Clinical Pharmacokinetics

, Volume 46, Issue 8, pp 681–696

General Framework for the Quantitative Prediction of CYP3A4-Mediated Oral Drug Interactions Based on the AUC Increase by Coadministration of Standard Drugs

Authors

  • Yoshiyuki Ohno
    • Department of Pharmacy, University of Tokyo Hospital Faculty of MedicineUniversity of Tokyo
    • Department of Pharmacy, University of Tokyo Hospital Faculty of MedicineUniversity of Tokyo
  • Hiroshi Suzuki
    • Department of Pharmacy, University of Tokyo Hospital Faculty of MedicineUniversity of Tokyo
    • Center for Advanced Medical Engineering and InformaticsOsaka University
Original Research Article

DOI: 10.2165/00003088-200746080-00005

Cite this article as:
Ohno, Y., Hisaka, A. & Suzuki, H. Clin Pharmacokinet (2007) 46: 681. doi:10.2165/00003088-200746080-00005

Abstract

Background

Cytochrome P450 (CYP) 3A4 is the most prevalent metabolising enzyme in the human liver and is also a target for various drug interactions of significant clinical concern. Even though there are numerous reports regarding drug interactions involving CYP3A4, it is far from easy to estimate all potential interactions, since too many drugs are metabolised by CYP3A4. For this reason, a comprehensive framework for the prediction of CYP3A4-mediated drug interactions would be of considerable clinical importance.

Objective

The objective of this study was to provide a robust and practical method for the prediction of drug interactions mediated by CYP3A4 using minimal in vivo information from drug-interaction studies, which are often carried out early in the course of drug development.

Data sources

The analysis was based on 113 drug-interaction studies reported in 78 published articles over the period 1983–2006. The articles were used if they contained sufficient information about drug interactions. Information on drug names, doses and the magnitude of the increase in the area under the concentration-time curve (AUC) were collected.

Methods

The ratio of the contribution of CYP3A4 to oral clearance (CRCYP3A4) was calculated for 14 substrates (midazolam, alprazolam, buspirone, cerivastatin, atorvastatin, ciclosporin, felodipine, lovastatin, nifedipine, nisoldipine, simvastatin, triazolam, zolpidem and telithromycin) based on AUC increases observed in interaction studies with itraconazole or ketoconazole. Similarly, the time-averaged apparent inhibition ratio of CYP3A4 (IRCYP3A4) was calculated for 18 inhibitors (ketoconazole, voriconazole, itraconazole, telithromycin, clarithromycin, saquinavir, nefazodone, erythromycin, diltiazem, fluconazole, verapamil, cimetidine, ranitidine, roxithromycin, fluvoxamine, azithromycin, gatifloxacin and fluoxetine) primarily based on AUC increases observed in drug-interaction studies with midazolam. The increases in the AUC of a substrate associated with coadministration of an inhibitor were estimated using the equation 1/(1 - CRCYP3A4 · IRCYP3A4), based on pharmacokinetic considerations.

Results

The proposed method enabled predictions of the AUC increase by interactions with any combination of these substrates and inhibitors (total 251 matches). In order to validate the reliability of the method, the AUC increases in 60 additional studies were analysed. The method successfully predicted AUC increases within 67–150% of the observed increase for 50 studies (83%) and within 50–200% for 57 studies (95%). Midazolam is the most reliable standard substrate for evaluation of the in vivo inhibition of CYP3A4. The present analysis suggests that simvastatin, lovastatin and buspirone can be used as alternatives. To evaluate the in vivo contribution of CYP3A4, ketoconazole or itraconazole is the selective inhibitor of choice.

Conclusion

This method is applicable to (i) prioritise clinical trials for investigating drug interactions during the course of drug development and (ii) predict the clinical significance of unknown drug interactions. If a drug-interaction study is carefully designed using appropriate standard drugs, significant interactions involving CYP3A4 will not be missed. In addition, the extent of CYP3A4-mediated interactions between many other drugs can be predicted using the current method.

Background

Cytochrome P450 (CYP) 3A4 is the most prevalent CYP enzyme in the human liver. It accounts for ≈30% of the total CYP enzymes in hepatic microsomes and is involved in the metabolism of >50% of the drugs currently on the market.[1,2] CYP3A4 is also the target enzyme for a number of drug interactions of significant clinical concern. Drug interactions are one of the major sources of adverse events, and some have actually led to drug withdrawals in the past.[35] Even though there are numerous reports of CYP3A4 drug interactions, it is far from easy to estimate all potential interactions, since too many drugs are metabolised by CYP3A4. For this reason, a comprehensive framework for the prediction of drug interactions would be of significant clinical importance. In addition, pharmaceutical companies are encouraged to carry out many in vivo drug-interaction studies during the drug development process, and the cost of these studies is increasing. Consequently, it is important to prioritise significant drug interactions to be confirmed as early as possible during the course of development. A reliable method for the prediction of CYP3A4 drug interactions would be advantageous in such circumstances.

A great deal of effort has already been devoted to establish a method for the accurate prediction of in vivo drug interactions using in vitro experimental data.[611] These predictions in principle rely on the [I]/Ki ratio, i.e. a ratio of the unbound concentration of the inhibitor at the interaction site to the in vitro inhibition constant. The results of these studies have increased our understanding of the mechanisms of drug interactions. Nowadays, both human liver specimens and expressed human CYP enzymes are commercially available, and it is not difficult to determine a profile of metabolic drug interactions in vitro. However, the proper interpretation and quantitative extrapolation of in vitro data to in vivo situations require a detailed understanding of the overall pharmacokinetics of the drugs involved. Consideration should be given to the site of interaction, the time-courses of the unbound drug concentration at the site, the effects of drug transporters on the pharmacokinetics, and the possible contribution of metabolites to the interaction.[12]

Moreover, the quantitative prediction of drug interactions is difficult for the following reasons:
  • Intestinal CYP3A4 plays a significant role in the first-pass metabolism of orally administered drugs. For example, several human in vivo studies have shown that midazolam, felodipine, ciclosporin and buspirone are extensively metabolised in the intestine.[13] Although Caco-2 cells are used in predicting the extent of intestinal absorption, it is difficult to predict the intestinal metabolism because of the very low expression of CYP3A4 in this cell line.[14] It is also known that CYP3A4 does not distribute uniformly along the length of the intestine — it is expressed more in the jejunum than in the ileum.[15] In addition, quantitative prediction of oral bioavailability is difficult because of the synergistic role of CYP3A4 and efflux transporters, such as multidrug resistance-1 (MDR1), in reducing the intestinal absorption of substrate drugs.[1620] MDR1 is expressed more in the ileum than in the jejunum.[21] Although the intestine is also considered an important site of drug interactions, the extent of intestinal metabolism is unpredictable in many cases.

  • CYP3A4 recognises a wide range of substrates, and some structural flexibility has been suggested at the substrate recognition site.[22] Consequently, the enzyme kinetics of CYP3A4 are sometimes complicated. Indeed, simple competitive inhibition theory has often failed to explain interactions via CYP3A4.[23]

  • A series of CYP3A4 substrates apparently act as mechanism-based inhibitors which covalently bind to the enzyme. For these drugs, the recovery of CYP3A4 activity depends on regeneration of the enzyme at the target site. For this reason, predictions of the mechanism-based interactions from in vitro data require more complicated kinetic models compared with reversible inhibitors.[7,8,13,2426]

Considering these complex factors, it is reasonable to conclude that, by using only in vitro experimental data, precise prediction of in vivo drug interactions is not easy for the variety of drugs that are metabolised by CYP3A4. One of the practical methods to overcome this problem is to use in vivo information on interaction data of probe drugs of CYP3A4. This approach would enable the prediction of various drug interactions from results of a small number of drug-interaction studies carried out early in the course of drug development.

The objective of the present study was to construct a framework for the prediction of various drug interactions mediated by CYP3A4 using minimum in vivo information on drug interactions. We selected midazolam as a standard substrate and ketoconazole or itraconazole as a standard inhibitor. We aimed to keep the method as simple as possible from a practical viewpoint while, at the same time, remaining theoretically accurate.

Methods

The analysis is based on 113 in vivo studies reported in 78 articles published over the period 1983–2006 (table I). Based on information from a comprehensive review,[9] we added some new data from the literature. Some substrates and inhibitors were removed from the original information because of the small contribution or low selectivity for CYP3A4. Studies were used if the report included information on the dosage regimen and the increase in the area under the concentration-time curve (AUC). A survey of a series of articles revealed that, in general, inhibitor drugs were administered consecutively for more than several days to attain steady-state conditions, and substrate drugs were given as a single-dose administration.
https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Tab1.jpg
Table I

Relationships between investigational drugs and references that reported clinical drug-interaction studies involving CYP3A4 used as the data source. The results of the referenced publications were used to evaluate the propriety of the present method

The oral clearance (CLoral) of drugs is given by equation 1, where CLtot, CLH, CLR, Fa, FG and FH are the total body clearance, hepatic clearance, renal clearance, absorption ratio, and gastrointestinal and hepatic availabilities, respectively. The reason why Fa is located at the left side in equation 1 is that the current analysis (right-side terms) should focus on events after absorption.
$${\rm{C}}{{\rm{L}}_{{\rm{oral}}}} \bullet {{\rm{F}}_{\rm{a}}} = {{{\rm{C}}{{\rm{L}}_{{\rm{tot}}}}} \over {{{\rm{F}}_{\rm{G}}} \bullet {{\rm{F}}_{\rm{H}}}}} = {{{\rm{C}}{{\rm{L}}_{\rm{H}}} + {\rm{C}}{{\rm{L}}_{\rm{R}}}} \over {{{\rm{F}}_{\rm{G}}} \bullet {{\rm{F}}_{\rm{H}}}}}$$
(Eq. 1)
Assuming the well-stirred organ model,[105] equation 1 can be converted to equation 2 where fu and CLint(H) are the unbound fraction of the drug in the blood and the intrinsic clearance of the liver, respectively. This equation represents a general relationship between the oral clearance and the intrinsic clearance of the liver.
$${\rm{C}}{{\rm{L}}_{{\rm{oral}}}} \bullet {{\rm{F}}_{\rm{a}}} = {{{{\rm{f}}_{\rm{u}}} \bullet {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}}} \over {{{\rm{F}}_{\rm{G}}}}} + {{{\rm{C}}{{\rm{L}}_{\rm{R}}}} \over {{{\rm{F}}_{\rm{G}}} \bullet {{\rm{F}}_{\rm{H}}}}}$$
(Eq. 2)
In the present study, two simplifications were made with respect to equation 2. First, we assumed that CLR can be ignored, which is frequently the case for lipophilic CYP3A4 substrates. Second, we assumed that FG can be regarded as 1.0 hypothetically. This simplification does not necessarily mean that the gastric metabolism is insignificant. Rather, it means that the gastric metabolism is assumed to occur in proportion to the hepatic metabolism as if it is a part of the liver. This may allow some overestimation of CLint(H) as a consequence. Significant gastric first-pass effects are well established facts for some CYP3A4 substrates; FG values of midazolam, felodipine, ciclosporin and buspirone are reported to be 0.57, 0.45, 0.39 and 0.21, respectively.[13,16] In the future, it may be advantageous to distinguish between gastric and hepatic metabolism by CYP3A4. At present, however, we have no alternative but to accept this simplification, since we do not know the extent of the in vivo gastric metabolism for all of the drugs analysed in the present study. In this connection, it is worth noting that incorporation of gut wall CYP3A4 inhibition did not result in a general improvement in drug-interaction predictions in a recent report.[106] Overall, equation 2 becomes equation 3 with these simplifications.
$${\rm{C}}{{\rm{L}}_{{\rm{oral}}}} \bullet {{\rm{F}}_{\rm{a}}}\tilde = {{\rm{f}}_{\rm{u}}} \bullet {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}}$$
(Eq. 3)
We then considered a relationship between the intrinsic clearance and metabolic drug interactions under an assumption of the rapid equilibrium of the drug concentration between the blood and the hepatocyte. It is often the case that a drug is metabolised by more than two pathways. In equation 4, we assume two intrinsic metabolic clearances, CLint(CYP3A4) and CLint(others), which represent the metabolism mediated by CYP3A4 and the sum of other metabolic pathways, respectively.[11,107]
$${\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}} = {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{CYP3A4}})}} + {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{others}})}}$$
(Eq. 4)
In the following equations, asterisks denote parameters altered by a drug interaction. When the metabolism of CYP3A4 is inhibited by a drug interaction, the altered clearance is given by equation 5.
$${\rm{C}}{{\rm{L}}^*}_{{\mathop{\rm int}} ({\rm{CYP3A4}})} = {{{\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{CYP3A4}})}}} \over {1 + {{[{\rm{I}}]} \over {{{\rm{K}}_{\rm{i}}}}}}}$$
(Eq. 5)

Equation 5 is applicable to both competitive and non-competitive inhibitions, since the drug concentration in vivo is usually much lower than the Michaelis-Menten constant (Km) value.

Here, we define the ratio of the contribution of CYP3A4 to oral clearance (CRCYP3A4) by equation 6:
$${\rm{C}}{{\rm{R}}_{{\rm{CYP3A4}}}} = {{{\rm{C}}{{\rm{L}}_{{\rm{oral}}}} - {\rm{C}}{{\rm{L}}_{{\rm{oral}}( - {\rm{CYP}}3{\rm{A}}4)}}} \over {{\rm{C}}{{\rm{L}}_{{\rm{oral}}}}}}$$
(Eq. 6)
where CLoral(−CYP3A4) is an altered in vivo oral clearance when CLint(CYP3A4) is blocked completely. CLoral(−CYP3A4) is given by equation 7, based on equations 3 and 4.
$${\rm{C}}{{\rm{L}}_{{\rm{oral}}( - {\rm{CYP}}3{\rm{A}}4)}} \bullet {{\rm{F}}_{\rm{a}}} = {{\rm{f}}_{\rm{u}}} \bullet ({\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}} - {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{CYP}}3{\rm{A}}4)}})$$
(Eq. 7)
From equations 6 and 7, CRCYP3A4 is given by equation 8, which indicates that the ratio of the in vivo contribution of CYP3A4 to oral clearance has the same value as the fraction metabolised by CYP3A4 to inhibition (fm(CYP3A4), which is determined by examining the effect of CYP3A4 selective inhibitors/antibodies on drug metabolism by human liver microsomes[108,109] when the urinary excretion is very low and rapid equilibrium is achieved in the liver. These relationships have been widely used for prediction of in vivo clearances in the presence of drug interactions or CYP enzyme polymorphisms, as reported by other groups.[11,107]
$${\rm{C}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}} = {{{\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{CYP}}3{\rm{A}}4)}}} \over {{\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}}}}$$
(Eq. 8)
Equation 5 can be converted to equation 9 using equations 6 and 8.
$${\rm{C}}{{\rm{L}}^*}_{{\mathop{\rm int}} ({\rm{H}})} = {{{\rm{C}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}} \bullet {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}}} \over {1 + {{[{\rm{I}}]} \over {{{\rm{K}}_{\rm{i}}}}}}} + (1 - {\rm{C}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}}) \bullet {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}} = (1 - {\rm{C}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}} \bullet {{[{\rm{I}}]} \over {[{\rm{I}}] + {{\rm{K}}_{\rm{i}}}}}) \bullet {\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}}$$
(Eq. 9)
To estimate AUC from equation 9, the equation needs to be integrated with time. For this purpose, the time-averaged apparent inhibition ratio (IRCYP3A4) is defined by equation 10:
$${\rm{I}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}} = {{[{{\rm{I}}_{{\rm{app}}}}]} \over {[{{\rm{I}}_{{\rm{app}}}}] + {{\rm{K}}_{\rm{i}}}}}$$
(Eq. 10)
where [Iapp] is the time-averaged apparent unbound concentration of the inhibitor in the liver. The increase in the ratio of the AUC caused by a drug interaction (equation 10) is derived from equations 9 and 10. We assumed here that the value of IRCYP3A4 for an inhibitor is the same for any substrate.
Equation 11 indicates that an AUC increase by a drug interaction between any oral drug via CYP3A4 can be predicted if the CRCYP3A4 of the substrate and the IRCYP3A4 of the inhibitor are available.
$${{{\rm{AU}}{{\rm{C}}^*}_{{\rm{oral}}}} \over {{\rm{AU}}{{\rm{C}}_{{\rm{oral}}}}}} = {{{\rm{C}}{{\rm{L}}_{{\rm{oral}}}} \bullet {{\rm{F}}_{\rm{a}}}} \over {{\rm{C}}{{\rm{L}}^*}_{{\rm{oral}}} \bullet {{\rm{F}}_{\rm{a}}}}} = {{{\rm{C}}{{\rm{L}}_{{\mathop{\rm int}} ({\rm{H}})}}} \over {{\rm{C}}{{\rm{L}}^*}_{{\mathop{\rm int}} ({\rm{H}})}}} = {1 \over {1 - {\rm{C}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}} \bullet {{[{{\rm{I}}_{{\rm{app}}}}]} \over {[{{\rm{I}}_{{\rm{app}}}}] + {{\rm{K}}_{\rm{i}}}}}}} = {1 \over {1 - {\rm{C}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}} \bullet {\rm{I}}{{\rm{R}}_{{\rm{CYP}}3{\rm{A}}4}}}}$$
(Eq. 11)

It needs to be mentioned that the above theory may not be directly applicable to mechanism-based inhibitors. However, the final form of equation 11 can be accepted even for mechanism-based inhibitors by regarding the IRCYP3A4 values as overall inhibition ratios of CYP3A4 at the equilibrium state between inactivation and generation of the metabolising enzyme. From this viewpoint, the definition of IR by equation 10 is invalid for mechanism-based inhibitors.

The values of CRCYP3A4 and IRCYP3A4 for various substrates and inhibitors were calculated sequentially according to the following steps based on AUC increases in the 53 interaction studies, which are indicated in table I.
  • We assumed that the CRCYP3A4 value of simvastatin is 1.0 since it has been reported that simvastatin is a selective substrate of CYP3A4,[32] and a search of the literature showed that the plasma AUC of simvastatin tends to be increased most markedly following inhibition of CYP3A4. It was confirmed that a reduction of the CRCYP3A4 value of simvastatin to 0.95 did not affect the overall outcomes of the present analysis.

  • Once we assumed the CRCYP3A4 value for simvastatin, the IRCYP3A4 value of itraconazole, a typical inhibitor, was obtained based on equation 11, using the result of a drug-interaction study involving simvastatin and itraconazole.

  • The CRCYP3A4 value of midazolam, a typical substrate, was calculated from studies with midazolam and itraconazole using the calculated IRCYP3A4 value of itraconazole with equation 11. An algebraic mean of the AUC increase was used for the calculation, whenever the results of multiple studies are reported for an interaction set of interest.

  • The IRCYP3A4 values of the other inhibitors including ketoconazole, another typical inhibitor, were calculated from interaction studies between the inhibitor and midazolam, using the calculated CRCYP3A4 value of midazolam with equation 11.

  • The CRCYP3A4 values of other substrates were calculated from interaction studies between the substrate and itraconazole or ketoconazole whenever possible, using the calculated IRCYP3A4 value of itraconazole or ketoconazole, respectively, with equation 11.

  • For nifedipine, no interaction study with itraconazole or ketoconazole has been reported. Accordingly, the CRCYP3A4 value of nifedipine was calculated from the study with nifedipine and diltiazem, using the calculated IRCYP3A4 value of diltiazem, with equation 11. Diltiazem was selected, since the AUC of nifedipine was most significantly increased by the administration of diltiazem.

Results

We surveyed 113 in vivo drug-drug interaction studies published in 78 articles (table I). The CRCYP3A4 values of 14 substrates and the IRCYP3A4 values of 18 inhibitors were calculated using equation 11 based on the results of 53 clinical studies (the estimation set), which are indicated in table I. As shown in table II, high CRCYP3A4 values of >0.95 were calculated for simvastatin, lovastatin, buspirone and nisoldipine, 0.85–0.94 for triazolam, midazolam and felodipine, and 0.70–0.84 for ciclosporin, nifedipine and alprazolam. High IRCYP3A4 values of >0.95 were estimated for ketoconazole (daily dose 200–400mg), voriconazole (400mg) and itraconazole (100–200mg), 0.85–0.94 for telithromycin (800mg), clarithromycin (500–1000mg), saquinavir (3600mg) and nefazodone (400mg), and 0.70–0.84 for erythromycin (1000–2000mg), diltiazem (90–270mg), fluconazole (200mg) and verapamil (240–480mg).
https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Tab2.jpg
Table II

Calculated ratios of the contribution of CYP3A4 to the oral clearance (CRCYP3A4) of substrates

The current method enabled predictions of the AUC increase caused by drug-drug interactions of any combination of the substrates and inhibitors. In order to validate the suitability of the present method, the extent of the increase in the AUC by drug interaction was predicted for the remaining 60 clinical studies (the validation set), which are indicated in table I, and the results were compared with the observed values (figure 1). This prediction was performed by substituting the values of CRCYP3A4 and IRCYP3A4 shown in table II and table III, respectively, in equation 11. As shown in figure 1, with the current method we could predict the increase in the AUC within 67–150% of the observed AUC increase for 50 clinical studies (83%) and within 50–200% for 57 clinical studies (95%).
https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Tab3.jpg
Table III

Calculated ratios of the time-averaged apparent inhibition ratio of CYP3A4 (IRCYP3A4) for inhibitors

https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Fig1.jpg
Fig. 1

Relationship between the observed and calculated increase in the area under the concentration-time curve (AUC) by drug interactions. Using the ratio of the contribution of cytochrome P450 (CYP) 3A4 to oral clearance (CRCYP3A4) and the time-averaged apparent inhibition ratio of the CYP3A4 (IRCYP3A4) values shown in table II and table III, respectively, the increase in the AUC of substrate drugs by drug interactions reported in 60 clinical studies (the validation set; table I) was predicted with equation 11. Each circle and vertical bar represents the mean + SD values of subjects reported in each article. A dashed bar represents the range. Where the SD values or the ranges were not reported in the article, the reported mean value is shown by a square. The solid and dotted lines represent 50–200% and 67–150% ranges, respectively, of the calculated increase.

In the calculation of CRCYP3A4 and IRCYP3A4 values, the algebraic mean of the increase in the AUC was used when more than one article was available for the same interaction set (table I). In these cases, however, significant deviation was observed in the AUC increase between or among reports. For the analysis, we often combined the data of clinical studies with different doses of the inhibitor. Since the lower doses frequently gave more AUC increase of a substrate, it is possible that the deviation of the inhibitor dose may not largely affect the results, as far as the dose of the inhibitor is set within the therapeutic range. The extent of this deviation is shown in figure 2, which was prepared in the same style as figure 1. Each circle and vertical bar in figure 2 represents the mean + SD reported in each article of the estimation set in table I. If the SD values were not reported in articles listed in table I, the reported mean values are shown by squares. As shown by the dotted lines in figure 2, for most of the articles the increase in the AUC of substrate drugs caused by drug inhibition deviated by 67–150% of the algebraic mean values. The predictions within 50–200% of the observed AUC increase in figure 1 were regarded as successful, since the corresponding variation of the AUC in figure 2(the estimation set) was within this range.
https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Fig2.jpg
Fig. 2

Relationship between the observed and calculated increase in the area under the concentration-time curve (AUC) by drug interactions. This figure was prepared in the same style as figure 1 for the purpose of demonstrating the deviation of AUC values among 53 clinical studies (the estimation set; table I), the mean values of which were used to determine the CRCYP3A4 and IRCYP3A4 values. Each circle and vertical bar represents the mean + SD values of subjects reported in each article. A dashed bar represents the range. Where the SD values or the ranges were not reported in the article, the reported mean value is shown by a square. The solid and dotted lines represent 50–200% and 67–150% ranges, respectively, of the calculated increase.

The data shown in figure 1 and figure 2 were then reorganised to indicate the relationships between the IRCYP3A4 values and the increase in AUC of each substrate (figure 3). It was found that the AUC increased steeply as the IRCYP3A4 value increased for highly CYP3A4-dependent substrates, such as simvastatin, lovastatin and buspirone, whereas only minimal increases were observed for poor CYP3A4 substrates, such as zolpidem and cerivastatin (figure 3). In the same manner, potent inhibitors, such as azole antifungals, increased the blood levels of a number of CYP3A4 substrates markedly, whereas no or only very minor increases were observed for weak inhibitors, such as azithromycin, gatifloxacin and fluoxetine (figure 4).
https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Fig3.jpg
Fig. 3

Increase in the area under the concentration-time curve (AUC) reorganised for each substrate drug as a function of the inhibition ratio (IR) of the inhibitors. The data shown in figure 1 and figure 2 were reorganised to show the increase in the AUC of each substrate drug as a function of the time-averaged apparent inhibition ratio of CYP3A4 (IRCYP3A4) values of the inhibitors. The closed and open symbols represent the dataset shown in figure 1 and figure 2, respectively. See figure 1 and figure 2 legends for details.

https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Fig4.jpg
Fig. 4

Increase in the area under the concentration-time curve (AUC) of the substrate drugs reorganised for each inhibitor as a function of the contribution ratio (CR) values of the substrate drugs. The data shown in figure 1 and figure 2 were reorganised for each inhibitor to show the increase in the AUC of each substrate drug as a function of the contribution ratio of CYP3A4 to the oral clearance (CRCYP3A4) values of the substrate drugs. The closed and open symbols represent the dataset shown in figure 1 and figure 2, respectively. See figure 1 and figure 2 legends for details.

Finally, the data were reorganised to show that the increase in the AUC in 251 kinds of drug interactions between 14 substrate drugs and 18 inhibitors could be predicted (figure 5; note that telithromycin is included with both the substrates and the inhibitors). The nomogram in figure 5 indicates that a very marked increase in the AUC is anticipated when substrate drugs with high CRCYP3A4 values are administered with potent inhibitors with high IRCYP3A4 values.
https://static-content.springer.com/image/art%3A10.2165%2F00003088-200746080-00005/MediaObjects/40262_2012_46080681_Fig5.jpg
Fig. 5

Predicted increase (log scale) in the area under the concentration-time curve (AUC) of substrate drugs by various drug interactions. The increase in AUC of substrate drugs by various drug interactions was predicted according to the ratio of the contribution of CYP3A4 to oral clearance (CRCYP3A4) and time-averaged apparent inhibition ratio of CYP3A4 (IRCYP3A4) values shown in table II and table III, respectively. Closed and open arrowheads indicate the dataset shown in figure 1 and figure 2, respectively.

Discussion

CYP3A4 is the most important drug-metabolising enzyme, which preferentially oxidises relatively large, lipophilic, neutral to basic molecules. Therefore, CYP3A4 is recognised as a key enzyme that determines the clearance of various drugs and, in some cases, has a major effect on their safety and efficacy. Although no major polymorphism in the CYP3A4 gene has been identified, marked interindividual differences have been reported in the activity of CYP3A4.[110] One possible reason for such differences in the activity is that CYP3A4 is inducible by various diets and drugs, such as rifampicin (rifampin) and carbamazepine, via the mechanism mediated by pregnane X receptor.[111113] Furthermore, CYP3A4 is the predominant metabolising enzyme not only in the liver but also in the intestine. It has been reported that intestinal metabolism is the major factor determining the bioavailability of some drugs.[16,114116] However, as far as we know, nobody has succeeded in predicting the extent of the first-pass effect on metabolism by intestinal CYP3A4 from in vitro data. Although there are some established methods to determine the activity of CYP3A4 in vivo, including evaluation of the metabolic ratio of selective substrates (midazolam, testosterone and cortisol) and the erythromycin breathe test, it has been reported that these methods do not offer consistent results,[117] possibly due to differences in the organ of metabolism (liver or intestine) and/or the presence of multiple recognition sites in the CYP3A4 molecule.[118]

Despite these issues regarding the in vivo evaluation of CYP3A4 activity, the current rather simple method gave satisfactory predictions in most cases. The following issues may contribute to this success. First, uncertain factors were avoided since the current method relies primarily on an overall in vivo evaluation. For example, the present method satisfactorily predicted drug interactions with mechanism-based inhibitors such as azithromycin, clarithromycin, diltiazem, erythromycin and roxithromycin (figure 4), which frequently exhibit complicated kinetics. Accurate predictions have been achieved recently from in vitro data for mechanism-based inhibitors by sophisticated analysis. For the successful prediction, it has been reported that evaluation of the unbound fraction of the drug in the incubation medium is important.[13,119] Moreover, the analysis requires a turnover rate of the metabolising enzyme and a rate constant for the irreversible reaction, both of which are not easy to estimate from in vitro experiments.

Second, we used simvastatin as a selective substrate and ketoconazole and itraconazole as selective inhibitors of CYP3A4, although these drugs are not absolutely specific for CYP3A4. For example, although we assumed that the CRCYP3A4 value of simvastatin is 1.0, this drug is also metabolised by CYP2C8 to a minor extent.[105] Ketoconazole is a well known selective inhibitor of CYP3A4, but this drug also inhibits the activities of CYP2C8,[120] CYP2C9[121] and MDR1,[122] which may also affect the disposition of substrate drugs analysed in the present study. Despite these defects, the success in the prediction of drug interactions with the present method (figure 1 and figure 2) suggests that CYP3A4 plays a crucial role in most of the drug interactions analysed in the present study.

A number of probe drugs have been used to study the activity of CYP3A4, including midazolam, nifedipine, simvastatin and erythromycin.[123] Among them, it is generally recognised that the most reliable probe drug is midazolam for CYP3A4.[108,124] The plasma AUC of midazolam is increased significantly by coadministration of various CYP3A4 inhibitors (figure 1, figure 2, figure 3 and figure 5). In our preliminary analysis, we found that the rank order of the AUC increase of typical substrates, such as simvastatin, lovastatin and buspirone, by a series of inhibitors was generally in good agreement with the rank order of the AUC increase of midazolam produced by these inhibitors. These results suggest that the extent of CYP3A4 inhibition after administration of each inhibitor is almost the same among substrates. From this analysis, we hypothesised that calculation of AUC increases from IRCYP3A4 values should be possible.

It has often been reported that in vitro Ki values vary significantly among the CYP3A4 substrates used,[10] which contradicts our hypothesis that the IRCYP3A4 value is the same for any substrate. For example, nifedipine was allocated to a different group from midazolam and triazolam by a cluster analysis of the victim profile of in vitro drug interactions.[125] However, as represented in figure 3, no clear discrepancy was observed for the predicted AUC increases of any particular substrate assuming a single IRCYP3A4 value for each inhibitor. It is therefore possible that the in vivo Ki value of each inhibitor is not affected by the substrate drugs analysed in the present study. This result may be because the number of available drug-interaction studies is limited for each inhibitor. Accordingly, we should be cautious in predicting the increase in the AUC for a novel substrate drug by using the IRCYP3A4 values determined in the present study.

In the validation process of our study, the method provided successful predictions in 57 of 60 cases. Telithromycin is a particular example of an accurate prediction. The CRCYP3A4 value of telithromycin was calculated to be 0.49 from the results of an interaction study with ketoconazole. The AUC increase produced by interaction with itraconazole, which has an IRCYP3A4 value of 0.95, was predicted to be 1.85, which was in good agreement with the observed increase of 1.60. Telithromycin also acts as an inhibitor of CYP3A4. The IRCYP3A4 value of telithromycin was 0.91 and an AUC increase of simvastatin produced by interaction of telithromycin was predicted to be 11.1, which was also in good agreement with the observed increase of 10.8.

In contrast, we had difficulties in predicting three reports of drug interactions, i.e. ciclosporin-voriconazole, triazolam-itraconazole, and one of two reports for a triazolam-erythromycin interaction. In reference[103], it was reported that the AUC of ciclosporin was increased 1.70-fold by the administration of voriconazole, whereas we predicted a 4.61-fold increase (figure 1 and figure 5). In the same manner, although the AUC of triazolam was reported to be increased 27.1-fold by administration of itraconazole,[88] we predicted an 8.85-fold increase (figure 1 and figure 5). Concerning the interaction between triazolam and erythromycin, there was a deviation in the increase in the AUC of triazolam by erythromycin. In reference[31], a 3.65-fold increase was reported whereas a 2.06-fold increase was reported in reference[62]. Our prediction was 4.32-fold (figure 1 and figure 5) and was in accord with the former article. The reason for these deviations is unknown. Further studies will help to investigate whether there was some mechanistic reason or whether there was simply some unavoidable variability. The factors that need to be considered include the contributions by other metabolising enzymes and transporters, and the variety of enzyme kinetics of CYP3A4 inhibition.

In the present study, midazolam, itraconazole and ketoconazole were used as a standard substrate or an inhibitor because they are used most commonly in drug-interaction studies. As a result, overall AUC increases were successfully predicted, indicating that the standard drugs were selected appropriately. It may be possible to use other commonly used substrates of CYP3A4 such as simvastatin, lovastatin and buspirone to calculate IRCYP3A4 values, because no deviation was observed in the predictability of AUC increase for these substrate drugs when coadministered with a wide range of inhibitors (figure 3).

To prioritise drug-interaction studies during the course of drug development, Obach et al.[126] have recently proposed a rank-order approach in which the mechanism of possible interactions is explored by in vitro experiments and then the most probable interactions are evaluated in vivo using typical substrates or inhibitors. The results of the present study support their approach. If a drug-interaction study is carefully designed using the appropriate standard drugs, significant interactions via CYP3A4 will not be missed. In addition, the extent of CYP3A4-mediated interactions between many other drugs will be able to be predicted using the current method, as suggested by the results in figure 5.

Conclusion

We have constructed a general framework for prediction of the increase in AUC mediated by CYP3A4. The precision and robustness of the method have been demonstrated satisfactorily. Several standard substrates and inhibitors are proposed for the evaluation of drug interactions via CYP3A4. This method would be applicable to (i) prioritise clinical trials for investigating drug interactions during the course of drug development and (ii) estimate the clinical significance of unknown drug interactions.

Acknowledgements

Dr Akihiro Hisaka now works in Pharmacology and Pharmacokinetics at the University of Tokyo Hospital Faculty of Medicine, University of Tokyo, Tokyo, Japan.

This study was supported by Health and Labor Sciences Research Grants for Research on Regulatory Science of Pharmaceuticals and Medical Devices from the Ministry of Health, Labor and Welfare, Japan. The authors have no conflicts of interest that are directly relevant to the content of this study.

Copyright information

© Adis Data Information BV 2007