There Are No Useful CYP3A Probes that Quantitatively Predict the In Vivo Kinetics of Other CYP3A Substrates and No Expectation that One Will Be Found

  1. Leslie Z. Benet
  1. Department of Biopharmaceutical Sciences, University of California, San Francisco, CA 94143-0446

Over thirty years ago, Garrett proposed dosing a drug cocktail to define the metabolic characteristics of a patient, allowing the selection of appropriate doses of drugs metabolized by the same enzymatic pathway as the probe compounds within the cocktail (1). As recently noted by Masica et al. (2), nonselective drugs, such as antipyrine, were originally proposed as probes to measure oxidation capacity (3), but the genetic elucidation of the cytochrome (CYP) P450 superfamily and the recognition of the polymorphic expression of certain enzymes made it possible to use more-specific probe drugs (4) to differentiate allele-dependent differences in the metabolism of substrates for CYP2D6 (e.g., debrisoquine and dextromethorphan) and CYP2C19 (e.g., mephenytoin and omeprazole), although quantitative predictions of metabolic ability and drug kinetics in individual patients have not been reported. Probes for enzymes that exhibit unimodal variability in metabolism (e.g., variability that cannot be ascribed to easily recognized genetic differences) have also been proposed with varying degrees of predictive success from study to study and laboratory to laboratory. Of particular interest has been the search for a probe for the CYP3A enzymes that are responsible for the metabolism of the majority of drugs in humans, and for which marked variability in drug elimination has been found. A small portion of this variability may be attributed to the polymorphic expression of CYP3A5 (5); however, the attempt to define probes for CYP3A4 has been “remarkably poor,” as reported by Masica et al. (2) who characterized “the correlations between pairs of the various presumed trait measures of CYP3A-mediated metabolism” [i.e., in vivo metabolism of dapsone, cyclosporine, endogenous cortisol, nifedipine, lidocaine, dextromethorphan, alprazolam, and midazolam, and most frequently exhaled radiolabeled CO2 utilizing the erythromycin breath test (Box 1).

Box 1.

Erythromycin Breath Test.

The erythromycin breath test is based on the observation that demethylation of erythromycin is mediated by CYP3A4, and the carbon in the methyl group is exhaled as CO2. A subject is given an intravenous dose of a trace amount of 14C-N-methyl erythromycin, and CYP3A4 activity is calculated based on the radiolabel collected in the exhaled breath.

Masica et al. (2) reviewed the poor performance of previous studies of in vivo probes of CYP3A activity and identified the flaws in some of these studies that went unrecognized when first performed. Most recently, Wu and Benet (6) demonstrated that categorizing drugs into the four classes represented by the Biopharmaceutics Classification System (BCS) (Box 2) (7, 8), employing solubility and permeability criteria, may be useful in predicting: 1) routes of drug elimination, 2) effects of efflux and absorptive transporters on oral absorption, 3) when transporter-enzyme interplay will yield clinically significant effects such as low bioavailability and drug-drug interactions, 4) the direction and importance of food effects, and 5) the effects transporters have on (post-absorption) systemic concentrations of drugs following oral and intravenous dosing. Wu and Benet (6) state, “Attempts to use markers of enzymatic processes (e.g., midazolam vs the erythromycin breath test) to predict metabolism of another substrate cannot be expected to work when the test drugs are in different BCS classes. Even when two enzymatic substrates are in the same class, there is little chance to detect a potential correlation when the two test compounds are substrates for different uptake and efflux transporters. Many, many papers have investigated the potential for one substrate to predict the metabolism of other substrates by the same enzyme. Almost all of these attempts have failed and we believe that the reason for the lack of correlation is due to differences in transporter susceptibilities. For example, erythromycin is a substrate for both uptake and efflux transporters as well as of CYP3A4. It is obvious that the ability of erythromycin metabolism to predict the metabolism of other CYP3A4 compounds will be compromised if differences in transport are not identified and fully taken into account. We believe, at this time, the administration of ‘cocktails’ of substrates (i.e., a mixture of small quantities of drugs that are specific substrates for particular metabolic enzymes) to characterize a patient’s metabolic profile will be of little use, except for the most obvious pharmacogenetic differences in enzyme capacity.”

Box 2.

Four Classes of Drugs as Categorized by the Biopharmaceutics Classification System.

Class 1: High solubility and high intestinal membrane permeability. Extensively metabolized. Examples: lidocaine, midazolam, nifedipine, and verapamil.

Class 2: Low solubility and high intestinal membrane permeability. Extensively metabolized. Examples: amiodarone, carbamazepine, cyclosporine, dapsone, and saquinavir.

Class 3: High solubility and low intestinal membrane permeability. Poorly metabolized. Examples: atenolol, erythromycin, fexofenadine, and ranitidine.

Class 4: Low solubility and low intestinal membrane permeability. Poorly metabolized. Examples: amphoteracin B, chlorothiazide, and furosemide.

Of the putative probe drugs for CYP3A considered by Masica et al. (2), lidocaine, midazolam, and nifedipine belong to BCS Class 1, where transporter effects are generally unimportant in vivo; dapsone and cyclosporine belong to Class 2, where efflux transporters have a marked effect on oral bioavailability and both uptake and efflux transporters can be important for hepatic elimination; and erythromycin belongs to Class 3, where uptake transporters are important for bioavailability, while uptake and efflux transporter effects are important for elimination, as suggested by Wu and Benet (6) Alprazolam, cortisol, and dextromethorphan are not classified, but are most likely members of Class 1. Midazolam and nifedipine were thought not to be substrates for the efflux transporter P-glycoprotein (P-gp), but recent work suggests that this may not be the case (9, 10). In vitro studies in Caco-2 cells comparing apical-to-basolateral and basolateral-to-apical flux do not exhibit differential transport, but studies in a high permeability resistant transfected cell line, such as MDCK-MDR1, show preferential basolateral-to-apical transport that can be decreased by P-gp inhibitors.

The purpose of Masica and colleagues’ study (2) was to determine the degree of predictability of one CYP3A substrate for another utilizing three drugs: 1) that have the same routes of metabolism, 2) that share structural characteristics sufficiently similar to suggest that they may be metabolized at the same domain within the CYP3A active site, 3) that apparently are unaffected in vivo by P-gp, and 4) that are likely BCS Class 1 compounds, unaffected in vivo by any transporter. Thus, Masica et al. (2) specifically investigated the correlation of the clearances between three closely related 1,4-benzodiazepines––alprazolam, midazolam, and triazolam (all presumably BCS Class 1 compounds)––that are predominantly metabolized through the same 1′- and 4-hydroxylation pathways by CYP3A. The investigators assumed that the drugs are not transported by P-gp, and “therefore clearance measures only reflect metabolism.” Pharmacokinetic parameters (11) and the elucidated metabolic pathways (12, 13) for the three drugs are presented in Table 1, together with the experimental values found by Masica et al. (2) in parentheses. Obviously, the mean half-lives found by Masica et al. match those found in the literature; the high CVs (coefficients of variation) observed for the experimental bioavailability (F) measurements (45% midazolam and 37% triazolam) suggest no difference from the literature reports; however, the experimental oral clearance (CL/F) for triazolam, with a CV of 24%, is probably different from that reported in the literature.

The correlation coefficients between systemic clearances (CLs) following iv administration of midazolam and triazolam, and the oral clearances (CLo calculated as oral dose divided by area under the curve) for all three drugs are given in Table 2. Masica et al. (2) found that the Spearman rank correlation coefficients were statistically significant for all of the relationships except CLo alprazolam vs CLs midazolam (Table 2). In contrast, no statistically significant relationship was found between the erythromycin breath test and any of the five measured clearance parameters. Finding a significant correlation coefficient between two parameters, however, does not guarantee that a clinically useful relationship exists; such a relationship depends on the measure of the coefficient of determination (r2). As Masica et al. (2) point out, r2 values between individual pairs of the five clearances in Table 2 were only 13 to 59% That is, only 13% to 59% of the variability in the clearance of one of drug could be explained by the clearance of another structurally similar compound. This careful analysis of the lack of quantitative predictability of one in vivo probe of CYP3A clearance for another very similar probe—not to mention the many other unsuccessful attempts to predict CYP3A substrate disposition––suggest that further investigations in this area will not be productive. In fact, it is my position, that even quantitative predictions from probe compounds where genetic polymorphisms in metabolism are most prominent (CYP2D6 and CYP2C19 substrates) are not sufficiently predictive for narrow therapeutic index drugs.

In a series of papers, Kharasch and coworkers have characterized the opioid alfentanil (likely a BCS Class 1 compound) as a CYP3A substrate and suggest that the drug may be a useful in vivo probe for hepatic and first-pass CYP3A activities and drug interactions. They demonstrate that miosis (i.e., pupil constriction) may be an acceptable noninvasive surrogate for plasma alfentanil concentrations [several of their articles are referenced in (14)]. These careful, excellent, and enlightening studies of Kharasch and coworkers are useful contributions; however, their work does not support the use of alfentanil as a reliable CYP3A probe drug that can be used to predict the kinetics of other CYP3A substrates. A statistically significant correlation (p<0.005) between alfentanil clearance and midazolam clearance in healthy young men and women yielded an r2 of 17% (15), suggesting no predictability. But, Kharasch et al. (14) report r2 values as high as 92% for the correlation of alfentanil iv clearance versus midazolam iv clearance and 97% for the oral clearances of these two drugs. The experimental values utilized in these correlations are, however, clearances obtained in the same individuals under four different conditions: a) control midazolam and alfentanil doses (where as stated above poor predictability is obtained), b) in the presence of a potent inhibitor of CYP3A, troleandomycin, c) following five days of dosing rifampin, a potent inducer of CYP3A, or d) with oral grapefruit juice, an inhibitor of intestinal CYP3A. Examination of the figures in this paper confirm that midazolam–alfentanil correlations under each condition do not approach these high r2 values, and that it is only the combination of marked inhibition and/or induction that yield the apparent excellent correlations. Combining studies in this manner, which do not overlap measured values, will not improve the predictability within any region of the correlation, even though the overall r2 will increase markedly.

Most recently, Kharasch et al. (16) suggested that alfentanil and fexofenadine in combination could be used as an in vivo probe of the first-pass activities of CYP3A and P-glycoprotein (P-gp). I have stated above that alfentanil is not a useful quantitative probe for other CYP3A substrates; however, I believe that fexofenadine would even be a poorer probe of P-gp activity for three reasons. First, in contrast to the research data for CYP3A, there are no studies that suggest that the clearance of one P-gp substrate can serve as a valid predictor of the kinetics of a second P-gp substrate. The complexity of the active site and the ability of one P-gp substrate to predict even qualitative P-gp effects for a second substrate are at least as complicated as that for CYP3A. Second, as pointed out by Wu and Benet (6), fexofenadine is a BCS Class 3 substrate and its pharmacokinetics are at least as dependent upon uptake transporters in the gut and the liver as they are for the efflux transporter P-gp. The decrease in fexofenadine plasma concentrations when the drug was dosed close to a meal as observed by Kharasch et al. (16) is consistent with fexofenadine being a Class 3 compound (6, 17). It is not valid to suggest that a drug can be a probe for a specific transporter when it is subject to the variability of multiple transporters in each subject. It is doubtful that any BCS Class 3 or Class 4 compound could be used as a probe for P-gp. Third, over the past four years our laboratory has emphasized the importance of enzyme-transporter interplay in defining drug metabolism [reviewed in reference (18), and more recent publications (6, 19, 20)]. We have shown that transporters, both uptake and efflux, can change the metabolism of drugs in ways that are not reflected in microsome studies where the transporter-enzyme architecture is not maintained.

In conclusion, as stated by Masica et al. (2): “it is considered unlikely that any practically useful CYP3A in vivo probe will ever be able to predict the enzyme’s activity in an individual subject so that quantitative metabolism of another CYP3A drug can be accurately and reliable predicted under constitutive conditions. This situation may, in fact, apply to a much broader context.” I believe that this will also be true for P-gp. In their final sentence, Masica et al. (2) modulate their conclusion somewhat by stating: “Finally, the present findings do not, of course, negate the case of the in vivo probe approach for characterizing a change in CYP3A activity mediated, for example, by a drug interaction involving inhibition or induction.” Useful insight in terms of potential drug-drug interactions and as to mechanistic explanations related to CYP3A and P-gp changes on drug disposition can be forthcoming from the studies of Kharasch and coworkers, as is obvious from their published work and as exemplified by the investigation of Lown et al. (21), in which I participated. We reported that the erythromycin breath test showed an r2=56% with steady-state oral clearance of cyclosporine in nineteen kidney transplant patients and that measures of gut P-gp protein levels on biopsy increased the forward regression (r2=73%), but that gut CYP3A protein levels provided no increase in the regression. These measurements, however, do not support the erythromycin breath test as a useful quantitative probe for predicting cyclosporine pharmacokinetics and, I see no evidence that inhibitory or inductive effects with one probe CYP3A substrate can quantitatively predict the extent of the in vivo interaction for a second CYP3A substrate in an individual subject. In this case, when an interaction between a substrate and CYP3A is only “metabolic” (not involving transporter-enzyme interplay), in vitro microsomal studies that demonstrate changes in a drug’s metabolism in the added presence of an interacting substrate will be predictive of an in vivo interaction. Subsequent in vivo studies, however, then must be carried out to quantitate the extent of the interaction.

Table 1.

Comparison of Published Mean Pharmacokinetic1 and Metabolic Pathway2 Parameters for Alprazolam, Midazolam, and Triazolam with Mean Experimental Values3

Table 2.

Spearman Rank Correlations for Relationships Between Five Clearance Values*


This Viewpoint in dedicated to Professor Grant R. Wilkinson, Vanderbilt University, senior author of Reference 2, for his significant seminal contributions to pharmacokinetics and human drug metabolism over the past four decades. Cited studies in the author’s laboratory were funded in part by NIH grants GM 61390 and HD 40543 and by unrestricted funds from Amgen, Inc. I thank Professors Wilkinson and Paul B. Watkins and Drs. Chi-Yuan Wu, Vincent Wacher and Hong Sun for their useful suggestions in the preparation and review of this Viewpoint.


Leslie Z. Benet, PhD, is Professor of Biopharmaceutical Sciences and Pharmaceutical Chemistry at the University of California, San Francisco. His recent work has focused on characterizing the importance of transporter-enzyme interplay in defining drug disposition in the liver and intestine. E-mail benet{at}; fax 415-476-8887.

| Table of Contents

Navigate This Article