric mean parameters. For perpetrator drugs clarithromycin, diltiazem, erythromycin, paroxetine, simvastatin, and verapamil, compound files were qualified by Simcyp. Azithromycin Simcyp file was developed by Certara using Simcyp V14 and published in Simcyp Global Health Repository. To maintain consistency with the use of input parameters generated internally (i.e., fu,p, KI, kinact, etc.), the steady-state volume of distribution (Vss) and oral clearance were adjusted to adequately recover the clinically observed AUC and Cmax at the dose used in the DDI study. In predicting the inhibitory effect of diltiazem administration, we have incorporated the inhibitory effects of its primary metabolite (MA). Sim-desmethyldiltiazem (MA) was qualified by Simcyp and used without adjusting Vss and CL. Input parameters of 23 perpetrators are summarized in Supplemental Tables 5–27. Briefly, physicochemical properties (pKa, log P) are obtained from in silico software (www.acdlabs.com; www.biobyte.com). Plasma binding (fu,p) was obtained from internal experimental data (Supplemental Table 3). The apparent permeability values used were either default values included as part of the qualified Simcyp files (https://members.simcyp.com/account/libraryFiles) or calculated from in silico models (Keefer et al., 2013) and extrapolated to effective permeability in human. ka, Vss, and oral clearance were fitted manually to adequately recover the clinically observed results at the doses used in the DDI studies (Supplemental Table 38). These values were internally consistent with those used in static projections (model 4).
For victim drugs, Sim-midazolam, SV-triazolam, and SV-simvastatin, Simcyp library compound files were qualified by Simcyp and used without modification. The terfenadine compound file was developed by Simcyp for QT prolongation prediction and was adjusted to reasonably recover the clinical fm(CYP3A) and Fg. The buspirone Simcyp model was developed based on physicochemical properties, plasma binding, distribution, and elimination obtained from literature (Kivisto et al., 1997, 1999; Mahmood and Sahajwalla, 1999). The buspirone model could reasonably recover both PK and fm(CYP3A). Using this model, the predicted AUCR after coadministration of itraconazole are in reasonable agreement with the observed values. Input parameters for the victim drugs are summarized in Supplemental Tables 28-33.
Confusion Matrix Analyses.
To determine the success of predicting DDI, contingency tables were analyzed separately for HLM and HHEP results. Cutoff values of AUCR were defined prospectively as ≥1.25-fold (bioequivalence) or ≥2-fold. As an example, in the pairing of clinical and predicted AUCR ≥1.25-fold, true positives (TP) were defined as drugs which resulted in an observed clinical AUCR ≥1.25-fold and were predicted to have an AUCR ≥1.25-fold. True negative (TN) drugs were those that resulted in an observed clinical AUCR <1.25 and were predicted to have an AUCR <1.25-fold. False positive (FP) drugs were defined as observed clinical AUCR <1.25-fold, but were predicted to have an AUCR ≥1.25-fold, whereas false negative (FN) drugs were those where an observed clinical AUCR was ≥1.25-fold, but the predicted AUCR was <1.25-fold.
Additional assessments of probability to describe model success included sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive predictive error (PPE), and negative predictive error (NPE). The sensitivity value describes the ability of the model to correctly identify observed positive DDIs (eq. 8), whereas the specificity value describes the ability of the model to correctly identify observed negative DDIs (eq. 9). The PPV is the portion of studies that were predicted to have an interaction in the clinic and a clinical interaction was observed (eq. 10). Conversely, the NPV represents the portion of studies that were predicted to not have an interaction in the clinic and a clinical interaction was not observed (eq. 11). The PPE describes the portion of studies that were predicted to have an interaction in the clinic, but a clinical interaction was not observed (eq. 12). Lastly, NPE describes the portion of studies that were predicted to not have an interaction in the clinic, but a clinical interaction was observed (eq. 13).
Accuracy of Predictions.
Accuracy of the various prediction models was assessed by the average fold error or bias (eq. 14). Precision of the predictions were evaluated using geometric mean absolute fold error (GMFE, eq. 15) and root mean square fold error (RMSFE, eq. 16). where N is the total number of predictions.
Results
CYP3A TDI in Human Liver Microsomes and Hepatocytes.
Among the 24 compounds evaluated (23 drugs and one metabolite), 23 and 20 demonstrated measurable inactivation kinetic parameters in HLM and HHEP, respectively. Values for KI ranged from 0.360 to 202 µM in HLM and 0.574 to 51.2 µM in HHEP. These were corrected for unbound fraction prior to utilizing them in projection of DDI (see below). Since hepatocytes are intact cells that possess a membrane barrier, availability of the inhibitor to the P450 enzyme to cause inactivation may be limited, and this barrier is not present in HLM. By estimating the intracellular free drug concentration, inhibitor availability to the active site can be determined using Kp,uu. Therefore, application of Kp,uu to HHEP KI was deemed to be a reasonable estimate of HHEP KI,u. For Kinact, values ranged from 0.00954 to 0.329 minute−1 in HLM and 0.00446 to 0.0978 minute−1 in HHEP. Values are listed in Tables 2 and 3, and the percent of control activity versus incubation time and kobs versus [I] plots for each drug are shown in the Supplemental Figures. On average, (i.e., average fold error) the KI,u, kinact, and kinact/KI,u values in HLM were 1.2-, 3.5-, and 3.2-fold greater than in HHEP (Fig. 1).
Projection of DDI from CYP3A TDI Data in Liver Microsomes and Hepatocytes: Static Method.
The equations described above were used to make projections of DDI, based on AUCR. Although the in vitro kinetic parameters are measured experimentally, several input values needed for these projections cannot be measured experimentally and must be estimated (e.g., Fa, ka, etc.). Furthermore, the correct in vivo concentrations of the inactivator in the liver and intestine are needed, and whether the best values to use are represented by maximum or average organ inlet or egress values is unknown. Based on the input parameters mentioned above, over 50 combinations of these values could be evaluated. These methods were narrowed down to three (referred to as Methods 2–4), selected because they represented internally consistent and physiologically meaningful combinations. That is, we did not further pursue random combinations that would make no sense or be illogically contradictory. These were also compared with the input parameters currently recommended in the United States Food and Drug Administration regulatory document on drug-drug interactions [referred to as Method 1; (FDA, 2020)].
When comparing the approaches, the input parameters of CYP3A degradation rates and organ flow rates were kept constant, as these are reasonably well established, and it was not an objective to test the validity of these values. Also, for many of the drugs used in the evaluation, their overall oral absorption is not known, and thus, a value of unity was used. This could be an overestimate for some of the drugs. The greatest source of variation in prediction methods lies in the selection of the most relevant in vivo concentrations of the perpetrator drug, in both liver and intestine. In these approaches, unbound concentrations were used, and the concentrations entering the organ (liver and intestine) were compared with concentrations exiting the organ. The concentrations entering the liver were considered as the hepatic inlet estimates (Cmax,hepatic inlet,u), whereas the concentrations entering the intestine were estimated using the absorption rate constant and oral doses (eq. 7a). The concentrations exiting the liver and intestine were considered as the systemic concentrations (Cmax,systemic,u or Cavg,systemic,u) and estimated portal vein concentrations (Cmax,portal,u, eq. 7b or Cavg,portal,u, eq. 7c), respectively. Calculated concentrations for [I]g and [I]h are listed in Supplemental Table 34.
A summary of performance characteristics of DDI prediction from TDI data are presented in Table 4. When using TDI data gathered in HLM in making estimates of DDI, the best input values for concentrations for the liver were the average organ exit values (i.e., Cavg,u), and the best input values for concentrations for the intestine were also the exit values described by the estimates of portal vein concentrations (Models 3 and 4). Model performance was somewhat better when employing estimated Cavg,u values than unbound maximum concentration values. Model 4 had a GMFE of 2.0, indicating that, on average, projections of DDI from in vitro data were within about 2-fold of the actual values, in this case the 2-fold error favored an overprediction of DDI (Fig. 2A). Using organ entrance values for perpetrator concentrations (Models 1 and 2) yielded poorer performance and marked overprojections of DDI. The consideration of using the estimated free concentration in the intestine (Models 2–4) is an important one. Model 1, as recommended in regulatory guidance, uses a very rapid rate of absorption (proximate to the rate of gastric emptying) and essentially states that the concentration of the perpetrator drug in the enterocyte is equal to the concentration of the total dose dissolved in the luminal fluids. This yields poor performance and consistent over-projection of DDI from TDI data. Consideration of the free concentration in the enterocytes, along with more realistic absorption rates (Models 2–4), yields a considerable improvement in DDI projection over Model 1. Individual estimates are listed in Supplemental Table 35.
A similar trend among the use of various input values for in vivo perpetrator drug concentrations was observed when using TDI data from HHEP (Table 4). Again, Models 3 and 4, which use unbound exit concentrations for liver and intestine, yielded the most reliable projections of DDI (Fig. 2B, Model 4). Furthermore, use of HHEP TDI data offered nominal performance improvements in projections relative to HLM TDI data, with Model 4 yielding a GMFE of 1.7, as compared with 2.0 for HLM data. Overall, projections of DDI were still higher than those observed in vivo, especially if organ entrance values for perpetrator concentrations were employed (Models 1 and 2). Individual estimates are listed in Supplemental Table 36.
Evaluation of AUCR projections relative to the extent of observed clinical AUCR revealed no clear relationship with overpredictions above and below the 2-fold criteria (Fig. 3). Of particular interest is significant overpredictions with observed AUCR below 2-fold for several drugs (e.g., tadalafil, midostaurin) irrespective of whether the in vitro data generated in HLM or HHEP. Further mechanistic evaluation may be required to clarify contributing factors. Drugs with underprediction of DDI using data generated in HHEP were diltiazem (Simcyp only), nelfinavir, and conivaptan (static and dynamic modeling).
In addition to evaluating these approaches for numerical accuracy of projection of DDI, it is also useful to evaluate them using categorical criteria and cutoff values for what is defined as a DDI and what is not (i.e., a confusion matrix analysis). This analysis was conducted using two criteria for DDI: 1.25-fold, as defined by bioequivalence boundaries, and 2-fold. The latter is offered as a reasonable value for DDI that may actually be meaningful for clinical outcome. This will always be dependent on the therapeutic window for the affected drug, as some drugs are very safe, and marked increases in exposure can still be well tolerated, but others could be dosed at levels such that smaller increases in exposure can yield meaningful side effects. Nevertheless, very few drugs go from safe to toxic with a mere 2-fold increase in exposure, and thus, a 2-fold DDI cutoff criteria was deemed a useful one by which to evaluate these DDI projection methods.
TDI experiments performed in HLM offer very high sensitivity, however, specificity can be low, depending on the model employed. When using the data in this binary fashion, the ten drugs (Fig. 4) that cause a greater than 2-fold DDI are readily identified irrespective of the model used yielding 100% sensitivity (Table 5). However, this comes at a cost of a high rate of false positives especially when using Models 1 and 2 (organ entrance concentrations as input for in vivo [I]), as 13 of 15 drugs that cause DDI below 2-fold are actually identified as causing greater than 2-fold DDI. The PPE, i.e., the portion of DDI studies that are identified as needing to be conducted but actually are negative and would not need to be conducted, is high at 57%. This attribute improves to 33%–50% if Models 3 or 4 are employed in that a greater number of the negative drugs are properly identified as such and this does not come at the expense of any false negative drugs: all positive drugs are identified. Overall, the performance of the models using HLM data are similar when using either 1.25- or 2-fold as the defining cutoff for DDI. When using TDI data generated in HHEP, the PPE improves to a minor extent, decreasing to as low as 25%. However, this comes at a cost of 1–3 instances of false negative outcomes. Overall, the combination that yields the lowest PPE while not giving any false negative is the use of TDI data from HLM entered into Model 4, which uses the average organ exit concentrations for in vivo [I]. This observation is independent of the DDI cutoff applied (i.e., 1.25- or 2-fold).
Projection of DDI from CYP3A TDI Data in Liver Microsomes and Hepatocytes: Dynamic PBPK Method.
These in vitro data were also used as input values for projecting DDI using Simcyp (individual estimates are listed in Supplement Table 37). In this case, various input parameters such as in vivo [I] values, absorption extent and rate values, and enzyme degradation rates are embedded in the algorithms. The output DDI projections were evaluated as above for both numerical accuracy of DDI projections (Table 6) as well as categorical assignment of the potential for DDI using 1.25- and 2-fold boundaries (Table 7). Use of Simcyp yielded overall accuracy for projections of DDI from in vitro data with GMFE of 1.7 and 1.6 if data were from HLM or HHEP, respectively (Fig. 2, C and D). These performance characteristics are slightly better than the best of the static models (Model 4). When evaluated in a confusion matrix, Simcyp modeling performed very well with TDI data from HLM, with no instances of false negatives (i.e., 100% sensitivity) and with high specificity (PPE of only 24% and 23% when using 1.25-fold and 2-fold cutoff values for defining a DDI (Table 7). However, with TDI data obtained from HHEP, the performance of Simcyp suffered slightly with occurrences of false negatives that decreased sensitivity to about 70%–85% depending on cutoff criteria (Fig. 5).
Discussion
The projection of clinical DDI from in vitro TDI data is challenging. Considerable progress has been made, yet it is still commonplace to project DDI from in vitro data and then conduct a clinical study based on that projection, only to observe no DDI. In drug development, such an occurrence (i.e., a false positive from the in vitro data), while demonstrating that it will be acceptable to coadminister the two agents based on the clinical observation, expends effort on a clinical DDI study that really did not need to be conducted. Efforts and resources would be better expended on other clinical studies. Our exploration of different model input values, especially the application of projected organ egress concentrations instead of entry concentrations, improves the overall fidelity of predictions of DDI while still avoiding the occurrence of false negative outputs. Use of PBPK modeling (Simcyp) yielded excellent performance with data from HLM, with no false negative outputs. It is false negative outputs (i.e., projection of no DDI from in vitro data but actual DDI in the clinic) which must be avoided as this would result in a deleterious impact on patient safety.
It should be noted that in vitro TDI assays, using HLM or HHEP, are highly sensitive. If one considers the lower limit of statistically detectable kobs values in vitro [0.002 minute−1 and 0.0015 minute−1 for HLM and HHEP, respectively; (Eng et al., 2021)] as compared with the estimate for the natural turnover rate for CYP3A in vivo (0.00032 minute−1), then any drug demonstrating measurable TDI would be expected to cause at least a 12-fold decline in CYP3A activity at the inhibitor concentration that yielded the lower limit value of kobs. That is, the rate of inactivation would exceed the natural rate of enzyme resynthesis by 12-fold. It should also be noted that varying values of kdeg,h for CYP3A4 have been reported in the literature (Yang et al., 2008), and that the accuracy of projections of DDI from in vitro data will be dependent on the kdeg,h value used. A value of 0.00032 minute−1 was selected for our static models to be consistent with the value embedded in the Simcyp algorithm.
The present efforts focused on three main objectives: 1) a comparison of in vitro TDI data generated in HLM and HHEP; 2) evaluation of different input values for estimates of in vivo concentrations of the inhibitor in static models for DDI prediction and identification of the best values; and 3) evaluation of the dynamic PBPK model Simcyp for prediction of DDI from TDI data. The in vitro dataset generated for these objectives represents the largest one reported using a fixed source of in vitro reagents and consistent methods run in one laboratory, paired with clinical DDI data gathered from the literature. Previously reported efforts had many fewer examples included (Mayhew et al., 2000; Obach et al., 2007; Mao et al., 2011; Kenny et al., 2012). CYP3A was the enzyme of focus, since the greatest number of TDI and DDI is known for this enzyme. Other P450 enzymes such as CYP2D6 or CYP1A2 have few examples of drugs that are TDI with corresponding clinical data for comparison and as such do not offer a large enough set of drugs from which conclusions could be drawn. It is proposed that the conclusions generated for CYP3A could be applicable to other enzymes. Also, compared with other P450s, the magnitude of CYP3A-based DDIs can be high for orally administered drugs because of the potential for inhibition of this enzyme not only in the liver but also in the small intestine during the absorption of the drugs after oral administration.
The data demonstrated that DDI are predictable from in vitro TDI data gathered in either in vitro system. Inactivation kinetic values in HLM were generally greater than in HHEP. In both cases, it was important to correct Ki and KI values generated from nominal concentrations of the inhibitor added to the incubations by binding and partition factors. Inhibitor concentrations available to inactivate CYP3A may actually be lower than those added to the assay due to nonspecific binding (HLM and HHEP) and unequal concentrations of unbound drug inside and outside the cell (i.e., Kp,uu in hepatocyte incubations). Most important is the selection of the most appropriate value for in vivo concentration of the inhibitor drug. Previous recommendations have advocated that the total concentration of drug in the lumen should be used for projection of DDI occurring in the intestine and unbound maximum concentrations estimated for the portal vein during absorption be used for projection of DDI occurring in the liver (Ito et al., 1998; Rostami-Hodjegan and Tucker, 2004; FDA, 2020). These are represented in Model 1, and lead to such marked over-projection of DDI (mean GMFE of 4–6, and high false positive rates) that the value of conducting in vitro TDI assays is essentially negated. The assay would be futile, as just about every drug tested will suggest the likelihood of DDI and the need for a clinical DDI study. However, moving to potentially more realistic values for in vivo [I] in intestine (enterocytes) and in liver greatly improves the accuracy of DDI projection. In Models 3 and 4, estimated unbound organ exit concentrations are used (i.e., free portal concentrations for intestine and free systemic concentrations for liver), and these may better reflect the concentrations of inhibitor available to affect the enzyme in these tissues. Use of Cavg,u values were shown to be best. Mean GMFE values of 2.0 and 1.7 were calculated when using HLM and HHEP data, respectively (Table 4). Simcyp modeling, in which the relevant in vivo concentration values are embedded in the software, yielded mean GMFE values of 1.7 and 1.6 for HLM and HHEP, respectively (Table 6). Furthermore, a high degree of correlation (r2 > 0.9) was observed when predicted AUCRs from HLM and HHEP using Model 4 and Simcyp were compared (Fig. 6). These performance characteristics for both static and Simcyp modeling demonstrate that DDI can be projected from in vitro TDI data provided that appropriate input values are entered. Additionally, the use of HHEP generated TDI values with Simcyp modeling would provide an estimated AUCR with the highest degree of numerical accuracy compared with clinical observations.
A second manner in which in vitro TDI data may be used is in a binary decision making fashion. Thus, instead of seeking numerical accuracy in projecting DDI, the data are used to determine whether a clinical DDI study is needed or not. This depends on the cutoff used to define a DDI, and in this analysis, the data were evaluated using two cutoff criteria: 2-fold representing a minimum of what may cause a clinically meaningful DDI for many drugs and 1.25-fold representing a conservative bioequivalence boundary. Utilizing the data in this manner strives for minimizing false positives (i.e., running a clinical DDI study which shows no DDI) while not allowing for any false negatives (i.e., not running a clinical DDI study when one was actually needed). As with the aforementioned numerical accuracy findings, evaluating the data in this manner yielded similar conclusions with regard to the most appropriate input values used for in vivo [I]. If using organ entry concentrations, over-projection of the frequency of actual DDI was high, rendering the use of in vitro TDI essentially uninformative for such decision making (Table 5). This observation was similar irrespective of the cutoff values used (1.25- or 2-fold). However, when organ exit concentrations were employed, the sensitivity remained high, and the selectivity increased (fewer false positives). Method 4 with HLM data also yielded a mean GMFE of 2-fold. Thus, the overall conclusion for static models is that the use of TDI data from HLM along with Cavg,u values for [I] (Model 4) is the recommendation for an optimal strategy when employing in vitro data for clinical study planning for CYP3A based DDI caused by TDI.
Similar observations were made from the data entered into Simcyp modeling (Tables 6 and 7). Simcyp modeling offers a more sophisticated approach than simple equations that assume a single static in vivo concentration of inhibitor and substrate when estimating DDI. In Simcyp, concentrations are correctly assumed to be changing over time, and this is accounted for when projecting the extent of CYP3A inactivation. Model performance characteristics for Simcyp were even better than Model 4. When performance was evaluated using the categorical criteria, Simcyp yielded no false negative outcomes for HLM TDI data, however, when employing TDI data from HHEP, there were instances of false negative outcomes (Table 7). Thus, the overall conclusion for Simcyp modeling for CYP3A DDI caused by TDI is that the best in vitro input data are from HLM.
Reasons behind why there are differences between TDI parameters generated in HHEP versus HLM, particularly kinact, have not been determined. Corrections for fu,mic and Kp,uu were made before making comparisons, removing these as possible factors. Various reasons can be speculated, such as the presence of conjugating enzyme activities present in HHEP but absent in HLM that could remove inactivating metabolites, or the existence of some unknown enzyme protection mechanism in HHEP that is absent in HLM. The catalytic cycle of P450 activity is complex and possesses steps wherein the process can be uncoupled, resulting in generation of reactive oxygen species without catalyzing metabolism of the substrate, and alteration of this cycle in one system versus the other could yield differences in enzyme inactivation. Furthermore, it has been proposed that there are multiple phases of inactivation in TDI experiments and that a more sophisticated method of deriving in vitro values from these experiments is needed as compared with the manner in which the present data were processed (Yadav et al., 2018, 2020). Preliminary data has been generated that shows that not only are kinact values different in the two systems, but that partition coefficients also differ (data on file). The partition coefficient is a measure of the ratio of the molecules of substrate consumed per molecule of enzyme inactivated and is an inherent biochemical property of an enzyme-inactivator pair. Thus, there appears to be a difference in the manner in which CYP3A behaves in the intact hepatocyte system versus microsome incubations. Further efforts to understand these differences at a fundamental biochemical level are underway and will be reported in due course. The drugs assessed here have not been demonstrated to be clinical inducers. However, this does not preclude them from possessing in vitro induction potential. Although the current analysis excludes in vitro induction parameters, work is ongoing to determine how in vitro induction may play a role in aligning compounds for which over predictions were observed.
Lastly, from the simplest basic and static models to more sophisticated and complex PBPK models, there is always a level of uncertainty that arises from the assumptions required for various in vivo input parameters (e.g., kdeg, fm, Fg, [I]in vivo, Kp,uu in vivo) and conceptual simplifications needed for modeling. Regarding the latter, it is necessary to simplify that the target organs, liver and intestine, are homogeneous compartments as opposed to the complex multicompartment tissues that they truly are. Furthermore, subcellular distribution of the inactivators and substrates is likely heterogeneous, which cannot be currently measured. Absorption is a more complex kinetic process than what has been addressed in the present treatment, and further work is planned to evaluate the impact of more complex and multiphasic absorption models driving in vivo input concentrations (e.g., the Advanced Dissolution, Absorption, Metabolism (ADAM) model in Simcyp). Finally, it must be appreciated that there is uncertainty and variability in important input parameters like kdeg for CYP3A4 and Fg for midazolam and other substrates, among others, as well as the potential impact of the role for CYP3A5 in the metabolism of the substrates in vivo. As the science and knowledge around these factors continue to develop, re-evaluation and refinement of DDI prediction methods will be necessary. With the present state of knowledge, the approaches described here offer the best performance for predicting DDI for CYP3A time-dependent inactivators that can be currently achieved.
In conclusion, it has been demonstrated that TDI data for CYP3A generated in either HLM or HHEP can be used effectively to predict the magnitude of DDI and to make decisions regarding the need for clinical DDI studies. Data from HLM appears to perform better overall regarding not missing any DDI, whether using static or dynamic modeling, however, use of these data tended to yield more over-projection of DDI as compared with TDI data generated using HHEP. Key to the success in this endeavor is the selection of the most appropriate value for the in vivo concentration of the inactivator at the most important sites of action, the liver and intestine. Projected organ exit concentrations were best, and the use of maximum or average values can work depending on the reagent used to generate the in vitro TDI data. PBPK modeling using the Simcyp module, wherein concentrations of the substrate and inhibitor are changing over time, performed well in predicting DDI.
Acknowledgments
We would like to thank Samantha Jordan and the scientists at York Bioanalytical Solutions for their contributions toward the generation of the binding data.
Authorship Contributions
Participated in research design: Tseng, Eng, Lin, Cerny, Goosen, Obach.
Conducted experiments: Tseng, Eng.
Performed data analysis: Tseng, Eng, Lin, Cerny, Tess, Goosen, Obach.
Wrote or contributed to the writing of the manuscript: Tseng, Eng, Lin, Cerny, Goosen, Obach.
Footnotes
- Received April 8, 2021.
- Accepted July 1, 2021.
This work received no external funding.
The authors are employees of Pfizer and report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.
↵This article has supplemental material available at dmd.aspetjournals.org.
Abbreviations
- AUC
- area under the concentration versus time curve
- AUCR
- area under the plasma concentration-time curve ratio in the inhibited and control state
- Cavg
- average concentration values
- Cavg,portal,u
- unbound average portal vein concentration
- Cavg,systemic,u
- unbound average systemic concentration
- Cavg,u
- unbound average concentration values
- Cmax,portal,u
- unbound steady state maximum portal vein concentration
- CYP3A
- cytochrome P450 3A
- DDI
- drug-drug interaction
- Fa
- fraction absorbed after oral dose
- FN
- false negative
- FP
- false positive
- fu
- fraction unbound
- fu,mic
- free fraction determined in microsomes
- fu,plasma
- plasma free fraction
- GMFE
- geometric mean absolute fold error
- HHEP
- human hepatocyte
- HLM
- human liver microsome
- [I]
- inhibitor concentration
- [I]g
- intestinal inhibitor concentration
- [I]h
- liver inhibitor concentration
- iQC
- inhibitor quality control
- ka
- inhibitor absorption rate constant
- kdeg
- enzyme degradation rate constant
- Ki
- reversible inhibition constant
- KI
- time-dependent inhibition constant
- KI,u
- time-dependent inhibition constant, unbound
- kinact
- maximal rate of enzyme inactivation
- kobs
- rate constant for inhibition
- Kp,uu
- unbound partition coefficient in hepatocyte
- LC-MS/MS
- liquid chromatography-tandem mass spectrometry
- MBI
- mechanism-based inactivator
- NPE
- negative predictive error
- NPV
- negative predictive value
- P450
- cytochrome P450
- PBPK
- physiologically based pharmacokinetic modeling
- PPE
- positive predictive error
- PPV
- positive predictive value
- RMSFE
- root mean square fold error
- TDI
- time-dependent inhibition\inhibitor
- TN
- true negative
- TP
- true positive
- Vss
- steady-state volume of distribution
- WEM
- Williams’ E medium
- Copyright © 2021 by The American Society for Pharmacology and Experimental Therapeutics