Mechanism-Based Pharmacokinetic and Pharmacodynamic ...166316/FULLTEXT01.pdf · variability in drug exposure and the sensitivity of the bone marrow. For two reasons ... Crude extract

Digital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Pharmacy 10

Mechanism-Based Pharmacokinetic and Pharmacodynamic Modelling of Paclitaxel

ANJA HENNINGSSON

ISSN 1651-6192ISBN 91-554-6232-4urn:nbn:se:uu:diva-5772

ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2005

Till min familj

Papers discussed

Th is thesis is based on the following papers, which will be referred to in the text by their Roman numerals.

I Henningsson Anja, Karlsson Mats O., Vigano Lucia, Gianni Luca, Verweij Jaap & Sparreboom Alex: Mechanism-based pharmacokinetic model for paclitaxel. J Clin Oncol 2001, 19, 4065-73

II Henningsson Anja, Sparreboom Alex, Sandström Marie, Freijs Agneta, Larsson Rolf, Bergh Jonas, Nygren Peter & Karlsson Mats O.: Population pharmacokinetic modelling of unbound and total plasma concentrations of paclitaxel in cancer patients. Eur J Cancer 2003, 39, 1105-14

III Henningsson Anja, Marsh Sharon, Loos Walter J., Karlsson Mats O., Garsa Adam, Mross Klaus, Mielke Stephan, Viganò Lucia, Locatelli Alberta, Gianni Luca, Verweij Jaap, Sparreboom Alex & McLeod Howard L.: Association of CYP2C8, CYP3A4, CYP3A5, and ABCB1 polymorphisms with the pharmacokinetics of paclitaxel. (Submitted)

IV Henningsson Anja, Sparreboom Alex, Loos Walter J., Verweij Jaap, Silvander Mats & Karlsson Mats O.: Population pharmacokinetic model for Cremophor EL. (Manuscript)

V Friberg Lena E., Henningsson Anja, Maas Hugo, Nguyen Laurent & Karlsson Mats O.: Model of chemotherapy-induced myelosuppression with parameter consistency across drugs. J Clin Oncol 2002, 20, 4713-21

VI Kloft Charlotte, Henningsson Anja, Wallin Johan, & Karlsson Mats O.: Population pharmacokinetic-pharmacodynamic model for neutropenia with patient subgroup identifi cation: comparison across anticancer drugs. (Manuscript)

Reprints were made with permission from the publishers: Th e American Society of Clinical Oncology and Elsevier Ltd.

Contents

Introduction ................................................................................................... 11Optimising cancer chemotherapy ..........................................................................11Paclitaxel ...............................................................................................................12

Mechanism of action ........................................................................................13Use in cancer chemotherapy .............................................................................13Side eff ects ........................................................................................................13Pharmacokinetics .............................................................................................13Polymorphism ..................................................................................................14

Cremophor EL ......................................................................................................15Main components .............................................................................................15Micelles ............................................................................................................15Pharmacological eff ects ....................................................................................16Pharmacokinetics .............................................................................................16

Haematological toxicity ........................................................................................17Physiology ........................................................................................................17Models of myelosuppression .............................................................................18Other drugs of interest in this thesis .................................................................19

Nonlinear mixed eff ects modelling .......................................................................20

Aim ................................................................................................................ 23

Patients and Methods ..................................................................................... 24Data description ....................................................................................................24Assays ....................................................................................................................30Pharmacogenetic analysis (Paper III) ....................................................................30Determination of critical aggregation concentration (CAC) in plasma (Paper IV) .31Pharmacokinetic model development ....................................................................31

Mechanism-based model (Paper I) ...................................................................31Validation and covariate relation investigation (Paper II) .................................32Association of genetic polymorphism with paclitaxel PK (Paper III) ................32Cremophor EL pharmacokinetics (Paper IV) ...................................................32

Pharmacodynamic model development .................................................................33Empirical haematological toxicity model (Paper I and II) .................................33Mechanism-based model of myelosuppression (Paper V) ................................ 34Subgroup identifi cation (Paper VI) .................................................................. 34Stepwise covariate model building within NONMEM ....................................35

Data analysis .........................................................................................................36Log likelihood profi ling ....................................................................................36

Results ........................................................................................................... 37Mechanism-based pharmacokinetic model (Papers I, II and III) ..........................37CrEL pharmacokinetics and CAC (Paper IV) .......................................................43

Predictive performance .....................................................................................45CAC measurements ..........................................................................................45

Empirical models for paclitaxel-induced neutropenia (Papers I and II) ................ 46Mechanism-based model of myelosuppression (Papers V and VI) .........................48

Sub-group analysis (Paper VI) ..........................................................................49

Discussion ...................................................................................................... 53

Conclusions ................................................................................................... 58

Populärvetenskaplig sammanfattning ............................................................ 60

Comments on my contribution ...................................................................... 61

Acknowledgements ........................................................................................ 62

References ...................................................................................................... 65

Abbreviations

AAG |1-acid glycoproteinALB Serum albuminALP Alkaline phosphataseALT Alanine aminotransferaseANC Absolute neutrophil countAST Aspartate aminotransferaseAUC Area under the concentration

time curveAUCEdir, 50 Duration of maximal direct

eff ect needed to produce half maximal observed

BCrEL Binding component proportional to [CrEL]

Blin, bc Linear binding to blood cellsBlin, p Linear binding to plasma

componentsBmax, bc Maximal binding to blood

cellsBmax, p Maximal binding to plasma

componentsBIL BilirubinBSA Body surface areaC ConcentrationCAC Critical aggregation

concentrationCb Total blood concentrationCDRUG Drug concentrationCirc0 Circulating cells at time 0Circt Circulating cells at time tCMC Critical micelle concentrationCp Total plasma concentrationCREA Serum creatinineCrEL Cremophor ELCu Unbound concentrationCYP450 Cytochrome P450DNA Deoxyribonucleic acid E Eff ectEC50 Drug concentration producing

half maximal eff ectEmax Maximal eff ectEB Empirical BayesEtOH EthanolFO First order estimation FOCE First order conditional

estimation

fu Fraction unboundG-CSF Granulocyte colony-

stimulating factorGGT Gamma glutamyl transferaseGM-CSF Granulocyte-monocyte colony-

stimulating factorH Hematocrit IIV Interindividual variabilityIL InterleukinIOV Interoccasion variabilitykcirc Elimination rate constant of

circulating cellsKm Concentration at half

maximum elimination rateKm, bc Concentration at half maximal

binding to blood cellsKm, p Concentration at half maximal

binding in plasmakprol Rate constant that determines

proliferationktr Rate constant from the transit

compartmentsLDH Lactate dehydrogenaseMTT Mean transit timen Number Na Not applicableNaCl Sodium chloridene Not estimatedOFV Objective function valuePD Pharmacodynamic PK PharmacokineticQ2, Q3 Intercompartmental clearancesSE Standard errorSFN Survival fraction at nadirV1 Volume of distribution of the

central compartmentV2, V3 Volume of distribution of the

peripheral compartmentsVmax Maximum elimination rateWBC White blood cell count~ Sigmoidicity factor in Paper

I and II, feedback power function in Paper V

} Sigmoidicity factor[CrEL] or CCrEL Concentration of CrEL

11

PK/PD Modelling of Paclitaxel

Introduction

Optimising cancer chemotherapyChemotherapy in cancer treatment is used in diff erent settings. It could be used as induction therapy for cure or symptom relief, adjuvant therapy after surgery or radiotherapy to prevent metastasis formation from cells that escaped from the primary site, or as preoperative and neoadjuvant therapy to achieve local or long-term tumour control. Very few solid tumours are curable with chemotherapy whereas various kinds of haematological cancers are1. Th e log-cell kill theory2-4 depicts that chemotherapy kills a proportion of the remaining tumour cells after surgery or radiation in each cycle to either cure the patient or keep the disease stable. Th e tumour cells regrow between the treatment cycles and the loss of tumour mass by surgery, cytotoxic therapy or radiotherapy induces the proliferation rate in the remaining tumour, which is then more sensitive to the next cycle of treatment5. If the next treatment is delayed because of excessive acute toxicity on normal tissues from the previous treatment cycle it could mean that the tumour regrowth is no longer controlled and the patient dies from the disease. A frequent acute toxicity following chemotherapy is myelosuppression, usually manifested as neutropenia. Th e variability in myelosuppression between individuals is generally large and is determined by the variability in drug exposure and the sensitivity of the bone marrow. For two reasons it would therefore be valuable to identify groups at risk for neutropenia; fi rst, the high risk of lethal infections during the neutropenic period and second, the risk of not achieving stable disease or cure due to a delay and/or unnecessary large dose reductions. It is also essential to achieve tumour control within a limited number of chemotherapy cycles because the total number of cycles is limited by cumulative exposure-dependent toxicities on tissues like heart, liver, kidney or nerve, which are only slowly or non-reversible5. With the use of pharmacokinetic/pharmacodynamic (PK/PD) models dosing can be improved both for the fi rst dose (a priori) and the following cycles when information from the fi rst cycle is available (a posteriori) and thereby the optimal treatment for each patient may be achieved more quickly. In this thesis the focus is set on the relations between PK and myelosuppression, especially neutropenia, since it is the most common dose-limiting toxicity for many anticancer drugs, including paclitaxel. Paclitaxel is an anticancer drug frequently used against various types of cancer e.g. ovarian, breast and non-small-cell lung cancer. Its PK has been characterised by capacity-limited elimination and distribution6-8 thus making

12

Anja Henningsson

dosing even more complex. Later studies have suggested that micelle entrapment in the formulation vehicle Cremophor EL (CrEL) in plasma could be responsible for this non-linear behaviour9,10. In this thesis a mechanism-based PK model for paclitaxel describing the relations between unbound and total concentrations in plasma and blood has been developed. PK/PD models that can describe and predict the time course of neutropenia have been developed and investigations of patient characteristics explaining variability in both PK and PD have been performed.

PaclitaxelPaclitaxel is a plant alkaloid extracted from the bark of the Western yew, Taxus brevifolia. Crude extract from the bark was tested against tumours already in the 1960s but it was Wani et al11 that identifi ed paclitaxel (Figure 1) as the active substance in the bark in 1971. Due to poor solubility in aqueous solutions paclitaxel is dissolved in 50% ethanol and 50% Cremophor EL (CrEL) in the commercially available formulation Taxol®. As CrEL itself has pharmacological eff ects (see below) other formulations of paclitaxel are under development, for example paclitaxel dissolved in another micelle forming surfactant (Pacliex)12 or in liposome formulations13. Prodrugs, e.g. polymer-conjugated paclitaxel14 or a covalent conjugate with docosahexaenoic acid paclitaxel15 are being evaluated in Phase I and II clinical trials.

Figure 1. Chemical structure of paclitaxel

13


Mechanism of actionTh e mechanisms behind the antitumour eff ects of paclitaxel are not fully known but the drug is known to bind to the beta subunit of tubulin and thereby promote polymerisation and formation of stable microtubules that are resistant to disassembly16. Th is prevents the normal growth and breakdown of microtubules required for cell division resulting in cell proliferation inhibition and apoptosis (programmed cell death)17,18

Use in cancer chemotherapyPaclitaxel is active in metastatic breast cancer and is under evaluation for adjuvant and neoadjuvant treatment of early breast cancer19,20. Th e combination of paclitaxel and a platinum (cisplatin or carboplatin) is used as fi rst line adjuvant chemotherapy in ovarian cancer21. Th e same combinations are used in non-small-cell lung cancer22. Paclitaxel has also shown to be eff ective in the AIDS related Kaposi’s sarcoma and has showed activity against a number of tumours (head and neck, gastric, colon, prostate)18.

Side eff ectsNeutropenia has been dose limiting in most clinical trials18 and has shown schedule-dependence with more pronounced neutropenia after longer infusion durations (24 h) compared with shorter infusion durations (3 h). Myalgia and neuropathy are also dose-related eff ects with the shorter infusions, while mucositis is more common with prolonged administration23. Due to severe hypersensitivity reactions associated with Taxol®, patients are generally pre-treated with histamine H1- and H2-receptor antagonists and corticosteriods e.g dexamethasone18.

PharmacokineticsTh e nonlinear PK behaviour of paclitaxel was demonstrated in the early 1990s7,24-26 and appeared to be more profound with the short infusions (3-hour). Both two- and three-compartment nonlinear models have been used for describing paclitaxel PK6,7,27. Th e nonlinearity was described by capacity limited elimination and distribution. Th e saturable distribution has been explained by saturable transport or tissue binding6,7,27. Th e fi rst study that suggested that the formulating vehicle could be causing the non-linear behaviour was performed by Sparreboom et al in 1996. Paclitaxel dissolved in Tween80-ethanol or dimethylacetamide was administered to mice. No PK nonlinearity in total plasma concentration profi les of paclitaxel was evident. When paclitaxel was dissolved in CrEL, tissue concentrations of paclitaxel increased linearly with increasing dose, suggesting linear PK for the unbound drug9. Th e free fraction of paclitaxel has been shown to be inversely related to CrEL concentrations in vitro10,28 and CrEL has also been shown to alter

14

Anja Henningsson

the blood to plasma ratio in vitro and in vivo by reducing the uptake in red blood cells28. In the absence of CrEL, paclitaxel is extensively bound in plasma where reported values of the fraction unbound (fu) have ranged from 5.4% to 13%10,29,30. Th e main part of the paclitaxel dose is metabolised by the cytochrome P450’s CYP2C8 and CYP3A4 and the metabolites are mainly excreted via the bile (Figure 2)31,32. Th e main metabolite in most patients is 6-alpha hydroxy paclitaxel33. Th e 3-p hydroxypaclitaxel showed some cytotoxic activity against a human A2780 ovarian carcinoma cell line while the 6|-hydroxy metabolites were inactive. All metabolites were toxic on haemopoietic progenitors, but much less toxic than paclitaxel, with mean concentrations at 50% inhibition of colony formation of 4, 60, 80 and 1000 nM for paclitaxel, 3-p hydroxypaclitaxel, 6|-hydroxypaclitaxel and 6|, 3-p-dihydroxypaclitaxel, respectively34. Elimination of unchanged drug is about fi ve percent through renal and fi ve percent through hepatobiliary elimination35. Th e hepatobiliary and intestinal secretion of the parent drug is mediated by the ABCB1 gene product, P-glycoprotein.

paclitaxel

CYP2C8 CYP3A4

6 -hydroxypaclitaxel 3´-p-hydroxypaclitaxel

CYP3A4 CYP2C8

6 , 3´-p-dihydroxypaclitaxel

Figure 2. Metabolic pathways of paclitaxel

Polymorphism Paclitaxel is mainly metabolized by CYP2C8 and CYP3A4 as described above. Th e association between drug PK and single nucleotide polymorphisms, SNPs, in genes encoding cytochrome P450s and membrane transport proteins is currently under evaluation. In vitro studies have demonstrated substantially decreased activity of the recombinant CYP2C8*2 and CYP2C8*3 enzymes in the metabolism of paclitaxel36-39. Recent clinical investigations also suggest that the CYP2C8*3 variant allele is associated with altered PK of the CYP2C8 substrate drugs repaglinide and ibuprofen40,41. CYP2C8*2 (805A>T; I269F) and CYP2C8*4 (792C>G; I264M) alleles have been found predominantly in African-Americans but not in Japanese

15


and Caucasians39. In contrast, the CYP2C8*3 variant (containing 416G>A; R139K) has an allele frequency of up to 14% in Caucasians42,43 but is very rare in African-Americans and Asians44. For CYP3A-mediated drug metabolism, the subfamily members CYP3A4 and CYP3A5 are the most important in adults45. A common transition in intron 3 of the CYP3A5 gene (CYP3A5*3C) introduces a frameshift during translation and results in a truncated, nonfunctional protein46,47. Th e expression of a functional CYP3A5 varies between races, suggesting that this subtype is more important for African-Americans than for Caucasians48. Although CYP3A4 and CYP3A5 have overlapping subtrate specifi city it is uncertain whether polymorphism in CYP3A5 is of importance in paclitaxel metabolism in vivo. For the membrane transport protein MDR1 P-glycoprotein, (encoded by ABCB1) the most extensively studied variant to date, is a common synonymous mutation at the transition C to T at nucleotide position 3435 (3435C>T; I1145I) at a wobble position† in exon 2649. Later fi ndings suggest that this variant is associated with altered protein expression in diff erent human tissues50, which may result in decreased hepatobiliary and/or intestinal secretion of substrate drugs like paclitaxel.

Cremophor EL

Main componentsCremophor EL (CrEL) is a heterogenous substance derived from the reaction between castor oil and ethylene oxide. Th is reaction gives rise to CrEL with usually highly variable mixture of polyoxyethyleneglycerol-triricinoleate, polyoxyethyleneglycerol, polyoxyethylene-triricinoleate, polyethyleneglycerol ricinoleate, polyoxyethyleneglycerol, polyoxyethylene-glycol (PEG)51. Recently a more detailed structural elucidation demonstrated the presence of polyoxyethyleneglycerol-S9,11-didehydrostearate and the results indicated that the major components of CrEL are polyoxyethyleneglycerol-monoricinoleate and polyoxyethylene-monoricinoleate and the polyoxyethyleneglycerol triricinoleate, which was previously claimed to be the major component, seems to only constitute a minor part52.

MicellesIn aqueous solutions, CrEL forms spherical micelles 8-22 nm in diameter as determined by quasi-elastic light scattering and cryogenic temperature transmission electron microscopy. Th e critical micelle concentration (CMC) in mixtures for parenteral use (NaCl, EtOH) is 0.064-0.071 mg/ml, which is greatly exceeded (> x 100) in a Taxol® infusion preparation with CrEL concentrations of 26-106 mg/

† Th ird nucleotide in a triplet, i.e., a silent mutation that do not change the aminoacid

16

Anja Henningsson

ml. After adding CrEL (in Taxol® formulation) to human plasma, formation of microdroplets (50-300 nm) has been seen. Th e larger lipid droplets were suggested to be formed from CrEL while small dense structures were suggested to be CrEL-enriched enlarged lipoproteins53.

Pharmacological eff ectsIn the later years it has been realized that CrEL exhibits several pharmacological eff ects. For example hypersensitivity reactions observed after intravenous administrations of drugs dissolved in CrEL could be attributed to the vehicle54,55. Th e mechanisms behind the acute hypersensitivity reactions associated with Taxol® administrations are not fully known but later studies have indicated that complement activation (C3) is involved. It has also been shown that the CrEL induced complement activation is concentration dependent, which could explain why hypersensitivity reactions are less frequently seen with lower infusion rates.

Peripheral neurotoxicity is another adverse eff ect observed after intravenous administration of drugs dissolved in CrEL56-58. Axonal swelling, degeneration and demyelinisation of rat dorsal ganglion neurons have been demonstrated in vitro58. CrEL has also been shown to alter lipoprotein patterns59,60. An interesting pharmacological eff ect of CrEL is its ability to inhibit the drug transporting membrane protein p-glycoprotein in vitro61,62 since this is one of the proteins excessively expressed in multidrug resistant (mdr) cancer cells. No eff ect on multidrug resistance has been shown in vivo, which could be due to the small volume of distribution of CrEL (se below).

PharmacokineticsSeveral studies of CrEL PK in plasma have been published in which diff erent assays have been used. In the early publications a bioanalysis method utilizing CrEL’s ability to inhibit mdr mediated daunorubicin effl ux was used63,64. Later, a more sensitive and reliable HPLC method was employed, in which ricinoleic acid after saponifi cation of CrEL was measured10,51. Due to time aspects and high costs of the previous assays a Coomassie Brilliant Blue colourimetric dye-binding assay was developed65,66. A method with electrochemical detection has also been presented67.

Regardless of assay used, CrEL has shown a peculiar schedule-dependent pharmacokinetic behaviour with higher clearances (lower AUCs) after longer infusions (24-hour)10,64,68. Th e volume of distribution is low, with reported values between 3 and 5 L/m2, suggesting limited distribution to peripheral tissues69,70. Th e elimination pathways of CrEL is not known but the renal elimination of CrEL has been reported to be < 0.1% and studies in patients with hepatic dysfunction have suggested that circulating concentrations of CrEL are not substantially altered compared to individuals with normal hepatic function71.

Recently a population PK three-compartment model with Michaelis-Menten elimination was presented (Table 1)72. Th e authors suggested that the saturable

17


elimination could be due to capacity-limited metabolism by carboxylesterases within the systemic circulation and that their three compartments could be explained by plasma, blood cells and micelles, respectively. However since total concentrations of CrEL were measured the micelles and monomers will all be part of the central compartment.

Table 1. Population pharmacokinetic model for CrEL72

Parameters Estimates V1 (L) 2.86 Q12 (L/h) 1.42 V2 (L) 1.75 Q13 (L/h) 0.154 V3 (L) 1.60 Km (mL/L) 0.197 Vmax (mL/h) 0.214 IIV V1 (%) 30.8 IIV V2 (%) 41.5 IIV Vmax (%) 33.9 Residual error: Additive (mL/L) 0.0985 Proportional (%) 6.83

V1, V2 and V3 volumes of distribution of the central and peripheral compartments; Q12 and Q13 intercompartmental clearances; Km plasma concentration at half maximum elimination rate; Vmax maximum elimination rate; IIV interindividual variability

Haematological toxicity

PhysiologyNeutrophils constitute 60-70% of the circulating leukocytes and the remaining cells are lymphocytes, monocytes, eosinophils and basophils, which all originate from a pluripotent stem cell. Th e stem cells have the ability to renew themselves and to produce committed haematopoietic progenitor cells, also called colony forming units. Neutrophil production occurs in the bone marrow. When the cells move through the diff erent development stages they are gradually becoming more diff erentiated (Figure 3). Th e myeloblast, promyelocyte and myelocyte are capable of cell division whereas the metamyelocyte, band and segmented cells are non-mitotic cells. Under normal conditions the mean transit time through the postmitotic stages was determined to be 158 hours73. Mature neutrophils are then released to the circulation and distributes between two pools; the circulating and the marginated (adhered to postcapillary venules), which are in equilibrium and of equal size74. Th e half-life of neutrophils in blood is around 7 hours75 and unlike erythrocytes and platelets, the disappearance is not related to their age but occurs randomly76.

18

Anja Henningsson

Several cytokines, for example interleukin (IL) 677, IL-3 and colony stimulating factors such as granulocyte colony stimulating factor (G-CSF), are involved in the regulation of the myelopoiesis. G-CSF increases proliferation and shortens the time of maturation. Th e clearance of G-CSF is mediated by receptor binding mainly on mature neutrophils (which have more receptors/cell than immature myeloid cells) and thus provide a negative feedback mechanism to maintain homeostasis78. After a single dose, anticancer drugs usually do not aff ect the stem cells, which have a low turnover rate, but the proliferating cells are very sensitive due to their high degree of mitotic activity.

Stem cell

CFU-GEMM

CFU-GMPro-

myelocyteMyelocyte

Meta-myelocyte

Bandcells

Segmentedcells

MyeloblastNeutrophilscirculating

Neutrophilsmarginated

Neutrophils

Bone marrowBlood Tissue

G-CSF, GM-CSF,

IL-3, IL-6 etc.

Differentiation

Proliferation

Regulatingfactors

Figure 3. Physiological features of neutrophil development and distribution

Models of myelosuppression

Empirical modelsSeveral models of paclitaxel myelosuppression have been presented in the literature. Relations between a summary variable of drug exposure and survival fraction at nadir has been described with threshold models, that is, all or nothing eff ects above and below a concentration of 0.05 or 0.1 µM24,27. Th e fact that these models assume no eff ect below the threshold and full eff ect above the threshold is not very physiological and a more general model has been suggested, where the observed drug eff ect is a function of the area under the direct eff ect time curve, which is dependent on the concentration of the drug79. Both the relation between the drug concentration and the direct eff ect and the relation between the area under the direct eff ect vs. time curve and the magnitude of the observed eff ect (the survival fraction of leukocytes) were described by sigmoidal Emax models for paclitaxel. In addition, the whole time course of decrease in leukocytes has been modelled with a

19


lag-time function, a cubic spline function determining the shape of the cell decrease, and a sigmoidal Emax model for the concentration-eff ect relation determining the magnitude of decrease from baseline.

Mechanism based modelsAn indirect eff ect model describing the whole time course of paclitaxel leukopenia with two compartments, representing bone marrow and peripheral blood, has been presented80. Th e drug eff ect was modelled as a fractional inhibition of leukocyte production depending on drug exposure of cells in a sensitive stage. Th e eff ect was modelled as % survival from baseline and a lag-time model was used to take into account the delay in decrease of leukocytes. Th e same model was later applied to another data set of paclitaxel81. Recently, a multiple-pool lifespan model with three compartments representing progenitor, maturation and circulating pools was suggested. Th e time spent in the diff erent compartments was estimated and the relationship between paclitaxel concentrations and the progenitor cell kill was described by a sigmoidal Emax model82. Semi-mechanistic models of myelosuppression have also been presented for other anticancer drugs. A model of topotecan-induced myelosuppression was based on the indirect-eff ect model presented by Minami et al.80 but included an additional eff ect compartment between plasma and bone marrow to account for a delay in drug distribution83. Models with transit compartments for proliferation and maturation have been developed for neutrophils after 2-deoxy-2-methylidenecytidine (DMDC) administration84 and for leukocytes after 5-fl uorouracil administration85. Th e use of transit compartments produce a more gradual delay compared with a fi xed time delay (lag-time) and is therefore more physiologic.

Other drugs of interest in this thesis

DocetaxelAlthough docetaxel and paclitaxel share mechanism of action, i.e. binding to tubulin, they are not two of a kind. Neutropenia is dose limiting for both drugs but it is reported to be AUC-dependent for docetaxel (not schedule dependent as for paclitaxel) and other toxicity patterns diff er86. Docetaxel causes fl uid retention and can cause skin and nail toxicity while paclitaxel is more neurotoxic23. Docetaxel is dissolved in Tween 80 and the PK of total drug concentration in plasma is described by a linear three- compartment model87. Docetaxel binds to |1-acid glycoprotein (AAG), albumin and lipoproteins to a high degree (92-94%)88. Docetaxel is mainly metabolised via CYP3A4 and excreted in the bile31. Th e drug is used against breast cancer, non-small-cell lung cancer and is being evaluated for numerous of other tumours, alone, or in combination with for example epirubicin, doxorubicin or cisplatin23.

20

Anja Henningsson

Vinfl unineVinfl unine is a vinca alkaloid under development by Pierre Fabre (Castres, France). Th e vinca alkaloids depolymerise microtubules and destroy the mitotic spindle and thereby block the mitosis89-91. Vinfl unine has shown activity in vitro against several tumours92 and has been tried clinically against non-small-cell lung cancer, breast cancer and bladder cancer93. Th e PK of vinfl unine has been described by a four-compartment model94.

DMDC2-deoxy-2-methylidenecytidine, DMDC, is an antimetabolite that can be orally administered. Elimination occurs mainly by metabolism in the liver. Th e PK has been described by a linear one-compartment model with fi rst order absorption95. A schedule-dependent haematological toxicity is dose limiting. Th e drug has shown activity in non-small-cell lung cancer and colorectal cancers but the development of DMDC has been discontinued.

EtoposideEtoposide is a topoisomerase II inhibitor used against small-cell and non-small-cell lung cancer, non-Hodgkin’s lymphoma, Hodgkin’s disease and cancer of the testis. Its antitumour eff ect is schedule dependent while its dose-limiting toxicity, myelosuppression, shows no or little dependence on schedule96. Th e PK of total and unbound drug has been described by a linear two-compartment model97. Etoposide is eliminated through renal excretion and metabolism. Clearance is related to age and creatinine clearance. Protein binding is high, between 80 and 97%98, and is associated with albumin and bilirubin concentrations99.

Irinotecan (CPT-11)Irinotecan, CPT-11, is a camptothecin derivative acting as topoisomerase I inhibitor which causes single strand breakage in the DNA and thereby block the DNA replication. CPT-11 is used as a single drug or in combination with 5-fl uorouracil in the treatment of colorectal cancer and has shown activity in small-cell lung cancer and non-small-cell lungcancer100. CPT-11 is metabolised to SN-38 in the liver. SN-38 has been shown to be at least a 100-fold more potent in vitro than the parent compound101. Th e PK of CPT-11 and its metabolite SN-38 have been described by three- and two-compartment models, respectively102. Plasma protein binding is about 65% for CPT-11 and 95% for SN-38103. Myelosuppression and diarrhoea are the dose-limiting toxicities104.

Nonlinear mixed eff ects modellingExperimental data arise from underlying processes, which are not always known. In nonlinear mixed eff ects modelling105, mathematical models are developed using

21


simultaneous estimation of fi xed eff ects that characterise the typical individual and random eff ects that quantify variability components. Mean and variances are thus estimated simultaneously and information from all individuals is merged, making it possible to use sparse sampling schemes. Th ere are two sources of random variability, one being the residual error or noise, caused for example by measurement errors in assay or reported time of observations. Th e jth observation in individual i can be described by

Observationij = PREDij + ßij (Eq. 1)

where ßij is the random eff ect describing the discrepancy between the individual pre-diction PREDij and the observation. ßij are assumed to be normally distributed with zero mean and an estimated variance «2. Th e other source of variability is due to real biological diff erences between individuals, i.e., the interindividual variability (IIV). For example one individual’s value of clearance could be described with the following equation

CLi = TVCL + ¢i (Eq. 2)

where ¢i denotes the diff erence between the ith individual’s value of clearance (CLi) and the population typical value of clearance, TVCL. Th e values of ¢i are assumed to be normally distributed and the estimated variance of ¢ is the variance of CL in the population (²2). However, in biology parameters are often assumed log-normally distributed and IIV is parameterised as follows:

CLi = TVCL·e¢i (Eq. 3)

Some parameters can vary over time within an individual and then give rise to interoccasion variability which can be estimated if observations are available from more than one course of treatment106. In this thesis the program NONMEM107 was used for population modelling.

NONMEM uses a parametric maximum likelihood method where the estimated parameters maximise the likelihood of the observations given the model. In NONMEM this is done by minimizing the extended least squares objective function value (OFV), which is proportional to –2 log likelihood of the data (-2 log L). In the extended least square OFV a penalty term is added, which increases with increasing variance and thereby counteracting the simultaneous decrease in the weighted squared error108.

Th e computation of the likelihood involves the evaluation of a multiple integral, which usually for nonlinear mixed eff ects models can not be solved exactly. Th erefore, diff erent approximations of the integral evaluation are available in the various NONMEM estimation methods. Th e fi rst-order (FO) methods implemented in NONMEM are based on fi rst-order Taylor series linearisation of the prediction with respect to the dependence on parameters. Th e derivative of the function can be evaluated around the typical value of the population, i.e., the expected value of

22

Anja Henningsson

the parameter, ¢i = 0 (FO method) or around the conditional Empirical Bayes (EB) estimate of the parameter (FOCE method). When the residual error is heteroscedastic (proportional to the observation) the residual error magnitude can be modelled as dependent on the population prediction (no INTERACTION) or on the individual prediction (INTERACTION). Th e choice and accuracy of the estimation method depends on the nature of the data and the degree of non-linearity in the model with respect to inter-individual random eff ects107.

23


Aim

Th e aim of the thesis was to develop predictive mechanism-based pharmacokinetic and pharmacodynamic models applicable for paclitaxel. Mechanism-based models promote a better understanding and may be used as tools in dosing individualisation and development of dosing strategies for new administration forms and new drugs in the same area.

24

Anja Henningsson

Patients and Methods

Data description

Paper IPatients with diff erent cancer types (ovary, breast, bladder, lung and adeno-carcinoma of unknown primary site) were included in Paper I. Th e patients, 23 females and 3 males, were between 32 and 69 years and had a median baseline neutrophil count of 4.9·109/L (range 1.55 to 21.4·109/L). Taxol® was administered as 3- or 24-hour infusions. Five individuals received concomitant therapy with doxorubicin (60 mg/m2) and two individuals received cisplatin (75 mg/m2). Th e data is described in Table 2. For CrEL and total plasma paclitaxel concentrations on average 12 samples (range 5 to 18) and 10 samples (range 6 to 13) per course were obtained for the 3- and 24-hour infusions respectively. Th e number of samples per course was on average 10 (range 7 to 12) and 11 (range 8 to 16) for unbound and blood concentrations respectively.

Table 2. Data included in Paper I

Observation type Patients (n)

(26 in total)

Courses (n)

(50 in total)

Observations

(n)

Dose range

(mg/m2)

Schedule

Cp 6 9 91 135-175 24-h infusion

Cp 20 41 484 135-225 3-h infusion

Cu 7 19 196 135-225 3-h infusion

Cb 8 20 221 135-225 3-h infusion

[CrEL] 26 50 575

ANC 24 40 80 Cu, unbound plasma concentrations Cp, total plasma concentrations Cb, blood concentrations [CrEL], total Cremophor EL plasma concentrations ANC, absolute neutrophil count

25


Paper IITh irty-one female and 14 male patients with diff erent types of solid tumours (colorectal, gastric, gall bladder, breast, uterine, ovarian, pancreas) receiving Taxol® as 3-hour infusion (112-233 mg/m2) were included in Paper II. Total paclitaxel concentrations (n=275), unbound paclitaxel concentrations (n=243) and CrEL concentrations (n=275) were obtained with an average of 3.5 (range 2 to 4) observations per individual, Plasma samples were collected according to a sparse sampling scheme; 1.5, 3 (before stop of infusion), 5 and 21 h after start of infusion. Only individuals with neutrophil observations for at least two weeks after administration, in total 106 observations from 36 individuals, were included in the toxicity analysis. Th e lowest observed neutrophil count was treated as nadir. Patient characteristics are described in Table 3.

Table 3. Patient characteristics Paper II.

Continuous covariates Courses (75 in total) Median (range)

Age (yr) 75 57 (23-71)

Height (cm) 73 169 (149-190)

Weight (kg) 68 68 (34-93)

BSA (m2) 72 1.76 (1.35-2.15)

CLCR (mL/min) 64 77 (41-166)

Albumin (g/L) 60 36 (25-47)

AAG (g/L) 69 1.35 (0.57-4.18)

Bilirubin ( M) 73 6.0 (3.0-41)

ALP ( kat/L) 72 4.85 (2.2-65)

AST ( kat/L) 73 0.51 (0.23-18.5)

ALT ( kat/L) 73 0.4 (0.09-10.6)

Categorical covariate

SEX 31 female patients (52 profiles)

14 male patients (23 profiles) BSA, body surface area; CLCr, creatinine clearance; AAG, 1-acid glycoprotein; ALP, alkaline phosphatase; AST, aspartate aminotransferase; ALT, alanine aminotransferase

Paper IIINinety-seven patients with diff erent primary tumour sites (bladder, breast, esophagus, lung, ovary, adenosarcoma of unknown primary site, head and neck, kidney, tongue and penis) were included. Th e patients received a 1-hour (n=42; dose range 50 to 100 mg/m2), a 3-hour (n=49; dose range 60 to 225 mg/m2) or a 24-hour (n=6; 135 or 175 mg/m2) infusion of Taxol®. Unbound paclitaxel concentrations (1025 observations) were obtained with an average of 11 (range 3 to 14) samples per individual. Patient characteristics are described in Table 4.

26

Anja Henningsson

Table 4. Patient characteristics Paper III

Courses (97 in total) Median (range)

BSA (m2) 97 1.78 (1.30-2.37)

Age (yr) 97 57 (24-81)

CREA ( M) 80 72 (20-145)

Bilirubin ( M) 97 7 (3-24)

ALP (U/L) 97 121 (38-1030)

AST (U/L) 97 21 (6.0-187)

ALT (U/L) 97 20 (2.0-264)

GGT (U/L) 96 35 (9.0-1567)

Hematocrit (L/L) 96 0.37 (0.20-0.45)

SEX 57 female patients

40 male patients

CCT carboplatin

1 hour infusion

15 patients

BSA, body surface area; CREA, serum creatinine; ALP, alkaline phosphatase; AST, aspartate aminotransferase; ALT, alanine aminotransferase; GGT, gamma glutamyl transpeptidase; CCT, concomitant medication

Paper IVA learning data set, with 1555 observations of CrEL plasma concentrations, was obtained from 116 patients who had received Taxol® as 1-, 3- or 24-hour infusions in previous PK studies of paclitaxel109-113 and CrEL concentration data from Paper I. Among these patients there were three individuals treated with a 3-hour infusion of Taxol® immediately followed by a 3-hour infusion of cisplatin (CDDP) and 19 patients receiving concomitant treatment with 5 mg/kg valspodar (SDZ PSC 833) orally 4 times a day, starting one day before a 3-hour infusion of 70 mg/m2 Taxol®. Schedules, doses and patient characteristics are described in Table 5. Number of observations per individual ranged from 2 to 15, with an average of 11 observations. CrEL concentration data from Paper II was used as validation data set.

Paper VLeukocyte and neutrophil data were available from six diff erent drugs (Table 6). All drugs were given as single drugs. Patients known to have received G-CSF were omitted. For paclitaxel Empirical Bayes (EB) parameter estimates from the population PK model developed in Paper II was used. Individual parameter estimates for cycles without PK information114 were calculated from individual covariate values and IIV estimates from the cycles where PK was measured (Paper II). For some individuals for whom pharmacokinetic information was lacking population parameter estimates and individual covariate values were used in the analysis. Th e other data sets contained PK information either as individual predicted concentration-time profi les (docetaxel)115, EB parameter estimates (DMDC)95 or actual concentrations (vinfl unine94,116-118, etoposide119,120, CPT-11104).

27


Table 5. Schedules, doses and patient characteristics Paper IV.

Schedule Number of patients (courses) Dose range CrEL (mL)

1 hour-infusion

(70-100 mg/m2 paclitaxel,

5.83-8.33 mL/m2 CrEL)

35 (35) 8.46 – 19.74

3 hour-infusion


5.83-18.8 mL/m2 CrEL)

76 (105) 8.28 – 37.6

24 hour-infusion


11.3-14.6 mL/m2 CrEL)

5 (7) 20.45– 26.86

Patient demographics Number of courses (147 in total) Median (range)

BSA (m2) 147 1.73 (1.28-2.37)

Age (yr) 147 55 (22-84)

CREA ( M) 129 71 (18-313)

Bilirubin ( M) 143 7 (3-24)

ALP (U/L) 87 122 (51-595)

AST (U/L) 142 22.5 (6.0-75)

ALT (U/L) 141 22 (2.0-138)

GGT (U/L) 85 30 (9.0-221)

Hematocrit 85 0.37 (0.24-0.48)

SEX 80 female patients (99 courses)

36 male patients (48 courses)

BSA, body surface area; CREA, serum creatinine; ALP, alkaline phosphatase; AST, aspartate aminotransferase; ALT, alanine aminotransferase; GGT, gamma glutamyl transpeptidase

Paper VIDocetaxel, paclitaxel and etoposide neutrophil data from Paper V was used. Patient characteristics are described in Table 7.

28 Anja Henningsson

Table 6. Data sets included in Paper V.

Drug Patients

(n)

Courses

(n)

WBC/ANC

(n of each)

Dose

(mg/m2)

Schedules WBC0

(·109/L)

ANC0

(·109/L)

Docetaxel 637 637 3553 75 or 100 1-h infusion 7.0 4.9

Paclitaxel 45 196 530 110-233 3-h infusion

every 3 weeks

7.6 5.5

Etoposide 71 118 682 225-789 72-h infusion 7.3 4.9

DMDC 147 147 823 12-50 Orally once/twice

daily for 7-14 days

8.9 6.2

Irinotecan

(CPT-11)

20 20 79 350 1.5-h infusion 7.8 5.2

Vinflunine 59 191 842

30-400

120-190

170-210

10-min infusions:

Day 1 every 3 weeks

Weekly

Day 1 & 8 every 3 weeks

7.0 4.8

WBC, white blood cells; ANC, absolute neutrophil count; WBC0 and ANC0 median baseline cell counts

29

PK/PD

Modelling of Paclitaxel

Table 7. Patient characteristics Paper VI

Drug Docetaxel Paclitaxel Etoposide 1st and [2nd] cycle Number (Fraction) Number (Fraction) Number (Fraction)

SEX malefemale

272365

(43%) (57%)

1133

(24%) (73%)

35 [21]36 [26]

(49%) [45%] (51%) [55%]

PERF 0 1

2

201347

84

(32%) (55%) (13%)

72810

(16%) (62%) (22%)

18 [16]30 [18]23 [12]

(25%) [34%] (42%) [38%] (32%) [26%]

PC yesno

287350

(45%) (55%)

3411

(76%) (24%)

PT yesno

13 32

(29%) (71%)

HT/IT yesno

12 33

(27%) (73%)

BLTR yes no

37 [23]34 [24]

(52%) [49%] (48%) [51%]

Median (Range) Median 1st cycle [all] (Range 1st cycle) [all] Median (Range) Age (yr) 56 (26-80) 57 [57] (31-71) [31-71] 60 [61] (27-77) [27-75] AAG (g/L) 1.34 (0.29-3.63) ALP ( kat/L) 5.7 [4.8] (1.9-46.0) [1.6-65] AST ( kat/L) 0.75 [0.49] (0.3-18.5) [0.22-18.5] ALT ( kat/L) 0.48 [0.38] (0.17-10.6) [0.09-10.6] LDH ( kat/L) 7.9 [7.3] (2.7-51.0) [2.7-118] CREA ( M) 76 [75] (57-153) [47-153] BIL (mg/dL) 0.4a [0.4] (0.2-2.2)a [0.1-2.4] 0.5 (0.1-1.8) BBIL (mg/dL) 0.5 (0.1-1.6) DBIL (mg/dL) 0 (-0.5-1.6) ALB (g/L) 36 [37] (28-47) [21-47] 37 (25-48) BALB (g/L) 37 (25-48) DALB (g/L) 0 (-16-8.0) PERF, Performance status according to WHO; PC, Previous chemotherapy; PT, Previous radiotherapy; HT/IT Previous hormone/immunotherapy; BLTR, Blood transfusion before cycle; AAG a1-acid glycoprotein; ALP, Serum alkaline phosphatase; ALT, Serum alanine aminotransferase; AST, Serum aspartate aminotransferase;LDH, Serum lactate dehydrogenase ; CREA, Serum creatinine; BIL, Serum bilirubin; ALB, Serum albumin. aRecalculated from M (MWbilirubin = 584.67 g/mol).

30

Anja Henningsson

AssaysIn Paper I total paclitaxel concentrations in plasma and blood were measured with a reversed phase high performance liquid chromatography method with UV detection at 230 nm as described earlier121. Th e lower limit of quantifi cation (LOQ) was 10 ng/mL (0.0117 HM). Th e quality controls (0.0468, 0.234, 0.468 and 17.56 HM) had an accuracy (measured/nominal) of 101.4, 100.9, 99.9, 102.75 and 100.5 % for total plasma concentrations. For blood concentrations the accuracy was between 96.7 and 103.3%. Between run precisions expressed as coeffi cients of variations (CVs) were <2.76 % for total plasma concentrations and <8.07 % for blood concentrations.

In Paper II total paclitaxel concentrations were measured with a reversed phase high performance liquid chromatography method with UV detection at 227 nm as described earlier79,122. LOQ for 500 HL plasma sample solution was tested at 0.017 HM (n=10, CV=13.4%, accuracy 99.7%). For between run precision, quality controls with paclitaxel concentrations of 0.0843, 5.057 and 13.49 Hmol/L were used, resulting in CVs of 7.3, 6.3 and 2.2% and accuracy of 99.1, 98.5 and 100.5%, respectively.

Unbound concentrations (Paper I, II and III) were obtained from an equilibrium dialysis method30 where 2 HL of a [3H] paclitaxel solution in ethanol was added to 300 HL of patient plasma prior to incubation. Th e ratio of drug concentrations measured in the buff er and plasma after dialysis was taken as an estimate of paclitaxel free fraction, fu. Because the volume shift during dialysis was negligible (<10%), the results were used directly without applying a correction factor. Tritiated paclitaxel was added to all samples, including the “blank” samples that were drawn from the patients prior to drug administration. Th e within-run and between-run variability were <9.2% and <3.3%, respectively, and the LOQ of the free fraction ratio was 1%. Unbound concentrations were obtained by multiplying measured total concentration with the estimated fu.

Th e analytical procedure for CrEL (study I, II and IV) was based on a colorimetric dye-binding assay using Coomassie Brilliant Blue G-25065,66. Th e LOQ of this procedure was 0.50 µL/mL, with an accuracy >93.67% and within-run and between-run precision of <9.47% and <1.83%, respectively. For inter-assay precision, quality controls with 2, 4 and 8 µL/mL of CrEL were used, resulting in a CV of 6.96, 3.88 and 11.3%, respectively, and an accuracy of 92.5, 92.0 and 90.1%, respectively.

Pharmacogenetic analysis (Paper III)Single nucleotide polymorphisms (SNPs) in CYP2C8 were identifi ed from the literature37or were mined using the publicly available SNP databases. Analyses of the CYP3A4, CYP3A5 and ABCB1 genes were performed as previously described123. Gentra PureGene Blood Kit (Gentra, Minneapolis, MN) or the QIAamp DNA Blood midi kit (Qiagen Inc, Valencia, CA) was used to extract genomic DNA from blood (n = 5) or plasma (n = 92), respectively. Th e single nucleotide polymorphisms (SNPs) in

31


the genes of interest [CYP2C8*2 (805A>T, I269F); CYP2C8*3 (416G>A, R139K); CYP2C8*4 (792C>G, I264M); CYP3A4*3 (1334T>C, M445T); CYP3A5*3C (6986A>G, splice variant); and ABCB1 3435C>T (I1143I)] were analysed using Pyrosequencing. Polymerase chain reaction (PCR) primers were designed using Primer Express version 1.5 (ABI, Foster City, CA). An Amplitaq Gold PCR master mix (ABI, Foster City, CA) was used for the PCR. Th e Pyrosequencing primers were designed using the Pyrosequencing SNP Primer Design Version 1.01 software (http://www.pyrosequencing.com) and a Pyrosequencing AB PSQ96 instrument and software (ABI, Uppsala, Sweden) were used for the Pyrosequencing. Th e genotype was called variant if it diff ered from the Refseq consensus sequence for the SNP position (http://www.ncbi.nlm.nih.gov/LocusLink/refseq.html). Genotype-frequency analysis of Hardy-Weinberg equilibrium was carried out using Clump version 1.9 124.

Determination of critical aggregation concentration (CAC) in plasma (Paper IV)Human drug-free fresh plasma was obtained from University Hospital Blood Bank, Uppsala, Sweden. Th e plasma was pooled and used within one day. Plasma samples were spiked with CrEL/EtOH (50:50) in saline, or with Taxol® infusion solution (0.6 mg/mL paclitaxel). Th e start-concentrations of CrEL ranged from 0.01 – 0.7 % (w/w), corresponding to 0.097 – 6.7 HL/mL. Th e spiked plasma was equilibrated at 37 ºC for approximately 1, 3 or 24 hours after administration to mimic the endpoints of diff erent durations of Taxol® infusions. Th e surface tension was determined by use of the droplet weight method125. Th e instrument was thermostatted to 37 ºC and the surface tension values presented are means of at least twelve data points (measured on twelve droplets).

Pharmacokinetic model development

Mechanism-based model (Paper I)Th e PK model was built to characterise unbound, total plasma and blood concentrations simultaneously, which required a disposition model for unbound paclitaxel, models for the relationship between unbound drug and drug bound to plasma constituents, including CrEL, and a model for distribution to blood cells which combined with the hematocrit predict total blood concentrations. Linear two- and three compartment models were tried for description of the unbound concentration. Diff erent combinations of linear binding to plasma constituents, nonlinear binding to plasma constituents and binding related to observed CrEL concentrations were

32

Anja Henningsson

tried for the model relating unbound to total plasma concentrations. Th e blood concentration model part was modelled with a linear and/or a nonlinear binding to blood cells. In the relation between blood and plasma concentration, the hematocrit (H) was used according to

Cb = H * Cbc + (1 - H) * Cp (Eq. 4)

Where Cb, Cbc and Cp are blood, blood cell and total plasma concentration of pacli-taxel, respectively.

Validation and covariate relation investigation (Paper II)Th e PK analysis was performed in two parts: 1) Validation of the model developed in Paper I, including previously determined parameter values, referred to as Model A, by: (i) predictions of unbound paclitaxel based on Model A and dose, (ii) predictions of unbound paclitaxel from EB parameter estimates based on Model A, dose and observed unbound concentrations of paclitaxel, (iii) predictions of unbound paclitaxel from EB estimates based on Model A, dose, observed total concentrations of paclitaxel and CrEL and (iv) an investigation whether the present data set requires all the structural model components of the previously developed model and whether any additional model components are necessary to achieve an adequate fi t to the data. 2) Whether any parameter-covariate relations could be identifi ed based on the present data set. Th e covariates listed in Table 3 were tested within NONMEM (method described below) on volumes of the central and peripheral compartments (V1, V2) and CL. Including highly correlated covariates such as BSA, height and weight in a stepwise procedure is not recommended126 so initially BSA was used as size measure. In addition, SEX, BIL, AGE, BSA, AST, ALT, ALP and the plasma proteins ALB and AAG were tested as covariates on the binding parameters. Missing values were replaced with the population median value.

Association of genetic polymorphism with paclitaxel PK (Paper III)Th e basic PK model development was based on the model developed in Paper I. Th e diff erent genotypes were tested as covariates on unbound clearance within NONMEM. In addition, individual parameter estimates (EB estimates) of unbound clearance were obtained and compared between the wild type group and the group of patients with variant allele present.

Cremophor EL pharmacokinetics (Paper IV)Two-, three- and four-compartment models with linear and/or capacity limited elimination (Michaelis-Menten) and capacity limited distribution or binding, were tried. Covariates (Table 5) were tested for statistical signifi cance in a stepwise procedure within NONMEM (see below). Missing covariate values were assigned

33


the population median value. Th e population PK model was developed using log-transformed data, i.e., the observed concentrations were log-transformed. Further, each course was treated as a separate individual (n=147). Th e estimated interindividual variability therefore includes the interoccasion variability.

Predictive performanceBased on the model for the learning data set, and covariates and dose information from the validation data set (CrEL concentrations from Paper II), population predictions of CrEL concentrations were obtained. Th e bias and precision of the prediction error (observed /predicted concentration) was compared with the 3-hour infusion data from which the model was developed. Population predictions for the studied populations were also obtained on the basis of the previously published three-compartment model with capacity limited elimination72 (Table 1) and the dose information.

Pharmacodynamic model development

Empirical haematological toxicity model (Paper I and II)Individual concentration-time profi les were obtained from the EB parameter estimates of the fi nal PK models. Th e exposure-toxicity (PK/PD) relationship was modelled with the general model79, which in its extremes correspond to a threshold or an AUC model, respectively. Th e direct eff ect, Edir, which is not observed, is a function of the drug concentration. Th e observed eff ect, Eobs is related to the area under the Edir-time curve, AUCEdir. Sigmoidal Emax models can be used to describe the relationships as follows:

Edir = C} / (EC50} + C}) (Eq. 5)

Eobs = Eobs, max * AUCEdir~ / (AUCEdir, 50

~ + AUCEdir~) (Eq. 6)

} and ~ are sigmoidicity factors. Th e parameter AUCEdir, 50 is the duration of maxi-mal direct eff ect needed to produce half maximal observed eff ect. Th e observed survival fraction at nadir (SFN) was used as observed eff ect in Paper I (Eq. 7 was applied) and in Paper II observed absolute neutrophil count (ANC) at baseline and nadir was used (Eq. 8 was applied).

SFN= 1 - Eobs, max * AUCEdir~ / (AUCEdir, 50

~ + AUCEdir~) (Eq. 7)

ANC= NEU0*(1 - Eobs, max * AUCEdir~ / (AUCEdir, 50

~ + AUCEdir~)) (Eq. 8)

In Paper II the baseline was estimated (NEU0). Eobs, max has a theoretical maximal value of 1 implying that if a high enough dose is given the blood cells will be wiped

34

Anja Henningsson

out entirely. Th e threshold and AUC models are special cases of the general model. For the threshold model, } is infi nitely high and EC50 is equivalent to the threshold concentration. For the AUC model } is 1 and EC50>>C. For the threshold model-ling, } was fi xed at a high value (G 20). When developing the AUC model, AUCEdir was set to the area under the plasma concentration–time curve. In these models individual predictions of Cp and Cu were used separately as predictors of toxicity. However, the models based on Cp or Cu are not nested models and no signifi cance can be attributed to diff erences in goodness of fi ts (OFV). In order to obtain this, in Paper II both models were compared to a full model including the sum of Cp and Cu, where the latter was scaled with a fi xed eff ects parameter estimated as part of the model.

Mechanism-based model of myelosuppression (Paper V)Th e data from each drug was analysed separately. Th e docetaxel, paclitaxel and etoposide data sets were used to develop the fi nal structural model (Figure 4) which consisted of one compartment representing proliferative cells, three transit compartments with maturing cells and a compartment of circulating cells (observed). A maturation chain, with transit compartments and rate constants (ktr), allowed prediction of a time delay between administration and the observed eff ect. Th e self-renewal of cells was dependent on the number of cells in the compartment, a proliferation rate constant (kprol) determining the rate of cell division and a feedback mechanism from the circulating cells to describe the overshoot from baseline (Circ0). Th e drug concentration (CDRUG) in the central compartment was assumed to reduce the proliferation rate or induce cell loss by a linear function (Slope) or an Emax-model. In the transit compartments it was assumed that the only loss of cells was into the next compartment. As the proliferative cells diff erentiate into more mature cell types, the concentration of cells is maintained by cell division. At steady state dProl/dt is equal to zero and therefore kprol=ktr. Since the elimination of cells from the circulation, described by kcirc, is not rate limiting for the observed decrease in cell counts after chemotherapy, the information on this parameter is limited and kcirc was set to ktr in the modelling. To improve interpretability, the mean transit time (MTT) was estimated which was defi ned as MTT=(n+1)/ktr where n is the number of transit compartments. Th us, the structural model parameters to be estimated were Circ0, MTT, ~ and Slope (or Emax and EC50). For consistency, interindividual variability was always (and only) estimated on Circ0, MTT and Slope (or EC50). To calculate Slopeu, i.e. Slope based on unbound concentrations, (or Slope for paclitaxel), unbound fractions of 0.0288 (docetaxel), 0.0529 (paclitaxel), 0.14127 (etoposide), 0.9795 (DMDC), 0.37103 (CPT-11) and 0.22 (vinfl unine; Pierre-Fabre, data on fi le) were used.

Subgroup identifi cation (Paper VI)Th e model developed in Paper V was used as the basic model with some minor

35


modifi cations. Th e drug eff ect was defi ned as an inhibitory linear slope model. Categorical and continuous patient-specifi c characteristics (Table 7) were analysed for their infl uence on the estimated parameters of the basic model, i.e. the three system-related parameters Circ0, MTT and ~ and the drug-specifi c parameter Slope. If baseline values of continuous variables were missing, they were replaced by the median of the population. If baseline data of categorical variables were missing the mode was used. Last observation carried forward was used within an individual. Performance status (PERF) was grouped as “unrestricted” performance (corresponding to PERF = 0 according to WHO128) and “restricted” performance (corresponding to PERF > 1 according to WHO. Covariates were tested for statistical signifi cance in an automated procedure within NONMEM (see below). For the covariate eff ects on system-related parameters, joint weighted means and standard errors for the estimates from the docetaxel, paclitaxel and etoposide analysis were calculated using the inverse variance method according to equations 9 and 10129,130.

Weighted mean = J(£P-COVi/SEi2) / J(1/SEi

2) (Eq. 9)

Weighted SE for this mean = √ [1/J(1/SEi2)] (Eq. 10)

where SEi is the standard error for the covariate eff ect £P-COVi for each of the three drugs Th e covariate eff ects were recalculated to show the percent diff erence from the typical value of the parameter with the alternative characteristic of the covari-ate (dichotomous covariate) or with the alteration of one covariate unit (continuous covariate).

Stepwise covariate model building within NONMEMCovariates were included in an automated stepwise procedure within NONMEM131. Th e process involves a forward inclusion where each covariate is included in a linear or nonlinear (two diff erent slopes, one for each side of the median value of the covariate) fashion and tested with the log likelihood ratio test (see below). Starting with a basic model the covariate relation that gives the largest drop in OFV is included in the fi rst step. In the next step each covariate is tested again on the intermediate model. For those relations already included, it is tested on the next level (nonlinear relation). When no more relations are statistically signifi cant on the p-level chosen (e.g. p<0.01), the forward search is complete and a backward elimination is performed. One relation at a time is excluded and the relation with the smallest increase in OFV is removed until all non-statistically signifi cant covariates are excluded. A stricter signifi cance criterion is usually used for the backward deletion (e.g. p<0.001). Th e evaluated covariate relations are centred around the median value of the covariate (MEDIAN), and are parameterised for continuous relations as

P=TVP*(1+COVEFF*(COV-MEDIAN)) (Eq. 11)

where the typical value of the parameter is TVP and the parameter of the magnitude

36

Anja Henningsson

in the covariate eff ect (COVEFF) describes the fractional change in the parameter (P) from TVP. For categorical relations the corresponding parameterisation is

P=TVP*(1+COVEFF) (Eq. 12)

where each level of the categorical covariate will have a separate value estimated for COVEFF. Th e degrees of freedom (df) are changed accordingly. A dichotomous variable corresponds to 1 df, a trichotomous variable to 2 df etc.

Data analysisTh e population analyses were all performed within NONMEM version V or VI beta107. Lognormal distributions were used for the interindividual and, if applicable, the interoccasion variability. Residual error was modelled with an additive and a proportional part, each of which could be excluded if not necessary to describe the data. If the model included a proportional error, FOCE with interaction was used. However, since this method is the most computationally complex and computer intensive, FOCE without interaction or FO was occasionally used in Paper V. Graphical diagnostics using the S-plus based program Xpose132, precision in the parameter estimates and the likelihood ratio test guided the model developments. Th e likelihood ratio test is one method available for statistical signifi cance testing of nested models. Th e -2 log (L1/L2), which expanded gives 2 log L2 – 2 log L1, corresponding to the diff erence in NONMEM OFV between two nested models, is approximately °2- distributed. For one degree of freedom a diff erence in OFV of >3.84, 6.63 and 10.83 corresponds to a statistical signifi cance level p < 0.05, 0.01 and 0.001, respectively. Th e degrees of freedom are the diff erence in the number of parameters estimated. In the stepwise covariate model building statistical signifi cance criteria of p<0.01 for covariate inclusion and p<0.001 for the backward elimination were generally used.

Log likelihood profi lingNONMEM reports estimates of the standard errors, which may be used to compute symmetrical confi dence intervals. Log likelihood profi ling (LLP) is one alternative method to estimate confi dence limits that do not assume symmetry around the estimate133,134. In brief, the parameters are fi xed, one at a time, to several values close to the estimates obtained from NONMEM when all parameters are estimated. Refi tting the other parameters of these reduced models generates a likelihood profi le. Th e model with the fi xed parameter estimate is competing with the full model (non-fi xed parameter) and the models are nested. Accordingly a likelihood ratio test can be used to assess the confi dence limits of the parameter. Th e upper and lower limits of the 95% confi dence interval are the fi xed parameter values generating a SOFV from the full model equal to 3.84 (assuming a °2 distribution).

37


Results

Mechanism-based pharmacokinetic model (Papers I- III)Th e structural model describing the disposition model for unbound drug and bindings in plasma and blood is shown in Figure 4. Observed versus predicted unbound concentrations from the diff erent studies are shown in Figure 5. Th e unbound concentrations (Cu) in Paper I were described with a linear three-compartment model referred to as Model A (Table 8). Although the data could be described with a two-compartment model without an obvious misfi t there was a preference for the three-compartment model (SOFV = 85.1). Th e parameters were generally well estimated (Table 8) and data well described (Figure 5(i)). Th e fraction unbound (fu) in plasma varied over time and was decreasing with increasing CrEL concentrations. However, the fu was higher during the infusion than post-infusion for CrEL concentrations of the same magnitude (Figure 6, upper panel).

Th ree diff erent binding components in plasma were identifi ed; a linear binding component, a capacity limited binding and a binding directly proportional to the observed concentration of CrEL. Omitting each of these binding components resulted in an increase in OFV by 58, 68 and 259 units, respectively. Total plasma concentrations (Cp) were described by the following equation:

Cp = Cu+ BCrEL · [CrEL]· Cu + Blin, p · Cu + Bmax, p · Cu / (Km, p + Cu) (Eq. 13)

where defi nitions and estimates are presented in Table 8. Th e fi nal model predicts that in the absence of CrEL the fu would be 7.5% and 11.6% for a Cu of 0.05 and 1 HM, respectively. Th e observed ratio of unbound concentration/concentration in blood cells was increasing with increasing unbound concentration suggesting a satu-rable process in blood cell binding.

Th e binding to blood cells (Paper I) was described with a linear and a capacity-limited binding component. Individual hematocrit (H) values were used in the modelling and total blood concentrations were described by the following equation:

Cb = H · (Blin, bc · Cu + Bmax, bc · Cu / (Km, bc + Cu)) + (1-H) · Cp (Eq. 14)

where defi nitions and estimates of the parameters are given in Table 8.

38 Anja Henningsson

Mean Transit Time =

E = 1-

Feedback =

Red blood cells

Platelets

Albumin

1-acid glycoprotein

Unboundpaclitaxel

CrEL

**

*

Peripheraltissues

Blin, p

Blin, bcBCrEL

Bmax,bcKm, bc

Bmax, p

Km, p

VmaxKm

Q2

Q3Q2

Q3

CL

V3

V3

V2V1

V1V2

Figure 4. Schematic overview of the mechanism-based PK and PK/PD models for paclitaxel including the PK model for CrEL (3-hour schedule). Defi nitions of parameters are described in Abbreviations. * Capacity-limited pathways

39


Parameters Model A (Cu,Cp,Cb Paper I)

Model B (Cu,CpPaper II)

Model C (Cu Paper III)

CL (L/h) 429 ( 6.6) 343 (3.5)a 301(4.3) V1 (L) 229 (17) 418 (7.1)b 225 (6.0) V2 (L) 134 (22) 1010 (4.2)c 3450 (7.8) V3 (L) 856 (15) 303 (6.6) Q2 (L/h) 213 (23) 188 (13) 132 (6.6) Q3 (L/h) 303 (21) 151 (6.5) Blin, p 7.59 (12) 3.31 (28) BCrEL 3.78 (10) 4.46 (7.2) Bmax, p ( mol/L) 0.0245 (27) 0.212 (34)d

Km, p ( mol/L) 0.000106 (43) 0.0225e Blin, bc 7.76 (8.7) Bmax, bc ( mol/L) 0.0897 (24) Km, bc ( mol/L) 0.00161 (61) BSA on V1 1.45 (14) BSA on V2 0.868 (18) BSA on CL 0.65 (22) BIL on CL -0.012 (21) AAG on Bmax 1.09 (14) IIVCL (CV %) 25 (31)f 20 (24)f 44 (18)f

IIVV1 (CV %) 46 (27)f 50 (27)f

IIVV2 (CV %) 40 (41)f 16 (37)f 38 (37)f

IIVQ2 (CV %) 65 (19)f

IIVQ3 (CV %) 37 (33)f

IIVCrELbind (CV%) 14 (31)f

IOVCL (CV %) 20 (26)f IOVV1, V2, V3,Q2, Q3 (CV %)

55 (23)f

Corr IIV CL-V1 0.41 (33)f

Corr IIV CL-Q2 0.71 (20)f

Corr IIV CL-Q3 0.55 (33)f

Corr IIV V1-Q2 0.44 (37)f

Corr IIV V1-Q3 0.38 (50)f

Corr IIV Q2-Q3 0.89 (22)f

1, Cu ( mol/L) 0.000697 (36)

2, Cu (%) 22.4 (8.6) 17.5 (8.9) 21.1(4.9)

1, Cp ( mol/L) 0.00296 (26)

2, Cp (%) 18.5 (7.2) 14.9 (8.8)

2, Cb (%) 17.5 (7.5) CL is clearance based on unbound plasma concentrations, V1, V2 and V3 are the volumes of the central and peripheral compartments of unbound drug, Q2 and Q3 are the intercompartmental clearances for unbound drug, BCrEL is binding directly proportional to CrEL concentration, Blin, p and Blin, bc are linear binding to plasma component and blood cells respectively, Bmax, p and Bmax, bc are maximal binding to plasma components and blood cells respectively. Km, p and Km, bc are concentrations at half maximal binding. RSE, relative standard error, in %; CV, coefficient of variation; Corr, correlation between individual estimates, 1

is the additive and 2 the proportional residual error related to Cu, Cp and Cb,a CL= 343 · (1+0.65· (BSA-

1.76))·(1-0.012·(BIL-6)), b V1 = 418·(1+1.45·(BSA-1.76)), c V2 = 1010·(1+0.868·(BSA-1.76)), d Bmax = 0.212*(1+1.09 · (AAG -1.35)), e Calculated from Slope 9.41 (9.4). Km=Bmax/slope, fRSE is related to the corresponding variance or covariance term.

Table 8. Final model parameter estimates with relative standard error (RSE, %) for the pharmacokinetic models for paclitaxel developed in Papers I-III.

40

Anja Henningsson

Th e paclitaxel concentrations obtained in Paper II were well described by Model A and dose information (Figure 5 (ii)). Unbound concentrations could be predicted from Model A and the in Paper II observed total paclitaxel plasma concentrations and CrEL concentrations or from Model B, dose and covariates (Figure 5 (iii and iv)). In Paper II the fu in the [3H] paclitaxel spiked pre-course samples were on average 7.9% (range 3.7 to 13 %) and decreased with increasing concentrations of CrEL (Figure 6, lower panel). When fi tting the model to the new data, the same binding components could be identifi ed, but only two compartments for the disposition model of unbound drug were supported (Model B, Table 8). In Paper III the random eff ects model developed in Paper I was modifi ed and correlations between the individual estimates were estimated (Model C, Table 8). Basic goodness of fi t plots for unbound concentrations in Paper III are shown in Figure 5 (v and vi).

In the covariate analysis in Paper II, BSA was found to be statistically signifi cant for CL, V1 and V2. Also, bilirubin for CL and AAG for Bmax, p were found statistically signifi cant. Th e typical values ranged from 169 to 654 L for V1 and 651 to 1352 L for V2 in this population. A change in BSA of 0.1 m2 typically caused a change in CL with 22.3 L/h and an increase in bilirubin (BIL) with 10 HM typically caused a decrease in CL of 41 L/h. Typical values of Bmax ranged from 0.032 to 0.866 HM. Th e IIVV2 decreased from 20% to 17% and IIVCL decreased from 26% to 19% when the covariates were included.

0.001 0.01 0.1 1Predicted Cu (i)

0.001

0.01

0.1

1

Obs

erve

d C

u (

M)

0.001 0.01 0.1 1Predicted Cu (ii)

0.001

0.01

0.1

1

Obs

erve

d C

u (

M)

0.001 0.01 0.1 1Predicted Cu (iii)

0.001

0.01

0.1

1

Obs

erve

d C

u (

M)

0.001 0.01 0.1 1Predicted Cu (iv)

0.001

0.01

0.1

1

Obs

erve

d C

u (

M)

0.001 0.01 0.1 1Predicted Cu (v)

0.001

0.01

0.1

1

Obs

erve

d C

u (

M)

0.001 0.01 0.1 1Predicted Cu (vi)

0.001

0.01

0.1

1

Obs

erve

d C

u (

M)

Figure 5. Goodness of fi t plots for the unbound concentrations of paclitaxel in Paper I-III. Shown are (i) observations (Paper I) vs. predictions based on Model A and Dose, (ii) observations (Paper II) predictions based on Model A and Dose, (iii) observations (Paper II) vs. EB predictions based on Model A, observed Cp of paclitaxel and [CrEL], (iv) observations (Paper II) vs. predictions based on Model B, dose and covariates (BSA and BIL), (v) observations (Paper III) vs. predictions based on Model C and Dose, (vi) observations (Paper III) vs. EB predictions based on Model C and observed Cu. Solid lines are lines of identity. Dotted lines are regression lines.

41


0 1 2 3 4 5 6 7

CCrEL ( L/mL)

0

4

8

12

f u (

%)

time after start < 2hall datatime after start >19h

0

1

2

3

4

5

6

7

8

9

0 1 2 3 4 5 6 7

CCrEL

( L/mL)

all data

time after start of infusion >20 h

time after start of infusion <2 h

f u (

%)

Th e IIV in unbound clearance in Paper III was not explained by the studied polymorphisms in CYP2C8, CYP3A4/5 or ABCB1. Observed frequencies of the diff erent genotypes are presented in Table 9. Th e ratio and 95% confi dence intervals of individual CLpresence of variant allele /CLwild type are presented in Figure 7. For the CYP3A5*3C the ratio represent presence of wild type allele/homozygous variant.

Figure 6. Observed fraction unbound (fu) of paclitaxel versus observed concentrations of CrEL in Paper I (upper panel) and Paper II (lower panel). In the lower panel a spline smooth for the indi-vidual predictions of fu is included.

42 Anja Henningsson

Table 9. Genotype and allele frequencies for the studied variant genes

Genotype frequencies a Allele frequencies b

Polymorphism c Nomenclature Effect d Wt e Het Var p q

CYP2C8 805A>T CYP2C8*2 I269F 69 (98.6) 1 (1.43) 0 (0) 0.993 0.007 CYP2C8 416G>A CYP2C8*3 R139K 78 (84.8) 11 (11.9) 3 (3.26) 0.908 0.092 CYP2C8 792C>G CYP2C8*4 I264M 90 (95.7) 4 (4.26) 0 (0) 0.979 0.021

CYP3A4 1334T>C CYP3A4*3 M445T 93 (98.9) 1 (1.06) 0 (0) 0.995 0.005

CYP3A5 6986A>G CYP3A5*3C Splice variant 1 (1.05) 11 (11.6) 83 (87.4) 0.068 0.932

ABCB1 3435C>T N/a I1145I 22 (25.3) 48 (55.2) 17 (19.5) 0.529 0.471

a Number represent number of patients with percentage in parenthesis; the difference in the total number of patients is due to the fact that not all samples yielded Pyrosequencing data or showed PCR amplification. b Hardy-Weinberg notation for allele frequencies (p, frequency for wild type allele and q, frequency for variant allele). c Number represents position in nucleotide sequence. d Number represents amino acid codon. e Wt, Wild type patient; Het, Heterozygous variant type patient; Var, Homozygous variant type patient; N/a, not available.

43


CrEL pharmacokinetics and CAC (Paper IV)A three-compartment model with capacity-limited elimination could describe the CrEL data from the one- and three-hour schedules. An additional linear CL component and a separate V2 were required for the 24-hour schedule to adequately describe all data. A model with linear elimination for the one- and three-hour schedule resulted in similar goodness of fi t plots, but the OFV was 18 units higher. BSA, which was the only size measure available in these data, was found to be a statistically signifi cant covariate on Vmax, V1 and V2. IIV decreased from 42% to 37%, 39% to 33% and 110% to 97% for Vmax, V1 and V2, respectively, when BSA was included. Th e corresponding increases in OFV were 19.1, 28.3 and 19.1 units, respectively, when BSA was excluded from the fi nal model. Parameter estimates with RSE are presented in Table 10 and basic goodness of fi t plots are shown in Figure 8 (left and middle panel).

0.5 1.0 1.5

Ratio

CYP2C8*4

CYP2C8*3

CYP3A5*3C

ABCB13435C>T

1.00 10.00

Predicted CCrEL

1.00

10.00

Obs

erve

d C

CrE

L (H

L/m

L)

1.00 10.00

Predicted CCrEL

1.00

10.00

Obs

erve

d C

CrE

L (H

L/m

L)

1.00 10.00

Predicted CCrEL

1.00

10.00

Obs

erve

d C

CrE

L (H

L/m

L)

Figure 8. Observed Cremophor EL concentrations (CCrEL) (Paper IV) versus population predictions based on; population parameter estimates, fi nal model, individual covariates and dose (left panel), individual predictions from EB parameter estimates based on the fi nal model, individual covariates and observed CCrEL(middle panel). Th e right panel shows predictions based on the published popu-lation model, parameter estimates (Table 1) and dose information.

Figure 7. Observed ratios (circles) of mean clearance values of unbound paclitaxel in the diff erent genotype groups (presence of variant allele/wild type) and the 95% confi dence inter-val (bars).

44

Anja Henningsson

Table 10. Parameter estimates with relative standard error(RSE,%) of the fi nal population PK model for CrEL.

Figure 9. Prediction errors (observed/predicted concentrations of Cremophor EL) for the 3-hour infusion data within the four time intervals around the scheduled sparse sampling points (1.5, 3, 5 and 21 h) of the validation data set are shown. Th e boxes represent 25th and 75th percentiles, the median (line within the box), whiskers 1.5*(inter-quartile range) and outliers (+) of prediction errors from fi nal model, learning data set (grey box, black line), fi nal model, validation data set (white box, black line), published model, learning data set (grey box, white line) and for reference the pre-diction errors from the published model, validation data set (black box, white line) are included.

1.3 - 2 h 2.3 - 4 h 4.1 - 6.1 h 19.3 - 24 hInterval

0.1

0.4

0.7

1.0

1.3

1.6

1.9

2.2

2.5

2.8

Obs

erve

d/P

redi

cted

Parameter Final model RSE (%)

V1 (L) 4.56a 4.3 BSA on V1 0.748a 16Q2 (L/h) 1.13 13 V21 & 3 hour infusion (L) 1.39b 12BSA on V21 & 3 hour infusion 1.64b 10V224 hour infusion 16.7 33 Q3 (L/h) 0.487 6.1 V3 (L) 3.55 6.5 Km (mL/L) 2.73 26 Vmax (mL/h) 0.682c 17BSA on Vmax 0.921c 27CL24-hour infusion 0.109 19 IIVV1 (CV%) 33 37d

IIVV2 (CV%) 97 22d

IIVVmax (CV%) 37 24d

Residual error: Additive (mL/L) 0.151 28 Proportional (%) 8.64 22 a V1 = 4.56* (1+0.748*(BSA-1.73)) b V2 = 1.39*(1+1.64*(BSA-1.73)) for 1 and 3-hour infusions c Vmax = 0.682*(1+0.921*(BSA-1.73)) d RSE is related to the corresponding variance term

45


Figure 10. Surface tension measurements in spiked human plasma, 1-24 hours after adding CrEL/EtOH/NaCl (empty symbols) and Taxol®infusion preparation (fi lled symbols). CrEL concentra-tiosn are shown in % (w/w). Critical aggregation concentration (CAC) is based on series F 3 hour after administration.

Predictive performance Th e prediction error for the validation data set was of the same magnitude as for the 3-hour infusion data in the learning data set with a median close to 1 (Figure 9). A clear disagreement was noted between observed and population predictions of CrEL concentration based on the previously published model72 and dose information (Figure 8, right panel). For the 3-hour infusion data there was a systematic error (Figure 9).

CAC measurementsTh e surface tension measurements of plasma spiked with CrEL/EtOH/NaCl, or the commercially available Taxol® prepared as infusion solution (paclitaxel/CrEL/EtOH/NaCl), versus start concentrations of CrEL are shown in Figure 10. No correction for degradation was done and therefore the CAC is referred to as apparent CAC. Th e degradation rate in plasma can be considered to be low, with a reported terminal half-life in vivo of 70-130 hours for 1-hour infusions135 and 37-141 hours for 3-hour infusions70. For low concentrations of CrEL the surface tension decreased substantially with increasing concentrations. Aggregates were formed as the surface tension levelled out and the apparent CAC was 0.04% (Figure 10),

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

CrEL(%)

40

45

50

55

60

65

Sur

face

Ten

sion

(m

Nm

-1)

1 hour3 hours24 hours2 hours (Taxol)4 hours (Taxol)24 hours (Taxol)

CAC = 0.04 %

46

Anja Henningsson

which corresponds to 0.39 HL/mL. Th e diff erence in surface tension between the sampling times was small at high concentrations and large at low concentrations. Th e apparent CAC was higher for the 24-hour samples, which could either be due to degradation (i.e. the real CAC is the same) or more likely due to time-dependent binding interactions with other surfaces.

Empirical models for paclitaxel-induced neutropenia (Papers I and II)In both Paper I and II, a threshold model best described the PK/PD relationship based on total plasma concentrations with threshold concentrations estimated to 0.197 HM and 0.0615 HM, respectively. Th e predicted times above this threshold causing a 50% decrease in neutrophil count from baseline were 14.4 and 7.5 hours, respectively. For unbound drug, 8.8 hours above a threshold concentration of 0.00615 HM caused a 50% decrease in neutrophils (Paper II) and no discrepancy could be made between the AUC, threshold and general models for unbound drug in Paper I (Table 11). Th e toxicity data in Paper II did not contain enough information to characterise all parameters of the general model.

Table 11. Empirical models of paclitaxel-induced neutropenia (Papers I and II).

Data Cp (Paper I) Cu (Paper I) Cp (Paper II) Cu (Paper II)

Model General AUC Threshold General AUC Threshold AUC Threshold AUC Threshold

OFV -83.6 -75.2 -83.5 -83.4 -79.2 -81.9 184.4 173.5 180.8 171.1

NEU0 (·109/L) 5.64 (7.9)

5.52 (7)

5.67 (7.6)

5.57 (8.5)

EC50 ( mol/L) 0.236 (40)

0.197 (5.3)

0.0241 (31)

0.0164 (3.8)

0.0615 (0.085)

0.00615 (3.2)

AUCE50 (a) 12.7

(38) 18.9 (21)

14.4 (18)

8.67 (33)

0.922 (19)

11.3 (18)

6.93 (14)

7.5 (16)

0.476 (13)

8.81 (20)

4.11 (64)

FIX 3 (27)

FIX FIX FIX

Additive residual error (·109/L)

0.0628 (26)

0.0533 (30)

0.0677 (19)

0.0562 (28)

0.053 (31)

0.0666 (21)

0.888 (29)

0.399 (56)

0.884 (23)

0.568 (105)

Proportional residual error (%)

11.4 (45)

8.99 (36)

10.6 (49)

12.6 (42)

9.26 43)

11.4 (49)

16.9 (52)

29.9 (22)

15 (53)

24.6 (74.8)

IIVAUCE50 (CV%) 58 (47)b

90(25)b

58(47)b

58 (47)b

79 (26)b

59(48)b

IIVBASE (CV%) 34 (32)b

41(34)b

35(31)b

38(52)b

Parameter estimates with relative standard error, RSE in % NEU0 baseline neutrophil counts; CV, coefficient of variation EC50 concentration at half maximal direct effect (general model) =threshold concentration in the threshold model, IIV interindividual variability a (h) for the general and threshold model, ( mol·h /L) for AUC model. bRSE is related to the corresponding variance term

47

PK/PD

Modelling of Paclitaxel

Table 12. Population parameter estimates with (RSE, %) for neutrophils with a linear concentration-eff ect model (Paper V).

Drug Estimation method Circ0

( 109/L)

IIVCirc0

(CV%)

MTT

(hours)

IIVMTT

(CV%)

Slope

( M-1)

Slopeu

( M-1)

IIVSlope

(CV%)

Add

( 109/L)

Prop

(%)

Docetaxel HYBRID (MTT, Slope) 5.05 (1.9) 42 (7.0)a 88.7 (2.5) 16 (24)a 0.161 (3.7) 8.58 (5.2) 429 60 (14)a 1.15 27.3

Paclitaxel FOCE INTER 5.20 (3.6) 35 (11)a 127 (2.1) 18 (30)a 0.230 (2.8) 2.21 44.2 (4.5) 43 (32)a ne 39.9

Etoposide FOCE 5.45 (7.3) 42 (20)a 135 (3.7) 14 (23)a 0.174 (6.6) 0.126 (14) 0.899 40 (78)a 0.671 45.3

DMDC HYBRID (MTT) 5.43 (3.9) 39 (16)a 123b 49 (23)a 0.160 (13) 0.782 (9.1) 0.806 63 (27)a ne 48.9

CPT-11 FOCE INTER 5.51 (3.4) 29 (19)a 113 (6.9) 29 (41)a 0.132 (9.8) 1.29 (15) 3.48 43 (61)a 0.434 34.1

Vinflunine FOCE INTER 4.72 (2.7) 41 (18)a 122 (3.7) 22 (21)a 0.162 (6.7) 0.00349 (7.8) 0.0159 41 (33)a ne 48.0

athe RSE is related to the corresponding variance term ; Add, additive residual error; Prop, proportional residual error; HYBRID (MTT) means that the FO method was used on IIVMTT; ne, not estimated; b fixed to estimate from the leukocyte analysis.

48

Anja Henningsson

Mechanism-based model of myelosuppression (Papers V and VI)All data were well described with a simple linear drug eff ect model (Slope*CDRUG), although for some of the drugs Emax models improved the fi t. Th e number of transit compartments was fi xed to three for all drugs. For paclitaxel, the system-related parameters Circ0, MTT and ~ estimated from the neutrophil data were 5.2 ·109/L, 127 hours and 0.23, respectively. Th e estimate of Slope for unbound concentrations was 44.2 HM-1. IIV were estimated to 35, 18 and 43% for Circ0, MTT and Slope, respectively. Estimated parameters for neutrophils for all studied drugs are presented in Table 12. Th e corresponding parameter estimates for leukocytes were similar but with, as expected, higher Circ0 and lower Slope values. Th e leukocyte and neutrophil observations, including successive treatment cycles of paclitaxel (Figure 11) were well described. To explore the generality of the model, the system-related parameters were fi xed to the same values for all drugs; 5 or 7 ·109/L (neutrophils or leukocytes), 125 hours and 0.17 for Circ0, MTT and ~, respectively, and IIV was fi xed to 35, 20 and 45 % for Circ0, MTT and Slope, respectively. Only the typical population value of Slope and the residual error were estimated. Th is resulted in similar Slopeu estimates (Figure 12) and individual predicted profi les (Figure 13) as when all parameters were estimated.

0

2

4

6

8

10

0

2

4

6

8

10

Neu

trop

hils

(10

9 /L)

0 50 100 150 200

Time (days)

0

2

4

6

8

10

Figure 11. Individual predicted profi les ( ) and observations (•) for the time-course of neutro-phils in three patients after at least nine successive cycles of paclitaxel treatment. Arrows denote times of administration.

49


Figure 12. Slopeu when system-related parameters were estimated versus Slopeu when system-related parameters were fi xed for leukocytes (•) and neutrophils ( ).

10-2 10-1 100 101 102

Slopeu with fixed system-related parameters

10-2

10-1

100

101

102

Slo

peu

with

est

imat

edsy

stem

-rel

ated

par

amet

ers

Paclitaxel

Vinflunine

Docetaxel

DMDC

CPT-11Etoposide

Sub-group analysis (Paper VI)In the sub-group (covariate) analysis (Paper VI), a few statistically signifi cant covariates on system-related parameters for paclitaxel were found. Patients who had received previous radiotherapy had a typically 34.8% lower baseline neutrophil count (NEU0) and MTT was decreased with 14.8 hours per 10 HM BIL. An elevated LDH predicted a decreased ~ while elevated AST levels increased ~. However, at the higher values of AST this relation caused physiologically implausible profi les. Individuals with these high covariate values also lacked data in the region of return to baseline and therefore this relation was excluded. No statistically signifi cant relations were found for the Slopeu estimate for paclitaxel, which might be due to the fact that unbound concentrations were used. For the other drugs, covariates explaining variability in drug binding were found statistically signifi cant for Slope, that is, AAG for docetaxel and ALB and BIL for etoposide. For docetaxel, SEX, previous chemotherapy and performance status were found statistically signifi cant covariates for NEU0 and, for etoposide, BIL was statistically signifi cant for NEU0 and ~. Population predictions based on the model, population PK after a standard dose and the individual sets of covariates during the fi rst three cycles of paclitaxel treatment are shown in Figure14. Only covariates included in both the PD and PK models were considered in the population PK model (i.e., BIL in CL, Paper II). All patients were assumed to have the same BSA (1.76 m2).

50

Anja Henningsson

Figure 13. Observations (DV), predictions based on typical population estimates (PRED) and indi-vidual predictions (IPRE) with estimated (e) and fi xed (f) system-related parameters for individuals selected based on their typical residual error magnitude (the individual’s average absolute weighted residual is at the population median).

Time (days)

Neu

trop

hils

(·1

09 /L)

0 5 10 15 20

0

4

8

12DVIPREePREDeIPREfPREDf

Docetaxel

0 20 40 60

0

4

8

12 Paclitaxel

0 20 40 60

0

4

8

12 Etoposide

0 5 10 15 20

0

4

8

12 DMDC

0 5 10 15 20

0

4

8

12 CPT-11

0 20 40 60

0

4

8

12 Vinflunine

From the observed sets of covariates the PK/PD model predicts that a patient who has had previous radiotherapy, a BIL of 32 HM and a LDH of 26 will suff er from grade 4 neutropenia. Th e patient will stay below 0.5·109 cells/L for approximately 4 days and below 1·109 cells/L for 7 days, and consequently, will have a high risk

51


of life-threatening infections. For the system-related parameters NEU0 and MTT, weighted means with 95%CI for the covariate eff ects are presented in Figure 15 and 16, respectively.

0 5 10 15 20

Day after dose

0.1

1.0

10.0

Pre

dict

ions

(10

9/L

)

Figure 14. Population predicted time-courses of neutrophil concentration based on the fi nal model, observed individual sets of covariates and population PK after a standard dose of paclitaxel (175 mg/m2). Horizontal lines represent the neutropenia grades 1-4 according to WHO.

*= significant†= not significant–= missing in datalegends corresponds to docetaxel,paclitaxel and etoposide in this order

Previous chemotherapy *†–

Liver methastasis †– –

Performance status*††

Previous radiotherapy–*–

Immuno/hormone therapy–†–

AAG*– –

AGE†††

ALB–††

ALP–†–

CREA–†–

AST –†–

ALT –†–

LDH –†–

Height –†–

Weight –†–

Sex *††

BIL–†*

Previous transfusion– –†

50% 100%-50%

Figure 15. Overall % change with 95% CI (whiskers) in baseline neutrophil concentration (NEU0) by diff erent covariates. Boxes represent % change from typical value for patients with extreme observed covariate values in cycle 1.

52

Anja Henningsson

Figure 16. Overall % change with 95% CI in mean transit time (MTT ) by diff erent covariates. Boxes represent % change from typical value for patients with extreme observed covariate values in cycle 1.

-50% 50%

Previous radiotherapy

Immuno/hormone therapy

AAG

AGE

ALB

BIL

ALP

CREA

AST

ALT

LDH

Weight

Height

Performance status

Liver methastasis

Previous transfusion

*= significant†= not significant–= missing in datalegends corresponds to docetaxel,paclitaxel and etoposide in this order

††–

†– –

†††

–†–

–†–

†– –

†††

–††

–*†

–†–

–†–

–†–

–†–

–†–

–†–

–†–

Sex†††

Previous chemotherapy– –†

53


Discussion

A mechanism-based model describing paclitaxel PK was successfully developed from blood concentrations, total plasma concentrations and unbound concentrations of paclitaxel and from observed concentrations of the formulation vehicle used in Taxol®, CrEL. Th e fraction unbound in plasma was highly associated with concentrations of CrEL (Figure 6). Th e model could explain the previously observed nonlinear PK of total concentrations after 3- and 24-hour infusions6,7,27 by paclitaxel binding to CrEL and to blood cells. Unbound paclitaxel concentrations were here well described with a linear disposition model. Th is is in accordance with the previously reported linear PK of paclitaxel when administered in another vehicle and the linear increase of tissue concentrations with increasing dose9. In addition, there was no statistically signifi cant relation between CL and dose in any of the data sets and thus, the non-linear PK of total paclitaxel seems at high concentrations mainly be attributed to binding to CrEL.

Th e higher fu values observed during infusion could be caused by capacity-limited binding to plasma proteins and in Paper II the non-linear binding component in plasma was associated with AAG concentrations. However, in the presence of CrEL, this association probably lack clinical relevance. It is also possible that the dependence of the non-linear binding on observed total CrEL concentrations is a simplifi cation. Th e CrEL micelles probably contain less other amphiphilic components from plasma during the infusion than at 17-72 hours post infusion and CrEL itself may be more likely to have interacted with other surfaces. Th e binding/trapping capacity of CrEL for paclitaxel might decrease with time although the observed total concentrations of CrEL are of the same magnitude. In Paper I a time-dependent binding to CrEL was tried but it did not improve the fi t (results not shown).

In vitro a saturable binding to tubulin in platelets has been seen for paclitaxel136. When normalising the in vitro values to normal levels of platelets in humans, the values are in good agreement with the estimates of Bmax,bc and the Km,bc. Unfortunately there was no information on platelet counts in our data so we could not test this hypothesis further. Th e estimated unbound CL was somewhat lower in Papers II and III compared to Paper I. IIVCL was higher in Paper III than in Paper I and II and correlations between individual estimates could be estimated in this larger population. Th e parameter estimates of volumes of distribution of the peripheral compartments were larger in Paper III. Th e diff erence in parameter estimates could perhaps be explained by the fact that the models in Paper I and II were based on unbound concentrations measured after 3-hour infusions only. Paper III included unbound concentrations

54

Anja Henningsson

from 1-, 3- and 24-hour infusions and the data thus contribute to diff erent degrees to the estimates of the volume parameters. For example, the short infusion contains more information on V1.

Th ere has in the last years been discussions regarding the dosing of anticancer drugs by BSA since for several drugs it has been shown that BSA is not correlated to CL137-139. In Paper II, BSA was found to be statistically signifi cant related to CL, V1 and V2. Th e relations were included as slope and intercept covariate models. Direct relations, as dosing by BSA suggests, performed worse but were better than not including BSA at all. In the fi nal model it was tried to replace BSA with weight or height. OFV was lowest for BSA, however, these models are not nested so statistical signifi cance can not be stated. Bilirubin was found signifi cant as covariate on clearance. Even if it is diffi cult to predict the liver function from markers like bilirubin, a high bilirubin value could be caused by liver damage or obstruction of the biliary passages which could result in a decreased elimination capacity of paclitaxel.

Th e development of new technology and the mapping of the human genome have provided possibilities to investigate the association of genetic diff erences with alteration in drug clearance. Until now most studies have focused on single-phased SNPs in specifi c genes. Currently, studies on understanding genetic variation within entire biological and pharmacological pathways are under evaluation140. In Paper III SNPs in genes of CYP450’s involved in paclitaxel metabolism as well as a transport protein (P-glycoprotein) involved in the hepatobiliary/instestinal elimination of paclitaxel were investigated. No statistical signifi cant relations were found and there may be several reasons. First, the possibility of other pathways of elimination being of such an quantitative importance so that overall CL for paclitaxel is essentially unchanged and the eff ects of the polymorphisms might not be obvious until administered with drugs inhibiting such pathways, e.g., it has been shown that 6|-hydroxypaclitaxel is not the main metabolite in all patients33,141. Second, in heterozygous patients it is possible that the expression of the unaltered allele is increased to compensate for the defi ciency of the altered allele, and few homozygous patients were available (Table 9). Taniguchi et. al. reported that metabolic activity for paclitaxel metabolism in human microsomes was related to mRNA and protein expression of the CYP3A4 and CYP2C8 and no altered activity was observed in the two livers expressing the CYP2C8*3 allele141.

Th e associations were weak (small covariate eff ects) and the 95% confi dence intervals were narrow, which suggest that the clinical relevance of the studied SNPs is limited for paclitaxel elimination. Diff erences in metabolic patterns between patients could however be interesting as the 3’-p hydroxypaclitaxel (product formed by CYP3A4) shows some cytotoxic activity while the 6|-hydroxy metabolites of paclitaxel do not34. If the CYP2C8 activity is decreased and the CYP3A4 thereby becoming a more important route of elimination, the increased amount of 3’-p hydroxypaclitaxel formed might lead to increased eff ect/toxicity.

55


It was shown in Paper II that the unbound concentrations could be well predicted from the model and observed total concentrations of paclitaxel and CrEL. To increase the usability of the developed PK models for paclitaxel it was desirable to have access to a PK model for CrEL that could be used when no CrEL concentrations have been measured or when sparse sampling schemes are to be used. A previously published population PK model72 based on CrEL data measured with a diff erent assay could not describe our data. For the three-hour infusion a systematic prediction error was evident. Th e diff erent assays used measures diff erent components of CrEL, which could contribute substantially to the variation in CrEL PK between the studies.

Th e critical micelle concentration has not previously been investigated in plasma but has been speculated to be higher than in aqueous solutions. Th e CMC is readily exceeded in solutions of Taxol® for parenteral infusions53. Paclitaxel is thus adminis-tered in micelles to the patients and to investigate whether CrEL could form aggre-gates in plasma at concentrations achieved after Taxol® infusions we estimated the critical aggregate concentration of CrEL in plasma in vitro. It was shown that surface tension was lowered with no time delay. Aggregates were formed within concentra-tion ranges observed after Taxol® administrations in vivo. Th e aggregates could be pure micelles, mixted micelles and/or larger aggregates of CrEL and lipoproteins. Even if the aggregates formed are not single component micelles they can still have a highly lipophilic core and paclitaxel is most likely distributed within such struc-tures.

Th e estimated CAC is below the lower limit of quantifi cation of the CrEL assay used in these studies why it is diffi cult to include this information in a mechanistic model. Also, the distribution and elimination of this heterogeneous substance could be very complex and only speculations about its nature could be done at present. Th e size and nature of the aggregates might determine their elimination pathways and can then have diff erent half-lives depending on these. Hydrolysis of lipoprotein bound triglycerides is a fast process142,143, Kupff er cells in the liver and spleen could eliminate larger aggregates similar to elimination of liposomes144 and as previously suggested a capacity-limited hydrolysis by esterases in plasma could be responsible for the elimination72. Diff erent elimination pathways for the diff erent components could cause the schedule dependence seen. Similarly, diff erent distribution patterns could possibly explain the larger volume of distribution for the prolonged infusions. Th e larger microdroplets observed in vitro53 might not be formed clinically at low concentrations and low infusion rates. Szebeni et al. also suggested that aggregates of a certain size were required to cause complement C3 activation53 and therefore this may explain why less hypersensitivity reactions are observed after lower infusion rates.

With the new PK information from unbound concentrations we wanted to re-evalu-ate the earlier published PK/PD threshold and general models for paclitaxel-induced neutropenia and compare between models based on unbound and total concentra-tions. Th e estimated threshold concentrations based on total concentrations of pacli-

56

Anja Henningsson

taxel were in good agreement with previously reported values24,27 but the time above the threshold concentrations were somewhat shorter, which could be explained by the diff erence in sigmoidicity factors. In Paper II the objective function value was lower for toxicity relations based on unbound concentrations and a statistically sig-nifi cant improvement was found when Cu was added as a predictor of toxicity to Cp (p=0.032), but not when tried the other way around (p=0.147), suggesting that Cu is a better predictor of toxicity than Cp. However, the diff erence was small and no real improvement of goodness of fi t plots could be seen.

Since it is not only the degree of neutropenia, but also the duration, that is related to the risk of infections145, modelling the whole time course of neutropenia is to be preferred. Physiology-based models are likely to be have a higher predictive capac-ity146 and especially models that separate system-related from drug-specifi c param-eters. Th e semi-physiological model could adequately describe all investigated drugs. System-related parameters showed consistency across drugs but were not completely void of drug infl uence. Th e fact that only the fi rst course of treatment was avail-able for docetaxel might contribute to the somewhat shorter MTT estimated for docetaxel in addition to the fact that diff erent drugs can have somewhat diff er-ent toxicity profi les. For example, nitrosureas have a more prolonged toxicity pro-fi le than vinblastine 100. Overall, the MTTs estimated were shorter than previously reported in healthy volunteers73. Th is is reasonable since a decrease in leukocytes induces release of G-CSF and interleukins which apart from promoting cell division also shortens transit time147. In our model all dividing cells are in the same compart-ment averaging out potential diff erence in drug eff ect on the diff erent proliferative cell types. To refl ect the repopulation of white blood cells a self-renewal and a feed-back mechanism were included in the model. A decrease in circulating cells induces diff erent growth factors such as G-CSF, which stimulates proliferation, diff eren-tiation and has a role in the survival of immature neutrophils through promotion of anti-apoptotic mechanisms. Homeostasis is maintained by a negative feedback where increasing number of circulating neutrophils results in an increased number of receptors for G-CSF and thereby increases clearance of G-CSF78. Th e parameter gamma could be regarded as a measure of the ability to recover from myelosuppres-sion and refl ects the fact that the leukocytes when highly stimulated can get above its normal baseline value before it returns. Estimated half-lives from the transit rate constants were between 15 and 24 hours. Th e fi ts were not improved by estimating the half-life of neutrophils in the circulation as a separate parameter or by fi xing it to a literature value of 6.7 hours75, suggesting that information on that parameter is lacking and the half-life in the circulation is not the rate limiting step for the delay in drug eff ect.

Both leukocytes and neutrophils could be described with the same model where the system-related parameters were approximately the same. Neutrophil baseline values were approximately 30% lower than for leukocytes, in agreement with the diff erence in the observed baseline values in the data sets. Th e larger Slope estimates for neutrophils points to the diff erence in sensitivity in neutrophil and lymphocyte

57


progenitors. Th e diff erent Slope values between paclitaxel and docetaxel shows that docetaxel is more toxic to the progenitor cells, which has previously been shown in vitro148.

Identifi cation of sub-groups in the population that are more sensitive to chemotherapy is of great interest to prevent excessive myelosuppression and delay of following cycles. Th e identifi cation of important drug specifi c eff ects of covariates like ALB and BIL on etoposide and AAG for docetaxel, in line with previous publications99,149 shows that relevant covariates for drug-specifi c eff ects can be identifi ed with this model. Minami et al. investigated the total plasma concentration and found a negative correlation between age and the toxic eff ect on paclitaxel exposure producing 50% inhibition of leukocyte production. Th ey discussed that diff erence in plasma binding with age as an alternative explanation to diff erences in sensitivity of the bone marrow could not be excluded . In this study no statistically signifi cant relations were found for the drug-related parameter for paclitaxel, which could partly be explained by the fact that unbound drug concentrations were used81. For the vast majority of covariates investigated, including height, weight, age, creatinine and several liver enzymes, we did not fi nd a statistically signifi cant infl uence on the system-related parameters. Th e across-drug comparison revealed that the confi dence intervals did not include relevant changes for these covariates when defi ned as +30% overall change (except liver enzymes on NEU0 where we generally investigated a large distribution of each covariate). For paclitaxel, bilirubin concentrations were statistically signifi cant negatively related to MTT. High bilirubin concentrations predicted a reduction in MTT which was of the same magnitude as previously found when G-CSF was given as concomitant treatment compared to when given without G-CSF treatment150.Th is relation and the other signifi cant relations found with system-related parameters (i.e. a lower NEU0 for females and previously chemo- (docetaxel) or radiotherapy (paclitaxel) treated patients or patients with unrestricted performance status (docetaxel)), need to be confi rmed with more data across diff erent drugs.

58

Anja Henningsson

Conclusions

For paclitaxel, the nonlinear PK has previously been described by empirical models without mechanistic foundation (saturable transport or tissue binding) that have not accounted for the events known to infl uence paclitaxel binding in vitro and CrEL characteristics. Th e developed mechanism-based model combined mechanistic insights gained from in vitro experiments with the clinical pharmacokinetics. Model components included components that corresponded to known mechanisms such as micellar trapping of paclitaxel by CrEL and saturable binding to platelets. Th us, a model has been developed that can jointly explain clinical observations of total, unbound and blood concentration-time profi les of paclitaxel and in vitro results. Further, the model was shown to have good predictive properties.

Th e developed population PK model for CrEL could be used to describe and predict CrEL concentrations after Taxol® administrations, which may be most useful when no or sparse CrEL concentration data is available.

Th e apparent CAC in plasma in vitro is within the range of concentrations observed after Taxol® infusions in vivo. Th is gives support to the theory of CrEL causing the nonlinear PK of paclitaxel in plasma either by micelle entrapment or by providing a preferable environment in larger lipophilic aggregates.

Th e studied SNPs in enzymes and transport proteins responsible for paclitaxel elim-ination are not likely to have a clinically relevant impact on overall unbound clear-ance for paclitaxel in Caucasians.

Haematological toxicity may be better related to unbound concentrations. PK/PD relations for myelosuppression based on total concentrations in plasma are not likely to be valid in the absence of CrEL, which should be considered when new routes of administration or new formulations of paclitaxel are planned.

Th e model of chemotherapy-induced myelosuppression is semi-physiologic since it contains transit compartments that mimic the maturation chain in the bone marrow, a self-renewal and a feedback mechanism. Th e model is applicable on varying doses and show consistency of system-related parameters across drugs. It can also be used on sparse and small data sets by keeping the system-related parameters fi xed. Th is will allow quantitative measures of myelosuppression from data derived in early phase I studies, which would aid design of later studies and could permit faster dose

59


escalation. Th is model has, since it was published, been shown to be a helpful tool in drug development of new formulations and prodrugs of paclitaxel151 and design of clinical trials. Th e models usefulness is further evident from the fact that it was used in the development of pemetrexed, recently approved for use in the USA. Further, information in the drug label regarding the clinical use of pemetrexed is based on this model (www.alimta.com). Th e model has also been shown to be extendable to combination drug therapy152.

Th e knowledge of patient characteristics signifi cantly infl uencing system and drug-related parameters may help to identify patients or subgroups of patients with higher risk of (life-threatening) neutropenia prior to start of the chemotherapy and thus individualize the dosing regimen accordingly. Th e awareness of patient factors that are not associated with system-related parameters may be of value when analysing new data sets or even with new compounds in clinical trials. Covariate analyses in future may thus be more effi cient.

In summary, these two models have connected the clinical behaviour of paclitaxel with knowledge about the mechanisms for drug distribution and immune system homeostasis. Th ey have shown appropriate properties regarding data description, prospective prediction and have already been of value in drug development.

60

Anja Henningsson

Populärvetenskaplig sammanfattning

Paklitaxel (Taxol®) är ett cellgift som används mot en rad olika cancertyper t ex bröstcancer, äggstockscancer och icke-småcellig lungcancer. Paklitaxel används framför allt som understödjande behandling efter kirurgi eller strålbehandling för att förhindra spridning av tumören och förlänga överlevnaden hos patienterna. Det kan även användas för att minska tumörbördan (ge symptomlindring) vid avancerad cancer. Även normala celler påverkas av paklitaxel och den dosbegränsande biverkningen är oftast en sänkning i vita blodkroppar som gör att patienterna blir infektionskänsliga. En överdriven sänkning i vita blodkroppar kan innebära en risk för livshotande infektioner och att nästa behandling fördröjs samt att dosen eventuellt måste sänkas. Detta kan innebära att man inte uppnår önskad eff ekt på tumören eftersom tumörcellerna växer till mellan behandlingstillfällena. Om vi med hjälp av matematiska modeller kan förutsäga och beskriva hur mycket biverkningar en viss dos ger upphov till i en viss individ, kan vi förhoppningsvis ge varje individ en så bra dos som möjligt vid första tillfället och/eller snabbare komma fram till vilken dos som är bäst. Vi har utvecklat farmakokinetiska modeller som beskriver samband mellan dos och halt paklitaxel i plasma eller blod. Modellerna är baserade på olika kända mekanismer för paklitaxels bindning i plasma och blod. Den ickelinjäritet som tidigare visats för paklitaxels farmakokinetik kunde till största delen förklaras av den micellbildare som tillsätts för att kunna lösa paklitaxel i vatten. Vi har också utvecklat en modell för sänkningen av vita blodkroppar som orsakas av paklitaxel och andra cellgifter. Den matematiska modellen efterliknar olika steg i de vita blodkropparnas reglering, bildning och mognad. Samband mellan olika patientkarakteristika t ex ålder, kön, kroppsyta, olika levervärden och modellparametrar har också undersökts. Modellerna bidrar till att öka förståelsen för paklitaxels farmakokinetik och biverkningar. Modellerna skulle kunna användas för dosindividualisering och har sedan de publicerats använts i läkemedelsutveckling av andra cellgifter och nya beredningsformer för paklitaxel.

61


Comments on my contribution

In paper I and II, I performed the PK/PD modelling (and did parts of the bioanalysis in paper II) and was responsible for writing the manuscripts.

In paper III, I performed the pharmacokinetic analysis of paclitaxel, statistical evaluation of the associations with genotypes and shared responsibility for writing the manuscript.

In paper IV, I was responsible for the pharmacokinetic modelling of the data and writing the manuscript.

In paper V, I was responsible for the model development of paclitaxel myelosuppression and partly responsible for writing the manuscript.

In paper VI, I was responsible for developing the covariate model for etoposide and was partly responsible for writing the manuscript.

62

Anja Henningsson

Acknowledgements

Th is work was performed at the Department of Pharmaceutical Sciences, Division of Pharmacokinetics and Drug Th erapy

I would like to express my sincere gratitude to:Professor Mats O. Karlsson, my supervisor, for being there with a million bright ideas on how to master obstacles and best present the results. Th ank you for being supportive when I needed it, for caring about me and my family and thank you for effi ciently taking away my panic. Th anks to your collaborations with other groups, I got clinical data to work with (no rat experiments) for which I am very grateful. I have really enjoyed working with you.

Professor Margareta Hammarlund-Udenaes, head of the division for providing excellent working facilities, for fun discussions about kids in the coff ee-room and for letting us know that our work is valued internationally –our place on the map.

Professor Lennart Paalzow, former head of the division, for your contributions to the development of PKPD and for providing excellent working facilities.

Dr Agneta Freijs, your true interest in the patients and belief that there were really things that could be done to improve drug treatments captured my interest. I still remember the phone call I had from you saying “now we have got money for you. When can you start?” Unfortunately you were not able to accompany me all the way.

Dr Lena Friberg, my co-supervisor, co-author and dear friend thank you for always being positive and for your invaluable comments on the manuscripts and summary, without you this work would not have been fi nished on time.

Dr Marie Sandström, co-author and friend, for your passion for the patients that inspired me to start my PhD work. You always take time to listen to small and large things, you are probably the only one who knew how things really were at times.

Dr Alex Sparreboom, co-author and data provider, it has been a pleasure working with you and your data. Th ank you for being so genuinely interested and encouraging, for all the help with the manuscripts and for our scientifi c discussions. Th ank you also for taking such good care of me while I was visiting your group in Rotterdam, it was a really great experience.

Dr Siv Jönsson, my former room-mate, for your support and very helpful advice in writing, modelling and for being a wonderful friend. Th ank you for being such a good travelling companion and for caring about me and my family. I think that for not being a supervisor, you have done a lot of supervising, especially during the last period. Th ank you for all help with proof-reading and all other things. I have enjoyed our discussions a lot. I don’t know how to thank you enough.

63


My co-authors Dr Charlotte Kloft and Johan Wallin for your really nice work on Paper VI and for keeping the spirit up, fun trips with not so nice food in Switzerland (at least if you are pregnant and suff er from motion sickness).

Co-author Hugo Maas, for your work on the myelosuppression model and for being a good friend. I hope I will see you again soon.

Co-author Mats Silvander for your late hours measuring the CAC and for your help with the manuscript.

My co-authors, Peter Nygren, Rolf Larsson, Jonas Berg, Laurent Nguyen, Jaap Verveij, Walter J. Loos, Lucia Viganò, Luca Gianni, Alberta Locatelli, Sharon Marsh, Adam Garsa, Klaus Mross, Stephan Mielke, and Howard L McLeod. Th ank you for your contribution.

Dr Karin Tunblad, my former room-mate, for being a wonderful friend, for listening and giving us all many great laughs, for going skiing without “valla” in Switzerland, being such a party-girl and for keeping secrets secret. Luckily we will keep working in the same corridor.

Jörgen Bengtsson and Elisabet Nielsen for being the best of friends, colleagues and neighbours. For taking such good care of Jakob and Ylva, making it possible for me to attend Voces rehearsals during the fall. Th ank you Jörgen, for bringing sunshine to the division and making everyone smile. “Ta fram det bästa humör du har var morgon när solen går upp…..”

Dr Niclas Jonsson for the continuous development of the Xpose program, for your help on PC issues and for providing a nice working atmosphere during your time as head of the division.

Britt Jansson and Karin Sjöberg for your work with the paclitaxel HPLC analysis, Eric Breuwer for assistance with the plasma analysis of CrEL.

Marianne Danersund, Marina Rönngren, Anna Sofi a and Kjell Berggren for always giving a helping hand. Birgit and Britt again for all the ”goda kakor” and little chats.

Lars Lindbom for all your time spent on the cluster issues making things work for the rest of us, for providing us with the time saving tools in the PsN tool kit.

Jakob R for handling cluster issues, for showing interest in all research and being such a party guy.

Pontus for assisting in PsN issues, taking care of the cluster and back-up.

Maria for being one of those happy persons, for help with the perl script. I hope I can be of assistance for you some time in the near future.

Hanna S for pulling me out of the dark moments. Th ank you also for fun days at your summer house.

Karin S for being a good friend and I hope we will have more time to meet after work now. Jessica and Kajsa for all the clothes for the kids and fun discussions in the coff ee room.

64

Anja Henningsson

Poppi, my present room-mate, for all the fun discussions and for putting up with me and my messy desk.

Ulrika S, Rikard, Anders V, Mats M, Kristin, Emma B, Andy, Anubha and Rada, for many enjoyable scientifi c and non-scientifi c discussions, for fun party company and week-end-work company and a special thank you to Andy for reading and commenting on the summary.

Ann-Marie, Anna, Jonatan, Emma L, Ulrika E, Eva, Hanna, Angelica, Gunnel, Sven, Anders G, Doaa, Eman, Daniel, and Justin for making coff ee-breaks and parties so enjoyable.

Dr Ulrika Wählby, Bengt H and P-H Z, for many fun parties, trips and conferences and all former colleagues at the division, especially Lena B, Toufi gh, René, Jacob B, Lotta, Håkan, Johan, Linus and Rujia for making my fi rst years so fun and interesting.

Mats Gullberg for commenting on the parts of the thesis where I am completely “novis”.

Magnus Jansson, for all the help on the layout of the manuscripts (dom är väldigt snygga!!) and putting the pieces of this thesis together.

Erik Bongcam-Rudloff for the excellent Figure 17.

To all cancer patients who participated in the studies.

Th e Swedish Cancer Society, the Swedish Academy of Pharmaceutical Sciences and the IF foundation for fi nancial support.

Henke, Millan, David, Patricia, Karin W, Malin och Karl för att ni är underbara vänner och tagit hand om oss lite extra.

Jane, för att du är så omtänksam och har en förmåga att alltid få mig att må mycket bättre.

Elisabet, Bodil, Karin och Anna-Karin för trevliga middagar och roliga partyn. Jag ser fram emot att ha mer tid att umgås.

Anneli, Petra och Sofi e, för att ni förblir så nära fast vi ses så sällan.

Mina underbara föräldrar som har stöttat mig i vått och torrt. Tack kära mamma för att du kommit och gjort fi nt i vårt hem och för att ni fi nns där för oss.

Mina älskade barn, Jakob och Ylva, som gör mitt liv så underbart. För att ni får mig att bli lycklig av att bli väckt kl. 5 med ett ”hejsan mamma” trots att jag lagt mig bara några timmar tidigare. Jag hoppas att ni snart kan säga något annat än att ”mamma jobbar” eller ”varför måste du jobba hela tiden”.

Till sist min älskade man, Anders, som har tagit hand om våra barn och ringt och hejat på, som stått ut med min otroliga disträhet och hjälpt till med läsning av valda delar av avhandlingen. Snart ska våra barn ha två föräldrar igen och du får din fru tillbaka.

65


References

1. Nygren P: What is cancer chemotherapy? Acta Oncol 40:166-74, 20012. Schabel FM, Jr.: Th e use of tumor growth kinetics in planning “curative” chemotherapy of

advanced solid tumors. Cancer Res 29:2384-9, 19693. Schabel FM, Jr.: Concepts for systemic treatment of micrometastases. Cancer 35:15-24, 19754. Skipper HE, Perry S: Kinetics of normal and leukemic leukocyte populations and relevance to

chemotherapy. Cancer Res 30:1883-97, 19705. McKinnel RG, Parchment RE, Perantoni AO, et al: Th e biological basis of cancer. Cambridge,

Cambridge university press, 19986. Karlsson MO, Molnar V, Freijs A, et al: Pharmacokinetic models for the saturable distribution

of paclitaxel. Drug Metab Dispos 27:1220-3, 19997. Sonnichsen DS, Hurwitz CA, Pratt CB, et al: Saturable pharmacokinetics and paclitaxel phar-

macodynamics in children with solid tumors. J Clin Oncol 12:532-8, 19948. Kearns CM, Gianni L, Egorin MJ: Paclitaxel pharmacokinetics and pharmacodynamics. Semin

Oncol 22:16-23., 19959. Sparreboom A, van Tellingen O, Nooijen WJ, et al: Nonlinear pharmacokinetics of paclitaxel in

mice results from the pharmaceutical vehicle Cremophor EL. Cancer Res 56:2112-5, 199610. van Tellingen O, Huizing MT, Panday VR, et al: Cremophor EL causes (pseudo-) non-linear

pharmacokinetics of paclitaxel in patients. Br J Cancer 81:330-5, 199911. Cragg GM, Newman DJ: A tale of two tumor targets: topoisomerase I and tubulin. Th e Wall

and Wani contribution to cancer chemotherapy. J Nat Prod 67:232-44, 200412. Hassan S, Dhar S, Sandstrom M, et al: Cytotoxic activity of a new paclitaxel formulation,

Pacliex, in vitro and in vivo. Cancer Chemother Pharmacol 55:47-54, 200513. Soepenberg O, Sparreboom A, de Jonge MJ, et al: Real-time pharmacokinetics guiding clinical

decisions; phase I study of a weekly schedule of liposome encapsulated paclitaxel in patients with solid tumours. Eur J Cancer 40:681-8, 2004

14. Meerum Terwogt JM, ten Bokkel Huinink WW, Schellens JH, et al: Phase I clinical and phar-macokinetic study of PNU166945, a novel water-soluble polymer-conjugated prodrug of pacli-taxel. Anticancer Drugs 12:315-23, 2001

15. Harries M, O’Donnell A, Scurr M, et al: Phase I/II study of DHA-paclitaxel in combination with carboplatin in patients with advanced malignant solid tumours. Br J Cancer 91:1651-5, 2004

16. Horwitz SB: Taxol (paclitaxel): mechanisms of action. Ann Oncol 5 Suppl 6:S3-6, 199417. Au JL, Li D, Gan Y, et al: Pharmacodynamics of immediate and delayed eff ects of paclitaxel: role

of slow apoptosis and intracellular drug retention. Cancer Res 58:2141-8, 199818. Spencer CM, Faulds D: Paclitaxel. A review of its pharmacodynamic and pharmacokinetic

properties and therapeutic potential in the treatment of cancer. Drugs 48:794-847, 199419. Crown J, O’Leary M, Ooi WS: Docetaxel and paclitaxel in the treatment of breast cancer: a

review of clinical experience. Oncologist 9 Suppl 2:24-32, 200420. Nowak AK, Wilcken NR, Stockler MR, et al: Systematic review of taxane-containing versus

non-taxane-containing regimens for adjuvant and neoadjuvant treatment of early breast cancer. Lancet Oncol 5:372-80, 2004

21. Johnston SR: Ovarian cancer: review of the National Institute for Clinical Excellence (NICE) guidance recommendations. Cancer Invest 22:730-42, 2004

66

Anja Henningsson

22. Rigas JR: Taxane-platinum combinations in advanced non-small cell lung cancer: a review. Oncologist 9 Suppl 2:16-23, 2004

23. Eisenhauer EA, Vermorken JB: Th e taxoids. Comparative clinical pharmacology and therapeu-tic potential. Drugs 55:5-30, 1998

24. Huizing MT, Keung AC, Rosing H, et al: Pharmacokinetics of paclitaxel and metabolites in a randomized comparative study in platinum-pretreated ovarian cancer patients. J Clin Oncol 11:2127-35, 1993

25. Hurwitz CA, Relling MV, Weitman SD, et al: Phase I trial of paclitaxel in children with refrac-tory solid tumors: a Pediatric Oncology Group Study. J Clin Oncol 11:2324-9, 1993

26. Schiller JH, Storer B, Tutsch K, et al: A phase I trial of 3-hour infusions of paclitaxel (Taxol) with or without granulocyte colony-stimulating factor. Semin Oncol 21:9-14, 1994

27. Gianni L, Kearns CM, Giani A, et al: Nonlinear pharmacokinetics and metabolism of paclitaxel and its pharmacokinetic/pharmacodynamic relationships in humans. J Clin Oncol 13:180-90, 1995

28. Sparreboom A, van Zuylen L, Brouwer E, et al: Cremophor EL-mediated alteration of paclitaxel distribution in human blood: clinical pharmacokinetic implications. Cancer Res 59:1454-7, 1999

29. Kumar GN, Walle UK, Bhalla KN, et al: Binding of taxol to human plasma, albumin and alpha 1-acid glycoprotein. Res Commun Chem Pathol Pharmacol 80:337-44, 1993

30. Brouwer E, Verweij J, De Bruijn P, et al: Measurement of fraction unbound paclitaxel in human plasma. Drug Metab Dispos 28:1141-5., 2000

31. Monsarrat B, Royer I, Wright M, et al: Biotransformation of taxoids by human cytochromes P450: structure-activity relationship. Bull Cancer 84:125-33, 1997

32. Monsarrat B, Chatelut E, Royer I, et al: Modifi cation of paclitaxel metabolism in a cancer patient by induction of cytochrome P450 3A4. Drug Metab Dispos 26:229-33, 1998

33. Sonnichsen DS, Liu Q, Schuetz EG, et al: Variability in human cytochrome P450 paclitaxel metabolism. J Pharmacol Exp Th er 275:566-75, 1995

34. Sparreboom A, Huizing MT, Boesen JJ, et al: Isolation, purifi cation, and biological activity of mono- and dihydroxylated paclitaxel metabolites from human feces. Cancer Chemother Phar-macol 36:299-304, 1995

35. Walle T, Walle UK, Kumar GN, et al: Taxol metabolism and disposition in cancer patients. Drug Metab Dispos 23:506-12, 1995

36. Soyama A, Saito Y, Hanioka N, et al: Non-synonymous single nucleotide alterations found in the CYP2C8 gene result in reduced in vitro paclitaxel metabolism. Biol Pharm Bull 24:1427-30, 2001

37. Dai D, Zeldin DC, Blaisdell JA, et al: Polymorphisms in human CYP2C8 decrease metabolism of the anticancer drug paclitaxel and arachidonic acid. Pharmacogenetics 11:597-607, 2001

38. Kerdpin O, Elliot DJ, Boye SL, et al: Diff erential Contribution of Active Site Residues in Sub-strate Recognition Sites 1 and 5 to Cytochrome P450 2C8 Substrate Selectivity and Regioselec-tivity. Biochemistry 43:7834-42, 2004

39. Bahadur N, Leathart JB, Mutch E, et al: CYP2C8 polymorphisms in Caucasians and their relationship with paclitaxel 6alpha-hydroxylase activity in human liver microsomes. Biochem Pharmacol 64:1579-89, 2002

40. Niemi M, Leathart JB, Neuvonen M, et al: Polymorphism in CYP2C8 is associated with reduced plasma concentrations of repaglinide. Clin Pharmacol Th er 74:380-7, 2003

41. Garcia-Martin E, Martinez C, Tabares B, et al: Interindividual variability in ibuprofen phar-macokinetics is related to interaction of cytochrome P450 2C8 and 2C9 amino acid polymor-phisms. Clin Pharmacol Th er 76:119-27, 2004

42. Weise A, Grundler S, Zaumsegel D, et al: Development and evaluation of a rapid and reliable method for cytochrome P450 2C8 genotyping. Clin Lab 50:141-8, 2004

43. Yasar U, Lundgren S, Eliasson E, et al: Linkage between the CYP2C8 and CYP2C9 genetic polymorphisms. Biochem Biophys Res Commun 299:25-8, 2002

67


44. Nakajima M, Fujiki Y, Noda K, et al: Genetic polymorphisms of CYP2C8 in Japanese popula-tion. Drug Metab Dispos 31:687-90, 2003

45. Koch I, Weil R, Wolbold R, et al: Interindividual variability and tissue-specifi city in the expres-sion of cytochrome P450 3A mRNA. Drug Metab Dispos 30:1108-14, 2002

46. Hustert E, Haberl M, Burk O, et al: Th e genetic determinants of the CYP3A5 polymorphism. Pharmacogenetics 11:773-9, 2001

47. Kuehl P, Zhang J, Lin Y, et al: Sequence diversity in CYP3A promoters and characterization of the genetic basis of polymorphic CYP3A5 expression. Nat Genet 27:383-91, 2001

48. Foti RS, Fisher MB: Importance of patient selection when determining the signifi cance of the CYP3A5 polymorphism in clinical trials. Pharmacogenomics J 4:362-4, 2004

49. Hoff meyer S, Burk O, von Richter O, et al: Functional polymorphisms of the human multidrug-resistance gene: multiple sequence variations and correlation of one allele with P-glycoprotein expression and activity in vivo. Proceedings of the National Academy of Sciences of the United States of America 97:3473-8, 2000

50. Sparreboom A, Danesi R, Ando Y, et al: Pharmacogenomics of ABC transporters and its role in cancer chemotherapy. Drug Resist Updat 6:71-84, 2003

51. Sparreboom A, van Tellingen O, Huizing MT, et al: Determination of polyoxyethyleneglycerol triricinoleate 35 (Cremophor EL) in plasma by pre-column derivatization and reversed-phase high- performance liquid chromatography. J Chromatogr B Biomed Appl 681:355-62, 1996

52. Meyer T, Waidelich D, Frahm AW: Polyoxyethylene-Delta(9,11)-didehydrostearate and glyc-erol-polyoxyethylene-Delta(9,11)-didehydrostearate: two new components of the non-ionic emulsifi er Cremophor EL. J Pharm Biomed Anal 30:263-71, 2002

53. Szebeni J, Alving CR, Savay S, et al: Formation of complement-activating particles in aqueous solutions of Taxol: possible role in hypersensitivity reactions. Int Immunopharmacol 1:721-35., 2001

54. Gelderblom H, Verweij J, Nooter K, et al: Cremophor EL: the drawbacks and advantages of vehicle selection for drug formulation. Eur J Cancer 37:1590-8., 2001

55. Szebeni J, Muggia FM, Alving CR: Complement activation by Cremophor EL as a possible con-tributor to hypersensitivity to paclitaxel: an in vitro study. J Natl Cancer Inst 90:300-6., 1998

56. de Groen PC, Aksamit AJ, Rakela J, et al: Central nervous system toxicity after liver transplanta-tion. Th e role of cyclosporine and cholesterol. N Engl J Med 317:861-6, 1987

57. Kuroi K, Shimozuma K: Neurotoxicity of taxanes: symptoms and quality of life assessment. Breast Cancer 11:92-9, 2004

58. Windebank AJ, Blexrud MD, de Groen PC: Potential neurotoxicity of the solvent vehicle for cyclosporine. J Pharmacol Exp Th er 268:1051-6., 1994

59. Kessel D, Woodburn K, Decker D, et al: Fractionation of Cremophor EL delineates compo-nents responsible for plasma lipoprotein alterations and multidrug resistance reversal. Oncol Res 7:207-12, 1995

60. Sykes E, Woodburn K, Decker D, et al: Eff ects of Cremophor EL on distribution of Taxol to serum lipoproteins. Br J Cancer 70:401-4., 1994

61. Woodcock DM, Jeff erson S, Linsenmeyer ME, et al: Reversal of the multidrug resistance phe-notype with cremophor EL, a common vehicle for water-insoluble vitamins and drugs. Cancer Res 50:4199-203, 1990

62. Schuurhuis GJ, Broxterman HJ, Pinedo HM, et al: Th e polyoxyethylene castor oil Cremophor EL modifi es multidrug resistance. Br J Cancer 62:591-4, 1990

63. Webster LK, Cosson EJ, Stokes KH, et al: Eff ect of the paclitaxel vehicle, Cremophor EL, on the pharmacokinetics of doxorubicin and doxorubicinol in mice. Br J Cancer 73:522-4, 1996

64. Rischin D, Webster LK, Millward MJ, et al: Cremophor pharmacokinetics in patients receiving 3-, 6-, and 24-hour infusions of paclitaxel. J Natl Cancer Inst 88:1297-301., 1996

65. Sparreboom A, Loos WJ, Verweij J, et al: Quantitation of Cremophor EL in human plasma samples using a colorimetric dye-binding microassay. Anal Biochem 255:171-5, 1998

66. Brouwer E, Verweij J, Hauns B, et al: Linearized colorimetric assay for cremophor EL: applica-tion to pharmacokinetics after 1-hour paclitaxel infusions. Anal Biochem 261:198-202., 1998

68

Anja Henningsson

67. Kunkel M, Meyer T, Bohler J, et al: Titrimetric determination of Cremophor EL in aqueous solutions and biofl uids: part 2: ruggedness of the method with respect to biofl uids. J Pharm Biomed Anal 21:911-22, 1999

68. van Zuylen L, Gianni L, Verweij J, et al: Inter-relationships of paclitaxel disposition, infusion duration and cremophor EL kinetics in cancer patients. Anticancer Drugs 11:331-7., 2000

69. Meerum Terwogt J, van Tellingen O, Nannan Panday VR, et al: Cremophor EL pharmaco-kinetics in a phase I study of paclitaxel (Taxol) and carboplatin in non-small cell lung cancer patients. Anticancer Drugs 11:687-94., 2000

70. Sparreboom A, Verweij J, van der Burg ME, et al: Disposition of Cremophor EL in humans limits the potential for modulation of the multidrug resistance phenotype in vivo. Clin Cancer Res 4:1937-42., 1998

71. ten Tije AJ, Verweij J, Loos WJ, et al: Pharmacological eff ects of formulation vehicles : implica-tions for cancer chemotherapy. Clin Pharmacokinet 42:665-85, 2003

72. van den Bongard HJ, Mathot RA, van Tellingen O, et al: A population analysis of the phar-macokinetics of Cremophor EL using nonlinear mixed-eff ect modelling. Cancer Chemother Pharmacol 50:16-24, 2002

73. Dancey JT, Deubelbeiss KA, Harker LA, et al: Neutrophil kinetics in man. J Clin Invest 58:705-15, 1976

74. Athens JW, Raab SO, Haab OP, et al: Leukokinetic studies. III. Th e distribution of granulocytes in the blood of normal subjects. J Clin Invest 40:159-164, 1961

75. Cartwright GE, Athens JW, Wintrobe MM: Th e kinetics of granulopoiesis in normal man. Blood 24:780-803, 1964

76. Mauer AM, Athens JW, Aschenbrucker H, et al: Leukokinetic studies. II. A method for label-ing granulocytes in vitro with radioactive diisopropylfl uorophosphate (DFP32). J Clin Invest 39:1481-1486, 1960

77. Suwa T, Hogg JC, English D, et al: Interleukin-6 induces demargination of intravascular neu-trophils and shortens their transit in marrow. Am J Physiol Heart Circ Physiol 279:H2954-60, 2000

78. Barreda DR, Hanington PC, Belosevic M: Regulation of myeloid development and function by colony stimulating factors. Dev Comp Immunol 28:509-54, 2004

79. Karlsson MO, Molnar V, Bergh J, et al: A general model for time-dissociated pharmacokinetic-pharmacodynamic relationship exemplifi ed by paclitaxel myelosuppression. Clin Pharmacol Th er 63:11-25, 1998

80. Minami H, Sasaki Y, Saijo N, et al: Indirect-response model for the time course of leukopenia with anticancer drugs. Clin Pharmacol Th er 64:511-21, 1998

81. Minami H, Sasaki Y, Watanabe T, et al: Pharmacodynamic modeling of the entire time course of leukopenia after a 3-hour infusion of paclitaxel. Jpn J Cancer Res 92:231-8., 2001

82. Krzyzanski W, Jusko WJ: Multiple-pool cell lifespan model of hematologic eff ects of anticancer agents. J Pharmacokinet Pharmacodyn 29:311-37, 2002

83. Zamboni WC, D’Argenio DZ, Stewart CF, et al: Pharmacodynamic model of topotecan-induced time course of neutropenia. Clin Cancer Res 7:2301-8, 2001

84. Friberg LE, Brindley CJ, Karlsson MO, et al: Models of schedule dependent haematological toxicity of 2’-deoxy-2’- methylidenecytidine (DMDC). Eur J Clin Pharmacol 56:567-74, 2000

85. Friberg LE, Freijs A, Sandström M, et al: Semiphysiological model for the time course of leuko-cytes after varying schedules of 5-fl uorouracil in rats. J Pharmacol Exp Th er 295:734-40, 2000

86. Verweij J, Clavel M, Chevalier B: Paclitaxel (Taxol) and docetaxel (Taxotere): not simply two of a kind. Ann Oncol 5:495-505, 1994

87. Bruno R, Vivler N, Vergniol JC, et al: A population pharmacokinetic model for docetaxel (Tax-otere): model building and validation. J Pharmacokinet Biopharm 24:153-72, 1996

88. Urien S, Barre J, Morin C, et al: Docetaxel serum protein binding with high affi nity to alpha 1-acid glycoprotein. Invest New Drugs 14:147-51, 1996

89. Okouneva T, Hill BT, Wilson L, et al: Th e eff ects of vinfl unine, vinorelbine, and vinblastine on centromere dynamics. Mol Cancer Th er 2:427-36, 2003

69


90. Ngan VK, Bellman K, Hill BT, et al: Mechanism of mitotic block and inhibition of cell prolif-eration by the semisynthetic Vinca alkaloids vinorelbine and its newer derivative vinfl unine. Mol Pharmacol 60:225-32, 2001

91. Ngan VK, Bellman K, Panda D, et al: Novel actions of the antitumor drugs vinfl unine and vinorelbine on microtubules. Cancer Res 60:5045-51, 2000

92. Kruczynski A, Hill BT: Vinfl unine, the latest Vinca alkaloid in clinical development. A review of its preclinical anticancer properties. Crit Rev Oncol Hematol 40:159-73., 2001

93. Jordan MA, Wilson L: Microtubules as a target for anticancer drugs. Nat Rev Cancer 4:253-65, 2004

94. Nguyen L, Retout S, Mentré F, et al: Population pharmacokinetics of vinfl unine from phase I data and evaluation of population sampling designs for further clinical development, 11th Population Approach Group Europe Meeting. Paris, France, 2002. http://userpage.fu-berlin.de/~page/

95. Brindley CJ, Morrison R, Gordon RJ, et al: Clinical pharmacokinetics of 2’-deoxy-2’-methy-lidenecytidine (DMDC), a deoxycytidine analogue antineoplastic agent. Clin Pharmacokinet 38:475-91, 2000

96. Slevin ML, Clark PI, Joel SP, et al: A randomized trial to evaluate the eff ect of schedule on the activity of etoposide in small-cell lung cancer. J Clin Oncol 7:1333-40, 1989

97. Nguyen L, Chatelut E, Chevreau C, et al: Population pharmacokinetics of total and unbound etoposide. Cancer Chemother Pharmacol 41:125-32, 1998

98. Liu B, Earl HM, Poole CJ, et al: Etoposide protein binding in cancer patients. Cancer Che-mother Pharmacol 36:506-12, 1995

99. Fleming RA, Evans WE, Arbuck SG, et al: Factors aff ecting in vitro protein binding of etopo-side in humans. J Pharm Sci 81:259-64., 1992

100. Tannock IF, Hill RP: Th e basic science of oncology. Toronto, McGraw-Hill Companies, Inc, 1998

101. Kawato Y, Aonuma M, Hirota Y, et al: Intracellular roles of SN-38, a metabolite of the campto-thecin derivative CPT-11, in the antitumor eff ect of CPT-11. Cancer Res 51:4187-91, 1991

102. Xie R, Mathijssen RH, Sparreboom A, et al: Clinical pharmacokinetics of irinotecan and its metabolites: a population analysis. J Clin Oncol 20:3293-301, 2002

103. Combes O, Barre J, Duche JC, et al: In vitro binding and partitioning of irinotecan (CPT-11) and its metabolite, SN-38, in human blood. Invest New Drugs 18:1-5, 2000

104. Kehrer DF, Sparreboom A, Verweij J, et al: Modulation of irinotecan-induced diarrhea by cotreatment with neomycin in cancer patients. Clin Cancer Res 7:1136-41, 2001

105. Davidian M, Giltinan DM: Nonlinear models for repeated measurements data. London, Chap-man & Hall, 1995

106. Karlsson MO, Sheiner LB: Th e importance of modeling interoccasion variability in population pharmacokinetic analyses. J Pharmacokinet Biopharm 21:735-50, 1993

107. Beal SL, Sheiner LB: NONMEM users guides. Hanover Maryland, GloboMax, Inc., 1989-1998

108. Sheiner LB, Beal SL: Evaluation of methods for estimating population pharmacokinetics param-eters. I. Michaelis-Menten model: routine clinical pharmacokinetic data. J Pharmacokinet Bio-pharm 8:553-71, 1980

109. Baker SD, Verweij J, Rowinsky EK, et al: Role of body surface area in dosing of investigational anticancer agents in adults, 1991-2001. J Natl Cancer Inst 94:1883-8, 2002

110. Gelderblom H, Mross K, ten Tije AJ, et al: Comparative pharmacokinetics of unbound pacli-taxel during 1- and 3-hour infusions. J Clin Oncol 20:574-81, 2002

111. ten Tije AJ, Synold TW, Spicer D, et al: Eff ect of valspodar on the pharmacokinetics of unbound paclitaxel. Invest New Drugs 21:291-8, 2003

112. Smorenburg CH, Sparreboom A, Bontenbal M, et al: Randomized cross-over evaluation of body-surface area-based dosing versus fl at-fi xed dosing of paclitaxel. J Clin Oncol 21:197-202, 2003

70

Anja Henningsson

113. Smorenburg CH, ten Tije AJ, Verweij J, et al: Altered clearance of unbound paclitaxel in elderly patients with metastatic breast cancer. Eur J Cancer 39:196-202, 2003

114. Berglund A, Glimelius B, Bergh J, et al: Selection of chemotherapy by ex vivo assessment of tumor sensitivity to cytotoxic drugs: results of a clinical trial. Med Oncol 19:151-9, 2002

115. Bruno R, Hille D, Riva A, et al: Population pharmacokinetics/pharmacodynamics of docetaxel in phase II studies in patients with cancer. J Clin Oncol 16:187-96, 1998

116. Armand J-P, Fumoleau P, Marty M, et al: Pharmacokinetics of vinfl unine, a Novel Vinca alka-loid, during the phase I Dose escalation Study (D1 Q3 weeks), American Association for Cancer Research. New Orleans, LA, 2001

117. Delord J-P, Stupp R, Pinel M-C, et al: A phase I study of vinfl unine given as 10 minute intra-venous (IV) infusion on a weekly schedule in patients with advanced solid tumours, American Society of Clinical Oncology. San Francisco, CA, 2001

118. Johnson P, Judson I, Ottensmeier C, et al: A phase I and pharmacokinetic study of vinfl unine given on days 1 and 8 every 3 weeks. Br J Cancer 83:P241, 2001 (suppl 1)

119. Ratain MJ, Schilsky RL, Choi KE, et al: Adaptive control of etoposide administration: impact of interpatient pharmacodynamic variability. Clin Pharmacol Th er 45:226-33, 1989

120. Ratain MJ, Mick R, Schilsky RL, et al: Pharmacologically based dosing of etoposide: a means of safely increasing dose intensity. J Clin Oncol 9:1480-6, 1991

121. Sparreboom A, de Bruijn P, Nooter K, et al: Determination of paclitaxel in human plasma using single solvent extraction prior to isocratic reversed-phase high-performance liquid chromatogra-phy with ultraviolet detection. J Chromatogr B Biomed Sci Appl 705:159-64, 1998

122. Beijnen JH, Huizing MT, ten Bokkel Huinink WW, et al: Bioanalysis, pharmacokinetics, and pharmacodynamics of the novel anticancer drug paclitaxel (Taxol). Semin Oncol 21:53-62, 1994

123. Mathijssen RH, Marsh S, Karlsson MO, et al: Irinotecan pathway genotype analysis to predict pharmacokinetics. Clin Cancer Res 9:3246-53, 2003

124. Sham PC, Curtis D: Monte Carlo tests for associations between disease and alleles at highly polymorphic loci. Ann Hum Genet 59 ( Pt 1):97-105, 1995

125. Tornberg E: A surface tension apparatus according to the drop volume principle. Journal of Col-loid and Interface Science 60:50-59, 1977

126. Derksen S, Keselman HJ: Backward, forward and stepwise automated subset selection algo-rithms: Frequency of obtaining authentic and noise variables. Br J Math and Stat Psych 45:265-82, 1992

127. Stewart CF, Pieper JA, Arbuck SG, et al: Altered protein binding of etoposide in patients with cancer. Clin Pharmacol Th er 45:49-55, 1989

128. Oken MM, Creech RH, Tormey DC, et al: Toxicity and response criteria of the Eastern Coop-erative Oncology Group. Am J Clin Oncol 5:649-55., 1982

129. Juul S: Epidemiologi og evidens. Copenhagen, Munksgaard, 2004130. Analysing and presenting results. In: Alderson P, Green S, Higgins J, editors. Cochrane Review-

ers’ Handbook 4.2.2 [updated December 2003]; Section 8., http://www.cochrane.dk/cochrane/handbook/hbook.htm(accessed 31st January 2004)

131. Jonsson EN, Karlsson MO: Automated covariate model building within NONMEM. Pharm Res 15:1463-8, 1998

132. Jonsson EN, Karlsson MO: Xpose-an S-PLUS based population pharmacokinetic/pharmaco-dynamic model building aid for NONMEM. Comput Methods Programs Biomed 58:51-64, 1999

133. Sheiner LB: Analysis of pharmacokinetic data using parametric models. III. Hypothesis tests and confi dence intervals. J Pharmacokinet Biopharm 14:539-55, 1986

134. Holford NH, Peace KE: Results and validation of a population pharmacodynamic model for cognitive eff ects in Alzheimer patients treated with tacrine. Proc Natl Acad Sci U S A 89:11471-5, 1992

71


135. Briasoulis E, Karavasilis V, Tzamakou E, et al: Pharmacodynamics of non-break weekly pacli-taxel (Taxol) and pharmacokinetics of Cremophor-EL vehicle: results of a dose-escalation study. Anticancer Drugs 13:481-9, 2002

136. Wild MD, Walle UK, Walle T: Extensive and saturable accumulation of paclitaxel by the human platelet. Cancer Chemother Pharmacol 36:41-4, 1995

137. Sawyer M, Ratain MJ: Body surface area as a determinant of pharmacokinetics and drug dosing. Invest New Drugs 19:171-7, 2001

138. Ratain MJ: Body-surface area as a basis for dosing of anticancer agents: science, myth, or habit? J Clin Oncol 16:2297-8., 1998

139. Felici A, Verweij J, Sparreboom A: Dosing strategies for anticancer drugs: the good, the bad and body-surface area. Eur J Cancer 38:1677-84, 2002

140. Marsh S, McLeod HL: Cancer pharmacogenetics. Br J Cancer 90:8-11, 2004141. Taniguchi R, Kumai T, Matsumoto N, et al: Utilization of human liver microsomes to explain

individual diff erences in paclitaxel metabolism by CYP2C8 and CYP3A4. J Pharmacol Sci 97:83-90, 2005

142. Rollan A, Guzman S, Pimentel F, et al: Catabolism of chylomicron remnants in patients with previous acute pancreatitis. Gastroenterology 98:1649-54, 1990

143. Bitzen U, Winqvist M, Nilsson-Ehle P, et al: Retinyl palmitate is a reproducible marker for chy-lomicron elimination from blood. Scand J Clin Lab Invest 54:611-3, 1994

144. Cabanes A, Briggs KE, Gokhale PC, et al: Comparative in vivo studies with paclitaxel and lipo-some-encapsulated paclitaxel. Int J Oncol 12:1035-40, 1998

145. Bodey GP, Buckley M, Sathe YS, et al: Quantitative relationships between circulating leuko-cytes and infection in patients with acute leukemia. Ann Intern Med 64:328-40, 1966

146. Sheiner LB, Steimer JL: Pharmacokinetic/pharmacodynamic modeling in drug development. Annu Rev Pharmacol Toxicol 40:67-95, 2000

147. Price TH, Chatta GS, Dale DC: Eff ect of recombinant granulocyte colony-stimulating factor on neutrophil kinetics in normal young and elderly humans. Blood 88:335-340, 1996

148. Braakhuis BJ, Hill BT, Dietel M, et al: In vitro antiproliferative activity of docetaxel (Taxotere), paclitaxel (Taxol) and cisplatin against human tumour and normal bone marrow cells. Antican-cer Res 14:205-8, 1994

149. Bruno R, Olivares R, Berille J, et al: Alpha-1-acid glycoprotein is an independent predictor of effi cacy and survival in NSCLC patients treated with docetaxel (Taxotere) [Abstract No. ?]. Proceedings of American Society of Clinical Onology, 1998, pp 471a

150. Karlsson MO, Anehall T, Friberg LE, et al: Pharmacokinetic/pharmacodynamic modelling in oncological drug development. Basic Clin Pharmacol Toxicol 96:206-11, 2005

151. Sparreboom A, Xie R, Baker SD, et al: Semi-mechanistic models for DHA-paclitaxel pharmaco-kinetics/pharmacodynamics. Clin Cancer Res 9(S):Abstract B253, 2003

152. Sandstrom M, Lindman H, Nygren P, et al: Model describing the relationship between phar-macokinetics and hematologic toxicity of the epirubicin-docetaxel regimen in breast cancer patients. J Clin Oncol 23:413-21, 2005

Acta Universitatis UpsaliensisDigital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Pharmacy 10

Editor: The Dean of the Faculty of Pharmacy

A doctoral dissertation from the Faculty of Pharmacy, Uppsala University, is usually a summary of a number of papers. A few copies of the complete dissertation are kept at major Swedish research libraries, while the summary alone is distributed internationally through the series Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Pharmacy. (Prior to January, 2005, the series was published under the title "Comprehensive Summaries of Uppsala Dissertations from the Faculty of Pharmacy".)

Distribution: publications.uu.seurn:nbn:se:uu:diva-5772

ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2005

Documents

Mechanism-Based Pharmacokinetic and Pharmacodynamic ...166316/FULLTEXT01.pdf · variability in drug exposure and the sensitivity of the bone marrow. For two reasons ... Crude extract