perspectives
`
`while ensuring the quality needed to draw
`reliable research conclusions and (ii)
`replacing the prevailing view of practice
`and research as separate activities with a
`“learning health system” methodology
`that incorporates research into practice
`as a routine element of clinical care. These
`changes will require significant adjust-
`ments to the ethical frameworks that span
`the spectrum of learning activities, from
`quality improvement to interventional
`research involving new therapies.10
`
`Conclusion
`Selker and colleagues have articulated a
`vision that is consistent with our evolv-
`ing understanding of therapeutic devel-
`opment. Before this vision can become
`a reality, numerous practical and con-
`ceptual barriers must first be overcome.
`However, revolutionary clinical research
`methods that are now being piloted have
`the potential to help make E2E a reality.
`
`CONFLICT OF INTEREST
`The author receives support from the National
`Institutes of Health and the Patient-Centered
`Outcomes Research Institute. He receives
`research grants that partially support his
`salary from Amylin Pharmaceuticals, Johnson
`& Johnson, Scios, Merck, Schering-Plough,
`Schering-Plough Research Institute, Novartis
`Pharma, Bristol-Myers Squibb Foundation,
`Aterovax, Bayer, Roche, and Lilly; all grants
`are paid to Duke University. He also consults
`for TheHeart.org, Johnson & Johnson, Scios,
`Kowa Research Institute, Nile, Parkview,
`Orexigen Therapeutics, Pozen, WebMD, Bristol-
`Myers Squibb Foundation, AstraZeneca,
`Bayer/Ortho-McNeil, Bristol-Myers Squibb,
`Boehringer Ingelheim, Daiichi Sankyo, Gilead,
`GlaxoSmithKline, Li Ka Shing Knowledge
`Institute, Medtronic, Merck, Novartis, Sanofi-
`Aventis, XOMA, University of Florida, Pfizer,
`Roche, Servier International, DSI–Lilly, Janssen
`R&D, CV-Sight, Regeneron and Gambro; all
`income from these consultancies is donated to
`nonprofit organizations, with most going to the
`clinical research fellowship fund of the Duke
`Clinical Research Institute. He holds equity in
`Nitrox LLC, N30 Pharmaceuticals, and Portola.
`Disclosure information for the author is also
`available at https://dcri.org/about-us/conflict-
`of-interest and at http://www.dukehealth.org/
`physicians/robert_m_califf.
`
`© 2014 ASCPT
`
`1. Selker, H.P. et al. A proposal for integrated
`efficacy-to-effectiveness (E2E) clinical trials. Clin.
`Pharmacol. Ther. 95, 147–153 (2014).
`2. Eapen, Z.J., Vavalle, J.P., Granger, C.B., Harrington,
`R.A., Peterson, E.D. & Califf, R.M. Rescuing clinical
`
`trials in the United States and beyond: a call for
`action. Am. Heart J. 165, 837–847 (2013).
`3. President’s Council of Advisors on Science and
`Technology (PCAST). Report to the President
`on propelling innovation in drug discovery,
`development, and evaluation <http://www.
`whitehouse.gov/sites/default/files/microsites/
`ostp/pcast-fda-final.pdf> (September 2012).
`4. Morris, S.A., Rosenblatt, M., Orloff, J.J., Lewis-
`Hall, F. & Waldstreicher, J. The PCAST report:
`impact and implications for the pharmaceutical
`industry. Clin. Pharmacol. Ther. 94, 300–302
`(2013).
`5. Thiers, F.A., Sinskey, A.J. & Berndt, E.R. Trends in
`the globalization of clinical trials. Nat. Rev. Drug
`Discov. 7, 13–14 (2008).
`6. Califf, R.M., Rasiel, E.B. & Schulman, K.A.
`
`Considerations of net present value in policy
`making regarding diagnostic and therapeutic
`technologies. Am. Heart J. 156, 879–885 (2008).
`7. Sjoerdsma, A. & Schechter, P.J. Eflornithine
`for African sleeping sickness. Lancet 354, 254
`(1999).
`8. McNeil, D.G. Jr. Cosmetic saves a cure for sleeping
`sickness. New York Times <http://www.nytimes.
`com/2001/02/09/world/cosmetic-saves-a-cure-
`for-sleeping-sickness.html> (9 February 2001).
`9. Kramer, J.M., Smith, P.B. & Califf, R.M.
`Impediments to clinical research in the United
`States. Clin. Pharmacol. Ther. 91, 535–541 (2012).
`10. Solomon, M.Z. & Bonham, A.C. (eds.). Ethical
`oversight of learning health care systems.
`Hastings Center Report Special Report 43, no. 1,
`S1–S44 (2013).
`
`See ARTICLES pages 179 and 189
`
`In Vitro Prediction of Clinical
`Drug Interactions With CYP3A
`Substrates: We Are Not There Yet
`DJ Greenblatt1
`
`In 1973, Malcolm Rowland and associates described an approach
`to predicting clinical pharmacokinetic drug–drug interactions
`(DDIs) using an inhibition constant determined in vitro (Ki)
`together with anticipated inhibitor exposure in vivo ([I]). Despite
`numerous modifications and refinements of the core model over
`the following 40 years, we still have not achieved a predictive
`paradigm having accuracy sufficient to justify bypassing all, or
`even most, clinical DDI studies in the course of drug development.
`
`The use of in vitro data to anticipate, pre-
`dict, or explain clinical pharmacokinetic
`drug interactions was first described
`by Rowland and Matin in 1973, in the
`context of the inhibition of tolbutamide
`clearance by coadministration of sulfa-
`phenazole.1 The core of the model was
`what is now commonly termed “[I] over
`Ki”—the ratio of inhibitor exposure in
`vivo ([I]) divided by an in vitro inhibi-
`tion constant (Ki) that reflects (in recip-
`rocal fashion) the quantitative potency
`of the inhibitor. The more [I] exceeds
`
`Ki, the greater is the [I]/Ki ratio, and the
`greater is the probability and/or magni-
`tude of a clinical pharmacokinetic DDI
`caused by the perpetrator’s (e.g., sulfa-
`phenazole) inhibition of clearance of the
`victim (e.g., tolbutamide). Rowland and
`Matin at that time also pointed out the
`importance of fm—the fraction of the
`dose metabolized via the target path-
`way—as a modulator of the predictive
`validity of the [I]/Ki ratio.1
`Clinical and scientific interest in
`DDIs intensified in the late 1980s and
`
`1Department of Molecular Physiology and Pharmacology, Tufts University School of Medicine, Boston,
`Massachusetts, USA. Correspondence: DJ Greenblatt (DJ.Greenblatt@Tufts.edu)
`
`doi:10.1038/clpt.2013.230
`
`CliniCal pharmaCology & TherapeuTiCs | VOLUME 95 NUMBER 2 | FEBRUARY 2014
`
`
`
`
`
`
`
`133
`
`1
`
`TEVA1023
`
`

`

`perspectives
`
`B = Buspirone
`N = Nifedipine
`M = Midazolam
`A = Alprazolam
`T = Triazolam
`S = Simvastatin
`
`Systemic unbound Cmax
`
`0.5 1
`
`2
`
`4
`
`7 10
`
`20
`
`40
`
`70
`
`70
`40
`20
`10
`
`7 4 2 1
`
`0.5
`
`Hepatic inlet unbound Cmax
`
`0.5 1
`
`2
`
`4
`
`7 10
`
`20 40
`
`70
`
`70
`40
`20
`10
`
`7 4 2 1
`
`0.5
`
`Systemic total Cmax
`
`1
`2
`4
`7 10
`20
`Hepatic inlet total Cmax
`
`40 70
`
`70
`40
`20
`10
`
`7 4 2 1
`
`AUCl /AUC0 observed
`
`0.5
`0.5
`
`70
`40
`20
`10
`
`7 4 2 1
`
`AUCl /AUC0 observed
`
`0.5
`
`0.5
`
`1
`
`2
`
`4
`
`7 10
`
`20 40 70
`
`AUCl /AUC0 predicted
`
`AUCl /AUC0 predicted
`
`Figure 1 Observed values of area under the curve (AUC)I/AUC0 ratios (explained in the text) from clinical drug–drug-interaction studies of six CYP3A substrate
`drugs (y-axis) vs. values predicted from the in vitro paradigm (x-axis), as described by Obach and associates.9 Four values of anticipated inhibitor exposure in vivo [I]
`are used in the prediction: systemic total maximum inhibitor concentration (Cmax; upper left); systemic unbound Cmax (upper right); hepatic inlet (portal) total Cmax
`(lower left); and hepatic inlet (portal) unbound Cmax (lower right). See text and Table 1 for analysis of the data. Reprinted from ref. 8.
`
`early 1990s, coincident with the regula-
`tory and media attention attracted by
`the Seldane (terfenadine) affair. Predic-
`tive in vitro–in vivo DDI scaling models
`resurfaced,2–4 again based on the [I]/Ki
`concept from Rowland and Matin.
`The models did not work well, even
`after numerous refinements and modifi-
`cations described by many authors in the
`late 1990s and up to the late 2000s (see
`Supplementary References online). The
`determination of Ki in vitro—even for a
`specific inhibitor vs. a specific substrate—
`was subject to technical and interpretive
`bias and inaccuracy5 and did not neces-
`sarily reflect the susceptibility of the met-
`abolic enzyme to chemical inhibition in
`vivo. Most importantly, the value of [I]
`in the [I]/Ki ratio—still the cornerstone
`of all scaling models—should reflect the
`concentration of inhibitor at the site of
`metabolic enzyme activity in vivo. We can
`measure the total or unbound inhibitor
`levels in the systemic circulation, and we
`can guess at what might be more relevant
`concentrations (e.g., intra-enteric, total
`portal, or unbound portal concentra-
`tions), but we cannot actually measure
`the quantitative exposure of the enzyme
`to the inhibitor in vivo.6–8
`
`We previously evaluated8 the validity
`of a predictive model reported in 2006
`by Obach and associates9 for a series of
`42 observed-vs.-predicted DDI pairs
`for six different CYP3A substrates. The
`model was based (as in 1973) on the
`[I]/Ki concept along with fm, but with
`additional assumptions: bioavailability
`of the substrate across the gastrointes-
`tinal tract mucosa (Fg), intestinal-wall
`inhibitor concentration ([I]g), apparent
`first-order absorption rate constant (ka),
`fraction of the inhibitor passing through
`the intestine unchanged (Fa), enteric
`blood flow (Qg), and hepatic blood flow
`(Qh). IC50 was used as a surrogate for Ki.
`The observed quantitative DDI in vivo
`was expressed as area under the plasma
`
`concentration curve (AUC) for the sub-
`strate (victim) during coadministration
`of the inhibitor (AUCI) divided by the
`corresponding AUC in the control state
`(AUC0).8 The predicted quantitative
`DDI was calculated from the model,
`using four possibilities for [I]: total sys-
`temic plasma Cmax, unbound systemic
`Cmax, total portal (hepatic inlet) Cmax,
`and unbound portal Cmax. Observed
`and predicted AUC ratios were plot-
`ted using logarithmic axes for clarity
`(Figure 1).
`Based on linear regression analy-
`sis of log-transformed values, all four
`[I] options yielded r2 values in a simi-
`lar range, with the most variability
`explained using the total systemic Cmax
`
`Table 1 Observed vs. predicted drug interactions for CYP3A substrates
`
`observed vs. predicted interaction
`
`Overall r²
`
`Observed > predicted
`
`Predicted > observed
`
`Percent differing by >50%
`
`Based on 42 data pairs from Table 9 in ref. 9.
`
`maximum inhibitor concentration
`
`systemic
` total
`
`systemic
` unbound
`
`portal total
`
`portal
`unbound
`
`0.75
`
`57%
`
`38%
`
`19%
`
`0.66
`
`76%
`
`24%
`
`24%
`
`0.67
`
`38%
`
`60%
`
`36%
`
`0.67
`
`57%
`
`38%
`
`19%
`
`134
`
`
`
`
`
`
`
`VOLUME 95 NUMBER 2 | FEBRUARY 2014 | www.nature.com/cpt
`
`2
`
`

`

`option (r2 = 0.75) (Table 1). Systemic
`unbound Cmax yielded a high fraction
`of underpredicted values, while total
`portal Cmax yielded a high fraction of
`overpredicted values. Unbound portal
`Cmax was no better than total systemic
`Cmax, either in overall r2, the frequency
`of under- and overprediction, or the per-
`centage of pairs for which observed and
`predicted values differed by more than
`50% (Table 1, Figure 1). Our conclusion
`at the time8 was that reasonable predic-
`tive accuracy was not achieved, and that
`no other estimate of [I] improved on that
`based simply on total systemic Cmax.
`The most recent iteration of CYP3A
`DDI prediction is described by Vieira and
`associates in this issue.10 Some data points
`from the 2006 paper9 are shared, and
`other data points were added (some of
`which come from regulatory submissions,
`with perpetrators not identified, and data
`not available to the public). The predictive
`model is more complex and refined, and
`it includes additional parameters that are
`measured or assumed: the intraluminal
`gastrointestinal concentration ([I]gut),
`the unbound in vitro inhibition constant
`(Ki,u), the unbound inhibitor concentra-
`tion causing half-maximal inactivation
`(KI,u), the maximal inactivation rate con-
`stant (kinact), and the enzyme degradation
`rate constant (kdeg). The last three named
`parameters are connected to perpetra-
`tors presumed to cause time-dependent
`(mechanism-based) inhibition. When
`induction is coincident with inhibition,
`the induction component is accounted for
`with an approach similar to that of Einolf
`and associates (described in this issue).11
`Figures 1–3 in the paper by Vieira and
`associates10 are disheartening, especially
`when the y = x lines (not drawn by the
`authors) are drawn in. The deviation of
`
`observed from predicted is extensive, and
`major overprediction is the rule. Model
`validity does not look improved since
`2006. We are not there yet.
`From a regulatory standpoint, it could
`be argued that the principal objective
`of DDI prediction should be the avoid-
`ance of false negatives—real clinical
`DDIs not predicted by the model. If
`so, major overprediction by the model
`protects the public, and the model is a
`“success.” But this is balanced by the
`high prevalence of negative clinical
`DDI studies, with the associated low,
`but still nonzero, risk to DDI study par-
`ticipants, as well as the cost burden to
`the drug development process which
`is passed on to the health-care system.
`From a scientific standpoint, we seem
`to be going in the wrong direction;
`model validity is not matching model
`complexity. Because more complex
`approaches are not leading to improved
`predictive capacity, we should look back
`to the core components of the scaling
`paradigm—the same [I] and Ki that
`Rowland’s group identified 40 years
`ago—rather than pursue increasingly
`complex models as proposed in cur-
`rent regulatory guidance. Improved
`validity of prediction may well be
`achieved through molecular physiology
`approaches to determining the inhibitor
`concentration that the enzyme actually
`“sees,” and a Ki value that reflects the
`effect of the inhibitor on the metabolic
`enzyme as it functions in vivo.
`
`SUPPLEMENTARY MATERIAL is linked to the
`online version of the paper at http://www.nature.
`com/cpt
`
`CONFLICT OF INTEREST
`The author is a scientific consultant to the Florida
`Department of Citrus, Lake Alfred, Florida.
`
`perspectives
`
`© 2014 ASCPT
`
`4.
`
`1. Rowland, M. & Matin, S.B. Kinetics of drug–drug
`interactions. J. Pharmacokinet. Biopharm. 1,
`379–385 (1973).
`2. Resetar, A., Minick, D. & Spector,
`T. Glucuronidation of 3ʹ-azido-3ʹ-
`deoxythymidine catalyzed by human liver
`UDP-glucuronosyltransferase. Significance of
`nucleoside hydrophobicity and inhibition by
`xenobiotics. Biochem. Pharmacol. 42, 559–568
`(1991).
`3. Tucker, G.T. The rational selection of drug
`interaction studies: implications of recent
`advances in drug metabolism. Int. J. Clin.
`Pharmacol. Ther. Toxicol. 30, 550–553 (1992).
`von Moltke, L.L., Greenblatt, D.J., Duan, S.X.,
`Harmatz, J.S. & Shader, R.I. In vitro prediction of
`the terfenadine–ketoconazole pharmacokinetic
`interaction. J. Clin. Pharmacol. 34, 1222–1227
`(1994).
`5. Greenblatt, D.J., Venkatakrishnan, K., Harmatz,
`J.S., Parent, S.J. & von Moltke, L.L. Sources of
`variability in ketoconazole inhibition of human
`cytochrome P450-3A in vitro. Xenobiotica 40,
`713–720 (2010).
`6. Greenblatt, D.J. & von Moltke, L.L. Clinical
`studies of drug–drug interactions: design and
`interpretation. In Enzyme and Transporter-Based
`Drug–Drug Interactions: Progress and Future
`Challenges (eds. Pang, K.S., Rodrigues, A.D. &
`Peter, R.M.) 625–649 (Springer, New York, 2010).
`von Moltke, L.L. & Greenblatt, D.J. Clinical
`drug interactions due to metabolic inhibition:
`prediction, assessment, and interpretation.
`In Enzyme Inhibition in Drug Discovery and
`Development (eds. Lu, C. & Li, A.P.) 533–547 (John
`Wiley & Sons, Hoboken, NJ, 2010).
`8. Greenblatt, D.J., He, P., von Moltke, L.L. & Court,
`M.H. The CYP3 family. In Cytochrome P450: Role in
`the Metabolism and Toxicology of Drugs and Other
`Xenobiotics (ed. Ioannides, C.) 354–383 (Royal
`Society of Chemistry, Cambridge, UK, 2008).
`9. Obach, R.S., Walsky, R.L., Venkatakrishnan, K.,
`Gaman, E.A., Houston, J.B. & Tremaine, L.M. The
`utility of in vitro cytochrome P450 inhibition data
`in the prediction of drug–drug interactions. J.
`Pharmacol. Exp. Ther. 316, 336–348 (2006).
`10. Vieira, M.d.L.T. et al. Evaluation of various static
`in vitro–in vivo extrapolation models for risk
`assessment of the CYP3A inhibition potential of
`an investigational drug. Clin. Pharmacol. Ther. 95,
`189–198 (2014).
`11. Einolf, H.J. et al. Evaluation of various static and
`dynamic modeling methods to predict clinical
`CYP3A induction using in vitro CYP3A4 mRNA
`induction data. Clin. Pharmacol. Ther. 95, 179–188
`(2014).
`
`7.
`
`CliniCal pharmaCology & TherapeuTiCs | VOLUME 95 NUMBER 2 | FEBRUARY 2014
`
`
`
`
`
`
`
`135
`
`3
`
`