Pharmacokinetic/Pharmacodynamic Modelling-Based Translational Approach Applied to the Anticancer Drug Gemcitabine in Advanced Pancreatic and Ovarian Cancer”

Departamento de Farmacia y Tecnología Farmacéutica Facultad de Farmacia y Nutrición UNIVERSIDAD DE NAVARRA

“Pharmacokinetic/Pharmacodynamic Modelling-Based Translational Approach applied to the Anticancer Drug Gemcitabine in Advanced Pancreatic and Ovarian Cancer”

María García-Cremades Mira Pamplona, 2017

Departamento de Farmacia y Tecnología Farmacéutica Facultad de Farmacia y Nutrición UNIVERSIDAD DE NAVARRA

TESIS DOCTORAL

“Pharmacokinetic/Pharmacodynamic Modelling-Based Translational Approach applied to the Anticancer Drug Gemcitabine in Advanced Pancreatic and Ovarian Cancer”

Trabajo presentado por María García-Cremades Mira para obtener el Grado de Doctor

Fdo. María García-Cremades Mira Pamplona, 2017

UNIVERSIDAD DE NAVARRA FACULTAD DE FARMACIA Y NUTRICIÓN Departamento de Farmacia y Tecnología Farmacéutica

D. JOSÉ IGNACIO FERNÁNDEZ DE TROCÓNIZ FERNÁNDEZ, Doctor en Farmacia y Catedrático del Departamento de Farmacia y Tecnología Farmacéutica. Certifica:

Que el presente trabajo, titulado “Pharmacokinetic/pharmacodynamic modelling-based translational approach applied to the anticancer drug gemcitabine in advanced pancreatic and ovarian cancer”, presentado por DÑA. MARÍA GARCÍA-CREMADES MIRA para optar al grado de Doctor en Farmacia, ha sido realizado bajo su dirección en los Departamentos de Farmacia y Tecnología Farmacéutica. Considerando finalizado el trabajo autorizan su presentación a fin de que pueda ser juzgado y calificado por el Tribunal correspondiente.

Y para que así conste, firma la presente:

Fdo.: Dr. José Ignacio F Trocóniz

Pamplona, 2017

“El mundo es de los que hacen de cada momento una gran aventura”

AGRADECIMIENTOS

Quisiera comenzar expresando mi agradecimiento a la Universidad de Navarra y al Departamentos de Farmacia y Tecnología Farmacéutica por haberme posibilitado la realización de esta tesis doctoral.

Al Dr. Iñaki Trocóniz me gustaría agradecerle la confianza y motivación que desde el primer día me ha dado. He aprendido muchísimo… y lo mejor de todo es que lo he hecho siempre disfrutando. Gracias, también, a Mª Jesús por formar parte de este aprendizaje.

Gracias a todas y cada una de las personas que han sido parte del grupo PSP estos años… muy especialmente a Víctor, Núria y JD, que en verdad nunca se fueron. Por supuesto a los doctores Zinnia y Eduardo, por sus consejos y toda su ayuda en los momentos del caos.

A las PKPDitas… Itziar, Leire, Vío, Belén. Gracias por toda la magia, espionaje y piratería. He aprendido mucho de cada una de vosotras. Ha sido una suerte haberos encontrado entre tantos modelos.

Thanks to Dr. Paolo Magni and his Lab for hosting me in Pavia! It was a great academic experience. Special thanks to Nicola for introducing me into Matlab and PBPK world, and for all the nice moments.

Special thanks to Celine, for being permanently caring and always willing to help.

A todos los compañeros y profesores del departamento. En especial a Ana y a Yolanda, por las múltiples risas y extraescolares, pero, sobre todo, por ser amigas además de compañeras. Gracias Pamplona por regalarme a uno de los mejores amigos, gracias Sergio (MTHLC). A Hilda, mi compañera, amiga y hermana… GRACIAS por tanto.

Muchas gracias a toda mi familia por el apoyo incondicional. Gracias mamá, papá, Gabri y Álex.

Sobre todo, gracias Raúl por haber estado a mi lado en todo este proceso. Lo haces todo más fácil, incluso los momentos más locos.

Han sido 4 años (¡5 años!) de aprender, llorar, reír… De TCs y viajes inesperados. De reuniones en inglés, pintxos y karaoke. De Bilman, blablas y hasta taxis. De crecer. GRACIAS por esta gran aventura.

TABLE OF CONTENTS

ABBREVIATIONS ...... 1

PREFACE ...... 3

INTRODUCTION ...... 5 1. PHARMACOMETRICS ...... 7 1.1 Pharmacokinetics and Pharmacodynamics models ...... 8 1.2 Physiologically-based models ...... 13 1.3 Population approach ...... 16 1.4 Model evaluation ...... 18 2. ROLE OF PHARMACOMETRICS IN DRUG DEVELOPMENT PROCESS ...... 22 3. MID3 APPLIED TO ONCOLOGY ...... 24 3.1 Preclinical stages ...... 24 3.2 Clinical stages ...... 27 4. GEMCITABINE ...... 31 4.1 PKPD modelling applied to gemcitabine ...... 33 5. REFERENCES ...... 35

AIM ...... 47

CHAPTER 1 ...... 51 Characterizing gemcitabine effects administered as single agent or combined with carboplatin in mice pancreatic and ovarian cancer xenografts: a semimechanistic pharmacokinetic/pharmacodynamics tumour growth-response model. ABSTRACT ...... 55 1. INTRODUCTION ...... 57 2. MATERIAL AND METHODS ...... 59 2.1 Experimental Data and Studies Design ...... 59 2.2 Data Analysis ...... 60 2.3 Model Building ...... 61 2.4 Pharmacokinetics ...... 61 2.5 Disease Progression Model ...... 62 2.6 Drug Effect Model ...... 63 2.7 Model Selection ...... 64 2.8 Model Evaluation ...... 65

2.9 Model Exploration ...... 65 3. RESULTS ...... 66 3.1 General Description of the data ...... 66 3.2 Modelling Tumour Profiles ...... 66 3.3 Model Exploration ...... 72 4. DISCUSSION ...... 74 5. REFERENCES ...... 79 6. SUPPLEMENTARY MATERIAL...... 85

CHAPTER 2 ...... 87 Predicting tumour growth and its impact on survival in gemcitabine-treated patients with advanced pancreatic cancer. ABSTRACT ...... 91 TRANSLATIONAL RELEVANCE ...... 93 1. INTRODUCTION ...... 95 2. PATIENTS AND METHODS ...... 97 2.1 Data and Studies Design ...... 97 2.2 Data analysis ...... 97 2.3 Gemcitabine Pharmacokinetics ...... 98 2.4 Tumour growth inhibition model ...... 98 2.5 Overall Survival ...... 99 2.6 Model selection ...... 99 2.7 Covariate analysis and selection ...... 100 2.8 Model evaluation ...... 100 2.9 External model validation ...... 100 2.10 Model exploration ...... 101 3. RESULTS ...... 102 3.1 General description of the datA...... 102 3.2 Joint Tumour growth inhibition and Survival model ...... 103 4. DISCUSSION ...... 109 5. REFERENCES ...... 113 6. SUPPLEMENTARY MATERIAL ...... 117

CHAPTER 3 ...... 121 Translational framework predicting tumour response and survival in gemcitabine-treated patients with advance pancreatic and ovarian cancer from xenograft studies. ABSTRACT ...... 125

TRANSLATIONAL RELEVANCE ...... 127 1. INTRODUCTION ...... 129 2. METHODS...... 131 2.1 Models and data ...... 131 2.2 Tumor volume prediction in humans ...... 132 2.3 Survival simulations ...... 133 3. RESULTS ...... 135 3.1 Pancreatic cancer ...... 135 3.2 Ovarian cancer...... 137 4. DISCUSSION ...... 140 5. REFERENCES ...... 145 6. SUPPLEMENTARY MATERIAL ...... 149

CHAPTER 4 ...... 151 Mechanistic multi-scale systems pharmacokinetics model applied to the anticancer drug gemcitabine in pancreatic cancer. ABSTRACT ...... 155 1. INTRODUCTION ...... 157 2. METHODS...... 159 2.1 System PK model ...... 159 2.2 PBPK model ...... 160 2.3 Clinical simulations ...... 163 3. RESULTS ...... 165 3.1 System PK model ...... 165 3.2 PBPK model ...... 167 3.3 Clinical simulations ...... 168 4. DISCUSSION ...... 171 5. REFERENCES ...... 175 6. SUPPLEMENTARY MATERIAL ......

GENERAL DISCUSSION...... 185 REFERENCES ...... 193

CONCLUSIONS/CONCLUSIONES ...... 197

ABBREVIATIONS

Abbreviature Definition -2LL -2xlog likelihood ADME Absorption, distribution, metabolism and excretion AIC Aikake information criteria AUC Area under the curve BW Body weight

Ce Concentration biophase CL Total clearance

CLi Individual clearance Cmax Maximum drug concentration Cp Plasmatic concentration CR Complete response CT Computerized tomography CV Coefficient of variation CWRES Conditional weighted residuals DV Observations ECOG Eastern Cooperative Oncology Group EFPIA European Federation of Pharmaceutical Industries and Associations EMA European Medicines Agency Emax Maximum effect F Bioavailability FDA Food and Drug Administration GOF Goodness-of-fit IIV Inter-individual variability IPRED Individual prediction IVIVE In vitro-in vivo extrapolation

Ka Absorption Kp Partition coefficient LRT Log-likelihood ratio test MID3 Model Informed Drug Discovery and Development MRI Magnetic resonance imaging

Abbreviature Definition NLSB Number of tumour lesions at baseline NODB Number of organs damaged with tumour lesions at baseline NONMEN NON-linear Mixed Effect Modelling NSCLC Nonsmall-cell lung cancer OFV Objective function value OS Overall survival PBPK Physiologically based pharmacokinetic PC-VPC Prediction-corrected visual predictive checks PD Pharmacodynamics PD´ Progressive disease PFS Progression free survival PK Pharmacokinetic PKPD Pharmacokinetic/pharmacodynamics PR Partial response PRED Population prediction R-Ce Drug-Receptor complex RECIST Response Evaluation Criteria in Solid Tumors RES Residuals SB Systems biology SD Stable disease SLD Sum of the longest diameters T1/2 Half-life TGI Tumour growth inhibition Tmax Time at which Cmax is achieved TR Tumour regression TRS Change of tumour regression TS Tumour size TTE Time-to-event TV Tumour volumen V Apparent volume distribution WRES Weighted residuals VPC Visual predictive checks

PREFACE

One of the major challenges facing drug development is the high attrition rate, with up to 95% associated to oncology drugs tested in phase I trials failing to progress to marketing authorisation. The FDA’s Critical Path Initiative promotes Model Informed Drug Discovery and Development (MID3) to optimize drug development (i.e., reducing attrition rates). Despite the adoption of MID3 by the majority of pharmaceutical companies, there are very few publicly available examples which integrate quantitative information across the phases of drug development.

Pharmacometrics is an emergent discipline focusing on the science-based quantitatively description of drug response. The integration of pharmacokinetic/pharmacodynamic (PKPD) modelling approaches among the different phases of drug development, integrating in vitro, preclinical and clinical data, promises to optimize translational research, by making it more efficient and reducing the current high attrition rates.

This thesis performs a retrospective evaluation of data generated for the cytotoxic/cytostatic antimetabolite drug gemcitabine along the different phases of drug development. Using these data, and through data analysis and simulations-based translational exercise, a model-based framework in oncology, relating in vitro and in vivo pre-clinical pharmacokinetic (PK) and tumour size (TS) information, with response outcome obtained in clinical trials, has been developed.

First, an overview of the current status of MID3, with special focus on the oncology area, along with a summary of the most relevant pharmacometrics concepts will be provided in the Introduction section. Then, in the following chapters, different PKPD modelling approaches along the different drug development phases are summarized.

Chapter 1 presents a preclinical semi-mechanistic model for describing tumour shrinkage effects of gemcitabine administered as a single agent or in combination with carboplatin in mice pancreatic or ovarian cancer xenografts.

Chapter 2 describes a PKPD model, linking exposure to gemcitabine active metabolite, TS and overall survival (OS) in advanced pancreatic cancer where gemcitabine was administered to patients from clinical trials phases II and III.

Chapter 3 presents a translational approach linking the preclinical model developed in Chapter1 with the PKPD models for gemcitabine developed for advanced pancreatic cancer patients, given as single agent (Chapter2), and for ovarian cancer patients, given in combination with carboplatin and previously developed and published by Eli Lilly.

Chapter 4 describes a mechanistic network of gemcitabine metabolic pathway, (built from in vitro data) that, coupled to a physiologically-based PK (PBPK) model, simulates different concentration profiles of gemcitabine active metabolite depending on different degrees of genetic polymorphisms affecting the enzymes´ expression. A framework linking the metabolisc network, the PBPK model and the clinical tumour-response PKPD model from Chapter 2 is also presented, enabling the simulation of different clinical outcomes based on different degrees of genetic polymorphisms.

Finally, the last section, General Discussion, integrates and highlights the most relevant aspects of the four chapters, to end with the Conclusions, summarising the main findings of this thesis.

INTRODUCTION

Introduction

Current drug development is associated to high attrition rates regardless the therapeutic area. The Food and Drug Administration (FDA) has expressed its concern about the rising in the development costs, with 40–50% of programs being discontinued even in clinical Phase III1 (1). In fact the FDA’s Critical Path Initiative2 recommends several changes and approaches aimed to optimize and accelerate the arrival of new therapeutic strategies to the patient. In one of those recommendations the use of model-based approaches is considered essential in the optimization of any drug development strategies, including decision making. The importance that different regulatory agencies have given to the model based approach [currently Model Informed drug Discovery and Development (MI3D)] has been summarized in the corresponding guides (2, 3), and has contributed significantly to promote MI3D within pharmaceutical companies.

1. PHARMACOMETRICS

Pharmacometrics is defined as “the science of developing and applying mathematical and statistical methods to: (i) characterize, understand and predict a drug´s pharmacokinetic and pharmacodynamics behaviour, (ii) quantify uncertainty of information about the behaviour, and (iii) rationale data-driven decision making in drug development process and pharmacotherapy.” (2) This concept has greatly evolved during these last four decades, but it is basically focused on the development of pharmacokinetic (PK), pharmacodynamics (PD) and disease progression models, integrating principles from the field of pharmacology and statistics for understanding drug effects over time, supporting drug research and, also, personalized medicine.

It should be highlighted that during these last decades, the pharmacologic response analysis has been mainly limited to the development of PKPD models, which in fact have resulted of an enormous utility on basic research, pharmaceutical industry, clinical practice and on regulatory field (3). Nevertheless, new challenges encouraged to the improvement of translational medicine, make it necessary the integration of traditional PKPD analysis together with computational analysis, systems biology (SB) and physiological-based PK (PBPK) models. Among the different phases of drug development, PKPD, PBPK and SB may play different roles. Coupling these disciplines across the development cycle of a compound, and integrating in vitro,

1 http://www-07.ibm.com/services/pdf/pharma_es.pdf 2 https://www.fda.gov/downloads/ScienceResearch/SpecialTopics/CriticalPathInitiative/CriticalPathOp portunitiesReports/ucm113411.pdf

Introduction pre-clinical and clinical data may assist on the development of multi-scale models that can optimize translational research, thus reducing the current high attrition rates.

Before illustrating the role of pharmacometrics in the drug development process, the main concepts regarding this emergent discipline, such as PKPD models, population approach and physiologically-based models, should be addressed.

PHARMACOKINETICS AND PHARMACODYNAMICS MODELS

1.1.1. Pharmacokinetics

In its initial stages, PK arose as a science aiming to characterize drug (and its metabolites) disposition in the organism. It describes the concentration-time profiles of the drug in the body, characterizing the absorption, distribution, metabolism and elimination (ADME) processes (4). These concentration-time profiles are explained by a set of parameters (which represent each of the above mentioned processes) that can be used to compare, evaluate and predict drug behaviour. In a simpler way, PK characterizes “what the body does to the drug”(5).

Traditionally, there have been two main approaches to analyse PK data: (i) Non- compartmental analysis, where descriptive statistics such as area under the drug concentration versus time curve (AUC), maximum drug concentration (Cmax) or time at which Cmax is achieved (Tmax) are summarized directly from observed individual profiles. (ii) Model-based compartmental analysis, in which a set of parameters, of unknown magnitude are estimated from the data and used to describe the PK profiles.

This model based-analysis quantifies primary (physiological related) PK parameters such as the first order rate constant of absorption (Ka), bioavailability (F), apparent volume of distribution (V) and total clearance (CL).

A common compartmental PK model typically has a central compartment, representing those organs to which the drug rapidly distributes, that can be linked to one (or more) peripheral compartment, representing organs with a slower drug distribution, via first rate order constants (k12, k21). One characteristic of the compartmental models is that the compartments do not represent any real tissue, organ or fluid of the body. A schematic representation of a two compartment model in case of extravascular administration is represented in figure 1.

Introduction

Depot

Ka k12 Central Per V k21

Figure 1. Schematic (left) and mathematical (right) representation of a two compartmental PK model in case of extravascular administration.

1.1.2. Pharmacodynamics

PD analysis aims to describe drug effect intensity as a function of its concentrations, once the drug reaches its site of action (biophase). It can be defined as the characterization of the binding of the active compound to its target(s) and the elicited response. Following the simpler description of PK analysis mentioned above, PD could be described as “what the drug does to the body”(5).

Similar to the case of the compartmental models in PK, the sigmoidal EMAX model (figure 2) and its variants are the most used models in PD (8,9).

Figure 2. Sigmoidal EMAX model representation. EMAX: Maximum response of a drug. C50: Concentration necessary to produce 50% of EMAX. n: Sigmoidicity factor or steepness of the curve; If n=1; hyperbolic curve; n>1 steeper curve; n<1 smoother curve, C: Concentration. E: Response.

Introduction

This PD model is based on the receptor theory and has several assumptions; (i) more than one molecule of drug can bind to the receptor, (ii) there is a single type of drug-receptor complex, (iii) the effect is proportional to drug-receptor complex, (iv) it presents equilibrium conditions with fast association and dissociation processes. Further examples of PD models are displayed in figure 3.

EMAX model Exponential model

Linear model EMAX + linear model Response

Concentration Figure 3. Graphical representation of additional PD models.

In figures 2 and 3 time is not considered, as the assumption of equilibrium between concentrations in biophase and time invariant PD are implicit. Therefore, to describe the time course of effect, both PK and PD have to be integrated into a PKPD model as represented by figure 4.

Figure 4. Representation of a PKPD model. Adapted from reference(4)

Introduction

Regarding the pharmacological response, with respect to the time course of the drug concentrations in plasma, we can difference between direct and non-direct responses. In figure 5 are represented both time profiles of PK and response, and response vs concentration relationship for the case of a direct (left) and non-direct (right) response; for the later the response vs concentration plot shows a phenomena call the hysteresis loop.

C C P P E E

Time Time E E

CP CP

Figure 5. Schematic representation of direct (panels on the left) and non-direct (panels on the right) response. CP: Plasmatic concentration. E: Response. (6)

Figure 3 and 4 represent the case of a direct relationship between effect and concentration (usually measured in plasma). The presence of a non-direct response vs concentration relationship may obey to one or more of the following mechanisms: (i) slow distribution to the biophase, (ii) indirect mechanism of action, (iii) slow receptor deactivation or (iv) signal transduction/maturation (7–9).

To visualize better the concept of non-direct response we might consider the general representation of the in vivo time course of drug action shown in figure 6, where disposition in plasma, distribution to the effect site, target engagement, signal propagation and feed-back mechanisms are included.

In general it is hard to identify all the processes (i.e., estimate all the associated parameters) from in vivo studies, and therefore the scheme shown in figure 6 collapses into simplified representations depending on the rate limiting step. However it is possible to build

Introduction mechanistic models integrating information gathered in vitro and literature data following the bottom approach of systems pharmacology.

It should not be forgotten that the biological system which represents the disease (i.e., tumour, cognitive capacity, motor capacity, etc) has its own dynamics which can be affected by the treatment and can modify either drug PK and/or PD. The dynamics of the pathology is called disease progression and, as in the case of PK and PD, several models have been proposed to account for the longitudinal changes in absence of treatment (10).

Feedback Tolerance Dose +/- kin

Biophase k on Biosignal Response CP Ce+ R Ce-R k k k k TR TR TR off kTR Transduction k e0 +/- kout

Circadian rhythms System Disease progression

Figure 6. Representation of the main processes responsible of the in vivo time course of drug response.

One concept of extreme importance in MI3D is the system related vs drug specific parameters (13).

This distinction supposes an important advantage in terms of translational research, due to the fact that (i) drug specific parameters are supposed to be similar between species and can be first derived from in vitro experiments, and (ii) system related parameters and system models characterized during a particular treatment can be used to understand the response dynamics for another drug incorporating its own PK and PD properties.

Until this point, the introduction of the pharmacometrics components has been focused on empirical and semi-mechanistic PKPD modelling. These models are commonly built by compartments or “building blocks” that, in the case of PK analysis, relate the plasmatic concentration of a given drug and time. However, in some cases a more granular prediction of

Introduction drug exposure in tissue is required, being physiologically-based pharmacokinetic (PBPK) models needed.

PHYSIOLOGICALLY-BASED MODELS

PBPK models aims to quantitatively describe processes in the mammalian body and its organs (11). PBPK comprises mechanistic models of principal ADME (absorption, distribution, metabolism and excretion) processes that are integrated into a physiologically-based whole- body compartmental model. They are built using a similar framework to empirical PK models, however, they consist of a larger number of compartments, representing the different real organs or tissues in the body and are parameterized using the known physiology (12). Like the compartmental PK models, they provide estimates of common PK parameters (CL, V, t1/2) and can predict plasma and tissue concentration-time profiles after a drug administration.

PBPK models consist of three major elements: (i) the model structure, representing the biological system, (ii) the organism parameters and (iii) the drug parameters (11).

Regarding the model structure, as it has been previously mentioned, these models are made up of compartments corresponding to the different organs of the body, linked by the circulating blood system (blood flow), that transports the drug. These set of compartments represent the biological system, for example, the human body, which can be built as detailed as necessary for the purpose of the model. Mass balance equations are used to characterize the concentration profiles in tissues and organs and the tissues can be either eliminating or non- eliminating.

Figure 7 shows the representation of a PBPK model, including the standard equations for describing non-eliminating and eliminating tissues and including also the structure of the gut compartment, important for oral administration, where the absorption process needs to be taken into account.

Introduction

Figure 7. PBPK model estructure. Adapted from reference (13). t represents the rest of the organs that composed the PBPK model. T= tissue; V=volume (L); C= concentration (mg/mL); A=Arterial; v= venous;

CL= intrinsic clearance of the compound; Q= blood flow (L/h); B:P=blood/plasma ratio; u= unbound; Kp= tissue/plasma partition coefficient.

Each tissue can be described as either perfusion or permeability rate limited. Perfusion rate limited models are used for small lipophilic molecules. In this case it is assumed that total drug in the tissue is in equilibrium with total drug in the circulation at steady state. Time to reach steady state depends on blood flow rate, tissue volume and the tissue partition coefficient for the particular tissue, parameters that are described below in this section.

Permeability rate limited models are applied for the characterization of larger polar molecules. The tissue is divided in two compartments representing the intracellular and the extracellular space which are divided by a diffusional barrier. In this case, time to reach the steady state depends on the permeability rate constant.

Regarding the parameters of the PBPK model, they can also be classified in organism or drug-specific parameters. Organism parameters account for the physiological parameters that describe the system in a quantitative manner. They are anatomic characteristics that depend on the species and that are independent from the considered drug. They are usually known and available in the literature for several species. The main organism parameters are blood fluxes in each tissue and the volume of each tissue. Tissue composition, surface areas or protein abundances, among others are also examples of these type of parameters.

Introduction

On the other hand, drug parameters are substance-specific parameters that describe the interaction between the drug and the system. There are two main groups of drug parameters, (i) the ones accounting for the physicochemical properties, such as permeability, lipophilicity, solubility, molecular weight or pKa, and (ii) the drug biological properties, such as fraction of drug unbound, intrinsic clearance of the compound and tissue-plasma partition coefficient (Kp). This Kp characterize the distribution of the drug into different tissues in the body and can be derived experimentally from steady-state studies or from the composition characteristics of each tissue and from the affinity of the drug with ethanol/oil.

To the best of our knowledge, the main advantage of PBPK modelling is that it provides a quantitative mechanistic framework by which scaled drug-specific parameters (using in vitro-in vivo extrapolation (IVIVE) techniques) can be used to predict the plasma and tissue concentration–time profiles of new drug. Nevertheless it has many other application such as: (i) pediatric extrapolations (14), extrapolation to disease populations(15), drug-drug interaction (16), scaling different treatment scenarios (17), test the behavior of different drug formulations (18), and some advanced applications such as cross-species extrapolation (19), multi-scale modelling (20) or statistical (Bayesian) modelling (21).

Future perspectives in PBPK modelling include combining these models with fully mechanistic PD models and variability in pharmacological response (including receptor genotype). In addition, it should be stated that systems pharmacology is likely to be considered as the next frontier of PKPD, developing fully mechanistic models, leading from dose to exposure ( PK or PBPK models) to response (PD models).

Nowadays, there are several softwares available for developing and applying PBPK models that assist on dealing with the high number of compartments and parameters. To the best of our knowledge, the most popular ones are Symcyp (Certara)3 and PK-Sim (Bayer)4. The PBPK model developed in chapter 4 is developed with Matlab5.

3https://www.certara.com/software/physiologically-based-pharmacokinetic-modeling-and-simulation/ 4 http://www.systems-biology.com/products/pk-sim.html 5 https://es.mathworks.com/products/matlab.html

Introduction

POPULATION APPROACH

The use of population approach to estimate the parameters of PKPD models supposes one of the critical parts in modelling analysis, given that it acknowledges the existence of variability in the observed concentrations and responses across individuals.

Traditionally, population analysis was performed by the two stage approach. By this method, individuals´ model parameters are first computed from individual model fits and then statistical summaries of population parameters (mean and variance) are computed. Its main limitation is that this approach required regular sampling per patient. Currently, nevertheless, the most widely method used to perform population modelling analysis is the so called “mixed- effects”.

Mixed effects population analysis integrates the PK and PD information of a whole population, taking into account the different individuals that comprise it and analysing all the data simultaneously. This approach has the advantage that: (i) sparse or rich data can be analysed, (ii) data do not have to be balanced, (iii) samples do not need to be taken at the same time for all subjects and (iv) rich and sparse data can be analysed simultaneously. Mixed effects population analysis aim is to describe the median tendency of the population (fixed effects) together with its associated variability (random effects), responsible of the different profiles between individuals (22) (figure 8).

PKPD model Covariate model CL KA CL STRUCTURAL PK V CRCL E C PART PD EMAX 50

C50 C Age

Interindividual variability

STATISTICAL CL PART Interocassion Residual variability variability C . . . . CL t Figure 8.Schematic representation of the components of population approach analysis.

Introduction

Fixed effects account for the structural part of the model and represent the typical parameters responsible of the description of the different process under study. These parameters might be influenced by several factors such as weight, sex or disease status among others that might explain some differences between typical profiles of the individuals. These factors are known as covariates and are also part of the fixed effects components of the model.

On the other hand, random effects refer to the statistical part of the model and describe the variability of the data that is not described by the structural model. Likewise, random effects account for (i) the inter-individual variability (IIV), (ii) the inter-occasion (intraindividual) variability (IOV) on each parameter and (ii) the residual error.

The IIV characterizes the dispersion of the individual model parameters around the typical value. This dispersion or “discrepancies” between the individual parameters and the population (typical) parameter () are represented by η. On the other hand, residual variability (ε) quantifies the deviations of the model predictions from the observations.

Using the PK clearance (CL) parameter as an example, the following expressions allow for a more clear explanation these different components.

, = ×

, = (, , ) + ,

Where CLi represents the CL corresponding to the ith individual and θCL corresponds to the typical population estimate (equal for every subject in the study population), the term ηCL,i represents the discrepancy between the CL and the individual CLi. For n subjects in the study, the set of ηCL,1,...n, values form a variable normally distributed around 0 with variance ω2CL. This variance characterizes the magnitude of the IIV associated to the CL parameter.

On the other hand, Yi,j represents the observation of the ith individual “i” at time “j”. This prediction is a function of the dose (D), the time (t), and the set of parameters (ф) plus the difference between the value observed and predicted by the model, represented by εi,j. The set of all the differences between observations and predictions also forms a random variable, normally distributed around the value 0 and with a variance σ2.

Figure 9 displays graphically, for one individual, the different variabilities, the observations and both the individual and the typical predictions.

Introduction

Observations Residual variability Individual prediction

Typical prediction Response

Interindividual variability

Time Figure 9. Main elements, displayed graphically, of the population approach.

Recapping, the objective of the population PKPD analysis is to provide accurate and precise estimates of (i) typical (fixed effect) parameters (taking into account PK & PD parameters and scale factors for covariate effects) and (ii) random effect parameters, accounting for IIV, IOV and residual variability), in order to successfully describe the population of study.

Different softwares and estimation methods have been developed during the last decades to perform population analyses. To the best of our knowledge, NONMEM (NON-linear Mixed Effect Modelling) (23) is the most widely used within the pharmacometrics community. NONMEM is a program that implements the non-linear mixed effects modelling, allowing the estimation of population parameter values and their variability to be obtained in a single step.

During the model development process and once the final model has been selected, it is extremely important to perform a proper evaluation to verify the adequacy of the model to the data.

MODEL EVALUATION

The EMA stated that model evaluation procedures should be submitted to prove that the final model displays a properly good description of the data, enabling the aim of the analysis (24).

Model evaluation can be defined as the procedure to: (i) asses the capability of the model to adequately describe the available data and (ii) determinate the appropriateness of the

Introduction underlying structural and statistical model assumptions. Model evaluation can be seen as a challenge process, in which it is tried to discover the weaknesses of the model.

This exercise is needed at the different levels of the model; (i) structural; exploring the typical PK, PD and system behaviour, and (ii) statistical; analysing the characterization of random effects, both IIV (i. e, distribution, subpopulations, covariance) and residual variability (i.e, subject/time dependency, covariance).

One particularity of the model evaluation exercise is that it does not exist a perfect single statistical/graphical tool that allows selecting and evaluating a population PKPD model. In fact, a combination of different diagnostic tools should be used (25,26).

Model diagnostic can be roughly divided in three groups: (i) numerical, (ii) graphical and (iii) simulation-based diagnostics.

1.1.1 Numerical diagnostics

The minimum value of the objective function (OFV), which is approximately equal to - 2xlog likelihood (-2LL), is used as a test for statistical significance during model development to discriminate between two nested models. Two nested models can be compared by using the log- likelihood ratio test (LRT) based on the difference of their OFVs. A difference in the OFV of 3.84, 6.63 and 10.83 corresponds to the <0.01, <0.05 and <0.001 level of significance, respectively, with the addition of one extra parameter (27). In case of comparing non-nested models, instead of the OFV, the Aikake information criteria (AIC) should be used (28).

The precision and reliability of the parameters should also be evaluated. This precision is expressed as coefficient of variation (CV) and it is assessed with the standard errors (29). It is stated that standard error for structural model parameters and random effects parameters should not exceed 25% and 50%, respectively. The standard errors can be obtained directly from NONMEM. In case standard errors could not been properly estimated, other strategies can be used to obtain confidence intervals (i.e, log-likelihood profiling (30), and parametric or non- parametric bootstrapping (31,32)).

Over-parameterizations of the model can also be assessed by the condition number, calculated from the ratio between the largest and smallest eigenvalues. A condition number exceeding 1000 is indicative of severe ill-conditioning (33).

Introduction

1.1.2 Graphical diagnostics

The most extended graphical diagnostics are the Goodness-of-fit (GOFs) plots. They consist on a set of graphics that allow assessing model performance at the structural and statistical level by exploring discrepancies between model predictions and real data. This graphical evaluation is performed based on the (i) typical population predictions (PRED), (ii) individual model predictions (IPRED), (iii) observations (DV), (iv) residuals (RES; difference between the predictions and the observations), (v) weighted residuals (WRES), (vi) conditional weighted residuals (CWRES) and time.

Some representative examples of these plots are the ones displaying the DVs versus the IPRED or PRED, in order to evaluate the general performance of the model and the discrepancies explained by IIV. If the model is appropriate, the IPRED will approximate to the DVs and the points will be uniformly distributed around the tendency line. Likely, the PRED will be similar to the median/mean of the observations of each study.

More examples of GOFs are (i) WRES or CWRES versus time or versus PRED, to check the structural part of the model. In this case, these residuals should be homogeneously distributed around the chart axis, without showing any apparent tendency. (ii) Histograms showing distribution of the parameters to evaluate the assumption of symmetry around the zero value, or (iii) IWRES versus IPRED or time plot, to evaluate model misspecification at the level of residual error.

In order to have reliable IPRED plots, individual data sufficiently informative on the parameter estimates should be available. Otherwise an over-fit will occur, with the individual parameter estimates distribution shrinking towards 0, IPRED towards OBS and IWRES towards 0, providing excellent agreement when actually, there is a model misspecification. Two diagnostic have been proposed in order to identify and quantify this over-fit, the η-shrinkage and the ε-shrinkage (34). It is stated that the individual plots become meaningless and therefore they should not be consider during model evaluation around shrinkage values of 20-30% higher.

1.3.3 Simulation-based diagnostics

Simulation is defined as the use of a model and its parameters to predict possible outcomes (35). Regarding simulation diagnostic, one of the most widely used instrument for evaluating model performance is the visual predictive check (VPC) (36). VPC evaluates both fixed and random effects parts of the PKPD model, assessing graphically if a model is able to reproduce the central trend and the variability in the observed data. Briefly, VPCs are

Introduction performed as follows: a number of datasets (usually 500 or 1000) of the same characteristics of the original data are simulated using the selected model and its parameter estimates. Then, for each simulated dataset, the 2.5th, 50th, and 97.5th are calculated and the 95% confidence intervals of the calculated percentiles are superimposed onto the 2.5th, 50th, and 97.5th raw data percentiles. VPC is widely used due to its simplicity, however heterogeneity in the design and in the model has to be low to be informative. In case of unbalanced data or differences in the sample time points, dosing schedules or covariate relationship, prediction-corrected VPC (PC- VPC) can be used, normalizing the observed and simulated dependent variable based on the PRED (37).

There are a growing number of tools to support model evaluation diagnostic, such as PsN (38), Xpose (39) or Piraña (40). Also, R6 is widely used for data manipulation, statistical calculation or graphical display.

6 https://www.r-project.org/

Introduction

2. ROLE OF PHARMACOMETRICS IN DRUG DEVELOPMENT PROCESS

At the beginning of this introduction, it was highlighted the high attrition rates in drug discovery, especially during the last phases of the process. This phenomenon occurs mainly due to the lack of efficacy, and/or because of unexpected safety issues (1). These last decades, PKPD modelling has been used to assist drug discovery and development, and its role is becoming more important as model development becomes less empirical and more mechanistic (4). It is used (i) to improve the understanding of PK and PD mechanisms (e.g., linear or saturable metabolism), (ii) dose selection , (iii) dose individualization, and (iv) design optimization (41).

Figure 10 shows some of the applications of PKPD modelling applied during the different stages of drug development, from the early discovery to market, and beyond.

Figure 10. Applications of the model-based approach during the different phases of drug development, adapted from reference (4).

The concept of PKPD modelling and simulation in drug development is under continuous evolution. During this last decade, the application of pharmacometrics, together with system pharmacology, in drug development process has been named as “Model-based drug development”, and defined as “a mathematical and statistical approach that constructs, validates, and utilizes disease models, drug exposure-response models, and pharmacometric models to facilitate drug-development”(42). Recently, this term has evolved to “Model Informed Drug Discovery and Development (MID3), which is defined as “a quantitative framework for prediction and extrapolation, centered in knowledge and inference generated from integrated models of compound, mechanism and disease level data and aimed at improving the quality, efficiency and cost effectiveness of decision making” (43). The replacement of the term “model-

Introduction based” to “model-informed” is driven by the fact that the decision-making was associated to the model instead of to the modeler expert. The addition of the term discovery looks to highlight the significant increase of quantitative approaches in the assessment of relevant pathways.

MID3 has been contextualized and defined by the European Federation of Pharmaceutical Industries and Associations (EFPIA), who have worked on establishing a white paper of good practices (43). MID3 is also landmarked by the “learn-confirm paradigm” proposed by Lewis Sheiner (44,45) which emphasizes the need for early development approaches to effectively inform later stages. The goal of learning is to quantitatively characterize the relationship between prognostic factors, dosage and outcomes. With confirming, it aims to confirm the findings obtained from previous learning phases (42).

2.1. TRANSLATIONAL APPROACHES

In the context of looking for predictive strategies, quantitative translational approaches stand out as one of the most challenging application of MID3. Translational PKPD modelling looks for a better understanding of drug efficacy and safety, optimizing the drug development process (46). Its goal is to develop a framework able to properly anticipate the PKPD properties of drugs in humans using prior information from pre-clinical in vitro and in vivo studies, after accounting for species differences.

When developing translational PKPD models, a first exercise of structuring the key development questions, a clear comprehension of the model assumptions taking during the analysis, and a recurrent refinement of the models during every phase of a drug development program is required (47).

Several PKPD translational modelling approaches, inside the MID3 paradigm, have been applied along different drug development stages and for different therapeutic indication over these last years. Among the different therapeutic areas, oncology stands out as one of the most discouraging field since it is associated with one of the highest attrition rates (95%) (48).

Introduction

3. MID3 APPLIED TO ONCOLOGY

Although a large number of new anticancer drugs have been developed over this last years (49), landscape of oncology drug development is particularly difficult (1).

Some of the current challenges with oncologic drug development include: (i) the difficult characterization of dose-response relationship, due to the fact that in oncology the use of placebo group is not feasible and that just one or two doses are given to the patient population, (ii) the narrow therapeutic index, drug concentrations causing tumour shrinkage usually goes in parallel with adverse effects(50), and (iii) the quantification of overall survival (which is the gold standard to asses clinical outcome of late phase clinical trial), because it involves large number of patients and long study durations, together with the censored information complications.

Given such practical difficulties, population PKPD modelling represents a key tool in the strategy of reducing these high attrition rates in oncology drug development by providing early understanding, identification and quantification of various dose-response relationships (51).

This PKPD modelling approach is based on the development of preclinical models that can act as better predictors of the clinical outcome, and the identification of model-based metrics that can act as predictors of drug response, providing efficient transition from early to late phase clinical trials.

Response endpoints in oncology include biomarkers and surrogate endpoints (i.e tumour marker levels, deemed as surrogate endpoint if those levels can substitute a clinical endpoint), and overall survival as gold standard for clinical endpoint (52).

PD endpoints can be obtained as continuous measurements (biomarkers levels), graded categories (adverse event grade) or categorical measures (response-no response or survival probability). Different type of models can be used to characterize these responses that can act as early predictors of the clinical response during the different stages of drug development process, assisting on the go-no go decision between these different phases.

PRECLINICAL STAGES

MID3 during the preclinical stage involves both in vitro-in vivo analysis and aims to characterize the PK and PD properties of compounds defining toxicity and antitumour efficacy (53).

Introduction

Different PD models can be used to explore the influence of exposure vs time and drug concentration on drug effect in vitro, which can later be used for compound selection and design experiments in vivo(54). From these preclinical in vitro models, the potency of a new drug can be first estimated and used to rank compounds and optimize designs (55). Likely, cell cycle models can be used to characterize the effects of chemotherapeutic agents on cancer cell lines (56), predicting dose and exposure regimens to maximize drug efficacy (57,58) and distinguishing cell cycle arrest versus cell killing in concentration–response curves (59).

A fundamental step of the preclinical development of oncology drugs is the in vivo evaluation of the anti-tumour effects, mostly in mouse bearing xenografts models (60). These models are comprised of human cells injected into host animals (normally mice) providing biological support for growth. Although these models do not consider cancer complexities such as metastasis, immunity, inter- or intratumour heterogeneity or resistance(61), and the fact that their predictive capacity has been largely discussed (62,63), their PKPD characterization has been shown to enhance the interpretation of the preclinical tumour growth inhibition induced by the drug and inform early clinical development (64).

Preclinical efficacy has been traditionally defined based on tumour volume measurements over time, by calculating the distance between the tumour growth curves in control and treated animals, reporting values of tumour growth delaies or tumour growth inhibition (TGI) (65). Unfortunately, these metrics highly depend on the experimental conditions and provide limited biological interpretation of drug efficacy (66). The applications of TGI PKPD models that link the plasma concentrations of the anticancer compounds to the effect on tumour growth, considering time variable, suppose a great advantage in drug efficacy interpretation.

TGI PKPD models are composed of two different parts: (i) one accounting for the unperturbed tumour growth (commonly described by empirical models such as gomperzt, linear, exponential or logistic) and (ii) a second part in which the drug exposure – normally represented by the PK-predicted model concentration or area under the curve- drive the dynamics of the drug effect as a negative effect on tumour growth over time. Several TGI PKPD models can be built depending on the data and on the drug mechanism of action. As an example, Simeoni TGI model (67) is one of the most popular and referenced preclinical TGI PKPD model.

Briefly, in the Simeoni TGI PKPD model, the growth term is composed by two different phases; an initial exponential growth phase (representing initial growth, when tumour cells have adequate nutrients to proliferate) and linear growth (when the availability of nutrients becomes limited due to excessive tumour volume). Also, a transit chain for the tumour affected

Introduction cells is incorporated to the model, to characterize the duration of the death process after drug action (68), emulating a cytotoxic effect.

Another example of TGI PKPD model developed for preclinical data is the one developed by Hahnfeldt and collages to account for the effect of antiangiogenic drugs (69). The particularity of this semi-mechanistic model is that nutrient supply represents a key element of the model structure. Therefore, tumour growth is governed by a gompertz model but incorporating a nutrients supply compartment, which acts as the limited process of the tumour behaviour. Within this model, drug effects of antiangiogenic drugs can also promote the decay of this nutrient compartment, resulting on a final reduction of tumour growth rate (cytostatic effect) and not directly by inducing decay in tumour size.

The development of this type of semi-mechanistic TGI PKPD models that differentiate between system and drug specific model parameters allows predicting the anti-tumour effect of a single compound as a function of the dosing regimen. They are recognized as a suitable tool for extracting the descriptors of these processes, which can be translated from preclinical to clinical settings (70,71).

Several recent publications highlight the benefits of applying the PKPD modelling framework to characterize anti-tumour effect at these preclinical studies in order to anticipate drug response in humans. As an example Wong and colleagues (72) established the relationship between TGI predicted responses using mouse derived PKPD parameters, dosing schedules used in the clinics together with human PK, to responses found in the clinic. Likely, Spilker and co-workers (73), calculate clinical relevant doses from preclinical TGI PKPD models, proposing new preclinical studies designs for optimizing the translation between pre-clinics and clinics.

Another opportunity for improving translation of preclinical phases through modelling and simulation is given by the growing field of systems pharmacology, which promises to link PK with mechanistic/semi-mechanistic models of biological pathway modulation by incorporating intracellular dynamics (74,75).

Recapping, regarding the development of models for preclinical data, most efforts, challenges and expectations are currently focused on the development of semi-mechanistic models (integrating them with possible biological and pharmacological mechanisms on the models) to translate preclinical results into clinical findings (68,76,77).

Introduction

CLINICAL STAGES

Figure 11 shows the typical model-based framework structure followed for characterizing the efficacy of an anticancer drug at clinical phases. First, a PK model is developed to describe the relationship between dose and exposure. By exposure we mean the predicted pharmacokinetic profile or the area under the plasma drug concentration vs time curve (AUC) among other possibilities. (50). Secondly, PKPD models are established to link the exposure profiles with dynamics of the tumor, identifying the dose range associated with clinical activity in early clinical studies. Finally, tumour metrics, derived from longitudinal modelling of tumour size (TS) data and linked to longer-term clinical outcome, which can be progression free survival (PFS) or overall survival (OS) (61).

Figure11. Modelling and simulation of anticancer drugs during clinical drug development.

Regarding the development of the PK model, some considerations have to be taking into account. Although PK evaluation is routinely carried out for anticancer drugs, during clinical trials its collection is really sparse, with high IIV associated to it.

Consequently, population PK models are the best option for its evaluation, allowing the identification of significant covariates, such as BSA, that may assist on guide dose individualization (78). Generally, the PK of anticancer drugs can be described with standard model-based approaches or with PBPK approaches. For developing PBPK models, in vitro-in vivo data integration is needed followed by model verification. One example is the development of a PBPK model for the anticancer drug capecitabine and its metabolites (79); this model coupled tissue specific information about metabolic enzyme activity between tumour and normal cells from in vitro data, allowing the prediction of the therapeutic index in terms of exposure/toxicity in target organs. Another point to take into account in the PK analysis is the fact that cancer therapy is usually handled by combination of drugs. In this context, drug interactions constitute an important problem in oncology due to the narrow therapeutic index on most of these drugs. For example, if a drug that is metabolized by CYP344 is given in

Introduction combination with strong inhibitors or inducers, dose adjustment based on its metabolism pathway are required. PBPK modelling has successfully been applied (and is increasingly used) to asses in this clinical situations, by integrating in vitro information (80,81).

Regarding PD endpoints, tumour size (TS) may be considered the gold standard biomarker for clinical outcome in oncology. At clinical trials, TS is typically measured using imaging techniques and recorded according to the Response Evaluation Criteria in Solid Tumors (RECIST (82)) as the sum of the longest diameters (SLD) across target lesions measured on the target organs. According to RECIST criteria, based on the reduction of SLD, tumour response to treatment is categorized in (i) complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD´). This categorization facilitates the clinical interpretation of the tumour response however, it also presents several limitations (83). As a response to those limitations, population PKPD models describing the time-course of TS, expressed as the sum of the longest tumour dimensions, allow describing drug efficacy over time and predict long-term clinical outcome (68). Among their possibilities, applying PKPD models for TS can (i) assist on go-no go decision of moving to phase III clinical trials based on TS response in phase II trials (84,85), (ii) evaluate the value of biomarkers in predicting tumour response (86), and (iii) can simulate and make predictions of innovative treatment in clinics during clinical routine (87,88).

In structure, TGI models in the clinics are nearly equal to ones used at preclinical stages. Briefly, they are based on ordinary differential equations, where change in TS is explained by a net tumour growth (usually described by a linear, exponential or gompertz function) minus tumour shrinkage (related to drug exposure such as drug concentrations or AUC) due to drug efficacy (52).

One example of TGI model that has been successfully applied for describing several cancer types and drugs(89,90), is the one developed by Claret and collages for describing TS data from colorectal cancer patients treated with capecitabine (85). This model is composed by (i) a disease specific part that follows an exponential unperturbed growth and (i) a drug-related part described by a metric of drug exposure (in this case, daily AUC accounting for off-treatment periods) which drives tumour eradication through a cell killing rate constant. This model also accounts for emerging resistance that decreases exponentially the cell killing rate constant, starting to act at the beginning of the treatment and being independent of dosing schedule and exposure.

One particularity that has to be taking into account in this type of modelling is that TS data are generally collected up to patient´s disease progression or treatment discontinuation and, in this situation, patients drop out from the study due to emerging drug effects, disease

Introduction progression or, in advanced cases, due to death. These missing data are informative and should be taken into account in model development and evaluation, being necessary in some cases to model together TS and drop outs (91).

One of the most important applications of TS models is that they can be used to derive tumour metrics useful to early predict clinical outcomes. For example, as a biomarker of early response, TS has been linked to OS using parametric survival analysis (84,92,93). This kind of approach may assists on go-no go decision in drug development process, by predicting OS in phase III studies by estimating TS dynamics from phase II clinical trials (85).

OS is evaluated in the pharmacometrics field with parametric time-to-event models (TTE) (94). These models allow the identification of the underlying hazard function, from which the survival function can be easily obtained by integrating the hazard with respect to time (95). One of the main advantages of using this type of models is the easy implementation of predictors (covariates) in the hazard function, being able to explain variability in OS and make prediction of different OS probability based on these identified covariates. Some examples of these covariates can be: (i) a baseline metric for each patient (i.e, baseline TS (84)), (ii)an individual parameter estimate (i.e, tumour growth rate (96)), patient baseline characteristic (i.e, ECOG (97)) or a time varying metric (i.e tumour marker or TS over time model-derived).

Likely, this type of TTE models can be applied not only to characterize OS, but also PFS, time of disease progression occurrence, time to complete response or time to toxicity event.

Although models for adverse events and for biomarkers have not been explicitly addressed in clinical PKPD modelling introduction, they should be highlighted, especially in field of Oncology. In the following two example of both type of models are briefly explained.

Although recent target therapies aim to decrease drug toxicity, it is still common to face adverse effects in the field of oncology, being haematological toxicities (myelosuppression) the most common ones. Through PKPD modelling these adverse effects can be characterized and can provide quantitative information to achieve a balance between risk of toxicity and drug efficacy. As an example, Friberg and coworkers developed a semi-mechanistic model to characterize drug-induced myelosuppression (88), that has become the most widely applied model to account for this type of toxicity (including neutropenia, leukopenia and thrombocytopenia) of different chemotherapy drugs given alone or in combination.

On the other hand, the identification of a predictive biomarker in oncology assists on the prediction of treatment efficacy or disease progression. Semi-mechanistic models to

Introduction characterized biomarker dynamics are often described by indirect response models(98). Buil- Bruna (99) characterize the time-course of two biomarkers in non-small cell lung cancer patients and related them to tumour progression levels assessed by RECIST categories. Subsequently, the probability of disease progression was related to the mentioned developed framework (100), providing a valuable tool for predict clinical success or failure.

In summary, pharmacometrics has a great implication in drug development process, especially in the oncology field. However there is much left to be done. Current modelling initiatives consist on integrating preclinical and clinical data and developing translational PKPD models that may anticipate, from early drug development stages, if a drug is going to successfully reach the market. The use of marketed drugs, with available preclinical and clinical data, to develop such an approach represents a great opportunity. In this work, the drug used to develop this analysis is gemcitabine, an anticancer drug indicated for the treatment of several solid tumours.

Introduction

4. GEMCITABINE

Gemcitabine (dFdC) is an antimetabolite cytotoxic-cytostatic anticancer pro-drug, analogue of the pyrimidine (101). It was synthesized in the 1980s by the pharmaceutical company Eli Lilly (Gemzar ®) and, although it was first thought as an antiviral drug, it was finally developed as an anticancer drug by presenting a greater promising activity in the oncology field (102).

Specifically, gemcitabine presents a high activity against solid tumours, being approved in the treatment of locally advanced or metastatic pancreatic cancer as single agent or in combination and in the treatment of nonsmall-cell lung (NSCL), breast, and ovarian cancer in combination (103,104).

The activity and toxicity of gemcitabine depend on dose and dosing schedule (105,106). It is usually administered intravenously (i.v), at doses of 1000-1250 mg/m2 as a 0.5-1 h infusion, given on days 1 and 8 or 1, 8 and 15 of a 21-days cycle and a 28-days cycle respectively (107). Its highest adverse effects are mostly haematological toxicities, manifested by leukopenia, thrombocytopenia and anemia, being myelosuppression the principal dose-limiting toxicity. In fact, patients should be monitored for myelosuppression during therapy. It may also cause, although with less frequency, flu-like symptoms, nausea, vomiting, and rash (103). However, one of the main problems of gemcitabine is the resistance to the treatment of several tumour types, either acquired of inherent (102,108,109).

Regarding its mechanism of action, as a pro-drug, the parent compound requires cellular uptake and intracellular phosphorylation to achieve its active form (gemcitabine triphosphate, dFdCTP) in order to exert its cytotoxic action, following a complex metabolic pathway (109), represented in figure 12.

Figure 12. Gemcitabine pathway. In blue, the natural metabolite pools reactions are shown. The development of the inactive metabolite dFdU, and its metabolites are displayed in yellow. Finally, the metabolism of gemcitabine, leading to the active metabolite, dFdCTP that binds to the DNA and RNA

Introduction compiting with the natural metabolites pools, are shown in green. The name of the enzymes and transporters are also indicated in the metabolism reactions.

First, its uptake into the cell requires specific carriage by active nucleoside transporters (NTs) located in the cell plasma membrane (hENTs and hCNTs transporters) (110). Once intracellularly, gemcitabine (dFdC) is phosphorylated by kinases enzymes (dCK, NMPK and DPK) to mono-(dFdCMP), di-(dFdCDP) and tri-phosphate (dFdCTP) to form the active metabolites that will be incorporated to the DNA (111).

dFdCTP competes with cell natural metabolites (dCTP) for incorporation to the DNA (main mechanism of action). Its incorporation promotes masked chain termination, causing apoptosis and inhibiting cellular tumour growth. The rate of dFdCTP incorporation depends on the activity of DNA polymerases, the competition with dCTP and the remove of gemcitabine from the DNA. Gemciabine is also incorporated into the RNA, however, although its incorporation has been related to gemcitabine sensitivity, it has not been proved that it plays an important role in the drug toxicity (112).

dFdCMP inhibits ribonucleotide reductase (RR), depleting dCTP pools (113). Besides, the reduction in the intracellular concentration of dCTP (by the action of the diphosphate) enhances the incorporation of gemcitabine triphosphate into DNA (self-potentiation). In figure 12, these self-potentiating mechanisms, that keep FdCTP and dFdCDP levels prolonged in time, can be appreciated.

Gemcitabine also suffers inactivation, by CDA enzyme, leading to the inactive metabolite dFdU, which is excreted in urine (113). Because of this rapid deamination, gemcitabine shows a short elimination half-life (t1/2), in the range if 10-30 minutes (107).

Although the main responsible of gemcitabine effects is the accumulation and intracellular retention of dFdCTP, sensitivity to the treatment is also determined by (i) the drug´s activity to enter into the cell, (ii) the activity of the catabolizing, anabolizing and target enzymes and (iii) cells´ ability to repair DNA damage (109).

Even though gemcitabine has been proved to be effective against several tumours, responses rates are still low, especially in relapsed tumours. In fact, in advanced pancreatic cancer, although treatment with gemcitabine constitutes the first line therapy (either given as single agent or in combination with paclitaxel), life expectancy is still discouraging, with responses rates of 10% and an OS of a 7% (114). Therefore, emphasis should be done in

Introduction investigating gemcitabine-based combination therapy, together with further research on its mechanism of action, given the huge number of target enzymes involved on it.

PKPD MODELLING APPLIED TO GEMCITABINE

Gemcitabine plasma concentration versus time has been described by a two compartment model, with drug distribution into central and peripheral compartments. This PK model (42) developed using combined data from seven PK studies performed in NSCLC and pancreatic cancer patients, also includes the characterization of the inactive metabolite dFdU and of the active metabolite dFdCTP. The model structure is displayed in figure 13.

dFdC dFdU Peripheral Peripheral

k12 k21 k34 k43

dFdCTP Vm dFdC kf dFdU Km Central Central

k50 k10 k30

Figure 13. Schematic representation of gemcitabine and its metabolites dFdU and dFdCTP PK. Adapted from reference (42). Vm & Km; Michaelis-Menten parameters driving dFdCTP development. k10, k30, k50; rate constant driving dFdC, dFdU and dFdCTP elimination, respectively. kf; apparent formation rate constant for dFdU. k12, k21, k34, k43; inter-compartment rate constant.

BSA, age, gender and duration of the infusion were identified as covariates of dFdC PK parameters. The inactive metabolite, dFdU, is excreted unchanged in urine. Apparent formation rate constant of this metabolite was determined by the total fraction of the gemcitabine dose excreted in urine.

Regarding the active metabolite, dFdCTP, its concentration is characterized in peripheral mononuclear cells, and its development is driven by a saturate metabolism, described by a Michaelis-Menten model, that was developed based on in vitro literature reports. In the development of this model, dFdCTP was suggested as a potential efficacy biomarker.

Introduction

Information regarding Gemcitabine dose-response relationship and the factors contributing to its magnitude of response are still limited. To the best of our knowledge, there is a lack of publicly available reports dealing with the PKPD characteristics of gemcitabine apart from the recent in vivo pre-clinical analysis in breast cancer xenografts (115), and the mechanistic models developed with in vitro data (56,116,117).

Regarding clinical PKPD models, the effect of gemcitabine administered in combination in NSCLC, breast and ovarian cancer patients have been described linking model predicted changes in TS with OS (89,90,118). With respect to pancreatic cancer, a model linking TS with OS has been recently published (119). However, drug exposure information was not reported (i.e., dose levels, treatment duration or dose reductions or discontinuations) thus precluding the characterization of gemcitabine effects and limiting the use of the model to evaluate outcome results from model-based simulations exploring alternative dosing scenarios. In addition, ignoring drug exposure prevents analysis of whether progression in the disease occurs due to treatment interruptions or emerging resistance.

Introduction

5. REFERENCES

1. Kola I, Landis J. Can the pharmaceutical industry reduce attrition rates? Nat Rev Drug Discov. 2004 Aug;3(8):711–6.

2. Ette EI, Williams PJ. Pharmacometrics : the science of quantitative pharmacology. John Wiley & Sons; 2007. 1205 p.

3. Milligan PA, Brown MJ, Marchant B, Martin SW, van der Graaf PH, Benson N, et al. Model- based drug development: a rational approach to efficiently accelerate drug development. Clinical pharmacology and therapeutics. 2013 Jun 14;93(6):502–14.

4. Bonate PL. The Art of Modeling. In: Pharmacokinetic-Pharmacodynamic Modeling and Simulation. 2011 p. 1–60.

5. Holford NH, Sheiner LB. Kinetics of pharmacologic response. Pharmacology & therapeutics. 1982;16(2):143–66.

6. Gabrielsson J, Dolgos H, Gillberg P-G, Bredberg U, Benthem B, Duker G. Early integration of pharmacokinetic and dynamic reasoning is essential for optimal development of lead compounds: strategic considerations. Drug Discovery Today. 2009 Apr;14(7–8):358–72.

7. Sheiner LB, Stanski DR, Vozeh S, Miller RD, Ham J. Simultaneous modeling of pharmacokinetics and pharmacodynamics: application to d-tubocurarine. Clinical pharmacology and therapeutics. 1979 Mar;25(3):358–71.

8. Dayneka NL, Garg V, Jusko WJ. Comparison of four basic models of indirect pharmacodynamic responses. Journal of pharmacokinetics and biopharmaceutics. 1993 Aug;21(4):457–78.

9. Kenakin TP. Pharmacologic analysis of drug-receptor interaction. Raven Press; 1987. 335 p.

10. Chan PL, Holford NH. Drug treatment effects on disease progression. Annual review of pharmacology and toxicology. 2001 Apr;41(1):625–59.

11. Dahl SG, Aarons L, Gundert-Remy U, Karlsson MO, Schneider YJ, Steimer JL, et al. Incorporating physiological and biochemical mechanisms into pharmacokinetic- pharmacodynamic models: A conceptual framework. Basic and Clinical Pharmacology and Toxicology. 2010;106(1):2–12.

Introduction

12. Jones H, Rowland-Yeo K. Basic concepts in physiologically based pharmacokinetic modeling in drug discovery and development. CPT: pharmacometrics & systems pharmacology. 2013 Aug 14;2(8):e63.

13. Jones H, Rowland-Yeo K. Basic concepts in physiologically based pharmacokinetic modeling in drug discovery and development. CPT: pharmacometrics & systems pharmacology. 2013 Aug;2(8):e63.

14. Maharaj AR, Edginton AN. Physiologically Based Pharmacokinetic Modeling and Simulation in Pediatric Drug Development. CPT: Pharmacometrics & Systems Pharmacology. 2014 Nov;3(11):148.

15. Edginton AN, Willmann S. Physiology-based simulations of a pathological condition: prediction of pharmacokinetics in patients with liver cirrhosis. Clinical pharmacokinetics. 2008;47(11):743–52.

16. Sinha VK, Snoeys J, Osselaer N Van, Peer A Van, Mackie C, Heald D. From preclinical to human - prediction of oral absorption and drug-drug interaction potential using physiologically based pharmacokinetic (PBPK) modeling approach in an industrial setting: a workflow by using case example. Biopharmaceutics & Drug Disposition. 2012 Mar;33(2):111–21.

17. Schaller S, Willmann S, Lippert J, Schaupp L, Pieber TR, Schuppert A, et al. A Generic Integrated Physiologically based Whole-body Model of the Glucose-Insulin-Glucagon Regulatory System. CPT: pharmacometrics & systems pharmacology. 2013 Aug 14;2(8):e65.

18. Willmann S, Thelen K, Becker C, Dressman JB, Lippert J. Mechanism-based prediction of particle size-dependent dissolution and absorption: cilostazol pharmacokinetics in dogs. European journal of pharmaceutics and biopharmaceutics : official journal of Arbeitsgemeinschaft fur Pharmazeutische Verfahrenstechnik eV. 2010 Sep;76(1):83–94.

19. Thiel C, Schneckener S, Krauss M, Ghallab A, Hofmann U, Kanacher T, et al. A systematic evaluation of the use of physiologically based pharmacokinetic modeling for cross- species extrapolation. Journal of pharmaceutical sciences. 2015 Jan;104(1):191–206.

20. Eissing T, Kuepfer L, Becker C, Block M, Coboeken K, Gaub T, et al. A Computational Systems Biology Software Platform for Multiscale Modeling and Simulation: Integrating Whole-Body Physiology, Disease Biology, and Molecular Reaction Networks. Frontiers in

Introduction

Physiology. 2011;2:4.

21. Krauss M, Burghaus R, Lippert J, Niemi M, Neuvonen P, Schuppert A, et al. Using Bayesian-PBPK modeling for assessment of inter-individual variability and subgroup stratification. In silico pharmacology. 2013;1(1):6.

22. Sheiner LB, Ludden TM. Population Pharmacokinetics/Dynamics*. Annual Review of Pharmacology and Toxicology. 1992 Apr;32(1):185–209.

23. Bauer R. NONMEM users guide: introduction to NONMEM 7.2.0. ICON Development Solutions: Ellicott City; 2011.

24. Ewp. COMMITTEE FOR MEDICINAL PRODUCTS FOR HUMAN USE (CHMP) GUIDELINE ON REPORTING THE RESULTS OF POPULATION PHARMACOKINETIC ANALYSES DRAFT AGREED BY EFFICACY WORKING PARTY ADOPTION BY CHMP FOR RELEASE FOR CONSULTATION. 2006 u.

25. Karlsson MO, Savic RM. Diagnosing Model Diagnostics. Clinical Pharmacology & Therapeutics. 2007 Jul;82(1):17–20.

26. Nguyen THT, Mouksassi M-S, Holford N, Al-Huniti N, Freedman I, Hooker AC, et al. Model Evaluation of Continuous Data Pharmacometric Models: Metrics and Graphics. CPT: pharmacometrics & systems pharmacology. 2017 Feb;6(2):87–109.

27. Wählby U, Jonsson EN, Karlsson MO. Assessment of actual significance levels for covariate effects in NONMEM. Journal of pharmacokinetics and pharmacodynamics. 2001 Jun;28(3):231–52.

28. Ette EI. Comparing non-hierarchical models: application to non-linear mixed effects modeling. Computers in biology and medicine. 1996 Nov;26(6):505–12.

29. Ette EI, Williams PJ. Population Pharmacokinetics II: Estimation Methods. Annals of Pharmacotherapy. 2004 Nov;38(11):1907–15.

30. Sheiner LB. Analysis of pharmacokinetic data using parametric models. III. Hypothesis tests and confidence intervals. Journal of pharmacokinetics and biopharmaceutics. 1986 Oct;14(5):539–55.

31. Ette EI. Stability and performance of a population pharmacokinetic model. Journal of clinical pharmacology. 1997 Jun;37(6):486–95.

Introduction

32. Ette EI, Williams PJ, Kim YH, Lane JR, Liu M-J, Capparelli E V. Model appropriateness and population pharmacokinetic modeling. Journal of clinical pharmacology. 2003 Jun;43(6):610–23.

33. Wilkins JJ. NONMEMory: a run management tool for NONMEM. Computer methods and programs in biomedicine. 2005 Jun;78(3):259–67.

34. Savic RM, Karlsson MO. Importance of shrinkage in empirical bayes estimates for diagnostics: problems and solutions. The AAPS journal. 2009 Sep;11(3):558–69.

35. Rombout F, Aarons L, Karlsson M, Man A, Mentré F, Nygren P, et al. Modelling and simulation in the development and use of anti-cancer agents: an underused tool? Journal of pharmacokinetics and pharmacodynamics. 2004 Dec;31(6):419–40.

36. Holford N. The Visual Predictive Check – Superiority to Standard Diagnostic (Rorschach) Plots. PAGE 14 (2005) Abstr 738 [www.page-meeting.org/?abstract=738]

37. Bergstrand M, Hooker AC, Wallin JE, Karlsson MO. Prediction-corrected visual predictive checks for diagnosing nonlinear mixed-effects models. The AAPS journal. 2011 Jun;13(2):143–51.

38. Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN)—a Perl module for NONMEM related programming. Computer methods and programs in biomedicine. 2004;75(2):85–94.

39. Jonsson EN, Karlsson MO. Xpose—an S-PLUS based population pharmacokinetic/pharmacodynamic model building aid for NONMEM. Computer methods and programs in biomedicine 1998 Jan;58(1):51–64.

40. Keizer RJ, Karlsson MO, Hooker A. Modeling and Simulation Workbench for NONMEM: Tutorial on Pirana, PsN, and Xpose. CPT: Pharmacometrics & Systems Pharmacology. 2013 Jun 26;2(6):e50.

41. Mould DR, Upton RN. Basic Concepts in Population Modeling, Simulation, and Model- Based Drug Development. CPT: Pharmacometrics & Systems Pharmacology. 2012 Sep 26;1(9):e6.

42. Zhang L, Sinha V, Forgue S, Callies S, Ni L, Peck R, et al. Model-Based Drug Development: The Road to Quantitative Pharmacology. Journal of Pharmacokinetics & Pharmacodynamics. 2006 Jul;33(3):369.

Introduction

43. Marshall S, Burghaus R, Cosson V, Cheung S, Chenel M, DellaPasqua O, et al. Good Practices in Model-Informed Drug Discovery and Development: Practice, Application, and Documentation. CPT: Pharmacometrics & Systems Pharmacology. 2016 Mar 1;5(3):93– 122.

44. Sheiner LB, Steimer JL. Pharmacokinetic/pharmacodynamic modeling in drug development. Annual review of pharmacology and toxicology. 2000 Apr;40(1):67–95.

45. Sheiner LB. Learning versus confirming in clinical drug development. Clinical pharmacology and therapeutics. 1997 Mar;61(3):275–91.

46. Danhof M, de Lange ECM, Della Pasqua OE, Ploeger BA, Voskuyl RA. Mechanism-based pharmacokinetic-pharmacodynamic (PK-PD) modeling in translational drug research. Trends in pharmacological sciences. 2008 Apr;29(4):186–91.

47. Chien JY, Friedrich S, Heathman MA, de Alwis DP, Sinha V. Pharmacokinetics/Pharmacodynamics and the stages of drug development: role of modeling and simulation. The AAPS journal. 2005 Oct 7;7(3):E544-59.

48. Moreno L, Pearson AD. How can attrition rates be reduced in cancer drug discovery? Expert opinion on drug discovery. 2013 Apr;8(4):363–8.

49. Trotta F, Leufkens HGM, Schellens JHM, Laing R, Tafuri G. Evaluation of Oncology Drugs at the European Medicines Agency and US Food and Drug Administration: When Differences Have an Impact on Clinical Practice. Journal of Clinical Oncology. 2011 Jun 1;29(16):2266–72.

50. Park K. A Review of Modeling Approaches to Predict Drug Response in Clinical Oncology. Yonsei medical journal. 2017 Jan;58(1):1–8.

51. Bender BC, Schindler E, Friberg LE. Population pharmacokinetic-pharmacodynamic modelling in oncology: a tool for predicting clinical response. British journal of clinical pharmacology. 2015 Jan;79(1):56–71.

52. Buil-Bruna N, Lopez-Picazo JM, Martin-Algarra S, Troconiz IF. Bringing Model-Based Prediction to Oncology Clinical Practice: A Review of Pharmacometrics Principles and Applications. The oncologist. 2016 Feb;21(2):220–32.

53. Barrett JS, Gupta M, Mondick JT. Model-based drug development applied to oncology. Expert opinion on drug discovery. 2007 Feb 16;2(2):185–209.

Introduction

54. Suggitt M, Bibby MC. 50 years of preclinical anticancer drug screening: empirical to target-driven approaches. Clinical cancer research : an official journal of the American Association for Cancer Research. 2005 Feb 1;11(3):971–81.

55. Karlsson MO, Anehall T, Friberg LE, Henningsson A, Kloft C, Sandström M, et al. Pharmacokinetic/pharmacodynamic modelling in oncological drug development. Basic & clinical pharmacology & toxicology. 2005 Mar;96(3):206–11.

56. Hamed SS, Straubinger RM, Jusko WJ. Pharmacodynamic modeling of cell cycle and apoptotic effects of gemcitabine on pancreatic adenocarcinoma cells. Cancer chemotherapy and pharmacology. 2013 Sep;72(3):553–63.

57. Florian JA, Eiseman JL, Parker RS. Accounting for quiescent cells in tumour growth and cancer treatment. Systems biology. 2005 Dec;152(4):185–92.

58. Sherer E, Hannemann RE, Rundell A, Ramkrishna D. Analysis of resonance chemotherapy in leukemia treatment via multi-staged population balance models. Journal of theoretical biology. 2006 Jun 21;240(4):648–61.

59. Hamed SS, Roth CM. Mathematical modeling to distinguish cell cycle arrest and cell killing in chemotherapeutic concentration response curves. Journal of pharmacokinetics and pharmacodynamics. 2011 Jun 27;38(3):385–403.

60. Rocchetti M, Poggesi I, Germani M, Fiorentini F, Pellizzoni C, Zugnoni P, et al. A Pharmacokinetic-Pharmacodynamic Model for Predicting Tumour Growth Inhibition in Mice: A Useful Tool in Oncology Drug Development. Basic Clinical Pharmacology Toxicology. 2005 Mar;96(3):265–8.

61. Venkatakrishnan K, Friberg LE, Ouellet D, Mettetal JT, Stein A, Trocóniz IF, et al. Optimizing oncology therapeutics through quantitative translational and clinical pharmacology: challenges and opportunities. Clinical pharmacology and therapeutics. 2015 97(1):37–54.

62. Ruggeri BA, Camp F, Miknyoczki S. Animal models of disease: pre-clinical animal models of cancer and their applications and utility in drug discovery. Biochemical pharmacology. 2014 Jan;87(1):150–61.

63. Kerbel RS. Human tumor xenografts as predictive preclinical models for anticancer drug activity in humans: better than commonly perceived-but they can be improved. Cancer

Introduction

biology & therapy. 2(4 Suppl 1):S134-9.

64. Stroh M, Duda DG, Takimoto CH, Yamazaki S, Vicini P. Translation of anticancer efficacy from nonclinical models to the clinic. CPT: pharmacometrics & systems pharmacology. 2014 Aug;3:e128.

65. Sausville EA, Burger AM, Becher OJ, Holland EC. Contributions of human tumor xenografts to anticancer drug development. Cancer research. 2006 Apr 1;66(7):3351–4, discussion 3354.

66. Steimer J-L, Dahl SG, De Alwis DP, Gundert-Remy U, Karlsson MO, Martinkova J, et al. Modelling the genesis and treatment of cancer: The potential role of physiologically based pharmacodynamics. European Journal of Cancer. 2010 Jan;46(1):21–32.

67. Simeoni M, Magni P, Cammia C, De Nicolao G, Croci V, Pesenti E, et al. Predictive Pharmacokinetic-Pharmacodynamic Modeling of Tumor Growth Kinetics in Xenograft Models after Administration of Anticancer Agents. Cancer research. 2004;64(3):1094– 101.

68. Ribba B, Holford NH, Magni P, Troconiz I, Gueorguieva I, Girard P, et al. A review of mixed-effects models of tumor growth and effects of anticancer drug treatment used in population analysis. CPT: pharmacometrics & systems pharmacology. 2014 May 7;3:e113.

69. Hahnfeldt P. Tumor development under angiogenic signaling: A dynamical theory of tumor growth, treatment response, and postvascular dormancy. Cancer research. 1999;59(19):4770.

70. Rocchetti M. Predicting the active doses in humans from animal studies: A novel approach in oncology. European journal of cancer. 2007;43(12):1862.

71. Terranova N, Germani M, Del Bene F, Magni P. A predictive pharmacokinetic– pharmacodynamic model of tumor growth kinetics in xenograft mice after administration of anticancer agents given in combination. Cancer Chemotherapy and Pharmacology. 2013 Aug 29;72(2):471–82.

72. Wong H, Choo EF, Alicke B, Ding X, La H, McNamara E, et al. Antitumor Activity of Targeted and Cytotoxic Agents in Murine Subcutaneous Tumor Models Correlates with Clinical Response . Vol. 18, Clin.Cancer Res. 2012. p. 3846–55. A

73. Spilker ME, Chen X, Visswanathan R, Vage C, Yamazaki S, Li G, et al. Found in Translation:

Introduction

Maximizing the Clinical Relevance of Nonclinical Oncology Studies. Clinical Cancer Research. 2017 Jan 4; A.

74. Zhang X-Y, Birtwistle MR, Gallo JM. A General Network Pharmacodynamic Model-Based Design Pipeline for Customized Cancer Therapy Applied to the VEGFR Pathway. CPT: pharmacometrics & systems pharmacology. 2014 Jan 15;3(1):e92.

75. Krippendorff B-F, Kuester K, Kloft C, Huisinga W. Nonlinear pharmacokinetics of therapeutic proteins resulting from receptor mediated endocytosis. Journal of Pharmacokinetics and Pharmacodynamics. 2009 Jun 25;36(3):239–60.

76. Haddish-Berhane N, Shah DK, Ma D, Leal M, Gerber H-P, Sapra P, et al. On translation of antibody drug conjugates efficacy from mouse experimental tumors to the clinic: a PK/PD approach. Journal of Pharmacokinetics and Pharmacodynamics. 2013;40(5):557– 71.

77. Shah DK, Haddish-Berhane N, Betts A. Bench to bedside translation of antibody drug conjugates using a multiscale mechanistic PK/PD model: a case study with brentuximab- vedotin. Journal of Pharmacokinetics and Pharmacodynamics. 2012 Dec 15;39(6):643– 59.

78. Mould DR, Walz AC, Lave T, Gibbs JP, Frame B. Developing Exposure/Response Models for Anticancer Drug Treatment: Special Considerations. CPT: pharmacometrics & systems pharmacology. 2015 Jan;4(1):e00016.

79. Tsukamoto Y, Kato Y, Ura M, Horii I, Ishitsuka H, Kusuhara H, et al. A physiologically based pharmacokinetic analysis of capecitabine, a triple prodrug of 5-FU, in humans: the mechanism for tumor-selective accumulation of 5-FU. Pharmaceutical research. 2001 Aug;18(8):1190–202.

80. Perdaems N, Blasco H, Vinson C, Chenel M, Whalley S, Cazade F, et al. Predictions of Metabolic Drug-Drug Interactions Using Physiologically Based Modelling. Clinical Pharmacokinetics. 2010 Apr;49(4):239–58.

81. Rowland Yeo K, Walsky RL, Jamei M, Rostami-Hodjegan A, Tucker GT. Prediction of time- dependent CYP3A4 drug–drug interactions by physiologically based pharmacokinetic modelling: Impact of inactivation parameters and enzyme turnover. European Journal of Pharmaceutical Sciences. 2011 Jun 14;43(3):160–73.

Introduction

82. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1). European Journal of Cancer. 2009 Jan;45(2):228–47.

83. Ratain MJ, Eckhardt SG. Phase II studies of modern drugs directed against new targets: if you are fazed, too, then resist RECIST. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2004 Nov 15;22(22):4442–5.

84. Wang Y, Sung C, Dartois C, Ramchandani R, Booth BP, Rock E, et al. Elucidation of relationship between tumor size and survival in non-small-cell lung cancer patients can aid early decision making in clinical drug development. Clinical pharmacology and therapeutics. 2009 Aug;86(2):167–74.

85. Claret L, Girard P, Hoff PM, Van Cutsem E, Zuideveld KP, Jorga K, et al. Model-Based Prediction of Phase III Overall Survival in Colorectal Cancer on the Basis of Phase II Tumor Dynamics. Journal of Clinical Oncology. 2009;27(25):4103–8.

86. Hansson EK, Amantea MA, Westwood P, Milligan PA, Houk BE, French J, et al. PKPD Modeling of VEGF, sVEGFR-2, sVEGFR-3, and sKIT as Predictors of Tumor Dynamics and Overall Survival Following Sunitinib Treatment in GIST. CPT: pharmacometrics & systems pharmacology. 2013 Nov 20;2(11):e84.

87. Ribba B, Kaloshi G, Peyre M, Ricard D, Calvez V, Tod M, et al. A Tumor Growth Inhibition Model for Low-Grade Glioma Treated with Chemotherapy or Radiotherapy. Clinical Cancer Research. 2012 Sep 15;18(18):5071–80.

88. Friberg LE, Henningsson A, Maas H, Nguyen L, Karlsson MO. Model of Chemotherapy- Induced Myelosuppression With Parameter Consistency Across Drugs. Journal of Clinical Oncology. 2002 Dec 15;20(24):4713–21.

89. Tate SC, Andre V, Enas N, Ribba B, Gueorguieva I. Early change in tumour size predicts overall survival in patients with first-line metastatic breast cancer. European journal of cancer . 2016 Oct;66:95–103.

90. Zecchin C, Gueorguieva I, Enas NH, Friberg LE. Models for change in tumour size, appearance of new lesions and survival probability in patients with advanced epithelial ovarian cancer. British journal of clinical pharmacology. 2016 Sep;82(3):717–27.

91. Hu C, Sale ME. A joint model for nonlinear longitudinal data with informative dropout.

Introduction

Journal of pharmacokinetics and pharmacodynamics. 2003 Feb;30(1):83–103.

92. Claret L, Gupta M, Han K, Joshi A, Sarapa N, He J, et al. Evaluation of tumor-size response metrics to predict overall survival in Western and Chinese patients with first-line metastatic colorectal cancer. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2013 Jun 10;31(17):2110–4.

93. Bruno R, Lindbom L, Schaedeli Stark F, Chanu P, Gilberg F, Frey N, et al. Simulations to Assess Phase II Noninferiority Trials of Different Doses of Capecitabine in Combination With Docetaxel for Metastatic Breast Cancer. CPT: pharmacometrics & systems pharmacology. 2012 Dec 26;1:e19.

94. Holford N. A time to event tutorial for pharmacometricians. CPT: pharmacometrics & systems pharmacology. 2013 May 15;2(5):e43.

95. Collett D. Collett David - Modelling survival data in medical research (Chapman&Hall-CRC, 2004)(ISBN 1584883251).pdf. 2003.

96. Quartino A, Claret L, Li J, Lum B, Visich J, Bruno R, et al. Evaluation of Tumour Size Metrics to Predict Survival in Advanced Gastric Cancer. PAGE 22 (2013) Abstr 2812 [www.page-meeting.org/?abstract=2812].

97. Claret L, Lu JF, Sun YN, Bruno R. Development of a modeling framework to simulate efficacy endpoints for motesanib in patients with thyroid cancer. Cancer chemotherapy and pharmacology. 2010 Nov;66(6):1141–9.

98. Sharma A, Jusko WJ. Characteristics of indirect pharmacodynamic models and applications to clinical drug responses. British journal of clinical pharmacology. 1998 Mar;45(3):229–39.

99. Buil-Bruna N, López-Picazo J-M, Moreno-Jiménez M, Martín-Algarra S, Ribba B, Trocóniz IF. A Population Pharmacodynamic Model for Lactate Dehydrogenase and Neuron Specific Enolase to Predict Tumor Progression in Small Cell Lung Cancer Patients. The AAPS Journal. 2014 May 17;16(3):609–19.

100. Buil-Bruna N, Sahota T, Lopez-Picazo J-M, Moreno-Jimenez M, Martin-Algarra S, Ribba B, et al. Early Prediction of Disease Progression in Small Cell Lung Cancer: Toward Model- Based Personalized Medicine in Oncology. Cancer Research. 2015 Jun 15;75(12):2416– 25.

Introduction

101. Mini E, Nobili S, Caciagli B, Landini I, Mazzei T. Cellular pharmacology of gemcitabine. Annals of Oncology. 2006 May 1;17(suppl 5):v7–12.

102. Teicher Godefridus Johannes BP. Gemcitabine. Deoxynucleoside analogs in cancer therapy : Cancer drug discovery and development. 2007. p. 225.

103. Hui YF, Reitz J. Gemcitabine: a cytidine analogue active against solid tumors. American Journal of Health-System Pharmac. 1997;54(2):162–70.

104. Toschi L, Finocchiaro G, Bartolini S, Gioia V, Cappuzzo F. Role of gemcitabine in cancer therapy. Future oncology. 2005 Feb;1(1):7–17.

105. Kaye SB. Gemcitabine: current status of phase I and II trials. Journal of Clinical Oncology. 1994 Aug;12(8):1527–31.

106. Boven E, Schipper H, Erkelens CA, Hatty SA, Pinedo HM. The influence of the schedule and the dose of gemcitabine on the anti-tumour efficacy in experimental human cancer. British journal of cancer. 1993 Jul;68(1):52–6.

107. Veltkamp SA, Beijnen JH, Schellens JHM. Prolonged versus standard gemcitabine infusion: translation of molecular pharmacology to new treatment strategy. The oncologist. 2008 Mar 1;13(3):261–76.

108. Binenbaum Y, Na’ara S, Gil Z. Gemcitabine resistance in pancreatic ductal adenocarcinoma. Drug resistance updates : reviews and commentaries in antimicrobial and anticancer chemotherapy. 2015 Nov;23:55–68.

109. Bergman AM, Pinedo HM, Peters GJ. Determinants of resistance to 2′,2′- difluorodeoxycytidine (gemcitabine). Drug Resistance Updates. 2002;5(1):19–33.

110. Mackey JR, Mani RS, Selner M, Mowles D, Young JD, Belt JA, et al. Functional nucleoside transporters are required for gemcitabine influx and manifestation of toxicity in cancer cell lines. Cancer research. 1998 Oct 1;58(19):4349–57.

111. Storniolo AM, AllerheiligenS R, Pearce HL. Preclinical, pharmacologic, and phase I studies of gemcitabine. Seminars in oncology. 1997;24(2 Suppl 7):S7–S2.

112. Bergman AM, Peters GJ. Gemcitabine. In: Peters GJ, editor. Deoxynucleoside Analogs In Cancer Therapy. Totowa, NJ: Humana Press; 2007. p. 225–51.

113. Plunkett W, Huang P, Xu YZ, Heinemann V, Grunewald R, Gandhi V. Gemcitabine:

Introduction

metabolism, mechanisms of action, and self-potentiation. Seminars in oncology. 1995 Aug;22(4 Suppl 11):3–10.

114. Fuchs CS, Azevedo S, Okusaka T, Van Laethem J-L, Lipton LR, Riess H, et al. A phase 3 randomized, double-blind, placebo-controlled trial of ganitumab or placebo in combination with gemcitabine as first-line therapy for metastatic adenocarcinoma of the pancreas: the GAMMA trial. Annals of oncology : official journal of the European Society for Medical Oncology. 2015 May;26(5):921–7.

115. Yuan Y, Zhou X, Ren Y, Zhou S, Wang L, Ji S, et al. Semi-Mechanism-Based Pharmacokinetic/Pharmacodynamic Model for the Combination Use of Dexamethasone and Gemcitabine in Breast Cancer. Journal of pharmaceutical sciences. 2015 Dec;104(12):4399–408.

116. Miao X, Koch G, Straubinger RM, Jusko WJ. Pharmacodynamic modeling of combined chemotherapeutic effects predicts synergistic activity of gemcitabine and trabectedin in pancreatic cancer cells. Cancer chemotherapy and pharmacology. 2016 Jan;77(1):181–93.

117. Zhu X, Straubinger RM, Jusko WJ. Mechanism-based mathematical modeling of combined gemcitabine and birinapant in pancreatic cancer cells. Journal of pharmacokinetics and pharmacodynamics. 2015 Oct;42(5):477–96.

118. Tham LS, Wang L, Soo RA, Lee SC, Lee HS, Yong WP, et al. A pharmacodynamic model for the time course of tumor shrinkage by gemcitabine + carboplatin in non-small cell lung cancer patients. Clinical cancer research : an official journal of the American Association for Cancer Research. 2008 Jul 1;14(13):4213–8.

119. Wendling T, Mistry H, Ogungbenro K, Aarons L. Predicting survival of pancreatic cancer patients treated with gemcitabine using longitudinal tumour size data. Cancer chemotherapy and pharmacology. 2016 May;77(5):927–38.

AIM

Aim

The aim of this work was to develop a model-based translational framework in oncology linking in vitro and in vivo pre-clinical pharmacokinetic (PK) and tumour size (TS) information, with response outcomes obtained in clinical trials. This retrospective evaluation used the drug gemcitabine, for which data from a broad spectrum of studies were available for this thesis.

To accomplish the above described objective, pharmacometrics concepts and approaches were applied to the following cases, representatives of the different stages in drug discovery and development:

1. IN VIVO PRECLINICAL MODELLING; Develop a semi-mechanistic tumour growth model to describe the effects of gemcitabine in both pancreas (administered as a single agent) and ovarian (given in combination with carboplatin) cancer mouse xenograft models, including different tumor cell lines for both types of tumor.

2. CLINICAL MODELLING; Develop a joint TS and OS population PKPD model of gemcitabine in patients with advanced pancreatic cancer including the time course of drug exposure and its effect on TS.

3. TRANSLATIONAL MODELLING; Anticipate the clinical response to gemcitabine in patients with advanced pancreatic and ovarian cancer using preclinical data obtained from xenograft tumour-bearing mice.

4. MULTI-SCALE MODELLING, FROM IN VITRO TO IN VIVO; Build a mechanistic multi scale model for gemcitabine based on its molecular pathway, integrating in vitro- in vivo data, in order to anticipate the different rate of responses to treatment in pancreatic cancer depending on the accumulation and retention of gemcitabine active metabolite.

CHAPTER 1

Garcia-Cremades M, Pitou C, Iversen PW, Troconiz I. Characterizing gemcitabine effects administered as single agent or combined with carboplatin in mice pancreatic and ovarian cancer xenografts: a semimechanistic pharmacokinetic/pharmacodynamics tumour growth response model. The Journal of Pharmacology and Experimental Therapeutics. March 2017; 360:445–456.

Characterizing gemcitabine effects administered as single agent or combined with carboplatin in mice pancreatic and ovarian cancer xenografts:

a semimechanistic pharmacokinetic/pharmacodynamics tumour growth- response model

Maria Garcia-Cremades1,2, Celine Pitou3, Philip W. Iversen4,Iñaki F. Troconiz1,2

J Pharmacol Exp Ther 360:445–456, March 2017

DOI: https://doi.org/10.1124/jpet.116.237610

1Pharmacometrics and Systems Pharmacology, Department of Pharmacy and Pharmaceutical Technology, School of Pharmacy 2Navarra Institute for Health Research (IdiSNA), University of Navarra, Pamplona, Spain 3Global Pharmacokinetic/Pharmacodynamics and Pharmacometrics, Eli Lilly and Company Windlesham, Surrey, United Kingdom 4Lilly Research laboratories, Eli Lilly and Company, Indianapolis, Indiana

Running Tittle: Gemcitabine Preclinical PKPD Tumour Growth-Response Model

CHAPTER 2

Predicting tumour growth and its impact on survival in gemcitabine-treated patients with advanced pancreatic cancer

Maria Garcia-Cremades1,2, Celine Pitou3, Philip W. Iversen4,Iñaki F. Troconiz1,2

Manuscript in preparation

Chapter 2

ABSTRACT

AIM: The aim of this evaluation was to characterize the impact of the tumour size (TS) effects driven by the anticancer drug gemcitabine on overall survival (OS) in patients with advanced pancreatic cancer by building and validating a predictive semi-mechanistic joint TS- OS model.

PATIENTS AND METHODS: TS and OS data were obtained from one phase II and one phase III study where gemcitabine was administered (1000-1500 mg/kg over 30-60 min i.v infusion) as single agent to patients (n=285) with advanced pancreatic cancer. Drug exposure, TS, and OS were linked using the population approach with NONMEM 7.3.

RESULTS: Pancreatic tumour progression was characterized by exponential growth (doubling time=67 weeks), and tumour response to treatment was described as a function of the area under the gemcitabine triphosphate concentration vs time curve (AUC), including treatment-related resistance development. The typical predicted percentage of tumour growth inhibition with respect to no treatment was 22.3% at the end of 6 chemotherapy cycles. Emerging resistance elicited a 57% decrease in drug effects during the 6th chemotherapy cycle. Predicted TS profile was identified as main prognostic factor of OS, with tumours responders´ profiles improving median OS by 30 weeks compared to stable-disease TS profiles Results of NCT00574275 trial were predicted using this modelling framework, thereby validating the approach as a prediction tool in clinical development.

CONCLUSION: Our analyses show that despite the advanced stage of the disease in this patient population, the modelling framework herein can be used to predict the likelihood of treatment success using early clinical data.

Chapter 2

TRANSLATIONAL RELEVANCE

Pharmacokinetic/Pharmacodynamic models play an increasingly important role in the process of new drug development since it allows the link between drug exposure and progression of the disease represented by either tumour size metrics and/or circulating biomarkers to be characterized, thus permitting prediction of the patient´s clinical outcome. The current study represents a valuable contribution to the field of oncology, where the difficulties of linking early clinical results with late clinical outcomes lead to high attrition rates. In this analysis we present a joint pharmacokinetic tumour size and overall survival model for gemcitabine in patients with advanced pancreatic cancer. The model identified tumour size as the main prognostic factor of overall survival, despite the advanced stage of the disease in the patient population studied, and the results of the NCT00574275 trial were successfully predicted using the modelling framework, thereby validating the approach as a prediction tool in clinical development.

Chapter 2

1. INTRODUCTION

Currently pancreatic cancer constitutes one of the most aggressive and lethal oncology diseases (1,2). Prognosis is particularly poor with an overall 5-year survival rate of less than 5% and a median survival of 6 months when patients are not treated (3). The potential cure for these patients is limited to surgery and this is only feasible at early stages of the malignancy, when the tumour is still resectable. However, only around 10–20% of patients present with resectable disease, and, even in this situation, only around 20% survive up to 5 years (4). Over the years, limited advances have been made in the treatment of pancreatic cancer, with very few anticancer drugs found to bring any real benefit. The first cytotoxic drug that was shown to produce a meaningful impact on survival and disease-related symptoms in pancreatic adenocarcinoma was gemcitabine (5,6).

Gemcitabine is a nucleoside antimetabolite anticancer pro-drug, indicated for the treatment of multiple types of cancer (6,7). The parent compound requires cellular uptake and intracellular phosphorylation to achieve its active form (gemcitabine triphosphate) in order to exert its cytotoxic action (8).

The standard dosing schedule used in clinic (30 minutes i.v infusion of 1000 mg/m2 given weekly for 3 weeks followed by 1 week rest (7,9)) has proved to be useful for prolonging survival and alleviating the symptoms of patients with advanced pancreatic cancer (5). However, the life expectancy of such patients is still discouraging, with an overall survival of 7.2 months and an objective response rate to treatment of only 10% (10). Moreover, apart from the highly variable nature of response rates (11) an even more important limitation is that the majority of patients develop resistances to the drug (12).

Although new therapeutic strategies are emerging for pancreatic cancer(2), treatment with gemcitabine still constitutes the first line therapy, either given in combination with nab- paclitaxel for patients with ECOG status 0-1, or as a single agent for advanced patients (ECOG >1) (13),and for those patients who cannot receive the combination treatments.

The paradigm of linking drug exposure and clinical endpoints through the incorporation of bio-/surrogate markers in a quantitative framework has been successfully applied in oncology, during a decade of efforts(14).These modelling efforts have allowed to identify metrics (such as the change of tumour size at a specific time point after the start of treatment) predicting overall survival (OS) (15,16), with important application in drug development and clinical use of anti-cancer drugs.

Gemcitabine PKPD Tumour size-Survival Model

Gemcitabine effects have been described under the model-based paradigm mentioned above for non-small cell lung cancer (NSCLC), metastatic ovarian cancer and metastatic breast cancer (17–19). In all these evaluations gemcitabine was given in combination with carboplatin (17,18) and with paclitaxel(19). With respect to pancreatic cancer, a model linking tumour size (TS) with OS has been recently published (20). However, drug exposure information was not reported (i.e., dose levels, treatment duration or dose reductions or discontinuations) thus precluding the characterization of gemcitabine effects and limiting the use of the model to evaluate outcome results from model-based simulations exploring alternative dosing scenarios. In addition, ignoring drug exposure prevents analysis of whether progression in the disease occurs due to treatment interruptions or emerging resistance.

Based on the above considerations, the main objective of this evaluation was to build a joint TS and OS population pharmacokinetic/pharmacodynamics (PKPD) model of gemcitabine in patients with advanced pancreatic cancer including the time course of drug exposure and its effect on TS. During the course of this analysis the contribution of TS over time as a driver of OS was evaluated beyond statistical significance. In addition, and due to the fact that, at least to the best of our knowledge, most models linking drug exposure, TS and OS have not been externally validated, we aimed to confirm the robustness of our approach challenging the current model against results reported from other clinical studies.

Chapter 2

2. PATIENTS AND METHODS

DATA AND STUDIES DESIGN

Data were obtained from two clinical studies in phase II (study JEAL; NCT00055250) and III (study JMES; NCT00035035) in which previously untreated patients with locally advanced or metastatic pancreatic cancer were treated with gemcitabine. The primary end-point of both studies was OS. Secondary endpoints included tumour response rate, progression free survival, or time to response among others. In the JEAL study, patients were treated with the current standard treatment, receiving an i.v infusion (over 30-60 minutes) of 1000 mg/m2 of gemcitabine in a 28-day cycle of treatment (given on days 1, 8 and 15). In the JMES study, gemcitabine was given intravenously (during 30-60 minutes) at the dose of 1250 mg/m2 in a 21 day-cycle (given on days 1 and 8). Dose adjustments were based on the absolute neutrophil and platelet counts measured on the corresponding day of therapy. For both studies, the planned duration of treatment, in the absence of disease progression, was up to 6 cycles of gemcitabine. Nevertheless, patients could continue receiving cycles in case of benefit. Treatment could also be discontinued and patients dropped from the studies mainly due to (i) progressive disease, (ii) unacceptable toxicity and (iii) patient or investigator decision. Even when such exclusions occurred, TS measurement and patient status were assessed every three months until death. Details of patient characteristics at the beginning of the studies can be found in Supplementary file I.

A CT (computerized tomography) or MRI (magnetic resonance imaging) scan was routinely repeated before drug administration at every cycle, and the sum of the longest diameters for all target lesions was calculated and reported.

Written informed consent was obtained from all patients or their legally authorized representatives, prior to the performance of any protocol-specific procedures. Studies were conducted in accordance with the ethical principles outlined in the Declaration of Helsinki, consistent with good clinical practices and the applicable laws and regulations, and were approved by the institutional review board of the ethics committee at each study site.

DATA ANALYSIS

Data from both studies were pooled, and in total data from 285 patients (58 of phase II & 227 of phase III) were analysed.

Gemcitabine PKPD Tumour size-Survival Model

Pharmacokinetic, TS and OS data were linked and described based on the population approach with NONMEM 7.3 (21) using the LAPLACE estimation method with the INTERACTION option. TS data were logarithmically transformed for the analysis. Inter-subject variability (ISV) was modelled exponentially, and the residual error for TS was described with an additive error model on the logarithmic domain of the transformed data. OS information was treated as a right-hand censored time to event variable.

GEMCITABINE PHARMACOKINETICS

The weekly area under the gemcitabine triphosphate concentration vs time curve in white blood cells (AUC), was predicted for each dose administration based on the individual dosing history (including dose reductions and delays in treatment administrations) and a pharmacokinetic model of gemcitabine developed previously (9), describing the formation of gemcitabine triphosphate which occurs intracellularly following a saturable mechanism, was also used. The covariate relationships between body surface area, gender and age, and total clearance identified in the mentioned model (19) was considered in the prediction of the AUC.

TUMOUR GROWTH INHIBITION MODEL

TS comprised the sum of the longest diameters (SLD) of every lesion in every organ. New lesions were also considered and their SLD added to TS. TS was characterized as the result of a balance between the processes of tumour cell proliferation and drug-induced tumour shrinkage, and shown in equation 1:

= × − × (1)

TS proliferation was characterized by the first order rate constant Kp. The term Edrug represents drug effects and is composed of two terms, one accounting for drug efficacy of the form Slope x

AUC and the other reflecting resistance to treatment as described in equation 2, where R2 accounts for the decrease in drug effects due to exposure to treatment. TS0, the value of TS at the beginning of each of the studies, corresponds to the initial condition of the system described by equation 1.

= × (2) (1+ )

Chapter 2

The set of equations 3 and 4 quantifies the dynamics of the resistance phenomena, characterized by the first order rate constant KR.

= × ( − ) (3)

= × ( − ) (4)

The resistance model described in equations 3 and 4 implies that resistance is reversible. That and other underlying assumptions included in the model for TS were tested as part of the model building process as shown in Supplementary file II, where a summary of the key models and an overview of corresponding results are presented.

OVERALL SURVIVAL

A parametric time-to-event model was used to describe OS. Different distributions (exponential, Gompertz and Weibull) were explored to characterise the hazard rate (hz). Equation 5 represents the model corresponding to the Weibull distribution.

ℎ = × × ( × ) (5)

Where λ and β are the base and shape parameters of the Weibull distribution model. The link between hz and OS is established through the cumulative hazard (HZ) as indicated by the following expression: = (22).

MODEL SELECTION

Models were selected based on the (i) minimum value of the objective function provided by NONMEM, which is approximately equal to -2 log (likelihood) (-2LL) where a decrease in - 2LL of 3.84, 6.63 and 10.83 points between two nested models is considered significant at p values of 0.05, 0.01 and 0.001, respectively, (ii) visual exploratory analysis of the goodness of fit plots and predictive checks and (iii) meaningfulness of the parameter estimates, as well as their precision.

Gemcitabine PKPD Tumour size-Survival Model

COVARIATE ANALYSIS AND SELECTION

Once the joint model for TS inhibition and OS was developed, a covariate analysis was performed. The following patient’s characteristics measured at baseline were taken into account for inclusion in the model: Number of organs damaged with tumour lesions (NODB), number of tumour lesions (NLSB), tumour lesion location, disease stage, and ECOG status. These covariates were tested on the TS0, Kp, Slope, λ and β parameters.

Selected covariates were finally included following the general covariate model described in equation 6:

= × × (6) ,

TVP represents the typical value of a model parameter and is described as a function of

th m continuous (covm) and p categorical (covp) covariates.describes the n typical parameter value for an individual with covariate values (covm ) equal to the reference values: [(covm

=covm,ref) and (covp =0)]. covm,ref refers to the median value across the populations studied. and θ are parameters quantifying the magnitude of the covariate-parameter relationship.

Covariate selection was performed using the stepwise covariate modelling implemented in the PsN software (v4.48) (23) by means of the −2LL ratio test with signiicance levels of 0.01 and 0.001 for the forward inclusion and for backward deletion approaches, respectively.

MODEL EVALUATION

TS and OS models were evaluated through simulation-based diagnostics, performing Visual Predictive Checks (VPCs) (24,25). In addition, the precision of the parameter estimates was evaluated from the analysis of 500 simulated bootstrap datasets.

EXTERNAL MODEL VALIDATION

The model was externally validated with the TS model and the OS data published by Wendling et al (20) using VPCs. As no raw TS data over time were available, 500 TS profiles were simulated using Wendling’s model, and used as the external data for validation purposes. We then generated VPCs by simulating 500 replicate datasets with the same characteristics as those of the external dataset (i.e, dosing schedule, TS0 and covariate information) using the currently developed model. For the validation of the OS model, survival profiles obtained from

100

Chapter 2

500 hundred studies simulated with the study conditions reported by Wendling and the current model, were summarized as a Kaplan Meier plot represented by the 95% confidence intervals and compared with the raw OS shown in Wendling et al. (20).

MODEL EXPLORATION

One of the main objectives of the current evaluation was to study whether or not TS in advanced pancreatic cancer proves to be a relevant predictor of OS. To evaluate the impact of TS on OS beyond statistical significance, we performed the following simulation exercise: (i) First, TS profiles were simulated for two hundred patients receiving the standard dosing schedule of gemcitabine (i.v infusion (over 30-60 minutes) of 1000 mg/m2 of gemcitabine in a 28-day cycle of treatment (given on days 1, 8 and 15)), then (ii) three individual simulated profiles, below the 2.5th percentile, similar to the 50th percentile, and above the 97.5 percentile of the simulated distribution, representing tumour response, stable tumour profile and tumour progression, respectively, were selected; (iii) Finally, for each of the selected TS profiles, 200 OS profiles were generated and compared visually.

The Perl-speaks-Nonmem (PsN v4.48), R (version 3.2.0), Simulx (http://simulx.webpopix.org/), Berkeley Madonna (version 8.3.18), and NONMEM 7.3 software were used to perform the required calculations for the bootstrap analysis, VPC, simulation exercises and graphical representation.

101

Gemcitabine PKPD Tumour size-Survival Model

3. RESULTS

GENERAL DESCRIPTION OF THE DATA

Raw data are shown in figure 1A.

Figure 1. Raw data & schematic representation of the model. A. In the left panel, TS profiles are represented over time in a logarithmic scale, where a few individual profiles are highlighted in red. The panel in the right corresponds to the Kaplan Meier plot showing the time profile of the OS probability, highlighting the median survival time and including the number of patients at risk during time. B. The green molecular structure corresponds to gemcitabine. WBC represents the development and accumulation of gemcitabine triphosphate in white blast cells. A description of the rest of terms and model parameters can be found in the Patients and Methods section.

When looking at the whole range of TS observations, it is difficult to observe a general trend in the data. However, when tumour profiles are observed individually, a slight response to

102

Chapter 2

the treatment, followed by tumour relapse during the treatment period can be observed. Some examples of these profiles are highlighted in red and displayed in figure 1A (left panel). In total, 904 TS measurements were analysed. 51% of the patients already presented metastasis at baseline in one or more organs, mostly in the liver (54%). Regarding OS, patients died during the course of study (figure 1A, right panel), with the median survival being about 33 weeks.

JOINT TUMOUR GROWTH INHIBITION AND SURVIVAL MODEL

The schematic representation of the selected model for the tumour shrinkage effects of gemcitabine and its impact on OS is represented in figure 1B. An exponential growth model best described the natural progression of the disease. The model predicted a time of a 100% increase in TS with respect to baseline of 67 weeks. Tumour response to treatment was best described (p<0.001) by the predicted AUC of gemcitabine triphosphate compared to the results using AUC corresponding to the parent drug. The individual predicted TS profiles shown in figure 2A represent at least three different behaviours: (i) responders, (ii) non-responders, and (iii) initial response to treatment followed by a relapse, which was well captured by the resistance term incorporated into drug effects.

As described in the methods section, the probability of OS in advanced pancreatic cancer was well described by a parametric survival model using the Weibull distribution. The impact of TS profile on OS was evaluated and proved significant (p<0.001). Equation 7 describes the model for the hazard rate selected where the term ×() describes the change in hz elicited by TS.

ℎ = × × ( × ) × ×() (7)

Among the covariates tested, NLSB and NODB were found to have a significant impact (P

< 0.001) on TS0. With respect to OS, ECOG was the only covariate that showed statistical significance (P < 0.001) on the parameter.

Table I lists the estimates of model parameters associated with their measures of precision corresponding to the final joint TS and OS model. With respect to parameter precision, the 95% confidence intervals (computed from the bootstrap analysis) of any of the parameters include the value of zero, indicating that the data analysed supported the degree of complexity of the model selected. A moderate 40% ISV in TS0 was obtained. Estimates of ISV corresponding to Kp and Slope were higher (87% and 86%, respectively).

103

Gemcitabine PKPD Tumour size-Survival Model

Table I. Population Parameters Estimates

Parameters Estimate (2.5-97.5th) TS model

() = × = 6.04 (5.65-6.66) [1 + × ( − 2)] × = 0.299(0.208-0.363) [1 + × ( − 2)] = 0.123(0.06-0.23)

Kp (wk-1) 0.0103 (0.003-0.0104) Slope (wk x AUC)-1* 0.00058 (0.00014-0.0067)

-1 KR (wk ) 0.0162(0.013-0.129)

ISV_TS0 (%) 40 (33-48) ISV_Kp (%) 87(82-176) ISV_Slope (%) 86(80-202) Residual error (log(cm)) 0.276 (0.191-0.332) OS model 0.0126 (0.009-0.015) -1 = 1.63 (1.5-2.02) (wk ) = × () = 1 () = 0.433 (0.112-0.652)

γ (log(cm)-1) 0.618 (0.419-1.02)

Parameters are listed as estimates with 95% confidence intervals obtained from 500 bootstrap analyses in parenthesis. Estimates of inter-subject variability (ISV) are shown as coefficients of variation. θNLSB,

θNODB, θECOG ,are parameters quantifying the covariate effects of the NLSB and NODB on TS0, and the ECOG status on the parameter .

*Units of AUC of gemcitabine metabolite in white blood cells (wk·picomol/106 cells).

The selected model showed good performance at the individual level as can be seen in figure 2A, where several individual TS profiles are displayed with their corresponding model predictions. Figures 2B and 2C show the VPCs corresponding to TS and OS, respectively, demonstrating good agreement between observed and simulated data. For TS, the dispersion at times greater than 30 weeks appears to be over-predicted. However it should be taken into consideration that at those later times, the number of patients remaining in the study decreased considerably.

104

Chapter 2

B C

Figure 2. Model evaluation. A. Individual TS observations (green points) and model predictions (red lines) in logarithmic scale vs time from a set of patients selected at random. Vertical grey areas represent duration of treatment. B. Visual predictive checks corresponding to the selected TS model. Dots represent the TS observations; the solid green line corresponds to the 50th percentile of the observations while the dashed green lines represent the 5th and 95th percentiles of these observations. Shaded green and grey areas are the 95% predicted intervals for the median, 5th and 95th percentiles obtained from 500 simulated datasets. C. Kaplan Meier plot of OS probability. The solid blue line represents raw data while the blue shaded area cover the 95% prediction interval calculated from the 500 simulated studies.

The results from the external validation of the model are shown in figure 3. Remarkably, the current model describes quite well the simulated data based on the model structure and parameters provided by Wendling et al (20) for the case of both response outcomes, TS and OS.

The design and patient population characteristics between training and validation clinical trials are shown in Supplementary file I.

105

Gemcitabine PKPD Tumour size-Survival Model

Figure3. External model validation. Left panel. Points represent the individual TS profiles associated with the external dataset summarized as 5th, 50th, and 95th percentiles (lines in orange). The area in green represents the 90% prediction interval of the 50th percentile of the simulated TS profiles using the current model. Right panel. Kaplan Meier plot of OS probability corresponding to the validation dataset (line in orange), and the 95% prediction interval calculated from the 500 simulated studies (blue shaded area).

Figure 4 displays the contribution of TS as a predictor of OS. The results from the simulation exercise indicate there is no overlap in the 95% prediction intervals in OS between patients exhibiting disease progression and those showing stable disease and a partial response. However, for partial and non-responders, although the graph shows a clear distinction between the median profiles, the corresponding 95% prediction intervals overlap slightly at initial times.

106

Chapter 2

Figure 4. Impact of TS profiles on the OS probability. Left panel. TS simulated profiles given the standard gemcitabine treatment schedule. Thin grey lines represent 200 TS simulated profiles, while thick green, orange and blue profiles represent disease progression, stable disease and response to treatment, respectively. Right panel. Kaplan Meier plots [lines (median); shaded areas (95% prediction intervals)] associated to TS profiles shown in left panel.

Figure 5A shows the impact of the selected covariates, ECOG, NODB and NLSB on the OS through deterministic simulations. The results displayed suggest no major impact of the above listed covariates. In figure 5B,C the effect of resistance on TS profiles and Edrug is explored, respectively, calculating different metrics (see below) derived from the TS profile of an untreated patient and two patients receiving six chemotherapy cycles of the standard treatment of gemcitabine (1000 mg/m2 iv, 30 minutes infusion, administered weekly x3 followed by a week of rest from treatment) incorporating or not resistance into treatment. Over 6 chemotherapy cycles, tolerance development elicits a maximal tumour regression ((TS0-

TS(t))/TS0) of 3.33% and maximal tumour growth inhibition ((TSuntreated(t)-TS(t))/Tuntreated(t)) of 22.3%, compared to the 10.28% and 29.9% produced in absence of tolerance, respectively. Thus, emerging resistance decreases Edrug on each drug administration, reaching a 50% decrease during the 5th chemotherapy cycle.

107

Gemcitabine PKPD Tumour size-Survival Model

Figure 5. Model exploration. A. OS probability profiles as a function of NLSB (number of tumour lesions at baseline), NODB (number of organs damaged with tumour lesions at baseline) and ECOG. B. Typical predicted tumor size for an untreated patient (points), and for two patients receiving 6 cycles of the standard treatment of gemcitabine (30 min i.v infusion of 1000 mg/m2 weekly x3 followed by a week of rest from treatment) considering (solid line) or not (dashed line) the development of resistance. Each vertical area corresponds to a cycle of treatment. C. Represents the percentage of change in Edrug (see equation 2 in the text), at the end of each chemotherapy cycle.

108

Chapter 2

4. DISCUSSION

The paradigm represented by sequentially linking drug exposure, bio-, and/or surrogate marker(s), and clinical response (progression free survival, OS) has been applied successfully over the last decade to optimize the development of new anticancer agents and improve patient care in the clinical setting. Starting from the work carried out in NSCLC (16), models using TS as a predictor variable of clinical response have been developed for various type of tumours such as colorectal (15), thyroid (26), breast (27) and ovarian (18) cancer, providing oncologists with quantitative tools capable of predicting disease progression using early clinical data, enabling go/no-go decision making before Phase 3 trials are initiated.

It is noteworthy that the examples cited above, except for NSCLC, correspond to tumour types associated with median-high 5 year relative survival rates which range between 46 and 98% (28). However, for tumours with even poorer prognoses, such as advanced pancreatic cancer, the potential for these predictive modelling frameworks is yet to be fully understood.

Our study focuses on the development of a predictive model for the effects of gemcitabine administered as single agent to patients with advanced pancreatic cancer. Pancreatic cancer is reported to have the lowest 5-year relative survival rate (5%) in cancer diseases(28), with a median survival time of 6 months for advanced disease(29), values that are in accordance with our data (figure 1A, right panel).

From a methodological point of view, model development in the current context is associated with several complexities and challenges that have been fully taken into account in the current evaluation. For example: (i)The continuous (TS dynamics) and the non-continuous time-to-event (OS) responses were analysed simultaneously due to the informative characteristics of the drop-outs (30), and (ii) to achieve the objective of developing a predictive tool, the model for TS should reflect the main processes affecting tumour dynamics (drug effects, progression of the disease, and possibly tolerance/resistance to treatment). The proposed TS model indicates exponential growth in the absence of treatment or for patients not responding to gemcitabine, assuming the ideal conditions in which tumour cells have sufficient nutrients to undergo continuous proliferation (31). The model was able to characterize the data well, providing satisfactory results in terms of individual and population data (figure 2A). Resistance to treatment, far from being just a gemcitabine treatment-related phenomenon (12,32), has been well documented in oncology (33). Of note, for the case of metastatic breast cancer patients treated with paclitaxel/gemcitabine combination therapy, emerging resistance occurs faster, with the mean tumour shrinkage rate dropping by half after 5.7 weeks (19) compared to

109

Gemcitabine PKPD Tumour size-Survival Model the 20 weeks (figure 5B,C) for the current analysis. In contrast, for advanced and/or metastatic colorectal cancer patients treated with capecitabine or FU, the tumour shrinkage rate decrease by half in 13 and 16 weeks (15), respectively, slightly faster but similar to our case.

Whereas the use of a TS-derived metric [change from baseline predicted at a fixed time point (i.e, 8 weeks after the start of the treatment (16))] has been used to predict OS with the advantage of its simplicity, in our analysis the full predicted time profile of TS was used, which allows a more realistic characterization of the hazard profile to be obtained. Results from simulation exercise and represented graphically in figures 4 and 5 (top panel), reveal that TS exerted an impact on OS far beyond the statistical significant improvement in the fit. The impact of the NLSB and NODB, and ECOG status is much more reduced as can be noted in figure 5A. For example, the median predicted survival time for ECOG 0 and ≥1 in the case of two organs affected showing two lesions was 35 and 32 weeks, respectively. Neither tumour location, nor development of metastasis showed statistical significance on OS. Our results could be explained by the fact that patients from our analysis already presented advanced and/or metastatic disease at the time of being enrolled in the trial.

Recently a model relating TS with OS but with the limitation of not considering drug exposure as a predictor of TS dynamics was established for gemcitabine effects on pancreatic cancer (20). We made use of the reported parameters to recreate virtual patients and externally validate our model (see figure 3). We emphasize the general need of integrating drug exposure for predictive population PKPD models, and in particular for the case of gemcitabine since clinical results suggest that the anti-tumour effects of gemcitabine are schedule dependent (11).

One pending issue that could not be addressed in the current analysis was the identification of patient-related factors responsible for the high variability associated with the rate of tumour proliferation and drug effects (87 and 86%, respectively). Several studies suggest that the different rates of responses associated with gemcitabine treatment could in part be explained by individual genetic factors affecting, among other processes, the gemcitabine metabolism pathway (34), leading to different amounts of the active metabolite.

In conclusion, this joint modelling exercise predicts the efficacy of gemcitabine in terms of tumour growth inhibition and OS of patients with advanced pancreatic cancer. The analysis is expected to have an impact on the development of new anticancer drugs as our results indicate that phase II and III data share the same model structure and model parameter estimates. Therefore, the current modelling framework could be used for predicting late clinical outcome from early phase II data, optimizing the design of late phase clinical trials for advanced

110

Chapter 2

pancreatic cancer. It will also serve to optimize the standard treatment of pancreatic cancer patients receiving the drug of interest, predicting the likelihood of treatment success and assisting with the selection of dosing regimens.

Despite of the advanced stage of the disease, this analysis has identified a significant impact of TS on clinical response. This result encourages the development of a translational approach (35) based on recent semi-mechanistic tumour growth inhibition models developed in mice pancreatic and ovarian xenografts (36).

111

Chapter 2

5. REFERENCES

1. Schober M, Javed MA, Beyer G, Le N, Vinci A, Sund M, et al. New Advances in the Treatment of Metastatic Pancreatic Cancer. Digestion. 2015;92(3):175–84.

2. Perkhofer L, Ettrich TJ, Seufferlein T. Pancreatic Cancer: Progress in Systemic Therapy. Gastrointest tumors. 2014 May;1(4):167–79.

3. Le N, Sund M, Vinci A. Prognostic and predictive markers in pancreatic adenocarcinoma. Dig Liver Dis. 2016 Mar;48(3):223–30.

4. Neoptolemos JP, Stocken DD, Friess H, Bassi C, Dunn JA, Hickey H, et al. A randomized trial of chemoradiotherapy and chemotherapy after resection of pancreatic cancer. N Engl J Med. 2004 Mar 18;350(12):1200–10.

5. Moore M. Activity of gemcitabine in patients with advanced pancreatic carcinoma. A review. Cancer. 1996 Aug 1;78(3 Suppl):633–8.

6. Oettle H, Arnold D, Hempel C, Riess H. The role of gemcitabine alone and in combination in the treatment of pancreatic cancer. Anticancer Drugs. 2000 Nov;11(10):771–86.

7. Toschi L, Finocchiaro G, Bartolini S, Gioia V, Cappuzzo F. Role of gemcitabine in cancer therapy. Future Oncol. 2005 Feb;1(1):7–17.

8. Mini E, Nobili S, Caciagli B, Landini I, Mazzei T. Cellular pharmacology of gemcitabine. Ann Oncol [Internet]. 2006 May 1;17(suppl 5):v7–12.

9. Zhang L, Sinha V, Forgue S, Callies S, Ni L, Peck R, et al. Model-Based Drug Development: The Road to Quantitative Pharmacology. J Pharmacokinet Pharmacodyn [Internet]. 2006 Jul;33(3):369.

10. Fuchs CS, Azevedo S, Okusaka T, Van Laethem J-L, Lipton LR, Riess H, et al. A phase 3 randomized, double-blind, placebo-controlled trial of ganitumab or placebo in combination with gemcitabine as first-line therapy for metastatic adenocarcinoma of the pancreas: the GAMMA trial. Ann Oncol Off J Eur Soc Med Oncol. 2015 May;26(5):921–7.

11. Ciccolini J, Serdjebi C, Peters GJ, Giovannetti E. Pharmacokinetics and pharmacogenetics of Gemcitabine as a mainstay in adult and pediatric oncology: an EORTC- PAMM perspective. Cancer Chemother Pharmacol. 2016 Jul;78(1):1–12.

113

Gemcitabine PKPD Tumour size-Survival Model

12. Bergman AM, Pinedo HM, Peters GJ. Determinants of resistance to 2′,2′- difluorodeoxycytidine (gemcitabine). Drug Resist Updat [Internet]. 2002;5(1):19–33.

13. Ruess DA, Görgülü K, Wörmann SM, Algül H. Pharmacotherapeutic Management of Pancreatic Ductal Adenocarcinoma: Current and Emerging Concepts. Drugs {&} Aging [Internet]. 2017;1–27.

14. Bender BC, Schindler E, Friberg LE. Population pharmacokinetic- pharmacodynamic modelling in oncology: a tool for predicting clinical response. Br J Clin Pharmacol. 2015 Jan;79(1):56–71.

15. Claret L, Girard P, Hoff PM, Van Cutsem E, Zuideveld KP, Jorga K, et al. Model- Based Prediction of Phase III Overall Survival in Colorectal Cancer on the Basis of Phase II Tumor Dynamics. J Clin Oncol [Internet]. 2009;27(25):4103–8.

16. Wang Y, Sung C, Dartois C, Ramchandani R, Booth BP, Rock E, et al. Elucidation of relationship between tumor size and survival in non-small-cell lung cancer patients can aid early decision making in clinical drug development. Clin Pharmacol Ther. 2009 Aug;86(2):167– 74.

17. Tham LS, Wang L, Soo RA, Lee SC, Lee HS, Yong WP, et al. A pharmacodynamic model for the time course of tumor shrinkage by gemcitabine + carboplatin in non-small cell lung cancer patients. Clin Cancer Res. 2008 Jul 1;14(13):4213–8.

18. Zecchin C, Gueorguieva I, Enas NH, Friberg LE. Models for change in tumour size, appearance of new lesions and survival probability in patients with advanced epithelial ovarian cancer. Br J Clin Pharmacol. 2016 Sep;82(3):717–27.

19. Tate SC, Andre V, Enas N, Ribba B, Gueorguieva I. Early change in tumour size predicts overall survival in patients with first-line metastatic breast cancer. Eur J Cancer. 2016 Oct;66:95–103.

20. Wendling T, Mistry H, Ogungbenro K, Aarons L. Predicting survival of pancreatic cancer patients treated with gemcitabine using longitudinal tumour size data. Cancer Chemother Pharmacol. 2016 May;77(5):927–38.

21. Bauer R. NONMEM users guide: introduction to NONMEM 7.2.0. ICON Development Solutions: Ellicott City; 2011.

22. Collett D. Modelling survival data in medical research. 1952;

114

Chapter 2

23. Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN)—a Perl module for NONMEM related programming. Comput Methods Programs Biomed [Internet]. 2004;75(2):85–94.

24. Bergstrand M, Hooker AC, Wallin JE, Karlsson MO. Prediction-corrected visual predictive checks for diagnosing nonlinear mixed-effects models. AAPS J. 2011 Jun;13(2):143– 51.

25. Nyberg J, Karlsson KE, Jönsson S, Simonsson USH, Karlsson MO, Hooker AC. Simulating large time-to-event trials in NONMEM.PAGE 23 (2014) Abstr 3166 [www.page- meeting.org/?abstract=3166]. 2014.

26. Claret L, Lu JF, Sun YN, Bruno R. Development of a modeling framework to simulate efficacy endpoints for motesanib in patients with thyroid cancer. Cancer Chemother Pharmacol. 2010 Nov;66(6):1141–9.

27. Bruno R, Lindbom L, Schaedeli Stark F, Chanu P, Gilberg F, Frey N, et al. Simulations to Assess Phase II Noninferiority Trials of Different Doses of Capecitabine in Combination With Docetaxel for Metastatic Breast Cancer. CPT pharmacometrics Syst Pharmacol. 2012 Dec 26;1:e19.

28. Miller KD, Siegel RL, Lin CC, Mariotto AB, Kramer JL, Rowland JH, et al. Cancer treatment and survivorship statistics, 2016. CA Cancer J Clin. 2016 Jul;66(4):271–89.

29. Muniraj T, Jamidar PA, Aslanian HR. Pancreatic cancer: a comprehensive review and update. Dis Mon. 2013 Nov;59(11):368–402.

30. Hu C, Sale ME. A joint model for nonlinear longitudinal data with informative dropout. J Pharmacokinet Pharmacodyn. 2003 Feb;30(1):83–103.

31. Ribba B, Holford NH, Magni P, Troconiz I, Gueorguieva I, Girard P, et al. A review of mixed-effects models of tumor growth and effects of anticancer drug treatment used in population analysis. CPT pharmacometrics Syst Pharmacol. 2014 May 7;3:e113.

32. Binenbaum Y, Na’ara S, Gil Z. Gemcitabine resistance in pancreatic ductal adenocarcinoma. Drug Resist Updat. 2015 Nov;23:55–68.

33. Longley DB, Johnston PG. Molecular mechanisms of drug resistance. J Pathol. 2005 Jan;205(2):275–92.

115

Gemcitabine PKPD Tumour size-Survival Model

34. de Sousa Cavalcante L, Monteiro G. Gemcitabine: metabolism and molecular mechanisms of action, sensitivity and chemoresistance in pancreatic cancer. Eur J Pharmacol. 2014 Oct 15;741:8–16.

35. Garcia-Cremades M, Pitou C, Iversen W P, Troconiz F I. A comparison of different model-based approaches to scale preclinical to clinical tumour growth inhibition in gemcitabine-treated pancreatic cancer. PAGE 25 (2016) Abstr 5704 [www.page- meeting.org/?abstract=5704]. 2016.

36. Garcia-Cremades M, Pitou C, Iversen PW, Troconiz IF. Characterizing Gemcitabine Effects Administered as Single Agent or Combined with Carboplatin in Mice Pancreatic and Ovarian Cancer Xenografts: A Semimechanistic Pharmacokinetic/Pharmacodynamics Tumor Growth-Response Model. J Pharmacol Exp Ther. 2017 Mar;360(3):445–56.

116

Chapter 2

6. SUPPLEMENTARY MATERIAL

Supplementary File I. Summary of design characteristics of the clinical trials used either for model development or external validation.

The overall survival data presented in Wendling et al1, and used for validation purposes, correspond to the control arm of a Phase III study2, in which patients with advanced metastatic pancreatic cancer, were treated with either gemcitabine + aflibercept or gemcitabine plus placebo (treatment arm analyzed by Wendling et al-).

MODEL DEVELOPMENT DATASET VALIDATION Study JEAL JMES DATASET Phase II III III Multicenter randomized, Multicenter randomized, Multicenter randomized, Design double blinded open label double blinded Disease Locally advanced or metastatic pancreatic cancer Gemcitabine i.v infusion Gemcitabine i.v infusion Gemcitabine i.v infusion Dose (30-60 min) 1000 mg/m2 (30-60 min) 1250 mg/m2 (30-60 min) 1000 mg/m2 Dosing Schedule 28 days/cycle. 21 days/cycle. 28 days/cycle. Cycle Given once a week x3 + Given once a week x2 + 1week Given once a week x3 + 1week rest rest 1week rest Duration 33 months 28 months 21 months Patients 58 227 275 Female 59% 46.50% 43% Gender Male 41% 53.50% 57% Median 62 63 61 Age Range 34-85 28-82 34-86 Caucasian 92% 89.70% N.R Origin Other 8% 10.30% N.R 0 32% 31.60% 37% ECOG PS 1 68% 68.40% 63% II 2% N.R 6% Disease Stage III 9% N.R 4% IV 89% 91.10% 90% 1 49% 42% NODB 2 51% 58% NLSB 1 38% N.R 2 62% N.R

ECOG=Eastern Cooperative Oncology Group; PS=Performance Status; NODB=Number of organs damaged with tumour lesions at baseline; NLSB=Number of tumour lesions at baseline; N.R=Not Reported

117

Gemcitabine PKPD Tumour size-Survival Model

REFERENCES SUPPLEMENTARY FILE I

Wendling T, Mistry H, Ogungbenro K, Aarons L. (2016) Predicting survival of pancreatic cancer patients treated with gemcitabine using longitudinal tumour size data. Cancer Chemother Pharmacol (2016) 77:927–938

Rougier P, Riess H, Manges R, Karasek P, Humblet Y, Barone C, Santoro A, Assadourian S, Hatteville L, Philip PA (2013) Randomised, placebo-controlled, double-blind, parallel-group phaseIII study evaluating aflibercept in patients receiving first-line treatment with gemcitabine for metastatic pancreatic cancer. Eur J Cancer 49(12):2633–2642

118

Chapter 2

Supplementary File II. Summary of model development based on -2LL values

Model Description -2LL

OS Model Exponential 2652.019 Gompertz 2576.734 Weibull 2523.067 TS + OS Joint Exponential tumour growth Model Drug effect: AUC metabolite 2009.633 Survival weibull. Exponential tumour growth Drug effect: AUC metabolite +resistance 1989.065 Survival weibull.

Exponential tumour growth Drug effect: AUC metabolite+resistance 1975.909 Survival weibull + link TS0

Exponential tumour growth Drug effect: AUC metabolite+resistance 1945.23 Survival weibull + link predicted TS

Covariate analysis Number of lesions on TS0 1778.758

Number of organ damaged on TS0 1751.865

Patient ECOG status on β 1733.784

119

CHAPTER 3

Translational framework predicting tumour response and survival in gemcitabine-treated patients with advance pancreatic and ovarian cancer from xenograft studies.

Maria Garcia-Cremades1,2, Celine Pitou3, Philip W. Iversen4,Iñaki F. Troconiz1,2

Manuscript in preparation

Chapter 3

ABSTRACT

AIM: The aim of this evaluation was to predict the clinical response to gemcitabine in patients using pre-clinical data obtained from xenograft tumour-bearing mice.

METHODS: The approach consisted of building a translational model combining preclinical pharmacokinetic/pharmacodynamic (PKPD) models and parameters, with dosing paradigms used in the clinics along with clinical human PK models to derive tumour profiles in humans driving overall survival. First, simulations of tumour growth inhibition (TGI) were performed using the (i)drug effect parameters obtained from the mice, (ii)system parameters obtained from mice after appropriate scaling, (iii)patient PK models for gemcitabine and carboplatin, and (iv)the corresponding standard dosing schedules given in the clinical scenario for both type of cancers. Tumour profiles in mice were scaled up by body weight to their equivalent values in humans. Then, as models for survival showed that tumour size was the main driver of the hazard rate, it was therefore possible to predict overall survival in pancreatic and ovarian cancer patients.

RESULTS: Tumour dynamics in both pancreatic and ovarian cancer patients were well captured, given the described translational modelling approach. Furthermore, calculated metrics showed values (%Tumour regression [0-17] and change of tumour size at week 12 [-9,- 4.5]) in the range of those predicted with the PKPD model developed with clinical data. Survival simulations for the six cell lines captured quite well the clinical outcome, at least until the end of the treatment period.

CONCLUSION: The Model-informed Drug Discovery and Development paradigm has been successfully applied retrospectively to gemcitabine data, through a mechanistic translational approach, describing the time-course of the tumour response and patient survival probability (the gold standard endpoint in oncology drug development) from pre-clinical data.

125

Chapter 3

TRANSLATIONAL RELEVANCE

Currently, preclinical modelling is often used in drug research and development for making go/no go decision according to qualitatively relationships (i.e Treated/Control=40%). However, high attrition rates in the field of oncology (95%) encourage the research in new strategies that provide a better understanding of drug efficacy from early development phases and support the translation from preclinical to clinical trials. In this work, a quantitative translational model-informed approach is proposed. The developed pharmacometric-based framework has successfully applied retrospectively to gemcitabine data linking pre-clinical and clinical response through a mechanistic translational approach, describing the time-course of the tumour response and patient survival probability (the gold standard endpoint in oncology drug development) from pre-clinical data. This analysis would help predict drug efficacy in humans from early development phases, assisting and optimizing the drug development and discovery process in oncology.

127

Chapter 3

1. INTRODUCTION

The high attrition rates in oncology drug development (95%) not only increases the cost of research but also represents a delay in patients receiving safe, effective medications in a timely manner (1–3).To address this problem, scientists from both academia and industry are making great efforts to search for new strategies that would allow the success of a drug to be predicted from the early pre-clinical stages (4). In 2006, the FDA published its Critical Path document where the use of pharmacokinetic and pharmacodynamic (PKPD) models were recognized as a relevant tool in the drug development process, due to their capacity to characterize drug effects and disease progression over time and to make predictions under different scenarios (4–6).

Briefly, PKPD modelling integrates the dose-concentration (PK) and the concentration effect relationships (PKPD) of a drug, enabling the description of the time-course response to a pharmacological treatment (7). Those PKPD models have evolved from being empirical to more mechanistic, incorporating drug and biological system-specific properties to their mathematical expressions (8). The development of these mechanism-based PKPD models facilitates the establishment of the relationship of model parameters with biological processes, providing a better understanding of drug efficacy and safety (9), and supporting the translation from preclinical studies to clinical trials (10).

The pre-clinical phase of development is critical in the characterization of drug effects and their correlation with clinical data. One of the most common pre-clinical experiments in oncology consists of using human tumour xenografts to describe anti-tumour drug efficacy (11). Although the predictive capacity of xenograft studies has been widely discussed (12,13), recent publications highlight the benefits of applying the PKPD modelling framework to data gathered from these types of studies at the time of anticipating drug response in humans. As an example, Wong and colleagues (14) established the relationship between tumour growth inhibition (TGI) predicted using mouse derived PKPD parameters, dosing schedules used in the clinic together with human PK, and clinical response. Additionally, Rochetti et al (15) predicted active doses in humans from pre-clinical PKPD models developed for animal TGI studies. One common characteristic of these previous examples is the use of pre-clinical and clinical data of marketed drugs to develop the translational links.

Gemcitabine (difluorodeoxycytidine; dFdC), a cytotoxic/cytostatic antimetabolite analog of nucleoside, is a pro-drug that has to be intracellularly metabolized to its active form (dFdCTP) to exert its action. Once metabolized, its main mechanism of action consists of its incorporation

129

Gemcitabine PKPD Translational Model-Based Approach into replicated DNA inducing the inhibition of cell-growth and causing apoptosis (16–18). Gemcitabine is indicated for the treatment of several solid tumours, mainly given as single agent in the treatment of locally advanced or metastatic pancreatic cancer, or in combination in non- small-cell lung cancer (NSCLC), breast and ovarian cancer (16).

The recent development of pre-clinical (19) and clinical gemcitabine PKPD models for advanced pancreatic (20) and ovarian (21) cancer opens up the possibility of establishing a translational modelling approach, which in fact represents the goal of this study, that is to predict clinical response to gemcitabine in patients using pre-clinical data obtained from xenograft tumour-bearing mice. The current evaluation aims to provide a general methodology of analysis and data interpretation beyond the specific case of gemcitabine.

130

Chapter 3

2. METHODS

Figure 1 provides a high-level overview of the translational approach followed in the current evaluation, which combines pre-clinical PKPD models with clinical dosing paradigms and PK models to predict tumour growth inhibition and thus overall survival (OS) for patients with advanced pancreatic cancer.

Figure 1. Schematic representation of the translational model-based approach strategy used in the current analysis. PKPD: Pharmacokinetic/Pharmacodynamic. TGI: Tumour Growth Inhibition. BW: body weight (65 Kg; human, 0.025 Kg; mouse). TV: Tumour volume. TS: Tumour size.

MODELS AND DATA

The pre-clinical PKPD characteristics of gemcitabine administered as a single agent or in combination with carboplatin in pancreatic or ovarian cancer xenografts, respectively, were extracted from a recently published PKPD model (19). The PKPD properties of gemcitabine in patients with ovarian or advanced pancreatic cancer were obtained from literature data, including the those reported in the companion paper to the present work (20,21).

Supplementary material 1 provides a brief description of the PKPD models used in this evaluation.

Raw tumour size (TS) and OS data used to develop the PKPD models in humans (20,21) were also available and served to evaluate the accuracy of the predictions obtained from the translational approach.

131

Gemcitabine PKPD Translational Model-Based Approach

TUMOR VOLUME PREDICTION IN HUMANS

First, gemcitabine (parent drug) and carboplatin concentrations in plasma were simulated based on previously developed clinical PK models (4,22) using the dosing schedules indicated below.

The dosing schedules used in both pancreatic and ovarian TGI simulations were the standard schedules recommendation for use in the clinic (23). (i) In the case of pancreatic cancer, a 30-minute infusion of 1000 mg/m2 of gemcitabine was given as a single agent once a week for 3 weeks for each 28 day cycle. (ii) For ovarian cancer, a 30-minute infusion of 1000 mg/m2 of gemcitabine was administered every week for 2 weeks of a 21 day cycle, in combination with carboplatin which was administered on the first day of each cycle at a dose corresponding to a target area under the curve (AUC) of 4 min.mg/mL. The treatment duration consisted of 6 cycles for both cancer types.

Second, TS profiles were predicted by coupling the simulated clinical exposure profiles to the pre-clinical PKPD models where the first order rate constants λ1 and B, and the parameter D (see appendix I) were scaled up to human equivalents as described in equation 1 (24):

. = × (1)

Where  represents each of the above mentioned first order rate constants, and BW is the body weight assumed to be 65 Kg and 25g, in an adult human and mouse, respectively.

To obtain the initial conditions for tumour volume (TV0) and the carrying capacity (K0) to be used during simulations, first the baseline TS (TS0) estimates obtained in the analysis of clinical data [61.7 mm, pancreatic (20); 69.7 mm, ovarian (21) cancer] were transformed to tumour volume (TV; mm3), under the assumption of a spherical tumour mass following the

equation = × (25) (TV0pancreas=113324 mm3; TV0ovarian=177295 mm3). Then, these

TV0 for both type of tumours were scaled down to mouse by multiplying them by the ratio of mouse-human BW. Finally, K0 were set to the values estimated during the pre-clinical PKPD analysis (K0mouse) (19) corrected for the difference between TV0 and the values reported in the

pre-clinical PKPD study (TV0mouse) (19) = × . Hence, while K0 show different values between cell lines, TV0 share the same values between cell lines of each type of cancer.

132

Chapter 3

Six TV profiles were generated in total, one for each of the tumour cell lines included in the pre-clinical analysis (19): KP4, ASPC1, MIAPACA2, and PANC1 (pancreas) and A2780, and SKOVxluc3 (ovarian) tumour cell lines.

Once simulations were performed, TV profiles were scaled up to human by multiplying by the ratio of human-mouse BW. These humanised TV values were then transformed into TS

(mm) following the equation: ()= × () to compare with raw data obtained in the clinical studies.

Different metrics [percentage of tumour growth inhibition (TGI) at the end of treatment, maximal tumour regression (TR) during treatment (in the case of pancreas tumour cell lines), or change of TS at week 12 (TSRwk12) (in the case of ovarian tumour cell lines)] were calculated using the equations indicated below:

= _ × 100 (2) _

= × 100 (3)

() () = × 100 (4)

Where TSend and TSControl_end represent the simulated TS value at the end of treatment of the TS profile, and the TS profile simulated without treatment, respectively.

TS simulations were performed using the R (version 3.2.0) package Simulx (http://simulx.webpopix.org/).

SURVIVAL SIMULATIONS

OS profiles were simulated for each of the TS profiles generated as explained in the preceding section using the published OS models (20,21) in which either the entire TS profile or

TSRwk12 were identified as main predictor factors of OS in patients with advanced pancreatic or ovarian cancer, respectively (20, 21). Supplementary material 1 shows the model structure of the OS models used in this evaluation for both type of cancers.

In the case of ovarian cancer, the survival model depends on TSR at each time until week 12 and TSR at week 12 for longer times. It also depends on the appearance of new lesions

133

Gemcitabine PKPD Translational Model-Based Approach during treatment (NewLes), on TS at baseline (TS0) and on ECOG status at enrolment (see supplementary material 1). To perform the simulations, these covariates were set to values similar to those found in the available clinical datasets.

Survival simulations were performed with NONMEM 7.3(26).

134

Chapter 3

3. RESULTS

PANCREATIC CANCER

The translational modelling approach for gemcitabine in pancreatic cancer was performed for the four pancreatic cell lines characterized in the mouse xenografts: KP4, ASPC1,

MIAPACA2, and PANC1. TV0 was 43.59 mm3 while K0 calculated showed values of 0.0056 mm3 (KP4), 22.52 mm3 (ASCP1), 289.22 mm3 (MIAPACA2) and 1.3x10-5 mm3 (PANC1) (see Methods). The left panel in Figure 2A shows that there is an agreement between typical tumour predictions obtained from the model developed with clinical data, and raw patient data, and those obtained based on the proposed simulation framework. Remarkably, the agreement is higher for the case of the PANC1 cell line. It also worth noting that, as expected, variability is greater in patients than in simulated individuals using human population PK parameters and mouse derived PD parameters. In the upper panels of figure 2B, simulations are shown for each cell line separately.

Figure 2. Translational TS simulated profiles. A. TS simulations performed for pancreas (left panel) and ovarian (right panel) cancer. Grey points represent TS raw clinical data for both cancers, while black line

135

Gemcitabine PKPD Translational Model-Based Approach is the simulated typical TS profile performed with the clinical PKPD models for each cancer, following the same dosing schedules as in the translational simulations. Coloured shaded areas represent the simulated translational TS profiles for each cell line. B. Translational TS simulations for each pancreatic (top panels) and ovarian (bottom panels) cell lines.

ASPC1 and especially KP4 cell lines appear to be the most sensitive at early times after the initiation of the treatment, showing faster subsequent disease progression. PANC1 and MIAPACA2 appear to be the most and least sensitive cell lines during the whole treatment period, respectively.

The simulated profiles shown in figure 2A (left) and 2B (upper) were translated into the clinical context calculating the metrics of tumour regression and tumour growth inhibition at the end of the treatment. In figure 3A, TR showed median values of 10 (overall), 17 (KP4), 5 (ASPC1), 0 (MIAPACA2), 14% (PANC1) that are within the range of values predicted with the PKPD model developed with the clinical data (4%). TGI values showed greater differences across cell lines: 77, 75, 46, and 14 % for KP4, PANC1, ASPC1, and MIAPACA, respectively, compared to a 22% TGI value for the model developed for the clinical data.

Figure 3. Boxplots of maximum %TR during treatment for pancreas cancer (A) and TSRwk12 for ovarian cancer (B). 1_Clinical is the descriptor calculated from the clinical PKPD model to each type of cancer, 2_PreclinicalCL is the descriptor calculated from the translational simulations without differing between cell lines for each type of cancer. 3_KP4, 4_ASPC1, 5_MIAPACA2 and 6_PANC1 indicate the %TR calculated from the translational simulations for each pancreatic cell line. 3_A2780, 4_ SKOVxluc3_1 and 5_ SKOVxluc3_2 indicate the TSRwk12 calculated from the translational simulations for each ovarian cell line.

136

Chapter 3

Finally, in figure 4A the results corresponding to OS are shown. In general, simulations from the four cell lines studied in xenografts adequately described clinical OS (both raw data or simulated based on the model developed with clinical data) for the intended treatment period, being PANC1 the best cell line predictor of the entire OS curve. Median OS of advanced pancreatic cancer patients is 33 weeks, compared to predictions for KP4, ASPC1, MIAPACA2 and PANC1 of 28, 29, 32 and 35 weeks, respectively. Extrapolations after the intended treatment period show deviations from observed profiles according to the aggressive and sensitive characteristics displayed in figure 4B.

Figure 4. A. Simulated OS probability for pancreatic cancer. Solid black line in all the figures represent the OS probability of the clinical raw data while dashed black line is the OS probability simulated with the clinical PKPD model using same conditions as the translational modelling approach. Grey vertical line indicates the end of the treatment. Coloured shaded areas represent the 95 % prediction interval of OS simulations for each cell line (KP4; green, ASPC1; yellow, MIA PACA2; red and PANC1; blue). B. shows the typical translational TS simulated profile for each cell line in pancreas (KP4; green, ASPC1; yellow, MIA PACA2; red and PANC1; blue). Black line is the typical profile simulated with the clinical PKPD model and the horizontal grey line indicates the end of the treatment.

OVARIAN CANCER.

Taking into account that during the pre-clinical analysis, TV0 and rate of proliferation exhibited inter-study variability in the case of the SKOVxluc3 cell line, two different sets of simulations were performed for this cell line. The calculated value of TV0 was 68.19 mm3.

Regarding K0, the calculated values were 75.2 (A2780), 92.62 (SKOVxluc3_1) and51.03

137

Gemcitabine PKPD Translational Model-Based Approach

(SKOVxluc3_2) mm3. The right panel in Figure 2A shows that there is an agreement between typical tumour predictions obtained from the model developed with clinical data, and those obtained based on the proposed simulation framework. The agreement is higher for the SKOVxluc3 cell line. As occurred in pancreatic cancer, variability is greater in patients that in simulated individuals using human population PK parameters and mouse-derived PD parameters. In the lower panels of figure 2B simulations are shown for each cell line separately. The cell line A2780 appears to be the most sensitive during the whole treatment period in comparison with the SKOVxluc3 cell line. The simulation results shown in figure 2 indicate that treatment effect on tumour shrinkage is more pronounced in ovarian cancer than in pancreatic cancer, as is the case in the clinical setting.

In Figure 3B, TSRwk12 showed median values of -0.6 (overall), -0.6 (SKOVxluc3_1), -0.45 (SKOVxluc3_2), -0.9 % (A2780) that are within the range of values predicted with the PKPD model developed with the clinical data (-0.5 %).

TGI values were 148%, 500%, 174% and 1118% for A2780, SKOVxluc3_1, SKOVxluc3_2 and for the model developed for the clinical data, respectively.

Figure 5 shows similar OS results to those obtained in the pancreatic cancer the results, where simulations for the two cell lines captured the clinical outcome during the treatment period quite well. SKOVxluc3_2 was identified as the best ovarian performing cell line, providing the best description of the entire OS curve. Median OS of ovarian cancer patients is 24 months, while the predicted value from A2780, SKOVxluc3_1, and SKOVxluc3_2 simulations are 27, 22, and 24 months, respectively.

138

Chapter 3

Figure 5. A. Simulated OS probability for ovarian cancer. Solid black line in all the figures represents the OS probability of the clinical raw data while dashed black line is the OS probability simulated with the clinical PKPD model using same conditions as the translational modelling approach. Coloured shaded areas represent the 95 % prediction interval of OS simulations for each cell line (A2780; purple, SKOVxluc3_1; pink, and SKOVxluc3_2 orange). B. shows the typical translational TS simulated profile for each cell line in ovarian (A2780; purple, SKOVxluc3_1; pink, and SKOVxluc3_2; orange) cancer, respectively. Black line is the typical profile simulated with the clinical PKPD models and the horizontal grey line indicates the end of the treatment.

139

Gemcitabine PKPD Translational Model-Based Approach

4. DISCUSSION

In this retrospective analysis, the Model-informed Drug Discovery and Development (MID3) paradigm has been successfully applied to gemcitabine data linking pre-clinical and clinical response through a mechanistic translational model-based approach. The methodology proposed which couples pre-clinical PKPD model, clinical PK models and dosing schedules, together with a physiologically-based mouse-human conversion, was able to provide a good description of tumour response and survival data for patients with pancreatic and ovarian cancer treated with gemcitabine alone or in combination with carboplatin. The purpose of this study was to develop a pharmacometric-based framework which would help predict drug efficacy in humans from early development phases, thus assisting and optimizing the drug development and discovery process in oncology.

The results obtained from the current analysis provide further support for the use of mouse xenograft experiments, despite the long discussion and controversy surrounding their predictive capacity (12,27), and show that they provide a wealth of information regarding tumour behaviour and drug effects. Moreover, the development of such translational approaches may assist in the design of pre-clinical analyses and selection of predictive xenograft cell lines, with the aim of optimising translation to the clinic. As an example, Spilker and colleagues (28) have recently highlighted the importance of identifying and selecting nonclinical doses to be used in mouse experiments that reflect clinical drug concentrations. This allows the efficacy of newly developed drugs to be compared against standards of care at concentrations which can be safely achieved in the clinic.

Similarly, Lindauer and colleagues (29) have developed a semi-mechanistic model including information regarding receptor binding characteristics, obtained from pre-clinical experiments, in order to assist with dosing selection during the clinical development of the antibody prembolizumab. Currently immune-oncology (IO) represents the most recent revolution in cancer treatment, nevertheless response rates are still below 50%, result that warrants drug combinations. To the best of our knowledge, there is a lack of information about PKPD model in IO, especially regarding combination therapy. We are aware of the differences regarding the immune system between animals and humans; however, from our perspective, the key point is to invest resources in developing mechanistic models in the pre-clinical arena, which is supported by the current work. In this regard, Ourdani et al (30) have recently shown that the structure of models developed using xenograft data can be translated to the analysis of tumour size human data. The work proposed in our analysis is a good example of a model-based framework that goes beyond the single mechanism based models, providing the flexibility

140

Chapter 3

required to incorporate the effect of different classes of anti-cancer drugs (as it is shown in the case of gemcitabine-carboplatin combination in ovarian cancer). In the same way, we foresee that well-stablished mechanistic models in the IO arena will help to optimize development and select the most suitable anticancer drug combinations.

In line with this, the development of pre-clinical semi-mechanistic PKPD models that differentiate between system and drug-related parameters is crucial for translational research (9), due to the fact that drug-specific PD parameters are likely to be similar across species (31) and therefore extrapolation is not required. In contrast, system specific parameters (in this case

T0, K0 and proliferation rate constants) should be adapted to each tumour type and species. The approach used in the current analysis for rescaling system-related rate constants from mouse to human has already been applied in non-small cell lung cancer (24), and for describing cancer progression in an allometric PKPD model developed for a novel IGF-1 receptor inhibitor (32).

Pre-clinical modelling has been applied in drug research for making go/no go decisions (33,34). However, even when pre-clinical quantitative analyses are performed, most decisions are usually made according to qualitative relationships (i.e Treated/Control=40% (27)).

The use of modelling and simulation in translational oncology focusing on the use of tumour growth dynamics measured in xenograft studies has been explored before (14,28,32). The current study presents both similarities and differences with respect previous reports in relation to: (i) drugs and data for which the approaches were developed, (ii) mouse-human scaling methodology, (iii) doses used for performing the simulations, (iv) comparison with clinical data and (v) their conclusions and contributions to translational research as summarized in table I.

Particularly noteworthy, is the use of already marketed drug information that allows a successful comparison between the outcome of translational-based simulations and the clinical outcome to be established (14,28). In the previous studies, differences in system parameters across species were not taken into consideration, and TGI profiles were simulated during the first cycle of the chemotherapy cycle, without considering disease progression or delay in drug response. However, Titze and collages (32) considered differences in system parameters for the different species and successfully applied their translational approach, predicting the optimal dose that would produce the maximum effect with minimum side effects. Unfortunately, the lack of clinical data so far makes it impossible to validate their approach.

141

Gemcitabine PKPD Translational Model-Based Approach

Table I. Comparison of model-based translational approaches applied to oncology

Drugs & data Scaling mice-human Doses for Clinical comparison methodology simulation

Wong - Retrospective. - Mouse PD-tumour - Clinical dosing - Maximum TGI at 21 (14) - Published - Human PK schedule days calculated from clinical data (standard for each their simulations vs - Develop mouse type of cancer) overall clinical PKPD TGI model response Spilker - Retrospective. - Mouse PKPD - Calculated Clinical - Maximum TGI at 21 (25) - Published Relevant Dose days calculated with clinical data (CRD) CRD vs overall clinical - Develop mouse response

PKPD TGI model -TGIWong vs TGISpilker Titze - Prospective. - Mouse PD - Different -Lack of clinical data, (27) - No clinical data - System simulations with no comparison with - Published parameter scaled different doses the clinics

mouse PKPD TGI - Human T0 looking for the model - Allometric scaled PK optimal dose (to human) Current - Retrospective. - Mouse PD - Clinical dosing - Comparison with work - Published - System schedule clinical TS over time clinical data parameter scaled (standard for each and OS clinical data

- Published - Mouse T0 type of cancer) and simulations with mouse PKPD TGI calculated from clinical PKPD models model human T - Comparison of 0 different markers - Human PK (predicted from the - Final TGI scale translational from mouse to modelling vs human by body calculated from weigh equivalence clinical PKPD model)

In our analysis different responses to treatment (mainly in the pancreas case study) were identified from the results of the translational TS simulations between cell lines. This situation is also observed in the clinical scenario, where tumour proliferation and drug effects on patients with advanced pancreatic cancer are associated with a high variability (20). In this context, it can be proposed that genetic factors associated with each cell line, affecting, among other processes, tumour proliferation and the gemcitabine metabolism pathway, could be responsible for the different tumour behaviour and response to treatment (35,36).

As it has already been highlighted, this type of model-based analysis is expected to optimise translation of new anticancer compounds to the clinic. Nevertheless, one limitation of the current translational approach in the context of pre-clinical compound selection is the requirement of PK models based on patient data, in order to perform the simulation exercises,

142

Chapter 3

which in principle implies that the drug has to be already administered to patients. One possibility to overcome such limitation which has not been addressed in the current work might be to develop a physiological based-PK model allowing a mechanistic scaled-up of animal to human drug exposure (37,38).

To conclude, while not the first example of translational tumour response from mice to humans, the current analysis adds important contributions to the drug development and discovery process, being, to the best of our knowledge, the first translational approach to describe the time-course of the tumour response and patient survival probability (the gold standard endpoint in oncology drug development) from pre-clinical data.

143

Chapter 3

5. REFERENCES

1. Kola I, Landis J. Can the pharmaceutical industry reduce attrition rates? Nat Rev Drug Discov . Nature Publishing Group; 2004;3:711–6.

2. Ait-Oudhia S, Mager DE. Array of translational systems pharmacodynamic models of anti-cancer drugs. Journal of Pharmacokinetics and Pharmacodynamics. 2016;43:549–65.

3. Zhang L, Pfister M, Meibohm B. Concepts and challenges in quantitative pharmacology and model-based drug development. The AAPS journal. Springer; 2008;10:552–9.

4. Zhang L, Sinha V, Forgue S, Callies S, Ni L, Peck R, et al. Model-Based Drug Development: The Road to Quantitative Pharmacology. Journal of Pharmacokinetics & Pharmacodynamics. 2006;33:369.

5. D. Breimer D. PK/PD Modelling and Beyond: Impact on Drug Development. Pharmaceutical research. 2008;25:2720.

6. Mould DR, Walz AC, Lave T, Gibbs JP, Frame B. Developing Exposure/Response Models for Anticancer Drug Treatment: Special Considerations. CPT: pharmacometrics & systems pharmacology. 2015;4:e00016.

7. Derendorf H, Meibohm B. Modeling of pharmacokinetic/pharmacodynamic (PK/PD) relationships: concepts and perspectives. Pharmaceutical research. 1999;16:176–85.

8. Danhof M, de Jongh J, De Lange EC, Della Pasqua O, Ploeger BA, Voskuyl RA. Mechanism-based pharmacokinetic-pharmacodynamic modeling: biophase distribution, receptor theory, and dynamical systems analysis. Annual Review of Pharmacology and Toxicology. 2007;47:357–400.

9. Danhof M, de Lange ECM, Della Pasqua OE, Ploeger BA, Voskuyl RA. Mechanism- based pharmacokinetic-pharmacodynamic (PK-PD) modeling in translational drug research. Trends in pharmacological sciences. 2008;29:186–91.

10. Simeoni M, Magni P, Cammia C, De Nicolao G, Croci V, Pesenti E, et al. Predictive Pharmacokinetic-Pharmacodynamic Modeling of Tumor Growth Kinetics in Xenograft Models after Administration of Anticancer Agents. Cancer research. 2004;64:1094–101.

145

Gemcitabine PKPD Translational Model-Based Approach

11. Rocchetti M. Predicting the active doses in humans from animal studies: A novel approach in oncology. European journal of cancer. 2007;43:1862.

12. Kerbel RS. Human tumor xenografts as predictive preclinical models for anticancer drug activity in humans: better than commonly perceived-but they can be improved. Cancer biology & therapy. 2003;2:S134-9.

13. Ruggeri BA, Camp F, Miknyoczki S. Animal models of disease: pre-clinical animal models of cancer and their applications and utility in drug discovery. Biochemical pharmacology. 2014;87:150–61.

14. Wong H, Choo EF, Alicke B, Ding X, La H, McNamara E, et al. Antitumor Activity of Targeted and Cytotoxic Agents in Murine Subcutaneous Tumor Models Correlates with Clinical Response. Clin.Cancer Res. 2012. 3846–55. A

15. Rocchetti M, Simeoni M, Pesenti E, De Nicolao G, Poggesi I. Predicting the active doses in humans from animal studies: a novel approach in oncology. European journal of cancer. 2007;43:1862–8.

16. Hui YF, Reitz J. Gemcitabine: a cytidine analogue active against solid tumors. American Journal of Health-System Pharmacy. 1997;54:162–70.

17. Storniolo AM, AllerheiligenS R, Pearce HL. Preclinical, pharmacologic, and phase I studies of gemcitabine. Seminars in oncology. 1997;24:S7–S2.

18. Teicher Godefridus Johannes BP. Gemcitabine. Deoxynucleoside analogs in cancer therapy : Cancer drug discovery and development. 2007. page 225.

19. Garcia-Cremades M, Pitou C, Iversen PW, Troconiz IF. Characterizing Gemcitabine Effects Administered as Single Agent or Combined with Carboplatin in Mice Pancreatic and Ovarian Cancer Xenografts: A Semimechanistic Pharmacokinetic/Pharmacodynamics Tumor Growth-Response Model. The Journal of pharmacology and experimental therapeutics. 2017;360:445–56.

20. Garcia-Cremades M, Pitou C, Iversen PW, Troconiz IF. Predicting tumour growth and its impact on survival in gemcitabine-treated patients with advanced pancreatic cancer.

21. Zecchin C, Gueorguieva I, Enas NH, Friberg LE. Models for change in tumour size, appearance of new lesions and survival probability in patients with advanced epithelial ovarian cancer. British journal of clinical pharmacology. 2016;82:717–27.

146

Chapter 3

22. Joerger M, Huitema ADR, Richel DJ, Dittrich C, Pavlidis N, Briasoulis E, et al. Population Pharmacokinetics and Pharmacodynamics of Paclitaxel and Carboplatin in Ovarian Cancer Patients: A Study by the European Organization for Research and Treatment of Cancer- Pharmacology and Molecular Mechanisms Group and New Drug Development Group. Clinical Cancer Research. 2007;13:6410–8.

23. https://pi.lilly.com/us/gemzar.pdf

24. Jeroen Elassaiss-Schaap. Allometric scaling in oncology disease progression from xenograft tumor growth to human non-small-cell lung cancer. PAGE 19 (2010) Abstr 1907 [www.page-meeting.org/?abstract=1907]%0A

25. Tomayko MM, Reynolds CP. Determination of subcutaneous tumor size in athymic (nude) mice. Cancer chemotherapy and pharmacology. 1989;24:148–54.

26. Bauer R. NONMEM users guide: introduction to NONMEM 7.2.0. ICON Development Solutions: Ellicott City; 2011.

27. Johnson JI, Decker S, Zaharevitz D, Rubinstein L V, Venditti JM, Schepartz S, et al. Relationships between drug activity in NCI preclinical in vitro and in vivo models and early clinical trials. British journal of cancer. 2001; 84:1424–31.

28. Spilker ME, Chen X, Visswanathan R, Vage C, Yamazaki S, Li G, et al. Found in Translation: Maximizing the Clinical Relevance of Nonclinical Oncology Studies. Clinical Cancer Research. 2017

29. Lindauer A, Valiathan CR, Mehta K, Sriram V, de Greef R, Elassaiss-Schaap J, et al. Translational Pharmacokinetic/Pharmacodynamic Modeling of Tumor Growth Inhibition Supports Dose-Range Selection of the Anti-PD-1 Antibody Pembrolizumab. CPT: pharmacometrics & systems pharmacology. 2017;6:11–20.

30. Ouerdani A, Struemper H, Suttle A, Ouellet D, Ribba B. Preclinical Modeling of Tumor Growth and Angiogenesis Inhibition to Describe Pazopanib Clinical Effects in Renal Cell Carcinoma. CPT: Pharmacometrics & Systems Pharmacology. 2015;4:660–8.

31. Mager DE, Jusko WJ. Development of translational pharmacokinetic- pharmacodynamic models. Clinical pharmacology and therapeutics. NIH Public Access; 2008;83:909–12.

147

Gemcitabine PKPD Translational Model-Based Approach

32. Titze MI, Schaaf O, Hofmann MH, Sanderson MP, Zahn SK, Quant J, et al. An allometric pharmacokinetic/pharmacodynamics model for BI 893923, a novel IGF-1 receptor inhibitor. Cancer Chemotherapy and Pharmacology. 2017;79:545–58.

33. van Kesteren C, Mathôt RAA, Beijnen JH, Schellens JHM. Pharmacokinetic- pharmacodynamic guided trial design in oncology. Investigational new drugs. 2003;21:225–41.

34. Barrett JS, Gupta M, Mondick JT. Model-based drug development applied to oncology. Expert opinion on drug discovery. 2007;2:185–209.

35. de Sousa Cavalcante L, Monteiro G. Gemcitabine: metabolism and molecular mechanisms of action, sensitivity and chemoresistance in pancreatic cancer. European journal of pharmacology. 2014;741:8–16.

36. Binenbaum Y, Na’ara S, Gil Z. Gemcitabine resistance in pancreatic ductal adenocarcinoma. Drug resistance updates : reviews and commentaries in antimicrobial and anticancer chemotherapy. 2015;23:55–68.

37. Thiel C, Schneckener S, Krauss M, Ghallab A, Hofmann U, Kanacher T, et al. A systematic evaluation of the use of physiologically based pharmacokinetic modeling for cross- species extrapolation. Journal of pharmaceutical sciences. 2015;104:191–206.

38. Eissing T, Kuepfer L, Becker C, Block M, Coboeken K, Gaub T, et al. A Computational Systems Biology Software Platform for Multiscale Modeling and Simulation: Integrating Whole-Body Physiology, Disease Biology, and Molecular Reaction Networks. Frontiers in Physiology. 2011;2:4

148

Chapter 3

6. SUPPLEMENTARY MATERIAL

Supplementary material 1

Summary of model and data characteristics of the PKPD analysis performed for pancreas cancer (19,20)

MOUSE HUMAN

Treatment Dose 15-200 mg/Kg 1000-1500mg/m2 Level

Dosing Q3dx4, q7dx3, q7dx4, q7dx2 28 day cycles (qwx3) schedules 21 day cycles (qwx2)

PK Model Parent tricompartmental Parent bicompartmental+ intracellular Michaelis-Menten metabolite development

TGI PD Model = − × × − × 1 1 = × − ×

=×−××23−2×

Drug effect ×() 1=Kd× 3 () = model 1 2=Kd×Cp

Resistance No resistance to Gem Resistance model

- ℎ = × × × × ×() Survival

149

Gemcitabine PKPD Translational Model-Based Approach

Summary of model and data characteristics of the PKPD analysis performed for ovarian cancer (19,21).

MOUSE HUMAN

Treatment Dose Gem single agent Carbo (target AUC mg/ml min) + Gem 1000 mg/m2 Level (15-160 mg/Kg) Combination: Gem (25-50 mg/kg). Carbo (25-50 mg/kg)

Dosing Gem single agent 21 day cycles schedules (q3dx7, q3dx4, q7dx2) Carbo iv d1 Combination (qwx3) Gem iv d1,8

PK Model Gem tricompartmental Gem tricompartmental+ intracellular Michaelis-Menten metabolite Carbo bicompartimental development Carbo bicompartimental

TGI PD Model = − × × − = × − + × 1 × = × − × × − ( + ) ×

Drug effect ×() =Kd ×Exposure = gem gem model 1 Carbo=Kdcarb×Exposurecarb 2=Kd×Cp

Carbo=KdCarbo×CpCarbo Resistance No resistance to Carbo or Gem No resistance to Carbo or Gem

×××× - ℎ = × × ( × ) × Survival

150

CHAPTER 4

Mechanistic multi-scale systems pharmacokinetics model applied to the anticancer drug gemcitabine in pancreatic cancer

Maria Garcia-Cremades1,2, Nicola Melillo3, Paolo Magni3, Iñaki F. Troconiz1,2

Manuscript in preparation

1Pharmacometrics and Systems Pharmacology, Department of Pharmacy and Pharmaceutical Technology, School of Pharmacy 2Navarra Institute for Health Research (IdiSNA), University of Navarra, Pamplona, Spain 3Laboratory of Bioinformatics, Mathematical Modelling and Synthetic Biology, Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, Italy

Chapter 4

ABSTRACT

AIM: A mechanistic multi-scale model for gemcitabine based on its molecular pathway, integrating in vitro- and in vivo data, was built with the aim of anticipate the different response rates to treatment in pancreatic cancer depending on the accumulation and retention of gemcitabine active metabolite (dFdCTP).

METHODS: A mechanistic network of gemcitabine metabolic pathway was developed using in vitro literature data and was coupled with a physiological pharmacokinetic (PBPK) model. Analyses were done with Matlab R2016b.

RESULTS: The network was able to describe the time course of extracellular and intracellular metabolites of gemcitabine for two different pancreatic cancer cell lines (normal-PK9 and resistant to gemcitabine-RPK9) using the same set of parameters, and including the ratio of protein concentration of the target metabolic enzymes (dCK and CDA) and transporters (hENT1) as covariates of the model. Once the system model was linked to a PBPK model, it was possible to generate concentrations profiles of gemcitabine in plasma and of dFdCTP in tumour in pancreas (AUCdFdCTP;tumour 1.11x10-4 mmolxh/mL; Cmax dFdCTP; tumour 1.37x10-6 mmol/mL) of the range of those reported in literature given the standard dose used for pancreatic cancer patients (3.34 mmol/m2 iv infusion (0.5h)). Finally, simulations of different dFdCTP exposures in pancreas tumour were generated assuming different degrees of polymorphisms that, coupled with the clinical PKPD model (previously developed for gemcitabine in patients with advanced pancreatic cancer), provided tumour size and survival profiles in agreement with published clinical results.

CONCLUSION: A multi-scale model characterizing the metabolic pathway of gemcitabine, and predicting the pharmacokinetics of dFdCTP has been developed. The computational platform, coupled with a clinical PKPD model provided different predictions of clinical responses to gemcitabine associated to individual genetic factors related to its metabolic pathway.

155

Chapter 4

1. INTRODUCTION

Gemcitabine (dFdC) is a nucleoside antimetabolite anticancer pro-drug effective against several solid tumours (1–4). As a pro-drug, it has to be intracellularly metabolized to its active metabolite, dFdCTP, to exert its cytotoxic action, following a complex metabolic pathway (5). Firstly, dFdC is taken into the cell by active transporters, (hENTs, hCNTs) (6), and then dFdC is phosphorylated by dCK enzyme to its monophosphate form, dFdCMP. It is subsequently metabolized by nucleoside kinases to the diphosphate (dFdCDP) and triphosphate (dFdCTP) nucleosides, which finally bind to the DNA promoting apoptosis (7). dFdC also suffers inactivation by cytidine deaminase (CDA) enzyme, leading to the inactive metabolite dFdU, which is excreted in urine (8).

Although the main responsible of gemcitabine effects is the accumulation and intracellular retention of dFdCTP, sensitivity to the treatment is also determined by the (i) activity of the drug to enter into the cell, (ii) activity of the catabolizing, anabolizing and target enzymes and (iii) ability of the tumor cells to repair DNA damage (5).

One of the biggest complications associated to treatment with gemcitabine is the variability in responses, ranging from lack of efficacy to severe toxicity (9).

Several studies suggest that these different rates of responses to gemcitabine could be in part explained by individual genetic factors affecting, among other processes, its metabolic pathway, leading to different amounts of dFdCTP. As an example, a high activity of CDA enzyme is related with a higher depletion of gemcitabine (10). It is also stated that treatment efficacy may be explained by a non-functional transporter of the pro-drug into the cell (6), and that cells lacking of dCK enzymes are resistant to dFdC cytotoxicity (11).

In addition, some clinical studies in patients with pancreatic cancer treated with gemcitabine, associated these different expressions of the transporters or the target enzymes activity with a high or low survival probabilities (12–14).

Pancreatic cancer constitutes one of the most aggressive and lethal oncology diseases, with an overall 5-year survival rate of less than 5% (15). Treatment with gemcitabine represents the first line therapy of this disease, either given in combination with nab-paclitaxel for patients with ECOG status 0-1, or as a single agent for advanced patients (ECOG >1), and for those patients who cannot receive combination treatments (16).

157

Gemcitabine Multi-Scale SPK& PBPK Model

Gemcitabine effects on pancreatic cancer have been described under the model- based PKPD paradigm in in vitro (17–19), preclinical in vivo (20) and clinical stages (21). However, to the best of our knowledge, those models do not consider variability in the metabolism pathway. In fact, the high variability associated with the rate of tumour proliferation and drug effects during the clinical characterization of drug efficacy (21) maight be explained by those individual genetic factors affecting gemcitabine metabolism pathway, leading to different concentration levels of the active metabolite.

Hence, an understanding of the systemic and cellular pharmacokinetics (PK) of dFdC could be used to develop a quantitative model that, including the characterization of its metabolism pathway, can help improving the efficiency of pancreatic cancer treatment.

The physiologically based PK (PBPK) modelling approach provides a framework for integrating drug-specific parameters and in vitro measurements with physiological system-specific parameters (22). This type of models for example allows simulating drug concentrations in a specific tissue, incorporating enzymatic information regarding drug metabolism.

Therefore, the aim of this work is to build a mechanistic multi-scale PK model for gemcitabine based on its molecular metabolism pathway and incorporate it into a PBPK model. This platform will allow simulating different concentrations of dFdCTP depending on the expression of the target metabolism enzymes, and therefore anticipating the different rate of responses to treatment in patients with pancreatic cancer.

158

Chapter 4

2. METHODS

The current work was performed in three different steps. First, a system PK model was developed for gemcitabine metabolism using published in vitro data. Secondly, a PBPK model was developed, incorporating the already built system PK model on it. Finally, this framework was coupled with a clinical PKPD model of gemcitabine for advanced pancreatic cancer (21), enabling the simulation of different responses to gemcitabine treatment in the clinics, depending of different gemcitabine metabolite exposures generating with the multiscale system PBPK model.

SYSTEM PK MODEL

Extensive literature research was performed looking for in vitro gemcitabine data to build and parameterize the drug metabolism pathway. This pathway, which has been fully defined over the years (5,7,8), is represented in figure 1.

Figure 1. Schematic representation of gemcitabine metabolism network, including metabolites (dFdCe, dFdCi, dFdCMP, dFdCDP, dFdCTP, dFdU and dFdUMP), transporters (hENT1) and target enzymes responsible of driving the metabolism reactions (dCK, NMPK, NDPK, dCDA and dCMPD).

Unknown enzymatic name reactions were named as KMPC, KDPMP, KTPDP and KINH.

The data used to build the model consist on: (i) In vitro concentration of gemcitabine metabolites profiles (dFdCextracellular, dFdCintracellular, dFdCMP, dFdCDP and dFdCTP) for 2 pancreatic cell lines (normal PK9 cell line and resistant RPK9 cell line) extracted from the literature (23), and (ii) the amount of target enzymes responsible of gemcitabine metabolism used in both cell lines experiments (23).

Gemcitabine metabolic network was optimized by modelling the metabolites profiles extracted from literature. Enzymatic reactions were parameterized using first- order rate constant, except for the target enzymes (CDA, dCK, hENT1) which were described by a Michaelish Menten reaction. For these enzymatic reactions, literature Km

159

Gemcitabine Multi-Scale SPK& PBPK Model

(Michaelis Menten constant) values were used (KMhent1= 160 µM (24), KMdck=4.6 µM (25)).

A global network was developed, estimating the same set of parameters for both cell lines and including the rate of amount of the target enzymes (CDA, dCK, hENT1) between cell lines as covariates. Those different covariates, COVCDA, COVdCK and COVhENT1, present (unitless) values of 1 for PK9 cell line and of 1.64, 0 and 1.35 for RPK9, respectively.

One critical aspect of the current analysis was the transformation of the metabolites profiles [in units of concentration as presented on the original article (pmol/mg of protein)] to units of amount (pmol)], facilitating the link to the PBPK model. Information regarding the in vitro experiments needed to stablish the correction of units was obtained from literature (23): Cells used in the in vitro experiments were seeded onto non-coated tissue culture dishes at the concentration of 1.5x104 cells/cm2, with a supposed culture area of 9.6 cm2.The cell number per protein amount of PK9 and RPK9 cell lines was 106 cells/mg. Original in vitro data were transformed to pmol by multiplying their values by a correction factor. This correction factor was calculated by dividing the total amount of cells (9.6 x 1.5x104) by cell number per mg of protein (106). Likewise, KM values obtained from literature for hENT1 and dCK were transformed to pmol by multiplying them by the total amount of cells and by the volume of a standard cell, which was assumed to be 187x10-15 L (23).

The analysis was performed with Matlab (R2016b), using CMA-ES estimation method (27). The R2, coefficient of determination of the estimation, was calculated for each of the estimated metabolite profiles, to asses the goodness of fit of the estimation.

PBPK MODEL

A PBPK model to describe dFdC distribution and metabolism was developed for a male mean subject (height 176 cm, weight 73 kg and age 30 years). The model consists on fourteen organs and tissues and it is represented in Figure 2.

160

Chapter 4

Figure 2. PBPK model structure. The tissues and organs modelled in the PBPK were: adipose tissue, bone, brain, gut, heart, kidneys, liver, lungs, muscle, pancreas, skin, spleen, arterial and venous blood. ti represents the organs modelled in the PBPK that are not shown in the figure. VB: Venous blood. AB: Arterial blood.

Each organ and tissue in PBPK (excluding arterial and venous blood) was described using a permeability limited model (28). This choice was supported by the hydrophilic nature of dFdC hampering distribution into the cells (6,29).

dFdC is transported inside the cell by concentrative nucleoside transporters (mainly hCNT1) and equilibrative nucleoside transporters (mainly hENT1) proteins. The activity of both, hCNT1 and hENT1, were modelled as a linear clearance on each organ (Equations 1 and 2). hENT1 was considered as a bidirectional transporter. To account for the different transporters expression on each organ, the term was multiplied for the relative expression of the transporter in the given organ. These relative expression values were taken from the Open Systems Pharmacology Suite version 7.17 (Supplementary material).

, = , (1)

, = (, − ,) (2)

is the relative expression of the enzyme, is the linear clearance, assumed equal in all organs and , and , are unbound extracellular and

7 https://github.com/open-systems-pharmacology

161

Gemcitabine Multi-Scale SPK& PBPK Model

intracellular dFdC concentrations, respectively. ℎ1 and ℎ1 represents the reactions concerning dFdC transport inside the cell, in time, by hCNT1 and hENT1, respectively.

Unbound fraction of dFdC was considered equal to one (30).

Each organ was divied in two compartments, representing intracellular and extracellular spaces. The generic dFdC tissue extracellular and intracellular unbound concentration dynamics are represented in Equations 3 and 4.

, , (3) , = − − , − , :⁄:

, (4) = + , , ,

Where is the tissue blood flow, corresponds to the arterial dFdC concentration, : is the tissue to plasma partition coefficient, calculated as in (28) and

: is the blood to plasma partition coefficient. , and , are calculated by multiplying the tissue volume for the extracellular and intracellular water fractions (fEW, and fIW, respectively): , = and , = . Supplementary material lists the values of the model parameters together with their references.

Finally, a linear clearance () was added in the liver compartment accounting for dFdC main metabolism (31). In Equation 5 represents the intracellular hepatic concentration dynamics.

, (5) = + − , , , ,

, , were optimized using CMA-ES method (32) on mean subject data generated from a PK model (33), assuming a 30 in infusion of3.34 / of dFdC .

To include the network described in the section System PK model, a compartment representing the pancreatic tumour was integrated into the pancreas compartment. This compartment was supposed to be spherical with a diameter of 5.3 (34). The metabolic network was included in this tumour compartment following the schematic representation of figure 3. The extracellular compartment of the network corresponds to pancreatic extracellular space and the main hypothesis was that the enzymatic concentration in vitro in normal pancreas cell line (PK9) was equal to that in vivo in pancreas organ.

162

Chapter 4

Supplementary material lists the equations accounting for gemcitabine and its intracellular metabolites concentration on the pancreas compartment.

Figure 3. Pancreatic organ compartment representation. Cyrcles represent the metabolites of gemcitabine network.

CLINICAL SIMULATIONS

Simulations of different dFdCTP PK exposures were generated assuming different degrees of polymorphisms (due to different expressions of the transporters or target enzymes), represented by different covariate values. Eight different dFdCTP concentration profiles in pancreas tumour compartment were generating assuming; (i) high expression of hENT (COVhENT1=2), dCK (COVdCK=2), and CDA (COVCDA=2), (ii) low expression of hENT

(COVhENT1=0.5), dCK (COVdCK=0.5) and CDA (COVCDA=0.5), (iii) high sensitivity to treatment

(COVhENT1=2, COVdCK=2, COVCDA=0.5), and (iv) high resistance to treatment (COVhENT1=0.5,

COVdCK=0.5, COVCDA=2).

The different dFdCTP concentration profiles generated in tumour compartment were then linked to the clinical tumour size response-survival model developed for gemcitabine (21) (given the standard dosing schedule of 6 chemotherapy cycles of 28 days [weeklyx3 + 1week rest]), and tumour size and survival probability profiles were simulated. This tumour size response- survival model is displayed briefly in the following equations:

= × − × (6)

163

Gemcitabine Multi-Scale SPK& PBPK Model

ℎ = × × ( × ) × ×() (7)

Where tumour size (TS) has an exponential growth driven by the first order rate constant Kp, and Edrug represents drug effects and accounts for gemcitabine efficacy, driven by dFdCTP exposure (AUC) in white blood cells, and a term accounting for gemcitabine emerging resistance (21). The risk or hazard is represented by hz, λ and β are the base and shape parameters of the Weibull distribution model, and the term ×() describes the change in hz elicited by TS. The link between hz and OS is established through the cumulative hazard (HZ) as indicated by the following expression: =

The main assumption was that the concentrations of dFdCTP generated in tumour compartment with the current PBPK model were equivalent to the concentrations of dFdCTP in white blood cells generated with the clinical compartmental PK model (33).

Tumour size and survival simulations were performed with Berkeley Madonna (34) (version 8.3.18).

164

Chapter 4

3. RESULTS

SYSTEM PK MODEL

The system network was able to describe the time course of extracellular and intracellular metabolites concentrations of gemcitabine for two different pancreatic cancer cell lines (normal-PK9 and resistant to gemcitabine-RPK9) using the same set of parameters, and including the ratio of protein concentration of the target metabolic enzymes (CDA (1.64), dCK (<0.02)) and transporters (hENT1(1.35)) as covariates of the model.

The estimated parameter values are displayed in Table I.

Table I. Estimates of model parameters

Parameters Values

VM hENT1 (pmol/h) (x COVhENT1) 5 7 VM dCK (pmol/h) (x COVdCK) 2.8x10 -1 6 KNMPK (h ) 2x10 -1 6 KNDPK (h ) 2x10 2 VMDCMPD (pmol/h) 4.9x10 -1 6 KDPMP(h ) 2.7x10 -1 6 KMPC (h ) 2.7x10 -1 6 KTPDP (h ) 2.7x10 3 VMCDAi (pmol/h) (x COVCDA) 4x10 3 VMCDAe (pmol/h) (x COVCDA) 10 3 KMCDAi (pmol) 1.6x10 3 KMCDAe (pmol) 4.4x10 -1 3 KINH (h ) 2x10 2 R dFdCe;PK9 0.9783 2 R dFdCi;PK9 -0.9063 2 R dFdCMP;PK9 0.9818 2 R dFdCDP;PK9 0.8018 2 R dFdCTP;PK9 0.8023 2 R dFdCe;RPK9 0.9676 2 R dFdCi;RPK9 -48.6116 2 R dFdCMP;RPK9 1 2 R dFdCDP;RPK9 1 2 R dFdCTP;RPK9 1 R2 is the coefficient of determination of the estimation of each metabolite of each cell line.

As a note, in order to estimate a lower amount of parameters, some parameters were fixed to the same values (KNMPK & KNDPK; KDPMP, KMPC &KTPDP). In Table I it is also indicated in which parameters a covariate has been included.

165

Gemcitabine Multi-Scale SPK& PBPK Model

Figure 4 displays the prediction results of applying the developed metabolisc network for each in vitro metabolite profile and each pancreatic cell line. As it can be observed, the observations were reasonably well described.

Figure 4. Metabolites profiles for PK9 and RPK9 cell lines. Points=raw data. Lines=model predictions.

First, parameters corresponding to the PK9 cell line (which contains greater battery of data) were estimated and used as initial values for the simultaneous analysis of the data of the two cell lines.

Because of the scarce data, it was not possible to describe the profiles regarding the inactive metabolites (dFdU, dFdUMP) when estimating both cell lines together, that were also available from the original article from which the in vitro data were extracted. Therefore, it was decided to reduce the network, characterizing only the pathway concerning the dFdC metabolites.

166

Chapter 4

PBPK MODEL

Regarding the PBPK model, the results obtained from the estimation of the clearances are displayed in Table II.

Table II. Parameter optimization values.

Parameters Values

[/ℎ] 423.0877

, [/ℎ] 208.6823

, [/ℎ] 16.3332

[/ℎ] 35.1293 R2 0.9186

Due to the fact that the organs were modelled as permeability limited, the system network for gemcitabine metabolism pathway was easily introduced into the pancreas organ. Once the system model was integrated with a PBPK model, it was also possible to generate plasma concentrations of gemcitabine and tumour active metabolite concentration of the range of those reported in literature. Figure 5 shows the simulation profiles of gemcitabine metabolites (both intracellular and in plasmatic concentration resembling dFdC extracellular). C (mmol/L) C (mmol/L) C C (mmol/L) C (mmol/L) C

Figure 5. Simulations of gemcitabine metabolites concentration profiles in plasma and in pancreas- tumour compartment, performed with the developed PBPK model coupled with the system network, given the standard dosing of 3.34 mmol/m2 infused over 30 min. dFdCi, dFdCMP, dFdCDP and dFdCTP plots show their concentration simulated profiles in in vitro compartment in pancreas. The profile in yellow is the dFdC venous plasma concentration profile simulated with the

167

Gemcitabine Multi-Scale SPK& PBPK Model current PBPK model. Grey profiles in dFdC Venous plasma and dFdCTP plots correspond to the respectively PK data profiles simulated with the clinical compartmental PK model (33).

In order to validate the results from the PK analysis, different metrics (total AUC and Cmax) were calculated from the compartmental PK model (33), and the current developed PBPK model. Results from both approaches were found equivalents and are displayed in Table III.

Table III. Predicted PK metrics obtained for the current PBPK model and the clinical PK model given gemcitabine standard dose (3.34 mmol/m2, infused during 30 min).

Parameter PK model PBPK model

AUCdFdC (mmolxh/L) 4.21x10-2 3.93x10-2

CmaxdFdC (mmol/L) 4.09x10-2 4.07x10-2

AUCdFdCTP (mmolxh/mL) 2.21x10-5 1.11x10-4

CmaxdFdCTP (mmol/mL) 1.5x10-6 1.37x10-6

CLINICAL SIMULATIONS

Simulations of different dFdCTP exposures (AUC) were generated assuming different degrees of polymorphisms (resembling different expressions of the transporters or target enzymes), adopting the values specified in Material and Methods. These AUCs were calculated accounting for the total exposure over time after drug administration. Once the simulations were performed, these simulated AUCs were transformed to pmolxweek/106 cells units in order to incorporate them to the clinical tumour-response PKPD model developed for gemcitabine for patients with advanced pancreatic cancer (21). Median AUC of dFdCTP in white blood cells, in advanced pancreatic cancer patients, after a standard dose administration of gemcitabine (3.34 mmol/m2, 30 min nfusion), is 34.06 pmol x week/106 cells (21). Table IV shows the simulated AUC of dFdCTP in tumour compartment resulted from each simulation performed with the current PBPK model.

168

Chapter 4

Table IV. Simulated AUCs of dFdCTP in pancreatic tumour given the standard administration of gemcitabine (3.34 mmol/m2, 30 min infusion), modifying the covariate values of

AUCdFdCTP;tumour COVARIATES MODIFICATIONS AUCdFdCTP;tumour (pmolxweek/106 (mmolxh/mL) cells) ↑ hENT expression; COVhENT1=2 4.39x10-4 373.299 ↓ hENT expression; COVhENT1=0.5 2.78x10-5 24.4 ↑ dCK expression; COVdCK=2 2.82x10-4 239.79 ↓ dCK expression; COVdCK=0.5 2.81x10-5 23.89 ↑ CDA expression; COVCDA=2 2.84x10-5 24.15 ↓ CDA expression; COVCDA =0.5 2.81x10-4 238.95 Sensitive; COVhENT1=2, COVdCK=2, COVCDA=0.5 1.56x10-3 1326.53 Resistance; COVhENT1=0.5, COVdCK=0.5, COVCDA =2 1.77x10-6 1.51 the network.

Finally, Figure 6 shows the concentration profiles over time of these dFdCTP simulated profiles in the tumour compartment after the single gemcitabine administration (standard dose of 3.34 mmol/m2,infused during 30 min), together with the TS and OS simulation profiles obtained after incorporating these set of simulated dFdCTP AUC values in the joint TS-OS PKPD model (21), given 6 chemotherapy cycles of the standard treatment of gemcitabine (3.34 mmol/m2 iv, 30 minutes infusion, administered weekly x3 followed by a week of rest from treatment) .

169

Gemcitabine Multi-Scale SPK& PBPK Model

Figure 6. A) dFdCTP concentration profiles in pancreas tumour compartment, simulated for each covariate modification. B) Tumour size simulations after incorporating each of the simulated dFdCTP AUC (identified with the same colour as in A) in the clinical PKPD model (21), given 6 chemotherapy cycles of the standard treatment of gemcitabine (3.34 mmol/m2 iv, 30 minutes infusion, administered weekly x3 followed by a week of rest from treatment). C) Survival simulated profiles, performed for each tumour profile in B.

170

Chapter 4

4. DISCUSSION

A multi-scale system PK model characterizing the metabolic pathway of gemcitabine, and predicting the PK of its active metabolite in patients together with their clinical outcome has been developed. The multiscale platform was done coupling different resources obtained from (i) published in vitro data (23), (ii) a PK compartmental model developed for gemcitabine pro-drug in plasma and active metabolite in white blood cells (33) and a (iii) PKPD model developed with data obtained from pancreatic cancer patients, from two clinical phase II and phase III trials (21), bridging together quantitative system pharmacology (in this case, quantitative system PK [QSPK]), PBPK and PKPD models accounting for disease progression.

The development of this type of platforms shows multiple advantages in terms of oncology drug discovery and development which currently shows one of the highest attrition rates (95%) despite the huge investments: (i) possibility of predicting clinical outcomes taking into account responsive patients, in terms of genetic predisposition (ii) understanding the mechanism behind drug efficacy/toxicity and (iii) accurate translation from in vitro to clinical data.

This work was performed in three steps, (i) characterization of the metabolism of gemcitabine, (ii) development of a PBPK model for gemcitabine together incorporating the metabolic network and, (iii) simulations of clinical outcomes corresponding to different degrees of genetic polymorphism associated to the enzymes responsible of gemcitabine metabolism.

Regarding gemcitabine metabolism network, QSPK modelling was successfully applied to describe, with one single model structure, both resistant and responder in vitro profiles to gemcitabine treatment as a function of the level of enzymatic expression. These results open the possibility of developing a platform able to predict clinical outcomes in patients exhibiting genetic polymorphisms and/or complex phenotypes, aiding dose individualization (35).

The in vitro experiments were performed after a single dose exposure of gemcitabine and a single profile per metabolite was collected for each cell line. In this context, too many parameters were aimed to be estimated with very few data. Thefore, the prediction was strongly biased, as it can be appreciated by the values of R2 for dFdCi metabolite for both PK9 and RPK9. However, due to the aim of the current work, the estimation was considered adequate.

171

Gemcitabine Multi-Scale SPK& PBPK Model

However, although the current QSPK network was able to describe the main in vitro intracellular gemcitabine metabolites, it failed at the time to describe the time course of the inactive metabolites of gemcitabine which are generated through the CDA pathway, responsible of gemcitabine fast elimination (25). Additional efforts are warranted to complete the network including that pathway, as it has been identified as a target metabolism, responsible of resistance to gemcitabine treatment in several cancers (mainly in pancreas) (5).

A permeability limited PBPK model was developed accounting for the activity of plasma membrane transporters mediating dFdC intracellular influx and efflux, given that dFdC shows hydrophilic characteristics and cannot cross plasma membrane by passive diffusion (6,29). In particular dFdC is transported inside the cell by CNT and ENT proteins, encoded by genes that belong to SLC28 and SLC29 families, respectively. CNT mediates unidirectional flux from outside to inside the cell, taking advance of transmembrane sodium gradient and is primarily expressed in kidney, liver and gut. ENT instead mediates bidirectional flux and the driving force is the substrate concentration gradient. dFdC is a high affinity substrate for hCNT1 and better substrate for hENT1 with respect to others transporters (e.g. hCTN2,hENT2) (36,37).

As it is specified in Methods section, a linear liver clearance was added to describe gemcitabine main metabolism supported by the high hepatic CDA activity.

One of the main advantages of PBPK in terms of multi-scale modelling is that it provides a quantitative workflow by which scaled drug-specific parameters (using in vitro- in vivo extrapolations) can be used to predict drug concentration on plasma and target tissues of new drugs. Therefore, in the current work it is provided a platform for translating in vitro gemcitabine metabolites profiles to in vivo concentrations in humans at target sites, coupling together system PK (QSPK) and PBPK modelling (38). One example of this integrated QSPK/PBPK workflow is the model developed by Melillo (39) for atorvastatin, in which a PBPK multi-scale model for describing human PK was built, considering a mechanistic metabolism network created with the use of in vitro data.

To the best of our knowledge, there are still very few reports on QSP/PBPK/PD models in translational and clinical oncology. As an example, a quantitative systems approach was developed for paclitaxel, given in combination with a liposome formulation of doxorubicin (40). In their model, the authors described and simulated plasma and tumour PK data, apoptosis induction profiles, as well as tumour growth profile for several treatments and dosing schedules. Similarly, the QSP model developed for rituximab (alone

172

Chapter 4 and in combination with fenretinide or recombinant human Apo2 ligand), was able to describe the CD20 signalling pathway in non-Hodgkin´s lymphoma (41). Their multi-scale models relate tumour responses to their mechanisms of action, and allow performing simulations of different new dosing regiments.

The final set of simulations performed in the current analysis brought together the systems, PBPK, and PKPD models developed for gemcitabine (33). OS profiles were generated based on different degrees of polymorphisms (resembling different expressions of the transporters or target enzymes). The resulting variability in the profiles generated agreed well with published clinical results (12–14) indicating that different rate of responses associated to gemcitabine treatment could be in part explained by individual genetic factors affecting, among other processes, the gemcitabine metabolic pathway. Nevertheless, these results should be taken carefully, since the lack of quantitative data regarding enzymatic expression only allows performing qualitative polymorphism simulations, doubling or halving the standard amount of enzymes estimated for PK-9 cell line. Our current results warrant new in vitro experiments covering several pancreatic cell lines, taking into account their rate of enzyme expression and the quantitative evaluation of the degree of target enzymes polymorphisms in the pancreatic cancer patient population receiving gemcitabine.

To conclude, we present an encouraging translational quantitative approach that bridges together QSPK/PBPK/PD modelling, capable of providing predictions of clinical response in oncology and understanding the sources responsible of patient variability to gemcitabine treatment.

173

Chapter 4

5. REFERENCES

1. Oettle H, Arnold D, Hempel C, Riess H. The role of gemcitabine alone and in combination in the treatment of pancreatic cancer. Anti-Cancer Drugs. 2000 Nov;11(10):771–86.

2. Heinemann V. Gemcitabine in metastatic breast cancer. Expert review of anticancer therapy. 2005 Jun 10;5(3):429–43.

3. Hansen SW. Gemcitabine in the treatment of ovarian cancer. International journal of gynecological cancer : official journal of the International Gynecological Cancer Society. 2001;11 Suppl 1:39–41.

4. Burkes RL, Shepherd FA. Gemcitabine in the treatment of non-small-cell lung cancer. Annals of oncology : official journal of the European Society for Medical Oncology. 1995;6 Suppl 3:S57-60.

5. Bergman AM, Pinedo HM, Peters GJ. Determinants of resistance to 2′,2′- difluorodeoxycytidine (gemcitabine). Drug Resistance Updates. 2002;5(1):19–33. A

6. Mackey JR, Mani RS, Selner M, Mowles D, Young JD, Belt JA, et al. Functional nucleoside transporters are required for gemcitabine influx and manifestation of toxicity in cancer cell lines. Cancer research. 1998 Oct 1;58(19):4349–57.

7. Storniolo AM, AllerheiligenS R, Pearce HL. Preclinical, pharmacologic, and phase I studies of gemcitabine. Seminars in oncology. 1997;24(2 Suppl 7):S7–S2.

8. Plunkett W, Huang P, Xu YZ, Heinemann V, Grunewald R, Gandhi V. Gemcitabine: metabolism, mechanisms of action, and self-potentiation. Seminars in oncology. 1995 Aug;22(4 Suppl 11):3–10.

9. Baker JAR, Wickremsinhe ER, Li CH, Oluyedun OA, Dantzig AH, Hall SD, et al. Pharmacogenomics of gemcitabine metabolism: functional analysis of genetic variants in cytidine deaminase and deoxycytidine kinase. Drug metabolism and disposition: the biological fate of chemicals. 2013 Mar 1;41(3):541–5.

10. Neff T, Blau CA. Forced expression of cytidine deaminase confers resistance to cytosine arabinoside and gemcitabine. Experimental hematology. 1996 Sep;24(11):1340–6.

175

Gemcitabine Multi-Scale SPK& PBPK Model

11. Ruiz van Haperen VW, Veerman G, Eriksson S, Boven E, Stegmann AP, Hermsen M, et al. Development and molecular characterization of a 2’,2’- difluorodeoxycytidine-resistant variant of the human ovarian carcinoma cell line A2780. Cancer research. 1994 Aug 1;54(15):4138–43.

12. Bengala C, Guarneri V, Giovannetti E, Lencioni M, Fontana E, Mey V, et al. Prolonged fixed dose rate infusion of gemcitabine with autologous haemopoietic support in advanced pancreatic adenocarcinoma. British journal of cancer. 2005 Jul;93(1):35–40.

13. Giovannetti E, Del Tacca M, Mey V, Funel N, Nannizzi S, Ricci S, et al. Transcription analysis of human equilibrative nucleoside transporter-1 predicts survival in pancreas cancer patients treated with gemcitabine. Cancer research. 2006 Apr;66(7):3928–35.

14. Sebastiani V, Ricci F, Rubio-Viqueira B, Kulesza P, Yeo CJ, Hidalgo M, et al. Immunohistochemical and genetic evaluation of deoxycytidine kinase in pancreatic cancer: relationship to molecular mechanisms of gemcitabine resistance and survival. Clinical cancer research : an official journal of the American Association for Cancer Research. 2006 Apr;12(8):2492–7.

15. Le N, Sund M, Vinci A. Prognostic and predictive markers in pancreatic adenocarcinoma. Digestive and liver disease : official journal of the Italian Society of Gastroenterology and the Italian Association for the Study of the Liver. 2016 Mar;48(3):223–30.

16. Ruess DA, Görgülü K, Wörmann SM, Algül H. Pharmacotherapeutic Management of Pancreatic Ductal Adenocarcinoma: Current and Emerging Concepts. Drugs {&} Aging. 2017;1–27.

17. Zhu X, Straubinger RM, Jusko WJ. Mechanism-based mathematical modeling of combined gemcitabine and birinapant in pancreatic cancer cells. Journal of pharmacokinetics and pharmacodynamics. 2015 Oct;42(5):477–96.

18. Hamed SS, Straubinger RM, Jusko WJ. Pharmacodynamic modeling of cell cycle and apoptotic effects of gemcitabine on pancreatic adenocarcinoma cells. Cancer chemotherapy and pharmacology. 2013 Sep;72(3):553–63.

19. Miao X, Koch G, Straubinger RM, Jusko WJ. Pharmacodynamic modeling of combined chemotherapeutic effects predicts synergistic activity of gemcitabine and

176

Chapter 4 trabectedin in pancreatic cancer cells. Cancer chemotherapy and pharmacology. 2016 Jan;77(1):181–93.

20. Garcia-Cremades M, Pitou C, Iversen PW, Troconiz IF. Characterizing Gemcitabine Effects Administered as Single Agent or Combined with Carboplatin in Mice Pancreatic and Ovarian Cancer Xenografts: A Semimechanistic Pharmacokinetic/Pharmacodynamics Tumor Growth-Response Model. The Journal of pharmacology and experimental therapeutics. 2017 Mar;360(3):445–56.

21. Garcia-Cremades M, Pitou C, Iversen PW, Troconiz IF. Predicting tumour growth and its impact on survival in gemcitabine-treated patients with advanced pancreatic cancer.

22. Mager DE, Woo S, Jusko WJ. Scaling pharmacodynamics from in vitro and preclinical animal studies to humans. Drug metabolism and pharmacokinetics. 2009;24(1):16–24.

23. Ohmine K, Kawaguchi K, Ohtsuki S, Motoi F, Egawa S, Unno M, et al. Attenuation of phosphorylation by deoxycytidine kinase is key to acquired gemcitabine resistance in a pancreatic cancer cell line: targeted proteomic and metabolomic analyses in PK9 cells. Pharmaceutical research. 2012 Jul;29(7):2006–16.

24. Mackey JR, Yao SY, Smith KM, Karpinski E, Baldwin SA, Cass CE, et al. Gemcitabine transport in xenopus oocytes expressing recombinant plasma membrane mammalian nucleoside transporters. Journal of the National Cancer Institute. 1999 Nov 3;91(21):1876–81.

25. Bouffard DY, Laliberté J, Momparler RL. Kinetic studies on 2’,2’- difluorodeoxycytidine (Gemcitabine) with purified human deoxycytidine kinase and cytidine deaminase. Biochemical pharmacology. 1993 May 5;45(9):1857–61.

26. Chapman EH, Kurec AS, Davey FR. Cell volumes of normal and malignant mononuclear cells. J Clin Pathol. 1981. 34:1083–90.

27. CMA-ES in MATLAB - File Exchange - MATLAB Central. https://es.mathworks.com/matlabcentral/fileexchange/52898-cma-es-in-matlab

28. Jamei M, Bajot F, Neuhoff S, Barter Z, Yang J, Rostami-Hodjegan A, et al. A mechanistic framework for in vitro-in vivo extrapolation of liver membrane transporters:

177

Gemcitabine Multi-Scale SPK& PBPK Model prediction of drug-drug interaction between rosuvastatin and cyclosporine. Clinical pharmacokinetics. 2014 Jan;53(1):73–87.

29. Veltkamp SA, Beijnen JH, Schellens JHM. Prolonged versus standard gemcitabine infusion: translation of molecular pharmacology to new treatment strategy. The oncologist. 2008 Mar 1;13(3):261–76.

30. Law V, Knox C, Djoumbou Y, Jewison T, Guo AC, Liu Y, et al. DrugBank 4.0: shedding new light on drug metabolism. Nucleic acids research. 2014 Jan;42(Database issue):D1091-7.

31. Veltkamp SA, Pluim D, van Eijndhoven MAJ, Bolijn MJ, Ong FHG, Govindarajan R, et al. New insights into the pharmacology and cytotoxicity of gemcitabine and 2’,2’-difluorodeoxyuridine. Molecular Cancer Therapeutics. 2008 Aug 1;7(8):2415–25.

32. Hansen N. The CMA Evolution Strategy: A Comparing Review. In: Towards a New Evolutionary Computation. Berlin, Heidelberg: Springer Berlin Heidelberg; 2006. p. 75–102.

33. Zhang L, Sinha V, Forgue S, Callies S, Ni L, Peck R, et al. Model-Based Drug Development: The Road to Quantitative Pharmacology. Journal of Pharmacokinetics & Pharmacodynamics. 2006 Jul;33(3):369.

34. Macey R, Oster G. Berkeley Madonna. University of California, Berkeley USA.; 2010.

35. Turner RM, Park BK, Pirmohamed M. Parsing interindividual drug variability: an emerging role for systems pharmacology. Wiley Interdisciplinary Reviews: Systems Biology and Medicine. 2015 Jul 1;7(4):221–41.

36. Plunkett W, Huang P, Searcy CE, Gandhi V. Gemcitabine: preclinical pharmacology and mechanisms of action. Seminars in oncology. 1996 Oct;23(5 Suppl 10):3–15.

37. Bergman AM, Peters GJ. Gemcitabine. In: Peters GJ, editor. Deoxynucleoside Analogs In Cancer Therapy. 2007. p. 225–51.

38. Rostami-Hodjegan A. Physiologically Based Pharmacokinetics Joined With In Vitro–In Vivo Extrapolation of ADME: A Marriage Under the Arch of Systems Pharmacology. Clinical Pharmacology & Therapeutics. 2012 Jul 30;92(1):50–61.

178

Chapter 4

39. Melillo N, Pasotti L, Magni P. Multiscale mechanistic models in Systems Pharmacology: development of a model describing Atorvastatin pharmacokinetics through integration of metabolic network in Physiologically Based Pharmacokinetic models. PAGE 26 (2017) Abstr 7147 [www.page-meeting.org/?abstract=7147].

40. Ait-Oudhia S, Straubinger RM, Mager DE. Systems Pharmacological Analysis of Paclitaxel-Mediated Tumor Priming That Enhances Nanocarrier Deposition and Efficacy. Journal of Pharmacology and Experimental Therapeutics. 2013 Jan 1;344(1):103–12.

41. Harrold JM, Straubinger RM, Mager DE. Combinatorial chemotherapeutic efficacy in non-Hodgkin lymphoma can be predicted by a signaling model of CD20 pharmacodynamics. Cancer research. 2012 Apr 1;72(7):1632–41.

179

Chapter 4

SUPPLEMENTARY MATERIAL

PBPK parameters value and model equations

PBPK PARAMETERS Table 1: Subject Subject characteristics Sex Male Height [] 176 Weight [] 73 BSA [] 1.8892

Table 2: Drug Drug parameters pKa 3.6 B:P (1) 1.94 Fup 1 Molecular weight [g/mol] 299.66 LogPow -1.2

Table 3: organs parameters value Organs Relative Percentage Neutral- Phospho- Extracellular Intracellular Lipoprotein Weight CO (1) lipids lipids water water to plasma (L/Kg) Volume Volume fraction fraction ratio (1) (2) (2) (2) (2) (3) Adipose 0.1196 0.05 0.0016 0.853 0.135 0.017 0.068 Bone 0.0856 0.05 0.0174 0.0016 0.1 0.346 0.05 Brain 0.0200 0.12 0.0391 0.0015 0.162 0.62 0.041 Heart 0.0047 0.04 0.0135 0.0106 0.32 0.456 0.16 Muscle 0.4000 0.25 0.01 0.0072 0.118 0.63 0.059 Skin 0.0371 0.05 0.0403 0.009 0.382 0.291 0.096 Spleen 0.0026 0.03 0.0071 0.0107 0.207 0.579 0.207 Kidney 0.0044 0.19 0.0121 0.024 0.273 0.483 0.137 Lung 0.0076 1 0.0215 0.0123 0.336 0.446 0.168 Gut 0.0171 0.15 0.0375 0.0124 0.282 0.475 0.141 Liver 0.0260 0.25 0.0135 0.0238 0.161 0.573 0.161 Pancreas 0.0013 0.01 0.0603 0.0044 0.12 0.664 0.06 Blood a 0.0771 0.0023 0.0013 a Blood is the sum of erythrocytes and plasma.

181

Gemcitabine Multi-Scale SPK& PBPK Model

Table 4: relative expression of hCNT1 and hENT1 Organs hCNT1 relative expression [4] hENT1 relative expression [4] Adipose 0 2.53a Bone 0.02 11.59 Brain 0.11 3.44 Heart 0.75 49.56 Muscle 0.62 100 Skin 0 2.53a Spleen 0.02 2.53 Kidney 100 4.69 Lung 0.03 5.9 Gut 4.8533 9.15 Liver 44.73 13.52 Pancreas 0.08 5.88 a In OSP this value was not provided and so was fixed as the lowest expression value.

PBPK EQUATIONS

hENT and hCNT transport reaction in a generic compartment.

, = ,

, = (, − ,)

Adipose tissue, bone, brain, gut, heart, kidneys, muscle, skin and spleen differential equations:

, , , = − − , − , :⁄:

, = + , , ,

182

Chapter 4

Liver differential equations:

, , , , , = − + + :⁄: :⁄: :⁄: , + − , − , :⁄:

, = + − , , , ,

Lungs differential equations:

, , , = − − , − , :⁄:

, = + , , ,

Venous blood differential equations:

, = − :⁄: ∈

Arterial blood differential equations:

, = − :⁄:

Pancreas differential equations:

, ,

, = − − , − , :⁄:

,, , − , , + ,

, = + , , ,

183

Gemcitabine Multi-Scale SPK& PBPK Model

, , , =−, − − , + , + + +

, = − − + + − 1+

= − − + ,

, = −

REFERENCES SUPPLEMENTARY MATERIAL

1. Poulin P, Theil FP. Prediction of pharmacokinetics prior to in vivo studies. 1. Mechanism-based prediction of volume of distribution. Journal of Pharmaceutical Sciences. 2002;91(1):129–56.

2. Rodgers T, Leahy D, Rowland M. Physiologically based pharmacokinetic modeling 1: Predicting the tissue distribution of moderate-to-strong bases. Journal of Pharmaceutical Sciences. 2005;94(6):1259–76.

3. Rodgers T, Rowland M. Physiologically based pharmacokinetic modelling 2: Predicting the tissue distribution of acids, very weak bases, neutrals and zwitterions. Journal of Pharmaceutical Sciences . 2006;95(6):1238–57.

(4) https://github.com/open-systems-pharmacology

184

GENERAL DISCUSSION

General Discussion

During these past decades, PKPD modelling approach has become a widely used methodology to guide decision making along preclinical and clinical stages of drug development (1–3). Although the well-established valuable contribution of MID3 across drug discovery, development, commercialization, and life-cycle management (4), the truth is that the success rate in the development of new drugs in the oncology area remains alarming, being the lowest compared to other therapeutic areas.

The complex nature of tumour cell biology and the limited translation of mouse- bearing tumour systems are some of the major factors contributing to the disappointing outcomes of anti-cancer lead compounds (5). Likely, the narrow therapeutic index of these compounds and the lack of placebo arm at clinical stages hamper the characterization and prediction of drug efficacy and clinical outcomes.

However, given the growing knowledge on tumour systems biology, genetics and disease predictive endpoints, the development of more (semi) mechanistic PKPD models provides a powerful tool to integrate all available multi-scale information throughout the process of anti-cancer drug development to predict drug effects on cancer systems (5).

Along the different chapters, the results obtained from applying PKPD models to retrospective data obtained from different stages of drug development for the anticancer pro-drug gemcitabine have been extensively discussed. Therefore, in this section a summary and general overview of the whole work will be given.

First, in Chapter 1 a semi-mechanistic population PKPD model for the tumour shrinkage effects of gemcitabine given as single agent or in combination with Carboplatin on human xenografts was established. The model was developed showing consistency in its structure across different tumour cell lines representing two tumour types in which gemcitabine treatment is approved (pancreatic and ovarian cancer).

One of the interesting features of this semi-mechanistic model is that the tumour progression model incorporates two main interconnected processes; tumour proliferation and nutrients supply or metabolite pools required for tumour growth, characterized by the carrying capacity/tumour volume relationship. The inclusion of this mechanism allows the characterization of two different effects described for gemcitabine, (i) decreasing tumour volume, associated to apoptosis by masked-chain termination (6), and (ii) decreasing the carrying capacity, associated to the self-potentiating mechanisms of gemcitabine metabolism, in which the natural metabolite pools necessary for DNA synthesis are diminished (7).

187

General Discussion

Besides, signal transduction model was included in the preclinical model for describing the active metabolite of gemcitabine, dFdCTP, development. The effective concentration in the last transit compartment, responsible for driving the cytotoxic effects, would represent the concentration of dFdCTP.

Due to its mechanistic nature, the model developed provides a robust preclinical platform that can be used in future analysis in terms of translational approaches. PD parameters tend to be independent between species so, this type of approaches could follow Wong (8) approach line, relating tumour metrics from preclinical analysis to clinical outcomes.

Along Chapter 2, a joint TS-OS PKPD model describing the efficacy of gemcitabine in terms of TGI and OS of patients with advanced pancreatic cancer is developed. This analysis presents several particularities such as the fact that patients died during the clinical analysis, making the missing data or drop outs informative, so, they should be modelled together, over all, in terms of model evaluation. TS derived metrics have been used as surrogate marker of survival evaluation during last decade (9–12), nevertheless, TS derived metric at one point instead of the whole profile is usually used. In this thesis, the full predicted time profile of TS was used, allowing a more realistic characterization of the hazard profile.

Gemcitabine cytotoxic effect on TS was driven by the active metabolite, dFdCTP exposure, simulated for each patient with the previously developed PK model (13). This exposure, represented by the area under the dFdCTP concentration vs time curve in white blood cells (AUC) was calculated taking into account dosing reductions and delays together with the covariates patient values (BSA, gender and age) identified in the mentioned model, allowing a proper and more accurate description of the patient PK.

The high variability observed in the clinical data must be highlighted. This is described by the IIV variability associated to the rate of tumour proliferation and drug effects (87 and 86%, respectively). It is suggested that the different rate of responses associated with gemcitabine treatment could be explained by individual genetic factors affecting, among other processes, gemcitabine metabolic pathway.

Despite the advanced stage of the disease, this analysis has identified a significant impact of TS on clinical response. This result, taking into account the model developed in Chapter 1, can be used for developing a translational methodology, trying to scale TGI in mice, to tumour response in humans through modelling and simulation.

188

General Discussion

Besides, a recently model for describing gemcitabine effects, in terms of tumour response and OS, in combination with carboplatin has been developed, following a similar approach as in pancreatic cancer, in patients with ovarian cancer (14).

These models developed with clinical data obtained from cancer patients treated with gemcitabine, together with TGI model developed for data obtained from mouse xenografts models, are used along chapter 3 for building a platform able to link both types of analyses (preclinical and clinical).

Along Chapter 3, MID3 was successfully applied retrospectively to gemcitabine, linking pre-clinical and clinical responses through a mechanistic translational model- based approach. The methodology proposed employs performing TS simulations coupling the model developed with the preclinical data presented in Chapter 1, clinical PK model and dosing schedule used in Chapter 2, together with a suitable system mice-human conversion.

This platform performs TS simulations that are able to describe the clinical panorama, in terms of TS and OS. The mouse-human conversion used supposes that drug parameters are drug specific and, therefore, equal between species. On this basis, system parameters should be rescaled from mouse to their equivalent value in human. In this analysis the rate constants are physiologically-based scaled (15), approach that has already been applied in NSCL cancer (16)

Besides, these results support the predictive value of mouse xenograft experiments, providing wealthy information regarding tumour behavior and drug effects from early development phases, despite the long discussion and controversy surrounding around them (17). In fact, this type of translational approaches may assist in the design of preclinical analyses and selection of predictive xenograft cell lines, with the aim of optimizing translation to the clinic. Furthermore, we believe that the current analysis adds important contributions to the drug development and discovery process, being, to the best of our knowledge, the first translational approach to describe the time-course of the tumour response and patient survival probability (the gold standard endpoint in oncology drug development) from pre-clinical data.

One limitation of this translational approach in the context of pre-clinical compound selection is the requirement of PK models based on patient data, in order to perform the simulation exercises. One proposal to overcome such problem could be to

189

General Discussion develop a PBPK framework allowing a mechanistic scaled-up of animal to human drug exposure.

While applying this translational approach, different responses to gemcitabine treatment (mainly in the pancreas case study) were identified from the results of the translational TS simulations between cell lines. This situation is also observed in the clinical scenario, where tumour proliferation and drug effects of patients with advanced pancreatic cancer, are associated with a high variability (18). In this context, it can be proposed that genetic factors associated to each cell line, affecting, among other processes, tumour proliferation and the gemcitabine metabolic pathway, could be the responsible for the different tumour behavior and response to treatment.

Therefore, next step of the work was trying to identify and describe this source of variability to gemcitabine responses. It has been widely stated that individual genetic factors affecting gemcitabine complex metabolism, leads to different amounts of its active metabolite, dFdCTP, and, consequently, promotes different rate of responses to its treatment. To carry out this project, the mentioned PBPK framework was applied.

In Chapter 4, a multi-scale system PK model for characterizing gemcitabine metabolic pathway is proposed. The model adequately described the metabolites profile concentration obtained from two different pancreatic cell lines in in vitro experiments. The advantage of this analysis was that one single network was developed for accounting for both resistant and responders profiles to gemcitabine treatment by including enzymatic expression levels as covariates of the models. This in vitro platform, coupled to a developed PBPK model for gemcitabine, allows simulating gemcitabine active metabolite profiles in humans from early development phases.

When the covariates that are introduced in the target enzymes of gemcitabine metabolism network were qualitative varied, active metabolites profiles in the same range of the ones observed in the clinical scenario were obtained. These simulated drug exposures, resembling genetic polymorphisms, were used with the model developed in Chapter 2 and different tumour responses with their associated survival probabilities were simulated. The resulting profiles agreed well with published clinical results (19–21) indicating that different rate of responses associated to gemcitabine treatment could in part be explained by individual genetic factors affecting, among other processes, the gemcitabine metabolism pathway.

190

General Discussion

The System model developed in Chapter 4 accounts for the variability in drug efficacy and response to gemcitabine treatment. However, it does not account for the variability associated to tumour proliferation mechanisms and apoptosis. Future perspectives may be focused on completing this work with a model accounting for the apoptosis mechanism, due to the fact that the PD pathway followed by gemcitabine has been widely described (22).

It should also be highlighted that these last results should be taken carefully, since the lack of quantitative data regarding enzymatic expression only allows performing qualitative polymorphism simulations. Future analysis should be done studying several in vitro pancreatic cell lines, taking into account the rate of enzyme expression in them and quantitative evaluate the degree of target enzymes polymorphisms of data obtained from pancreatic cancer patients.

In summary, an encouraging translational quantitative approach, applied to the anticancer pro-drug gemcitabine, capable of providing accurate predictions of clinical response in oncology and understanding the sources responsible of patient’s variability is presented in this thesis. We used systems and PKPD models built from in vitro and preclinical in vivo data, together with human derived drug independent relationships which in this case was the link between tumour dynamics and OS.

191

General Discussion

REFERENCES

1. Barrett JS, Gupta M, Mondick JT. Model-based drug development applied to oncology. Expert opinion on drug discovery. 2007 Feb 16;2(2):185–209.

2. D. Breimer D. PK/PD Modelling and Beyond: Impact on Drug Development. Pharmaceutical research. 2008;25(12):2720.

3. Sheiner LB, Steimer JL. Pharmacokinetic/pharmacodynamic modeling in drug development. Annual review of pharmacology and toxicology. 2000 Apr;40(1):67–95.

4. Marshall S, Burghaus R, Cosson V, Cheung S, Chenel M, DellaPasqua O, et al. Good Practices in Model-Informed Drug Discovery and Development: Practice, Application, and Documentation. CPT: Pharmacometrics & Systems Pharmacology. 2016 Mar 1;5(3):93–122.

5. Ait-Oudhia S, Mager DE. Array of translational systems pharmacodynamic models of anti-cancer drugs. Journal of Pharmacokinetics and Pharmacodynamics [Internet]. 2016;43(6):549–65.

6. Mini E, Nobili S, Caciagli B, Landini I, Mazzei T. Cellular pharmacology of gemcitabine. Annals of Oncology 2006 May 1;17(suppl 5):v7–12.

7. Plunkett W, Huang P, Searcy CE, Gandhi V. Gemcitabine: preclinical pharmacology and mechanisms of action. Seminars in oncology. 1996 Oct;23(5 Suppl 10):3–15.

8. Wong H, Choo EF, Alicke B, Ding X, La H, McNamara E, et al. Antitumor activity of targeted and cytotoxic agents in murine subcutaneous tumor models correlates with clinical response. Clinical Cancer Research. 2012;18(14):3846–55.

9. Claret L, Girard P, Hoff PM, Van Cutsem E, Zuideveld KP, Jorga K, et al. Model-Based Prediction of Phase III Overall Survival in Colorectal Cancer on the Basis of Phase II Tumor Dynamics. Journal of Clinical Oncology. 2009;27(25):4103–8.

10. Claret L, Lu JF, Bruno R, Hsu CP, Hei YJ, Sun YN. Simulations using a drug- disease modeling framework and phase II data predict phase III survival outcome in first- line non-small-cell lung cancer. Clinical pharmacology and therapeutics. 2012 Nov;92(5):631–4.

193

General Discussion

11. Claret L, Gupta M, Han K, Joshi A, Sarapa N, He J, et al. Evaluation of tumor- size response metrics to predict overall survival in Western and Chinese patients with first-line metastatic colorectal cancer. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2013 Jun 10;31(17):2110–4.

12. Bruno R, Lindbom L, Schaedeli Stark F, Chanu P, Gilberg F, Frey N, et al. Simulations to Assess Phase II Noninferiority Trials of Different Doses of Capecitabine in Combination With Docetaxel for Metastatic Breast Cancer. CPT: pharmacometrics & systems pharmacology. 2012 Dec 26;1:e19.

13. Zhang L, Sinha V, Forgue S, Callies S, Ni L, Peck R, et al. Model-Based Drug Development: The Road to Quantitative Pharmacology. Journal of Pharmacokinetics & Pharmacodynamics. 2006 Jul;33(3):369.

14. Zecchin C, Gueorguieva I, Enas NH, Friberg LE. Models for change in tumour size, appearance of new lesions and survival probability in patients with advanced epithelial ovarian cancer. British journal of clinical pharmacology. 2016 Sep;82(3):717–27.

15. Titze MI, Schaaf O, Hofmann MH, Sanderson MP, Zahn SK, Quant J, et al. An allometric pharmacokinetic/pharmacodynamics model for BI 893923, a novel IGF-1 receptor inhibitor. Cancer Chemotherapy and Pharmacology. 2017 Mar 27;79(3):545–58.

16. Jeroen Elassaiss-Schaap. Allometric scaling in oncology disease progression from xenograft tumor growth to human non-small-cell lung cancer. PAGE 19 (2010) Abstr 1907 [www.page-meeting.org/?abstract=1907]%0A

17. Kerbel RS. Human tumor xenografts as predictive preclinical models for anticancer drug activity in humans: better than commonly perceived-but they can be improved. Cancer biology & therapy. 2003;2(4 Suppl 1):S134-9.

18. Garcia-Cremades M, Pitou C, Iversen PW, Troconiz IF. Predicting tumour growth and its impact on survival in gemcitabine-treated patients with advanced pancreatic cancer.

19. Giovannetti E, Del Tacca M, Mey V, Funel N, Nannizzi S, Ricci S, et al. Transcription analysis of human equilibrative nucleoside transporter-1 predicts survival in pancreas cancer patients treated with gemcitabine. Cancer research. 2006 Apr;66(7):3928–35.

194

General Discussion

20. Sebastiani V, Ricci F, Rubio-Viqueira B, Kulesza P, Yeo CJ, Hidalgo M, et al. Immunohistochemical and genetic evaluation of deoxycytidine kinase in pancreatic cancer: relationship to molecular mechanisms of gemcitabine resistance and survival. Clinical cancer research : an official journal of the American Association for Cancer Research. 2006 Apr;12(8):2492–7.

21. Bengala C, Guarneri V, Giovannetti E, Lencioni M, Fontana E, Mey V, et al. Prolonged fixed dose rate infusion of gemcitabine with autologous haemopoietic support in advanced pancreatic adenocarcinoma. British journal of cancer. 2005 Jul;93(1):35–40.

22. de Sousa Cavalcante L, Monteiro G. Gemcitabine: metabolism and molecular mechanisms of action, sensitivity and chemoresistance in pancreatic cancer. European journal of pharmacology. 2014 Oct 15;741:8–16.

195

. CONCLUSIONS/CONCLUSIONES

Conclusions

(1) A comprehensive semi-mechanistic model to account for pancreas and ovarian tumour progression in mice as well as the PD effects of gemcitabine alone and in combination has been developed and internally validated under different in vivo experimental conditions.

(2) The model is expected to have an impact in translational approach, predicting tumour response in humans from models and parameters obtained from experimental settings, as it contributes timely to the recent, but still scarce quantitative information available from gemcitabine, with a robust and consistent model across cell lines and tumour types.

(3) The clinical modelling exercise, predicts the efficacy of Gemcitabine in terms of tumour growth inhibition and survival of patients with pancreatic cancer, identifying, despicte the advanced stage of the disease, patient tumour size profile as one of the main predictors of overall survival.

(4) The Model-informed Drug Discovery and Development (MID3) paradigm has been successfully applied retrospectively to gemcitabine data linking pre-clinical and clinical response through a mechanistic translational approach in two different cancers. Tumour size simulations performed with preclinical PD model, clinical PK model and clinical dosing schedule were linked to survival, using a suitable mice-human conversion, providing a good description of overall survival.

(5) A multi-scale system PK model characterizing the metabolic pathway of gemcitabine, and predicting the PK of its active metabolite (dFdCTP) has been developed. The model is able to generate different concentrations of dFdCTP depending on individual’s enzyme levels, which would explain the different rate of responses to gemcitabine treatment observed in patients with pancreatic cancer.

(6) The developed platform has the potential of being used together with PKPD models providing different predictions of clinical response to gemcitabine associated to individual genetic factors affecting, among other processes, gemcitabine metabolic pathway.

199

Conclusiones

(1) Se ha desarrollado un modelo semi-mecanístico exhaustivo que caracteriza la progresión de tumores de páncreas y ovario en ratón, así como los efectos farmacodinámicos que produce la gemcitabina, administrado como agente único y en combinación con carboplatino. Además, estos modelos han sido validados internamente bajo diferentes condiciones experimentales in vivo.

(2) El modelo preclínico desarrollado es robusto y consistente entre distintas líneas celulares y distintos tipos de tumores. Se espera que el modelo tenga un impacto traslacional, y pueda predecir la respuesta tumoral en humanos a partir de un ejercicio de modelado y simulación posterior.

(3) Se ha desarrollado un modelo PKPD que describe la eficacia de la gemcitabina en términos de inhibición del crecimiento tumoral y de supervivencia con datos extraídos de pacientes con cáncer de páncreas avanzado, identificando, a pesar del avanzado estado de la enfermedad, los perfiles tumorales de los pacientes como uno de los principales predictores de la supervivencia.

(4) El paradigma “Model-informed Drug Discovery and Development” (MID3) se ha aplicado con éxito, en un estudio retrospectivo, con datos de gemcitabina que permiten vincular la respuesta preclínica y clínica a través de un enfoque mecanístico y traslacional en dos distintos tipos de cáncer. Las simulaciones de tamaño tumoral llevadas a cabo con el modelo preclínico PD, el modelo clínico PK y el régimen de dosificación clínica, se relacionan con la supervivencia. Para ello, se ha utilizado una apropiada conversión ratón- humano, obteniendo una descripción satisfactoria de la supervivencia en pacientes a partir de fases tempranas de desarrollo de medicamentos.

(5) Por otra parte, se ha desarrollado un modelo multi-escalado de sistemas farmacocinéticos que caracteriza la ruta metabólica de gemcitabina y que además predice la farmacocinética de su metabolito activo (dFdCTP). El modelo es capaz de generar diferentes concentraciones de dFdCTP en función de los niveles enzimáticos individuales de los pacientes, lo cual explicaría la variabilidad en la tasa de respuesta al tratamiento con gemcitabina que se observa en pacientes con cáncer de páncreas.

(6) La plataforma desarrollada, acoplada al modelo de PBPK, presenta un importante potencial para poder ser utilizada conjuntamente con modelos PKPD, pudiendo predecir diferentes respuestas clínicas a gemcitabina asociadas a factores genéticos individuales que afecten, entre otros procesos, a la ruta metabólica de la gemcitabina.

201