Comprehensive Summaries of Uppsala Dissertations from the Faculty of Pharmacy 286

Pharmacokinetic-Pharmacodynamic Modelling of Anticancer Drugs

Haematological Toxicity and Tumour Response in Hollow Fibres

BY

LENA E. FRIBERG

ACTA UNIVERSITATIS UPSALIENSIS UPPSALA 2003 Contents

Abbreviations and Nomenclature ...... 7 1. Introduction ...... 9 1.1 Chemotherapy in cancer treatment ...... 9 1.1.1 Schedule dependence...... 10 1.2 Haematology ...... 10 1.2.1 Chemotherapy-induced myelosuppression...... 12 1.3 Animal models in anticancer research...... 12 1.3.1 Animal tumour models...... 13 1.4 PKPD modelling...... 14 1.4.1 Population analysis...... 14 1.4.2 Measurements of drug exposure...... 15 1.4.3 Models based on the Hill equation ...... 15 1.4.4 Models of Myelosuppression...... 17 1.4.4.1 Semi-physiological models of myelosuppression...... 17 1.4.5 Models of chemotherapy-induced effects on tumour cells ...... 19 1.5 Drugs of interest in this thesis ...... 20 1.5.1 The FEC Regimen ...... 20 1.5.2 Drugs in clinical studies ...... 22 1.5.3 The novel agent CHS 828...... 23 2 Aims of the Thesis...... 24 3. Materials and Methods ...... 25 3.1 Experimental procedures (Papers I, III, V and VI)...... 25 3.1.1 Animals ...... 25 3.1.2 Drugs ...... 25 3.1.3 Pharmacokinetic monitoring...... 26 3.1.3.1 Chemical assays ...... 26 3.1.4 Pharmacodynamic monitoring...... 27 3.1.4.1 General toxicity ...... 27 3.1.4.2 Haematological toxicity...... 27 3.1.5 Hollow fibre model...... 28 3.1.5.1 Fibre preparation ...... 28 3.1.5.2 Surgical procedure...... 28 3.1.5.3 Cell-density evaluation...... 29 3.1.6 Study designs...... 29

5 3.2 Clinical data...... 31 3.2.1 Schedules and measurements ...... 31 3.2.1.1 DMDC (2'-deoxy-2'-methylidenecytidine)...... 31 3.2.1.2 Docetaxel...... 31 3.2.1.3 Paclitaxel ...... 31 3.2.1.4 Etoposide...... 32 3.2.1.5 CPT-11 (Irinotecan) ...... 32 3.2.1.6 Vinflunine...... 33 3.3 Model building ...... 33 3.3.1 Pharmacokinetics...... 33 3.3.2 Pharmacodynamics...... 34 3.3.2.1 Model development (Paper I)...... 34 3.3.2.2 Model development (Paper II)...... 34 3.3.2.3 Model development (Paper III) ...... 35 3.3.2.4 Model development (Paper IV) ...... 36 3.3.2.5 Model development (Paper V) ...... 36 3.3.2.6 Model development (Paper VI) ...... 37 3.3.3 Data analysis...... 37 3.3.3.1 Model validation (Paper III)...... 38 4 Results and Discussion ...... 39 4.1 Pharmacokinetics...... 39 4.1.1 Results ...... 39 4.1.1.1 5-FU, epirubicin and 4-OHCP...... 39 4.1.1.2 CHS 828 ...... 39 4.1.2 Discussion ...... 40 4.2 PKPD modelling of haematological toxicity...... 42 4.2.1 Results ...... 42 4.2.1.1 Schedule dependence of 5-FU and epirubicin...... 42 4.2.1.2 Schedule dependence of DMDC ...... 42 4.2.1.3 Semi-physiological model for DMDC...... 44 4.2.1.4 Semi-physiological model for 5-FU...... 45 4.2.1.5 Semi-physiological model with parameter consistency across drugs...... 46 4.2.2 Discussion ...... 48 4.3 The hollow fibre model in immunocompetent rats...... 51 4.3.1 Results ...... 51 4.3.1.1 Development of the hollow fibre model in rats ...... 51 4.3.1.2 Schedule dependence of CHS 828...... 52 4.3.2 Discussion ...... 53 5 Conclusions ...... 56 6 Acknowledgement...... 58 7 References ...... 61

6 Abbreviations and Nomenclature

ABC-7 area between the baseline and effect curve up to day 7 AUC area under the concentration-time curve

AUC50 AUC producing half maximal effect AUCEdir integral of the direct effect from zero to infinite time AUCEdir50 AUCEdir producing half the maximal observed effect C concentration

Cp plasma concentration CFU-GEMM colony forming unit, capable of forming granulocytes, erythrocytes, monocytes and megakaryocytes (platelets) CHS 828 N-(4-chlorophenoxyhexyl)-N´-cyano-N´´-4-pyridylguanidine Circ circulating cell count at any time point

Circ0 baseline circulating cell count CL total body clearance CLL chronic lymphocytic leukaemia

Css concentration at steady state CV coefficient of variation DMDC 2'-deoxy-2'-methylidenecytidine DMSO dimethyl sulfoxide DPD dihydropyrimidine dehydrogenase E effect

EC50 drug concentration producing half maximal effect EC50,0 EC50 at time zero EC50,24h EC50 at 24 hours after first administration EC50,dir drug concentration producing half maximal direct effect EC50,max EC50 at infinite time Edir direct effect EDrug elimination rate constant governed by drug concentration EDTA ethylenediaminetetraacetic acid

Emax maximal effect

Eobs observed effect Eobs,max maximal observed effect F bioavailability

F1 bioavailability the day of administration F2 bioavailability of a second absorption phase observed after a lag time FEC the 5-fluorouracil-epirubicin-cyclophosphamide regimen FO first-order estimation method in NONMEM FOCE first-order conditional estimation method in NONMEM

7 fu fraction unbound 5-FU 5-fluorouracil G-CSF granulocyte colony stimulating factor HPLC high performance liquid chromatography IIV inter-individual variability unexplained by the model iv intravenous ip intraperitoneal ka absorption rate constant ka,1 absorption rate constant on the day of administration ka,2 absorption rate constant of a second absorption phase, after a lag time kcirc elimination rate constant of circulating cells kEC50 rate constant of time-dependent decrease in EC50 kF rate constant of dose-dependent decrease in bioavailability kGrow growth rate of tumour cells kin production rate constant of proliferative cells Km drug concentration at which the elimination rate is half its maximum ktr rate constant from the transit compartments kprol rate constant that determines proliferation (self-renewal) MTT mean transit time (Paper IV); [3-4,5-dimethylthiaxol-2-yl]-2,5- diphenyltetrazolium bromide (Papers V and VI) n number (general) ne not estimated in the final model NCI National Cancer Institute OD optical density OFV objective function value (produced by NONMEM) 4-OHCP 4-hydroxycyclophosphamide PBS phosphate buffered saline PD pharmacodynamic PHSC pluripotent haematopoietic stem cell PK pharmacokinetic

R1 zero-order rate input on the day of administration RSE relative standard error s sigmoid factor SCID severe combined immunodeficiency SD standard deviation SE standard error Slope coefficient of linear drug effect

SlopeCL coefficient of linear time-dependent decrease in clearance THF tetrahydrofuran UV ultraviolet

Vmax maximum elimination rate Vss volume of distribution at steady state t1/2,circ half-life of circulating cells β sigmoid factor γ sigmoid factor (Papers I, II and VI ); feedback power function (Paper IV)

8 1. Introduction

1.1 Chemotherapy in cancer treatment Anticancer drugs generally affect all types of rapidly proliferating cells and, consequently, their therapeutic indices are narrow. It is commonly held that “more is better”, i.e. a high dose is more likely to be successful in tumour treatment than a low dose. Severe side effects are therefore often observed and clinical doses are more or less empirically determined by the risk of severe toxicity. Until drugs with more tumour-specific cell-killing effects with improved therapeutic indices are available, more effective use of both new and established chemotherapeutic drugs is searched for. One step in the optimisation process is to establish models that will help in decisions on the choice of dosage and treatment schedules for anticancer drugs. Observed effects are in general better related to plasma concentrations than the drug dose, and in cancer treatment, a few studies have also shown a correlation between pharmacokinetics and tumour effect.1-4 The development of models that describe the interrelationships between the dosage of the drug, its plasma concentration and the response, i.e. pharmacokinetic- pharmacodynamic (PKPD) models, would be of value to assist in the improvement of therapeutic indices. Optimally, a PKPD model would allow investigation of both dose-limiting toxicity and antitumour effects. Frequent measurements of tumour size are, however, usually not achievable and there are often large inter- and intra-individual variations in tumour sensitivity. Most attention in this field has therefore been given to models of dose- limiting toxicities. As myelosuppression, with the associated risk of developing infections, is the most frequent dose-limiting toxicity in chemotherapy, the focus in this thesis is on PKPD modelling of myelosuppression. However, the thesis also describes the development of an animal model that is suitable for PKPD modelling of tumour response over time. The relevance of PKPD modelling in the optimisation of cancer therapy was already pointed out a decade ago.5

9 1.1.1 Schedule dependence Many anticancer drugs show schedule dependence,6 i.e. the magnitude of response or toxicity depends on the rate of administration. In other therapeutic areas, such as the treatment of migraine, it is expected that the choice of administration schedule will affect the response-time profile. For anticancer drugs, this expectation appears to be less common, perhaps because drug concentrations and the observed effects are dissociated in time. Mechanisms possibly underlying the schedule dependence of anticancer drugs include cell-cycle specificity, saturable transport into or out of cells, time-dependent cellular repair mechanisms, co-factor depletion, on-off mechanisms in critical steps, drug sensitisation/resistance development over time, or multiple mechanisms of drug action.7,8 The dependence of haematological toxicity on administration schedule may not necessarily be associated with the schedule dependence of antitumour effect or of other toxicities. The choice of administration schedule is therefore important, since changing the schedule may have a greater effect than increasing the dose, on the outcome. However, there is a lack of PKPD models with the documented capability of predicting the effect of new schedules and the search for more effective administration regimens has largely relied on trial and error. The use of models would allow more efficient detection and characterisation of schedule dependence for drugs under development. An apparent schedule dependence could be due to dose- dependencies and circadian rhythms in the PK. Characterisation of the PK of a drug is therefore of importance for understanding the influence of different schedules on the effect. There is no schedule dependence in the PD when the magnitude of effect is dependent on the actual exposure, i.e. the area under the concentration-time curve (AUC).

1.2 Haematology Anticancer drugs can affect levels of leukocytes, erythrocytes and platelets; however, as myelosuppression is the dominating toxicity for most anticancer drugs, only the production and regulation of leukocytes, with focus on neutrophils, will be described here. The circulating leukocytes in blood comprise neutrophils (approximately 60-70% in humans), lymphocytes (30%), monocytes (5%), eosinphils (2%) and basophils (<1%). Neutrophils, eosinophils and basophils constitute the granulocytes. Maturation of eosinophils, basophils and monocytes basically follows the same maturation steps in the bone marrow as that of neutrophils; however, the mechanisms of early lymphoid development remain poorly -lymphocytes differentiate in the bone marrow of adults, while T-lymphocytes develop from bone marrow 10 progenitors that migrate to the thymus, where they undergo maturation and a careful selection process. Most of the description below relates to humans, but leukocytes are produced and regulated similarly in rodents. However, in rodents, the turnover is faster and lymphocytes are the dominating cell type. In Sprague Dawley rats, 85% of the leukocytes are lymphocytes and only approximately 10% are neutrophils (Charles River, Uppsala, Sweden). In addition, in rats haematopoiesis also occur in the spleen, but to a relatively small extent under normal conditions.10 All blood cells originate from a common type of cell, the pluripotent haematopoietic stem cell (PHSC), which has a high capacity for self- renewal, i.e. such cells can give rise to “identical” daughter cells.11 PHSCs can also differentiate to form lymphoid stem cells or myeloid multipotent stem cells (CFU-GEMMs), which give rise to granulocytes, erythrocytes, monocytes and platelets. During maturation, the cells remain at a certain stage for a while, and then move to the next stage (Figure 1), usually on a first in, first out basis.12 Most cells in the bone marrow are lineage specific precursors with little self-renewal capacity and high mitotic activity.11 The proliferating stage from the myeloblast to the last myelocyte stage takes approximately 5 days in healthy volunteers.13 Non-mitotic cells include metamyelocytes, band cells and segmented neutrophils that are released into the blood. The mean transit time for progression through the non-mitotic stages is 6.6 days in humans.14 After labelling of rat myelocytes the average emergence time for mature neutrophils is approximately 3 days.10

Proliferating precursor cells Non-mitotic cells

Pluri- Committed Band Segmented Bone potent Myeloblast Pro- Myelocyte Meta- stem cell stem cell myelocyte myelocyte cells cells marrow

Marginated Circulating Blood cells cells

Cells in Tissues tissues

Figure 1. Schematic illustration of granulopoiesis.

The circulating neutrophils in blood are in rapid equilibrium with a pool of neutrophils that are marginated along the vessel walls. Both pools are of approximately equal size.15 Neutrophils disappear from the blood in a random (first-order) process16 with a half-life in blood of approximately 6.7

11 hours in humans17 and 5.7 hours in rats.18 Thereafter, the neutrophils enter the tissues, where they undergo apoptosis in the next few days. Lymphocytes have a transit time in blood of less than 30 minutes in both rats and humans.19,20 Regulation of leukocyte levels is only partly understood.21 Cytokines and extracellular matrix components regulate proliferation and differentiation;22 G-CSF (granulocyte colony stimulating factor) is the most important of these factors. G-CSF stimulates extra cell divisions and shortens the transit time, but does not change the half-life of circulating neutrophils.23,24 G-CSF levels and the neutrophil count in blood are negatively correlated, implying that mature neutrophils mediate one of the major pathways for G-CSF clearance.25,26

1.2.1 Chemotherapy-induced myelosuppression Anticancer drugs generally have less effect on non-proliferative precursor cells than on proliferating precursor cells, which have high mitotic activity.27 As the stem cells cycle slowly under normal conditions, they are rarely affected by a single administration of a cytotoxic drug, although the first dose might stimulate the cycling activity of the primitive cells. Therefore, these cells can be vulnerable to doses administered 3 to 5 days after the first dose.28 The grade of chemotherapy-induced neutropenia correlates with elevated levels of colony stimulating factors, known to increase neutrophil production.29 Different drugs have different effects on the granulopoietic cells. For example, compared with most other chemotherapeutic drugs, methotrexate and vinblastine cause a more rapid fall in blood counts, while recovery after treatment with melphalan and nitrosureas is delayed.30 It is believed that drugs causing prolonged myelosuppression affect the slowly cycling stem cells. The large inter-individual variability (IIV) in chemotherapy-induced myelosuppression appears to depend on differences in sensitivity to anticancer drugs.31

1.3 Animal models in anticancer research Studies in animals are a necessary step that falls between in vitro studies and studies in patients in the development of a drug. Compared with in vitro studies, animal studies can (i) set the effect in relation to host toxicity, (ii) better mimic the tumour environment in the clinical situation, (iii) provide a similar concentration-time profile as in patients, and (iv) to some degree account for drug protein binding and metabolism. Disparate protein binding

12 or active metabolites in animals from those in patients can, however, disguise comparable concentration-effect relationships, while similar (unbound) plasma concentrations often produce the same effect in experimental animal models and humans.32,33 PKPD models of different schedules may therefore allow successful extrapolation of preclinical studies in order to predict both effective and toxic drug concentrations for clinical investigations.34 In addition, animal models allow investigation of a range of dose levels and schedules that would not be possible in clinical studies of anticancer drugs. Traditionally, the tolerability and PK of a new drug are studied in non- tumour-bearing animals, either separately or simultaneously.35 However, drug disposition may differ between rats with and without tumours and surgical procedures may affect the PK of a drug.36 Important information regarding IIV in drug activity may therefore be wasted, if PK and effect is not studied simultaneously. PKPD relationships established with PK and PD information from the same animal may aid in the development of human administration regimens37 and knowledge of individual PK parameters has 38 been shown to reduce variability in the estimation of EC50 in rats. Rats are preferable to mice for PKPD studies, as rats can withstand repeated blood sampling. The rat is also well suited for haematology investigations.10,39 In fact, rodents can predict the myelosuppressive effects of many anticancer drugs.40,41

1.3.1 Animal tumour models Tumour regression in animal experimental tumour models is considered an important endpoint of clinical relevance.42 However, drug-activity patterns in tumour models of murine origin often differ from those in clinical tumours. Animal tumour models with cancer cells of human origin are more attractive,43 as they are better in predicting specific activity in clinical diagnoses.44,45 A recent analysis demonstrated that activity in at least one third of the xenograft models correlated with ultimate activity in at least some Phase II trials.46 However, in this study, activity in a xenograft model with a particular histology did not closely correlate with activity against the same cancer histology in patients. Implantation of human cancer cells induces undesirable immune responses in animals. Therefore, nude mice that lack the thymus, and SCID (severe combined immunodeficiency) mice that lack both T- and B-cells, are often used as xenograft models in which to culture human cell lines. Human tumour cell lines are available for a wide range of malignant diagnoses. Nonetheless, there are drawbacks with xenograft models: (i) immunosuppressive animals cannot be used for concurrently run

13 myelosuppression studies, (ii) these animals are expensive and time consuming, as they need special care, and (iii) generally only one tumour can be implanted per animal. To reduce development costs, the National Cancer Institute (NCI) has developed an alternative in vivo model, based on cultivation of tumour cells in semi-permeable hollow fibres.47 This original hollow fibre model in mice is now used routinely at the NCI for in vivo drug screening, before further xenograft testing.48

1.4 PKPD modelling

The use of PKPD modelling optimally allows data to be summarised, relevant predictions to be made, and a better understanding of the underlying physiology to be reached. Relevant PKPD models can guide the search for the best dose, based on patient characteristics, actual concentration measurements and/or observed effects. PKPD modelling is also being used more often by the drug industry to reduce the number of clinical trials.

1.4.1 Population analysis Population analysis using non-linear mixed effects modelling offers great advantages over standard data analysis procedures, since (i) more complex models can be applied, as each individual “borrows” information from the others, and (ii) the same model can be applied to all individuals. Compared with standard methods, population modelling is more likely to detect non- linearity49 and is better able to handle imbalances in the data. In populations where it is preferable to reduce the number of samples per individual from an ethical viewpoint and/or to limit the blood volume taken (e.g. infants, cancer patients and animals), a sparse sampling technique (generally 2-4 samples) offers advantages. Sparse sampling in combination with population analysis can provide accurate estimates and reliable predictions.50,51 Characterisation of IIV in the PK leads to better PKPD relationships, even if there are only a couple of concentration samples per individual.52 However, a potential drawback of population analysis is that more individuals are required than in a standard analysis. The value of preclinical population modelling has been stressed,53,54 but few PK models have been established for anticancer drugs.55-57 Further, despite the advantages associated with simultaneous evaluation of PK and PD (fewer animals and lower workload, the use of PK to explain extreme PD findings in an animal, the possibility of assessing individual concentration/effect relationships,58 as well as the possibility to use all

14 available data), no example of a preclinical population PKPD model for anticancer drugs was reported when the present studies started. In this thesis, the NONMEM59 (nonlinear mixed effects modelling) software was used for the population analysis. The population model parameters include both fixed effects, related to the typical individual, and random effects, with magnitudes of IIV in the parameters, and magnitudes of residual variability between individual predictions and observations.

1.4.2 Measurements of drug exposure Since observed drug concentrations and the observed effects, i.e. myelosuppression and tumour effects, are dissociated in time, summary 60 61 variables of exposure like AUC, steady-state concentration (Css), and time above a threshold concentration,62 are often used. Summary variables such as these can be associated with counterintuitive predictions. For example, when AUC is used as a summary variable, it would actually mean that a bolus injection will produce the same effect as a very extended infusion with an extremely low concentration, as long as the AUC is the same. On the other hand, when the time for which the concentration is above a threshold level is related to the effect, an all-or-none relationship is predicted. This means that if the concentration is just below this level, there will be no effect, while if the concentration is just above this level; the effect is at its maximum. It would be preferable to estimate the best summary of the concentration-time profile63 or to directly apply the whole concentration-time profile, so that different dosage regimens and possible schedule-dependent effects can be accounted for with accuracy.

1.4.3 Models based on the Hill equation 64,65 The Hill equation (the sigmoid Emax model), or simplifications thereof, are often used to describe the relationship between the concentration (Cp) and the effect (E):

γ γ γ E = Emax · Cp / (EC50 + Cp ) (Eq. 1) where EC50 is the exposure that produces 50% of the maximal effect (Emax), and is inversely related to the potency. In the basic Emax model, γ = 1; in the threshold model, γ = ∞; and for linear models, γ = 1 and EC50 » Cp. However, a more general model for time-dissociated effects has also been published.63 In this model, the concentration of drug results in a direct effect (Edir) in a non-linear manner

β β β Edir = Cp / (Cp + EC50,dir ) (Eq. 2) 15 where β is a sigmoid factor and EC50,dir describes the concentration at which half of the maximal direct effect is achieved (Figure 2). The accumulated direct effect (AUCEdir), i.e. the integral of the direct effect from zero to infinite time, is included in a second sigmoid Emax model for the observed effect (Eobs)

γ γ γ Eobs = Eobs,max · AUCEdir / (AUCEdir + AUCEdir50 ) (Eq. 3)

When the direct effect is not observed, the maximal direct effect can be arbitrarily fixed to 1 and hence only scales AUCEdir50, i.e. the duration of maximal direct effect needed to obtain half of the maximal observed effect (Eobs,max).

AB ) ) p dir

Eq. 2 Direct effect (E effect Direct Concentration (C Concentration

Time Concentration (Cp)

CD ) dir

Eq. 3 AUCE AUCEdir Observed effect Direct effect (E

Time Cumulative direct effect (AUCEdir) Figure 2. Illustration of the concentration – direct effect – observed effect relationships for the general model. The drug produces a concentration-time profile (A). The concentration is related to a direct effect by e.g. a sigmoid Emax model (B), described in Eq. 2. The integral of the direct effect over time (C) is then related to the observed effect by e.g. a second sigmoid Emax model (D), described in Eq. 3. For an AUC-dependent model, panel (B) would show a linear relationship, while for a threshold model it would be a step function.

16 There are two extreme cases of the general model. Equation 2 collapses to a linear model, i.e. the observed effect is AUC-dependent, if the estimate of EC50 » Cp and β=1. Or, if β is infinite, the model predicts an all or nothing response, i.e. the time above a threshold concentration (EC50) determines the extent of the direct effect. The advantage of the general model is that it can also characterise relationships between these two extremes.

1.4.4 Models of Myelosuppression PKPD models of haematological toxicity have been described for most anticancer drugs. The majority of these relate a summary variable of drug exposure to the absolute or relative decrease of blood cells at nadir (the lowest cell count during a course of chemotherapy), using one of the models described above. As measurements are made only once or twice a week in the clinic, it is unlikely that the real nadir will be observed. The nadir is therefore likely to be overestimated, measurement errors are not appropriately accounted for, IIV may be inflated and information on the duration of myelosuppression is wasted. Other summary variables of myelosuppression such as “area between the curves”,66 time to nadir27 and number of days below a certain neutrophil value66 have been used to account for the time aspect. It would be more attractive to model the whole time- course of myelosuppression, since then both the degree and duration of toxicity could be predicted. There are a few such empirical models in the literature.63,67,68 These models incorporate a lag time before the decrease and recovery phases, which are described either by a linear slope connected to a logistic curve,67 or by cubic spline functions.63,68 The pretreatment blood cell count is a predictor of subsequent leukopenia.69,70 Survival fractions71 or % decrease60 are therefore often used, although these assume that the baseline does not vary with time and that a single measurement can adequately describe it. In addition, since it is the absolute cell counts that are related to the incidence of neutropenia and neutropenic fever,72 it might be preferable to model the absolute counts, which in practice also means estimating the baseline count.

1.4.4.1 Semi-physiological models of myelosuppression Mechanistic models, i.e. models based on physiology and pharmacology, are preferable to empirical models. Because prior information about the structure of the model and potentially also about the parameter values can be used, they are of more predictive value and can be used to learn something beyond the design on which the model was developed. The need for and relevance of mechanistic models is even more evident when clinical trial simulations have become an important tool during clinical drug

17 development. Extrapolation to untested doses and schedules or for subgroups of patients favour a model based on physiology. A few physiology-based mathematical models of chemotherapy-induced myelosuppression have been developed.73-77 However, PK has been neglected, the models were not developed to allow for estimations, and they require a number of parameters that are not available when analysing clinical data. To be able to develop PKPD models suited for estimations and consecutive predictions, it has been necessary to simplify the models, i.e. to make them semi-mechanistic. A semi-physiological model of myelosuppression has been published.78 To account for the time delay, a lag time was incorporated into an indirect inhibitory model and the drug was assumed to inhibit leukocyte production while the cells were in sensitive stages. In general, delays in time between drug administration and effect have been described by lag times,63 effect compartment models for direct effects,79 indirect effect models,80 turnover models for irreversible effects81 and transit compartments or signal transduction models.82,83 Preferably, knowledge of the rate-limiting processes involved should guide the choice of model. Transit compartments82 (i.e. compartments in series) are attractive for modelling of myelosuppression because they mimic the different cell stages within the bone marrow and because maturation is a gradual process rather than occurring at an absolute time-delay (Figures 1 and 3). In the transit compartments it is assumed that the only loss of cells is into the next compartment. Transit compartments delay and spread out the effects of cell production and drug effects from the proliferative compartments, as the cells propagate to the circulating neutrophil compartment. More compartments in the delay chain can be included without adding extra parameters since the same rate constants can be used for all the compartments. Increasing the number of compartments reduces the variability in the cellular transit time and changes the shape and distribution of the observed effect curve (Figure 3). The recovery from nadir is expedited because of feedback mechanisms, as discussed earlier. An overshoot of blood cells, i.e. a rebound phenomena, are therefore often observed, which should be taken into account in the model building to be able to predict consecutive courses accurately. Negative feedback mechanisms from circulating cells to proliferative cells, as well as feedback mechanisms within the delay chain, have previously been applied in mathematical models of myelosuppression.76,77

18 1.4.5 Models of chemotherapy-induced effects on tumour cells Exponential growth of tumour cells is often assumed, i.e. the rate of cell production and cell loss is assumed to be proportional to the number of tumour cells, with a constant rate of doubling. In reality, the growth rate slows down when the tumour grows larger, and therefore a Gompertzian growth rate might be preferable84 as it combines exponential growth with an exponentially decaying growth constant. In the early 1970s, Jusko developed models for irreversible effects of chemotherapeutic agents on cell-cycle dependent and cell-cycle independent drugs (Figure 4).85,86 Two different PKPD models have been developed for predicting the time course of tumour effects in vivo.87,88 Both studies were built on simulations without estimations.

k k k k Circulating Proliferative tr tr tr tr cells in cells ...... blood

Transit compartments

1.2 Zero transit compartments

0.8

0.4

0.0

Cell count Cell 1.2 Five transit compartments

0.8

0.4

0.0

0 5 10 15 20 Time

Figure. 3. Transit compartments between a proliferative cell compartment and a compartment of circulating blood cells (top). Graphs show simulated profiles after zero and five transit compartments (dotted lines) between the proliferative (thin line) and circulating cell (thick line) compartments. The mean transit time (MTT) in the simulations were 10 time units and the profiles were simulated to reach the same nadir value.

19 To characterise schedule dependencies found in vitro, Levasseur et al.89 (Eq. 90 4) and Adams (Eq. 5) applied models where EC50 changes over time:

s s s EC50(t) = EC50,0 + (EC50,max - EC50,0) · Time / (Time + T50 ) (Eq. 4)

1/n EC50(t)= (K/Time) (Eq. 5) where EC50,0 is the baseline value, and EC50,max is the EC50 at infinite exposure time. K and n are constants specific to the drug. However, models where the mechanism of action can explain the schedule dependence are preferable over models with parameters dependent on time, as they are more reliable for both intrapolations and extrapolations. In summary, a PKPD model should ideally take the whole concentration- time profile into account, use the absolute values, describe the whole time course of effect, be applicable to different administration schedules and be based on physiology. Additionally, it is preferable that the model can differentiate drug-specific parameters from system-related parameters that are common across drugs.

kGrow

Resting Proliferating k cells cells

EDrug

Figure 4. Models of the cytotoxic effect of chemotherapeutic agents presented by Jusko (1971, 1973). The cell growth is dependent on the amount of cells in the proliferating, sensitive, compartment, and the first-order rate constant kGrow. k determines the natural cell death and EDrug is the killing rate constant produced by drug exposure. For cell-cycle independent drugs, no resting compartment is needed.

1.5 Drugs of interest in this thesis

1.5.1 The FEC Regimen One of the most commonly used drug regimens in breast cancer is the 5- fluorouracil (5-FU) - epirubicin – cyclophosphamide (FEC) regimen. Clinical studies designed to optimise this regimen have been performed in 20 collaboration with our department.91 All component drugs are administered within a few hours at the convenience of the patient. As for many other combinations, quantitative information about the relative contribution of component drugs to toxicity is missing, and the effect of changed dosage schedules within the combination is not defined. 5-FU is an antimetabolite that induces multiple biochemical lesions; it inhibits thymidylate synthase, which is necessary for the synthesis of DNA, and it can be incorporated into RNA and DNA.92 5-FU is mainly inactivated in the liver and shows non-linear elimination due to the capacity-limited metabolising enzyme, dihydropyrimidine dehydrogenase (DPD).93-95 A schedule-dependent toxicity pattern characterises 5-FU: bone-marrow depression is dose limiting after iv bolus doses, while gastrointestinal events and hand-foot syndrome are dose limiting after infusions.92 In contrast, fractionated bolus injections of 5-FU over a 3-5 day period lead to more bone marrow toxicity than a single injection.96 The dose intensity can be increased by three to four times when given as a continuous infusion compared with a bolus injection.8 This is partly due to the non-linear PK, but even when the total AUC is calculated for different schedules,8 using previously published PK parameters,91 a higher AUC is tolerated after infusions than after bolus injections, i.e. schedule dependence in myelosuppression is evident. The schedule dependence has been attributed to two different mechanisms of action, where high concentrations (as after bolus injections) are needed for 5-FU to incorporate into RNA, while at low concentrations the cytotoxicity parallels thymidylate synthase inhibition.8 Epirubicin is an anthracycline that shows linear three-compartment PK both in patients and in rats.56,97 Epirubicin is mainly eliminated by the liver and epirubicinol is produced, which is an active metabolite associated with a low degree of cytotoxic activity.98,99 The dose-limiting toxicity after a single course is myelosuppression, which is believed to be AUC dependent, while cumulative cardiotoxicity is possibly related to the peak concentration.100,101 Cyclophosphamide is classified as an alkylating drug that primarily undergoes hepatic transformation to form 4-hydroxy-cyclophosphamide (4- OHCP).102 4-OHCP is in equilibrium with its tautomer, aldophosphamide, which degrades within the cells to yield the alkylating agent, phosphoramide mustard, and acrolein. 4-OHCP exhibits a mono-exponential decline in plasma.91 The maximum tolerated dose for bolus and protracted infusions of cyclophosphamide is approximately the same, with myelosuppression as the dose limiting toxicity.6 A fractionated schedule of 4-OHCP have, however, been reported to improve the therapeutic index in mice.103 As previously discussed,104 different schedules of the same total cyclophosphamide dose may result in different concentration-time profiles of 4-OHCP; bolus injections might saturate the metabolising enzymes and the autoinduction

21 phenomena, observed in the metabolism of cyclophophamide, will mainly influence the concentration-time curve after prolonged administration schedules.

1.5.2 Drugs in clinical studies DMDC is an antimetabolite that is relatively resistant to inactivation and can therefore be administered by the oral route.105 A linear one-compartment model with first order absorption has been used to describe the PK.106 DMDC is mainly eliminated in the liver.106 Haematological toxicity is dose limiting and appears to be schedule dependent, as twice-daily administration is associated with greater toxicity than once-daily administration.107-109 The elimination of the taxanes is characterised by hepatic metabolism. For both docetaxel and paclitaxel, three-compartment models have been applied.110,111 While docetaxel shows linear disposition, the non-linear distribution of paclitaxel is largely explained by binding to the vehicle Cremophor EL.111 The dose-limiting toxicity is myelosuppression for both drugs. Docetaxel has been classified as an AUC-dependent drug,112 while paclitaxel shows less myelosuppression after 3-hour infusions than after 24- hour infusions of the same dose.113 The duration above a threshold concentration has therefore been reported to be the best predictive measure for toxicity.114 However, a more general model appears to be superior over threshold and AUC-dependent models.63 Etoposide is a topoisomerase II inhibitor, whose disposition is characterised by bi-exponential decay.115 Approximately 20 to 45% is excreted as unchanged drug in the urine.115 The dose-limiting toxicity, myelosuppression, shows no or little dependence on schedule, while the antitumour effect is highly schedule dependent.116 CPT-11 (irinotecan) is a camptothecin analogue that is extensively metabolised in the liver into the metabolite, SN-38,117 which has a 100- to 1000-fold higher potency than CPT-11 in vitro.118 Three-compartment and two-compartment models have been used to describe the disposition of CPT- 11 and SN-38, respectively.119 Dose limiting toxicities are myelosuppression and diarrhoea.117 No evidence of schedule-dependent toxicity has been presented to date, apart from a recent study in which a lower incidence of neutropenia was observed when the drug was administered on days 1 and 10, every 21 days, than when the whole dose was administered on day 1.120 Vinflunine is under development by Pierre Fabre (Castres, France) and has demonstrated superior activity to other vinca alkaloids in preclinical tumour models.121 A four-compartment model characterises the PK.122 Two- thirds of radiolabelled vinflunine is eliminated through bile and one-third by

22 the kidneys.123 The dose-limiting toxicity is haematological, essentially neutropenia, with no obvious schedule dependency.124-126

1.5.3 The novel agent CHS 828 The cyanoguanidine CHS 828 is an anticancer agent under development by Leo Pharma (Ballerup, Denmark). The mechanism of action has not been fully elucidated but high activity has been found against many tumour cell lines in vitro127,128 and in primary cell cultures from chronic lymphocytic leukaemia (CLL), acute leukaemia and some solid tumour samples.129 Activity against CLL was also shown in the hollow fibre assay in vitro130 and in vivo.131 The xenograft models of MCF-7 and NYH small lung cancer in nude mice were both sensitive to CHS 828.128 A schedule-dependent antitumour effect of CHS 828 has been demonstrated both in vitro132 and in vivo.128 Schedule dependence in toxicity was found in phase I studies aiming to establish a maximum tolerated dose, as a much lower total dose was tolerated for a five-day dosage regimen than for a single-dose regimen.133,134 Thrombocytopenia and gastrointestinal complications were the dose-limiting toxicities after the five-dose regimen.133 CHS 828 has a low solubility (0.1 mg/ml in 10% DMSO solution) and effective doses can only be administered orally.

23 2 Aims of the Thesis

The aims of this thesis were:

• To better understand schedule dependence in haematological toxicity of anticancer drugs through preclinical pharmacokinetic-pharmacodynamic studies.

• To develop pharmacokinetic-pharmacodynamic models that can describe the time course and schedule dependence of myelosuppression in animals and patients.

• To develop and apply the hollow fibre model in immunocompetent rats as a convenient experimental procedure for simultaneous assessment of pharmacokinetics, haematological toxicity and tumour response.

24 3. Materials and Methods

3.1 Experimental procedures (Papers I, III, V and VI)

3.1.1 Animals Male Sprague Dawley rats [CRL:CD(S-D)BR-rats; Charles River, Uppsala, Sweden] were used. They were acclimatised for at least one week prior to the start of the experiments. Throughout the studies, the animals had free access to water and a standardised pellet diet and were kept in a room illuminated from 7 a.m. till 7 p.m. All blood samples were drawn from a hind paw vein after the rats had been on a heating pad for at least 10 minutes. The samples were collected in EDTA-prepared Microtainer tubes (Becton Dickinson, Franklin Lakes, NJ), except when serum was to be prepared (CHS 828), in which case eppendorf tubes were used. The Animal Ethics Committee in Uppsala (Tierps Tingsrätt) approved the studies.

3.1.2 Drugs The local pharmacy prepared syringes containing 5-FU (Flurablastin, Pharmacia & Upjohn; 50 mg/ml), epirubicin (Farmorubicin, Pharmacia & Upjohn; dissolved in sterile water to 2 mg/ml), and cyclophosphamide (Sendoxan, ASTA Medica; dissolved in sterile water to 20 mg/ml). The syringes were used within 24 h for intraperitoneal (ip) injections. For preparation of standards and quality controls, 5-FU and 5-bromouracil (internal standard) were purchased from Sigma Chemical Co. (St Louis, MO) while epirubicin was obtained from Pharmacia & Upjohn. CHS 828 [N-(4-chlorophenoxyhexyl)-N´-cyano-N´´-4-pyridyl-guanidine] was provided by Leo Pharmaceutical Products (Ballerup, Denmark). The drug was prepared in dark glass bottles as 5-50 mg/ml suspensions with methyl cellulose (1-11%) and Millipore water. The suspensions were refrigerated for a maximum of 1 week. Before oral administration by gavage,

25 the suspensions underwent ultrasonic radiation for 30-60 minutes and were mixed thoroughly with a magnetic stirrer. Control animals received corresponding volumes of normal saline or, in the CHS 828 studies, a mixture of methyl cellulose and water.

3.1.3 Pharmacokinetic monitoring Two to four blood samples (250-500 µl) were drawn from each rat. Sampling intervals were 15-360 min for 5-FU, 15-360 min for epirubicin, 15-240 min for 4-OHCP and 10 min to 48 h for CHS 828. Samples were immediately chilled on ice and, for plasma preparation, centrifuged in a chilled centrifuge for 5 minutes at 7 200 g (5-FU and epirubicin) or 6 minutes at 2 600 g (4-OHCP). For 5-FU and epirubicin, the plasma was frozen on dry ice. For 4-OHCP, 150 µl plasma was mixed with 300 µl cold acetonitrile and centrifuged again for 6 minutes at 2 600 g. Thereafter 300 µl of the supernatant was directly frozen on dry ice. After collection of the CHS 828 samples, the tubes were kept cold for 30-60 minutes and were then centrifuged for 10 minutes at 3 500 g. The serum was transferred to new tubes. All samples were stored at –70°C until analysis.

3.1.3.1 Chemical assays All chemicals and solvents used in the chemical analyses were of high- performance liquid chromatography (HPLC) or analytical grade. Reversed phase HPLC with ultraviolet (UV) detection was used to determine plasma concentrations of 5-FU in 150 µl plasma, according to a method described earlier.135 In Papers III and V, 0.75 ml tetrahydrofuran (THF; Merck, Darmstadt, Germany) was added per 1000 ml of mobile phase to increase the selectivity in rat plasma. 5-Bromouracil was used as internal standard. Intra- and inter-assay precisions were less than 3% and 11%, respectively, in the linear range of 0.080-50 mg/L (after addition of THF, the limit of quantification was reduced to 0.025 mg/L). Samples with concentrations expected to be over 50 mg/L were diluted with blank rat plasma prior to work-up. Epirubicin was quantified on a reversed phase HPLC system with a fluorescence detector after solid phase extraction of 150 µl plasma on Sep- Pak cartridges (Waters Corp., Milford, MA). Minor modifications of previously described extraction methods136-138 were made according to Sandström et al.91 Within-run and between-run precision was less than 7% in the range of 0.02-0.6 mg/L. At the limit of quantification, 0.005 mg/L, the coefficient of variation was 13%.

26 A reversed phase HPLC method with UV detection was used to determine the 4-OHCP concentrations after a derivation step with 2,4- dinitrophenylhydrazine (Merck, Darmstadt, Germany). The method has been described in detail elsewhere,139 but was validated for 150 µl rat plasma (300 µl supernatant) within the range of 0.034-10 mg/L. The within-run and between-run precisions were less than 5 and 9%, respectively. The accuracy was between 95 and 106%. At the limit of quantification, the recovery was 88% and the coefficient of variation was 11%. Samples predicted to be above the validated range of the analysis method were diluted with plasma/supernatant prior to work-up. CHS 828 serum concentrations were determined at LEO Pharma, by a reversed phase HPLC method with UV detection. Prior to work-up, 50 µl of the rat serum sample was mixed with 950 µl blank serum. EO 859-000 (LEO Pharma) was added to each sample as internal standard. The analytical range was 0.050-20 mg/L for 50 µl rat serum. The in-process precision ranged from 8 to 15%, while the accuracy ranged from 91 to 97%.

3.1.4 Pharmacodynamic monitoring

3.1.4.1 General toxicity At the start of the studies, the average weights of the rats were 288-295 grams. Weights were registered every day of handling. In order to avoid suffering from extravasation and the influences of surgical procedures, ip injection was the preferred route of administration. In rats that received ip injections (Papers I, III and V), the abdominal cavities were examined for signs of toxicity on the day of sacrifice. Peritoneal lesions limited the observation time in the epirubicin studies.

3.1.4.2 Haematological toxicity Blood for toxicity measurements (250 µl) was taken at baseline and followed for 11-15 days (Papers I, V and VI) or 23-25 days (Paper III) (Figure 5). In Paper V, an additional blood sample was taken 28 days after the first dose in some of the rats that received CHS 828. The blood was analysed for haemoglobin and for the number of leukocytes and platelets in a Coulter MD II Series (Coulter Electronics Ltd., Luton, England; Papers I, III, V and VI) or in a Sysmex F-800 (Toa Medical Electronics Co. Ltd., Kobe, Japan; Paper I).

27

Fibres Fibre Fibre Fibre Fibre Papers V and VI filled implantation Retrieval Retrieval Retrieval

Days after -3 -2 -1 0 1 2 3 4 5 6/7 8/9 11/12 first dose

Paper I

Paper III

Papers V and VI

Haem. Haem. Haem. Haem. Haem. Haem. Haem. Haem. Papers I and III Tox. Tox. Tox. Tox. Tox. Tox. Tox. Tox.

Haem. Haem. Haem. Haem. Haem. Papers V and VI Tox. Tox. Tox. Tox. Tox.

Figure 5. Study design in animal studies. Syringes denote times of drug administration. In some groups of animals, haematological toxicity (Haem. Tox.) was studied for up to 23-28 days (for details, see text). Words in italics and unfilled symbols denote procedures that only occurred in some of the animals.

3.1.5 Hollow fibre model The in vivo hollow fibre model presented in this thesis was modified from the method developed at NCI.47

3.1.5.1 Fibre preparation The human breast cancer cell lines MDA-MB-231140 (Papers V and VI) and MCF-7141 (Paper V), and the human leukaemia cell line CCRF-CEM142 (Paper VI), were used in the studies. The cells were maintained in RPMI- 1640 culture medium (MCF-7 in minimum essential medium), supplemented with 10% calf serum, glutamine and streptomycin/penicillin (Sigma Chemical, St Louis, MO), passaged twice a week, and harvested with trypsin / EDTA (Biochrome, Berlin, Germany). Cell suspensions of 2⋅106 cells/ml were flushed into polyvinylidine fluoride (PVDF) hollow fibres (500 kDa molecular weight cut-off, 1 mm inner diameter; Spectrum Medical Industries, Los Angeles, CA), the ends were heat-sealed and cut at 20-mm intervals. Thereafter they were incubated in vitro at 37°C in supplemented medium for two days prior to implantation.

3.1.5.2 Surgical procedure The fibres were inserted subcutaneously under anaesthesia with 2.5% enflurane (Efrane; Abbot, Stockholm, Sweden) mixed with nitrous oxide and oxygen (1.5 L/min). Two parallel rows were shaved on the back of each animal and eight small cuts were made in each row. A suture was tied

28 around one end of each fibre and, with the help of a trocar, the fibre was inserted between the skin incisions. Tissue glue (Indermil; Kendall Medical, Mansfield, MA) and skin staples were used to close the incisions and keep a short end of the suture fixed. Each rat carried 7-8 fibres from two different cell lines (one cell line in experiments where fibres were retrieved on multiple occasions, Paper VI). Fibres were retrieved from anaesthetised animals by opening the incisions and pulling the suture. After retrieval, incisions were closed with tissue glue and the fibres were incubated in supplemented medium for a maximum of 4 hours at 37°C until cell staining.

3.1.5.3 Cell-density evaluation Metabolically active cells convert MTT [(3-4,5-di-methylthiazol-2-yl)-2,5- diphenyltetrazolium bromide; Sigma Chemical] to blue formazan crystals. Fibres were incubated with 3 ml medium and 200 µl of MTT stock solution (5 mg/ml in phosphate-buffered saline, PBS) for 4 h at 37°C to evaluate the living cell density. When staining was stopped, the fibres were washed overnight by incubation at 4°C in a buffer containing PBS with 2.5% of a protamine sulphate stock solution (1%; Sigma Chemical). Each fibre was rubbed carefully, cut in half and allowed to dry for a maximum of two weeks, at which point all fibres within one experiment were analysed simultaneously. The formazan was extracted with 250 µl DMSO (dimethylsulfoxide; Sigma Chemical) per well over 4 hours, and the optical density (OD) in 150 µl was read at 570 nm. The OD from blank wells containing only DMSO was subtracted from each reading. For wells showing ODs out of the readable range (> 2.5), 75 µl extracts were diluted with 75 µl DMSO, and these values were doubled in the data analysis. To estimate the time zero cell mass in each experiment, the mean viable cell mass from three to five fibres was determined on the day of implantation. For graphical presentation, the relative cell density on the day of retrieval was calculated for each fibre:

ODretrieval day – ODimplantation day Net Growth (%) = ______(Eq. 6) ODimplantation day

Hence, a net growth of –100% represents total cell kill, while a value greater than 0% represents cell growth in the fibre since implantation day.

3.1.6 Study designs For Paper I, the doses and schedules are presented in Table 1. Days of experimental procedures are presented in Figure 5. The doses were

29 administered as single injections (half of the rats in the morning and half in the afternoon), as two injections six hours apart, or as three injections four hours apart. In the fractionated groups, the PK were studied after the first dose in half of the rats and after the last dose in the other half. A control group of animals received ip injection(s) of saline.

Table 1. Dosing schedules for 5-FU and epirubicin in Paper I.

5-FU Epirubicin Group Dose (mg/kg) Group Dose (mg/kg) 100Sa 1 x 100 7S 1 x 7.0 100D 2 x 50 7D 2 x 3.5 100T 3 x 33 7T 3 x 2.3 50S 1 x 50 3.5S 1 x 3.5 50T 3 x 17 3.5T 3 x 1.2 33S 1 x 33 2.3S 1 x 2.3 a Doses were administered as single (S), double (D) or triple (T) injections.

In Paper III, three different ip injection schedules of 5-FU were studied. One group of animals received a single injection of 127 mg/kg, the double group received two injections of 63 mg/kg each, two days apart, and the triple group was given three injections of 49 mg/kg each, on days 0, 2 and 4. A control group was also included. In Paper V, each animal carried three to four fibres containing each of the two breast cancer cell lines and one fibre, containing medium only. In every experiment, six rats were randomised into three groups, which allowed the study of two different drug regimens together with a control group in each experiment. Every drug was run in two or three independent experiments. 5- FU (125 mg/kg), epirubicin (10 mg/kg) and cyclophosphamide (120 mg/kg) were given as single ip injections, while CHS 828 was administered orally as a single dose of 375 mg/kg or as 75 mg/kg for five consecutive days. These schedules were the same as the two phase I studies of CHS 828 that were 133,134 planned at the same time these experiments started. In Paper VI, the single- and five-dose regimens of CHS 828 were studied further. Two rats of each schedule received total doses of 125 mg/kg, 250 mg/kg and 500 mg/kg. Each rat carried four fibres of MDA-MB-231 and four fibres of CCRF-CEM, and the fibres were retrieved five days after the first dose. In a separate experiment of each cell line, at a dose level of 375 mg/kg, fibres were retrieved one, three and five days after first

30 administration. The MDA-MB-231 data from Paper V were also included in the data analysis.

3.2 Clinical data The data were derived from clinical studies in the treatment of various solid cancers. All patients gave informed consent, and local human investigation committees at each participating institution approved the studies. Patients known to have received G-CSF were excluded.

3.2.1 Schedules and measurements

3.2.1.1 DMDC (2'-deoxy-2'-methylidenecytidine) Sixty-five (66 in the leukocyte and platelet analyses in Paper II) patients received an oral once-daily regimen and 85 patients received a twice-daily regimen of DMDC (Table 2).107-109 Haematological measurements were taken prior to the start of treatment, and again one, two, three and four weeks after, in the once-daily regimen and generally four, six, nine, thirteen, seventeen and twenty days after the first dose in the twice-daily regimen. A total of 823-825 observations were used in the semi-physiological model building. Median baseline cell counts were 8.9, 6.2 and 318⋅109/L for leukocytes, neutrophils and platelets, respectively.

3.2.1.2 Docetaxel Leukocytes and neutrophils (3553 observations of each type) were obtained from 601 patients in 24 phase II studies143 during the first cycle of treatment. Median baseline cell counts were 7.0 and 4.9⋅109/L for leukocytes and neutrophils, respectively. Patients received a dose of 75 or 100 mg/m2, most of them as a 1-hour infusion; however, in a few patients, two or three short infusions were given.

3.2.1.3 Paclitaxel Leukocytes and neutrophils (530 observations of each type) were obtained from 45 patients receiving paclitaxel in a total of 196 cycles (1 to 18 cycles per patient, median 3 cycles).144,145 Median baseline cell counts were 7.6 and 5.5⋅109/L for leukocytes and neutrophils, respectively. Paclitaxel was administered as a 3-hour infusion with an initial dose of 175 mg/m2 every third week. Doses were adjusted according to haematological and non- haematological toxicity levels, which resulted in a final dose range of 110- 2 232 mg/m .

31 3.2.1.4 Etoposide Leukocytes and neutrophils (682 observations of each type) were obtained from a total of 71 patients receiving a 3-day continuous infusion of etoposide in two studies.71,146 Median baseline cell counts were 7.3 and 4.9⋅109/L for leukocytes and neutrophils, respectively. The standard total dose was 375 mg/m2, but in the individualised groups, the total delivered dose ranged from 225 to 789 mg/m2. A second course of treatment was administered to 47 of the patients at least four weeks after the first course. Because information was missing when the next course started, we assumed that the leukocyte and neutrophil counts had returned to baseline at that time point, i.e. that there was no carry-over effect from the previous dose.

Table 2. Number of patients and dosage regimens included in the PD modelling of DMDC

Duration of Dose Once/Twice Total dose Number of Group administration (mg/m2) daily (mg/m2) patients (days) 1 12 10 Once 120 5 1 2 18 10 Once 180 6 3 30 10 Once 300 12 4 40 10 Once 400 4 5 30 7 Once 210 6 6 30 14 Once 420 6 7 40 7 Once 280 7 1 8 50 7 Once 350 7 2 9 3 40 7 Once 280 13 1 10 12 10 Twice 240 9 11 9 10 Twice 180 23 1 12 12 7 Twice 168 6 13 15 7 Twice 210 47 1 One patient was excluded from these groups. 2 One patient was excluded in the neutrophil analysis. 3 This group was exposed to the same schedule as group 7, but included only patients with colorectal cancer.

3.2.1.5 CPT-11 (Irinotecan) Leukocytes and neutrophils were obtained from 20 patients (79 observations of each type) during the first 21 days after receiving 350 mg/m2 CPT-11 as a 1.5-hour infusion.147 Median baseline cell counts were 7.8 and 5.2⋅109/L for

32 leukocytes and neutrophils, respectively. Total drug concentrations (lactone and carboxyl acid forms) of CPT-11 and the active metabolite SN-38 were inserted into the data set.

3.2.1.6 Vinflunine Fifty-nine patients receiving a total of 191 courses (842 observations each of leukocytes and neutrophils) were included.124-126 Median baseline cell counts were 7.0 and 4.8⋅109/L for leukocytes and neutrophils, respectively. Three different dosage schedules were studied: one administration every three weeks (30-400 mg/m2, 30 patients), weekly administrations for a total duration of four weeks (120-190 mg/m2, 14 patients) and administration on days 1 and 8 every three weeks (170-210 mg/m2, 15 patients). All administrations were 10-minute infusions.

3.3 Model building

3.3.1 Pharmacokinetics Compartment models with linear, non-linear and mixed linear and non-linear elimination were investigated. Where appropriate, absorption models, some of which predicted dose dependency, were also tried. Individual pharmacokinetic parameters and individual AUCs were obtained from doses and empirical Bayes estimates. Additional concentration-time data were included in some PK analyses to improve the models. Paper I included 5-6 samples collected from each of four additional epirubicin rats over 48 hours or less. Paper III included the 5- FU PK data from Paper I. Paper V included 34 additional rats which had received cyclophosphamide, and 12 rats with sampling times up to 30 h which had received CHS 828. Paper VI included 4-6 PK samples from each of 20 rats, after doses of CHS 828 in the range of 20-500 mg/kg. Eight of these were given a five-dose regimen where concentrations were also monitored after the fifth dose. The subroutine PRIOR in NONMEM, which allowed inclusion of information from a prior rat study (in which epirubicin was given iv),56 was used to stabilise the pharmacokinetic epirubicin analysis (Paper I). Individual PK parameters for 5-FU and epirubicin in Paper V were determined by Bayesian analysis with the population PK models presented in Paper I. All predictions in the PD modelling were based on individual concentration-time profiles, except for the analysis in Paper V, where the

33 presence of drug was included as a covariate effect on the growth rate, i.e. different growth rates were allowed for fibre OD in treated and control animals. These profiles were obtained using (i) doses and empirical Bayes estimates derived from the presented PK analyses (Papers I, III, VI), (ii) individual parameters from previously presented PK analyses106,122,145 (DMDC, paclitaxel and vinflunine in Papers II and IV), (iii) population PK models110 (docetaxel in Paper IV) or (iv) actual concentrations (etoposide and CPT-11 in Paper IV). Total concentrations were used in the modelling, except for paclitaxel, where unbound concentrations were used.

3.3.2 Pharmacodynamics

3.3.2.1 Model development (Paper I)

Basic or sigmoid Emax models were applied to relate the effect on leukocytes to the AUC. The AUC was used as the measurement of exposure because when the general model for time-dissociated relationships63 was used, a linear relationship between concentration and the direct effect was obtained, i.e. AUC described the PKPD equally as well as more complex models. To find potential schedule dependence, the AUC50 was allowed to vary between the single, double and triple dose groups. The fractional decrease in leukocytes from baseline at nadir was used in the modelling:

Nadir decrease = (Leukocytebaseline - Leukocytenadir) / Leukocytebaseline (Eq. 7)

Another summary variable for toxicity, ABC-7 (i.e. the area encompassed by the baseline and observed % decrease in leukocyte count versus time curve up to day 7 after dose administration) was also used. ABC-7 was calculated using the trapezoidal method.

3.3.2.2 Model development (Paper II) In the first analysis, the general model63 was used to relate the concentration- time profile of DMDC to the survival fraction at nadir for each haematological parameter (leukocytes, neutrophils and platelets). The survival fraction was calculated as:

Survival fraction = Cell countnadir / Cell countbaseline (Eq. 8)

Consequently, one observation per individual was modelled. Eobs,max was fixed to 1 (minimal survival fraction is zero) in the model building. A difference in EC50 or AUCEdir50 between once-daily and twice-daily schedules was also allowed.

34 In the second PD analysis, the influence of DMDC on neutrophils over time was modelled with transit compartments (Figure 6). The zero-order rate constant into the first compartment, kin, was determined as the product of the estimated circulating cell count in blood at baseline (Circ0) and the rate constant for the disappearance of neutrophils from the compartment of circulating cells, kcirc. Different numbers of transit compartments were investigated, as well as different rate constants (ktr) for proliferating (sensitive) and non-mitotic (insensitive) compartments. The concentration- time profile of DMDC was hypothesised to affect kin or drain the sensitive compartments. EDrug was a function of drug concentration. A marginating compartment in equilibrium with the circulating compartment (Circ) was investigated as well as different feedback mechanisms. Absolute counts were used but in order to satisfy the assumption of symmetrical residual errors, the observations were Box-Cox transformed with power 0.2

(Observation0.2 – 1) / 0.2 (Eq. 9)

γ Circ0 kprol IV Feedback = III,IV Circ

IV EDrug

ktr ktr ktr ktr Circulating kin Proliferative cells in II,III cells ...... blood II,III EDrug II,III kcirc = (ln2/t1/2,circ)

Transit compartments

Figure 6. Schematic illustration of the semi-physiological models of myelosuppression in Papers II, III and IV. Arrows with dashed lines denote processes that are included in only some of the final models/Papers; the respective Papers are referred to by the appropriate Roman numeral. In Papers II and III, both drug sensitive and drug insensitive transit compartments were included.

3.3.2.3 Model development (Paper III) In the model building process, transit compartments were coupled to a circulating leukocyte compartment (Figure 6). kin was determined as in Paper II. Different numbers of 5-FU-sensitive compartments and 5-FU-insensitive compartments were evaluated as well as different rate constants from the sensitive and insensitive compartments. The order of the rate constants was considered. In addition, a marginating compartment was investigated.

35 Different sensitive compartments were allowed to have different parameters for the 5-FU-dependent elimination. To characterise the rebound, feedback mechanisms which allowed the number of circulating leukocytes or number of cells in different parts of the delay chain to affect the rate constants of the delay chain were considered.

3.3.2.4 Model development (Paper IV) Each drug was fitted separately. The docetaxel, paclitaxel and etoposide data sets were used to develop the structural model (Figure 6) which consisted of one proliferative compartment, transit compartments with maturing, drug insensitive cells and a compartment of circulating blood cells. The generation of new cells in the proliferative compartment was dependent on the number of cells in the compartment, i.e. self-renewal or mitosis, as well as a proliferation rate constant determining the rate of cell division (kprol). To account for rebound phenomena, a feedback mechanism from the circulating cells to the proliferating cells was applied. The drug concentration in the central compartment was assumed to reduce the proliferation rate or induce cell loss by a linear function or an Emax model. The data contained little information regarding the elimination rate of circulating cells as that is not the rate limiting step. Therefore, it was assumed in the modelling that kcirc = ktr. At steady state, the number of cells in the proliferative compartment is constant and, the number of proliferative cells was arbitrarily set to the number of cells in the circulating compartment. Consequently, kprol was equal to ktr. To improve interpretability, the mean transit time was estimated, which was defined as MTT = (n + 1) / ktr where n is the number of transit compartments. Thus, the structural model parameters to be estimated were Circ0, MTT, γ and Slope (or Emax and EC50). For consistency, IIV was always (and only) estimated on Circ0, MTT and Slope (or EC50). Absolute cell counts were used but in the vinflunine analysis the data were log-transformed.

3.3.2.5 Model development (Paper V) Extended least squares within NONMEM was used to evaluate whether drug-treated rats had a significantly different absolute change in OD from the day of implantation (baseline) to the day of retrieval, compared to control rats;

Absorbance (OD) = Baseline + Growth (Eq. 10)

Absolute ODs were used in the modelling. Inter-experimental differences in Baseline and Growth were considered. Each drug and cell line was modelled separately. Only a few rats of each drug were included and they all received

36 the same dose and, therefore, no further PKPD modelling was performed in this study.

3.3.2.6 Model development (Paper VI) All cell density-time profiles, from all rats, were modelled simultaneously, including the cell density data from control animals. Each rat within an experiment was modelled to have the same baseline value, i.e. all three to five OD measurements on the day of implantation were added for each rat in the data set. Consequently, inter-experimental variability in baseline was estimated instead of IIV. Log-transformed OD values were used. During the model-building process, the effect of the drug (EDrug) was assumed to directly drain the cell compartment,

dOD/dt = kGrow · OD – EDrug · OD (Eq. 11) or to affect the growth rate constant,

dOD/dt = kGrow · (1 – EDrug) · OD (Eq. 12) where kGrow was the first order rate growth constant. Models with zero order and second order growth rates, as well as models estimating the amplification factor (i.e. the order) were evaluated. Different functions of effect delay were evaluated, as well as a model similar to the model for cell- cycle specific drugs presented by Jusko.86 A linear model, a basic or sigmoid Emax model, and a threshold model were investigated to test the dependency of EDrug on CHS 828 serum concentration. In addition, the general model was also investigated.63 To further improve the fit, EC50 was allowed to decrease from a baseline value (EC50,0) with time from first exposure or with cumulative AUC (AUC0-t), either linearly or exponentially with the parameter kEC50

-kEC50 ⋅ Time EC50(t) = EC50,0 · e (Eq. 13) or to decrease in a non-linear fashion (Eq. 4)89 or according to the relationship presented by Adams et al. (Eq. 5).90

3.3.3 Data analysis The software NONMEM (Versions V or VI beta)59 was used to fit non- linear mixed effects models to the data, except for the PKPD models presented in Paper I, where the Solver function in Excel was used in combination with the F-test on the sum of squares of residuals (significance level of p<0.05). In the population analyses in NONMEM, IIV (CV%) were included as exponential variance models. The residual error was modelled with an additive and/or proportional component, and allowed to vary

37 between individuals,148 when indicated by the data. Model development was guided by comparing the objective function values (OFV) produced by NONMEM of competing models in the likelihood ratio test, as well as by using graphical diagnostics in the S-PLUS based program Xpose (Versions 2 or 3).149 To discriminate between two nested models that differed by one parameter, drops in OFV of 3.84, 6.63 and 10.83 corresponding to significance levels of p<0.05, p<0.01 and p<0.001 (if χ2-distributed differences in OFV are assumed) were used. More strict criteria were generally used with the first-order (FO) method than with the first-order conditional estimation (FOCE) method with INTERACTION, as it has become evident that actual significance levels are higher than the nominal with the former method.150 A reduction in OFV of 3.84 was adopted as the criterion for inclusion of an additional transit compartment. This did not require an extra parameter.

3.3.3.1 Model validation (Paper III) A posterior predictive check151 was performed on the semi-physiological model described in Paper III. One hundred data sets were simulated from the original data set and the final model parameter estimates. The means of the minimum leukocyte counts, maximum leukocyte counts and times to the minimum and maximum leukocyte counts were calculated for each group in each simulated data set. The corresponding observed means were compared with the 90% predicted intervals of the simulated means. The 5-FU leukocyte data from Paper I was used for external validation. From the final semi-physiological model and its parameter estimates, one hundred data sets were simulated for each of the dosage groups of the validation data set. Within a dosage group, the mean observed leukocyte value at each time point was compared with the simulated means and the 90% predicted intervals of the simulated means.

38 4 Results and Discussion

4.1 Pharmacokinetics

4.1.1 Results

4.1.1.1 5-FU, epirubicin and 4-OHCP The 5-FU data in both Papers I and III were best described by a one- compartment model with non-linear elimination that followed Michaelis- Menten kinetics (Table 3). Rats with high Km also appeared to have a high Vmax. The epirubicin data were best described by a three-compartment model (Table 3). The initial dilution volume (RSE %) was 0.36 (14) L with an IIV of 39%. The half-lives of the three phases were 1.7 minutes, 43 minutes and 42 hours. In Paper V, the Bayesian analyses performed in NONMEM were assumed to be valid since the residual error magnitudes in the rats receiving 5-FU and epirubicin were lower than or within the confidence limits of the original PK models. The 4-OHCP data were best described by a one- compartment model with first order absorption (Table 3). The half-life of 4- OHCP after administration of cyclophosphamide was 37 minutes.

4.1.1.2 CHS 828 The PK models of CHS 828 in Papers V and VI were both characterised by a linear one-compartment model with the dissolution and absorption described by a zero-order rate (R1) and a consecutive first-order rate input (ka,1; Table 4). The pharmacokinetics were not dose dependent in Paper V, and the total AUC seemed to be the same in both schedules. However, in Paper VI, bioavailability decreased with dose by an exponential function with the rate constant kF and was highest for the lowest dose of 5.9 mg (F1,max). In addition, increasing serum CHS 828 concentrations were observed in Paper VI in four of five rats with at least one sample containing measurable drug in the range of 13 to 21 hours and one sample in the range of 23 to 30 hours. To explain this, a fraction of the dose (F2) was modelled to be absorbed after a

39 lag time. F2 decreased with dose, in the same manner as F1. The total bioavailability (F1 + F2) for the 5.9 mg dose was set as the reference bioavailability, i.e. 100%, and F2 = 1 - F1,max for this dose. Oral clearance increased from time zero, and was modelled to be linearly dependent on cumulative AUC (AUC0-t). Because of the dose-dependent bioavailability, the five-dose regimens produced larger AUCs than the single-dose regimens of the same total dose and the model predicted similar AUC values for the three highest single doses.

Table 3. Estimated PK parameters (RSE %) of 5-FU, epirubicin and 4-OHCP.

5-FU 5-FU Epirubicin 4-OHCP (Paper I) (Paper III) (Paper I) (Paper V) CL (L/h) 1.1 (10) 2.0 (2.3) e - - IIV 40% ne V (mg/L/h) 103 (8.0) a 108 (5.5) b max - - IIV 16 % 21 % K (mg/L) 16 (12) 22 (7.7) m - - IIV 16 % 21 % d e Vss (L) 0.24 (2.6) 0.24 (2.4) 54 1.8 (4.4) IIV ne ne - 22 % F (%) - c - 71 (10) - k 2.4 (10) 11 (3.0) a - c - IIV 33 % ne Prop error (%) 10 14 18 16 Add error (mg/ml) 0.32 0.38 ne ne a 85 mg/L/h in the afternoon. b 121 mg/L/h in the morning. c Instantaneous input and complete absorption were indicated when compared with iv data. d Calculated value. e Oral clearance and volume, not corrected for bioavailability or fraction metabolised.

4.1.2 Discussion PK models were successfully developed from preclinical data using sparse sampling. The clinical PK of 5-FU have also been described by a one- 91 compartment model with capacity-limited elimination. Estimates of Vmax and Km in patients were similar to the values reported here. A two- compartment model with parallel non-linear and linear elimination of 5-FU has been suggested in rats.94 However, such a model did not fit the data from

40 our studies any better than the presented model. The value of V reported here (0.24 L) is in accordance with previous estimates in rats of 0.25 and 94,152 0.32 L. The higher Vmax in the morning compared with the afternoon is also in accordance with an earlier study where a 10-15% higher elimination rate of 5-FU was encountered one hour after the onset of light compared with seven hours after, in perfused rat livers.153 The PK of epirubicin and 4- OHCP have been described in patients receiving the FEC regimen, as well as in rat studies, by linear three- and one-compartment models, respectively, as 56,91,154 reported here.

Table 4. Estimated PK parameters (RSE %) of CHS 828.

Paper V Paper VI CL (L/h) 0.65 a (11) 0.24 (12) c,d IIV ne 52 % a d Vss (L) 4.0 (19) 0.81 (8.7)

R1 (mg/h) 11 (17) 2.5 (6.9) ka,1 (h-1) 2.2 (32) 8.2 (30) IIV ne 170 %

F1,max - 0.62 (5.7) kF (mg-1) - 0.0089 (6.8) Lag time (h) - 20 (0.28) -1 ka,2 (h ) 0.054 (20) IIV - 160 %

SlopeCL (L/mg/h) - 0.0063 (25) Residual error (%) 82 b 39 (6.9) Half-life (h) 4.3 2.3 a Oral clearance and volume. b Combined IIV and residual variability. c At time zero. d Oral clearance and volume for the lowest dose (5.9 mg).

More data were available in the PK analysis of CHS 828 in Paper VI than in Paper V, and this allowed a more complex model to be applied. Approximately one third of the systemically available dose was absorbed at 20 hours after administration, probably as a result of coprophagy, since the gut transit time in non-fasting rats is approximately 21 hours.155 The phenomenon that clearance increases with exposure has not previously been shown for CHS 828 and requires confirmation. Autoinduction phenomena have been modelled for other drugs before;156,157 however, more data are

41 needed before a more complex autoinduction model for CHS 828 can be applied.

4.2 PKPD modelling of haematological toxicity

4.2.1 Results

4.2.1.1 Schedule dependence of 5-FU and epirubicin The reduction in leukocytes in rats receiving 5-FU was dose-dependent and was greater for the single dose groups than for the respective fractionated dose groups. The sigmoid Emax model gave a significantly better fit than the basic Emax model for ABC-7, but not for nadir decrease (Figure 7 and Table 5). Allowance for schedule-related AUC50s did not improve any of the models significantly. The apparent difference in toxicity of 5-FU was therefore attributed to non-linear PK. The platelets decreased in a pattern similar to that seen with the leukocytes. Haemoglobin also decreased in an exposure-dependent manner. Sigmoid Emax models best described the epirubicin data for both nadir decrease and ABC-7 (Figure 7 and Table 5), with no improvement when schedule-dependent AUC50s were allowed. The platelet counts after epirubicin administration tended to increase rather than decrease. Haemoglobin levels fell only slightly.

Table 5. Basic and sigmoid Emax models of nadir decrease and ABC-7.

5-FU Epirubicin Nadir decrease ABC-7 Nadir decrease ABC-7

Emax 82 % 280 84 % 452 γ 1a 2.4 1.9 1.7

AUC50 (mg·h/L) 30 20 0.75 1.2 a Fixed to 1 in the basic Emax model.

4.2.1.2 Schedule dependence of DMDC

For all haematological parameters, sigmoid Emax models with separate AUC50s for once- and twice-daily regimens were significantly better for relating the survival fractions to AUC than models with schedule- independent AUC50s. The relative increases in potency with twice-daily

42 regimens were 62 to 74% (Table 6). For all haematological parameters, the OFVs obtained using the general model63 were higher than those from the final model.

Figure 7. Model predictions of leukocyte toxicity after exposure of different dosing schedules. Upper row applies to 5-FU and shows the best fit of a basic Emax model for the nadir decrease (A) and a sigmoid Emax model to the ABC-7 (B), versus AUC. Model predictions were made at variable AUC50 for each individual dosage regimen and at constant AUC50 when combining all the dosing schedules. Lower row shows corresponding fits of sigmoid Emax models for epirubicin.

Table 6. Parameter estimates (RSE %) for once- and twice-daily regimens of DMDC.

AUC50 (mg·h/L) γ Once-daily Twice-daily Leukocytes 6.7 (23) 2.6 (24) 0.48 (25) Neutrophils 3.7 (32) 0.97 (42) 0.58 (25) Platelets 10.8 (19) 2.8 (20) 0.62 (21)

43 4.2.1.3 Semi-physiological model for DMDC The final model successfully described the data (Figure 8). It consisted of nine sensitive compartments, including the first proliferative compartment, with mean transit times of 0.98 days (RSE=6.2%, IIV=77%) each, and five non-sensitive compartments with mean transit times of 1.2 days (RSE=9.2%) each. The estimated neutrophil half-life was 21 hours (RSE=9.0%) and baseline was 5.6·109/L (RSE=3.6%, IIV = 37 %). First- order elimination of cells from the first progenitor compartment, proportional to the DMDC concentration, was adopted. The DMDC-related elimination rates were significantly different (p<0.001) for the once-daily and twice-daily regimens, with estimates of 76·ConcentationDMDC cells/day (RSE=29%) and 210·ConcentationDMDC cells/day (RSE=27%), respectively. The unexplained IIV in the effect was 130%. The drug-related effects on the other proliferating compartments were 1.8% of the effect on the first progenitor compartment. Allowance for feedback mechanisms did not improve the model significantly.

Figure 8. Typical predicted profiles (—) and observed cell counts (□) for the time- course of neutropenia for four different dosing regimens. The observations are randomly displaced (± 0.5 days) around the true observation day for illustrative purposes.

44

4.2.1.4 Semi-physiological model for 5-FU All rats receiving 5-FU showed a transient decrease and rebound in leukocyte (Figure 9) and platelet levels. The fractionated schedules of 5-FU resulted in decreases in leukocyte levels similar to or more pronounced than the decreases in the single-dose group, despite the fact that the total AUC values in the fractionated-dose groups were only 52-56 % of those in the single-dose group. On average, haemoglobin levels decreased to 59-67 % of baseline. Two 5-FU-sensitive and two 5-FU-insensitive transit compartments were included in the semi-physiological model (Table 7). Second-order delay rate constants (ktr), which were the same from all transit compartments, markedly improved the fit over first order rate constants. A negative feedback from the circulating leukocytes was applied to kin and modelled as the ratio of the estimated leukocyte baseline value (Circ0) and the estimated circulating leukocyte level at any time point (Circ). The 5-FU- dependent elimination rate of cells from the sensitive compartments was expressed as a slope value multiplied by the concentration of 5-FU (Slope⋅Conc5-FU). The observed means of the investigated leukocyte levels and their respective timing were well included in the 90% predicted intervals of the means from the simulated data sets, except for the maximum leukocyte levels in the triple group, which were underestimated. The final model also performed well in predicting the time-course of leukopenia after the dosing schedules of 5-FU in Paper I, in spite of the lower doses in that study.

Table 7. Population parameter estimates (RSE %) for the semi- physiological model of 5-FU

Parameter Estimate IIV (%) 9 -1 Circ0 (cells⋅ 10 ⋅L ) 14.0 (5.9) 12 -1 -9 -1 -5 ktr (cells ⋅ 10 ⋅day ) 5.7 ⋅ 10 (4.9) ne

t1/2, circ (min) 16 (3.6) ne Slope (L⋅mg-1⋅day-1) 0.99 (45) 120 Residual error (%) 24 -

45 Placebo group Single group 40

30

20 /L) 9 10 10 ⋅

0

Double group Triple group 40 Leukocyte count ( 30

20

10

0 0 5 10 15 20 25 0 5 10 15 20 25 Time after first dose (days) Figure 9. Observed number of leukocytes (o), a Loess smoother (S-PLUS) through the observations (---), and the model predicted number of leukocytes over time for an individual with typical PK (—). Arrows indicate days of 5-FU administration.

4.2.1.5 Semi-physiological model with parameter consistency across drugs The toxicity profiles were well characterised for all drugs, even when different schedules and several treatment cycles were modelled continuously with time (Figure 10). In order to compare results, and for the sake of simplicity, the number of transit compartments was fixed to three for all drugs. Increasing or reducing the number of transit compartments led to little improvement in OFV. The estimated half-lives of circulating cells (ln2/ktr) were 15 to 24 hours. There was no improvement in the model fit whether this half-life was estimated as a separate parameter (range 8 to 32 hours) or was fixed to a literature value of 6.7 hours.17 System-related parameters and their corresponding IIV were similar for the different drugs (Table 8). Parameter estimates for leukocytes were close to those estimated for neutrophils, except that leukocytes had higher baseline values (Circ0) and lower Slope-values. The DMDC neutrophil model was unstable, and estimates of mean transit times (MTT) were dependent on which estimation method and residual error model that was used. Therefore,

46 MTT was fixed to the estimate in the leukocyte analysis (123 hours). For CPT-11, the OFV was lower when the effect was assumed to be from the parent drug rather than from the metabolite, SN-38.

DV 12 IPREe Docetaxel 12 Paclitaxel PREDe IPREf 8 PREDf 8

4 4

0 0

0 5 10 15 20 0204060 /L) 9 12 Etoposide 12 DMDC

8 8

4 4

Neutrophils (·10 Neutrophils 0 0

0 204060 0 5 10 15 20

12 CPT-11 12 Vinflunine

8 8

4 4

0 0

0 5 10 15 20 0 204060

Time (days) Figure 10. Observations (DV), predictions based on typical population estimates (PRED), and individual predictions (IPRE) with estimated (e) and fixed (f) system- related parameters for individuals selected on the basis of their typical residual error magnitude (the individual’s average absolute weighted residual is at the population median).

To evaluate the model and assess the generality in describing 9 myelosuppression, system-related parameters were fixed: Circ0 to 7⋅10 /L 9 (5⋅10 /L for neutrophils), MTT to 125 hours, and γ to 0.17. IIV on Circ0, MTT and Slope was fixed to 35%, 20% and 45%, respectively. These fixed 47 parameter values were based on the results of the estimations. Similar Slope estimates (Figure 11), residual errors (not shown) and goodness of fit (Figure 10) were derived as when all parameters were estimated.

Table 8. Parameter estimates (RSE %) for neutrophils with a linear concentration-effect model.

a b γ c c a Circ0 MTT Slope Slopeu Add Drug d IIV IIV IIV Prop 5.05 42 88.7 16 0.161 8.58 60 429 1.2 27 Docetaxel (1.9) (7.0) (2.5) (24) (3.7) (5.2) (14)

5.20 35 127 18 0.230 2.21 43 44.2 ne 40 Paclitaxel (3.6) (11) (2.1) (30) (2.8) (32) (4.5)

5.45 42 135 14 0.174 0.126 40 0.899 0.67 45 Etoposide (7.3) (20) (3.7) (23) (6.6) (14) (78)

5.43 39 123* 49 0.160 0.782 63 0.806 ne 49 DMDC (3.9) (16) (23) (13) (9.1) (27)

5.51 29 113 29 0.132 1.29 43 3.48 0.43 34 CPT-11 (3.4) (19) (6.9) (41) (9.8) (15) (61)

4.72 41 122 22 0.162 .00349 41 .0159 ne 48 Vinflunine (2.7) (18) (3.7) (21) (6.7) (7.8) (33) a unit: ⋅109/L; b unit: hours; c unit: µM-1; d unit: % Abbreviations: Add, additive residual error; Prop, proportional residual error; *, fixed.

4.2.2 Discussion Relationships between AUC and the effect on leukocytes were established for 5-FU and epirubicin. It was apparent that a small increase in exposure can have a large impact on the degree of toxicity. The approximate AUCs achieved after 600 mg/m2 5-FU and 60 mg/m2 epirubicin in the FEC regimen in patients are 13 mg·h/L and 1.8 mg·h/L, respectively. If similar PKPD relationships are assumed in patients, 5-FU is likely to have less contribution than epirubicin to myelosuppression. An AUC of 18 mg·h/L resulted in a 50% risk of toxicity (WHO toxicity > grade 1) after a single injection of 5- 158 FU to patients, which is close to the AUC50 for ABC-7 in rats (20 mg·h/L; Paper I). The AUC-dependent Emax model for 5-FU injections in Paper I predicted the nadir values in Paper III to be 33, 41 and 42 % of baseline for the single-, double- and triple-dose groups, respectively. The observed means were 37, 36 and 28%. The decrease in leukocytes in the triple-dose group was greater than expected, probably explainable by the hypothesis that primitive

48 haematopoietic stem cells cycle too slowly to be sensitive to a single dose of 5-FU.28 However, if the first dose stimulated the cycling activity of the primitive stem cells, the cells would be vulnerable to doses administered 3 to 5 days after the first dose. It is obvious that the presence and characteristics of 5-FU’s schedule dependence relies not only on the rate of delivery (bolus vs infusion), but also on the time between administrations.

Docetaxel

102 Paclitaxel

101

estimated Etoposide CPT-11 0 10 DMDC with u

10-1 Slope system-related parameters parameters system-related

10-2 Vinflunine

10-2 10-1 100 101 102

Slopeu with fixed system-related parameters Figure 11. Slopeu when system-related parameters were estimated vs Slopeu when system related parameters were fixed for leukocytes (•) and neutrophils (ο).

For 5-FU and epirubicin, no difference between once- and twice-daily dosing was found but, for DMDC, the survival fractions at nadir were significantly lower after twice-daily than after once-daily dosing. The general model could not explain this schedule dependence without including schedule-dependent EC50 values. The mechanism of action of DMDC could be of such a nature that the general model cannot explain it. Furthermore, the treatment lengths also varied, and the nadir was only evaluated weekly. Nadir values were then generally overestimated, to a different extent for different treatment lengths. It is thus desirable to model the whole time course of myelosuppression, especially in the evaluation of schedule dependence. In agreement with the aim of mechanistically based models, many features of the models discussed here mimic physiological theories of the structure and regulation of the granulopoietic system. A model that can mimic physiological processes is eligible for predictions, but since physiology changes during stresses such as sickness and drug therapy, fixing parameters to values from the literature was not an attractive option. The complexity of the models is limited by the fact that they are developed from

49 only one type of observation, i.e. the number of circulating cells. Even though the models contain relatively few parameters, especially the models in Papers III and IV, they are able to describe the data well. The model in Paper IV estimated similar system-related parameters for six different drugs with different mechanisms of action, and it was shown that relatively small and sparse data sets could be successfully modelled. This, together with the evaluation of the model, implies that this model could be used with even more sparse data sets using prior information on the system-related parameters. In the semi-physiological model of DMDC, the estimated transit time through the proliferative compartments was 8.8 days, whereas a value of 5 days for myelocytes to pass the precursor proliferation pool has been reported.13 The mean transit time for the non-proliferating pool was 6.2 days in Paper II, similar to the 6.6 days reported in the literature for healthy volunteers.14 In Paper IV, the estimated MTTs were somewhat shorter. However, myelosuppression induces release of growth factors that stimulate extra cell divisions in the proliferating compartments, thus shortening the time for mature cells to reach the circulation.159 Time to emergence can also be decreased by the presence of infections.160 The half-lives of circulating neutrophils estimated in Paper II (21 hours) and Paper IV (15 to 24 hours) were longer than those reported in the literature.14,16,17 Other modellers have also reported longer neutrophil half- lives than expected.161 The estimated half-life of 16 minutes for circulating leukocytes in rats was, on the other hand, somewhat shorter than expected.20 However, with only daily to weekly leukocyte counts, the temporal resolution is limited; also, the elimination of cells from the circulation is not the rate-limiting process. In contrast to the AUC-dependent model derived for 5-FU in Paper I, the more mechanistic model presented in Paper III could predict the observed schedule-dependent decrease in leukocytes when the intervals between the doses were extended to two days. The fractionated schedules in Paper III resulted in a huge rebound of leukocytes, which has also been shown after repeated short infusions of 5-FU in patients.162 However, the model underestimated the maximum leukocyte level in the triple dose group. It is likely that more information about the system is required, since the leukocyte rebound would depend on the functional capacity of the system, rather than on the drug. The second-order rate constants included in Paper III could be interpreted as a part of the regulation of leukocyte production, since haematopoietic growth factors cause extra cell divisions, and they also shorten the transit times within the maturation chain.159 When the number of cells within a transit compartment falls below baseline, an amplification factor higher than

50 one will lead to the prioritisation of increasing the number of cells within the compartment, before transferring the cells to the next compartment. When the cell number is higher than baseline, transit time is increased. In this way the rapid and pronounced change in the number of circulating cells was better characterised. In vitro and/or preclinical in vivo data could be helpful in predicting the clinical concentration-effect relationship. Approximate Slope estimates could be derived from animal data in vivo and Slopeu estimates could be derived by comparing the relative potencies of drugs on bone marrow progenitor cells in vitro. In combination with the suggested values of system-related parameters in Paper IV, the whole time-course of myelosuppression in patients could be simulated for new drugs. The model has been applied to preclinical in vivo data,163 and been used for evaluation of drug combinations.163,164 In addition, the model has been extended to be valid for platelets165 and leukocytes after G-CSF administration.166 Thus, the model developed in Paper IV, has several potential applications in drug development and therapeutic optimisation. Other semi-physiological models for myelosuppression have been reported recently, in addition to the model reported by Minami et al.78 Zamboni et al. added an effect compartment to Minami’s model and exchanged the lag time for one transit compartment.167 The addition of an effect compartment did not, however, improve the fit of our models. Krzyzanski and Jusko developed a multiple-pool lifespan model with a zero- order input of cells.121 The transit times through mitotic and non-mitotic compartments, as well as in the circulation, were fixed lifespans without variation between the cells.

4.3 The hollow fibre model in immunocompetent rats

4.3.1 Results

4.3.1.1 Development of the hollow fibre model in rats Paper V illustrated that the hollow fibre model in rats was successful in obtaining PK, toxicity and antitumour effect data simultaneously within the same animal. The rats receiving 5-FU, epirubicin and cyclophosphamide showed a pronounced decrease in haematological parameters, while the rats receiving either regimen of CHS 828 were practically unaffected. In Paper VI, the highest dose level of CHS 828 (500 mg/kg) produced a decrease in leukocytes, which was most pronounced in the five-dose regimen. However,

51 the myelosuppression was limited (30% decrease from baseline). The five- dose regimen also produced the greatest weight loss in the rats (11% decrease from baseline). The drug effect on the fibres was generally low in Paper V (Figure 12). Neither 5-FU nor epirubicin had significant effects on the growth of either cell line. With cyclophosphamide, the drug effect was low but was overall statistically significant compared with controls for both cell lines. When given as a single dose, CHS 828 exerted a low but significant effect on MDA-MB-231, but not on MCF-7. For the five-dose regimen, the net growth of MCF-7 was decreased and the drug produced a cell-killing effect on MDA-MB-231 (Figure 12). a)

600

500 Ve h ic le 400 Treated

300

200

100 Net growth (%) 0

-100 5-FUEP ICP CHS 828 CHS 828 x1 x5

b)

300 Ve h ic le Treated 200

100

Net growth (%) growth Net 0

-100 5-FUEPICP CHS 828 CHS 828 x1 x5 Figure 12. Effects of different drugs on MDA-MB-231 (a) and MCF-7 (b) in the hollow-fibre model in immunocompetent rats. Net growth means (± SE) are presented from one to three experiments of 5-FU, epirubicin (EPI), cyclophosphamide (CP) or CHS 828 with corresponding controls (vehicle).

4.3.1.2 Schedule dependence of CHS 828 Depression of the net growth of MDA-MB-231 cells was seen in fibres from only one rat in the single-dose regimen (500 mg/kg) while, for the five-dose regimen, the cell kill increased in a dose-dependent manner (Figure 13A). The CCRF-CEM cells were more sensitive than the MDA-MB-231 cells. However, the five-dose regimen still resulted in a larger cell kill, at least for 52 the lower doses (Figure 13C). The single dose of 375 mg/kg produced similar net cell growths over time to those seen in the controls (MDA-MB- 231; Figure 13B) or with the five-dose regimen (CCRF-CEM; Figure 13D). Exponential growth with an effect of drug on the growth rate (Eq. 12) was adopted (Figure 14). The final effect models of CHS 828 were a threshold model (γ=20) for MDA-MB-231 and a basic Emax model (γ=1) for CCRF- CEM (Table 9). EC50 was modelled to decrease exponentially over time from first exposure for both cell lines.

Five dose schedule Single dose Control

300 A 300 B

200 200

100 100

0 0

-100 -100 0 100 200 300 400 500 600 0 1 2 3 4 5 6

Net growth (%) growth Net 300 C 300 D

200 200

100 100

0 0

-100 -100 0 100 200 300 400 500 600 0 1 2 3 4 5 6

CHS 828 total dose (mg/kg) Time (days)

Figure 13. Net growth for different dose levels after fibre retrieval, five days after first administration (left) and net growth over time for the 375 mg/kg doses (right) for MDA-MB-231 (A, B) and CCRF-CEM (C, D). Means +/- SE.

4.3.2 Discussion Our extended hollow-fibre model in rats had the potential for studying antitumour effects, toxicity and PK, within the same animal. The use of immunocompetent animals reduces the costs, makes handling easier and allows determination of haematological toxicity. The time frame of six days for the fibres to remain inside the rats is comparable to that used in the NCI experiments.47 Paper VI also showed that this model is suitable for studying the antitumour effects of drugs over time.

53 Table 9. Parameter estimates (RSE%) for the PD model of the effect of CHS 828 on tumour cells in hollow fibres.

MDA-MB-231 CCRF-CEM Estimate IIV (%) Estimate IIV (%)

a ODBaseline 0.463 (5.3) 20 (84) 0.486 (3.4) ne -1 kGrow (Day ) 0.178 (5.8) ne 0.171 (8.4) 38 (53) -1 Emax (Day ) 3.39 (3.2) ne 3.80 (12) ne

EC50,24h (mg/L) 11.2 (57) 200 (38) 19.5 (42) ne -1 kEC50 (Day ) 0.620 (18) ne 1.43 (16) 80 (54) Residual error (%) 29.8 (7.7) - 34.8 (9.2) - a Inter-experimental variability.

kGrow EDrug γ Emax ·C EDrug = γ γ C + EC50(t)

Viable cells -kEC50 ·Time EC50(t) = EC50,0 ·exp (OD)

EC50(t) for MDA-MB-231 EC50(t) for CCRF-CEM Conc Single dose Conc Five dose schedule 10 (t) (mg/L) (t) 50

1

0.1 Concentration/EC 0.01

012345 Time (Days)

Figure 14. PD model of the effect of CHS828 on hollow fibres. EDrug was a threshold model (γ=20) for the MDA-MB-231 cell line, and a basic Emax model (γ = 1) for the CCRF-CEM cell line. EC50 decreased exponentially with time of exposure. EC50(t) is illustrated graphically with concentration-time profiles for both schedules in typical rats after a total dose of 500 mg/kg.

54

CHS 828 exerted higher antitumour activity than the standard drugs in the FEC regimen despite the lack of significant toxicity associated with the 375 mg/kg dose of CHS 828, while the other drugs were used close to their maximum tolerated dose. However, from the phase I study of CHS 828, it is apparent that patients have a much lower tolerability to CHS 828 than rats.133 The maximum tolerated AUC was around 1 mg·h/L in patients after a five- day schedule while the rats receiving a five-day schedule of a total of 375 mg/kg, achieved an AUC of approximately 225 mg·h/L. It is therefore likely that there is a species difference in sensitivity to CHS 828. The FEC regimen may produce clinical response rates of up to 90% in the treatment of metastatic breast cancer,168 but only cyclophosphamide had a significant effect on tumour growth in the hollow-fibre model here. This may be because MDA-MB-231 and MCF-7 are relatively resistant cell lines, because a combination of two or three of the drugs is necessary to produce an effect or, because drug delivery is dependent on blood vessel development.169 Low effects of standard chemotherapeutic agents on MDA- MB-231 cells have also been reported from the original hollow-fibre model.47 Paper VI showed that the lack of effect on the net growth of MDA-MB- 231 after the 375 mg/kg single dose was not due to re-growth (Fig. 13B) and that the schedule-dependent antitumour effect of CHS 828 in vivo is partly due to dose-dependent bioavailability and partly due to schedule dependence in the PD. As a consequence, even when formulated to avoid the dose- dependent bioavailability, CHS 828 will exert an antitumour effect that is dependent on the rate of administration. The rat hollow fibre model was shown to be suitable for population PKPD modelling of antitumour effects over time in vivo. However, a time- dependent EC50 is an empirical explanation for the schedule dependence. Indeed, it is likely that the true EC50s are overestimated as, in this model, EC50(t) also accounts for the effect delay. Knowledge of the cell-killing mechanisms of CHS 828 would have allowed information from in vitro studies to be included, possibly contributing to a more mechanistic model. It has been suggested that the dependence of effect on dosage schedule may be due to two different mechanisms of action of different potencies, of which the lower potency mechanism is independent of exposure time while the more potent mechanism only acts at longer exposure times.132 The combination of this relatively simple tumour model with computer modelling and simulation has potential for assistance in guiding schedule decisions in the development of new drugs.

55 5 Conclusions

• It is of importance to characterise schedule dependence in the optimisation of anticancer therapy since the administration rate can have a large impact on the therapeutic index. Apparent schedule dependence can be due to dose-dependent PK and if the effect is not dependent on AUC, there is schedule dependence in the PD. To understand schedule dependence, the PK need to be characterised. Population PK models were successfully established in rats using sparse sampling (2-4 samples per rat). 5-FU showed capacity-limited elimination and CHS 828 demonstrated a dose-dependent bioavailability and a potential autoinduction phenomenon. Epirubicin and 4-OHCP showed linear PK.

• Population PKPD is valuable in both preclinical and clinical studies of anticancer drugs, as it allows characterisation of both PK and PD within the same individual. Quantitative relationships between AUC and summary measurements of haematological toxicity were established in rats for 5-FU and epirubicin. The models suggest that 5-FU contributes little to myelosuppression in the FEC-regimen, compared with epirubicin. Fractionation of a single injection into two or three injections, administered within eight hours, did not change the haematological toxicity for either 5-FU or epirubicin, when the non-linear PK of 5-FU was accounted for. However, when the time interval between the injections was extended to two days for 5-FU, the fractionated schedules were more toxic.

• PD models should preferably mimic underlying physiology and describe the whole time course of drug effect, as there are several drawbacks of using summary measurements of PD variables. (Semi-) physiological models were developed that were based on knowledge of the haematopoietic system. Such models successfully described myelosuppression in patients (DMDC) and in rats (5-FU), after different administration schedules. Transit compartments were applied to account for the observed lag time and feedback mechanisms from circulating to proliferative cells were included to characterise the rebound.

56 • A PKPD model useful in drug development is relatively simple, have interpretable parameters and can separate drug-related and system-related parameters, where the latter shows consistency across drugs. Small and sparse data sets should also be possible to apply. Such a model of myelosuppression was developed which included a self-renewal mechanism, compared with a zero-order rate input of cells in previous models. The model successfully characterised myelosuppression after administration of each one of six different drugs. Successive courses and different schedules were also well described.

• The developed hollow fibre model in immunocompetent rats can be used to evaluate PK and PD (tumour response and/or toxicity) simultaneously. Several tumours of different types can be studied in the same animal and the time course of drug effect can be monitored. The use of immunocompetent animals reduces the cost, have practical advantages and haematological toxicity can be evaluated simultaneously. This model was also shown useful for obtaining data for PKPD modelling.

• CHS 828 showed superior effect of five-dose regimens over single doses of the same total size. The developed PKPD-model included exponential growth and the drug potency increased with time from first exposure. This model showed that the schedule dependence in rats is partly due to dose-dependent bioavailability and partly due to a schedule-dependent effect in the PD.

57 6 Acknowledgement

The studies in this thesis were carried out at the Department of Pharmaceutical Biosciences, Division of Pharmacokinetics and Drug Therapy, Faculty of Pharmacy, Uppsala University, Sweden. I would like to express my sincere gratitude to all who have contributed to this thesis and especially to:

Professor Mats Karlsson, for being an excellent supervisor with extensive scientific knowledge, for your enthusiastic and inspiring guidance through the field of PKPD-modelling and for always being there. It is also very fun to work with you, “the man with the big brain”, and all your bright ideas!

Professor Margareta Hammarlund-Udenaes, head of the division, for providing excellent working facilities, and for your support, also in the equation NONMEM cannot solve; work + family.

Professor Lennart Paalzow, former head of the division, for your contributions to the division and for putting Uppsala on the “map of PKPD”, for providing excellent working facilities and for your stories and smiles.

My co-authors in the “cancer-group”: Dr Agneta Freijs for inspiring me for PKPD and encouraging me to become a PhD-student. Dr Marie Sandström for your positive attitude and for being a great mentor of science, teaching and life in general! Anja Henningsson for always having time for discussions of all types of matters, for all support during my last maternity leave and for proof-reading my thesis introduction.

My co-authors at the Department of Clinical Pharmacology: Professor Rolf Larsson for your enthusiasm and generosity. Dr Elin Jonsson/Lindhagen for fruitful collaboration, discussions and introduction to the mystery of cells and fibres, any time a day (and night)! Saadia Hassan for your skilful contribution to the results and for discussions of science and cultural matters.

My other co-authors for good collaboration, especially Dr Charles Brindley for support and for providing challenging DMDC data (I hope to meet you

58 some time!), Laurent Nguyen for your contributions to the vinflunine analysis, and Hugo Maas, for making myelosuppression models progress so nicely during maternity leaves!

My co-author and room-mate, Dr Ulrika W, and Dr/”Tant” Ulrika S, for your friendship, for all our more or less constructive discussions of science, relations and other important matters in life…, for being great travelling company, for fun detours (“var är vi nu?”), for concerns of my blood sugar… and for keeping me in shape during my first PhD years!

My degree project students and “summer workers”, Bodil, Niclas, Lena, Petra, Elin, Anna, Henrik, Magnus, Mattias and Therese for your contributions to my work and for good company.

My first room-mates, Dr Lena Sj/B and Pia for scientific and non-scientific discussions about everything and nothing in the “big belly” room.

Jessica and Ing-Marie for skilful technical assistance and for making life enjoyable “down under”. Britt for your invaluable knowledge of HPLC. Marianne, Karin S, Anna Sofia, Marina, Annica and Kjell for valuable practical help. Lena, Carina and Kicki at Clinical Pharmacologyyou’re your patience with my students and me during the blood cell analysis the first years.

Dr Niclas, Lasse and Jakob for providing Xpose and arranging the cluster (to make modelling even more fun!) and for taking time with my computer- related problems…

Siv, for your concern for everyone, Dr Rikard, the last Macintoshianian, for trying to bring my Cricket graphs back to life, Karin T for all your outstanding remarks. All other present and former colleagues at the division, for contributing to the great atmosphere and unforgettable Christmas parties; Kristin, Emma x 2, Jacob, Ulrika E, Anubha, Mia, Fredrik, MaMa, Jörgen, Anders V, Ann-Marie, Johan, Elisabet, Sven, Anders G, Birgit, Kati, Evur, Hasse, Anders L, Michael, Mohammed, Torgny, Gustaf, Johan, Linus, Håkan, Lotta, Irene, René, Mariann, Toufigh, Rujia, Marie G and Urban.

The Swedish Cancer Society, LEO Pharma, Quintiles Scotland Ltd, the Swedish Academy of Pharmaceutical Sciences and The IF foundation for financial support.

59 All other friends for all good times and great dinners. A special thank to the Wulff, Gullbo and Bergman-Persson families, for giving Sofia and Tilda (and Patrik) a great time when I had to work on my “book”.

All relatives for your support and care, especially, my sister in law, Johanna, for being a great aunt and helping us with our two miracles, and my mother in law, Kristina, for your thoughtfulness and invaluable help.

My parents, Martha and Leif, for everything… for always supporting me and for being there for me anytime, anywhere. Words cannot express what you mean to me!

My precious Miracles, Sofia and Tilda, for bringing me so much joy, everyday! Your hugs and laughs really remind me of what is most important in life!

Patrik, my love and my best friend, for all your support and patience, for being a great daddy and for being the one you are! Now the roe deer, elks and fishes need to watch out again!

Lena

60 7 References

1. Newell DR: Can pharmacokinetic and pharmacodynamic studies improve cancer chemotherapy? Ann Oncol 5:9-14; discussion 15, 1994 2. Desoize B, Robert J: Individual dose adaptation of anticancer drugs. Eur J Cancer 30A:844-51, 1994 3. van Warmerdam LJC, van den Bemt BJF, Ten Bokkel Huinik WW, et al: Dose individualisation in cancer chemotherapy: Pharmacokinetic and pharmacodynamic relationships. Cancer Research, Therapy and Control 4:277-291, 1995 4. Gurney H: Dose calculation of anticancer drugs: a review of the current practice and introduction of an alternative. J Clin Oncol 14:2590-2611, 1996 5. Ratain MJ: Therapeutic relevance of pharmacokinetics and pharmacodynamics. Semin Oncol 19:8-13, 1992 6. Lokich J, Anderson N: Dose intensity for bolus versus infusion chemotherapy administration: review of the literature for 27 anti neoplastic agents. Ann Oncol 8:15- 25, 1997 7. Gurney H, Dodwell D, Thatcher N, et al: Escalating drug delivery in cancer chemotherapy: a review of concepts and practice--Part 1. Ann Oncol 4:23-34, 1993 8. Sobrero AF, Aschele C, Bertino JR: Fluorouracil in colorectal cancer--a tale of two drugs: implications for biochemical modulation. J Clin Oncol 15:368-81, 1997 9. Lee WM, Dang CV: Control of cell growth and differentiation, in Hoffman R, Benz EJ, Sanford J, et al (eds): Hematology. Basic principles and practice (ed 3rd). New York, Churchill Livingstone, 1999, pp 57-71 10. Ulich TR, del Castillo J: The hematopoietic and mature blood cells of the rat: their morphology and the kinetics of circulating leukocytes in control rats. Exp Hematol 19:639-48, 1991 11. Botnick LE, Hannon EC, Hellman S: Nature of the hemopoietic stem cell compartment and its proliferative potential. Blood Cells 5:195-210, 1979 12. Maloney MA, Patt HM: Granulocyte transit from bone marrow to blood. Blood 31:195-201, 1968 13. Finch CA, Harker LA, Cook JD: Kinetics of the formed elements of human blood. Blood 50:699-707, 1977 14. Dancey JT, Deubelbeiss KA, Harker LA, et al: Neutrophil kinetics in man. J Clin Invest 58:705-15, 1976 15. Athens JW, Raab SO, Haab OP, et al: Leukokinetic studies. III. The distribution of granulocytes in the blood of normal subjects. J Clin Invest 40:159-164, 1961

61 16. Mauer AM, Athens JW, Aschenbrucker H, et al: Leukokinetic studies. II. A method for labeling granulocytes in vitro with radioactive diisopropylfluorophosphate (DFP32). J Clin Invest 39:1481-1486, 1960 17. Cartwright GE, Athens JW, Wintrobe MM: The kinetics of granulopoiesis in normal man. Blood 24:780-803, 1964 18. Gerecke D, Schultze B, Maurer W: Kinetics of neutrophilic granulocytes in the blood of rats. Cell Tissue Kinet 6:369-78, 1973 19. MacLennan ICM, Drayson MT: Normal lymphocytes and non-neoplastic lymphocyte disorders, in Hoffbrand AV, Lewis SM, Tuddenham EGD (eds): Postgraduate haematology (ed 4th). Oxford, UK, Butterworth Heinemann, 1999, pp 296 20. Westermann J, Puskas Z, Pabst R: Blood transit and recirculation kinetics of lymphocyte subsets in normal rats. Scand J Immunol 28:203-10, 1988 21. Haurie C, Dale DC, Mackey MC: Cyclical neutropenia and other periodic hematological disorders: a review of mechanisms and mathematical models. Blood 92:2629-40, 1998 22. Verfaillie CM: Anatomy and physiology of hematopoiesis, in Hoffman R, Benz EJ, Sanford J, et al (eds): Hematology. Basic principles and practice (ed 3rd). New York, Churchill Livingstone, 1999, pp 139-154 23. Lord BI, Bronchud MH, Owens S, et al: The kinetics of human granulopoiesis following treatment with granulocyte colony-stimulating factor in vivo. Proc Natl Acad Sci U S A 86:9499-503, 1989 24. Price TH, Chatta GS, Dale DC: Effect of recombinant granulocyte colony-stimulating factor on neutrophil kinetics in normal young and elderly humans. Blood 88:335-340, 1996 25. Takatani H, Soda H, Fukuda M, et al: Levels of recombinant human granulocyte colony-stimulating factor in serum are inversely correlated with circulating neutrophil counts. Antimicrob Agents Chemother 40:988-91, 1996 26. Ericson SG, Gao H, Gericke GH, et al: The role of polymorphonuclear neutrophils (PMNs) in clearance of granulocyte colony-stimulating factor (G-CSF) in vivo and in vitro. Exp Hematol 25:1313-25, 1997 27. Moore MJ, Theissen JJ: Cytotoxics and Irreversible Effects, in van Boxtel CJ, Holford NHG, Danhof M (eds): The in vivo study of drug action. Amsterdam, Elsevier Science Publisher B. V., 1992, pp 377-400 28. Harrison DE, Lerner CP: Most primitive hematopoietic stem cells are stimulated to cycle rapidly after treatment with 5-fluorouracil. Blood 78:1237-40, 1991 29. Cebon J, Layton JE, Maher D, et al: Endogenous haemopoietic growth factors in neutropenia and infection. Br J Haematol 86:265-74, 1994 30. Tannock IF: Experimental chemotherapy and concepts related to the cell cycle. Int J Radiat Biol Relat Stud Phys Chem Med 49:335-55, 1986 31. Sundman-Engberg B, Tidefelt U, Paul C: Toxicity of cytostatic drugs to normal bone marrow cells in vitro. Cancer Chemother Pharmacol 42:17-23, 1998

62 32. Levy G: The case for preclinical pharmacodynamics, in Yacobi A (ed): Integration of pharmacokinetics, pharmacodynamics, and toxicokinetics in rational drug development. New York, Plenum Press, 1993, pp 7-13 33. Breimer DD, Danhof M: Relevance of the application of pharmacokinetic- pharmacodynamic modelling concepts in drug development. The "wooden shoe' paradigm. Clin Pharmacokinet 32:259-67, 1997 34. Lieberman R, McMichael J: Role of pharmacokinetic-pharmacodynamic principles in rational and cost- effective drug development. Ther Drug Monit 18:423-8, 1996 35. Burtles SS, Newell DR, Henrar RE, et al: Revisions of general guidelines for the preclinical toxicology of new cytotoxic anticancer agents in Europe. The Cancer Research Campaign (CRC) Phase I/II Clinical Trials Committee and the European Organization for Research and Treatment of Cancer (EORTC) New Drug Development Office. Eur J Cancer 31A:408-10, 1995 36. Kennedy JM, Riji AM: Effects of surgery on the pharmacokinetic parameters of drugs. Clin Pharmacokinet 35:293-312, 1998 37. Peck CC, Barr WH, Benet LZ, et al: Opportunities for integration of pharmacokinetics, pharmacodynamics, and toxicokinetics in rational drug development. Clin Pharmacol Ther 51:465-73, 1992 38. Aarons L, Mandema JW, Danhof M: A population analysis of the pharmacokinetics and pharmacodynamics of midazolam in the rat. J Pharmacokinet Biopharm 19:485- 96, 1991 39. Newell DR, Burtles SS, Fox BW, et al: Evaluation of rodent-only toxicology for early clinical trials with novel cancer therapeutics. Br J Cancer 81:760-8, 1999 40. Schurig JE, Florczyk AP, Bradner WT: The mouse as a model for predicting the myelosuppressive effects of anticancer drugs. Cancer Chemother Pharmacol 16:243- 6, 1986 41. Arbuck SG: Workshop on phase I study design. Ninth NCI/EORTC New Drug Development Symposium, Amsterdam, March 12, 1996. Ann Oncol 7:567-73, 1996 42. Plowman J, Dykes DJ, Hollingshead M, et al: Human tumor xenograft models in NCI drug development, in Teicher BA (ed): Anticancer drug development guide. Totowa, NJ, Human Press, 1997, pp 101-125 43. Goldin A, Venditti JM, Macdonald JS, et al: Current results of the screening program at the Division of Cancer Treatment, National Cancer Institute. Eur J Cancer 17:129- 42, 1981 44. Mattern J, Bak M, Hahn EW, et al: Human tumor xenografts as model for drug testing. Cancer Metastasis Rev 7:263-84, 1988 45. Langdon SP, Hendriks HR, Braakhuis BJ, et al: Preclinical phase II studies in human tumor xenografts: a European multicenter follow-up study. Ann Oncol 5:415-22, 1994 46. Johnson JI, Decker S, Zaharevitz D, et al: Relationships between drug activity in NCI preclinical in vitro and in vivo models and early clinical trials. Br J Cancer 84:1424- 31, 2001

63 47. Hollingshead MG, Alley MC, Camalier RF, et al: In vivo cultivation of tumor cells in hollow fibers. Life Sci 57:131-41, 1995 48. Hollingshead M, Plowman J, Alley M, et al: The hollow fiber assay, in Fiebig H, Burger A (eds): Contributions to Oncology, Volume 54: Relevance of tumor models for anticancer drug development. Freiburg, Germany, Karger, 1999, pp 109-120 49. Jonsson EN, Wade JR, Karlsson MO: Nonlinearity detection: Advantages of nonlinear mixed-effects modeling. AAPS PharmSci 2:article 32, 2000 50. Collart L, Blaschke TF, Boucher F, et al: Potential of population pharmacokinetics to reduce the frequency of blood sampling required for estimating kinetic parameters in neonates. Dev Pharmacol Ther 18:71-80, 1992 51. Sheiner L, Wakefield J: Population modelling in drug development. Stat Methods Med Res 8:183-93, 1999 52. Hashimoto Y, Sheiner LB: Designs for population pharmacodynamics: value of pharmacokinetic data and population analysis. J Pharmacokinet Biopharm 19:333-53, 1991 53. Burtin P, Mentre F, van Bree J, et al: Sparse sampling for assessment of drug exposure in toxicological studies. Eur J Drug Metab Pharmacokinet 21:105-11, 1996 54. Nedelman JR, Gibiansky E, Tse FL, et al: Assessing drug exposure in rodent toxicity studies without satellite animals. J Pharmacokinet Biopharm 21:323-34, 1993 55. Looby M, Linke R, Weiss M: Pharmacokinetics and tissue distribution of idarubicin and its active metabolite idarubicinol in the rabbit. Cancer Chemother Pharmacol 39:554-6, 1997 56. Sandström M, Simonsen LE, Freijs A, et al: The pharmacokinetics of epirubicin and docetaxel in combination in rats. Cancer Chemother Pharmacol 44:469-74, 1999 57. Rouits E, Guichard S, Canal P, et al: Non-linear pharmacokinetics of irinotecan in mice. Anticancer Drugs 13:631-5, 2002 58. van Bree J, Nedelman J, Steimer JL, et al: Application of sparse sampling approaches in rodent toxicokinetics: A prospective view. Drug Inf J 28:263-279, 1994 59. Beal SL, Sheiner LB: NONMEM Users Guides, NONMEM project group, University of California at San Francisco, 1992 60. Egorin MJ, Van Echo DA, Whitacre MY, et al: Human pharmacokinetics, excretion, and metabolism of the anthracycline antibiotic menogaril (7-OMEN, NSC 269148) and their correlation with clinical toxicities. Cancer Res 46:1513-20, 1986 61. Trump DL, Egorin MJ, Forrest A, et al: Pharmacokinetic and pharmacodynamic analysis of fluorouracil during 72-hour continuous infusion with and without dipyridamole. J Clin Oncol 9:2027-35, 1991 62. 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 63. Karlsson MO, Molnar V, Bergh J, et al: A general model for time-dissociated pharmacokinetic-pharmacodynamic relationship exemplified by paclitaxel myelosuppression. Clin Pharmacol Ther 63:11-25, 1998

64 64. Hill A: The possible effects of the aggregation of the molecules of haemoglobin on its dissociation curves. J Physiol (Lond) 40:IV-VIII, 1910 65. Wagner JG: Kinetics of pharmacologic response. I. Proposed relationships between response and drug concentration in the intact animal and man. J Theor Biol 20:173- 201, 1968 66. Crawford J, Ozer H, Stoller R, et al: Reduction by granulocyte colony-stimulating factor of fever and neutropenia induced by chemotherapy in patients with small-cell lung cancer. N Engl J Med 325:164-70, 1991 67. Rosner GL, Müller P: Pharmacodynamic analysis of hematologic profiles. J Pharmacokinet Biopharm 22:499-524, 1994 68. Karlsson MO, Port RE, Ratain MJ, et al: A population model for the leukopenic effect of etoposide. Clin Pharmacol Ther 57:325-34, 1995 69. Ackland SP, Ratain MJ, Vogelzang NJ, et al: Pharmacokinetics and pharmacodynamics of long-term continuous-infusion doxorubicin. Clin Pharmacol Ther 45:340-7, 1989 70. Miller AA, Tolley EA, Niell HB, et al: Pharmacodynamics of prolonged oral etoposide in patients with advanced non-small-cell lung cancer. J Clin Oncol 11:1179- 88, 1993 71. Ratain MJ, Schilsky RL, Choi KE, et al: Adaptive control of etoposide administration: impact of interpatient pharmacodynamic variability. Clin Pharmacol Ther 45:226-33, 1989 72. Bodey GP, Buckley M, Sathe YS, et al: Quantitative relationships between circulating leukocytes and infection in patients with acute leukemia. Ann Intern Med 64:328-40, 1966 73. Blumenson LE: A comprehensive modeling procedure for the human granulopoietic system: Over-all view and summary of data. Blood 42:303-313, 1973 74. Rubinow SI, Lebowitz JL: A mathematical model of the chemotherapeutic treatment of acute myeloblastic leukemia. Biophys J 16:1257-71, 1976 75. Blumenson LE, Bross ID: Assessment of myelotoxic effects of chemotherapy from early leukopenic response: application of a mathematical model for granulopoiesis. J Surg Oncol 11:171-6, 1979 76. Smeby W, Benestad HB: Simulation of murine granulopoiesis. Blut 41:47-60, 1980 77. Steinbach KH, Raffler H, Pabst G, et al: A mathematical model of canine granulocytopoiesis. J Math Biol 10:1-12, 1980 78. Minami H, Sasaki Y, Saijo N, et al: Indirect-response model for the time course of leukopenia with anticancer drugs. Clin Pharmacol Ther 64:511-21, 1998 79. Sheiner LB, Stanski DR, Vozeh S, et al: Simultaneous modeling of pharmacokinetics and pharmacodynamics: application to d-tubocurarine. Clin Pharmacol Ther 25:358- 71, 1979 80. Dayneka NL, Garg V, Jusko WJ: Comparison of four basic models of indirect pharmacodynamic responses. J Pharmacokinet Biopharm 21: 457-78, 1993

65 81. Gisleskog PO, Hermann D, Hammarlund-Udenaes M, et al: A model for the turnover of dihydrotestosterone in the presence of the irreversible 5 alpha-reductase inhibitors GI198745 and finasteride. Clin Pharmacol Ther 64:636-47, 1998 82. Kahn MG, Fagan LM, Sheiner LB: Combining physiologic models and symbolic methods to interpret time-varying patient data. Methods Inf Med 30:167-78, 1991 83. Sun YN, Jusko WJ: Transit compartments versus gamma distribution function to model signal transduction processes in pharmacodynamics. J Pharm Sci 87:732-7, 1998 84. Norton L, Simon R: Tumor size, sensitivity to therapy, and design of treatment schedules. Cancer Treat Rep 61:1307-17, 1977 85. Jusko WJ: Pharmacodynamics of chemotherapeutic effects: dose-time-response relationships for phase-nonspecific agents. J Pharm Sci 60:892-5, 1971 86. Jusko WJ: A pharmacodynamic model for cell-cycle-specific chemotherapeutic agents. J Pharmacokin Biopharm 1:175-200, 1973 87. Harashima H, Tsuchihashi M, Iida S, et al: Pharmacokinetic/pharmacodynamic modeling of antitumor agents encapsulated into liposomes. Adv Drug Deliv Rev 40:39-61, 1999 88. Gardner SN: Modeling Multi-drug Chemotherapy: Tailoring Treatment to Individuals. J Theor Biol 214:181-207, 2002 89. Levasseur LM, Slocum HK, Rustum YM, et al: Modeling of the time-dependency of in vitro drug cytotoxicity and resistance. Cancer Res 58:5749-61, 1998 90. Adams DJ: In vitro pharmacodynamic assay for cancer drug development: application to crisnatol, a new DNA intercalator. Cancer Res 49:6615-20, 1989 91. Sandström M, Freijs A, Larsson R, et al: Lack of relationship between systemic exposure for the component drug of the fluorouracil, epirubicin, and 4- hydroxycyclophosphamide regimen in breast cancer patients. J Clin Oncol 14:1581-8, 1996 92. Pratt WB, Ruddon RW, Ensminger WD, et al: Antimetabolites, In: The anticancer drugs (ed 2nd). New York, Oxford University Press, 1994, pp 69-107 93. Collins JM, Dedrick RL, King FG, et al: Nonlinear pharmacokinetic models for 5- fluorouracil in man: intravenous and intraperitoneal routes. Clin Pharmacol Ther 28:235-46, 1980 94. Jarugula VR, Lam SS, Boudinot FD: Nonlinear pharmacokinetics of 5-fluorouracil in rats. J Pharm Sci 86:756-8, 1997 95. Milano G, Etienne MC: Dihydropyrimidine dehydrogenase (DPD) and clinical pharmacology of 5-fluorouracil (review). Anticancer Res 14:2295-7, 1994 96. Looney WB, Macleod MS, Hopkins HA: Solid tumour models for the assessment of different treatment modalities: VII: single vs fractionated doses of 5-fluorouracil on two solid tumours and their hosts. Br J Cancer 37:841-8, 1978 97. Robert J: Clinical pharmacokinetics of epirubicin. Clin Pharmacokinet 26:428-38, 1994

66 98. Coukell AJ, Faulds D: Epirubicin. An updated review of its pharmacodynamic and pharmacokinetic properties and therapeutic efficacy in the management of breast cancer. Drugs 53:453-82, 1997 99. Schott B, Robert J: Comparative activity of anthracycline 13-dihydrometabolites against rat glioblastoma cells in culture. Biochem Pharmacol 38:4069-74, 1989 100. Workman P: Infusional anthracyclines: is slower better? If so, why? [editorial; comment]. Ann Oncol 3:591-4, 1992 101. Plosker GL, Faulds D: Epirubicin. A review of its pharmacodynamic and pharmacokinetic properties, and therapeutic use in cancer chemotherapy. Drugs 45:788-856, 1993 102. Boddy AV, Yule SM: Metabolism and pharmacokinetics of oxazaphosphorines. Clin Pharmacokinet 38:291-304, 2000 103. Voelcker G, Wagner T, Wientzek C, et al: Pharmacokinetics of "activated" cyclophosphamide and therapeutic efficacies. Cancer 54:1179-86, 1984 104. Yule SM, Price L, Cole M, et al: Cyclophosphamide metabolism in children following a 1-h and a 24-h infusion. Cancer Chemother Pharmacol 47:222-8, 2001 105. Matsuda A, Azuma A, Nakajima Y, et al: Design of antitumor nucleosides: the synthesis and antitumour activity of 2?-deoxy-(2?-C-substituted) cytidines, in Chung CK BD (ed): Nucleosides and nucleotides as antitumour and antiviral agents. New York, Plenum Press, 1993, pp 1-22 106. Brindley CJ, Morrison R, Gordon RJ, et al: Clinical pharmacokinetics of 2'-deoxy-2'- methylidenecytidine (DMDC), a deoxycytidine analogue antineoplastic agent. Clin Pharmacokinet 38:475-91, 2000 107. van der Gaast A, Woll PJ, Planting A, et al: An adaptive phase I study of oral 2'- deoxy-2'-methylidenecytidine (DMDC). Presented at 10th NCI-EORTC Symposium on New Drugs in Cancer Therapy, Amsterdam, The Netherlands, June 15-19, 1998 108. van der Gaast A, Woll PJ, Planting A, et al: A phase 1 and pharmacokinetic study of oral 2´-deoxy-2´-methylidenecytidine (DMDC). Presented at Proceedings of the American Society of Clinical Oncology, Athens, Greece, May 16-19, 1998 109. Pluzanka A, Rastogi R, Devlin AJ, et al: Phase I and pharmacokinetic (PK) study of an orally active drug, 2'-deoxy-2'-methylidenecytidine (DMDC) in patients with advanced malignancies. Presented at 23rd European Society of Medical Oncology Congress, Athens, Greece, October 6-30, 1998 110. Bruno R, Vivler N, Vergniol JC, et al: A population pharmacokinetic model for docetaxel (Taxotere): model building and validation. J Pharmacokinet Biopharm 24:153-72, 1996 111. Henningsson A, Karlsson MO, Vigano L, et al: Mechanism-based pharmacokinetic model for paclitaxel. J Clin Oncol 19:4065-73, 2001 112. Cortes JE, Pazdur R: Docetaxel. J Clin Oncol 13:2643-55, 1995 113. Eisenhauer EA, ten Bokkel Huinink WW, Swenerton KD, et al: European-Canadian randomized trial of paclitaxel in relapsed ovarian cancer: high-dose versus low-dose and long versus short infusion. J Clin Oncol 12:2654-66, 1994

67 114. 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 115. Clark PI, Slevin ML: The clinical pharmacology of etoposide and teniposide. Clin Pharmacokinet 12:223-52, 1987 116. Hande KR: The importance of drug scheduling in cancer chemotherapy: etoposide as an example. Stem Cells 14:18-24, 1996 117. Moore MJ, Erlichman C: Pharmacology of anticancer drugs, in Tannock IF, Hill RP (eds): The Basic Science of Oncology (ed 3rd). New York, McGraw-Hill, 1998, pp 370-391 118. Kawato Y, Aonuma M, Hirota Y, et al: Intracellular roles of SN-38, a metabolite of the camptothecin derivative CPT-11, in the antitumor effect of CPT-11. Cancer Res 51:4187-91, 1991 119. Xie R, Mathijssen RH, Sparreboom A, et al: Clinical pharmacokinetics of irinotecan and its metabolites in relation with diarrhea. Clin Pharmacol Ther 72:265-75, 2002 120. Jansman FG, Sleijfer DT, Coenen JL, et al: Risk factors determining chemotherapeutic toxicity in patients with advanced colorectal cancer. Drug Saf 23:255-78, 2000 121. Kruczynski A, Etievant C, Perrin D, et al: Characterization of cell death induced by vinflunine, the most recent Vinca alkaloid in clinical development. Br J Cancer 86:143-50, 2002 122. Nguyen L, Retout S, Mentre F, et al: Population pharmacokinetics of vinflunine from phase I data and evaluation of population sampling designs for further clinical development. Presented at 11th Population Approach Group Europe Meeting, Paris, France, 2002. http://userpage.fu-berlin.de/~page/ 123. Focan C, van Heugen JC, Kreutz F, et al: Vinflunine metabolism and disposition in cancer patients. Presented at 38th Annual Meeting of the American Society of Clinical Oncology, Orange County CC, Orlando, 2002 124. Armand J-P, Fumoleau P, Marty M, et al: Pharmacokinetics of vinflunine, a Novel Vinca alkaloid, during the phase I Dose escalation Study (D1 Q3 weeks). Presented at American Association for Cancer Research, New Orleans, LA, March 24-28, 2001 125. Delord J-P, Stupp R, Pinel M-C, et al: A phase I study of vinflunine given as 10 minute intravenous (IV) infusion on a weekly schedule in patients with advanced solid tumours. Presented at American Society of Clinical Oncology, San Francisco, CA, May 12-15, 2001 126. Johnson P, Judson I, Ottensmeier C, et al: A phase I and pharmacokinetic study of vinflunine given on days 1 and 8 every 3 weeks. Br J Cancer 83:P241, 2001 (suppl 1) 127. Schou C, Ottosen ER, Petersen HJ, et al: Novel cyanoguanidines with potent oral antitumour activity. Bioorganic & Medicinal Chemistry Letters 7:3095-3100, 1997 128. Vig Hjarnaa PJ, Jonsson E, Latini S, et al: CHS 828, a novel pyridyl cyanoguanidine with potent antitumor activity in vitro and in vivo. Cancer Res 59:5751-7, 1999

68 129. Åleskog A, Bashir-Hassan S, Hovstadius P, et al: Activity of CHS 828 in primary cultures of human hematological and solid tumors in vitro. Anticancer Drugs 12:821- 7, 2001 130. Hassan SB, de la Torre M, Nygren P, et al: A hollow fiber model for in vitro studies of cytotoxic compounds: activity of the cyanoguanidine CHS 828. Anticancer Drugs 12:33-42., 2001 131. Jonsson E, Friberg LE, Karlsson MO, et al: In vivo activity of CHS 828 on hollow- fibre cultures of primary human tumour cells from patients. Cancer Lett 162:193-200., 2001 132. Hassan SB, Jonsson E, Larsson R, et al: Model for Time Dependency of Cytotoxic Effect of CHS 828 in Vitro Suggests Two Different Mechanisms of Action. J Pharmacol Exp Ther 299:1140-7, 2001 133. Hovstadius P, Larsson R, Jonsson E, et al: A Phase I Study of CHS 828 in Patients with Solid Tumor Malignancy. Clin Cancer Res 8:2843-50, 2002 134. Ravaud A, Morant R, Skov T, et al: An EORTC/ECSG Phase I Study of CHS 828, a cyanoguanidine in patients with refractory solid tumors. Presented at Proc Am Assoc Cancer Rec, New Orleans, LA, 2001 135. Koks CH, van der Kam HJ, Brouwers JR: A simple high pressure liquid chromatographic method for the determination of fluorouracil to monitor patients on regional infusion for hepatic metastases. Pharm Weekbl Sci 12:154-7, 1990 136. Launay MC, Milano G, Iliadis A, et al: A limited sampling procedure for estimating adriamycin pharmacokinetics in cancer patients. Br J Cancer 60:89-92, 1989 137. Robert J: Extraction of anthracyclines from biological fluids for HPLC evaluation. J Liquid Chromatogr 3:1561-1572, 1980 138. Israel M, Pegg WJ, Wilkinson PM, et al: Liquid chromatographic analysis of adriamycin and metabolites in biological fluids. J Liquid Chromatogr 1:795-809, 1978 139. Johansson M, Bielenstein M: Determination of 4-hydroxycyclophosphamide in plasma, as 2,4-dinitrophenylhydrazone derivative of aldophosphamide, by liquid chromatography. J Chromatogr B Biomed Appl 660:111-20, 1994 140. Cailleau R, Young R, Olive M, et al: Breast tumor cell lines from pleural effusions. J Natl Cancer Inst 53:661-74, 1974 141. Soule HD, Vazguez J, Long A, et al: A human cell line from a pleural effusion derived from a breast carcinoma. J Natl Cancer Inst 51:1409-16, 1973 142. Foley GE, Lazarus H, Farber S, et al: Continuous culture of human lymphoblasts from peripheral blood of a child with acute leukemia. Cancer 18:522-529, 1965 143. 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 144. 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

69 145. Henningsson A, Sparreboom A, Sandström M, et al: Population pharmacokinetic modelling of unbound and total plasma concentrations of paclitaxel in cancer patients. 2003 (In Press) 146. 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 147. 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 148. Karlsson MO, Jonsson EN, Wiltse CG, et al: Assumption testing in population pharmacokinetic models: illustrated with an analysis of moxonidine data from congestive heart failure patients. J Pharmacokinet Biopharm 26:207-46, 1998 149. Jonsson EN, Karlsson MO: Xpose-an S-PLUS based population pharmacokinetic/pharmacodynamic model building aid for NONMEM. Comput Methods Programs Biomed 58:51-64, 1999 150. Wählby U, Jonsson EN, Karlsson MO: Assessment of actual significance levels for covariate effects in NONMEM. J Pharmacokinet Pharmacodyn 28:231-52, 2001 151. Gelman A, Carlin JB, Stern HS, et al: Bayesian data analysis. London, Chapman and Hall, 1995 152. Dilloway MR, Lant AF: Effect of H2-receptor antagonists on the pharmacokinetics of 5-fluorouracil in the rat and monkey. Biopharm Drug Dispos 12:17-28, 1991 153. Harris BE, Song RL, Soong SJ, et al: Circadian variation of 5-fluorouracil catabolism in isolated perfused rat liver. Cancer Res 49:6610-4, 1989 154. Hong PS, Srigritsanapol A, Chan KK: Pharmacokinetics of 4- hydroxycyclophosphamide and metabolites in the rat. Drug Metab Dispos 19:1-7, 1991 155. Bungay PM, Dedrick RL, Matthews HB: Enteric transport of chlordecone (Kepone) in the rat. J Pharmacokinet Biopharm 9:309-41, 1981 156. Boddy AV, Cole M, Pearson AD, et al: The kinetics of the auto-induction of ifosfamide metabolism during continuous infusion. Cancer Chemother Pharmacol 36:53-60, 1995 157. Hassan M, Svensson US, Ljungman P, et al: A mechanism-based pharmacokinetic- enzyme model for cyclophosphamide autoinduction in breast cancer patients. Br J Clin Pharmacol 48:669-77, 1999 158. van Groeningen CJ, Pinedo HM, Heddes J, et al: Pharmacokinetics of 5-fluorouracil assessed with a sensitive mass spectrometric method in patients on a dose escalation schedule. Cancer Res 48:6956-61, 1988 159. Testa NG, Dexter TM: The regulation of haemopoietic cell production, in Hoffbrand AV, Lewis SM, Tuddenham EGD (eds): Postgraduate haematology (ed 4th). Oxford, UK, Butterworth Heinemann, 1999, pp 1-12 160. Fliedner TM, Cronkite EP, Killmann SÅ, et al: Granulocytopoiesis. II. Emergence and pattern of labeling of neutrophilis granulocytes i humans. Blood 24:683-700, 1964

70 161. Wang B, Ludden TM, Cheung EN, et al: Population pharmacokinetic- pharmacodynamic modeling of filgastrim (r-metHuG-CSF) in healthy volunteers. J Pharmacokinet Pharmacodyn 28:321-342, 2001 162. Moore GE, Bross ID, Ausmans R, et al: Effects of 5-fluorouracil (NSC-19893) in 389 patients with cancer. Eastern Clinical Drug Evaluation Program. Cancer Chemother Rep 52:641-53, 1968 163. Friberg LE, Karlsson MO: A semi-physiological model for the pharmacodynamic interaction on leukocytes after 5-fluorouracil and epirubicin administration. Presented at 10th Population Approach Group Europe Meeting, Basel, Switzerland. http://userpage.fu-berlin.de/~page/, June 7-8, 2001 164. Sandström M, Lindman H, Bergh J, et al: PK/PD of the epirubicin-docetaxel therapy in breast cancer patients. Presented at 10th Population Approach Group Europe Meeting, Basel, Switzerland. http://userpage.fu-berlin.de/~page/, June 7-8, 2001 165. van Kesteren C: Population pharmacokinetics and pharmacodynamics of two novel anticancer agents: Ecteinascidin-743 and E7070. Utrecht, Universiteit Utrecht, 2002 166. Sandström M: Pharmacokinetics and pharmacodynamics of anti-cancer regimens. Emphasis on busulphan and the combination therapies epirubicin-docetaxel and fluorouracil-epirubicin-cyclophosphamide, Div. of Pharmacokinetics and Drug Therapy. Uppsala, Uppsala University, 2002 167. 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 168. van der Wall E, Rutgers EJ, Holtkamp MJ, et al: Efficacy of up-front 5-fluorouracil- epidoxorubicin-cyclophosphamide (FEC) chemotherapy with an increased dose of epidoxorubicin in high-risk breast cancer patients. Br J Cancer 73:1080-5, 1996 169. Phillips RM, Pearce J, Loadman PM, et al: Angiogenesis in the hollow fiber tumor model influences drug delivery to tumor cells: implications for anticancer drug screening programs. Cancer Res 58:5263-6, 1998

71