Welcome to the joint Pharmacokinetics UK 2005 and Rosenön Meeting

Wednesday 23rd – Friday 25th November, 2005

De Vere Grand Hotel King’s Road Brighton BN1 2FW

Programme and Abstract Book

Published with the support of John Wiley & Sons Ltd, publishers of the journal BIOPHARMACEUTICS AND DRUG DISPOSITION

Wednesday 23rd November

12.30 Arrival, Coffee & Buffet Welcome & Session 1: Pharmacogenetics on PK/PD 13:50 Welcome (Steve Toon & Amin Rostami) 14:00 Introduction to the First Session: Geoff Tucker and Margareta Hammarlund-Udenaes 14:05 Geoff Tucker Pharmacogenetics on PKPD – expectations and reality 14:35 Duncan McHale Applications of Pharmacogenomics to the drug discovery and development pipeline 15:05 Leif Bertilsson Interethnic differences in drug disposition 15.30 Coffee break Viewing Posters & Exhibitions 16:00 Munir Pirmohamed Pharmacogenetics of adverse drug reactions Simulations as a tool to assess the propagation of genetic polymorphisms in drug Gemma Dickinson 16:20 metabolising enzymes into PK and PD outcomes 16:40 Marja-Liisa Dahl Pharmacogenetics in PK and PD of warfarin 17:30 Break 18:30 Poster Session – Free Bar 20:30 Dinner th Thursday 24 November

Session 2: Microdosing in Drug Development

09:00 Introduction to Second Session: Anders Grahnen and Steve Toon 09:10 Colin Garner Human PK studies on microdoses of drugs - scientific and regulatory perspectives Seeing through the MIST: abundance versus percentage commentary on metabolites 09:50 Dennis Smith safety testing 10:30 Coffee break Viewing Posters & Exhibitions 11:00 Mats Bergstrom PET imaging as a biomarker in discovery medicine – pros and cons 11:40 Nenad Sarapa Microdosing in clinical drug development: recommendations and debatable issues 12:30 Lunch

Session 3: Advances in PK/PD Modelling

14:00 Introduction to Third Session: Peter Milligan and Mats Karlsson 14:10 Rune Overgaard Stochastic Differential Equations in PKPD Modelling Adrian Dunne & 14:50 Modelling binary data with inter-subject variability Valda Murphy 15:30 Coffee break Viewing Posters & Exhibitions Comparison using simulated data and blinded analysis of approximate and exact 15:45 Pascal Girard likelihood parametric estimation methods for population PK model 16:25 Andrew Hooker New diagnostics for the FO/FOCE methods in NONMEM 17:00 Break 20:00 PKUK Banquet th Friday 25 November

Open Session

08:45 Introduction to Open Session: Alison Thomson and Siv Jonsson 08:50 Monica Edholm Current (and future) role of modeling and PK/PD analysis in regulatory assessments 09:20 Esther Schmid Innovation in the Pharmaceutical Industry – Myths versus Reality Increasing the Strategic Value of Kinetic and Dynamic Data Through Unified Modeling, 09:50 Simon Davis Higher Productivity, and Regulatory Compliance 10:15 Coffee break Viewing Posters & Exhibitions Plasma protein binding: prediction of albumin levels in the adult population and ligand 10:30 Mark Baker displacement Prediction of time-dependent CYP3A4 drug-drug interactions - impact of enzyme 10:50 Aleksandra Galetin degradation and intestinal inhibition 11:10 Shuying Yang Evaluation of Methods for PK Modeling with Data below Quantification Limit 11:30 Radojka Savic Evaluation of a nonparametric estimation method in NONMEM VI Optimal design for first in human studies to investigate the pharmacokinetic/ 11:50 Ivelina Gueorguieva pharmacodynamic behaviour of TGF-beta RI kinase inhibitor Lorea Bueno Mechanistic PK/PD modelling for signal transduction modulators. Application to TGF-b RI 12:10 Burgos antagonists 12:30 Final conclusions and Closing remarks, Buffet Lunch

The PKUK Organising Committee wishes to express their thanks to Mr Ben Meakin, Mrs Dolores Wellens, and the Swedish Academy of Pharmaceutical Sciences for their role in the administration of advertising, registration, and organising this year’s meeting.

Poster Session Titles (Wednesday 23rd, 18.30)

Please note that for the Poster Session (not coffee breaks), the presenters should be by their respective posters to answer questions.

The application of structural identifiability analysis to the design of pharmacokinetic experiments, S. Y. A Cheung, L. Aarons, I. Gueorguieva, N. D. Evans, K. R. Godfrey and M. J. Chappell.

The influence of experimental methods on the estimation of kinact and KI values of a time-dependent inhibitor. L. M. Van, J. Hargreaves, K. R. Yeo, G. T. Tucker, and A. Rostami-Hodjegan.

A retrospective PBPK approach to assess Lung metabolism for SR, E. Manolis, P. Delrat, and T. Shepard.

Comparative Pharmacokinetics of Five Drugs Administered to Humans at a Therapeutic and Microdose, G. Lappin and R. C. Garner.

CYP450 Reductase Knockout Mice – A Useful Tool in Understanding Drug Disposition? A. O’Hara, G. Hughes, H. Meigh, C. Tyman, and K. Watson.

Estimation of dosing strategies for cefuroxime by minimisation of a risk function, A. Viberg, O. Cars, M. O. Karlsson, and S. Jönsson.

Prediction of Xenobiotic Clearance in Humans: In vitro – in vivo extrapolation (IVIVE) vs allometric scaling (AS). M. R. Shiran, N. J. Proctor, E. M. Howgate, K. R. Yeo, G. T. Tucker, and A. Rostami-Hodjegan.

Enterohepatic recirculation of norfloxacin – a whole body physiologically based pharmacokinetic model, C. L. Rodrigues, I. Gueorguieva, M. Channel, and D. A. Berk.

Cure Rate Models in Depression Trials, G. Santen, R. Gomeni, M. Danhof, and O. Della Pasqua.

A Critical Evaluation of Experimental Designs used in studies of Mechanism-Based Enzyme Inhibition (MBI) with Implications for In Vitro-In Vivo Extrapolation (IVIVE). F. Ghanbari, K. R. Yeo, M. S. Lennard, G. T. Tucker, and A. Rostami-Hodjegan.

A Single Dose Study to Investigate the Effect of the CCR5 Antagonist maraviroc on the QTc Interval in Healthy Subjects, J. Davis, F. Hackman, D. Sudworth, and G. Weissgerber.

Identification of hOCT2 substrates and prediction of renal clearance in vivo using stably transfected cell lines, N. Attkins, I. Gardner, and K. Duffy.

Genetic Algorithms and Their Applications in PK/PD Data Analysis. M. Jamei, J. Yang, K. R. Yeo, G. T. Tucker, and A. Rostami-Hodjegan.

Inter-Individual Variability in the Catalytic Activity of CYP3A4 per unit of Enzyme (kcat). Y. Lei, Z. E. Wilson, K. H. Crewe, G. T. Tucker, and A. Rostami-Hodjegan.

The Quantification of Residual Blood in Excised Tissue A. M. Quigley and M. B. Baker

Inter- and intra-individual variability in gastro-intestinal physiology has significant effects on the prediction of the fraction of dose absorbed (fa). M. Jamei, J. Yang, A. Rostami-Hodjegan, and G. T. Tucker.

Building Physiologically-Based Pharmacokinetic Models in NONMEM using the PRIOR functionality. G. Langdon, I. Gueorguieva, L. Aarons, and M. Karlsson.

Dynamic Systems as Intractable Systems of Differential Equations: Design of Experiments. J. Cornebise and B. Boulanger.

Prediction of biliary clearance using Caco-2- a future screening assay? M. Baker, L. King, H. Hailu, T. Parton, S. Bartlett, and S. Carter.

Presentation Abstracts

Session 1: Pharmacogenetics on PK/PD Wednesday 23rd November (PM)

Pharmacogenetics: Expectations and Reality

Geoffrey Tucker

Academic Unit of Clinical Pharmacology University of Sheffield

Genomic medicine has generated high expectations with regard to the advent of ‘personalised medicine’. This has been fuelled by the pace of technological advances in genotyping, enthusiasts extrapolating beyond small proof of principle and retrospective studies, and a few apparent success stories. However, drug response is a complex function of genetic, environmental and behavioural factors. Given that genetic factors need to be put into perspective, the challenge is to assemble large, prospective, multidisciplinary projects to assess the real clinical value of predictive genetic testing in drug therapy. This presentation will focus on: current evidence for the significance of genetic variation response; factors that enhance the cost- effectiveness of pharmacogenetic testing; criteria for the conduct of prospective studies to develop genetic tests of drug response; and potential constraints on implementing the ‘promise of precise prescriptions’.

Tucker, GT: Pharmacogenetics – expectations and reality. Br Med J 329: 4-6, 2004.

Applications Of Pharmacogenomics To The Drug Discovery And Development Pipeline

Duncan McHale

Pfizer Limited

Following the success of the late 80s and early 90s the last few years has seen rising failure rates, a marked reduction in the number of drug approvals and a sharp rise in the cost of drug development. This combined with patent expiries and external pricing pressures has led to a critical need to improve the overall survival rate of compounds and reduce the time and resources spent on compounds which will not survive. Pharmacogenomics is the application of genomic technologies to the drug discovery and development process. There is considerable expectation around this emerging science but the demonstrable value to date has been limited. Whilst the drive to develop more personalised or precise medicines is often discussed near term value to the drug portfolio is also required. Pharmacogenomics is one of the tools which can be used to help in industry’s endeavour to overcome the current challenges. Examples of where Pfizer has used pharmacogenomics to aid target choice and early decision making creating near term value will be presented.

Interethnic differences in drug disposition

Leif Bertilsson

Professor, Clinical Pharmacology at Karolinska Institutet Karolinska University Hospital

At a given dose of such drugs as antidepressants and neuroleptics, the therapeutic response is highly variable. To a major extent this is due to variation between patients in drug metabolism and pharmacokinetics. Many antidepressants and most neuroleptics are metabolized by the polymorphic cytochrome P450 enzyme CYP2D6 (review by Bertilsson et al 2002). Seven % of Caucasians are poor metabolisers (PM) mainly due to the presence of the CYP2D6*4 allele encoding no CYP2D6 enzyme. In Asians and Africans there are two SNPs, CYP2D6*10 and 17, respectively coding for enzymes with decreased activity. Interestingly the CYP2D6*17 allele encodes an enzyme with decreased rate of metabolism of debrisoquine and dextromethorphan, but not of the two other CYP2D6 substrates codeine or metoprolol (Wennerholm et al 2002).

We discovered a new principle in drug metabolism ultrarapid metabolism due to CYP2D6 gene duplication and multiduplication. There is pronounced interethnic variation in the frequency of CYP2D6 gene duplication: 1 % in Sweden, 10 % in Italy and Spain and as much as 29 % in Ethiopia. We recently showed that in nonresponders to antidepressant therapy in Sweden, 10 % had a CYP2D6 gene duplication (Kawanishi et al 2004), which is a 10-fold increase compared to the general Swedish population. Thus ultrarapid metabolism seems to be one important factor for antidepressant non-response.

Another polymorphic enzyme CYP2C19 catalayzes the N-demethylations of citalopram, clomipramine and diazepam. In Caucasians 3 % are PM of CYP2C19 probe drugs e.g. omeprazole, while as many as 12-20 % of Asians are PM. This higher incidence is due to the presence of CYP2C19*3 in addition to the CYP2C19*2 present in all populations. Japanese patients had higher levels of clomipramine compared to Swedish patients due to the high incidence of PM and heterozygotes in the former compared to the later group (Shimoda et al 1999). Most neuroleptics e.g. haloperidol, perphenazine and risperidone are metabolized by CYP2D6, but not clozapine (Jerling et al 1994). We found that the plasma concentration of clozapine increased several-fold when fluvoxamine was added to the treatment. Later it has been confirmed that clozapine is metabolized by the tobacco inducible CYP1A2.

For drug metabolism in total CYP3A4 is one of the most important enzymes. It has a sister enzyme CYP3A5, which has similar substrates, inhibitors and inducers as CYP3A4. CYP3A5 is highly polymorphic. Among 136 healthy Swedes only 17 expressed CYP3A5 (1 with CYP3A5*1/*1 and 16 with CYP3A5*1/*3) (Mirghani et al 2005). In 143 Tanzanians, however, as many as 38 and 68 subjects had 2 and 1 functional CYP3A5 alleles, respectively, when genotyping for *3, *6 and *7. In these Tanzanians there was a relationship between the metabolism of the CYP3A4/5 substrate quinine and the CYP3A5 genotype. CYP3A5 seems to be an important enzyme for certain populations i.e. black Africans, but not in others like Europeans.

References: Bertilsson, L., et al. (2002). Br J Clin Pharmacol, 53, 111-122. Jerling, M., et al. (1994). Ther Drug Monit, 16, 368-74. Kawanishi, C., et al. (2004). Europ J Clin Pharmacol, 59, 803-807. Mirghani, R., et al. (2005) Pharmacogenetics and Genomics, submitted. Shimoda, K., et al. (1999). J Clin Psychopharmacol, 19, 393-400. Wennerholm, A., et al. (2002). Clin Pharmacol Ther, 71, 77-88.

Pharmacogenetics of adverse drug reactions

Munir Pirmohamed

Department of Pharmacology The University of Liverpool

Adverse drug reactions (ADRs) are a major clinical problem accounting for 6.5% of hospital admissions. Clearly not all these ADRs are due to genetic predisposition, and many of these will be preventable through improved prescribing practices. Nevertheless, the occurrence of ADRs within families together with evidence suggesting that many of these ADRs may be due to metabolism by polymorphic enzymes, implies that pharmacogenetics may have an important role in improving drug safety. Genetic variation predisposing to ADRs may reside in the pharmacokinetic pathways (absorption, distribution, metabolism and excretion) or in the pharmacodynamic targets (receptors, ion channels, enzymes, etc.). For type A ADRs (predictable ADRs with a readily discernible dose-response relationship), it may be possible to use pharmacogenetics to guide drug dosage and choice, while for type B (idiosyncratic ADRs without a clear dose-response relationship), pharmacogenetics may guide drug choice. While individual SNPs that predispose to certain type A ADRs have been identified with a number of drugs, they have not been of sufficient predictive value to be incorporated into clinical practice. A typical example here is the variability in anticoagulation with warfarin – CYP2C9 polymorphisms are known to affect dose requirements, but have a low predictive value. Interestingly, even with type A reactions, variability in pharmacodynamic targets may have a greater quantitative effect on drug response, as has recently been shown with warfarin and VKORC1 polymorphisms. By contrast, with idiosyncratic ADRs, major gene effects have recently been identified for both abacavir hypersensitivity and carbamazepine-induced Stevens-Johnson syndrome. Importantly, these remarkable findings have been achieved with small numbers of patients, and with regard to abacavir, it may also be cost-effective to genotype patients prior to its use in HIV positive patients. There are likely to be many more advances in this area over the next few years, and pharmacogenetics thus may have huge potential in improving the safety of old drugs, as well as of new drugs currently in development.

Simulations as a tool to assess the propagation of genetic polymorphisms in drug metabolising enzymes into PK and PD outcomes: Warfarin and dextromethorphan as examples

Gemma Dickinson, Martin Lennard, Geoffrey Tucker, Amin Rostami-Hodjegan

Academic Unit of Clinical Pharmacology University of Sheffield

Background: Due to an increase in interindividual variability in response compared to the metabolism/PK of drugs it is expected that most studies of the impact of genetic polymorphism in drug metabolism will be more successful in assessing PK rather than PD outcome (e.g. dextromethorphan (DEX)). However, some studies have linked PD to enzyme phenotype/genotype but fail to establish a relationship between phenotype/genotype and PK (e.g. warfarin). A simulation approach may help to clarify the reasons for these contrasting outcomes.

Methods: Simcyp® algorithms (www.simcyp.com) were used to simulate virtual populations with respect to: (1) the PK of DEX and its antitussive response in CYP2D6 phenotypes, and (2) the relationship between CYP2C9 genotype and the PK and PD of S-warfarin. To mimic the design of studies reported in the literature, the populations in case (1) were enriched with the poor metaboliser phenotype, and, in case (2), subjects unselected for CYP2C9 genotype were studied.

Results: While 5 subjects of each phenotype were required to achieve 80% power to discriminate the PK of DEX between extensive and poor metabolisers (EMs and PMs respectively), the corresponding number to detect the difference in antitussive effects was >1000. With a study size of 550, the power to detect a difference in warfarin clearance between CYP2C9 wild type (*1/*1) and some of the less frequent genotypes was higher than that for the more frequent genotypes (e.g. 90% power for *2/*3 vs 45% power for *1/*2). This is because of the combined effects of relative enzyme activity (*2/*3 = 40%; *1/*2 = 85% of wild type activity) and genotype frequency (*2/*3 = 1.4%; *1/*2 = 20.4% in Caucasians).

Conclusions: The simulations explain the failure of reported studies with regard to defining relationships between genetic polymorphism of drug metabolising enzymes and PK and PD. Integration of prior information on enzyme kinetics is helpful in optimising study design and in avoiding costly and unsuccessful clinical studies.

Pharmacogenetics in PK and PD of warfarin

Marja-Liisa Dahl

Department of Medical Sciences, Clinical Pharmacology Uppsala University, Sweden

Warfarin is the most widely prescribed oral anticoagulant for treatment and prevention of thromboembolic events. It is characterized by a narrow therapeutic range and a large interindividual variation in dose requirement, requiring close monitoring of effect through measurement of the protrombin time, expressed as the international normalized ratio (INR). Both over- and underdosing have severe clinical consequences in the form of increased risk of bleeding and thromboembolic events, respectively. Bleeding complications during warfarin therapy still remain one of the major causes of drug-related death in many western countries.

Warfarin is a racemic compound with complex pharmacokinetics and pharmacodynamics. Its more potent enantiomer, S-warfarin, is metabolised by the polymorphic cytochrome P450 CYP2C9, and significant associations between the CYP2C9 genotype and dose requirement as well as adverse effects of warfarin have been shown in many studies. Warfarin exerts its effect by reducing the regeneration of vitamin K from vitamin K epoxide, through inhibition of vitamin K epoxide reductase. Polymorphisms in the vitamin K epoxide reductase complex subunit 1 gene (VCORC1) have recently been shown to greatly influence warfarin dose requirement. Thus, variation in genes involved in both pharmacokinetics and pharmacodynamics of warfarin are of importance for the dosing and therapeutic outcome of the drug.

Apart from genetic variation, age, bodyweight, concomitant interacting drugs and treatment indication have been shown to influence warfarin dose requirement. A number of dosing algorithms and computer-generated dosing regimen have been developed with the aim to reduce the risk of overcoagulation during the initiation of warfarin therapy. Inclusion of genotype information in these algorithms offers the so far best documented example of the potential use of pharmacogenetic data in the clinic.

Session 2: Microdosing in Drug Development Thursday 24th November (AM)

Human PK studies on microdoses of drugs – scientific and regulatory perspectives

Colin Garner

CEO, Xceleron Ltd

Science only progresses as fast as new technologies and methods are developed and applied. The introduction of new enabling technologies by the pharmaceutical and biotechnology industry is essential if the costs of drug discovery and development are to be reduced and better, cheaper and safer medicines are to be developed. One enabling technology, which is being applied to problems of drug discovery and development is accelerator mass spectrometry (AMS). Although developed over 25 years ago for carbon dating, AMS has only recently been applied in biomedical research. The attraction of the AMS technique stems from its ultrasensitivity for the analysis of elemental isotopes which can be measured at the single atom level. 14C is the isotope of most interest to the pharmaceutical community since this can be incorporated into all organic molecules. AMS is up to one million times more sensitive than decay counting and hence radioactive doses in preclinical and clinical studies can be reduced to levels which are equivalent to background radioactive exposures. Historically AMS instruments have been large (up to two tennis courts in size) but recently smaller more compact instruments have become available. These smaller instruments will become widely used by the biomedical community over the coming years; a top five pharma company is currently installing such an instrument in their facility.

AMS has a range of applications and potential applications in biomedicine including (1) microdose human AME studies (2) conversion of high dose to low dose human radioactive mass balance studies (3) absolute bioavailability studies (4) metabolite profiling (5) AME studies of biomolecules including proteins, peptides and DNA (6) biomarker analysis including turnover of endogenous metabolic pools and bone turnover studies (7) clinical diagnosis and (8) toxicology studies of protein and DNA adducts. In my presentation I will focus on AMS applications in Human Phase 0 microdosing studies giving examples of how microdosing has been used to determine human PK. In addition I will present in summary form the results of the CREAM trial, an industry collaboration in which microdose and therapeutic dose PK was compared.

References

Barker J and Garner R C (1999) Biomedical applications of accelerator mass spectrometry – isotope measurements at the level of the atom. Rapid Commun Mass Spectrom, 13, 285-293.

Garner R C (2000) Accelerator mass spectrometry in pharmaceutical research and development – a new ultrasensitive analytical method for isotope measurement. Current Drug Metabolism, 1, 205- 213.

Garner R C, Barker J, Flavell C, Garner J V, Whattam M, Young G C, Cussans N, Jezequel, S and Leong D (2000) A validation study comparing accelerator MS and liquid scintillation counting for analysis of 14C-labelled drugs in plasma, urine and faecal extracts. J Pharm Biomed Analysis, 24, 197- 209.

Garner R C, Garner JV, Gregory S, Whattam M, Calam A and Leong D (2002) Comparison of the absorption of micronised (Daflon 500®mg) and non-micronised 14C-diosmin tablets after oral administration to healthy volunteers by accelerator mass spectrometry and liquid scintillation counting. J Pharm Sci, 91, 32-40.

Garner R C, Goris I, Laenen A A E, Vanhoutte E, Meuldermans W, Gregory S, Garner J V, Leong D, Whattam M, Calam and Snel C A W (2002) Mass balance and pharmacokinetic study – experience with 14C-(R)-6-[amino(4-chlorophenyl)(1-methyl-1h-imidazol-5-yl)methyl]-4-(3-chlorophenyl)-1- methyl-2(1h)-quinolinone (R115777), a farnesyl transferase inhibitor, Drug Metab Disposn, 30, 823- 830.

Lappin G and Garner R C (2003) Big physics, small doses: the use of AMS and PET in human microdosing of development drugs. Nature Reviews – Drug Discovery, 2, 233-240.

Lappin G and Garner R C (2004) Current perspectives of 14C-isotope measurement in biomedical accelerator mass spectrometry. Anal Bioanal Chem, 378, 356-364.

Sarapa N, Hsyu P H, Lappin G and Garner R C (2005) The application of accelerator mass spectrometry to absolute bioavailability studies in humans: simultaneous administration of 14C- nelfinavir mesylate solution and oral nelfinavir to healthy volunteers. J Clin Pharmacol, 45, 1198- 1205.

Williams P, Zannikos P, Weaner L, Stubbs R J, Van Marle S P, Garner R C and Hutchison J (2002) Accelerator mass spectrometry (AMS) and ‘microdosing’ to evaluate the clinical pharmacokinetics (PK) of a new agent. Clin Pharm Therap, 71, p75 Abs TOIII-B-4.

Seeing through the MIST: Abundance Verses Percentage. Commentary on Metabolites in Safety Testing

Dennis Smith and R. Scott Obach

Pharmacokinetics, Dynamics, and Metabolism, Pfizer Limited

Recent attention has been given to the potential roles that metabolites could play in safety evaluations of new drugs. In 2002, a proposal was published on "metabolites in safety testing" ("MIST") [T. A. Baillie, M. N. Cayen, H. Fouda, R. J. Gerson, J. D. Green, S. J. Grossman, L. J. Klunk, B. LeBlanc, D. G. Perkins, and L. A. Shipley (2002) Toxicol Appl Pharmacol 182:188–196], which suggested some guidelines regarding when it is necessary to provide greater assessment of the safety of metabolites. However, this proposal was based on relative abundance values, i.e., the percentage that a metabolite comprises of total exposure to drug-related material. In the present commentary, we propose that absolute abundance criteria be used rather than relative abundance. The absolute abundance of a metabolite in circulation or excreta in humans should be combined with other information regarding the chemical structure of the metabolite (e.g., similarity to the parent drug, presence of chemically reactive substituents) and potential mechanisms of toxicity (e.g., suprapharmacological effects, secondary pharmacological effects, nonspecific effects). Decision trees are described that can be used to address human metabolites in safety testing.

PET imaging as a biomarker in discovery medicine – pros and cons

Mats Bergstrom

No abstract submitted.

Microdosing in clinical drug development: recommendations and debatable issues

Nenad Sarapa

Translational Medicine & Clinical Pharmacology Daiichi-Sankyo, US

Introducing innovative ways of doing first-in-human studies earlier is currently the most attractive option for pharmaceutical companies to optimizing drug development by improving the quality and speed of drug selection and reducing attrition in later phases of clinical development.

Selection of compounds in drug discovery and preclinical development is undergoing rapid changes coupled with a shift in emphasis in the early clinical development. Where appropriate, the registration-driven classic Phase 1 study supported by the “full IND package” could be preceded by scientifically and mechanistically based screening studies where early generation of human data would drive selection between optimized leads (‘Phase 0’ studies). First-in-human studies with extremely low single doses (‘microdoses’) or with low pharmacologically active doses (‘Exploratory IND’) of one or several lead candidates, supported by highly sensitive analytical methods, may accelerate and improve internal decision making before formal Phase 1 studies even begin. Of late, the regulatory agencies accept that microdosing and Exploratory IND studies be supported by abbreviated nonclinical safety and CMC data because the risk to humans is considered commensurate with the very low exposure and short duration of treatment. The EMEA Position Paper on microdosing (2004) and the US FDA Draft Guidance on Exploratory IND (2005) define the design of microdosing and other screening human studies and prescribe the requirements for nonclinical supporting data that are less comprehensive than ICH M3 and even allow customized non-GMP approaches. When introduced into wider clinical research practice, the human screening studies will facilitate more reliable selection of drug candidates potentially resulting in more effective safer therapeutic agents and (by means of identifying failures earlier) allow for quicker and more economical drug development.

Session 3: Advances in PK/PD Modelling Thursday 24th November (PM)

Stochastic Differential Equations in PK/PD modelling

Rune Overgaard

Technical University of Denmark, Denmark

Stochastic Differential Equations (SDEs) could potentially aid many parts of PK/PD modelling by a more complete description of the variations, better simulation properties, as a diagnostic tool that can pinpoint model deficiencies, etc. These models offer a general intra-individual error structure, where the residuals are decomposed into system noise from the SDEs and uncorrelated measurement noise.

Over the past few years, increasing interest has emerged for SDEs in PK/PD modelling, and some selected projects shall be reviewed: 1) CTSM is a stand-alone software for estimation in SDEs with measurement noise, and has been used for individual PK/PD modelling, e.g. to deconvolute the rate of appearance for linear and non-linear disposition models1. 2) The Extended Kalman Filter can be merged with the FOCE algorithm for estimation of non-linear mixed-effects models based on SDEs. MATLAB was used to confirm that inter-individual variability, measurement- and system noise can be separated, such that these models can be treated meaningfully2. 3) The Extended Kalman Filter was implemented in NONMEM to facilitate SDEs in more general PK/PD models and to increase estimation speed and performance3.

The implementation of SDEs in NONMEM has enabled more recent applications of SDEs in population PK/PD modelling, and we shall summarize three unpublished models for SDEs in PK/PD: 1) A model for thermoregulation, where SDEs offer a more complete description of the variations and better simulation properties, 2) A model for secretion, where SDEs changes the individual estimates, and 3) A model for haemoglobin, where SDEs decrease the sensitivity to model misspecification and increase the predictive performance.

References

Kristensen NR, Madsen H, Ingwersen SH. A Deconvolution Method for Linear and Nonlinear Systems based on Stochastic Differential Equations. PAGE poster presentation 2004.

Overgaard RV, Jonsson N, Tornoe CW, Madsen H. Non-linear mixed-effects models with stochastic differential equations: implementation of an estimation algorithm. J.Pharmacokinet.Pharmacodyn. 2005; 32(1):85-107.

Tornoe CW, Overgaard RV, Agerso H, Nielsen HA, Madsen H, Jonsson EN. Stochastic differential equations in NONMEM: implementation, application, and comparison with ordinary differential equations. Pharm.Res. 2005; 22(8):1247-1258.

Modelling Binary Data with Inter-subject Variability

Adrian Dunne and Valda Murphy

Department of Statistics, School of Mathematical Sciences, University College Dublin, Ireland.

The question of whether or not a mixed effects model is required when modelling binary data with inter-subject variability and within subject correlation was reported by Yano, Beal and Sheiner in 2001. That report used simulation experiments to demonstrate that, under certain circumstances, the use of a fixed effects model produced more accurate estimates of the fixed effect parameters than those produced by a mixed effects model. The Laplace approximation to the likelihood was used when fitting the mixed effects model. This presentation repeats one of those simulation experiments, with two binary observations recorded for every subject, and uses both the Laplace and the adaptive Gaussian quadrature approximations to the likelihood when fitting the mixed effects model. The results show that the estimates produced using the Laplace approximation include a small number of extreme outliers. This was not the case when using the adaptive Gaussian quadrature approximation. Further examination of these outliers shows that they arise in situations in which the Laplace approximation seriously overestimates the likelihood in an extreme region of the parameter space.

It is also demonstrated that when the number of observations per subject is increased from two to three, the estimates based on the Laplace approximation no longer include any extreme outliers.

The root mean squared error is a combination of the bias and the variability of the estimates. Increasing the sample size is known to reduce the variability of an estimator with a consequent reduction in its root mean squared error. The estimates based on the fixed effects model are inherently biased and this bias acts as a lower bound for the root mean squared error of these estimates. Consequently, it might be expected that for data sets with a greater number of subjects the estimates based on the mixed effects model would be more accurate than those based on the fixed effects model. This is borne out by the results of a further simulation experiment with an increased number of subjects in each set of data.

The difference in the interpretation of the parameters of the fixed and mixed effects models is discussed. It is demonstrated that the mixed effects model and parameter estimates can be used to estimate the parameters of the fixed effects model but not vice versa.

Comparison using simulated data and blinded analysis of approximate and exact likelihood parametric estimation methods for population PK model

Pascal Girard and France Mentré

INSERM and University of Lyon, France

Background: In 1994, a blind evaluation of several estimation algorithms on one population PK simulated data set was presented at American Statistical Association meeting (1). Since this date NONMEM software and nlme function in Splus™ and R have allowed large diffusion of population PK-PD in industry and academic research. NONMEM (FO and FOCE option) proc nlmixed in SAS™(FO) and nlme (FOCE) in Splus™ shares the same methods based on an approximation of the likelihood (2). Limits of those algorithms have been raised by several authors (3-5). More recently new algorithms avoiding likelihood approximation have been proposed, in particular the use of Gauss Hermitte quadrature (GQ) (6) and stochastic approximation EM algorithm (7).

Objectives: To perform a comparison between gold standard population methods and newly developed ones using several simulated data sets that would be serially, automatically and blindly analysed (i.e. ignoring the true value of the parameters), according to bias, precision and standard error estimates.

Methods: Simulate 100 population PK data sets using a one compartment model with first order absorption and first order elimination, the parameters being volume of distribution, elimination rate constant (Ke) and absorption rate constant (Ka). To avoid flip-flop during simulation, Ka is expressed as Ke+θ, θ being the parameter to be estimated. Three random effects, a full covariance matrix and an exponential error model complete the model. Each data set contains approximately one hundred patients receiving one single dose and having 1 to 4 concentration points. Proportion of patients having 3 or 4 points is on average 74% dataset. The datasets and exact true pharmaco-statistical model and rough initial estimates of fixed effect parameters were provided to several statisticians or pharmacometricians, well-known for their skills or having developed algorithms for non-linear mixed effect models. Each participant was supposed to fit each of the one hundred data sets, using an automated fitting procedure in order to avoid the potential effect of the analyst’s skills and experience on the results. Results were analyzed by computing relative bias and RMSE as well as by comparing standard errors with empirical SE defined as SD of the 100 estimated parameters. A crude comparison of CPU times is also provided. Following algorithms are included in this blind comparison (statistician in parenthesis): NONMEM™ V & VI with FOCE (N. Jonsson); nlme Splus™ function (J. Pinheiro and Chyi-Hung Hsu); Iterative Two-Stage Bayesian method (J. H. Proost); PEM (Bob Leary); MCPEM (S. Guzy); SAEM (M. Lavielle); SAS™ proc nlmixed (FO and GQ) (Russ Wolfinger).

Results: All algorithms converged and provided estimates for all simulated data sets except FOCE (NONMEM and nlme) which had difficulties, for about 30 to 40 datasets, either to reach their criterion convergence or to provide precision covariance matrix. In terms of fixed effect, FO from SAS nlmixed showed large biases. For random effect parameters, once again, FO provided biased variance estimates for all parameters; other method correctly estimated it when it was large. For small random variability (true value=15%), all methods overestimated it except SAEM. The structure of the covariance matrix was correctly estimated by SAEM and nlme and incorrectly for other methods. Globally, the 3 best methods in terms of bias were PEM, SAEM and GQ from SAS. In terms of RMSE, methods SAEM, PEM and MCPEM provided the best results. The worst methods in term of bias and RMSE were FO, nlme and Two-Stage Bayesian method. In terms of CPU, approximate likelihood based methods were the fastest ones, closely followed by GQ and SAEM.

Discussion: As expected, all methods using a non-approximated likelihood provided better estimation results in terms of bias and root mean square errors. In addition, software implementing the FOCE algorithms had convergence difficulties with a non negligible fraction of data sets. The interesting point is that those excellent results for exact likelihood methods were provided in a timely efficient manner very close to the one of FOCE, at least for SAEM and Gaussian quadrature. However this latter method requested first to estimate the parameters using FO as starting values for GQ. Those results need to be further completed with new real and simulated datasets. References: 1. Roe DJ. Comparison of population pharmacokinetic modeling methods using simulated data: results from the population modeling workgroup. Statistics in Medicine 1997;16:1241-62. 2. Lindstrom MJ, Bates DM. Nonlinear mixed effects models for repeated measures data. Biometrics 1990;46:673-87. 3. Comets E, Mentre F. Evaluation of tests based on individual versus population modeling to compare dissolution curves. J Biopharm.Stat. 2001;11:107-23. 4. Wahlby U, Jonsson EN, Karlsson MO. Assessment of actual significance levels for covariate effects in NONMEM. J Pharmacokinet.Pharmacodyn. 2001;28:231-52. 5. Gobburu JV, Lawrence J. Application of resampling techniques to estimate exact significance levels for covariate selection during nonlinear mixed effects model building: some inferences. Pharm.Res. 2002;19:92-8. 6. Pinheiro JC, Bates DM. Approximations to the Log-Likelihood function in the nonlinear mixed-effects model. J.Comput.Graph.Stat. 1995;1:12-35. 7. DeLyon B, Lavielle M, Moulines E. Convergence of a stochastic approximation version of the EM algorithm. Ann.Statist. 1999;27:94-128.

New diagnostics for the FO/FOCE methods in NONMEM

Andrew Hooker, Justin Wilkins, and Mats Karlsson

Division of Pharmacokinetics and Drug Therapy, Department of Pharmaceutical Biosciences, Uppsala University, Sweden.

There is a vast array of diagnostics for model building and evaluation, and it is often not clear which diagnostics to use and when to use them. This problem is compounded by the fact that development and analysis of population PK/PD models has shifted from using the first-order (FO) approximation to the more complex first- order with conditional estimation (FOCE) approximation to the true model. The FO method linearizes the model about the population mean of the random model parameters whereas the FOCE approximation conditions the linearization of the model about each individual’s empirical Bayes (post-hoc) estimates. Often model diagnostics appropriate for the FO approximation are not appropriate when using the FOCE approximation and vice versa. For example, the ratio likelihood test (used for hypothesis testing in model building) is often only strictly applicable when using the FOCE approximation. We present an overview of the types of model diagnostic tools available when building population PK/PD models, and discuss when and where different types of diagnostics should be used. We then present some new model diagnostics and demonstrate their utility.

One new diagnostic presented is the conditional weighted residuals (CWRES), an update to the weighted residuals (WRES), which are commonly used to evaluate model misspecification during the model building process. However, the WRES are always calculated using the FO approximation (even when using the FOCE method), leading to the possibility of misguided model development. The CWRES are calculated based on the FOCE approximation using each individual’s post-hoc inter- ηˆ individual variability estimates ( i ):

y − Ef() CWRES = ii Cov() y i (1)

y Ef() where i is the vector of data from an individual, i is the expectation or mean of Cov() y the model at those data values and i is the covariance of the data conditional on the model

Cov() y=Σ+Ω diag ( H H ') L L ' i (2)

with L the derivative of the model with respect to the inter-individual variability η η =ηˆ evaluated at i (WRES are evaluated at η = 0 ). We demonstrate that utilization of CWRES may improve model development and evaluation, and provide a more accurate method of detecting model misspecification when using the FOCE approximation. Open Session Friday 25th November (PM)

Current (and future) role of modelling and PK/PD analysis in regulatory assessments

Monica Edholm

Pharmacokinetic Assessor, The Medical Products Agency.

The role of pharmacokinetics, PK/PD, modelling and simulations in the regulatory assessment of new drug applications will be discussed. Information regarding PK/PD relationship both for efficacy and safety endpoints are important when developing dosing recommendations in sub-groups with altered pharmacokinetics. The presentation will give some examples of PK/PD analyses provided in recent applications and how modelling and simulations are used in developing dosage recommendations in sub-populations, using renal impairment as an example.

Innovation in the Pharmaceutical Industry - Myths versus Reality

Esther Schmid and Dennis Smith

Pfizer Global Research and Development Pfizer Limited

Over the recent years, the pharmaceutical industry's ability to bring new products to market has come under increasing scrutiny. Lack of "innovation" and rapidly increasing costs have been cited, as well as the huge investments in new technologies over the last decade, without apparent pay-back.

This presentation will use case studies of new medicines in cancer, HIV, cardiovascular disease and pain, which were driven by advances in molecular biology, high throughput screening, ADME structure-activity understanding and imaging technologies, to explode some of the myths associated with the ongoing public debate about science and technology's influence on pharmaceutical innovation. Moreover, the impact of pharmaceutical and biomedical innovations on life-expectancy, as well as the flow of new drug approvals will be discussed.

Increasing the Strategic Value of Kinetic and Dynamic Data Through Unified Modeling, Higher Productivity, and Regulatory Compliance

Simon Davis

Pharsight

Clinical development costs continue to escalate, leading to a constipated pipeline. Inefficient decision making and poor knowledge management are root causes of this inefficiency. Decisions are made without the necessary inputs, are not quantified, are focused on the wrong issues, and suffer from loss of knowledge when staff transfers.

FDA acknowledged these problems in its publication last year of "Innovation and Stagnation," which proposes utilization of model-based approaches to improve drug development knowledge management and decision making. The move to more reliance on model-based drug development at FDA is in fact several years old. FDA has published "Guidance for Industry, Population Pharmacokinetics," "Guidance for Industry, "Exposure-Response Relationships: Study Design, Data Analysis, and Regulatory Applications," "Guidance for Industry, Investigators, and Reviewers: Exploratory IND Studies," and "Center Proposes End-of-Phase 2A meeting to Identify Optimal Doses." In addition, the head of FDA has indicated that "we must find a way of illuminating the drug's progress by using modelling, simulation and quantitative disease models to create more informative and predictive development designs."

A recent survey of 244 NDA by Bhattaram showed that 42 used modeling to some degree. Pharmacometric analyses were pivotal in regulatory decision making in over half of the 42 NDAs. Of the 14 reviews that were pivotal to approval related decisions, 5 identified the need for further trials while 6 reduced the burden of conducting additional trials. Pharmacometric analyses influenced the approval basis, labeling, trial design, and evaluation of primary analysis methods. Given these guidances, we believe FDA OCP also recognizes the role that efficient management and reporting of clinical pharmacology data plays in supporting modeling and simulation activities.

FDA’s Bob Powell presented on the topic of FDA’s view of its own plan for a modern data management architecture in October, 2005. In that vision, FDA outlined the need for a centralized data repository to create a productive and validated workflow supporting all analyses. Such a system can reduce quality assurance overhead to uncover unavoidable data mismatches caused by manual data transformation steps. It can reduce long data preparation cycles for PK analysis, interpretation and presentation, due to fragmented source data from multiple heterogeneous systems with incompatible formats, and can reduce extensive lead-out time due to cumbersome manual data transcription and transfer steps for the preparation of report tables and graphs.

Plasma protein binding: prediction of albumin levels in the adult population and ligand displacement

Mark Baker (1, 2), Hege Christensen (2, 3), Amin Rostami-Hodjegan (2, 4), Geoffrey Tucker (2, 4)

1: UCB, Slough 2: Academic Unit of Clinical Pharmacology, University of Sheffield 3: School of Pharmacy, University of Oslo, Norway 4: Simcyp Limited, Sheffield

Circulating albumin concentrations determine to what extent ligands bind to multiple binding sites on albumin and the resulting compound free fraction (fu). Albumin has been described as a negative acute phase protein, with its levels declining in response to a wide range of factors including acute and chronic disease, injury, and nutrition status. A literature consensus has not been reached with regards to the influence of age, gender, and body composition with conflicting reports as to their effects on albumin concentrations. More than six thousand individuals participated in the 1999- 2000 NHANES health survey providing albumin concentrations and a broad range of covariates. This data was used to investigate the predictive ability of demographic covariates used in the Simcyp software. These covariates included age, gender, body weight, height, and ethnic group. A model was obtained that indicated that albumin concentrations were predicted by gender, age and body mass index. The model indicated age dependent decreases in albumin concentration of 0.57 g/L/decade age and 0.37 g/L/decade age in males and females with increasing age demonstrating a significant age effect.

An additional mediator of plasma protein binding is the displacement of bound compound by a competing ligand. Although displacement from plasma protein binding (dPB) is usually of little clinical significance, it should be taken into account when interpreting changes in total plasma concentrations of drugs subject to metabolically-based drug-drug interactions (mDDI). An approach was developed to predict changes in the free fractions (fu) of pairs of drugs that compete for albumin binding, knowing their binding affinity constants. This was explored using one-site and two-site albumin binding models for antiepileptic agents and non-steroidal antiinflammatory drugs.

Experimental fu values of valproic acid and phenytoin in the presence of ibuprofen, diflunisal or naproxen were predicted successfully (within 0.99-1.36 fold) by the model.

To understand the impact of the changes in fu, as a consequence of altered protein concentrations or displacement, a full physiologically-based pharmacokinetic model is required encompassing the wider set of compound binding plasma proteins including alpha-1 acid glycoprotein and lipoproteins. The approach developed in this study is a significant step towards the development of such a model.

Prediction of Time-Dependent CYP3A4 Drug-Drug Interactions – Impact of Enzyme Degradation and Intestinal Inhibition

Aleksandra Galetin

School of Pharmacy and Pharmaceutical Sciences, University of Manchester

Over the years main focus of the quantitative in vitro-in vivo drug interaction prediction was on reversible interactions, whereas irreversible or time-dependent interactions (TDI) were assessed on an individual case-to-case basis [1,2]. General under-prediction of an observed TDI using the simple [I]/Ki approach [3,4] indicated a need for a more systematic analysis of this type of inhibition interaction. In the current study, 37 in vivo cases of irreversible inhibition were collated, focusing on macrolides (erythromycin, clarithromycin and azithromycin) and diltiazem as inhibitors. The interactions included 17 different CYP3A substrates showing up to 7- fold increase in AUC (13.5 % of studies were in the range of potent inhibition).

In contrast to previous predictions of TDI [1,2], the current analysis is based on the human CYP3A4 degradation half-life (t1/2deg) estimated from either induction studies or in vitro investigations in liver slices. The impact of inter-individual variability observed in estimated decay of CYP3A4 activity (mean t1/2deg= 3 days, ranging from 1-6 days) on the prediction of TDI potential was investigated for compounds with a differential contribution from CYP3A4 to the overall elimination (defined by fmCYP3A4).

Significant intestinal first-pass metabolism may contribute to the inter-individual variability observed in bioavailability and magnitude of drug-drug interactions for some CYP3A4 substrates. Therefore, the effect of maximal inhibition of intestinal metabolism (assessed by the FG ratio) on the precision and accuracy of the TDI predictions was investigated.

The sensitivity of the predicted TDI to the differential CYP3A4 t1/2deg was highly co-dependent on fmCYP3A4, as minimal effects are observed when CYP3A4 contributes less than 50% to the overall elimination and when the parallel elimination pathway is not subject to inhibition. The current analysis indicates the suitability of the mean CYP3A4 t1/2deg of 3 days in the assessment of TDI potential as it provided low bias and the highest precision of TDI predictions.

Maximal FG ratio has proved to be a useful initial indicator of the extent of intestinal inhibition and it minimized the number of false negative predictions. However, the increased number of over-predictions indicated a need for further evaluation of this approach and assessment of the possible interplay with the efflux transporters and variability in FG estimates.

References:

1. Ito K, Ogihara K, Kanamitsu S-I and Itoh T (2003) Drug Metab Dispos 31: 945- 954. 2. Wang Y-H, Jones DR and Hall SD (2004) Drug Metab Dispos 32: 259-266. 3. Ito K, Brown HS and Houston JB (2004) Br J Clin Pharmacol 57: 473-486. 4. Rostami-Hodjegan A and Tucker GT (2004) Drug Discovery Today: Technologies 1: 441-448.

Evaluation of Methods for PK Modeling with Data Below Quantification Limit

Shuying Yang, Chuanpu Hu, Rashmi Mehta, Jixian Wang, Frank Hoke, Misba Beerahee, Mark Sale

Clinical Pharmacokinetics, Modelling and Simulation GlaxoSmithKline, UK

Pharmacokinetic data may contain concentration measurements below the quantification limit (BQL). BQLs are essentially censored data since no information other than they lie within an interval between 0 and the lower limit of quantification. While specific values cannot be assigned to these observations, nevertheless these observed BQL levels are informative. Using these observations creates challenges for modeling. Several potential likelihood-based methods have been considered in the literature for using the information contained in BQLs and have shown some advantages over the simple deletion or substitution methods. However complexity of implementing the likelihood-based approach has limited the usage of such approaches. In addition, the behaviour of these methods have not been well understood when the frequency of BQLs is high. On the other hand, censored data are straightforward to handle with Bayesian framework which has become increasingly popular due to the advance of computational power. Some studies have implemented the Bayesian approach in population PK models. The main focus of this work is to assess the performance of maximum likelihood-based method as implemented in NONMEM, and Bayesian approach using Markov Chain Monte Carlo (MCMC) as implemented in WinBUGS/PKBUGS for cases when some data are BQLs. These methods are evaluated and compared based on simulated data from a one- compartment first order elimination model and two real clinical trial data sets. Several quantification limits are applied to each of the simulated data to generate data with BQL ranging from average of 20 to 60 percent. These methods are applied to the simulated data sets and the real data, the influence of the amount of BQLs on the behaviour of the methods are then examined. For comparison, the simple deletion method, where all BQLs were removed, is used as well. The results show that removing BQLs can severely bias the parameter estimations, overall Bayesian approach with censoring results in least biased and most accurate parameter estimations, while the likelihood-based approach behaves reasonably well when only small proportion of BQLs exist in data.

Evaluation of the nonparametric estimation method in NONMEM VIβ

Radojka M. Savic, Maria C. Kjellsson, Mats O. Karlsson

Division of Pharmacokinetics and Drug Therapy, Department of Pharmaceutical Biosciences, Uppsala University, Sweden.

Background: In NONMEM VIβ, an option exists for estimation of a nonparametric parameter ( ) distribution. In this the parameter distributions are characterized as a discrete probability density function at a number of parameter values (support points) equal to or lower than the number of individuals in the data set. It is thus similar to other nonparametric methods. However, the support points are obtained from the empirical Bayes estimates (EBE) from a preceding parametric estimation step, which could be run with any of the standard methods available (e.g. FO or FOCE). The present study aims at exploring the performance of this nonparametric estimation method through a Monte Carlo simulation study with special emphasis on the analysis of data with non-normal distribution of random effects.

Methods: Pharmacokinetic data sets (100 sets for each condition) were simulated from a one compartment iv bolus model in which the CL/V parameter distributions were (i) log-normal, (ii) multimodal or (iii) heavy-tailed. Simulation settings were altered with respect to the number of individuals (200 vs 50 subjects) and informativeness of the data. Each simulated data set was analyzed by models assuming (i) the true (ii) log-normal and (iii) nonparametric distribution. In each case both FO and FOCE was used and the nonparametric estimation was always preceded by an estimation assuming a log-normal parameter distribution. The EBEs, which were used as support points for the nonparametric estimation step, were obtained using (i) the estimated parametric variance (ii) moderately inflated variance and (iii) highly inflated variance. Based on the estimated models, parameter distributions were evaluated at the 10th, 25th, 50th, 75th and 90th percentile of the CL and V distribution. Comparisons were made between the estimated distributions and the true one.

Results: With FOCE, estimated nonparametric parameter distributions matched the true parameter distributions in all studied data types, even when the preceding parametric step incorrectly assumed a log-normal parameter distribution. With FO, the estimated nonparametric parameter distribution matched the log-normal and heavy tailed true parameter distributions well. In the case of a true multimodal parameter distribution, there were observable differences between the estimated nonparametric distribution and the true one. However, these differences were smaller than for the parametric assumption of a log-normal distribution. The estimated nonparametric distributions were not influenced by the chosen set of EBEs which served as the support points.

Conclusion: The nonparametric estimation method in NONMEM has shown promising properties when analyzing different type of PK data with both FO/FOCE methods.

Optimal design for first in human studies to investigate the pharmacokinetic/pharmacodynamic behaviour of TGF-beta RI kinase inhibitor

Sophie Glatt and Ivelina Gueorguieva

Global Pharmacokinetics/Pharmacodynamics , UK.

To suggest optimal sampling times for a prospective study to characterize pharmacokinetics and pharmacodynamics in cancer patients following administration of TGF-beta RI kinase inhibitor.

Data from tumor growth kinetics Xenograft model in mice and from in vivo target inhibition (IVTI) in rats and in mice were incorporated in our PK/PD analysis. The PK/PD model in mice integrated the plasma time course of the compound, the inhibition of SMAD phosphorylation (biomarker) and tumor growth data, which were our pharmacodynamic measures. An indirect response model was used to describe plasma concentrations and observed pSMAD data. The model integrated factors within the tumor cell responsible for the synthesis and degradation of pSMAD. The tumor growth curve was described by a Gompertz model, which was extended to further understand the relationship between the time course of the tumor growth and the time course of TGF-beta RI kinase inhibitor. Additionally a PK/PD model, incorporating plasma concentrations and inhibition of pSMAD from the IVTI studies was fit to available rat data. The model used TGF-beta RI kinase inhibitor concentrations, relating those to the inhibition of the SMAD phosphorylation. The developed PK/PD models in the two species, mice and rat, were extrapolated to human. This enabled prediction in human of the targeted inhibition of SMAD phosphorylation under different dosing regimens. Using the extrapolated model we calculated the optimal sampling scheme for both concentration and effect measurements, to be used in patients. We used D-optimal design criterion, which minimises the volume of the joint confidence region by maximising the determinant of the Fisher information matrix (FIM) (inverse of variance-covariance matrix). It was further assumed that measurements made at distinct times are independent, but measurements made of each concentration are correlated with a constant variance- covariance matrix. Optimal sampling times were suggested for both individual and population designs. To determine the D-optimal design the determinant of the FIM has to be maximised over the whole design space using a number of optimisation methods (downhill simplex, simulated annealing, modified Fedorov).

Potentially, the relationship between time course of TGF-beta RI kinase inhibitor concentrations and clinical tumor response was defined. Following analysis of Phase I data we will update the PK/PD model and use optimal design as a guidance tool in designing future studies.

Mechanistic PK/PD modelling for signal transduction modulators. Application to TGF-β RI antagonists

Lorea Bueno (1), Celine Pitou (2), Sophie Glatt (2), J. Yingling (3), Dinesh de Alwis (2), Iñaki Fernández de Trocóniz (1)

1: Department of Pharmacy, University of Navarra, Spain 2: Eli Lilly Global PK/PD Trial Simulation, UK 3: Lilly research Laboratories, Eli Lilly and Company, US

Objectives. To develop a mechanistic PK/PD Model for a new Type I Receptor TGF- β Kinase inhibitor (TI-T KI) using human xenografts, and use the selected model to predict the time course of tumour growth of a series of follow-up TI-T KI compounds.

Methods. Human xenografts (MX1-breast and Calu6-NSCLC) were implanted sc. to nude mice. Experiments started 7 to 10 days after tumor implantation. Two different type of experiments were performed: (i) the PK/PD experiment providing information about the plasma levels of the compound, and the percentage change (inhibition) with respect to baseline of phospho-SMAD2,3 (PSMAD) in tumor, and (ii) the tumor growth experiment where the kinetics of tumor growth was followed during 25 to 30 days after the first drug administration. In both experiments drug or saline were administered orally in a single dose or in a multiple dosing design. Three TI-TβKI compounds were studied: LY2157299, LY2148675 and LY2109761.

Model development. Data obtained for LY2157299 were used to establish the model. An indirect response model was used to relate the predicted plasma concentrations with the observed inhibition in PSMAD. The model assumes the existence of factors within the tumor cell responsible of the synthesis and degradation of PSMAD. Tumor size (TS) in animals receiving saline did not reach a plateau and therefore a variant of the Gompertz model allowing for the switch from an exponential to a linear growth was used. Tumor growth inhibition observed in the animals receiving the new compound was linked to the inhibition of PSMAD through a delay in the propagation of the inhibitory signal which was modelled as a chain of transit compartments, and quantified by the mean signal propagation time (MSPT) parameter.

Model validation. The model selected was validated externally simulating the tumor growth kinetics in different studies where the three compounds were given under different dosing schedules. Bias in the model simulations with respect to the observed data was accounted computing the mean prediction error over time. Previously to the simulations, and for each compound the kinetic and PSMAD parameters had to be estimated.

Results: LY2157299 showed very similar PSMAD effects for the two cell lines. Estimates of IC50 (μM) were 0.79 and 0.70 for Calu6-NSCLC and MX1-breast, respectively. The model predicted a complete inhibition of PSMAD at high drug concentrations, and a very rapid turnover rate [t1/2 (min) = 18.6 (Calu6) and 32.0 (MX1)]. MSPT was estimated in 6.17 days for Calu6-NSCLC, which means that the drug will reach its steady-state effects after three weeks of continuous administration. For MX1 the estimate of MSPT was 28.7 days. Estimates of IC50 for the other two TI-TβKI compounds ranged from 0.0942 to 0.187 μM. Simulations obtained from the selected model were capable to describe very well the data for the different studies and compounds.

Conclusion: The integrated model was validated externally, and provided a tool to investigate different experimental scenarios as well as giving insights regarding the mechanisms of signal transduction in the cascade of events associated to the TGF-β membrane receptor. In its current status the model allows, once the pharmacokinetic and biomarker properties are established, to simulate tumor growth kinetic profiles. A natural expansion of the work, which is now under development, is to model the biomarker properties integrating in vitro drug information.

Poster Abstracts

The application of structural identifiability analysis to the design of pharmacokinetic experiments.

S. Y. A. Cheung1, L. Aarons1, I. Gueorguieva2, N. D. Evans3, K. R. Godfrey3 and M. J. Chappell3

1 School of Pharmacy and Pharmaceutical Sciences, University of Manchester, Manchester, Manchester , M13 9PL, U.K. 2 Eli Lily, Erl Wood Manor, Windlesham, Surrey, GU20 6PH, U.K. 3 School of Engineering, University of Warwick, Coventry CV4 7AL, U.K.

Structural identifiability analysis [1] of models is part of the experimental design process. The analysis should be performed prior to the planned experiment to identify whether the internal parameter values can be uniquely globally determined by the proposed input-output experiments. An unidentifiable model is such that an infinite set of parameter values would reproduce the observed data. This is important for population PK/PD experimental design where population variability is analysed via the parameter estimates. These concepts are demonstrated through the investigation of the structural identifiability analysis of two linear pharmacokinetic models.

The exhaustive modelling or similarity transformation approach [2] was chosen from other existing methods [3] to use for the structural identifiability analysis for both linear systems due to its robustness and the efficiency in handling a large number of input-output measurements. The complete controllability and observability of the models was verified before applying the similarity transformation approach. Analysis was carried out on variant models with different model structures to ratify the dependence of model assumptions in relation to the identifiability results. The similarity transformation approach was carried out using the symbolic calculation software, MATHEMATICA.

The results show the impact of modification and simplification of the model in relation to the level of structural identifiability. The prior consideration of the structural identifiability is shown to be an important part of PK/PD experimental design and leads to an understanding of the relationship between input-output experiments and the internal structure of the proposed model. This allows development of the model before any actual experiment is carried out.

References [1] R. Bellman and K.J Åström. On structural identifiability. Mathematical Biosciences. 7: 329-339 (1970) [2] Evans, N. D. et al., An Identifiability of a Parent-Metabolite Pharmacokinetic Model for Ivabradine, Journal of Pharmacokinetics and Pharmacodynamics, 28: 93-105 (2001). [3] Godfrey, K. R. and DiStefano III., J. J., Identifiability of model parameters. In E. Walter, editor, Identifiability of parametric models, pp1-20 Pergamon Press, Oxford, (1987).

The influence of experimental methods on the estimation of kinact and KI values of a time-dependent inhibitor.

Linh M. Van∗1, Judith Hargreaves2, Karen Rowland-Yeo3, Martin Lennard1, Geoff Tucker 1,3 and Amin Rostami-Hodjegan1,3 1University of Sheffield, Academic Unit of Clinical Pharmacology. Sheffield S10 2JF, UK. 2AstraZeneca, DMPK, Alderley Park, Macclesfield, Cheshire SK10 4TG, UK. 3Simcyp Ltd, Blades Enterprise Centre, Sheffield S2 4SU, UK.

Mechanism-based inhibition (MBI) is associated with irreversible/quasi-irreversible loss of enzyme activity. In vitro assays to characterise MBI involve two steps: (1) pre-incubation of enzyme with inhibitor (enzyme inactivation), and (2) incubation (measurement of residual enzyme activity with a probe substrate). To prevent further inactivation during step 2, enzyme and inhibitor are diluted into a solution containing high substrate concentration. We have investigated the impact of the extent of this dilution and probe substrate concentration on estimates of kinact (maximal inactivation rate) and KI (inactivation constant) for 3,4-methylenedioxy- methamphetamine (MDMA), which causes potent MBI of CYP2D6. Enzyme activity was measured by the conversion of dextromethorphan (DEX) to dextrorphan (DOR) in recombinant CYP2D6. Dilution factors of 1.25, 2, 5, 10, 25 and 50 (DEX at 30 mM) gave kinact (min-1) values of 0.20±0.06, 0.21±0.05, 0.31±0.06, 0.37±0.11, 0.51±0.10, 0.59±0.08 and KI (mM) of 3.32±2.84, 4.19±2.00, 8.63±3.09, 9.49±4.38, 5.95± 2.34, 7.26±2.04 respectively. DEX concentrations (10 - 100 mM) were not associated with statistically significant systematic changes in kinact and KI values when a 5-fold dilution was used (except for lower KI at 10 mM DEX concentration). Uncritical application of reported kinact and KI values when predicting MBI effects may compromise the reliability of in vitro-in vivo extrapolation. (LMV is supported by an AstraZeneca PhD Studentship).

A retrospective PBPK approach to assess Lung metabolism for SR E. Manolis, P. Delrat, T. Shepard

For the SR compound poor hepatic in vitro - in vivo extrapolation suggested the existence of extra-hepatic metabolism. A PBPK model was developed and additional lung and gut metabolism from in vitro experiments was incorporated into the model. This ascertained that both in vitro intrinsic clearances could not account for the difference between observed and simulated concentrations. Since intravenous administration had been performed in healthy young volunteers, the total clearance (in vivo CLtotal) of the compound was known. PBPK modelling was used to incorporate this clearance into a model taking into account both lung and liver metabolism and explore their roles on the overall metabolism of the compound (gut metabolism was not tested because in vitro it had a negligible effect).

Based on phase I clinical data, we allocated the in vivo CLtotal to lung and liver. In a first step the additivity of clearances after IV administration was tested. The contribution of each organ was expressed as a fixed percentage (f, 1-f) of the CLtotal with f corresponding to the contribution of the liver and 1-f the contribution of the lung. Based on the in vivo CLtotal observed value and using the well-stirred model equations, we generated intrinsic clearances for both organs. In a second step, the impact of various values of each organ contribution was investigated after intravenous and oral administration.

The impact was evaluated by superimposing in vivo and simulated profiles as well as by comparing both simulated and observed Cmax, tmax and exposure (AUCs) after each type of administration.

In a third step, Monte-Carlo simulations were performed. Known inter-individual variability was allocated to each organ intrinsic clearances. The influence was then assessed on the different combinations of intrinsic clearances.

The use of PBPK modelling allowed us to discriminate between lung and liver contributions on the overall total oral clearance. However due to the high observed inter and intra individual variability, it was difficult to draw a definitive conclusion. But a trend was definitely shown.

Comparative Pharmacokinetics of Five Drugs Administered to Humans at a Therapeutic and Microdose.

Graham Lappin and R Colin Ganer Xceleron Ltd., The Biocentre, Innovation Way, York, YO10 5NY, UK

Administering a sub-pharmacologic dose (microdose) to humans very early in drug development has been proposed to predict likely pharmacokinetics at therapeutic doses1, 2. This hypothesis was tested using five compounds, each chosen to represent a situation where in vitro and animal data might represent difficulties in scaling to therapeutic doses.

Although there is an account of pharmacokinetic linearity between a microdose and a therapeutic dose in the dog 3, the data presented here represent the first time the concept has been tested in humans. Subjects received one test compounds as a 14C- labeled microdose (100μg) alone and together with a therapeutic dose, with plasma analysis by Accelerator Mass Spectrometry. Good accord for t½, Cl, V and bioavailability between the microdose and the therapeutic dose was observed for diazepam, midazolam and a development compound, ZK253. For warfarin, clearance was reasonably well predicted, but a discrepancy in distribution was probably due to high affinity low capacity tissue binding. Oral microdose erythromycin failed to provide detectable levels, perhaps caused by a combination of acid instability and intestinal metabolism and efflux. Overall, used intelligently, microdosing offers the potential to aid in early drug candidate selection.

1. Lappin, G. and R.C. Garner, Big physics, small doses: the use of AMS and PET in human microdosing of development drugs. Nat Rev Drug Discov, 2003. 2(3): p. 233-40. 2. Combes, R.D., T. Berridge, J. Connelly, M.D. Eve, R.C. Garner, S. Toon, and P. Wilcox, Early microdose drug studies in human volunteers can minimise animal testing: Proceedings of a workshop organised by Volunteers in Research and Testing. Eur J Pharm Sci, 2003. 19(1): p. 1-11. 3. Sandhu, P., J.S. Vogel, M.J. Rose, E.A. Ubick, J.E. Brunner, M.A. Wallace, J.K. Adelsberger, M.P. Baker, P.T. Henderson, P.G. Pearson, and T.A. Baillie, Evaluation of microdosing strategies for studies in preclinical drug development: demonstration of linear pharmacokinetics in dogs of a nucleoside analog over a 50-fold dose range. Drug Metab Dispos, 2004. 32(11): p. 1254-9.

CYP450 Reductase Knockout Mice – A Useful Tool in Understanding Drug Disposition?

Alison O’Hara, Gareth Hughes, Heather Meigh, Christine Tyman and Kenny Watson Pfizer Ltd, Sandwich Labs, Sandwich, Kent, UK.

Recent advances in molecular biology now allow the routine breeding of genetically modified animals. Significant developments in the understanding of fundamental biological pathways have been driven by the increased success of this transgenic technology.

Cytochromes P450 are the principal and most widely profiled of the metabolising enzymes. However, deletion of all CYP450s is impossible due to the enormous diversity both within and across species, and the varying genetic locations which have evolved for this family of proteins. However, all CYP450s receive electrons from a single source – cytochrome P450 reductase (CPR). A CPR knockout mouse has been raised by the conditional deletion of this enzyme.

Investigation of an hepatic CPR knockout mouse is reported here, the aim of this work being to evaluate the feasibility of using the CPR knockout mouse to elucidate the role of CYP450 in the disposition of project compounds. Pharmacokinetics in the CPR KO mouse are presented for of a number marketed compounds, whose disposition in the mouse in well understood. Interpretation of these profiles is provided demonstrating the value of this model in the drug discovery environment.

Estimation of dosing strategies for cefuroxime by minimisation of a risk function

Anders Viberg1, Otto Cars2, Mats O Karlsson1 and Siv Jönsson1 1 Division of Pharmacokinetics and Drug Therapy, Department of Pharmaceutical Biosciences, Uppsala University, Uppsala. 2 Department of Medical Sciences, Infectious Disease, Uppsala University Hospital, Uppsala

Background Cefuroxime is a beta-lactam antibiotic mainly eliminated by renal excretion. Treatment with cefuroxime aims at exposing the patient to concentrations above a Minimum Inhibitory Concentration (MIC) for at least 50 % of the dosing interval. To achieve this in 90 % of the population, a previously described method (1) for optimising dosing strategies for individualisation with a discrete number of dosing categories was utilised.

Methods The dosing strategy was estimated by minimisation of a therapeutically defined risk function given known target and known characteristics of the population. The individualisation was made on the basis of creatinine clearance (CLcr). The target was defined as the fraction of time of the dosing interval above MIC (FTAU) and the value was set to 0.5. Combined risk functions were developed in which deviations from the target FTAU, the amount of drug administered in excess to reach the target and the time being below MIC during each dosing interval were considered and weighted against each other. A population covariate pharmacokinetic model for cefuroxime (2) and empirical distributions of CLcr and weight were used for description of the target population. Dosing intervals and CLcr-values at which doses should be in(de)cremented were estimated, while dose sizes were fixed to those available at the market (750 mg and 1.5 g). Dosing strategies with 2, 3, 4 or 5 different dosing categories were estimated for a series of pre-defined MIC-values. A dosing strategy was considered clinically acceptable if at least 90 % of the treated population was exposed to a concentration above MIC for at least 50 % of the dosing interval. This fulfilment was assessed using empirical distributions of individual MIC values for E.coli and S.pneumoniae obtained from the EUCAST database (3).

Results Target fulfilment for treatment of infections caused by E.coli was possible to achieve when the MIC was set to 16 mg/L, and for individuals with good renal function the estimated dosing interval was estimated to be less than 4 hours. For S.pneumoniae, several of the estimated dosing schedules resulted in target fulfilment, however, some dosing strategies involved extreme dosing intervals (>48 hours). By considering the time below MIC in the risk function dosing intervals with reasonable dosing intervals were obtained.

Discussion The present work illustrates how a risk function combining different aspects of drug treatment could be used for dose estimation. The results indicate that the use of cefuroxime for treatment of infections caused by E.coli bacteria may be inappropriate, since impractical dosing intervals were required to achieve target fulfilment. On the other hand, for S.pneumoniae, our results indicate that lower dose rates than currently used could be employed and still be clinically acceptable.

References 1. Jönsson S, Karlsson MO. Clin Pharmacol Ther 2003;73(1):7-19. 2. Viberg A et al., Br J Clin Pharmacol, in manuscript 3. EUCAST - European Committee on Antimicrobial Susceptibility Testing, www.eucast.org

Prediction of human xenobiotic clearance: Allometric scaling (AS) vs in vitro - in vivo extrapolation (IVIVE) M.R. Shiran1, N. J. Proctor1, E. M. Howgate1, K. Rowland-Yeo1, G. T. Tucker1,2 and A. Rostami-Hodjegan1,2 1Academic Unit of Clinical Pharmacology, The University of Sheffield, 2 Simcyp Limited, Sheffield, UK

The objective of this study was to compare IVIVE and AS in predicting human clearance of xenobiotics for which adequate data were available on in vitro metabolism, animal clearance in different species and human clearance values. The examples included alprazolam (oral), caffeine (oral), clozapine (oral), cyclosporine (oral and iv), dextromethorphan (oral), midazolam (oral and iv), omeprazole (oral and iv), sildenafil (oral), tolbutamide (oral), tolterodine (oral) and warfarin (oral). The prediction of human clearances via IVIVE carried out with Simcyp® methods. At least three animal species and four different methods were used for AS studies. The methods involved: (a) simple allometry (clearance versus body weight); (b) correction for maximum life-span potential (CL × MLP); (c) correction for brain weight (CL × BRW); and (d) use of body surface area. Overall accuracy of the predictions was determined by using mean fold error and pearson product moment correlation coefficient. Predictions considered successful if the absolute value of the mean fold error was ≤2. Accordingly, IVIVE predictions were accurate in 13 out of 14 cases (fold mean error ranging from 0.85 and 1.98) with sildenafil (4.1 fold over prediction) being the only outlier. AS methods (a) - (d) were accurate in 11, 9, 11 and 8 of 14, respectively. Some of the predicted clearances by AS were in error by more than 5 fold (cyclosporine (iv - method (a)); omeperazol (iv methods (b & c): tolterodine (iv method (c & d); sildenafil (methods (c & d)).

The results show that IVIVE could be more reliable than AS in predicting human clearance values and economical implications of switching from IVIVE to AS methods in drug development should be considered.

Norfloxacin – a whole body physiologically based pharmacokinetic model C. L. Rodrigues1*, I. Gueorguieva4, Marylore Channel2 and D. A. Berk3

1Centre for Applied Pharmacokinetic Research, School of Pharmacy and Pharmaceutical Sciences, University of Manchester, Oxford Rd., Manchester M13 9PL, UK. 2 Clinical Pharmacokinetics Department, Institut de Recherches Internationales Servier, 6 place des Pleiades, 92415 Courbevoie, France 3School of Pharmacy and Pharmaceutical Sciences, University of Manchester, Oxford Rd., Manchester M13 9PL, UK 4 Global PK/PD, Lilly Research Centre, Erl Wood Manor, Sunninghill Road, Windlesham, Surrey GU20 6PH.

*To whom correspondence should be addressed. Email: [email protected]

The extent of tissue penetration, distribution and the size of the therapeutic window of antibiotics are among the most important issues confronting antibiotic therapies. Modelling is an important applied tool in drug discovery and development for the prediction and interpretation of drug pharmacokinetics. Whole body physiologically based pharmacokinetic (WBPBPK) models are increasingly used to predict pharmacokinetic behaviour of drugs. Norfloxacin is a widely distributed fluoroquinolone antibiotic used mainly in genitourinary infections. The aim of this study is to develop a WBPBPK model to study norfloxacin kinetics in rats. Using tissue concentration and arterial blood data, a WBPBPK model for norfloxacin is derived. Initially, it is assumed that each tissue is represented by a single, well-stirred compartment and tissue affinities were estimated from the model. Permeability versus perfusion rate limitation of the norfloxacin distribution process for each tissue was tested. These alternative assumptions were judged based on fits of the tissue concentration data to the WBPBPK model by using nonlinear least squares regression. Additionally, data were available following the administration of three different doses. It has been reported that norfloxacin has nonlinear pharmacokinetics, which was investigated using the developed WBPBPK model.

Cure Rate Models in Depression Trials Gijs Santen(1), Roberto Gomeni(2), Meindert Danhof(1), Oscar Della Pasqua(1,3)

(1) Division of Pharmacology, LACDR, Leiden University (2) Department of Clinical Pharmacokinetics/Modeling & Simulation, GlaxoSmithKline, Verona, Italy (3) Department of Clinical Pharmacokinetics/Modeling & Simulation, GlaxoSmithKline, Greenford, UK

Efficacy studies of anti-depressant drugs often fail. This failure can be attributed not only to true lack of efficacy, but also to other factors, such as placebo effect, study design and endpoint selection. Despite growing evidence that the HAMD score is not a sensitive measure of drug effect, its change from baseline has been used as primary endpoint in efficacy trials. Thus it is of interest to investigate alternative methodology to evaluate treatment response in depression. In the current investigation, we propose a Bayesian parametric approach based on time of response to characterise drug effect of SSRIs, the so-called cure rate model.

The Bayesian parametric cure rate model is a modified survival approach which accounts for non-responders in the population. A 50% change from baseline was defined as clinical response. The data used in model building were retrieved from clinical databases. The proposed approach includes the use of reference placebo data as prior information to model parameter estimation. Not only does this allow the model to be fitted more accurately, it also allows comparison between the current study placebo and historical data. This can provide important insight in the study dynamics, such as estimates of differences in placebo effect size, which could lead to a failed study.

The Cure Rate Model was able to fit the data adequately in all clinical trials investigated. The transformation of the endpoint from ‘continuous’ to binary enabled the characterisation of drug effect over time and highlighted which parameter was most sensitive in capturing treatment effect. The parameterisation yielded an increase in sensitivity to drug response and hence increased statistical power.

The Cure Rate Model offers an opportunity to evaluate treatment effect with smaller sample sizes and could be used as an Interim Analysis Tool.

A critical evaluation of the experimental design used in reports of mechanism based inhibition (MBI) studies and the assessment of their implication for in vitro- in vivo extrapolation F. Ghanbari1, K. Rowland Yeo2, M.S. Lennard1, G.T. Tucker1, and A. Rostami- Hodjegan2 1University of Sheffield, Academic Unit of Clinical Pharmacology, Sheffield, United Kingdom; 2Simcyp Limited, Sheffield, United Kingdom

An increasing number of drugs are reported to cause MBI of cytochrome P450 enzymes (CYPs). Determination of the kinetics of MBI involves a pre-incubation followed by an incubation stage (Silverman, 1995), although experimental design differs considerably between laboratories. We have evaluated the experimental protocols reported in the literature for >100 mechanism based inhibitors of human CYPs in vitro. Wide variability in the timing of the first sample during the ‘‘pre- incubation’’ stage (range: 0.5–6 min; mode: 2 min), the dilution factor (range: no dilution—100 fold; mode: 20 fold), and the incubation time for measurement of the remaining enzyme activity (range: 2–20 min; mode: 10 min) was apparent. No correction was made for the natural decline in enzyme activity in 9% of the studies, and 14% of the reports used the slope of the log to base 10 instead of the natural log for calculating inactivation constants. The use of non-linear fitting was popular, although 55% of the data analyses were carried out using Kitz-Wilson plots. Because study design can have a substantial effect on the accuracy of the resulting MBI kinetic parameters, literature values should only be used for in vitro–in vivo extrapolation after due attention to experimental design issues.

A Single Dose Study to Investigate the Effect of the CCR5 Antagonist maraviroc on the QTc Interval in Healthy Subjects. J. Davis1, F. Hackman1, D. Sudworth1 & G. Weissgerber2 1Pfizer Global R&D, Sandwich Labs, UK & 2Pfizer Research Clinic, Hôpital Erasme, Brussels, Belgium. Corresponding author: [email protected] Background Maraviroc is a novel CCR5 receptor antagonist for the treatment of HIV infection. Phase 3 studies in both treatment experienced and treatment-naïve patients are currently ongoing. This study was designed to estimate the effect of maraviroc on the QTc interval in healthy subjects. Methods Single doses of maraviroc (100, 300 & 900mg) and comparator (moxifloxacin 400mg) were given in a randomised, placebo-controlled 5-period crossover study. Twelve-lead ECG’s were recorded at regular intervals. There was a washout of 7 days between periods. Sixty one subjects entered the study, of which 30 males and 27 females received all treatments. Their average age was 29.9 (range 19-44) years and their average weight was 66.1(range 50-85) kg. For each subject, the QT:RR relationship was evaluated from measurements made pre-dose in the 5 study periods and on a run-in day in period 3. A non-linear mixed effect model was used to

estimate the correction factor for each subject (bs). QTcI was calculated per subject as bs QTcIs = QT/(RR) . The primary endpoints analysed were; 1) QTcI at median Tmax (assumed to be 2h for moxifloxacin); 2) Maximum increase from baseline in QTcI from 1-4h post-dose; 3) Average QTcI from 1-4h post-dose. Comparisons against placebo were carried out for all 3 primary endpoints using ANOVA. The ANOVA for each pair-wise comparison allowed for variation due to sequence, subject within sequence (random effect), period and treatment. Baseline was used as a covariate. Results The mean difference in QTcI from placebo for all primary endpoints was <4ms for maraviroc (100, 300 & 900mg). The upper limit of the 90% confidence interval was below 7ms for all primary endpoints. There were no maraviroc-treated subjects showing maximum QTcI values > 450ms (males) and 470ms (females), or maximum increases from baseline in QTcI >60ms. Moxifloxacin caused a mean increase in

QTcI of 12-14ms for all 3 primary endpoints. Geometric mean Cmax and AUC following the 900mg dose were 1.15 mg/L and 5.26 mgh/L, respectively. No serious adverse events were reported. The most commonly reported adverse event was dizziness, following maraviroc 900mg. Other reported adverse events included headache, postural hypotension, nausea and cystitis. Conclusions Single maraviroc doses up to and including 900mg have no clinically significant effect on the QTc interval in healthy subjects.

Identification of hOCT2 substrates and prediction of renal clearance in vivo using stably transfected cell lines. Neil Attkins, Iain Gardner & Kevin Duffy* Pfizer Ltd, Sandwich Labs, Sandwich, Kent, UK. *University of Aberdeen, Aberdeen, Scotland AB252ZD

Renal excretion of drugs is a major route of elimination from the body, and most drugs are eliminated to some extent by the kidneys. Organic cation transporters (OCTs) are a major family of transporters and the OCTs play an important role in the kidney to excrete both foreign and endogenous substances. From this transporter family OCT2 is the primary human transporter involved in the active uptake of organic cations from the blood into the renal tubules to facilitate the clearance of compounds out via the urine. These studies have involved the use of a stably transfected cell line expressing hOCT2 (HEK293) to identify substrates for the OCT2 transporter. In addition the utility of using kinetic parameters (e.g. Km, Vmax) from the in vitro system to predict human renal clearance was assessed. The results from these studies will be presented.

Genetic algorithms and their applications in PK/PD data analysis M. Jamei1, J Yang1,2, A. Rostami-Hodjegan1,2, G. T. Tucker1,2 [email protected]

1- Simcyp Ltd, Blades Enterprise Centre, John St, Sheffield, S2 4SU, UK 2- Academic Unit of Clinical Pharmacology, University of Sheffield, Sheffield, UK

Drug discovery and development demands versatile and efficient prediction and optimisation tools which can effectively handle complex, multi-variant problems (van de Waterbeemd et al., 2003). Recently, evolutionary optimisation algorithms have been developed and Genetic Algorithms (GA) are amongst the most popular algorithms. GAs mimic natural selection (Goldberg, 1989) and have been applied successfully in various fields, including chemistry, biology, and many engineering disciplines.

GAs can be used as alternatives to conventional optimisation techniques in PK/PD data analysis. As a case study, we have applied GAs to determine the kinetic values for Mechanism-Based Enzyme Inhibition (MBEI). The simulation study was carried- out on 81 virtual MBEIs with known starting values. In 80% of the cases, GA predicted the values with less than 2% bias and the bias never exceeded 15% in any cases. However, the conventional experimental protocol introduced more than 100% bias for at least one of the kinetic parameters of 15 virtual MBEIs. The use of GA in other PK/PD problems warrants further investigation.

References:

GOLDBERG, D.E. (1989). Genetic algorithms in search, optimisation, and machine learning. Wokingham: Addison Wesley.

VAN DE WATERBEEMD, H. & GIFFORD, E. (2003). ADMET in silico modelling: towards prediction paradise? Nature Reviews Drug Discovery, 2, 192-204.

Inter-individual variability in the catalytic activity of CYP3A4 per unit of enzyme (kcat) Y Lei, ZE Wilson, KH Crewe, GT Tucker, A Rostami-Hodjegan

Academic Unit of Clinical Pharmacology, University of Sheffield, Royal Hallamshire Hospital, Sheffield, UK, S10 2JF

Inter-individual variation in in vitro CYP3A catalytic activity (metabolite formation per minute per mg microsomal protein; /mmp) is considerable. Differences in individual levels of hepatic CYP3A4 protein are considered to be the main source of this variation, and significant correlations between hepatic CYP3A4 abundance and in vitro catalytic activity (/mmp) have been reported. Our aim was to re-evaluate the variability based on the intrinsic activity of the enzyme (kcat). Microsomal CYP3A4 abundance was measured using a validated ELISA in duplicate samples from 53 livers. Testosterone 6β hydroxylase activity (nmol/min/pmol CYP3A4 at 200 μM testosterone) was available for 35 of these livers.

Inter-individual variation in kcat was lower than that of catalytic activity (/mmp): 22 vs 42 fold, respectively. The apparent variability in CYP3A4 related kcat may be due to additional variable contribution from CYP3A5 in different livers and differences in the levels of accessory proteins such as cytochrome b5. However, the CYP3A5 contribution to CYP3A mediated liver metabolism is believed to be small in Caucasians, and correlation of cytochrome b5 levels with catalytic activity (/mmp) as part of a multiple regression analysis of 26 of our livers did not achieve statistical significance.

The Quantification of Residual Blood in Excised Tissue Andrew M. Quigley & Mark B Baker

Blood and plasma are convenient for following the progress of a drug a body, but it is often desirable to know the concentration of the drug in specific tissues or organs, particularly brain and liver. Arguments rage over the merits of total concentration and free concentration, but in either case there are significant difficulties in determining drug concentration in organs. Many methods of quantifying drug in organs require radiolabelling and are also complicated by the presence of metabolites.

In early research studies, it is customary to homogenise the organ, extract the drug and measure its concentration using a specific assay such as LCMSMS. For brain in particular, effective drugs can have very low brain:plasma ratios. The residual blood in the brain can therefore have a major impact on the accurate determination of drug in brain.

Methods to measure the residual blood in brain have used 125I-labelled albumin, Ga- labelled white blood cells, 51Cr-labelled red blood cells and Evans blue dye as markers.

We describe a photometric method for the determination of haemoglobin in homogenates and blood, as a marker of residual blood that can be used to correct tissue concentrations. This method provides suitable throughput for use during pre- clinical discovery.

Inter- and intra-individual variability in gastro-intestinal physiology has significant effects on the prediction of fraction of dose absorbed (fa) M. Jamei1, J Yang1,2, A. Rostami-Hodjegan1,2, G. T. Tucker1,2 [email protected]

3- Simcyp Ltd, Blades Enterprise Centre, John St, Sheffield, UK 4- Academic Unit of Clinical Pharmacology, University of Sheffield, Sheffield, UK

Predictive physiologically-based PK models are being applied increasingly in drug development. However, the application of such models without consideration of inter- and intra-individual variability of the physiological parameters in the target population may lead to flawed conclusions, especially since early clinical data are often obtained from relatively small study populations.

The Compartmental Transit and Absorption (CAT) model (1) and literature values defining the variability of relevant aspects of gastrointestinal physiology were implemented in Simcyp®. The parameters were: gastric emptying time, Ts, small intestinal transit time, Tsi, and the radius and length of the small intestine, R and L respectively. Model drugs with a wide range of permeability characteristics were studied, including enalaprilat (a poorly permeable compound; Peff= 0.079 cm/h) and antipyrine (a highly permeable compound; Peff= 2.02 cm/h) (2).

Simulations were done for a population of 100 individuals repeated for 10 separate trials. For example, the predicted fa values of enalaprilat and terbutaline varied from 0.12 (5th percentile) to 0.42 (95th percentile) and 0.34 (5th percentile) to 0.71 (95th percentile), respectively. The variability was greater for less permeable compounds. Inconsistency between point estimates of fa and observed values from small clinical studies may be expected if inter- and intra-individual variability is not considered.

1. YU, L.X. & AMIDON, G.L. (1999). A compartmental absorption and transit model for estimating oral drug absorption. Internat J Pharmaceutics, 186, 119-125.

2. LENNERNAS, H., AHRENSTEDT, O. & UNGEL, A.-L. (1994). Intestinal drug absorption during induced net water absorption in man: A mechanistic study using antipyrine, atenolol and enalaprilat. Brit J Clin Pharmacol, 37, 589–596.

Building Physiologically-Based Pharmacokinetic Models in NONMEM using the PRIOR functionality

Grant Langdon1, Iva Gueorguieva2, Leon Aarons3, Mats Karlsson1

1 Department of Pharmaceutical Biosciences, Division of Pharmacokinetics and Drug Therapy, Uppsala University, Sweden 2 Eli Lilly & Company Ltd, Surrey, United Kingdom 3 School of Pharmacy and Pharmaceutical Sciences, University of Manchester, UK.

Objectives To formulate a physiologically-based pharmacokinetic (PBPK) model, using informative priors to describe diazepam disposition in the rat, and to extrapolate further to describe human pharmacokinetics (PK) using Phase I clinical data.

Methods Diazepam PBPK model in the rat Tissue and arterial blood concentration-time profiles following an intravenous infusion were collected from 24 rats, and used as input for the PBPK model, which comprised 12 tissue compartments and 2 blood compartments.1 Clearance was assumed to occur entirely from the liver compartment. Equilibrium tissue-to-venous concentration ratio bound (Kpb), intrinsic clearance (CLint) and residual error components were estimated from the model. Prior information included the fraction unbound in plasma (fu), the blood-to-plasma concentration ratio (R), and 2 experimentally measured Kpb values.

Diazepam PBPK model in man Venous blood samples following a single intravenous dose were collected from 12 healthy volunteers. The structure of the PBPK model from the rat was retained. Kpb, CLint and residual error components were estimated using the model. Prior information included the fu and R values in man and equilibrium tissue-to-venous concentration ratio unbound (Kpub) values from the rat model. Kpub values in man were assumed to be identical to Kpub values in the rat.

Results and Conclusions Kpb and CLint values were estimated simultaneously in each model and parameter estimates were in close agreement with experimental prior values2 and posterior mean values from a Bayesian analysis1 of the same data. The model provided a good overall description of the tissue concentrations in the rat although there was a degree of over- prediction in the arterial compartment. Plasma concentrations in man were well described by the model. Use of prior information allowed parameter estimation from a full PBPK model with limited data and may allow for a continuous flow of information through the different stages of drug discovery in the future.

References 1. Gueorguieva I, Aarons L, Rowland M. Diazepam Pharmacokinetics from preclinical to Phase I using Bayesian population physiological model with informative prior distributions in WinBugs. J Pharmacokinet Pharmacodyn. 2005 (accepted). 2. Igari Y, Sugiyama Y, Sawada Y, Iga T, Hanano M. Prediction of diazepam disposition in the rat and man by a physiologically based pharmacokinetic model. J Pharmacokinet Biopharm. 1983 Dec;11(6):577-93.

Dynamic Systems as Intractable Systems of Differential Equations: Design of Experiments

Julien Cornebise1, Bruno Boulanger2 1. University Paris VI and ESIEA 2. European Early Phase Statistics, MSG

Many bioprocesses, ranging from chemo-kinetics to glucose rate regulation, can be modelled as a system of ordinary differential equations with unknown parameters that are to be estimated, based on measurements achieved at several sampling times. This kind of system most frequently does not have a closed-form solution.

The aim of this poster is to present design of optimal experiments, i.e. how to determine the optimal sampling times (for a given criterion) which, based on a priori information (taken either in the classical sense of an assumed fixed value of the parameters or in the Bayesian sense of a prior distribution), allow for the most accurate estimation of the parameters.

Once the theoretical background has been explored (mostly nonlinear regression and asymptotic results based on the Fisher Information Matrix), many practical issues arise, due to the lack of closed-form solution: numerical algorithms to approximate the derivatives, numerical optimisation problems, convergence issues, etc. We here introduce both the theory and the practical aspects leading to an automated satisfactory solution.

Prediction of biliary clearance using Caco-2- a future screening assay? Mark Baker*, Lloyd King*, Hanna Hailu, Ted Parton, Sarah Bartlett and Simon Carter

At , part of the early DMPK screening cascade comprises human Colonic Adeno CarcinOma (Caco)- 2 cell line incubations, hepatocyte incubations and i.v. dosing of bile duct cannulated (BDC) rats. Compounds determined to have low metabolic turnover are progressed into the BDC for further characterisation.

The Caco-2 screen is used as an in vitro system for the investigation of intestinal epithelia permeability (apical to basolateral flux). In addition to permeability measurements, Caco-2 cells are known to express the dipeptide transporter (PEPT1), amino acid transporter, monocarboxylic acid transporter, and P-glycoprotein (P-gp) on the apical membrane corresponding to the brush border. This makes it suitable for the study of some transporter effects/events (basolateral to apical flux).

The bile duct cannulated rat allows collection of plasma, bile and urine following an iv administration of drug. It provides quantification of the hepatic (metabolic and biliary) and renal clearance of the compound. The samples from the BDC study can also provide valuable information as to the metabolic fate of the compound. As attempts to reduce metabolism have improved, biliary clearance has replaced it as a major contributor to total clearance. At present there is no simple in vitro screen for biliary clearance; however we report a recent trend observed at Celltech which may indicate a screening role for Caco-2 incubations. Exhibitors

PKUK 2006

PKUK 2006 has been provisionally booked at the Marriott Hotel, Sheffield, from the 17th to the 19th of November 2006. ______Notes