Vous êtes sur la page 1sur 84

Modelling Pressure Induced Cell Movement and

Tumour Invasion in a Cellular Automata

Darren Hassan

This report is submitted as partial fulllment of

the requirements for the Master of Computer Science Programme of the

School of Computer Science and Software Engineering

The University of Western Australia

2010
Abstract

Inter-cellular pressure has a signicant eect on a tumour's growth and

malignancy. When the cells in one region of a tumour proliferate more

quickly than the cells in another, due to dierences in nutrient concen-

tration, density dierences result. Density dierences lead to a build-up

of pressure that pushes cells from densely populated regions to sparsely

populated regions. This pushing eect grows the tumour outward, into

the surrounding healthy tissue. Inter-cellular pressure therefore causes

pressure-induced cell movement that promotes tumour invasion and should

thus be included in a model of tumour growth.

However, it is inherently dicult to model inter-cellular pressure in

traditional Cellular Automata (CA) tumour models. Traditional CA

tumour models are formulated on a xed, regular lattice where each

lattice site represents a cell. The restrictions imposed by the model's

lattice require a dividing cell be near an empty site in which the new cell

can be placed. It is not possible for density dierences to develop in the

bulk of the tumour, when only cells near empty sites can divide.

A CA tumour model that captures realistic inter-cellular pressure has

been developed. This was achieved by introducing local cellular density

information into the model by allowing several cells to occupy a lattice

site. Based on physical principles and porous media theory, a relation-

ship between cellular density and inter-cellular pressure is proposed that

has allowed us to model pressure-induced cell movement and the inho-

mogeneous diusion of nutrient molecules.

When applied to in vitro conditions, the model has successfully sim-

ulated tumour growth that exhibits the emergence of the typical layered

structure with expanding radius seen experimentally.


Keywords: Tumour model, mathematical model, Cellular Automata,

pressure-induced cell movement

CR classication: I.6.5 [Computer Methodologies]: Simulation and

Modeling - Model Development

2
Acknowledgements
I would like to acknowledge the contribution of the project's supervi-

sors, Dr Pascal Buenzli and Dr Peter Pivonka of the Engineering and

Computational Biology Group, School of Computer Science and Soft-

ware Engineering, The University of Western Australia.

3
Contents

1 Introduction 10
1.1 Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.2 On the Nature of Mathematical Models . . . . . . . . . . 11

1.3 Why Model? . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.4 Models in Biology . . . . . . . . . . . . . . . . . . . . . . 13

1.5 Types of Mathematical Models . . . . . . . . . . . . . . 13

1.6 Thesis Motivation . . . . . . . . . . . . . . . . . . . . . . 15

2 A Biological Background to Cancer 18


2.1 Cells, Genes and DNA . . . . . . . . . . . . . . . . . . . 18

2.2 Cancer's Cause . . . . . . . . . . . . . . . . . . . . . . . 20

2.3 Malignant Tumour Development . . . . . . . . . . . . . . 21

3 Literature Review 23
3.1 Brain Tumour Growth . . . . . . . . . . . . . . . . . . . 23

3.2 Breast Tumour Growth . . . . . . . . . . . . . . . . . . . 24

3.3 Tumour Evolution . . . . . . . . . . . . . . . . . . . . . 25

3.4 Tumour Evolution and Inheritance . . . . . . . . . . . . 25

3.5 Avascular Tumour Growth . . . . . . . . . . . . . . . . . 26

3.6 Tumour Metabolism . . . . . . . . . . . . . . . . . . . . 27

4 Model Design 29
4.1 Modelling Cells . . . . . . . . . . . . . . . . . . . . . . . 30

4.1.1 A Biological Background to Cell Metabolism . . . 31

4.1.2 Modelling Cell Metabolism . . . . . . . . . . . . . 32

4.1.3 Cell Proliferation . . . . . . . . . . . . . . . . . . 33

4.2 Inter-Cellular Pressure . . . . . . . . . . . . . . . . . . . 35

4.3 Modelling Cell Migration . . . . . . . . . . . . . . . . . . 42

4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . 42

4.3.2 Modelling Pressure-Induced Cell Movement . . . 43

4.3.2.1 Displacement in 1D . . . . . . . . . . . 44

4
4.3.2.2 Displacement in 2D . . . . . . . . . . . 45

4.4 The Chemical Mass-Balance Equations . . . . . . . . . . 46

4.4.1 Glucose . . . . . . . . . . . . . . . . . . . . . . . 46

4.4.2 Oxygen . . . . . . . . . . . . . . . . . . . . . . . 48

4.5 Modelling Chemical Diusion . . . . . . . . . . . . . . . 49

4.5.1 The Adiabatic Approximation . . . . . . . . . . . 51

5 Results 52
5.1 Simulation Parameters . . . . . . . . . . . . . . . . . . . 52

5.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . 53

5.3 Comparison with in vitro Experiments . . . . . . . . . . 65

6 Conclusion 67
7 Appendices 73
7.1 An Alternative Relationship Describing Cell Volume as a

Function of Pressure . . . . . . . . . . . . . . . . . . . . 73

7.2 Discretising a Continuous Chemical Concentration . . . . 79

7.3 The Adiabatic Approximation . . . . . . . . . . . . . . . 80

7.3.1 A Theoretical Analysis of the Adiabatic Approxi-

mation . . . . . . . . . . . . . . . . . . . . . . . . 80

7.3.2 A Numerical Analysis of the Adiabatic Approxi-

mation Limit . . . . . . . . . . . . . . . . . . . . 81

5
List of Tables

4.1 The metabolic threshold values. . . . . . . . . . . . . . . 33

4.2 Glucose mass-balance equation's parameters. . . . . . . . 47

4.3 Oxygen mass-balance equation's parameters. . . . . . . . 48

5.1 The simulation's parameters. . . . . . . . . . . . . . . . 52

7.1 A description of parameters describing a spherical cell's

shape as a function of pressure. . . . . . . . . . . . . . . 73

7.2 The glucose steady state time in hours. . . . . . . . . . . 82

7.3 The oxygen steady state time in hours. . . . . . . . . . . 82

6
List of Figures

2.1 The idealised avascular tumour's structure. . . . . . . . . 22

4.1 A simplied representation of the model, viewed from above. 29

4.2 The model's assumed lattice structure, reproduced from

[30]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4.3 A cell's metabolic response to chemical stimuli described

as a phase diagram. . . . . . . . . . . . . . . . . . . . . . 33

4.4 The cell volume as a function of pressure, described by

Equation (4.11). The vertical and horizontal grey lines

denote the IFP of normal tissue and intrinsic volume re-

spectively. . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.5 The regular packing of spheres in 3D. The left-hand side

describes the hexagonal close-packing formation, whereas

the right-hand side describes the face-centred cubic pack-

ing [42]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.6 The density-porosity relationship for densities greater than

the critical density. . . . . . . . . . . . . . . . . . . . . . 40

4.7 The relationship between ρ̄ and a(ρ̄). . . . . . . . . . . . 40

4.8 The average inter-cellular pressure as a function of density. 41

4.9 Migration of cells due to pressure inhomogeneities between

lattice sites. Cells in the light grey-shaded area are moved

to the dark-shaded area during the time increment 4t.


The cell counts in each lattice site are updated accordingly. 44

4.10 Migration of cells due to pressure inhomogeneities between

the sites of a 2D lattice. Cells in the light grey-shaded area

are moved to the dark grey-shaded area during the time

increment 4t. The cell count in each lattice site is updated


0
accordingly, as the −0 and
0
+0 signs suggest. . . . . . . . 45

7
5.1 The cross-sections describe the tumour's growth with the

original metabolic thresholds. Going from top to bottom,

left to right, the cross-sections describe the tumour at day

ten, twelve, fourteen, sixteen, eighteen, nineteen, twenty

and twenty-one. . . . . . . . . . . . . . . . . . . . . . . 53

5.2 The new metabolic phase diagram describing a cell's metabolic

response to chemical stimuli. . . . . . . . . . . . . . . . . 54

5.3 The tumour's composition at 10 days. . . . . . . . . . . . 55

5.4 The tumour's composition at twelve days. . . . . . . . . 55

5.5 The tumour's composition at fourteen days. . . . . . . . 56

5.6 The tumour's composition at sixteen days. . . . . . . . . 56

5.7 The tumour's composition at eighteen days. . . . . . . . 57

5.8 The tumour's composition at nineteen days. . . . . . . . 57

5.9 The tumour's composition at twenty days. . . . . . . . . 58

5.10 The tumour's composition at twenty-one days. . . . . . . 58

5.11 A density plot of the PAT layer at twenty-one days, viewed

from above; cells in the other metabolic states are not shown. 59

5.12 The change in size of the tumour's metabolic populations

with time. . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.13 The change in size of the tumour's metabolic populations

with time, plotted on a logarithmic scale. . . . . . . . . . 60

5.14 The growth in tumour radius with time. . . . . . . . . . 61

5.15 The growth in tumour radius with time, plotted on a log-

arithmic scale. . . . . . . . . . . . . . . . . . . . . . . . . 62

5.16 The change in maximum cellular density with time. . . . 62

5.17 The change in tumour radius with time, for a maximum

site density of 300 (cells)(100 µm)−3 , plotted on a loga-

rithmic scale. . . . . . . . . . . . . . . . . . . . . . . . . 64

5.18 The change in cellular population with time, for a max-

imum site density of 300 (cells)(100 µm)−3 , plotted on a

logarithmic scale. . . . . . . . . . . . . . . . . . . . . . . 64

7.1 The change in cell volume with spontaneous curvature. . 75

7.2 Cell volume as a function of pressure, according to Equa-

tion (7.8). . . . . . . . . . . . . . . . . . . . . . . . . . . 76

7.3 Cell volume as a function of pressure, according to Equa-

tion (7.8) for a large range of pressure values. . . . . . . 76

8
7.4 A comparison of the alternative approaches for describing

cell volume as a function the pressure experienced on its

membrane. The green line denotes the inverse relationship

between cell volume and pressure, described by Equation

(4.11). The blue lines denote the volume of an empty lipid

bi-layer as a function of pressure, described by Equation

(7.1). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

7.5 The average inter-cellular pressure as a function of density,

from Equation (7.4), assuming cells deform under increas-

ing pressure. . . . . . . . . . . . . . . . . . . . . . . . . 78

7.6 A comparison of the alternative approaches for describing

average inter-cellular pressure as a function of density. The

green line denotes the pressure as a function of density for

the inverse relationship between cell volume and pressure,

described by Equation (4.11). The blue lines denote the

pressure as a function of density for an empty lipid bi-

layer, described by Equation (7.1). . . . . . . . . . . . . 78

7.7 The steady state contours for the glucose concentration eld. 83

7.8 The steady state contours for the oxygen concentration eld. 83

9
Chapter 1

Introduction

1.1 Systems
Systems are ubiquitous throughout nature, and seemingly integral to life

itself. At this moment systems are at work in your body to maintain

its homoeostatic state. Systems also result from human endeavour, some

preordained such as a mechanical clock, while others seem to emerge

independently such as the economy of a country.

A system is commonly dened as a set of independent, interacting en-

tities that together form an integrated whole. They share several common

characteristics: a structure, inter-connectivity and behaviour. Systems

are composed of parts that are structured in an ordered way. The inter-

connectivity of the various parts of a system have a functional as well as

a structural relationship. The relationships between the individual parts

of a system result in system wide behaviour.

System complexity is the most important factor in our ability to un-

derstand and predict a system's behaviour. There is no consensus on

the denition of a complex system. Boccara, in his 2004 book Modeling

Complex Systems [4] lists three characteristic he considers necessary for a

system to be considered complex: a system must consist of a large num-

ber of interacting agents, the system must exhibit emergent behaviour,

the emergent behaviour must not result from the existence of a central

controller. Boccara denes emergent behaviour as the large scale eects

of locally interacting agents that are often surprising and hard to predict

even in the course of simple interactions. An example of an emergent

property would be the cooperative behaviour of ants in a colony that

results from the local interactions between its members and not from the

existence of a central controller.

10
1.2 On the Nature of Mathematical Models
Philosophically, the universe itself can be considered the most complex of

systems, where it is reasonable to suppose that everything, to a greater

or lesser degree, has a causal inuence over everything else. However,

understanding anything would be impossible from such a holistic view

point.

Wheldon, in his 1988 book Mathematical Models in Cancer Research

[41], describes the nature of a mathematical model derived from a scien-

tic theory. He explains that the understanding of any phenomenon relies

on simplication, the picking out of the causal inuences essential to un-

derstanding the phenomenon. Once the dominant causal inuences have

been identied, they may be embodied as a set of assumptions stating

how they are believed to operate. If those assumptions form a connected

whole that deal with important questions for the subject concerned, that

collective can be described as a theory. If the theory is quantitative in na-

ture, a mathematical model can be developed describing how the causal

inuences combine to produce the phenomenon of interest.

Mathematical models of complex systems are created in a similar

way to the mathematical models of scientic theories. In fact, the phe-

nomenon a scientic theory seeks to explain, can be considered the be-

haviour of a complex system on some level. Understanding a complex

system relies on simplifying reality into an idealised system that can be

better understood and manipulated. An idealised system is formed by

considering only those entities that have a dominant inuence over the

system. These essential entities are usually determined through exper-

imentation. Once the essential entities have been identied and their

relationships understood that knowledge can be encapsulated in a set

of assumptions. If those assumptions completely describe an idealised

system and can explain the system's behaviour observed in reality, those

assumptions may be collectively known as a theory. It is possible to un-

derstand the system in a quantitative way when the theory is concerned

with measurable quantities and changes in some way, as a result of causal

inuences.

When the system is understood quantitatively, mathematical formu-

lation of the assumptions is suitable. The assumptions are said to con-

stitute a mathematical model that explains how the causal inuences

combine to produce the system's behaviour observed. A mathematical

model is therefore an exact form of a scientic theory, with particular

11
focus on the theory's quantitative aspects. A theory in turn, is a simpli-

ed representation of a complex system that captures those entities and

their relationships essential to understanding the system's behaviour. If

the assumptions are loosely dened, as with many complex systems, it is

possible for several models to be compatible with one scientic theory.

1.3 Why Model?


The main reason to create a mathematical model is to discover precisely

what phenomena ought to occur if the causal inuences taken as domi-

nant are the only ones present. If the assumptions regarding dominant

causal inuences are encapsulated in a scientic theory, a model allows

derivation of the theory's implications. When a model captures those

aspects believed to be essential to a complex system, the model allows

discovery of the idealised system's behaviour. The implications of a sci-

entic theory or description of an idealised system's behaviour may be

sought for many dierent reasons.

Karl Popper, in his 1963 book Conjectures and Refutations: The

Growth of Scientic Knowledge [31], described his philosophy of science:

Critical Rationalism. Popper believed scientic theories to be abstract

in nature, and only testable indirectly, via their implications. The truth

of a scientic theory, he contended, could only be conrmed, but never

proved, by comparing its implications to reality. However, it is possible,

and in fact essential for a theory to be falsiable for it to be consider

scientic. The modelling of a theory describing the workings of a complex

system is often the only way that theory's implications can be discovered.

When a theory's implications can only be discovered by a mathematical

model, the mathematical model is the only tool we have for conrming

or refuting the underlying scientic theory.

Mathematical models are integral to the scientic process. Science

progresses through a cyclic methodology. A theory is developed to ex-

plain a phenomenon. A mathematical model of the theory is then created.

The mathematical model allows derivation of the theory's implications,

and the experimental testing of those implications. If the implications

are found not to agree with reality, the scientic theory is refuted in

its original form. The theory may then be abandoned, or modied to

improve its predictive powers before embarking on the process again.

12
1.4 Models in Biology
In biology, models can serve other roles with more medical relevance. A

medical practitioner will use models when attempting to treat a patient.

The appropriate course of action for the medical practitioner will depend

on the model used. For example, the broad theory that some diseases may

be caused by invading micro-organisms implies that attempts should be

made to kill the micro-organisms concerned. This is a recent idea, when

disease was thought to be caused by a plethora, or overabundance, in the

blood, then bloodletting would be seen as the correct course of action.

Historically, biological models have been qualitative in nature, such

as that of diseases being caused by micro-organisms. The limited use of

mathematical models in biology is in part due to the nature of biological

theories. Wheldon [41] expressed what was needed for a phenomenon to

be amenable to mathematics, when he stated 'the phenomenon of inter-

est should involve measurable quantities for which the main interest of

the subject consists in the magnitudes of the quantities or the relation-

ship between them'. Biological theories have tended to be based upon

composite quantities, such as a species, or an organism. These are enti-

ties with a myriad of dierent qualities, not amenable to a mathematical

description or manipulation.

Recently, in the eld of oncology, a biological unit fundamental to the

topic and amenable to mathematics has emerged: the cancer cell. Cancer

researchers are now asking: how do cell populations change? how does

a tumour's radius grow? Mathematics has the potential to answer these

and many other questions of this type.

1.5 Types of Mathematical Models


Mathematical models can be classied in many dierent ways. In the

context of this thesis, we are interested in the dichotomy between con-


tinuous and discrete mathematical models. This thesis will describe the

development of a cellular automata (CA) that models both continuous

and discrete phenomena and can therefore be considered a hybrid of the

two. An understanding of our model will be improved through an exam-

ination of models from both categories.

Continuous models deal with average quantities that have been shown

representative of phenomena by statistical and stochastic theories. Con-

tinuous models are usually formulated in terms of a system of dierential

13
equations. A dierential equation is an equation relating some function

to one or more of its derivatives of various orders. Dierential equations

occur in many areas of science and technology, whenever a determinis-

tic relation involves some continuously varying quantities (modelled by

functions) and their rates of change in space and time (expressed as

derivatives).

Dierential equations can be categorised as Ordinary dierential equa-

tions (ODEs) or Partial dierential equations (PDEs). An ODE is a rela-

tion that contains functions of only one independent variable, and one or

more of their derivatives with respect to that variable. Models consisting

of a system of ODEs, whose independent variable is time, ignore local,

spatial interactions, replacing them with uniform long-range ones. Such

models are rather a crude representation of complex systems. A PDE is

a relation that contains function of two or more independent variables,

and one or more of their derivatives with respect to those variables. Sys-

tems of PDEs can be formulated with respect to space and time, thus

enabling the modelling of local interactions over time, resulting in a more

sophisticated representation.

In contrast, discrete models deal with discrete quantities. A discrete

model of a complex system explicitly models those entities believed to

have a dominant inuence, and the causal relationships between them,

though in a simplied way.

CAs are a class of discrete model that were rst proposed by John

Von Neumann [27]. CAs are formulated on a spatial lattice, which can

be considered a type of graph where the lattice sites are represented by

graph nodes and the relationships between lattice sites are represented by

edges. The dynamics of the model are dened by local rules of interaction

among neighbouring nodes, or adjacent lattice sites. The rules governing

interactions can be deterministic or stochastic. In cancer modelling, each

lattice site, of a traditional CA, represents a cell.

CAs can be used to study phenomena where agents, each following

a set of simple rules results in the sophisticated emergent behaviour of

the whole system. A CA's ability to replicate emergent behaviour makes

them suitable for modelling self organising, complex systems, such as

cancer.

One of the main advantages of an CA for modelling biological systems

is the ease with which dierent populations of cells can be modelled. By

assigning internal state variables to the sites depending on cell type a

heterogeneous population of cells can be modelled. The fact that cancer

14
arises from mutations in single cells and the ability to incorporate cell-

based data into the model makes discrete models well suited to modelling

cancer [26].

1.6 Thesis Motivation


It is our long-term aim to investigate prostatic cancer growth in bone,

by creating a CA model of the bone environment. During the last stage

of a malignant prostate tumour's development, which will be described

in more detail in Chapter 2, prostate metastases migrate to the bone via

blood vessels in the bone marrow. There are large dierences in stiness

between the mineralised bone and marrow. In-order to account for dif-

ferences in stiness we would need to model pressure. From a review of

the literature, it appears pressure is yet to be modelled by a traditional

CA, which may be due to the diculties traditional CA's have modelling

a representative cellular density. We intend to rst create a computa-

tional model of an in vitro tumour. Modelling an in vitro tumour will

allow us to ignore many of the complications involved in modelling the

bone environment, and instead focus on rening a technique for mod-

elling pressure-induced changes in cellular density. The wealth of models

and experimental data describing in vitro tumours will provide ample

opportunity to compare and test our model's results. Opportunities to

extend the model to the bone environment will be explained to the reader

as they are encountered throughout this thesis.

A cancer cell's movement is governed by a pressure-induced force

transferred through the cells it is in contact with and chemotaxis. Chemo-

taxis is a phenomenon in which a cell directs its movement according to

chemicals within its environment. Cancer cells have been known to move

from low to high oxygen and glucose regions. However, we have decided

to ignore chemotaxis in the rst iteration of the model, and instead

focus on movement caused by pressure gradients. A pressure-induced

force, experienced by a cell on its membrane, will be related to the cellu-

lar density in its vicinity. If a cell experiences a pressure gradient across

its membrane, due to a dierence in density, the cell's movement will be

directed down that pressure gradient. Of the CA models reviewed, all of

them ignored pressure. It is inconvenient to dene a smooth local density

in a traditional CA model as it requires averaging over a volume large

enough to comprise many cells, resulting in a loss of resolution.

Modelling cell movement due to proliferation is inherently dicult

15
in traditional CA models with a xed lattice. On a xed lattice, cell

division is taken to occur instantaneously, resulting in two cells where

there was one. One of the cells usually occupies the original site of the

dividing cell, but the other requires an extra space, and so must move

to a dierent site. Commonly, CA models, such as that by Kansal et


al. [23], place the new cell on an empty site within the dividing cell's

neighbourhood. If an empty site cannot be found the dividing cell is pre-

vented from proliferating. Rules such as this, fail to model the 'pushing'

eect caused by a buildup in cellular density. We intend to replace this

algorithmic prescription with a more realistic scheme grounded in the

physical considerations of the local pressure experienced by a cell.

We believe pressure-induced force has a dominant inuence over a

tumour's malignancy, and is therefore deserving of inclusion within our

model. I will now outline our approach, which will be described in more

detail in Chapter 4. We propose to model the tumour at the mesoscopic

scale, between that of the micro (≈


10 µm), in which individual cells are
2
modelled and that of the macro (≈ 10 mm ), in which average values for

the whole tumour are modelled. We intend to model packets of cells at

each site, where a packet is a cube of known volume and cell population.

To my knowledge, Piotrowska and Angus [30] were the rst to propose

this idea within the context of an in vitro tumour model. However, this

thesis intends to take this idea to the next level, by including the following

improvements:

(i) When modelling packets of cells, the volume over which the density

is measured becomes a model parameter. By selecting an appropriate

packet size a representative cellular density can be calculated. The in-

clusion of a representative cellular density within our model has many

benets. With an appropriate scheme, which will be described in Chap-

ter 4, the density of a packet of cells can be used to calculate the average

pressure experienced by the cells within the packet; information that can

then be used to model cell movement due to pressure inhomogeneities.

Modelling packets will also allow us to implicitly move cells, by subtract-

ing a cell from one packet and adding a cell to another, thus simplifying

the practicalities of cell movement.

(ii) In reality, the diusion of a substance into the tumour is aected

by the density of the tissue through which the substance is diusing.

By neglecting to model density dierences, traditional CA models have

16
assumed the homogeneous diusion of chemicals. The calculation of a

representative cellular density for a packet of cells, will enable the calcu-

lation of that packet's porosity. A packet's porosity can then be used to

modify the diusion of chemicals through that packet, with a technique

adapted from porous media transport theory. When this technique is

applied to all the packets within our model, the inhomogeneous diusion

of chemicals through the tumour can be modelled; further enhancing our

model's realism.

As with most design decisions, the modelling of packets of cells as

opposed to individual cells does have its drawbacks. The modelling of

packets of cells will result in the loss of cellular information. We will

no-longer be able to track individual cells within the model. This would

present a problem if, for instance, we were attempting to model the

tumour's sub-clonal evolution. However, at this stage, it is phenomena

at the mesoscopic scale that we are interested in, such as the tumour's

morphology, radius, and total population. These are properties that are

accessible, regardless of whether the state of every cell is known or not.

17
Chapter 2

A Biological Background to
Cancer

This chapter will discuss those aspects of cancer biology that form the

background to this thesis.

We, and all other higher species, are composed of millions of cells that

cooperate to produce a composite entity. The cells of our body dier in

their form and function depending on the role they play in maintaining a

homoeostatic state. Cooperation amongst cells is maintained by a num-

ber of checks and balances, some external to the cell, such as chemical

signals, whereas others are internal to the cell. The failure of any single

cell to cooperate does not necessarily have a systemic eect. However,

when a cell gains control over its reproductive behaviour, producing in-

denite numbers of progeny, the survival of the organism, to which that

cell belongs, can be threatened. As will be seen, cancer is a complex dis-

ease, but at its essence it can be described as the acquisition of autonomy

by a cell, in a multi-cellular organism.

2.1 Cells, Genes and DNA


A cell can be thought of as a chemical automata, or biochemical factory,

whose behaviour is programmed by the genes it contains. The genome

refers to the entirety of a cell's genetic information. Genes provide in-

structions of the sort encountered in a computer programming language.

As with computer programs, genes may be arranged in packages to pro-

vide whole sets of complex instructions, similar to a computer program's

functions. A gene tells the cell to make a protein of a particular type, by

assembling it from its constituent parts (amino acids). The protein will

18
then participate in the biochemistry of the cell according to its chemical

nature. There are many dierent types of proteins, each with a particu-

lar function. Enzymes, for example, control the rate at which chemical

reactions occur within the cell, other proteins constitute cellular building

material, while other proteins act as signals for communication amongst

cells.

Like a computer program, genetic instructions require a physical em-

bodiment. The genetic instructions of all organisms on Earth are em-

bodied and encoded in the same way; which strongly suggests a common

ancestry for all life. The genetic instructions in all species are encoded

from small molecules (nucleotides) of four types, strung together to form

a long 'supermolecule' called deoxyribonucleic acid (DNA). It is the or-

der in which the nucleotides occur that dictates the code. Each group of

three nucleotides provides an instruction specifying a particular amino

acid. A gene is a collection of these instructions that tells the cell to put

together amino acids to form a protein.

The execution, by the cell, of a gene instruction proceeds in two

phases. In the rst phase, known as transcription, the genetic instruc-

tions are recorded in a slightly dierent chemical form that results in an-

other long molecule, called ribonucleic acid (RNA). The genetic instruc-

tions contained within DNA can be thought of as being in a high-level

programming language, while the transcription into RNA converts the

high-level language into machine code, from which the cellular machin-

ery can act more directly. In the second phase, known as translation, the
genetic instructions are executed to assemble the amino acids to form the

appropriate protein. The reading and execution of genetic instructions

is a highly complex process that has been greatly simplied here.

Each step of transcription and translation are prone to error, such as

when instructions are not carried out correctly. However, these errors

are insignicant provided the genetic instructions or DNA remain intact,

the opportunity will exist to assemble the protein correctly next time the

instructions are executed. It is when the instructions have to be copied,

during cell division, that the genetic memory is at its greatest risk.

The cell cycle is a series of events that take place in a cell leading to

its division. During the rst phase of the cell cycle (interphase), the cell

enlarges by intake of the nutrients needed for cell division (mitosis) and

duplicates its DNA. During mitosis the cell splits itself into two distinct

'daughter' cells. Under ideal circumstances, the two daughter cells are

identical to each other and miniature replicas of the parent cell whose

19
ssion created them.

2.2 Cancer's Cause


Within biology, it is commonly believed that a cell's behaviour (pheno-

type) is determined by the cell's genetic instructions (genotype). Within

this context, it can be understood that corruption of a cell's DNA or geno-

type can lead to that cell having a cancerous phenotype. Specic genes

ensure uncontrolled growth is prevented. When these genes are mutated,

cells become prone to developing a cancerous phenotype. These genes

can be broadly divided into three categories [37]: oncogenes, tumour

suppressor genes and repair genes.

A multi-cellular organism is composed of many dierent types of spe-

cialised cells that are arranged to form tissues and organs. Most tissues

and organs have a recognisable physiological function for which the cells

have adapted by evolution. Depending on their function, some of these

cells will need to be continually regenerated, such as skin (epithelial tis-

sue). Oncogenes promote the proliferation of cells in the presence of

appropriate growth signals that leads to the regulated regeneration of

such cells. When oncogenes become mutated they induce the cell to

divide continuously, irrespective of growth signals.

Tumour suppressor genes are responsible for stopping growth in nor-

mal tissue. Cell proliferation is stopped by either the repression of genes

that are essential for continuing the cell cycle (senescence), or by induc-

ing programmed cell death (apoptosis), such as when the cell's DNA has

been damaged. When tumour suppressor genes become inactivated, a

cell can proliferate uncontrollably.

In human cells, both normal metabolic activities and environmen-

tal factors such as ultraviolet radiation can cause DNA damage. This

damage can eliminate the cell's ability to transcribe the genes that the

aected DNA encodes. Other types of damage induce potentially harmful

mutations in the cell's genome, which aect the survival of its daughter

cells. Repair genes encode for proteins that work to repair damage to the

DNA. When normal repair processes fail, and apoptosis does not occur,

irreparable DNA damage may result. A cell whose DNA is irreparably

damaged may acquire autonomy leading to unregulated cell division.

20
2.3 Malignant Tumour Development
Cancer can loosely be dened as a condition in which inappropriate cell

proliferation leads to an excess of cells (known as a tumour), which causes

disruption of normal tissue architecture. Not all tumours are described

as cancerous. Tumours, such as a wart, that are self-limiting in growth

and whose cells stay within the tumour are said to be benign. Malignant

tumours are those that seem to have no built-in limitations to growth and

whose cells are prone to spread to other parts of the body (metastasise).

Malignant tumours are said to be cancerous.

Once a cell has gained autonomy, and begun proliferating uncontrol-

lably, a clump of cancerous cells will form over time that is initially known

as an avascular tumour. Avascular tumours are fed by the diusion of

glucose and oxygen from the surrounding vascularture. An avascular tu-

mour's size is limited by its supply of glucose and oxygen. As tumour

cells accumulate, the diusion of glucose and oxygen becomes insucient

to sustain all the cells within the tumour mass. The tumour will then de-

velop a layered structure that depends on the distribution of glucose and

oxygen; refer to Figure 2.1 for an idealised description of an avascular tu-

mour's structure. Cells undergo un-programmed cell death (necrosis) in

the low glucose and oxygen environment at the tumour's centre. Around

the necrotic core, a layer of quiescent cells will form where there is a

sucient glucose and oxygen concentration to sustain the cells without

dividing. On the tumour's outer edge, a rim of proliferating cells will

emerge where the glucose and oxygen concentration is highest.

For an avascular tumour to progress to the next stage it must ac-

quire its own blood supply. Cells under stressful conditions release pre-

angiogenic growth factors to drive angiogenesis, the process by which

existing blood vessels are stimulated to grow from the main circulatory

system to supply tissue with blood. The accumulation of necrotic cells at

an avascular tumour's core promotes the growth of blood vessels towards

the tumour.

Once the tumour has acquired its own blood supply it is said to be in

the vascular stage of development. Once the tumour is vascularised, it

can grow larger and shed cells into blood vessels. Tumour cells can now

escape the primary tumour site via the circulatory system and set up

satellite tumours in distant parts of the body (metastases). Metastasis

is the predominant cause of mortality due to cancer.

21
The quiescent layer.

The tumour's necrotic core.


The proliferating outer rim.

Figure 2.1: The idealised avascular tumour's structure.

22
Chapter 3

Literature Review

A diverse range of CA tumour models exist. Dierences in CA mod-

els are, primarily, a result of the choices made by the modeller regarding

those aspects of tumour growth deemed essential, and therefore deserving

of inclusion within the model. Often, the modelling of one phenomenon

over another necessitates a dierent approach. A representative set of

models have been reviewed that demonstrates dierent phenomena com-

monly modelled and typical approaches taken. The reviewed models

will briey be introduced, followed by a critical analysis of the approach

employed.

3.1 Brain Tumour Growth


Kansal et al. [23] have developed a model of brain tumour growth. Tissue

is modelled using a three-dimensional (3D) lattice. Lattice sites take

on one of four distinct states corresponding to: proliferate, quiescent,

necrotic and healthy. The tumour is initially seeded in the centre of the

lattice and then allowed to evolve in space and time.

Site transition rules are non-local. A site's state depends on its loca-

tion within the tumour, for example, sites within a given distance of the

tumour's perimeter are proliferating. The authors select input param-

eters, such as the width of the proliferating rim, that quantitatively t

experimental situations. By deciding on the tumour's structure a priori,


through their choice of input parameters, the authors are prevented from

discovering new insights into the tumour's development. CAs should be

used to model systems at the microscopic level. It can then be discovered

what aect microscopic interactions have at the macroscopic level. The

transition rules were designed to model a nutrient gradient that dimin-

ishes towards the tumour's core. However, the model does not possess

23
a nutrient parameter, the tumour's morphology could just as easily be

attributable to other factors.

It appears the authors have built a simulation to t the Gompertz

function [16], a sigmoid function in which the right-hand asymptote of

the function is approached much more gradually by the curve than the

left-hand asymptote. The modellers have forgone an opportunity to im-

prove understanding and stimulate insights into cancer's development by

deciding to build a model to t the Gompertz function. The simulation

is too specic and simplistic to be useful beyond its original purpose.

3.2 Breast Tumour Growth


A model that adheres more closely to CA convention is that by Pi-

otrowska and Angus [30], who have developed a two-dimensional (2D)

model of an in vitro tumour. Lattice sites represent a unit cube contain-

ing a packet of homogeneous cells. Sites can exist in one of ve dierent

states: proliferate, quiescent, in either the aerobic or anaerobic metabolic

form and necrotic. A site's state depends on the local pH, glucose, oxy-

gen and metabolic waste levels. Nutrient diusion into the tumour and

waste diusion out are modelled with a numerical diusion algorithm.

The modelled tumour's composition and morphology were validated by

comparison to experimental data derived from the literature, for a range

of cells per site ratios.

The many cells per site approach can be used to scale the model to

sizes beyond what would be computationally possible otherwise. Models

employing the more typical one cell per site approach, which require

a dividing site to have a free near neighbour to place the new cell, are

restricted to modelling proliferate rims one cell thick, whereas this model

can increase the size of the proliferate rim by increasing the number of

cells per site.

The many-to-one approach comes at the expense of model resolution.

Automaton updates occur at the level of the site, rather than the cellu-

lar level, resulting in cellular information loss. The individual dierences

between cells becomes signicant when you consider the role sub-clonal

evolution plays in tumour malignancy [28]. By neglecting to model in-

dividual cells, the authors have chosen to ignore the eects of sub-clonal

evolution.

24
3.3 Tumour Evolution
A model that does take into account the eects of sub-clonal evolution, is

that by Anderson et al. [2]. Tumour cells are modelled as discrete entities

that occupy a single lattice site on a 2D lattice. Individual tumour cells

can have dierent genotypes that results in them having dierent pheno-

types. A cell's proliferation, movement and death potential depend on

the interaction between its genotype and local environmental conditions.

The tissue environment is described with Extracellular Matrix (ECM)

concentration (host tissue), Matrix Degrading Enzyme (MDE) concen-

tration and oxygen concentration. When a cell proliferates the new cell

randomly takes on a dierent genotype, which models mutations during

cell division, thus enabling the tumour's sub-clonal evolution.

The authors were not able to quantitatively validate the model, but

were able to show that it qualitatively resembles those tumours observed

experimentally by replicating experimental situations and observing that

the simulated tumour evolved in a similar way. In a harsh environment,

characterised by either an heterogeneous ECM or low oxygen concentra-

tion, the simulated tumour develops an invasive nger-like morphology

and is composed of cells with aggressive phenotypes, characterised by

having a high proliferation potential, high oxygen consumption rate, high

MDE production rate and high movement potential. The model supports

the clonal dominance theory [24] of cancer metastasis. The clone dom-
inance theory argues that metastasis occurs when the tumour becomes

dominated by invasive (aggressive) cells under harsh environments.

The model suggests that conventional techniques for combating can-

cer, such as radiotherapy, which increase environmental hostility, are

promoting tumour evolutionary development. A better approach would

be to make the environment benign and stable so as to limit selective

pressure and retard tumour development, such is the qualitative insight

brought about by modelling.

3.4 Tumour Evolution and Inheritance


The model by Gerlee et al. [15] can be thought of as an extension of [2].

Both models investigate the eect environment has on sub-clonal evolu-

tion and tumour morphology. The most signicant dierence between

this model and the one previously reviewed, is the modelling of a tu-

mour cell's phenotype with a feed-forward neural network. Stimuli from

25
a cell's micro-environment are processed by the neural network to calcu-

late its phenotype. When a cell proliferates there is a small probability

that its daughter cell's neural network will be slightly dierent. In the

previous model, a new cell's phenotype may bear little resemblance to

that of its parent's. The modelling of slight changes in phenotype during

proliferation, better simulates the tumour's evolution, which evolves by

minor changes from one generation to the next. Over successive genera-

tions, those phenotypes with a selective advantage grow to dominate the

tumour.

The authors had originally intended to model many dierent aspects

of tumour growth, other than those directly eecting its evolution, such

as aerobic and anaerobic metabolism. The dierent types of metabolism

are only indirectly related to the subject under investigation. The au-

thors eventually came to the conclusion that the completeness, resulting

from modelling these additional aspects was not worth the additional

complexity they incurred. The authors gathered results using a minimal

model, which only considered aerobic metabolism.

The authors can only oer supportive evidence for the model by high-

lighting experimental results that agree with its predictions, which are

similar to those of [2]: harsh environments cause the tumour to form a

ngering morphology and achieve a high level of sub-clonal diversity.

3.5 Avascular Tumour Growth


Ferreira et al. [11] have created a model of avascular tumour growth.

Nutrients diuse across a xed, regular lattice from the rst row. Lattice

sites hold one normal or necrotic cell, or many tumour cells. Nutrient con-

sumption is proportional to site population. Cells become necrotic with

a probability dependent on the nutrient per cell concentration, which

models cell death caused by nutrient starvation. Cells in the tumour's

core exhibit a higher necrotic potential than cells on the periphery.

Tumour cells divide with a probability proportional to local nutrient

per cell concentration, resulting in cells on the tumour's periphery ex-

hibiting a higher proliferation potential than cells in the core. If on the

tumour's perimeter, the new cell will randomly supplant a neighbouring

site's non-tumour cell, causing the tumour to grow, otherwise the new

cell will add to the density of the parent cell's site. Cell proliferation has

the eect of causing tumour invasion and increasing core density while a

sucient quantity of nutrients persists.

26
A tumour cell moves with a probability proportional to its local den-

sity and nutrient concentration. Cells that move, do so to a randomly

chosen neighbouring site, without regard for the density of the destina-

tion site, which could result in the unrealistic situation of a cell moving to

a more densely populated site than the site it had previously inhabited.

If the cell moves to a site containing a non-tumour cell it will supplant it,

otherwise it will exchange places with other tumour cells. The movement

rule could be improved by taking into account the density of neighbouring

sites, so that the movement potential is proportional to the density dif-

ferential between sites. As it is, cell movement has the eect of dispersing

cells from densely populated regions, albeit in an unrealistic, inecient

manner.

The model can replicate the typical layered structure caused by nu-

trient diusion into an avascular tumour. The model is also able to

reproduce the ngering morphology described in [2, 15]. The nger-

ing morphology is caused by a phenomenon known as diusion-limited


growth. Diusion-limited growth occurs in any system in which diusion

is the primary means of transporting a substance needed for growth into

an aggregate, the tumour. Diusion-limited growth occurs because the

cells arrange themselves so as to maximise their exposure to the diusing

substance. Results were gathered from a model with no cell movement,

which leads me to believe that with cell movement the model was not

able to reproduce this phenomenon.

3.6 Tumour Metabolism


From the observations of Warburg [40, 39], it was seen that tumour cells

preferentially convert glucose to lactic acid, even under aerobic condi-

tions, which results in excess acid production; this behaviour is termed

the Warburg eect. Patel et al. [29] have developed a CA model to

study the qualitative relationship between native tissue vascular density

and avascular tumour lactic acid production and invasion.

The fabric of the model is a 2D lattice, with one cell per site. Sites

are characterised by the type of cell they contain, hydrogen ion (pH)

and glucose concentrations. Sites can contain a normal cell, tumour

cell, micro-vessel or vacant space. Micro-vessels are evenly distributed

across the lattice. Glucose enters the system by diusion across micro-

vessel walls and is consumed by normal or tumour cells. Hydrogen ions

are produced by tumour cells metabolising glucose, and are removed by

27
diusion across the micro-vessel wall.

Site pH and glucose concentration determines cell behaviour. If the

pH is below a threshold the cell dies. If the pH is above the death

threshold, but below a proliferation threshold the cell is quiescent and

if above the proliferation threshold and there exists sucient glucose

concentration, the cell will proliferate if it has a free near neighbour.

Tumour cells have a higher tolerance for pH than normal cells and are,

therefore, more likely to survive and proliferate.

The authors found that a small clone of tumour cells will develop

into a malignant tumour if the clonal phenotype alters the environment

so that it is favourable to tumour cells and hostile to normal cells. Alter-

ations in the environment are dependent on the amount of acid produced

and the density of micro-vessels. If the amount of acid produced is more

than can be removed by the micro-vessels an excess of acid will accumu-

late, creating a hostile environment for normal cells.

28
Chapter 4

Model Design

An in vitro tumour growing on an articial substrate will be mathemati-


cally modelled. The tumour is assumed to be growing in isolation, on an

extra-cellular matrix (ECM). The ECM is the extra-cellular part of tissue

that provides structural support to the cells. A constant concentration of

glucose and oxygen will be maintained at the ECM's perimeter. Glucose

and oxygen will diuse across the ECM and into the tumour, where it

will be consumed by the tumour's cells; as described by Figure 4.1.

The extra-cellular matrix.

The tumour.

The constant chemical concentration.

Figure 4.1: A simplied representation of the model, viewed from above.

During a model update, the chemical (glucose and oxygen) and cel-

lular distributions are calculated for the entire lattice from the state of

the lattice at the end of the previous update. Under an adiabatic ap-

29
proximation (see Section 4.5.1), model updates proceed by calculating

the chemical equilibrium (steady state) concentration from the bound-

ary concentration and cellular distribution. Once a lattice site's chemical

steady states is known, the metabolic state of the tumour cells at that

site are updated. The cells in a lattice site update their metabolic state

depending on the glucose and oxygen concentration found at their site.

Sites containing tumour cells in a proliferating state update their cellular

density. Eventually, cell migration will redistribute cells amongst lattice

sites so as to smooth out pressure inhomogeneities.

4.1 Modelling Cells


The model consists of a 2D, xed, regular, square lattice. Each lattice

site is assumed to be a 3D cube that may contain tumour cells. The

entire lattice represents a 3D slice through the centre of the tumour.

Figure 4.2 describes the lattice.

Figure 4.2: The model's assumed lattice structure, reproduced from [30].

Lattice site geometry will be based on the model by Piotrowska et


al. [30], i.e. having sides of length of 100 µm, resulting in a volume of

(100 µm)3 per site. Lattice sites are large enough to calculate a represen-

tative tissue density, but also small enough that interesting phenomena,

such as the tumour's structure can be observed.

30
A lattice site may contain tumour cells of several type. Tumour cell

types are based on the cell's phenotype (behaviour) and metabolic state.

Cells can be proliferating, quiescent or necrotic. Cells are further cate-

gorised by their metabolic state, which can be either aerobic or anaerobic.

When the phenotype and metabolic categories are taken together, there

are ve dierent tumour cell types: proliferating aerobic (PAT), prolifer-

ating anaerobic (PANT), quiescent aerobic (QAT), quiescent anaerobic

(QANT) and necrotic.

An introduction to cell metabolism that forms the background to

modelling cells is presented below.

4.1.1 A Biological Background to Cell Metabolism

Within a human cell, energy is stored and released in the form of adeno-

sine triphosphate (ATP), which is converted from glucose, fats and amino

acids. ATP is the body's fundamental unit of energy storage. ATP's

breakdown powers muscle contraction, the generation of electrical signals,

the pumping of ions and other molecules against concentration gradients,

and much more.

The vast majority of energy used in the human body is derived from

glucose. Energy from glucose is released during glucose oxidation [9]. Ox-

idation is chemically dened as the removal of electrons from a molecule.

In oxygen rich conditions, aerobic glucose oxidation progresses through

three stages. In oxygen poor environments, anaerobic glucose oxidation

is halted after the rst stage.

Almost all cancer cells preferentially consume glucose and produce

lactic acid under aerobic conditions, a behaviour termed the Warburg

eect [40, 39]. The Warburg eect is believed to aid the rapid prolifera-

tion of cancer cells, by increasing a cell's uptake of the nutrients needed

for proliferation [18, 7] and by altering the cell's local environment, by

making it hostile to native tissue [14]. It is common for CA tumour

models to dierentiate between aerobic and anaerobic metabolism. The

main physiological dierence between these two forms of metabolism is

the number of ATP molecules produced per glucose molecule. Anaerobic

metabolism produces only two ATP molecules per molecule of glucose,

whereas aerobic metabolism produces more than thirty ATP molecules

per molecule of glucose.

Stage one, glycolysis, occurs within the cytoplasm of cells. The ly-

sis, or degradation, of a glucose molecule into two C3 fragments yields

31
pyruvate, accompanied by a reduction of N AD+ to NADH. N AD+ , or

nicotinamide adenine dinucleotide, acts catalytically by being continually

re-oxidised to N AD+ and reduced to NADH, and in doing so transfers

electrons to the mitochondria.

Anaerobic glycolysis occurs in the absence of oxygen. For glycolysis

to proceed, NADH must be recycled back to N AD+ . When a cell's

oxygen supply is insucient, the capacity of its mitochondria to recycle

NADH back to N AD+ is inadequate. NADH can then be re-oxidised by

reducing pyruvate to produce lactic acid. Anaerobic glycolysis produces

two ATP molecules per glucose molecule [9].

When the oxygen supply is plentiful, aerobic glycolysis occurs and the

cell's metabolism enters the second stage of glucose oxidation. Stage two,

the citric acid cycle or Krebs cycle (named after its discoverer), occurs

within the cell's mitochondria. Aerobic glycolysis produces pyruvate and

NADH in the cytoplasm. To be further oxidised, the pyruvate must

enter the mitochondria. A molecule of pyruvate from the cytoplasm is

converted in the mitochondria by pyruvate dehydrogenase and the citric

acid cycle to three molecules of CO2 , during this process three molecules
of N AD+ and one molecule of FAD are reduced. At this point, two

molecules of ATP have been created in glycolysis and two in the citric

acid cycle per molecule of glucose, it is in the nal stage that the majority

of ATP is created.

The nal stage of glucose oxidation, the electron transport system,


occurs in the inner mitochondrial membrane. The electron transport

system is a hierarchy of electron carriers of dierent energy potentials.

Starting from NADH and FADH, electrons move from one carrier to

the other down the energy gradient until they are eventually passed to

oxygen, resulting in the formation of water. The energy released during

electron transport is ultimately trapped as ATP. The glucose oxidation

process can result in the synthesis of more than thirty molecules of ATP

from one molecule of glucose [9].

4.1.2 Modelling Cell Metabolism

A modelled cell responds to chemical stimuli by changing its metabolic

state, in a way similar to that implemented by Piotrowska et al. [30]. A

tumour cell's metabolic phase space is depicted in Figure 4.3. Metabolic

states are delineated by the metabolic thresholds described in Table 4.1.

32
Glucose concentration @molH100 ΜmL3 D
7. ´ 10-16
PANT
6. ´ 10-16

5. ´ 10-16
QANT
4. ´ 10-16 PAT

3. ´ 10-16 Βpant
Γqant
2. ´ 10-16 Necrotic
Γpant
1. ´ 10-16 QAT
Γpat
0
4. ´ 10-17 5. ´ 10-17 6. ´ 10-17 7. ´ 10-17 8. ´ 10-17 9. ´ 10-17 1. ´ 10-16

Oxygen concentration @molH100 ΜmL3 D

Figure 4.3: A cell's metabolic response to chemical stimuli described as


a phase diagram.

Parameter Value Units Ref.

βpat 8.3 × 10−17 (mol)(cell)−1 (s)−1 [13]


γpat 18 × 10−17 (mol)(cell)−1 (s)−1 [13]
γpant 52 × 10−17 (mol)(cell)−1 (s)−1 [13]
γqant 43 × 10−17 (mol)(cell)−1 (s)−1 [13]

Table 4.1: The metabolic threshold values.

A cell's metabolic state depends on where its lattice site's chemical

environment falls in Figure 4.3. The most important distinction to draw,

is between those cells that are alive and those that are necrotic. Referring

to Figure 4.3, a cell becomes necrotic when its local glucose and oxygen

concentration is below the proliferate aerobic oxygen (βpat ) and quies-

cent anaerobic glucose (γqant ) thresholds. Once necrotic, tumour cells

cease to consume chemicals and proliferate, however, their mass still ex-

ists within the model; and plays an important role in creating pressure

inhomogeneities.

4.1.3 Cell Proliferation

Cells in a proliferating state add to the cellular density of their lattice

site at each model update. The model's proliferation scheme is based on

the mass-balance equation [10, 3]. Within our model, the cellular density

of a lattice site is dened as:

33
Ncells
ρ= (4.1)
Vsite
Where:

• ρ is the cellular density of a lattice site, with units of (cells)(100 µm)−3 .

• Ncells is the number of cells at the lattice site, with units of (cells).

• Vsite is the volume of the lattice site, with units of (100 µm)3 .

The change in cellular density of a lattice site can be described by a PDE

as:

∂ρ
=Pρ (4.2)
∂t
Where P is the proliferation rate. Additional terms, such as those de-

scribing the migration of cells into and out of a lattice site, can be added

to Equation (4.2) without loss of generality. Equation (4.2) has a con-

tinuous analytical solution of:

ρ(t) = ρ(0) e(P t) (4.3)

A cell's proliferation rate is commonly characterised by the time T needed

for a population to double, known as the doubling time. In the absence

of additional terms in Equation (4.2), the proliferation rate P can be

related to T by:

Ln(2)
P (T ) = (4.4)
T
A lattice site's cellular density will be updated at discrete time intervals

while the site's porosity is greater than zero. The porosity of tissue

is dened as the fractional volume of tissue not occupied by cellular

material:

Vcells
0≤φ=1− ≤1 (4.5)
Vsite
Where:

• φ is the tissue porosity, which is dimensionless.

• Vcells is the total tumour cell volume at a lattice site, with units of

(100 µm)3 .

A discrete approximation of Equation (4.2) can be found with an explicit

scheme as:

34

ρ(t) + 4t P ρ(t) , φ > 0
ρ(t + 4t) = (4.6)
ρ(t) , φ=0

During the time (cellular update) interval 4t, an additional 4t P ρ(t)


cells are added to the site. When the porosity of a lattice site equals

zero, cellular material is completely occupying the site's volume, leaving

no space for additional cells.

In discretising Equation (4.2), we have assumed the cellular update

interval 4t, is small in comparison to the typical cellular time scale of

the model, the doubling time T. Numerical experiments were conducted


to ascertain a cellular update interval that balances stability with e-

ciency. The cellular update interval has to be short enough to calculate

a stable cellular density, but long enough to simulate tumour growth in

a reasonable amount of time. As 4t → 0, the stability of our numer-

ical scheme improves and the discrete approximation converges to the

continuous solution.

The proliferation scheme can be extended to model sub-populations of

cells with dierent proliferation rates. The modelling of sub-populations

of cells can be achieved with a separate discrete approximation for each

sub-population. Consider sub-population ρ1 and ρ2 , their progression

could be modelled as:

ρ1 (t + 4t) = ρ1 (t) + 4t P1 ρ1 (t) (4.7)

and

ρ2 (t + 4t) = ρ2 (t) + 4t P2 ρ2 (t) (4.8)

For now, we have assumed all proliferating cells have the same doubling

time. There are, however, grounds for modelling a cell's doubling time

dependent on its metabolic state, due to the diering amount of ATP

produced in the anaerobic and aerobic states, an extension that will be

considered for the planned in vivo model.

4.2 Inter-Cellular Pressure


We want to nd a phenomenological relationship between a lattice site's

cellular density and the average pressure experienced by the site's cells

on their membrane, a quantity we will term the average inter-cellular

35
pressure. This aim forms the core of this thesis, as it is the average

inter-cellular pressure that will drive cell movement away from densely

populated regions into sparsely populated regions, thus promoting the

tumour's invasion of healthy tissue. It is possible to relate Equation

(4.1) and (4.5) as:

Ncells Vcells 1−φ


ρ= = = (4.9)
Vsite Vcell Vsite Vcell
Where Vcells is the total volume of cellular material at the lattice site,

given by:

N
Xcells

Vcells = Vcelli = Ncells Vcell (4.10)


i=1

Where Vcell is the average volume of a single cell at the lattice site. Some

preliminary observations can now be made by rearranging Equation (4.9).

Assuming that a cell's shape deforms under the pressure experienced

on its membrane, it can be observed that as the ratio Vcells /Vsite → 1,


Vcell → 1/ρ. When the cells within a lattice site are at full contact, so that

the site's volume is completely occupied by cellular material, geometric

considerations tell us that the average volume of a cell is equal to the

inverse of the cellular density or the site volume divided by the number

of cells. Also, as the ratio Vcells /Vsite → 0, Vcell → 0, the average volume

of an individual cell goes to zero.

It is now possible to relate a site's cellular density to its average

inter-cellular pressure. The average volume of a site's cells is inversely

proportional to the average inter-cellular pressure. We considered two

alternatives for describing a cell's volume as a function of the pressure

experienced on its membrane. The simpler relationship, which was used

in the experiments described in Chapter 5, will be described here, refer to

the appendices of a description of the alternative relationship. A simple

inverse relationship between average cell volume and pressure:

α
Vcell (p) = (4.11)
p
Where:

• Vcell is the average volume of a site's cells, with units of (100 µm)3 .

• p is the average inter-cellular pressure, with units of (kg)(100 µm)−1 (s)−2 .

• α is a constant, with units of (kg)(100 µm)2 (s)−2 .

36
Average inter-cellular pressure can be retrieved from the inverse of Equa-

tion (4.11).

In sparsely populated lattice sites, the probability of a cell coming

into contact with other cells is low, thus the majority of the pressure

experienced on a cell's membrane will be caused by the interstitial uid

pressure (IFP) present in the ECM. Within our model the IFP of nor-

mal tissue (p0 ) is assumed constant and equal to 7 mmHg (millimetres


−1 −2
of mercury) or 0.09333 (kg)(100 µm) (s) , from Jain's review of the

transport of molecules in the tumour interstitium [21]. Cells are assumed

to form the intrinsic shape of a sphere under the IFP of normal tissue, as

this is the lowest pressure the cell will experience. The intrinsic shape of

a cell is assumed to have a radius of 10 µm [1], from which the intrinsic

volume (Vcell0 ) can be calculated as 0.00419 (100 µm)3 . The constant α,


in Equation (4.11) can now be found algebraically from our understand-

ing of a cell's volume under IFP, as:

α = Vcell0 p0 = 0.00039 (kg)(100µm)2 (s)−2 (4.12)

It is now possible to solve Equation (4.11) and described cell volume as

a function of pressure, as in Figure 4.4.

0.12

0.10
Cell volume @H100 ΜmL3 D

0.08

0.06

0.04

0.02

0.00
0.00 0.02 0.04 0.06 0.08 0.10

Pressure @kg  H100 ΜmLHsL2 D

Figure 4.4: The cell volume as a function of pressure, described by Equa-


tion (4.11). The vertical and horizontal grey lines denote the IFP of
normal tissue and intrinsic volume respectively.

In sparsely populated lattice sites, where the pressure is assumed

constant and equal to the IFP (p = p0 ), relation (4.9) species how the

37
porosity depends on the density as:

φ(ρ) = 1 − Vcell (p0 ) ρ (4.13)

As the pressure is constant, the cell volume is also constant and equal to

the intrinsic cell volume.

Vcell0 = Vcell (p0 ) (4.14)

However, in densely populated lattice sites, a cell has a high proba-

bility of being in contact with other cells, thus the majority of the force

experienced on its membrane will be due to the cells it is in contact with.

Within densely populated regions, pressure is expected to increase with

density as the probability of a cell coming into contact with other cells

increases.

In-order to describe the average inter-cellular pressure of densely pop-

ulated sites we will start by considering an idealised tumour, composed of

impermeable cells, which are perfectly spherical and uniform in size. The

cells are imagined to be growing within an ECM. Our idealised tumour is

analogous to spheres packed in 3D. As the cells within a site proliferate,

the cellular density increases until a maximum is reached. Figure 4.5

describes the two regular packing arrangements that achieve a minimum

porosity of ≈ 0.26, according to the Hales proof of the Kepler conjecture

[17].

Figure 4.5: The regular packing of spheres in 3D. The left-hand side
describes the hexagonal close-packing formation, whereas the right-hand
side describes the face-centred cubic packing [42].

38
By assuming cells are spheres of uniform size, we can determine the

range of densities for which the average pressure experienced by the cells

in a site is equal to the IFP, and for which Equation (4.13) provides a

valid description. It is assumed that the cells within a site will form an

irregular arrangement. It is known that the irregular packing of spheres

can achieve a minimum porosity of 0.36 [33], which we will term the

critical porosity (φc ). The close packing of cells will therefore occur when

Vcells /Vsite ≈ 0.64. The density at which all the cells within a site are

in contact with their neighbours, which we will term the critical density

(ρc ), can now be found as approximately 153 (cells)(100 µm)−3 .


For densities above the critical density, we expect there to be an

interplay between a cell changing its volume, but maintaining its shape

and a cell changing its shape with a constant volume, under increasing

pressure on its membrane. We expect that a cell's shape would change

rst at the critical density, due to the uneven pressure distribution across

its membrane. When spherical cells rst come into contact it will be at

a small point across their surfaces. As the contact area enlarges, due

to deformation of the cell's shape, the pressure distribution will even

out, which will cause the cell's volume to shrink. Then as the cells

shrink the contact point will become small again, thus increasing the

pressure. The relationship between cell volume, shape and pressure could

be investigated by constructing a cell structural model that included such

phenomena as cell membrane and cytoskeleton, however such a model

would be beyond the scope of this project. At the site level, we would

expect the porosity to converge to zero at a slower rate for densities

greater than the critical density, as described in Figure 4.6.

The change in porosity for densities greater than the critical density

can be described by a relationship of the form:

φρ̄ (ρ) = a(ρ̄)(1−Vcell0 ρ) − a(ρ̄)(1−Vcell0 ρc ) − φc


 
(4.15)

Where:

• ρ̄ is the site density at which the porosity equals zero, with units

of (cells)(100 µm)−3 .

The term a(ρ̄) ≥ 1 is chosen such that φρ̄ (ρ̄) = 0. The change in a(ρ̄)
with ρ̄ is described in Figure 4.7.

39
1.0

0.8

HΡc , Φc L
0.6
Porosity Φ

0.4

0.2 HΡ, 0L

0.0
0 100 200 300 400 500 600 700

Density Ρ @cells  H100 ΜmL3 D

Figure 4.6: The density-porosity relationship for densities greater than


the critical density.

3
aHΡL

0
0 200 400 600 800 1000 1200 1400

Figure 4.7: The relationship between ρ̄ and a(ρ̄).

Referring to Figure 4.7, as the value of ρ̄ decreases, the value of the

term a(ρ̄) increases to innity at the critical density. For the experiments
described in Chapter 5, ρ̄ was chosen to be near the upper bound of the
−3
densities recorded experimentally by [13], ρ̄ = 691 (cells)(100 µm) .

Cells will eventually die due to the pressure experienced on their mem-

40
brane and be converted into necrotic material, which is uid in nature.

Under pressure, the necrotic material would be forced out of high density

regions if it has channels through which it can drain. The necrotic ma-

terial would then be replaced by surrounding cells. When the pressure

reaches the range of pressures for which cells become necrotic, we would

expect to see a further slowing in the decline of porosity and perhaps

even an increase in porosity with density, this is a modication that will

be considered for the planned in vivo model.

Cell volume as a function of density can be found from Equation (4.9)

as:

1 − φ(ρ)
Vcell (ρ) = (4.16)
ρ
The average inter-cellular pressure as a function of cellular density can

be found by substituting Equation (4.16) into the inverse of Equation

(4.11), which yields:

α αρ
p(ρ) = = (4.17)
( 1−φ(ρ)
ρ
) 1 − φ(ρ)

The density-pressure relationship is described in Figure (4.8), according

to the density-porosity relationship described by Equation (4.15).

0.30

0.25
Pressure @kg  H100 ΜmLHsL2 D

0.20

0.15

0.10

0.05

0.00
0 100 200 300 400 500 600 700

Density Ρ @cells  H100 ΜmL3 D

Figure 4.8: The average inter-cellular pressure as a function of density.

Referring to Figure 4.8, the average inter-cellular pressure in a site

is constant and equal to the IFP for densities less than the critical den-

41
sity. Once the density equals the critical density, Average inter-cellular

pressure increases in a curve until the density equals ρ̄, at which point

the porosity equals zero. Average inter-cellular pressure then increases

linearly for densities greater than ρ̄.


Having described the average inter-cellular pressure as a function of

density, it is now possible to model pressure-induced cell movement.

4.3 Modelling Cell Migration


4.3.1 Introduction

Within our model, a cell's movement is driven by a pressure force trans-

ferred through the cells it is in contact with. Cell movement due to

pressure is assumed independent of cell's metabolic state. A pressure

force will act on cells in a region where the cellular density is above the

critical density. A cell's velocity due to pressure will be assumed propor-

tional to the pressure gradient across its membrane (see Equation (4.21)

to (4.25)), expressed as:

vp = −ωp ∇p (4.18)

Where:

• vp is a cell's velocity induced by a pressure gradient, with units of

(100 µm)(s)−1 .

• p is the average inter-cellular pressure, with units of (kg)(100 µm)−1 (s)−2 .

• ∇p is the pressure gradient along which the cell is moving.

• ωp is a proportionality factor with units of (100 µm)2 (s)(kg)−1 .

For now we have chosen to ignore cell movement due to a chemotatic

attraction to higher concentrations of glucose and oxygen. However, we

anticipate modelling chemotaxis induced movement in the planned in


vivo model. Unlike pressure induced cell movement, chemotatic move-

ment would be dependent on a cell's metabolic state. A cell would need

to be alive for it to move. Generally, a cell's velocity due to chemotaxis

would be proportional to the chemical concentration gradient across its

membrane. Glucose-induced chemotaxis can be expressed as:

vg = ωg ∇g (4.19)

42
Where:

• vg is a cell's velocity induced by a glucose concentration gradient,

with units of (100 µm)(s)−1 .

• g is the glucose concentration within a site, with units of (mol)(100 µm)−3 .

• ∇g is the glucose concentration gradient along which the cell is

moving.

• ωg is a proportionality factor with unit of (100 µm)4 (mol)(s)−1 .

Chemotaxis due to oxygen would be similar to Equation (4.19). We

then anticipate combining the separate equations describing the dierent

velocity components, to form a single equation describing changes in cell

density due to movement, as:

∂ρ
= ∇ · (ρ v) (4.20)
∂t
Where:

• ∇ is the gradient of the velocity components, with units of (100 µm)−1 .

• v = v p + vg + vo is the sum pressure, glucose and oxygen velocity

components respectively, with units of (100 µm)(s)−1 .

4.3.2 Modelling Pressure-Induced Cell Movement

Below, the details of the cell migration algorithm are described. Cell

migration is the result of an inhomogeneous pressure eld that results

from dierences in cellular density. The inhomogeneous pressure eld is

assumed to generate a driving force that gives rise to a cellular velocity.

An object of volume V, surface S, embedded in a pressure eld p(r)


experiences a total force:

Z
F =− dσ n(r) p(r) (4.21)

Where n(r) is an outward-pointing unit vector normal to the surface and


dσ a line integral. If the pressure eld can be extended to the region

occupied by the object, the divergence theorem can be used to express

the force as a volume integral:

Z
F =− d3 r ∇p(r) (4.22)

43
For small objects, Equation (4.22) is approximated by:

F ≈ −V ∇p(r) (4.23)

For an object experiencing a force F in a viscous uid, the equation of

motion reads:

m v̇ = F − λ v (4.24)

If friction is strong or the object's mass small, inertial eects are negligible

(terminal velocity is reached very quickly) resulting in:

1
v= F (4.25)
λ
Where v is the cell's velocity, λ is viscous drag coecient and F the

pressure force experienced by a cell. A cell's velocity is, therefore, pro-

portional to the pressure gradient experienced across its membrane. Ad-

ditional terms, such as those describing velocity driven by chemotactic

attraction can be added to Equation (4.25) without loss of generality.

4.3.2.1 Displacement in 1D

Δx Δx

pi pi+1 pi + pi+1

i i+1 i i+1

Figure 4.9: Migration of cells due to pressure inhomogeneities between


lattice sites. Cells in the light grey-shaded area are moved to the dark-
shaded area during the time increment 4t. The cell counts in each lattice
site are updated accordingly.

Within our model, pressure is an average quantity over lattice sites. Pres-

sure inhomogeneities, therefore, only occur between neighbouring sites.

Consider a patch of cells that occupies an area overlapping two neigh-

bouring lattice sites, site i and site i+1 (Figure 4.9). During the small

time increment 4t, these cells are displaced by a distance 4x ≈ v 4t.


Equation (4.23) and (4.25) can be combined to describe the displacement

of our patch of cells between neighbouring sites:

44
h2
4x ≈ − (pi+1 − pi ) 4t (4.26)
λ
Where h is the length of a lattice site, and pi+1 and pi are the average

pressure experienced by cells in site i+1 and i respectively. Knowing the

displacement of our patch of cells, enables us to calculate the change in

cellular density of the two sites straddled by our patch of cells. In 1D,

this method can be extended to a strip of lattice sites.

4.3.2.2 Displacement in 2D

pi,j+1 pi+1,j+1 pi,j+1 pi+1,j+1


+ + +
j+1 j+1

+
Δr
j j

pi,j pi+1,j pi,j pi+1,j


i i+1 i i+1

Figure 4.10: Migration of cells due to pressure inhomogeneities between


the sites of a 2D lattice. Cells in the light grey-shaded area are moved
to the dark grey-shaded area during the time increment 4t. The cell
0 0
− and 0 +0 signs
count in each lattice site is updated accordingly, as the
suggest.

Now consider a patch of cells occupying an area overlapping four neigh-

bouring lattice sites, sites (i, j), (i+1, j), (i, j +1) and (i+1, j +1) (Figure
4.10). From Equation 4.21, we can calculate the pressure-induced force

exerted along the i and j axis as:

h2
Fi = −e1 [(pi+1,j − pi,j ) + (pi+1,j+1 − pi,j+1 )] (4.27)
2
and

h2
Fj = −e2 [(pi,j+1 − pi,j ) + (pi+1,j+1 − pi+1,j )] (4.28)
2
Where e1 and e2 are unit vectors along the i and j axes respectively.

Again, the pressure force induces a velocity component along the i and

j axis as:

45
1
vi = Fi (4.29)
λ
and

1
vj = Fj (4.30)
λ
During the time increment 4t, the displacement of our patch of cells in

the x and y directions can be calculated as:

4x = vi 4t (4.31)

and

4y = vj 4t (4.32)

which can be combined to form the vector:

! !
4x vi
4r = = 4x e1 + 4y e2 ≈ 4t (4.33)
4y vj
Knowing the displacement of our patch of cells in 2D allows us to calcu-

late the change in cellular density of the lattice sites it straddles. This

method can be applied to the lattice to model cellular migration for the

entire tumour.

4.4 The Chemical Mass-Balance Equations


The glucose and oxygen concentration of a lattice site dictates the metabolic

state of the site's cells. A site's glucose and oxygen concentrations are

described by the following mass-balance equations.

4.4.1 Glucose

The glucose mass-balance equation describing the rate of change of glu-

cose entering and leaving a lattice site:

∂g X
= ∇(Dg ∇ g) − γσ ρσ g + γv ρv (4.34)
∂t
σ∈{pat,pant,qat,qant,dt}

Refer to Table 4.2 for a description of the equation's parameters. The

rst term on the right-hand side of the equation describes the change in

46
Description Symbol Value Units
Glucose diusion parameters
Concentration gradient ∇ (100 µm)−1
Apparent diusion coecient Dg (100 µm)2 (s)−1
Concentration g (mol)(100 µm)−3
Scaled tumour cell glucose consumption/production rates
Proliferate aerobic γpat
¯ 1.9 × 10−4 (100 µm)3 (cell)−1 (s)−1
Proliferate anaerobic γpant
¯ 1.9 × 10−4 (100 µm)3 (cell)−1 (s)−1
Quiescent aerobic γqat
¯ 1.9 × 10−4 (100 µm)3 (cell)−1 (s)−1
Quiescent anaerobic γqant
¯ 1.9 × 10−4 (100 µm)3 (cell)−1 (s)−1
Dead γ¯dt 0 (100 µm)3 (cell)−1 (s)−1
Blood vessel γv (mol)(cell)−1 (s)−1
Cell densities
Proliferate aerobic ρpat (cell)(100 µm)−3
Proliferate anaerobic ρpant (cell)(100 µm)−3
Quiescent aerobic ρqat (cell)(100 µm)−3
Quiescent anaerobic ρqant (cell)(100 µm)−3
Dead ρdt (cell)(100 µm)−3
Blood vessel ρv (cell)(100 µm)−3

Table 4.2: Glucose mass-balance equation's parameters.

glucose concentration due to the diusion of glucose into and out of a

site, this term will be discussed in more detail in Section 4.5.

The second term on the right-hand side describes the cellular con-

sumption of glucose. Note, the chemical consumption term takes on a

dierent form from that common to most CA tumour models, such as

[30, 15, 29], which consist of a consumption rate multiplied by a cellular

density. We reason that for a cell to consume glucose, glucose must be

present in the cell's local environment, which justies inclusion of the

local glucose concentration as a factor in the consumption term.

The form of our consumption term necessitates we adapt the reported

consumption rates (in the literature), for our consumption denition.

Conventional models vary the consumption rate depending on the cell's

metabolic state, to account for their lack of local glucose concentration in

the consumption term, so as to simulate a decrease in consumption going

from the edge of the tumour to its core. Due to the form of our consump-

tion term, the glucose consumption rate is proportional to the glucose

concentration, which more closely resembles reality, where tumour cells

will attempt to consume glucose at the highest possible rate, regardless

of metabolic type. The form of our consumption term is more akin to a

chemical reaction than that of [30, 15, 29], which are more algorithmic

in nature.

47
The third term on the right-hand side of Equation (4.34) describes the

production of glucose from a blood vessel within the model. The glucose

production term has more relevance to the anticipated in vivo model,

and has been included here for the completeness of our description. The

production term was set to zero in the experiments conducted for Chapter

5.

4.4.2 Oxygen

The oxygen mass-balance equation describing the rate of change of oxy-

gen entering and leaving a lattice site:

∂o X
= ∇(Do ∇ o) − βσ ρ o + βv ρv (4.35)
∂t
σ∈{pat,pant,qat,qant,dt}

Description Symbol Value Units


Oxygen diusion parameters
Concentration gradient ∇ (100 µm)−1
apparent diusion coecient Do (100 µm)2 (s)−1
Concentration o (mol)(100 µm)−3
Scaled tumour cell oxygen consumption/production rates
Proliferate aerobic β¯pat 2.4 × 10−3 (100 µm)3 (cell)−1 (s)−1
Proliferate anaerobic ¯
βpant 2.4 × 10−3 (100 µm)3 (cell)−1 (s)−1
Quiescent aerobic β¯qat 2.4 × 10−3 (100 µm)3 (cell)−1 (s)−1
Quiescent anaerobic ¯
βqant 2.4 × 10−3 (100 µm)3 (cell)−1 (s)−1
Dead β¯dt 0 (100 µm)3 (cell)−1 (s)−1
Blood vessel βv (mol)(cell)−1 (s)−1

Table 4.3: Oxygen mass-balance equation's parameters.

Refer to Table 4.3 for a description of the equation's parameters.

Equation (4.35) is similar in form to Equation (4.34). The rst term

on the right-hand side describes the diusion of oxygen into and out

of a lattice site, which will be discussed in Section 4.5. The second

term describes the cellular consumption of oxygen, while the third term

accounts for the production of oxygen from a blood vessel located within

the model and was again set to zero for the experiments performed in

Chapter 5.

As before, the oxygen consumption rate is taken to be constant ir-

respective of a cell's metabolic state. Indeed, oxygen diuses through a

cell's membrane without the need for active uptake on the cell's behalf.

48
Cells consume oxygen at a rate proportional to their local concentration.

This behaviour is reected in the oxygen consumption term (the second

term on the right-hand side) of Equation (4.35). Again, this is in contrast

to conventional CA tumour models, such as those by [30, 15, 29], which

vary a cell's oxygen consumption rate depending on its metabolic state.

4.5 Modelling Chemical Diusion


The diusion of a chemical species is described by the rst term on the

right-hand side of Equation (4.34) and Equation (4.35), for an apparent

diusion coecient. The apparent diusion coecient (D ), can be de-

ned phenomenologically as the proportionality factor between the ux

(J ) and the gradient of the density (∇n) of a diusing species:

J = −D ∇n (4.36)

In a medium with several phases, we can either consider each phase

separately, or speak of averaged quantities over a representative volume

element large enough to comprise these phases, yet small enough to pre-

serve media/tissue heterogeneity. Within our model, a chemical species

can either diuse through the ECM or tumour phase. We will consider

the average quantities of a chemical diusing through a lattice site, which

can contain either ECM or tumour cells.

The porosity of tissue, is dened as the fractional volume of the tis-

sue in which the diusing substance can move. Molecular transportation,

under diusion, through tissue, depends on the porosity of the tissue. If

porosity is high, a molecule can diuse further, within a certain time,

than when porosity is low. We imagine an idealised tumour, made of

impermeable cancer cells living in an ECM. Chemical molecules are as-

sumed to diuse through the ECM, but not through the cells. Molecules

can be captured by a cell and metabolised.

Tortuosity (τ ) is the property of a curve describing its twists. Within

our model, we are concerned with the tortuosity of a chemical molecule's

path between cells, which is related to the tissue's porosity. Tortuosity af-

fects the distance a molecule can diuse within a certain time. The more

tortuous (twisted) the path a molecule takes, the shorter the straight line

distance it will travel under diusion, within a given time and the lower

the eective diusion of the species to which it belongs.

In 2D, the tortuosity of a curve can be dened as the ratio of the

49
distance (d) between the end points of a curve, and (l ) the length of a

curve [8], more formally:

d
τ= ≤1 (4.37)
l
In 3D, for spherical cells, the tortuosity can be related to the porosity by

a technique borrowed from porous media transport theory [8], as:

p
τ= φ (4.38)

The tortuosity is a proportionality factor relating the apparent diusion

coecient to the diusion coecient of the uid phase (D0 ), which cor-

responds to diusion in the absence of cells. Within our model, the ECM

is assumed to be free of cells and therefore has the diusion coecient

of the uid phase. The diusion coecient of the uid phase is related

to the apparent diusion coecient by:

D = τ D0 (4.39)

It is now possible to calculate a lattice site's diusion coecients as a

function of its density. A lattice site's porosity can be calculated as a

function of its density from Equation (4.13) for sparsely populated lattice

sites and Equation (4.15) for densely populated lattice sites. A lattice

site's tortuosity can then be found from Equation (4.38). A lattice site's

diusion coecients can be related to the diusion coecients in the

uid phase and tortuosity by Equation (4.39), thus allow us to calculate

a chemical specie's ux through the lattice site as:

J = −τ D0 ∇n (4.40)

When applied to the entire lattice, this technique allows us to model the

inhomogeneous diusion of a chemical species. The continuous equations

describing the inhomogeneous diusion of glucose and oxygen were dis-

cretised on to the lattice using the Forward Euler method in time [5],

refer to the Appendices for a description of their discretisation. Once dis-

cretised, the diusion equation were solved by the method of Successive

Overrelaxation (SOR) [32].

Note that if the cells are assumed not to deform under pressure, but

remain spherical, the porosity will not reach zero with increasing density

in either 2D or 3D. However, in 2D the tortuosity will become zero at the

maximum packing density when a path no-longer exists between the cells

50
because all the cells are in contact, thus causing the ux to equal zero. In

3D the tortuosity will not reach zero as a path always exists around the

cells even when all the cells are in contact, which can be seen in Figure

4.5. In 3D, a ux will exist allowing the transportation of molecules

between the cells, albeit at a restricted rate. Although our model is

composed on a 2D lattice, lattice sites represent a 3D slice through the

tumour. A ux will, therefore, exist even when all the cells in a lattice

site are in contact.

4.5.1 The Adiabatic Approximation

Chemical elds are, in principle, time dependent and therefore described

by a PDE, which determines their spatio-temporal evolution. However,

as chemical molecules are negligible in mass compared to the mass of a

typical cell, they reach their equilibrium state on a much shorter time

scale than a cellular distribution. Cell doubling time is of the order of 22

hours [13, 22, 12], whereas the inter-vessel diusion time for lactic acid

and glucose is only of the order of 1-10 seconds [29] (with a mean inter-

vessel distance 10−2 cm and lactic acid diusion coecient of 1.08 × 10−5
(cm)2 (s)−1 ).
The dierence in representative time scales allows us to assume that a

chemical concentration eld will have reached equilibrium (steady state)

between cellular updates. As it is the cells and their evolution that we

are interested in, this observation allows us to ignore the chemical eld's

evolution and directly calculate its steady state, upon which the next

cellular update can be based.

This assumption has also been made by Patel et al. [29] to greatly im-
prove their model's eciency, they termed it the adiabatic approximation.

The adiabatic approximation, however, has certain limits beyond which

it cannot be used, refer to the Appendices for a study of the adiabatic

approximation.

51
Chapter 5

Results

5.1 Simulation Parameters


Tumour growth was simulated from a single cell at the centre of the

lattice. The lattice consisted of 101 × 101 sites, corresponding to a total

area of ≈ 1 cm2 . The tumour is assumed to grow within an extra-cellular

matrix (ECM), a circular region with a radius of 49.5 (100 µm), see Figure
4.1. The simulation terminated when the tumour grew to the edge of the

ECM. Constant concentrations of glucose and oxygen were maintained

outside of the ECM. The simulation's parameter can be found in Table

5.1.

Parameter name Symbol Value Units Ref.

Chemical parameters
−12
Glucose boundary concentration 2.8 × 10 (mol)(100 µm)−3 [13]
Oxygen boundary concentration 3.5 × 10−14 (mol)(100 µm)−3 [13]
Scaled glucose consumption rate γ̄ 1.9 × 10−4 (100 µm)3 (cell)−1 (s)−1
Scaled oxygen consumption rate β̄ 2.4 × 10−3 (100 µm)3 (cell)−1 (s)−1
ECM parameters
IFP p0 0.9333 (kg)(100 µm)−1 (s)−2 [21]
Glucose diusion coecient Dg 0.91 (100 µm)2 (s)−1 [6]
Oxygen diusion coecient Do 0.182 (100 µm)2 (s)−1 [36]
Tumour parameters
Mitotic cycle time 22 hours [13]
Cellular update interval 11.25 minutes
Maximum site density ρ̄ 691 (cells)(100 µm)3 [13]
Cell migration parameters
Viscous drag coecient λ 10 (kg)(s)−1

Table 5.1: The simulation's parameters.

52
5.2 Simulation Results
Initially, the simulation was conducted with the metabolic threshold val-

ues described in Table 4.1. We found that is was not possible to maintain

a stable quiescent layer with the original metabolic thresholds, which can

be seen in Figure 5.1.


Cellular density [(cells)/(100 micrometer)3]

Cellular density [(cells)/(100 micrometer) ]


160

3
PAT 200
PANT
140
PAT
120 150
100

80 100
60

40 50

20

0
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer] Lattice site [100 micrometer]
Cellular density [(cells)/(100 micrometer)3]

Cellular density [(cells)/(100 micrometer) ]


3
250 350
PANT PANT
PAT 300 PAT
200
250
150 200

100 150

100
50
50

10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer] Lattice site [100 micrometer]
Cellular density [(cells)/(100 micrometer)3]

Cellular density [(cells)/(100 micrometer) ]


3

600 600
Dead Dead
PANT PANT
500 PAT 500 PAT
QANT QANT
400 400

300 300

200 200

100 100

10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer] Lattice site [100 micrometer]
Cellular density [(cells)/(100 micrometer)3]

Cellular density [(cells)/(100 micrometer) ]


3

700 700
Dead Dead
600 PANT 600 PANT
PAT PAT
500 QANT 500 QANT

400 400

300 300

200 200

100 100

10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer] Lattice site [100 micrometer]

Figure 5.1: The cross-sections describe the tumour's growth with the
original metabolic thresholds. Going from top to bottom, left to right, the
cross-sections describe the tumour at day ten, twelve, fourteen, sixteen,
eighteen, nineteen, twenty and twenty-one.

Referring to Figure 5.1, the quiescent layer (yellow bar) is seen to form

on day eighteen, then vanish on day nineteen, only to re-emerge again

on day twenty. It was felt that the oscillating behaviour of the quiescent

layer was due to the narrow range of glucose concentrations for which cells

are quiescent (see Figure 4.3). To improve the quiescent layer's stability,

the γqant threshold was lowered to be less than the γpant threshold by

53
an order of magnitude, so that γqant = 5.2 × 10−17 (mol)(cell)−1 (s)−1 ;
Figure 5.2 describes the new metabolic phase diagram.

Glucose concentration @molH100 ΜmL3 D


7. ´ 10-16

6. ´ 10-16 PANT

5. ´ 10-16
PAT
4. ´ 10-16
Βpant

3. ´ 10-16 Γqant
QANT
-16 Γpant
2. ´ 10
Γpat
1. ´ 10-16 QAT
Necrotic
0
4. ´ 10-17 5. ´ 10-17 6. ´ 10-17 7. ´ 10-17 8. ´ 10-17 9. ´ 10-17 1. ´ 10-16

Oxygen concentration @molH100 ΜmL3 D

Figure 5.2: The new metabolic phase diagram describing a cell's


metabolic response to chemical stimuli.

With the new metabolic thresholds, the tumour reached the edge

of the ECM, i.e. a radius of ≈ 5 mm, after 21.28125 days of growth,

which were simulated in approximately 129 minutes. The model was

implemented in the C++ programming language. The simulations were

conducted on an Intel Core 2 Duo CPU (2.40 GHz, with 4 GB RAM)

running a Ubuntu 9.04 virtual machine.

The tumour achieved a maximum site density of 626 (cells)(100 µm)3 ,


which, according to Equation (4.15), equates to a minimum porosity

of 0.03613. The corresponding apparent glucose and oxygen diusion

coecient can be calculated from Equation (4.38) and Equation (4.39)

as 0.17297 (100 µm)2 (s)−1 and 0.03459 (100 µm)2 (s)−1 respectively. The

ratio of the diusion coecient in the uid phase and tumour, for both

glucose and oxygen is 5.26, which, according to Figure 7.7 and 7.8, is

within the limits of the adiabatic approximation.

We will begin by analysing the tumour's growth at a microscopic level,

which will be followed by a macroscopic interpretation. The following

gures describe the composition of a strip of lattice sites, running through

the centre of the lattice, for the period between the tenth and twenty-

rst day of growth. The bars in the plots correspond to the cellular

density and composition of a lattice site. The bar's magnitude describes

54
the lattice site's cellular density and its colour the metabolic state of the

cells located at that lattice site.

Cellular density [cells/(100 micrometer)3]


160
PAT
140

120

100

80

60

40

20

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.3: The tumour's composition at 10 days.

During the rst ten days of growth, the relatively small tumour has

remained in a proliferate aerobic (PAT) state. After ten days, the tumour

has a radius of 2.5 (100 µm).


Cellular density [cells/(100 micrometer)3]

180
PANT
160 PAT
140
120
100
80
60
40
20

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.4: The tumour's composition at twelve days.

After twelve days, the tumour has developed a proliferate anaerobic

(PANT) core surrounded by a rim of PAT cells. Due to a density and

pressure build-up and pressure-induced cell movement, the tumour has

become large enough to cause the oxygen concentration in the centre of

the tumour to drop below the βpat threshold (see Fig. 5.2), thus causing

the cells at the tumour's core to transition to a PANT metabolic state.

55
Cellular density [cells/(100 micrometer)3]
250
PANT
PAT
200

150

100

50

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.5: The tumour's composition at fourteen days.

After fourteen days, the tumour has grown to form a dome shaped

density, centred on the origin of the tumour, with a radius of approxi-

mately 8 (100 µm).


Cellular density [cells/(100 micrometer)3]

350
PANT
300 PAT

250

200

150

100

50

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.6: The tumour's composition at sixteen days.

Referring to Figure 5.6, after sixteen days, the tumour's dome shaped

density is more exaggerated and its radius has grown to approximately

14 (100 µm). Unlike conventional CA tumour models, cells in the bulk

of the tumour are still proliferating: the height of the 'dome' is seen

to increase from fourteen to sixteen days. This increase 'pushes' cells

outwards, causing the tumour to grow in radius. The tumour's radius

grows, primarily, because of the internal proliferation of its bulk.

56
Cellular density [cell/(100 micrometer)3]
600
Dead
PANT
500 PAT
QANT
400

300

200

100

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.7: The tumour's composition at eighteen days.

After eighteen days of growth, a dead (necrotic) core has emerged

at the tumour's centre, surrounded by a layer of quiescent anaerobic

(QANT) cells. The density distribution curve has developed a depression

at the tumour's core, due to the inability of the necrotic and QANT cells

to divide, while cells at the rim of the necrotic core are still dividing.

The mass of PAT and PANT cells has become large enough to cause

the glucose concentration in the centre of the tumour to drop below the

γqant threshold, thus causing the cells at the tumour's core to become

necrotic. As we move from the centre of the tumour to its edge, the

glucose concentration increases. The QANT layer has formed where the

glucose concentration is between the γqant and γpant thresholds.


Cellular density [cells/(100 micrometer)3]

600
Dead
PANT
500 PAT
QANT
400

300

200

100

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.8: The tumour's composition at nineteen days.

Referring to Figure 5.8, unlike the previous simulation, with the origi-

nal metabolic thresholds, the quiescent layer has maintained a consistent

57
width of ve lattice sites between day eighteen and nineteen.

Necrotic and QANT cells are seen to occupy a lattice site on the edge

of the necrotic core. QANT cells have been pushed into the necrotic
1
site .
Cellular density [cells/(100 micrometer)3]

600
Dead
PANT
500 PAT
QANT
400

300

200

100

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.9: The tumour's composition at twenty days.

The nal composition of the tumour can be seen in Figure 5.10.


Cellular density [cells/(100 micrometer)3]

700
Dead
600 PANT
PAT
500 QANT

400

300

200

100

10 20 30 40 50 60 70 80 90 100
Lattice site [100 micrometer]

Figure 5.10: The tumour's composition at twenty-one days.

After the emergence of the necrotic core, the PAT and PANT layers

grow out across the lattice maintaining a relatively consistent width,

approximately 2 and 15 (100 µm) respectively. While the cellular density

of the PANT layer steadily increases, the PAT rim maintains a relatively

consistent cellular density around the critical density.

1 Smaller bars have been superimposed over larger bars, thus obscuring some of

the larger bar.

58
100
200
80

150
60

40 100

20
50

20 40 60 80 100

Figure 5.11: A density plot of the PAT layer at twenty-one days, viewed
from above; cells in the other metabolic states are not shown.

Figure 5.11 describes the extreme of the tumour's spatial expansion

across the lattice. It can be seen that the tumour has a circular shape,

due to the constant concentration of glucose and oxygen on the lattice

perimeter in this simulation; as expected the tumour's growth is not

biased in any one direction.

We will now analyse the tumour morphology at a macroscopic level,

to ascertain what eect the tumour's microscopic evolution had on the

overall system.

3.5 ´ 106
QANT
6
3.0 ´ 10 Dead
PANT
2.5 ´ 106
Population @cellsD

PAT
6
2.0 ´ 10 Total

1.5 ´ 106

1.0 ´ 106

500 000

0
0 5 10 15 20
Time @daysD

Figure 5.12: The change in size of the tumour's metabolic populations


with time.

The QANT and PAT populations grow relatively slowly in comparison

59
to the PANT and dead populations that exhibit rapid growth. At the

end of the simulation, the dead population has grown to be greater than

the PANT population.

106 QANT
Dead
PANT
PAT
Population @cellsD

4
10 Total

100

1
0 5 10 15 20
Time @daysD

Figure 5.13: The change in size of the tumour's metabolic populations


with time, plotted on a logarithmic scale.

Referring to Figure 5.13, the PAT population size increases exponen-

tially until, approximately, day twelve (obscured by the blue line). The

end of the PAT population's exponential growth corresponds with the

creation of cells in a PANT state. The PANT population then grows

exponentially, from (approximately) day thirteen to eighteen. The end

of the PANT population's exponential growth corresponds with the cre-

ation of cells in a QANT and dead state.

The tumour exhibits an exponential growth rate in its cellular popu-

lation. The growth rate slows at approximately eighteen days, with the

development of a necrotic core. The tumour achieves a maximum cellular

population of approximately 4.8 × 106 , after twenty-one days of growth.

60
50

40
Tumour radius @H100 ΜmLD

30

20

10

0
0 5 10 15 20
Time @daysD

Figure 5.14: The growth in tumour radius with time.

Referring to Figure 5.14, at the start of the simulation all the cells

are located in the central lattice site that was seeded with the original

cell. While the central site's density is less than the critical density, the

pressure is not sucient to push cells into neighbouring sites, thus all

the cells remain in the central site; the tumour has a radius less than our

lattice resolution of 100 µm. When the density of the central site reaches

the critical density the tumour begins to grow out across the ECM.

61
50

Tumour radius @H100 ΜmLD


20

10

1
10 15 20
Time @daysD

Figure 5.15: The growth in tumour radius with time, plotted on a loga-
rithmic scale.

Referring to Figure 5.15, the tumour's radius is seen to exhibit ex-

ponential growth for the rst seventeen days, before slowing to linear

growth.

600
Density @HcellsLH100 ΜmL3 D

500

400

300

200

100

0
0 5 10 15 20
Time @daysD

Figure 5.16: The change in maximum cellular density with time.

Referring to Figure 5.16, initially, the maximum cellular density de-

scribes the density of the central site. The maximum cellular density

grows exponentially for the rst seven days, which has been checked by

62
plotting the change in maximum cellular density with time on a logarith-

mic axis (gure not included). The point at which the maximum cellular

density equals the critical density, the seventh day, corresponds with the

point at which the tumour radius begins to grow beyond the central site,

see Figure 5.14.

After exceeding the critical density, the maximum cellular density

grows slowly. The large dierence in pressure between the central site

and its neighbours causes most new cells, created in the central site, to

be pushed into its neighbouring sites. Cell movement acts to homogenise

the pressure. As pressure dierences equalise, less cells are pushed out of

the central site and the maximum cellular density begins to grow more

rapidly.

Growth in the maximum cellular density then slows at, approxi-

mately, eighteen days corresponding to the onset of cells in a quiescent

and necrotic state. The tumour eventually achieves a maximum cellu-

lar density of 626 (cells)(100 µm)−3 , which is below the maximum site
density (ρ̄) of 691 (cells)(100 µm)−3 . An additional experiment was per-
formed, to observe the simulated tumour's behaviour on reaching the

maximum site density.

The new maximum site density (ρ̄) was set to equal 300 (cells)(100 µm)−3 ,
all other variables were the same as in the previous experiment. A density

of 300 (cells)(100 µm)−3 is on the lower bound of the range of tumour

densities recorded experimentally by [13].

63
50

Tumour radius @H100 ΜmLD


20

10

1
10 15 20 25
Time @daysD

Figure 5.17: The change in tumour radius with time, for a maximum site
−3
density of 300 (cells)(100 µm) , plotted on a logarithmic scale.

106

105
Population @cellsD

104

1000

100

10

1
0 5 10 15 20 25
Time @daysD

Figure 5.18: The change in cellular population with time, for a maximum
−3
site density of 300 (cells)(100 µm) , plotted on a logarithmic scale.

Some sites achieve the maximum site density at approximately four-

teen days, which has been seen by plotting the maximum site density

with time (gure not included). On reaching the maximum site den-

sity, the growth rate of both the tumour's radius, Figure 5.17, and the

tumour's cellular density, Figure 5.18, transitions from exponential to

64
linear growth.

5.3 Comparison with in vitro Experiments


In the later stages of tumour growth, once a nutrient concentration gra-

dient has formed, the simulated tumour's morphology resembles those

seen in experimental histologies [38], and described by Figure 2.1. The

tumour is composed of a necrotic core, surrounded by cells in a quiescent

state, with proliferating cells located on its rim.

The tumour's morphology was also compared on quantitative aspects,

with data found in the literature. Freyer and Sutherland [13], report on

assays of multicellular spheroid of the EMT6/Ro mouse mammary tu-

mour cell line. Many of the model parameters' values were referenced

from Freyer and Sutherland, such as maximum site density, medium nu-

trient concentrations and the mitotic cycle time. It is therefore appropri-

ate to compare the simulated tumour's morphology with those reported

in [13].

Freyer and Sutherland report that after ve days of growth, the av-

erage diameter of their tumour spheroid grew linearly, at a rate of 75

µm per day [13, Page 519]. In comparison, our tumour's diameter grew

linearly after approximate thirteen days (Fig. 5.15), at a rate of approxi-

mately 400 µm per day. The larger growth rate of our tumour has meant

we are not able to quantitatively replicated Freyer and Sutherland's nd-

ing, but by demonstrating a linear growth rate in diameter, our model

has been able to qualitatively replicate the behaviour of those tumours

observed experimentally.

The average number of cells per spheroid, as a function of time, were

also described in [13]. The number of cell per spheroid as a function of

time, plotted on a logarithmic scale, indicates that the growth rate of

the cellular population goes from exponential growth to linear growth

at, approximately, twelve days [13, Fig. 1B]. Again, our tumour's larger

growth rate has caused its cellular population to exceed those of Freyer

and Sutherland. Our model was, however, able to replicate the transition

from an exponential growth rate in cellular population to a linear growth

rate, which is conrmed by Figure 5.18.

Freyer and Sutherland also report data regarding the emergence of

a necrotic core. They report that necrosis rst developed in tumour

spheroids with a diameter of 400 µm, but it was not until the tumour
had achieved a diameter of 1000 µm that the necrotic fraction constituted

65
a signicant portion. Further growth of their tumour spheroids resulted

in a large increase in the relative amount of necrotic material [13, Page

521]. In contrast, our tumour's diameter was approximately 5000 µm


when a signicant necrotic core rst developed (see Figure 5.7). Like

those tumours studied by Freyer and Sutherland, the relative amount

of necrotic material then steadily increased until it eventually came to

dominate the tumour, see Figure 5.12.

Our model has been able to qualitatively replicate many of the phe-

nomena observed experimentally, such as a linear growth rate in diame-

ter, a transition from exponential to linear growth in cellular population

and the growth of a necrotic core that eventually dominates the tumour.

66
Chapter 6

Conclusion

The aim of this project was to develop an approach to modelling pressure-

induced cell movement and tumour invasion. When the cells in one region

of a tumour proliferate more quickly than the cells in another, density

dierences result. Density dierences lead to a build-up of pressure that

pushes cells from densely populated regions to sparsely populated regions.

This pushing eect grows the tumour outwards into the surrounding

healthy tissue, thus promoting tumour invasion.

It is inherently dicult to model pressure-induced cell movement and

proliferation in a traditional CA tumour model. Traditional CA tumour

models are formulated on a xed, regular lattice, where each lattice site

represents a cell. The restrictions imposed by the model's structure re-

quire a dividing cell be near an empty site in which the new cell can be

placed. It is not possible for density dierence to develop, except at the

tumour's boundary, when only cells near empty sites can divide.

In this project, a CA tumour model that captures realistic pressure-

induced cell movement and tumour growth has been developed. By allow-

ing several cells to occupy one lattice site, we have been able to introduce

local cell density information into a CA model. Based on physical prin-

ciples and porous media theory, a relationship between pressure and cell

density has been proposed that has allowed us to model pressure-induced

cell movement and the inhomogeneous diusion of nutrient molecules.

Applying this model to in vitro conditions, we have simulated tumour


growth that qualitatively replicates experimentally observed phenomena,

such as the emergence of the typical layered structure with expanding

radius. However, quantitative aspects of simulated tumour growth were

found not to reect reality with the current model parameter estimates.

The simulated tumour grew much faster than those observed by Freyer

and Sutherland [13], having a larger growth rate in diameter and cellular

67
population. Performing a parametric study to ascertain what eect vary-

ing the model's parameters has on the simulated tumour's growth would

provide a more quantitative understanding of the mechanisms underlying

the model, and enable us to identify a set of parameters that will cause

the simulated tumour's growth to better reect reality.

Designing the model, we neglecting to model chemotaxis induced cell

movement and the production of hydrogen ions, as a by-product of cell

metabolism. Chemotaxis directly eects a cell's movement and hydrogen

ion concentration (pH) eects a cell's viability. We intend to incorporate

these factors into our model, as they both inuence tumour growth.

This model constitutes the rst stage of a larger goal to create a CA

model of solid tumour growth in the bone environment, which can be

used to simulate prostatic tumour growth in bone. The present model

will need to be adapted to investigate prostatic tumour growth in bone,

by modelling pressure proles in patterns found in the bone environment.

It would then be possible to seed the model with prostate metastases and

simulate tumour growth under varying conditions.

It is expected that hard tissue, such as bone, will constrain a tu-

mour's growth, causing a buildup of inter-cellular pressure within the

tumour. Under such circumstances, inter-cellular pressure may have a

dominant inuence over the tumour's growth, and therefore needs to be

taken into account. It is hoped that this CA tumour model will become

an invaluable tool for studying solid tumour growth.

68
Bibliography

[1] T. Alarcon, H. M. Byrne, and P. K. Maini. A cellular automaton

model for tumour growth in inhomogeneous environment. Journal


of Theoretical Biology, 225(2):257274, 2003.
[2] Alexander R.A. Anderson, Alissa M. Weaver, Peter T. Cummings,

and Vito Quaranta. Tumor morphology and phenotypic evolu-

tion driven by selective pressure from the microenvironment. Cell,


127:905915, 2006.

[3] R.B. Bird, W.E. Stewart, and E.N. Lightfoot. Transport Phenom-
ena. New York: Wiley, 2002.

[4] Nino Boccara. Modeling Complex Systems. Springer, 2004.

[5] J.C. Butcher. Numerical methods for ordinary dierential equations.


Wiley, 2008.

[6] J.J. Casciari, S.V. Sotirchos, and Sutherland R.M. Glucose diu-

sivity in multi-cellular tumor spheroids. Cancer Res., 48:39053909,


1988.

[7] Ralph J. DeBerardinis, Julian J. Lum, Georgia Hatzivassiliou, and

Craig B. Thompson. The biology of cancer: Metabolic reprogram-

ming fuels cell growth and proliferation. Cell Metabolism, 7:1120,

2008.

[8] L. Dormieux and E. Lemarchand. Macroscopic modeling of the dif-

fusive transport: Contribution of upscaling techniques. Oil & Gas


Science and Technology - Rev. IFP, 55:1534, 2000.
[9] William H. Elliott and Daphne C. Elliott. Biochemistry and
Molecule Biology. Oxford University Press, 2009.

[10] D.J. Evans and G. Morriss. Statistical mechanics of nonequilibrium


liquids. Cambride University Press, 2008.

69
[11] S. C. Ferreira, M. L. Martins, and M. J. Vilela. Reaction-diusion

model for the growth of avascular tumor. Phys. Rev. E, 65(2):021907,


Jan 2002.

[12] J.P. Freyer and R.M. Sutherland. Selective dissociation and char-

acterization of cells from dierent regions of multicell tumour

spheroids. Cancer Res., 40:39563965, 1980.


[13] J.P. Freyer and R.M. Sutherland. A reduction in the in situ rates of

oxygen and glucose consumption of cells in emt6/ro spheroids during

growth. J. Cell. Phys., 124:516524, 1985.


[14] Robert A. Gatenby and Robert J. Gillies. Why do cancers have high

aerobic glycolysis? Nature reviews Cancer, 4:891899, 2004.


[15] P. Gerlee and Alexander A.R. Anderson. An evolutionary hybrid

cellular automaton model of solid tumour growth. J. Theor. Biol.,


246(4):583603, 2007.

[16] B. Gompertz. On the nature of the function expressive of the law

of human mortality, and on a new mode of determining the value

of life contingencies. Phil. Trans. Roy. Soc. London, 123:513585,

1832.

[17] T. C. Hales. A computer verication of the kepler conjecture. Pro-


ceedings of the Internation Congress of Mathematicians, 2:795804,
2002.

[18] Matthew G. Vander Heiden, Lewis C. Cantley, and Craig B. Thomp-

son. Understanding the warburg eect: The metabolic requirements

of cell proliferation. Science, 324:10291033, 2009.


[19] W. Helfrich. Elastic properties of lipid bilayers - theory and possible

experiments. Z. Naturforsch, C 28:693703, 1973.


[20] Fritz Jahnig. What is the surface tension of a lipid bilayer mem-

brane? Biophysical Journal, 71:13481349, 1996.


[21] R.K. Jain. Transport of molecules in tumor interstitium: a review.

Cancer Res, 47:30393051, 1987.


[22] Yi Jiang, Jelena Pjesivac, Charles Cantrell, and James Freyer. A

multiscale model for avascular tumor growth. Biophysical Journal,


89:38733883, 2005.

70
[23] A. R. Kansal, S. Torquato, G. R. Harsh IV, E. A. Chiocca, and

T. S. Deisboeck. Cellular automaton of idealized brain tumor growth

dynamics. Biosystems, 55(1-3):119127, 2000.

[24] R.S. Kerbel. Growth dominance of the metastatic cancer cell: cel-

lular and molecular aspects. Adv. Cancer Res., 55:87132, 1990.

[25] Daniel R. Lynch. Numerical Partial Dierential Equations for


Environmental Scientists and Engineers: a rst practical course.
Springer, 2005.

[26] Joana Moreira and Andreas Deutsch. Cellular automaton models of

tumor development: A critical review. Advances in Complex Sys-


tems, 5:247267, 2002.

[27] John von Neumann. Design of Computers, Theory of Automata and


Numerical Analysis. Pergamon Press, 1948.

[28] P.C. Nowell. The clonal evolution of tumor cell populations. Science,
194 194:2328, 1976.

[29] Aalpen A. Patel, Edward T. Gawlinski, Susan K. Lemieux, and

Robert A. Gatenby. A cellular automaton model of early tumor

growth and invasion: The eects of native tissue vascularity and

increased anaerobic tumor metabolism. Journal of Theoretical Bi-


ology, 213(3):315331, 2001.

[30] Monika Joanna Piotrowska and Simon D. Angus. A quantitative

cellular automaton model of in vitro multicellular spheroid tumour

growth. Journal of Theoretical Biology, 258(2):165178, 2009.

[31] Karl R. Popper.Conjectures and Refutations: The Growth of Sci-


entic Knowledge. Routledge, 1963.

[32] W.H. Press, S.A. Teukolsky, W.T. Vetterling, and B.P. Flannery.

Numerical Recipes in C. Cambridge University Press, Cambridge,

UK., 1992.

[33] C. Song, P. Wang, and H.A. Makse. A phase diagram for jammed

matter. Nature, 453:629632, 2008.

[34] Z.C. Tu. Elastic theory of membranes. AAPPS Bulletin, 16:3033,


2006.

71
[35] Z.C. Tu and Z.C. Ou-Yang. Elastic theory of low-dimensional con-

tinua and its applications in bio- and nano-structures. J. Comput.


Theor. Nanosci., 5:422448, 2008.
[36] Raja Venkatasubramanian, Michael A. Henson, and Neil S. Forbes.

Incorporating energy metabolism into a growth model of multicel-

lular tumor spheroids. Journal of Theoretical Biology, 242(2):440 


453, 2006.

[37] B. Vogelstein, D. Lane, and A.J. Levinne. Surng the p53 network.

Nature, 408:307310, 2000.


[38] S. Walenta, J. Doetsch, W. Mueller-Klieser, and L.A. Kunz-

Schughart. Metabolic imaging in multicellular spheroids of

oncogene-transfected broblasts. J. Histochem. Cytochem,


48(4):509522, 2000.

[39] O. Warburg. The metabolism of tumors. The American Journal of


Sciences, 182(1):123, 1930.
[40] O. Warburg, F. Wind, and E. Negelein. The metabolism of tumors

in the body. Journal of General Physiology, pages 519530, 1927.


[41] T.E. Wheldon. Mathematical models in cancer research. Adam

Hilger, 1988.

[42] Wikipedia. Sphere packing. 2010.

[43] Joshua Zimmerberg and Michael M. Kozlov. How proteins produce

cellular membrane curvature. Nature Reviews|Molecular Cell Biol-


ogy, 7:919, 2006.

72
Chapter 7

Appendices

7.1 An Alternative Relationship Describing


Cell Volume as a Function of Pressure
Cell volume, as a function of the pressure experienced on its membrane,

has been investigated theoretically on robust grounds. The surface of

a cell is formed by lipids that constitute the membrane bi-layer, the

shape of a cell is therefore determined by its membrane's shape [43]. The

common understanding of the energetics of membrane shaping are largely

based on the seminal Helfrich theory of membrane bending elasticity

[19], from which Helfrich was able to describe a spherical cell's shape

as a function of pressure, refer to Table 7.1 for a descriptions of the

parameters.

k c c0 2 2λ
p(R) = ( − c0 − ) (7.1)
R R k c c0

Description Symbol Value Units Ref.


−1 −2
Osmotic pressure p (kg)(100 µm) (s) [35]
Cell radius R (100 µm) [35]
Bending moduli kc 20 kB T J [34]
Spontaneous curvature c0 1/R0 (100 µm)−1 [34]
Surface tension λ 0 (J)(100 µm)−1 [20]
Scaled Boltzmann constant kB 1.38 × 10−15 (kg)(100 µm)2 (s)−2 (K)−1 [35]
Room temperature T 298 K [35]
Spontaneous radius R0 (100 µm)

Table 7.1: A description of parameters describing a spherical cell's shape


as a function of pressure.

The membrane composition controls the non-deformed unstressed

73
shape of the membrane, which is referred to as the 'spontaneous shape',

which is characterised by the spontaneous curvature. The spontaneous

curvature of the whole membrane depends on the spontaneous curvature

of the inner and outer membrane monolayers [43].

Note, Equation 7.1 describes the osmotic pressure across a cell's mem-

brane, which is the dierence between the external and internal cell pres-

sure. By using this equation to describe only the external pressure, we

are making the implicit assumption that the internal pressure is zero,

which is a valid assumption if the internal cell pressure is very low.

Assuming that cells are spherical vesicles that conserve their spheri-

cal shape, but are allowed to modify their radius and therefore volume,

according to the pressure experienced on their membrane allows us to

determine a cell's volume as a function of pressure. The volume of a

sphere, as a function of its radius:

4 π R3
Vsphere (R) = (7.2)
3
Resolving Equation (7.1) for R and combining with Equation (7.2), yields
a relationship for cell volume as a function of pressure:

1
Vcell (p) = p (7.3)
(β1 + β12 + β2 p)3
Where:

β4
• β1 = √
3
β3
, with units of (100 µm)−1 .

• β2 = √1
( 3 β3 )2
, with units of (J)(100 µm)−1 .
q √
• β3 = 3 4π
3
( 2 kc c0 )3 , with units of (J)3/2 (100 µm)−3/2 .
√ c0
2 k c c0 + k λc
• β4 = 2
2 c 0
, with units of (J)1/2 (100 µm)−3/2 .

The pressure as a function of cell volume can be found from the inverse

of Equation (7.3) as:

−1 −1
V 3 (V 3 − 2 β1 )
p(Vcell ) = cell cell (7.4)
β2
In-order to solve Equation (7.3) it is necessary to calculate the sponta-

neous curvature (c0 ). As only the spontaneous curvature is unknown,

it is, in principle, possible nd the spontaneous curvature algebraically.

However, we found that no real solutions for c0 exist for the intrinsic

volume.

74
1.4 ´ 10-10

Cell volume @H100 ΜmL3 D 1.2 ´ 10-10

1. ´ 10-10

8. ´ 10-11

6. ´ 10-11

4. ´ 10-11

2. ´ 10-11

0
0 2000 4000 6000 8000 10 000

c0 @H100 ΜmL-1 D

Figure 7.1: The change in cell volume with spontaneous curvature.

Indeed, referring to Figure 7.1, it can be seen that the curve describing

cell volume as a function of c0 never reaches the intrinsic cell volume

(Vcell0 ) of 0.00419 (100 µm)3 , no real solutions of c0 exist. It appears

that Equation (7.1) describes the shape of a empty lipid bi-layer, not one

containing cellular material, as the intrinsic shape is much smaller than

that of a cell.

Alternatively the shape of a cell can be described phenomenologically.

Equation (7.3) describes a cell's volume as a function of β1 , β2 and p. We

wish to nd values for β1 and β2 , so that Vcell (p0 ) = Vcell0 , the cell volume
equals the intrinsic cell volume for the IFP of normal tissue. This can

be achieved by solving Equation (7.3) for β2 , and then substituting the

intrinsic cell volume (Vcell0 ) for Vcell and IFP of normal tissue (p0 ) for p:

−2 −1
(V 3 − 2 β1 Vcell30 )
β2 (β1 ) = cell0 (7.5)
p0
Since β2 ≥ 0 (see Eq. (7.3)), the range of acceptable values for β1 can be

found from Equation (7.5) as:

−2 −1
Vcell30 − 2 β1 Vcell30 > 0 (7.6)

which can be solved as:

−1
V 3
β1 < cell0 = 3.10175 (100 µm)−1 (7.7)
2
The change in cell volume with β1 can be found by substituting Equation

75
(7.5) into Equation (7.3):

1
Vcell (p) = p
2
(7.8)
(β1 + β1 + β2 (β1 ) p)3
Figure (7.2) and (7.3) describe how cell volume changes with pressure

when β1 is in the range 1 − 3, the most dramatic change in cell volume

occurs when β1 = 1.

0.0050
Β1 = 1
0.0048
Cell volume @H100 ΜmL3 D

Β1 = 2
0.0046

0.0044
Β1 = 3
0.0042

0.0040

0.080 0.085 0.090 0.095 0.100

Pressure @kg  H100 ΜmLHsL2 D

Figure 7.2: Cell volume as a function of pressure, according to Equation


(7.8).

0.12

0.10
Cell volume @H100 ΜmL3 D

0.08
Β1 = 1
0.06

0.04

0.02 Β1 = 2
Β1 = 3
0.00
0.00 0.02 0.04 0.06 0.08 0.10

Pressure @kg  H100 ΜmLHsL2 D

Figure 7.3: Cell volume as a function of pressure, according to Equation


(7.8) for a large range of pressure values.

76
0.12

Cell volume @H100 ΜmL3 D 0.10

0.08

0.06

0.04

0.02

0.00
0.00 0.02 0.04 0.06 0.08 0.10

Pressure @kg  H100 ΜmLHsL2 D

Figure 7.4: A comparison of the alternative approaches for describing


cell volume as a function the pressure experienced on its membrane.
The green line denotes the inverse relationship between cell volume and
pressure, described by Equation (4.11). The blue lines denote the volume
of an empty lipid bi-layer as a function of pressure, described by Equation
(7.1).

Referring to Figure 7.4, it can be seen that the two equations de-

scribing a cell's volume as a function of pressure, Equation (4.11) and

Equation (7.8), are similar when β1 = 1 for Equation (7.8).

The average inter-cellular pressure as a function of cellular density

can be found by substituting Equation (4.16) into Equation (7.4), which

yields:

2 1
( 1−φ
ρ
)− 3 − 2 β1 ( 1−φ
ρ
)− 3
p(ρ) = (7.9)
β2
Figure 7.5 describes how average inter-cellular pressure changes with

pressure when β1 is in the range 1 − 3, the most dramatic change in

average inter-cellular pressure occurs when β1 = 3.

77
7
Β1 = 3
Pressure @kg  H100 ΜmLHsL2 D 6

2
Β1 = 2
1
Β1 = 1
0
0 200 400 600 800 1000 1200 1400

Density @cells  H100 ΜmL3 D

Figure 7.5: The average inter-cellular pressure as a function of density,


from Equation (7.4), assuming cells deform under increasing pressure.

2.0
Pressure @kg  H100 ΜmLHsL2 D

1.5

1.0

0.5

0.0
0 200 400 600 800 1000 1200 1400

Density @cells  H100 ΜmL3 D

Figure 7.6: A comparison of the alternative approaches for describing


average inter-cellular pressure as a function of density. The green line
denotes the pressure as a function of density for the inverse relationship
between cell volume and pressure, described by Equation (4.11). The
blue lines denote the pressure as a function of density for an empty lipid
bi-layer, described by Equation (7.1).

78
7.2 Discretising a Continuous Chemical Con-
centration
Below, our method of discretising a continuous chemical concentration

on to the model lattice will be discussed. The method will be described

for the glucose concentration, which is the same for that of the oxygen

concentration. When the apparent glucose diusion coecient (Dg ) is

invariant with position, such as for an inhomogeneous cellular density,

the continuous glucose concentration is expressed as:

∂g
= ∇ · (Dg ∇ g) (7.10)
∂t
or

∂g(x, y, t) ∂ ∂g(x, y, t) ∂ ∂g(x, y, t)


= Dg (x, y) + Dg (x, y) (7.11)
∂t ∂x ∂x ∂y ∂y

Using the Forward Euler method in time [5], the derivative on the

left-hand side can be approximated by:

∂g(x, y, t) g(x, y, t + 4t) − g(x, y, t)


≈ (7.12)
∂t 4t
Using the central dierence approximation for the second derivatives,

taken from [25, Eq. 4.23], the rst term on the right-hand side of Equa-

tion (7.11), can be expressed as:

h

∂x
Dg (x, y) ∂g(x,y,t)
∂x
≈ i
1 Dg (xi+1 ,y)+Dg (x,y) g(xi+1 ,y,t)−g(x,y,t) Dg (x,y)+Dg (xi−1 ,y) g(x,y,t)−g(xi−1 ,y,t)
4x
(( 2
)( 4x
)) − (( 2
)( 4x
))

Simplifying:


∂x
Dg (x, y) ∂g(x,y,t)
∂x
≈ (0.5 (Dg (xi+1 ,y)+Dg (x,y))
(4x)2
(g(xi+1 ,y,t)−g(x,y,t))

(0.5 (Dg (x,y)+Dg (xi−1 ,y)) (g(x,y,t)−g(xi−1 ,y,t))
(4x)2

Equation (7.11) can be discretised as:

g(x, y, t + 4t) =
4t
g(x, y, t)+ (4x)2 [((0.5 (Dg (xi+1 , y) + Dg (x, y)) (g(xi+1 , y, t) − g(x, y, t)))−
((0.5 (Dg (x, y) + Dg (xi−1 , y)) (g(x, y, t) − g(xi−1 , y, t)))] +
4t
(4y)2
[((0.5 (Dg (x, yj+1 ) + Dg (x, y)) (g(x, yj+1 , t) − g(x, y, t))) −
((0.5 (Dg (x, y) + Dg (x, yj−1 )) (g(x, y, t) − g(x, yj−1 , t)))]

79
Press et al. [32] describe a heuristic stability criterion of:

(4x)2
 
4t ≤ mini (7.13)
2 (0.5 (Dg (xi+1 , y) + Dg (x, y))
or

(4y)2
 
4t ≤ minj (7.14)
2 (0.5 (Dg (x, yj+1 ) + Dg (x, y))

7.3 The Adiabatic Approximation


7.3.1 A Theoretical Analysis of the Adiabatic Ap-

proximation

Below, the theoretical basis of the adiabatic approximation is explored.

Glucose is chosen as our representative chemical concentration for this

discussion, equally oxygen or any other chemical consumed or produced

by tumour cells could have been used. The spatio-temporal evolution of

glucose is described by Equation (4.34), where the consumption rate γσ


is the same for all cell types. Dividing Equation (4.34) by γσ yields:

   
1 ∂g Dg γv
=∇ ∇ g − ρσ g + ρv (7.15)
γσ ∂t γσ γσ
Assuming γσ and γv are of the same order of magnitude, the terms on

the right-hand side will be of order 1 whereas the term on the left-hand

side will be of order 1/γσ . If is it assumed that γσ  1, then the change

in glucose concentration with respect to time will be negligible, so that:

   
Dg γv
0=∇ ∇ g − ρσ g + ρv (7.16)
γσ γσ
Thus, the glucose concentration eld will instantaneously reach a steady

state in our model when the tumour cell consumption rate is high.

Let us now assume that the glucose diusion coecient is low, such

that Dg  1, which will occur when the cellular density is high. Now,

only the glucose production and consumption terms will be of order 1.

The glucose concentration and diusion terms will be of order 1/γσ , if

we again assume γσ  1:

γσ ρ g = γv ρv (7.17)

When the cellular density is high, the diusion into and out of a

lattice site with respect to time will be negligible, as will be the change

80
in glucose concentration. Glucose consumption will be equal to glucose

production, if glucose is not produced within a lattice site, cells within

that lattice site will not consume any glucose and will eventually become

necrotic. This nding supports the hypothesis that an avascular tumour's

necrotic core is caused by its increased cellular density.

Even with a small diusion coecient, diusion into a lattice will

occur given a long enough time interval. However, when the time interval

is increased so that it approximates that of the cellular time scale, the

adiabatic approximation will be invalidated, because we can no-longer

assume the chemical concentration has reached equilibrium.

7.3.2 A Numerical Analysis of the Adiabatic Ap-

proximation Limit

The theoretical analysis of the adiabatic approximation does not apply

to a necrotic tumour. Necrotic tumours do not consume nutrients and

therefore have a consumption rate of zero. A numerical analysis enabled

us to quantitatively assess what eect a necrotic tumour's cellular density

and radius has on the time taken for a glucose and oxygen concentration

eld to reach equilibrium.

The numerical analysis was conducted on a square lattice with con-

stant chemical concentration at the perimeter and zero concentration

within the tumour growth region. A tumour composed of necrotic tissue

was modelled at the lattice centre. Over a series of tests, the tumour's

cellular density was exponentially increased while the tumour's radius

was incrementally increased. The increase of a lattice site's cellular den-

sity was realised through a decrease in its diusion coecients. The

time take for the chemical concentration elds to reach equilibrium were

recorded for each test.

As a necrotic tumour does not consume glucose or oxygen, given a

long enough time, a homogeneous concentration of glucose and oxygen

would form across the entire lattice. To reduce simulation time, a chemi-

cal steady state is dened as a state where the chemical concentration of

any lattice site does not change by more than 10−6 % from one iteration

to the next.

Table 7.2 and 7.3 describes the experiment's results for the glucose

and oxygen concentration elds respectively.

81
Tumour radius [100 um]
Dg0 /Dg 5 10 15 20 25 30

1 1.46 1.46 1.46 1.46 1.46 1.46


2 1.52 1.67 1.88 2.11 2.32 2.5
4 1.64 2.14 2.78 3.42 4 4.47
8 1.93 3.14 4.56 5.93 7.14 8.13
16 2.58 5.12 7.92 10.6 12.96 14.88
32 3.96 8.88 14.17 19.19 23.6 27.2
64 6.65 15.85 25.6 34.81 42.91 49.5

Table 7.2: The glucose steady state time in hours.

Tumour radius [100 µm]


Do0 /Do 5 10 15 20 25 30

1 6.26 6.26 6.26 6.26 6.26 6.26


2 6.51 7.16 8.05 9.01 9.89 10.61
4 7.04 9.12 11.79 14.45 16.81 18.73
8 8.26 13.28 19.09 24.68 29.6 33.61
16 10.98 21.4 32.72 43.43 52.83 60.48
32 16.66 36.55 57.5 77.17 94.37 108.38
64 37.59 64.04 101.77 136.91 167.52 192.46

Table 7.3: The oxygen steady state time in hours.

Where Dg0 /Dg and Do0 /Do are the ratio of the glucose and oxygen

diusion coecients in the uid phase (in the absence of cells) to the

diusion coecients within the tumour. Figure 7.7 and 7.8 graphically

describe how the time taken for the glucose and oxygen concentration

elds to reach equilibrium changes with the tumour's density and radius.

82
The Glucose Steady State Time for a Necrotic Tumour
30 45

40
Tumour radius [100 micrometers] 25
35

20 30

25
15 20

15
10
10
Steady state time [h]
5 5
1 2 4 8 16 32 64
D0/D1

Figure 7.7: The steady state contours for the glucose concentration eld.

The Oxygen Steady State Time for a Necrotic Tumour


30 180

160
Tumour radius [100 micrometer]

25
140

20 120

100
15 80

60
10
40
Steady state time [h]
5 20
1 2 4 8 16 32 64
D0/D1

Figure 7.8: The steady state contours for the oxygen concentration eld.

Figure 7.7 and 7.8 describes an upper-bound on the adiabatic approx-

imation assumption that underlies the modelling of chemical diusion.

Provided the time for a chemical eld to reach equilibrium does not ex-

ceed the model's representative time, the mitotic cycle time (cell doubling

time), the adiabatic approximation will hold and enable us to ignore a

chemical concentration eld's evolutions and directly calculate its steady

state.

83

Vous aimerez peut-être aussi