CLINICAL TRIALS AND TRANSLATIONAL MEDICINE
`COMMENTARIES
`
`Lead PK Commentary: Predicting Human Pharmacokinetics
`
`Malcolm Rowland,1 Leslie Z Benet2
`
`1School of Pharmacy and Pharmaceutical Sciences, University of Manchester, Manchester M13 9PT, UK
`
`2Department of Bioengineering and Therapeutic Sciences, School of Pharmacy, University of California San Francisco,
`San Francisco, California 94143
`
`Received 6 May 2011; accepted 6 May 2011
`
`Published online 31 May 2011 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/jps.22637
`
`Keywords: 2-dimensional gel electrophoresis; 5-Aminolevulinic Acid; Ab initio calculations;
`ABC transporters; Absorption
`
`Selecting the best compounds, often from among
`many, to bring forward for evaluation in humans is a
`difficult and challenging task. Recognizing this prob-
`lem, the Pharmaceutical Research Manufacturers As-
`sociation (PhRMA) established a Pharmaceutical In-
`novations Steering Committee (PISC) with the aim
`of improving the chances of success. In addition to
`the two task forces focusing on safety and efficacy,
`another focused on prediction of human pharmacoki-
`netics (PK). Published in this issue of the Journal, in
`the form of five papers, is the outcome of the research
`undertaken by this last group. Although there have
`been substantial improvements in our understand-
`ing of factors controlling the PK of a compound, such
`that it is often now not the primary reason for fail-
`ure during clinical drug development, there is still
`a great need to improve our prediction methods for
`several reasons. One is the need to reduce the waste
`of resources (material, time, and cost) spent unnec-
`essarily on poor compounds that could have other-
`wise been better spent. Another is the ethical one of
`exposing animals to distress and trauma with poor
`compounds that will be dropped subsequently during
`development. A third is that better prediction of hu-
`man PK will not only aid in the better estimate of the
`doses needed to ensure adequate systemic exposure of
`active principle in first-in-human (FIH) studies, but
`also subsequently facilitate the improved ease of de-
`velopment and optimal therapeutic use of compounds.
`There are several unique features of this study by
`PhRMA. It is the first wherein PhRMA member com-
`
`Correspondence to: Malcom Rowland (E-mail: mrow190539@
`aol.com)
`Journal of Pharmaceutical Sciences, Vol. 100, 4047–4049 (2011)
`© 2011 Wiley-Liss, Inc. and the American Pharmacists Association
`
`panies collectively provided nonclinical and FIH PK
`data on low molecular weight lead clinical candidate
`compounds (108 following oral administration in hu-
`mans, of which 19 had also been given intravenously),
`exhibiting a wide array of physicochemical and struc-
`tural properties. The data were then anonymized by
`PhRMA prior to data analysis. Furthermore, the hu-
`man data were initially withheld from the data ana-
`lyst, and only made available after all the predictions
`of human PK had been made, thereby mimicking the
`prospective situation faced by drug developers, while
`minimizing the likelihood of bias. This approach con-
`trasts with that employed in the vast majority of
`previous studies in which the investigator knew the
`human data and tested, and sometimes “adjusted,”
`the prediction methodology retrospectively. Further-
`more, there are a multitude of published methodolo-
`gies, each often claiming to be superior to others,
`so a critical part of the current project was to com-
`prehensively evaluate all published methodologies on
`the same data set, an onerous task. Finally, the com-
`pounds that comprised the current study are those
`under recent development (as only these have all the
`necessary physicochemical, in vitro, as well as in vivo
`data), in contrast to many published reports that eval-
`uated decades old compounds. This is confirmed by
`the distribution of PhRMA compounds within the Bio-
`pharmaceutics Classification System, with 15%, 57%,
`12%, and 23% in class 1, 2, 3, and 4, respectively,
`compared with 39%, 31%, 23%, and 7% for marketed
`products cited by the authors. In keeping with this
`distribution, the majority of PhRMA compounds were
`primarily eliminated by metabolism. Accordingly, the
`current findings are likely to be more relevant to those
`engaged in drug development today.
`
`JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 10, OCTOBER 2011
`
`4047
`
`Apotex v. Novartis
`IPR2017-00854
`NOVARTIS 2103
`
`

`

`4048
`
`ROWLAND AND BENET
`
`Broadly, the prediction methodologies have been di-
`vided into three areas: allometry, in vitro–in vivo ex-
`trapolation (IVIVE), and physiologically based phar-
`macokinetics (PBPK). Allometry, involving scaling to
`humans PK data in test animals (often rat and dog,
`and in this study also frequently in mouse and non-
`human primate), based on body size, has a long his-
`tory and is the mainstay approach by industry. IVIVE,
`which has been applied to predict a parameter, such as
`hepatic metabolic clearance, based on human-derived
`tissues or expression systems, is more recent and aims
`to address particularly the known species differences
`in active processes. The latest newcomer to see in-
`creasing application is PBPK, which incorporates a
`combination of physicochemical and in vitro biological
`data of a compound into a whole body physiologically
`based model, comprising independent data on tissue
`size, composition, blood flow, and physiologic function,
`to predict the temporal pattern of a compound in body
`fluids and tissues. Animal data are not inherently re-
`quired in PBPK prediction of human PK, although
`development and verification of the methodology are
`often undertaken in animals. IVIVE may be regarded
`as a component of PBPK.
`The findings of the project are illuminating. They
`show that although some prediction methods are def-
`initely better than others, there is currently no uni-
`versally outstanding method. Also, as expected, the
`prediction of human disposition kinetics following in-
`travenous administration is much better than the pre-
`diction of events following oral administration. In-
`deed, even the best methods could only predict events
`after oral administration, within a factor of two, in
`the order of 45% of the time, with a tendency to under
`predict the area under the curve, probably primar-
`ily due to an underprediction of oral bioavailability.
`There are likely to be many reasons for this failure, in-
`cluding the complex physiology of the gastrointestinal
`tract coupled with the complex processes occurring
`during absorption, especially following administra-
`tion of sparingly soluble compounds, with markedly
`different formulations often used in preclinical de-
`velopment than tested in humans. Thus, it would be
`expected that predictions should be better for class
`1, highly soluble–highly permeable compounds, as
`was found and reported in paper 5 for bioavailability
`predictions.
`Although popular, interanimal scaling allometri-
`cally is essentially empirical, which probably explains
`the myriad of modifications proposed by various in-
`vestigators over the years, and evaluated in the cur-
`rent project. The distinction is also brought out clearly
`between prediction of a parameter value, such as
`clearance or volume of distribution and, ultimately
`more important, prediction of a concentration–time
`profile, given that concentration over time drives re-
`sponse, even though originally attention was given
`
`to prediction of temporal events using the Dedrick
`plot and its modifications. Perhaps one reason for the
`more common evaluation of a parameter rather than
`a temporal profile is the lack of a universally accept-
`able method to evaluate similarity of shape. The au-
`thors of the project have addressed this problem in
`a novel way based on a series of reasonable criteria,
`which, although informative and pragmatic, may be
`questioned by those interested in statistical rigor, al-
`though some of us recall the many debates of a sim-
`ilar nature in bioequivalence testing, without clear
`resolution. That said, attempts to predict the con-
`centration–time profile allometrically were generally
`poor. Interestingly, when predicting human concen-
`tration–time profiles, no one preclinical species was
`found to be superior to the others nor were multi-
`ple species data shown to be superior to data from
`one species. Biologics aside, this finding raises the
`question as to the value of using non-human primates
`for such purposes. The results here also suggest that
`rodent data, usually available at an early stage of
`the process, for example, discovery, could be equally
`(but probably not sufficiently) as reliable as the data
`obtained in dog and monkey in the late candidate
`selection stage. However, considering the poor pre-
`dictability of bioavailability, one must recognize that,
`especially for the many poorly soluble drug candi-
`dates, the later stage candidate selection studies in
`large animals almost always involve formulations
`that are closer to that used in FIH studies than used
`in the discovery stage rodent studies.
`Physiologically based pharmacokinetics was eval-
`uated using a generic model. Perhaps surprisingly to
`some, on average, this methodology fared no better
`than allometry in predicting human PK, and some-
`times proved inferior. However, this needs to be put in
`perspective. Unlike allometry, which fundamentally
`can progress no further, PBPK, which is still in its in-
`fancy, is highly mechanistic, and current failures help
`highlight a lack of understanding of particular pro-
`cesses. Other advantages of this bottom-up approach
`are that it explains the observed in vivo behavior of
`drugs, requires minimal resources, including being
`animal sparing, and can be employed upstream, by
`linking processes to physicochemical and structure
`properties, to help guide the drug discovery teams in
`the better design of compounds. In addition, unlike
`allometry (with the possible exception of its use in pe-
`diatrics), PBPK models extend beyond dose selection
`in FIH studies, by helping to guide various aspects of
`clinical development, such as prediction of the impact
`of drug–drug interactions, age, disease, and so forth
`on PK, design of experiments, as well as subsequent
`therapeutic use of compounds.
`Although the PhRMA project will hopefully help in
`evidence-based choice of prediction methodologies, it
`has its limitations, which need to be kept in mind.
`
`JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 10, OCTOBER 2011
`
`DOI 10.1002/jps
`
`

`

`First, although the ranges of structures and physic-
`ochemical properties are quite wide, these undoubt-
`edly do not cover all chemical space encountered in
`drug development, so that companies concentrating
`in particular areas of chemical space will need to place
`the current findings into perspective. Although many
`companies provided all the expected in vitro physio-
`chemical and biological data, this was by no means
`universal, which reflects the state of acceptance of
`such data in decision-making. In the area of PBPK, a
`generic model was employed, which while very help-
`ful in gaining some insights does not incorporate all
`the many features found in commercial PBPK soft-
`ware, which if employed may affect the success rate
`of this methodology. In addition, in all cases evalu-
`ated, allometry, IVIVE, and PBPK, analysis and pre-
`diction dealt with mean data, whereas information
`regarding variability can be equally important when
`considering FIH studies, and beyond.
`So where are things likely to go from here? Despite
`its limitations, allometry is likely to remain the main-
`tained approach by industry for prediction of human
`PK in candidate selection for some years yet, although
`the benefit of using non-human primates is question-
`able. However, as our understanding of processes con-
`trolling PK improves, and the in vitro human bio-
`logic systems become increasingly more predictive of
`in vivo events, we will see the increasing adop-
`tion of the more mechanistic PBPK as the first-
`line approach. Pharmaceutical scientists are encour-
`aged to collaborate to progress these developments.
`Hopefully, the path taken by PhRMA, to act as an
`anonymized repository of data and as a facilitator of
`such collaboration, will be the forerunner of future
`developments in other areas of the pharmaceutical
`sciences.
`
`LEAD PK COMMENTARY
`
`4049
`
`Today, much is written and discussed in work-
`shops and consensus panels about the barriers in the
`drug approval process. Concerns are expressed about
`the money available, the discovery and development
`process itself, and the adequacy of the regulatory
`agencies. There is a great interest in developing mech-
`anisms for cooperative methodologies at the precom-
`petitive stage. We highly commend PhRMA and the
`scientists from the 12 companies, and the data ana-
`lysts, who carried out the present initiative and be-
`lieve that it serves as a model for other future ap-
`proaches. A barrier question was identified, that is,
`how useful are animal data in predicting human PK?
`Real data in multiple animal species for a large num-
`ber of drugs were assembled, anonymized, and evalu-
`ated in terms of the many methodologies proposed for
`using such data to predict human results. Then, these
`analyses were compared with the observed human PK
`findings. There were also analyses of the adequacy or
`lack thereof for in silico and in vitro predictive pa-
`rameters. Thus, PhRMA and the regulatory agencies
`now have very useful information about the adequacy/
`inadequacy of animal PK studies. But, in our opinion,
`of even greater credit to PhRMA and the 12 companies
`is their willingness to make this anonymized data set
`freely available. Here is the real potential for advanc-
`ing drug discovery and development science. Making
`the anonymized data generally available required a
`significant additional effort from the PhRMA scien-
`tists. When the barriers to improve the drug discovery
`and development process are discussed, the hurdles
`added by corporate legal policies need to be included
`together with money, scientific approaches, and regu-
`latory requirements. In this case, with much effort, a
`very valuable set of data has become available to the
`pharmaceutical sciences community.
`
`DOI 10.1002/jps
`
`JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 10, OCTOBER 2011
`
`