Computational modelling of

embryogenesis driven by empirical

evidence from 4D microscopy images

Joel Dokmegang

PhD 2020
 Manchester Metropolitan
University

Computational modelling of
embryogenesis driven by empirical
evidence from 4D microscopy images

Joel Dokmegang

A thesis submitted in partial fulfilment of the requirements of

Manchester Metropolitan University for the degree of Doctor of
Philosophy

Centre for Advanced Computational Science (CfACS)

Department of Computing and Mathematics
Faculty of Science and Engineering
Manchester Metropolitan University

2020
 Abstract

The development of multicellular organisms remains one of the most

enduring puzzles of science. While wet lab methods have proven effective in
unravelling multiple mechanisms, much is yet to be discovered. Computer
simulations are increasingly used in this context and present over wet lab
experiments the advantages of simplicity, reduced risk and total control over
experimental conditions and parameters. Hence the need for more computa-
tional models and studies establishing their usefulness for biologists. In this
work, we present a novel agent-based computational model of cell and tissue
mechanics (MG#) of the family of Deformable Cell Models, able to simulate
various phenomena in morphogenesis. Furthermore, we show that MG# can
be extended to couple mechanical and chemical variables describing the dy-
namics of a cell within a unified framework. Using MG#, we reproduce key
morphological events of mouse implantation and, for the first time, provide
theoretical evidence that trophectoderm morphogenesis can regulate epiblast
shape upon implantation. Moreover, enriching centre-based models with a
polarity term, we show that directed cell behaviours are a sufficient drive
for zebrafish fin development. Together, the results presented in this thesis
offer key insights into morphogenesis, highlight the usefulness of agent-based
modelling methods in the study of embryogenesis, and propose new math-
ematical models, and computational tools purposed for the investigation of
early development.
Contents

Abstract iii

Contents iii

1 Introduction 5

2 Computational models of biological development 11

2.1 A brief history of mathematical and computational models of
development . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Models of biological development . . . . . . . . . . . . . . . 13
2.2.1 Models of cell mechanics . . . . . . . . . . . . . . . . 14
2.2.2 Coupling mechanical and chemical variables . . . . . 20
2.3 The validation feedback loop between observations and models 21

3 MG#: A modelling framework and simulation platform for

cell and tissue mechanics 25
3.1 Computational implementation and Preliminary results . . . 30
3.1.1 Cell shapes . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.1.1 Red blood cell shapes . . . . . . . . . . . . 33
3.1.1.2 Cell elongation . . . . . . . . . . . . . . . . 34
3.1.1.3 Epithelial cell shapes . . . . . . . . . . . . . 36
3.1.2 Apoptosis . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1.3 Cell polarity . . . . . . . . . . . . . . . . . . . . . . . 40
3.1.4 Cell cycle and division . . . . . . . . . . . . . . . . . 41
3.2 MG#: The open simulation framework . . . . . . . . . . . . 43

4 Modelling of epithelial morphogenesis 47

4.1 Morphological changes in single epithelial cells . . . . . . . . 49
4.1.1 Planar Polarised Constriction . . . . . . . . . . . . . 49
4.1.1.1 Biological motivation . . . . . . . . . . . . . 49

v
Contents vi

4.1.1.2 Computational modelling . . . . . . . . . . 50

4.1.2 Apical Constriction & Apical Expansion . . . . . . . 53
4.1.2.1 Biological motivation . . . . . . . . . . . . . 53
4.1.2.2 Computational modelling . . . . . . . . . . 54
4.1.3 Apical constriction with volume conversation: basal-
lateral modulation . . . . . . . . . . . . . . . . . . . 59
4.1.3.1 Biological motivation . . . . . . . . . . . . . 59
4.1.3.2 Computational modelling . . . . . . . . . . 59
4.1.4 Integrating chemical variables: 𝐶𝑎-controlled Apical
Constriction . . . . . . . . . . . . . . . . . . . . . . . 67
4.1.4.1 Biological motivation . . . . . . . . . . . . . 67
4.1.4.2 Computational modelling . . . . . . . . . . 68
4.2 Morphogenesis in epithelial tissues . . . . . . . . . . . . . . . 71
4.2.1 Folding of epithelial sheets . . . . . . . . . . . . . . . 72
4.2.1.1 Biological motivation . . . . . . . . . . . . . 72
4.2.1.2 Computational modelling . . . . . . . . . . 72
4.2.2 Morphogenesis of multi-cellular rosettes . . . . . . . . 74
4.2.2.1 Biological motivation . . . . . . . . . . . . . 74
4.2.2.2 Computational modelling . . . . . . . . . . 75

5 Computational modelling unveils how epiblast remodelling

and positioning rely on trophectoderm morphogenesis dur-
ing mouse implantation 79
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2.1 Repulsion at the apical surface of the epiblast is suffi-
cient for lumenogenesis . . . . . . . . . . . . . . . . . 86
5.2.2 Mechanical constraints imposed by TE morphogenesis
on the epiblast drive cup shape acquisition . . . . . . 88
5.2.3 Trophectoderm morphogenesis fosters epiblast move-
ment towards the uterine tissue . . . . . . . . . . . . 92
5.2.4 Sensitivity analysis . . . . . . . . . . . . . . . . . . . 95
5.3 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . 99

6 Quantification of cell behaviors and computational modelling

show that cell directional behaviors drive zebrafish pectoral
fin morphogenesis 105
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.2.1 Data acquisition workflow . . . . . . . . . . . . . . . 108
Contents vii

6.2.1.1 Zebrafish husbandry . . . . . . . . . . . . . 110

6.2.1.2 Imaging pectoral fin growth . . . . . . . . . 110
6.2.1.3 Image processing and reconstruction . . . . 110
6.2.2 Tracking zebrafish pectoral fin growth along PD, AP
and DV axes . . . . . . . . . . . . . . . . . . . . . . 111
6.2.3 Computational model . . . . . . . . . . . . . . . . . . 112
6.2.3.1 Model description . . . . . . . . . . . . . . 113
6.2.3.2 Computational implementation . . . . . . . 115
6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.3.1 Zebrafish pectoral fin morphogenesis is proximal distal
oriented . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.3.2 Distal tip-based growth does not account for zebrafish
pectoral fin morphogenesis . . . . . . . . . . . . . . . 116
6.3.3 Zebrafish pectoral fin cells exhibit preferential direc-
tional behaviors . . . . . . . . . . . . . . . . . . . . . 119
6.3.4 Directional cell behaviors are essential to drive ze-
brafish pectoral fin morphogenesis . . . . . . . . . . . 121
6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

7 Conclusion 131
7.1 Future perspectives . . . . . . . . . . . . . . . . . . . . . . . 133

Bibliography 135

List of Figures 156

Acknowledgements 1

Acknowledgements

I would like to express my sincere gratitude to my former Director of Studies

and supervisor Dr Rene Doursat. He saw potential in me right from the
application stage, and offered me the amazing opportunity to pursue these
doctoral studies. Indeed during this journey, his continuous support, enthu-
siasm, patience, motivation, and wisdom have guided my endeavours and
nurtured my embryonic research skills. I could not have imagined having a
better advisor and mentor for my PhD studies.

I would also like to express my sincere gratitude to my current Director

of Studies Dr. Matteo Cavaliere. Even before joining my supervisory team,
we had a good working relationship. After joining my supervision as Director
of Studies, he quickly stepped up to the role of a mentor. His practical
wisdom, administrative skills, and research experience significantly helped
me get through the demanding last months of my studies.

I would like also to thank the rest of my supervisory team in the persons
of Prof. Liangxiu Han, Dr. Moi Hoon Yap, and Dr. Sylvester Czanner.
They have often provided timely encouragements, fresh eyes, out-of-the-box
critical thinking, but also for challenging questions that have incited me to
widen my research from various perspectives.

This thesis project is one of fourteen within the EU Horizon 2020 Inno-
vative Training Network ImageInLife. I would like to thank Prof. Georges
Lutfalla of the University of Montpellier, our Project Leader, along with
all Principal Investigators for initiating the ImageInLife framework, thus
contributing in my thesis in multiple ways. My gratitude goes particularly
to Prof. Magdalena Zernicka-Goetz and Prof. Nadine Peyrieras, who re-
spectively provided me with an opportunity to visit their laboraties at the
Acknowledgements 2

Mammalian and Stem Cell Group of the University of Cambridge, and the
BioEmergences Lab of the French National Centre for Scientific Research.
Without their precious support and the insightful discussions I had with
them and their lab members, it would not have been possible to conduct
this research. To conclude this series, my gratitude go to Mrs. Veronika
Peciarova, our project manager, for always being there for us and organising
everything so well.

I would then like to thank all my fellow Early Stage Researchers within
ImageInLife. I have often quoted to you this thought by Eric Schmidt, former
CEO and Executive Chairman of Google: “The easiest way to have a good
idea is to have a lot of ideas”. A lot of ideas that have shaped this thesis
sprout from the many meetings and conversations we have had together. I
would particularly like to thank Hanh Nguyen, our collaboration has birthed
a manuscript that we co-authoured, and which constitutes a chapter of this
thesis; and Antonia Weberling, with whom I had stimulating discussions that
have led to interesting results.

I thank fellow PhD students at Manchester Metropolitan University.

Our multiple discussions on various topics, frequent lunches, and once-in-a-
while parties have kept this studies a fun adventure.

Last but not the least, I would like to thank my family: my parents,
brother and sisters; and my friends, for supporting me spiritually and emo-
tionally throughout this thesis and my life in general.
Statement of originality

I, Joel Dokmegang, hereby declare that this dissertation and the work de-
scribed in it are my own work, unaided except as may be specified below.
The work presented in chapter 6 is the result of a joint collaboration with
colleagues at the French CNRS BioEmergences laboratory, our fellow part-
ners within the EU Horizon 2020 Innovative Training Network Imageinlife.
In this collaboration, BioEmergences provided imaging datasets consisting
of tracked cell trajectories in a developing zebrafish pectoral fin (as shown
in figure 6.1a). Every other analysis in this chapter and in other chapters
were performed by me. The findings presented in chapter 6 were also jointly
published in the form of an article in the scientific review bioinformatics.

3
Chapter 1

Introduction

Embryogenesis, which designates the formation and development of an em-

bryo, is home to spectacular and species-consistent morphological events that
gradually transform an organism from a single egg to a fully functional liv-
ing being [81]. Morphological changes in embryonic tissues are driven by the
coordinated dynamics of myriads of cells, the basic unit of biological life.
At the subcellular stage, an impressive machinery deploys intense activity
dominated by biomechanics, gene expression and molecular signalling that
regulates cell states. On the one hand, cell shapes change and their move-
ments generate the physical forces that shape the embryo. On the other
hand, differences in genetic activity lead to the synthesis of various proteins
that explain differentiation into distinct cell types [152]. Development is
therefore the result of the interplay between two mutually influencing pro-
cesses: “mechanical morphogenesis” and “chemical morphogenesis” [145]. To
adopt an artistic metaphor, these tightly coupled processes can also be de-
scribed as shaping, by which the embryo “sculpts” itself, and atlassing, by
which it “paints” itself [35].

5
Introduction 6

Following this description, modelling biological development becomes

a question of how much each of these processes can be integrated into a
unified mathematical and computational formulation to account for what
is being observed experimentally. The many different modelling techniques
currently in existence can be construed as different compromises to this prob-
lem. While approaches based solely on either mechanism have been proposed
[61, 116, 146], models that couple both mechanical and chemical variables
have been on the rise in recent years [15, 35, 64, 85, 90, 115, 125], and their
promising results justify further investigation. Grounded on the latter, we
aim in this project to build computational models that can account for both
mechanical and chemical aspects of tissues dynamics, and are able to repro-
duce key morphological events in the early development of living organisms.

Computer simulations present numerous advantages over traditional ex-

perimental techniques, one of which is their extreme modulability, both in
input and output [18]. “In silico”, the experimentalist has more control over
environmental conditions, and can measure any metrics of interest. Although
this desirable feature eliminates risks, ethical concerns and ideally allows a
more thorough investigation of the studied system, one however runs the dan-
ger of falling out of the biological domain. As a matter of fact, bio-inspired
models can be pushed beyond classical biology to explore alternative forms
of life [40]. This is the programme of Artificial Life, a discipline extending
theoretical biology by exploring “life-as-we-know-it” within the broader con-
text of “life-as-it-could-be”, and deriving from there innovative technologies,
e.g. robotic or biomedical. In real-world biological modelling however, it is
crucial to keep the model outcomes in accordance with experimental observa-
tions. In this context, microscopy imaging plays a central role: observations
and simulations interact in a feedback loop, whereby imaging data informs
Introduction 7

and refines the model, which in turn raises new questions and suggests new
data acquisitions [18].

In this thesis, we aim to apply this workflow to the study of key episodes
of vertebrate development: mouse embryo implantation, and zebrafish pec-
toral fin development. For each of these studies, and taking into account
their peculiarities, we propose novel approaches for modelling the biome-
chanics involved. For the former, we develop a novel model of cell and tissue
mechanics based on established approaches in which cells can exhibit various
shapes. This enables us to schematically reproduce the observed biological
phenomena and quest for new insights. For the latter, we extend an existing
modelling paradigm of cell mechanics to accommodate the specific require-
ments of the study. In order to showcase the work carried out, we propose
the following outline for the rest of the thesis.

Chapter 2: Computational models of biological development

A great diversity of models exist to simulate biological cells and tissues.
Depending on the biological realism they include, the spatio-temporal scale
they capture, or the nature of variables they manipulate (physical, chemical),
models exhibit different properties and suit different purposes. Here we
review computational models of biological development in the literature.

Chapter 3: MG#: A modelling framework and simulation

platform for cell and tissue mechanics Building on the fundamen-
tal principles of Sub-cellular Element Models, we introduce a novel model
(MG#) of cells and tissues biomechanics. In this model, a cohort of particles
on the membrane and a single intracellular particle represent the biological
cell. Non-linear elastic potentials between particles mimic the plasticity of
cell membranes and the activity of the cytoskeleton within the cell. With
Introduction 8

this model, cells are able to exhibit bio-realistic cell shapes and cell popula-
tions can reproduce biological tissue behaviours such as invagination of an
epithelial layer or the formation of multicellular rosettes.

Chapter 4: Modelling of epithelial morphogenesis Not many

biological entities are as much regulators of biological development in or-
ganisms as are epithelial tissues and their characteristic behaviours. In this
chapter, we explore different mechanisms of epithelial morphogenesis. In
particular, we use MG# to model and simulate epithelial processes at the
scale of individual cells and tissues, including planar polarised constriction,
apical constriction and its variants, epithelial folding and rosette formation.
This chapter also aims at showcasing what is possible with MG# and lays
foundations for the study in the next chapter.

Chapter 5: Computational modelling unveils how epiblast re-

modelling and positioning rely on trophectoderm morphogenesis
during mouse implantation Implantation is a critical milestone in mouse
embryogenesis. Upon implantation, the blastocyst undergoes significant re-
modelling from an oval ball to an egg-cylinder. A key feature of this transi-
tion is the symmetry breaking in the epiblast and its shaping into a “cup”. We
hypothesise that this event is the result of mechanical constraints originat-
ing from the trophectoderm, also going through significant transformations
in this time. Using MG#, we investigate this hypothesis. Our results sug-
gest that trophectoderm morphogenesis indeed dictates the cup shaping of
the epiblast, and fosters its movement towards maternal tissue.

Chapter 6: Quantification of cell behaviors and computational

modelling show that cell directional behaviors drive zebrafish pec-
toral fin morphogenesis How the zebrafish grows its pectoral fin from a
2D layer to a 3D structure remains a challenge in embryology. By analysing
Models of biological development 9

single cell dynamics from live imaging of zebrafish embryos, we find out
that during this development cells gradually align their long axis to the
proximal-distal axis (PD) of the zebrafish fin. Building on this observation,
we enhance the basic centre-based cell model with a polarity term and sim-
ulate fin growth, allowing proliferation. Our simulations results in 3D fins
similar with shape to real ones, suggesting that directed cell behaviours, are
essential to drive fin morphogenesis in zebrafish.

Chapter 7: Conclusion Here we review the landscape of the work

carried out, identify its strengths and weaknesses, and present future per-
spectives.
Chapter 2

Computational models of
biological development

2.1 A brief history of mathematical and com-

putational models of development

Only a handful, if any, of natural mechanisms could be considered as fascinat-

ing and inspiring as morphogenesis. The formidable challenge presented by
biological development, as described by Enrico Coen, is to “understand how
the complex pattern and arrangement of different cell types that make up a
mature organism can arise from a single cell in a consistent way each gener-
ation” [30]. In their efforts solving this self-organising puzzle, embryologists
have formulated theories spanning a vast range of the natural sciences. Early
attempts described it in terms of intuitive principles such as heat, wetness,
and “solidification”. After several centuries of enlightenment and progress,
D’Arcy Thompson formally postulated that anatomical complexity emerges
from the principles of physics and chemistry [143]. A major breakthrough
11
Models of biological development 12

occurred when in 1938, Nicolas Rashevsky pioneered the field of mathe-

matical biophysics, emphasising quantitative, metric aspects of the physical
manifestations of life. In particular, he proposed a framework to describe
organisms as networks in which vertices represented biological functions and
oriented edges the interactions between them [123], and showed that cell
polarity was possible even for spherical cells [122]. Another landmark of
mathematical modelling in biology was reached in 1952 when Alan Turing
showed that pattern formation in natural systems could be simulated using
reaction-diffusion equations [145]. Expanding on Turing’s hypothesis, Lewis
Wolpert introduced the concept of positional information, which postulates a
hidden geographical organisation of the embryo into regions of cells following
different fates [152]. Over the years, more theoretical hypotheses, for exam-
ple Holtfreter’s “selective affinity” further refined by Steinberg’s differential
adhesion hypothesis (DAH), appeared and inspired further translation into
mathematical biophysics [79, 138].

Parallel to the birth of mathematical formulations of developmental the-

ories, computers also appeared and their use for all sorts of scientific purposes
grew rapidly. Whilst handwritten equations could help solve the dynamics
of a small number of cells, or evaluate a few state variables characteristic of
a tissue, this new huge computational power allowed for the simulation of
more complex phenomena. One of the earliest computerised simulations of
vertebrate development can be attributed to Jacobson and Gordon, who de-
signed and simulated a mathematical model of neural plate formation [77].
Their virtual experiment, backed by mathematical analysis, validated the
hypothesis postulated upon experimental observation which stipulated that
shrinkage of the surface of the neural plate and displacement of the entire
sheet are necessary and sufficient to produce the transformation of the neural
Models of biological development 13

plate from a hemispheric sheet of cells to a keyhole shape. Since then, com-
putational modelling has gained considerable ground and positioned itself as
an instrument of choice in the biologist’s toolbox.

2.2 Models of biological development

Anatomical complexity in embryonic tissues emerges from the multiple in-

teractions in which cells are involved. The dynamics of a cell originate from
within itself, from the influence of surrounding cells, and from the extra-
cellular matrix (ECM). Cells have the intrinsic ability to initiate complex
movements and deformations, which they achieve through constant redeploy-
ments of their internal structures [152], first and foremost their cytoskele-
ton. As cell populations usually reside in densely packed arrangements,
any changes in shape and position has a direct influence on neighbouring
cells, which themselves undergo (passively) or trigger (actively) morpholog-
ical changes in response. Several studies have established that genes are
ultimately responsible for orchestrating the internal machinery of the cell by
defining the properties that drive its activity [84]. Through multiple molec-
ular pathways featuring complex networks of inhibitor/activator agents, ge-
netic regulation establishes an anisotropic functional atlas of cells and tis-
sues, e.g. inducing asymmetries in the distribution of force-related molecules
within a single cell, thus driving deformations and migrations. In the same
way that genetic activity produces forces, mechanical signals received by the
cell can also be translated back into chemical signal via a mechanism called
mechanotransduction.
Models of biological development 14

The above description highlights the double dynamics of cellular activ-

ity, biomechanics and genetic regulation, and their tight links. Computa-
tional modelling of development is deeply grounded in this framework. On
the one hand, models of cell mechanics investigate how physical forces shape
the embryo. On the other hand, models of gene expression and molecular
signalling examine how cells determine their states and behaviours. In the
next paragraphs, we review these different paradigms, including some of their
fundamental principles and results achieved.

2.2.1 Models of cell mechanics

Generally, there are two approaches when it comes to modelling cell me-
chanics: global or “continuum” approaches, and agent-based approaches.
Continuum models are well suited for capturing large-scale dynamics in de-
veloping organisms. These models have proven useful in studying develop-
ment mechanisms of various interest. Recent studies include investigating
the role of topology and mechanics in uniaxial growing cell networks [59],
unravelling how lumenal pressure and tissue mechanics control the embryo
size [22], showing that human organoids development rely upon the contrac-
tion of the inner core of the organoid and the microstructural remodeling of
its outer cortex[6], and combining muscular differentiation and differential
growth to reproduce morphological patterns observed in the vilification of
several species [137]. In continuum systems, however, single cells are over-
looked, and focus is put on larger regions of tissues. Hence, the emergence
of phenomena initiated by single cells acting in a cohort (the very essence of
morphogenesis) can therefore not be fully appreciated. Agent-based models
of tissue mechanics, on the other hand, allow the representation of single
Models of biological development 15

cells and their heterogeneities. They can be classified into three categories:
lattice-based models, centre-based models and deformable cell models.

Lattice-based models

In lattice-based models, cells reside on a fixed lattice and their dynamics

consists of swapping lattice sites under certain conditions with the goal of
minimising some defined global energy specific to the simulated phenomena.
Simple rules prescribed at lattice level guarantee that large populations of
cells can be simulated at a computationally low cost. Surprisingly, lattice-
based models have been able to successfully simulate complex tissue-level
behaviours including cell sorting, proliferation, cell death, differentiation
and polarisation [146]. Further distinctions can be made within this
category: models featuring one cell per lattice site in Cellular Automata
(CA); multiple cells per lattice site; or one cell occupying several lattice sites
in the Cellular Potts Models (CPM). CPM arguably constitutes the most
widely used class of lattice-based models today. Inspired by the success of
the large-q Ising model in reproducing the topological changes of metals
and soap foams, Graner et al. [60] adapted this approach to biological cells
and simulated cell sorting driven by differential adhesion, giving birth to
CPM. Niculescu et al. [104] showed that CPM could realistically reproduce
shape-driven migration of biological cells, in particular the crawling motion
and deformation of amoeboid cells, and gliding of half-moon shaped
keratocyte-like cells. Lattice-based models can also be extended to 3D, as
exemplified by Belmonte et al. [10] who used CPM in 3D to simulate the
in-plane elongation (in one direction) and simultaneous shortening (in the
perpendicular direction) of a planar-polarised epithelial tissue, an event also
known as convergent extension. Nevertheless, this class of models exhibit
Models of biological development 16

a few drawbacks. Due to their stochastic nature, time scales are not well
defined. Moreover, the simple rules that govern cell movements in these
models generally do not represent biomechanics in a physically meaningful
way and are hard to interpret.

Figure 2.1: A population of cells, initially randomly scattered, gradually

sorts itself into two regions due to differences in adhesion strength between
cell types.

Centre-based models

Another class of models, centre-based models (CBMs), represent cells or sim-

ply their centres as single particles embedded in a highly viscous 3D envi-
ronment. These models assume that, in analogy to physical particles, cell
trajectories in space can be described by an equation of motion [118, 146].
Similar to lattice-based models, CBMs enable physical properties such as
volume, surface area, internal pressure and mechanical stress to be defined
for simulated cells. Moreover, the gained physical realism allows the explicit
introduction of forces, well-defined time scales, and intuitive ways of mod-
elling cell-cell interactions. With these models, however, the evaluation of
mesoscale properties such as cell shapes remains a problem. Cells generally
have a constant geometric shape, in most cases spheres. Although cell bod-
ies usually display perfect symmetry, it is possible to simulate asymmetric
cell behaviours. In order to achieve this, Delile et al. [35] define ‘active’
Models of biological development 17

behavioural forces underlying at once polarisation, mesenchymal cell pro-

trusion, and epithelial cell-junction remodelling. In CBMs, cells generally
interact with one another through a spring-damper system applied in their
centre. This constitutes of course a crude simplification of biology, where
in reality adhesion molecules act at the surface of the cell membrane, but is
nonetheless sufficient in most simulations. In an attempt to improve precise
cell-cell adhesion and collisions, Disset et al. [39] proposed a force applied at
the centre of the contact surface of cell bodies and proportional to the area of
this surface. Further enhancements employ established methods of contact
mechanics, namely Hertzian or Johnson-Kendall-Roberts theories [118, 146].
Interesting experimentally observed phenomena have been simulated using
CBMs, ranging from proliferation within monolayers [41, 88] to complex
cellular rearrangements induced by intercalation observed in the enveloping
layer of the zebrafish during gastrulation [34, 35].

The advantages of force-based models are a well-defined time scale, and

a more intuitive way of taking into account complex interactions of cells
with other cells or their environment which is why they became the standard
approach.

Deformable cell models

At the higher end of biological realism integrated within a single cell are
deformable cell models (DCMs). In DCMs the cell body is discretised into a
number of vertices or ‘nodes’ connected by viscoelastic edges and interacting
via opposite conservative forces, creating a flexible scaffolding structure with
multiple degrees of freedom per cell [146].
Models of biological development 18

In some DCMs, it is common to make a distinction between internal

nodes, which mimic the activity of intracellular organelles, and external
nodes, corresponding to the cell membrane. Newmann et al. put forward
this modelling approach by introducing the subcellular element model (SEM)
where subcellular elements (SCEs) are subjected to forces derived from a
Morse potential [102]. Multiple extensions to the SEM have been proposed.
Of notable interest is the work of Milde et al. [98] which introduces several
new features: a modification of the potential function to increase adhesion
between SCEs; cell polarity; the smooth insertion and removal of SCEs to
simulate polymerisation and depolymerisation of the actin complex (leading
to shape change, migration and growth); novel methods for detection of the
cell membrane elements; and a special SCE representing the cell nucleus.

Although pairwise interactions between particles can accommodate cell-

cell adhesion and collisions, other ways of modelling these behaviours have
also been used. In addition to potentials, and to avoid overlapping between
cell membranes, some authors employ specular reflections of cell particles on
the membranes of other cells [50]. More accuracy may be attained through
the use of established theories of contact mechanics. This is the case of
Odenthal et al. [107] who rely on the Maugis-Dugdale theory of contacts
between elastic bodies to investigate cell-cell and cell-substrate adhesion.
However, SEM descriptions of the cell are generally too broad to take into
account the specificities of epithelial cells and tissues in terms of shape and
polarisation. Marin-Riera et al. [90] tackled this challenge by proposing a
unified model of development enabling the simulation of both mesenchymal
and epithelial cells. In their model, epithelial subelements, in contrast to
mesenchymal subelements represented by spheres, are assumed to be cylin-
drical entities, each consisting of an apical and a basal part that can move
Models of biological development 19

independently. More recently, Revell et al. [127] developed a SEM imple-

mentation able to predict the mechanical drivers of cell sorting in multicel-
lular aggregates. Combining this model with experimental observations, the
group was able to show that dynamic cell surface fluctuations, in addition to
static mechanical properties, played a crucial role in the spatial segregation
of the founding lineages in mammalian embryo [153]. Other SEM applica-
tions span a wide range of biological processes including the dynamics of red
blood cells [50, 107], cell sorting driven by differential adhesion [150], and
sea urchin gastrulation [90].

A particular class of DCMs that capture epithelial dynamics remark-

ably well consists of vertex models (VMs). VMs were created in response
to the poor performance of CBMs regarding the flexibility of cell shapes in
dense packings [101]. In VMs, a tissue is represented by a tiling made of
non-overlapping connected polygons (each polygon corresponding to a cell),
whose vertices are free to move. Vertex motion is driven by the minimisa-
tion of a potential energy of the tissue. Depending on the phenomenon to
simulate and the dimensionality of the system, this potential energy may
contain a certain number of terms such as: total edge length, area or volume
conservation, cavity volume in blastocyst formation, or directed elongation
in convergent extension [70]. VMs probably constitute the most popular
class of agent-based models currently in use. Extending the work of Honda
and coworkers, Okuda et al. thoroughly investigated morphogenesis using
VMs [110–114].

In general, the accuracy of the mechanics in DCMs is closely correlated

with the number of nodes in each cell. Depending on the application, a low
number of nodes can create unrealistic dynamics, whereas a high number of
nodes considerably increases the computational complexity of the model.
Models of biological development 20

2.2.2 Coupling mechanical and chemical variables

Many authors have identified the need for models accounting for both me-
chanical and genetic/molecular aspects of cellular life, as these are tightly
correlated in a feedback loop. In the context of VMs, Yu et al. [154] ex-
plain that forces might be proportional to the concentration of molecules
whose activity create them, as is the case with actin-myosin driven migra-
tion forces and E-cadherin driven adhesion forces. Additionally, they advo-
cate the necessity of using reaction-diffusion systems to model the dynamics
and transport of substances through the tissue. Okuda et al. [114, 115]
applied Turing’s activator/inhibitor system to 3D vertex models, correlat-
ing the concentration of activator with the growth rate of individual cells.
They showed that such coupling could reproduce spontaneous multicellular
morphogenetic patterns including undulation, tubulation, branching, arrest,
expansion, invagination and evagination.

Investigating how sharp boundaries between regions of different gene

expression appear in tissues, Wang et al. [150] simulate the dynamics of
an epithelium with a unified model coupling cell biomechanics through the
SEM approach (model S) and gene regulation via reaction-diffusion equa-
tions (model P). The tissue is composed of two cell archetypes characterised
by the preponderance of two genes (A and B), and a morphogen (M) dif-
fusing through the tissue. Gene expression levels influence the mechanical
properties of cells, and cells can ultimately differentiate into the other type.
Using a new “sharpness index” metric, they show that coupling models S
and P produces a better characterisation of boundaries than models S and
P applied separately. Another notable work combining SEMs to GRNs was
carried out by Marin-Riera et al. [90]. In their model, both mechanical and
chemical variables act directly at the subcellular element level. The model
Models of biological development 21

further distinguishes between enzymes, adhesion molecules and transcription

factors, the latter being restricted to the nuclear SCE of each cell.

Delile et al. [34, 35] implemented mechanogenetic behaviours by in-

tegrating mechanics, gene regulation and molecular signalling in a CBM.
Their model called MecaGen defines a cell behaviour ontology that relates
cell behaviours to specific levels of concentration of proteins within cells. In
this context, attraction forces are controlled by surface densities of adhesion
molecules; differentiation and specialised behaviours are influenced by gene
regulation; polarisation is determined by ligands and neighbouring cells; and
mechanotransduction affects gene regulation.

2.3 The validation feedback loop between ob-

servations and models

As mentioned in the introduction, the crucial difference between biologi-

cal modelling and bioinspired or artificial life approaches is the ultimate
comparison of the model’s predictions with actual biological observations.
During most of the history of biological modelling, models of development
have been validated or falsified based on a mere qualitative evaluation of
their predictions. For a model to be judged satisfactory, it had to visually
reproduce well-known morphological phenomena such as the neural plate
formation [77], cavity formation or convergent extension [70], DAH driven
cell sorting [58], or the sea urchin gastrulation [90]. While qualitative as-
sessment remains an important criterion, a major paradigm shift is now un-
derway. With the recent explosion of microscopy imaging data, models are
increasingly rated on their ability to infer results that quantitatively match
Models of biological development 22

experimental data. Generally, achieving acceptable levels of accuracy is a

matter of optimisation of model parameters with the goal of minimising a
given “fitness” energy. Different authors deal with this challenge in different
ways.

Farhadifar et al. [47] use metrics based on the topology of the tissue
to validate their simulations. Based on a vertex model, they investigate
the influence of cell mechanics, cell-cell interactions and proliferation on the
epithelial packing of the Drosophila wing disc. Simulating the dynamics
of a proliferating epithelial layer undergoing eight cycles of division, they
analyse the proportions of polygonal cell shapes and their average surface
areas in the resulting tissue, and compare them with experimental data.
This data consists of a network of apical junctions of 1,738 cells segmented
from microscopy images of wing discs using an automated image-processing
algorithm. By searching the parameter space, they discover regions where
calculated network morphologies match the experimentally observed ones.

Delile et al. [35] employ topological distances measured on the tissue to

validate their model and hypothesis. In examining the impact of protrusive
forces and cell polarisation on the dynamics of zebrafish enveloping layer
during epiboly, a parameter space search is conducted. At the end of each
iteration, they make use of a fitness function to compare key properties of the
simulated embryo (embryo height, yolk height, margin height, and margin
diameter) to those measured on a real embryo. Parameters for which calcu-
lated data fit experimental data are found, thus corroborate their hypothesis
that oriented protrusion of deep cells in the enveloping layer drives zebrafish
epiboly.

Kursawe et al. [83] propose a more informed approach for parameter

Models of biological development 23

space search. With the goal of determining whether vertex model parame-
ters can be calculated from imaging data, they use summary statistics from
cell packing and laser ablation experiments. With their method, the au-
thors show that data gathered across multiple experiments generate valid
parameter estimates. This data driven approach has the benefit that it fur-
ther allows an estimate of the uncertainties related to these parameters, thus
providing a means of evaluating confidence intervals associated with model
predictions.

In summary, and despite these advances, the validation of developmen-

tal models is still a grey area of research: understanding how to best in-
tegrate and interpret experimental data with cell-based models remains a
major challenge for the modelling community. In order to make the most
of the available data, Fletcher et al. [53] propose an end-to-end workflow
including data acquisition, analysis and fusion; model development, reduc-
tion and parametrisation; model validation/selection and guidance for future
experiments.

In the area of model development, although many computational stud-

ies still rely on tailored implementations of models, there have been ad-
vances in the uniformisation in the field with the emergence in recent years
of multipurpose modelling and simulation software tools. These software usu-
ally offer implementations of either single or multiple model families within
open-source frameworks which modellers are able to use and extend for
the benefit of their studies. The most notable packages include Chaste[99],
CompuCell3D[142], VirtualLeaf[96], LAMMPS[36], Yalla [57], MecaGen[35].
With the exception of a few, most notably CompuCell3D [142], these plat-
forms are often aimed at audiences with strong mathematical and program-
ming skills, hence limiting the participation of stakeholders across the board
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 24

in model development. Subsequent advances will require standardisation

and simplification of communication practices between scientists of various
domains contributing to the field, including biologists, physicists, mathe-
maticians and computer scientists.
Chapter 3

MG#: A modelling framework

and simulation platform for cell
and tissue mechanics

In this chapter, we present our model of cell and tissue mechanics, the open
source simulation framework developed, and preliminary experiments of sim-
ulated morphological events. The necessity of a model featuring deformable
cells arose from the need to simulate drastic cell shape changes involved
morphological events such as mouse embryonic implantation (Chapter 5).
As discussed in the previous chapter, two family of models are suitable for
these specific modelling requirements: Subcellular Elements Models (SEM)
and Vertex Models (VM). These classes of models however presented obvious
limitations for our targeted use cases. On the one hand, SEM approaches
often use a large number of particles, to discretise the cell, increasing the
computational time and power requisite for simulations. On the other hand,
Vertex Models, by using fewer particles representing in most cases only apical
and basal faces of cell, are unable to account for a vast breadth of biological
25
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 26

cell shapes. We therefore aimed at developing a model which metaphorically

bridges the gap between the two frameworks.

MG#: A Computational Model

Based on the fundamental principles of DCM, our abstraction of the biolog-

ical cell features particles in interaction under the influence of conservative
forces. Emphasis is put on particles at the surface of the cell membrane,
bringing our model close to VM [71], while at the same time we also include
a single intracellular particle reminiscent of the cell’s microtubule organising
centre (Fig. 3.1A,B).

On the cell membrane, we define a topological neighbourhood based

on a triangulation of vertices. Two same cell particles are deemed internal
neighbours if they both belong to one of the mesh triangles (Fig. 3.1C,F). We
also define an external neighbourhood based on distances between particles
of different cells (Fig. 3.1C,F). To minimise the computation time required,
we introduce cell-cell neighbourhood relationships where particles of different
cells are tested for external neighbour links only when the cells to which they
belong were already approved as neighbours. The implementation of cell-cell
neighbourhoods depends on the geometry of the problem at hand. In dense
tissues with little variation of cell positions, depending on the geometry of
cells, we opt for either a Moore neighbourhood, where a cell is surrounded by
eight neighbours in its plane, or a hexagonal neighbourhood, a configuration
in which most cells have six neighbours. In other cases, a metric-based cell-
cell neighbourhood may be favoured.
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 27

Figure 3.1: Computational model (MG#). A,B. 3D representa-

tion of cells: The cell is abstracted by an agglomeration of particles (small
white spheres), whose triangulation (white edges) forms the membrane,
and by an intracellular particle (big white sphere). Interactions between
the intracellular and membrane particles (blue lines) mimic the cytoskele-
ton. Membrane particle 𝑖 is under forces from neighbouring particles (𝑗,
𝑘, ...), and from the intracellular particle 𝜒. A. Cell with 42 vertices. B.
Columnar cell with 34 vertices. C. A view of particle neighbourhoods in
a square arrangement of four cells (nuclei not displayed). white: internal
neighbourhood links. Yellow: external neighbourhood links. In this case,
diagonal cells are not touching. D,E. Forces acting on a cell. Membrane
particle 𝑖 is under forces from neighbouring particles (𝑗, 𝑘, ...), and from
the intracellular particle 𝜒. F. External forces acting on a cell via its
→
− →
− ext →
− int →
− int →
−𝜒
particles. Here, 𝐹 ext
𝑖2 = 𝐹 𝑗2 𝑖2 = ( 𝐹 𝑗1 𝑗2 + 𝐹 𝑗3 𝑗2 ) + 𝐹 𝑗2 . G. Plots of the
magnitude of Morse forces under different values of 𝐽, with 𝜌 = 1 and
𝑟eq = 0.5.
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 28

In order to induce intrinsic mechanical behaviours within cells, we as-

similate internal particle neighbourhood links to non-linear springs, which
have been shown to faithfully emulate living matter [108]. These springs
mimic the activity of actomyosin and microtubule networks in the cytoskele-
ton, and forces are derived from their elastic potential (Fig. 3.1D,E). In the
cell’s resting state, the equilibrium distance of each spring coincides with
the length of the segment formed by its nodes. Cell dynamics arise from
alterations to these equilibrium distances.

Equation of motion

Acting on a given membrane particle 𝑖, we distinguish four main types of

→
− →
−𝜒 →
− ext
forces: internal forces 𝐹 int
𝑗𝑖 , cytoskeleton forces 𝐹 𝑖 , external forces 𝐹 𝑗𝑖 ,
→
−
and specific forces 𝐹 spe
𝑖 . Biological media are often characterised by a low

Reynolds number, due to their high viscosity, which minimises the effects of
inertia [146]. We therefore subject particles to an over-damped, first-order
equation of motion:
⎛ ⎞ ⎛ ⎞
∑︁ →
− int ⎠ →
− →
− ext ⎠ → −
= 𝜆med →
−
∑︁
⎝ 𝐹 𝑗𝑖 + 𝐹 𝜒𝑖 + ⎝ 𝐹 𝑗𝑖 + 𝐹 spe
𝑖 𝑣𝑖 (3.1)
𝑗∈𝒩int (𝑖) 𝑗∈𝒩ext (𝑖)

Here, 𝒩int (𝑖) and 𝒩ext (𝑖) represent the sets of internal and external neigh-
bours of particle 𝑖, and 𝜆med is the coefficient of friction exerted on all par-
ticles.

In line with Newton’s third law of motion, membrane particles enter-

tain reciprocal forces equal in magnitude and opposite in direction with the
intracellular particle. Therefore, the dynamics of the intracellular particle
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 29

are dictated by:

∑︁ →
−
− 𝐹 𝜒𝑖 = 𝜆𝜒 →
−
𝑣𝜒 (3.2)
𝑖

Here, 𝜆𝜒 is the coefficient of friction exerted on the intracellular particle.

Internal and cytoskeleton forces

The internal force created by a particle 𝑗 on a neighbouring particle 𝑖 derives

from a Morse potential (Fig. 3.1G). Previous studies have used Morse po-
tentials to represent forces in a biological context [98, 102]. The expression
of this force is given by:

→
− int
𝐹 𝑗𝑖 = 2𝐽𝜔 𝜌 (𝑒2𝜌(𝑟−𝑟eq ) − 𝑒𝜌(𝑟−𝑟eq ) ) →
−
𝑢 𝑖𝑗 (3.3)

Here, 𝐽𝜔 represents the interaction strength between particles 𝑖 and 𝑗, both of

cell type 𝜔, 𝑟 is the equilibrium of the spring force between 𝑖 and 𝑗, and →
eq
−
𝑢 𝑖𝑗

is the unit vector along the direction formed by 𝑖 and 𝑗. Similar forces dictate
interactions between the intracellular particle and the membrane particles.

External forces

As discussed in the previous chapter, there are several approaches to im-

plementing cell-cell interactions in the wider family of sub-cellular elements
models [108]. In the scope of this work, inspired by vertex models [47], we
developed a simple method to account for cell-cell interactions in epithelial
tissues.

Given the tight packing in epithelial tissues, a cell membrane is always in

contact with neighbouring cell membranes. Thus local action on a membrane
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 30

produces an equivalent deformation on the surrounding cells. In other words,

a particle always transmits the force received to its external neighbours. To
account for this behaviour, we submit particles and their external neighbours
to equal forces. This is done by setting the external force acting on a particle
to be equal to the sum over all its external neighbours of their internal and
nucleus forces:
→
− ext ∑︁ →
− ext
𝐹𝑖 = 𝐹 𝑗𝑖 (3.4)
𝑗∈𝒩ext (𝑖)
⎛ ⎞
→
− ext − int ⎠ →
∑︁ → −
𝐹 𝑗𝑖 =⎝ 𝐹 𝑘𝑗 + 𝐹 𝜒𝑗 (3.5)
𝑘∈𝒩int (𝑗)

Specific forces

Generally speaking, it is possible to include specific forces in DCM to account

for desired behaviours. A few studies have taken advantage of this possibility
to enable for example cell surface bending resistance [108] or cell surface area
and volume conservation [49]. Specific forces can be defined to complement
the other three main forces when a particular behaviour is expected. Their
implementation depends on the modelled event.

3.1 Computational implementation and Pre-

liminary results

In summary, cells are represented by an intracellular particle and membrane

particles. The cell shape emerges from a triangulation of membrane parti-
cle positions. Nonlinear springs between particles obey a Morse potential
between nodes (membrane particles and the intracellular particle) and drive
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 31

shape changes. These springs mimic the activity of actomyosin networks in

the cytoskeleton. In the cell’s resting state, the equilibrium distance of each
spring coincides with the length of the segment formed by its nodes. Cell
dynamics arises from alterations to these equilibrium distances. Based on
this principle, a number of cellular and tissue behaviours were simulated and
are described here.

The computational implementation of the described model requires a

→
− →
−
time discretisation of the equation of motion. Given 𝑋 𝑡𝑖 , 𝜆 and 𝐹 𝑡𝑖 =
(︁→
− int )︁𝑡 (︁→
− )︁𝑡 (︁→
− )︁𝑡 (︁→ − spe )︁𝑡
𝐹𝑖 + 𝐹 𝜒𝑖 + 𝐹 ext𝑖 + 𝐹𝑖 , respectively the position, friction
coefficient and total force applied on a particle 𝑖 at time point 𝑡, an explicit
→
−
Euler scheme is used to calculate the position 𝑋 𝑡+1
𝑖 of the particle at the next
time point. If 𝑑𝑡 is the time lapse between instants 𝑡 and (𝑡 + 1), particles
update their configuration based on the following rule (Eq. 6.2).

→
− 𝑡+1 →
− →
−
𝑋 𝑖 = 𝑋 𝑡𝑖 + 𝑉 𝑡𝑖 × 𝑑𝑡 (3.6)

→
−𝑡 →
−
𝑉 𝑖 is neither given, nor determined in (Eq. 6.2). An expression of 𝑉 𝑡𝑖
can however be drawn from the equation of motion (Eq. 6.3).

→
− →
−
𝜆 𝑉 𝑡𝑖 = 𝐹 𝑡𝑖 (3.7)

Hence we update particle positions in simulations according to the for-

mula in equation (Eq. 6.4).

→
−
→ − 𝑡 𝐹 𝑡𝑖
− 𝑡+1 →
𝑋𝑖 = 𝑋𝑖 + × 𝑑𝑡 (3.8)
𝜆

In the next section, we provide preliminary results of simple biology

MG#: A modelling framework and simulation platform for cell and tissue
mechanics 32

phenomena that can be simulated with our model. We also use one of the
examples (red blood cells) to show that the adopted Euler scheme is stable
and converges when the time laps 𝑑𝑡 is refined.

3.1.1 Cell shapes

As the embryo develops, cells gradually lose the perfect symmetry of the ini-
tial zygote. Active phenomena such as protrusion, Epithelial-Mesenchymal
Transition (EMT), or Mesenchymal-Epithelial transition (MET) [128] among
others, and passive behaviours such as mechanical response to physical stress
in multicellular settings foster single shape changes. For example, during gas-
trulation, several mechanisms of large-scale morphological change are due to
cell intercalation driven by the elongation of individual cells along a partic-
ular axis [35, 102].

Accounting for various cell shapes constitutes an important requirement for

our model. Globally, specific cell shapes can be obtained in two ways: either
by setting particles’ positions to preferential coordinates and triangulating
the cell membrane accordingly, or by transitioning from an initial cell shape
to a target shape. For the latter, we distinguish between two cases. Regular
shapes like spherical cells, cylindrical cells or red blood cells can be modelled
with analytical equations that set new equilibrium lengths for cell springs. In
the next sections, we show how new equilibrium lengths for desired regular
shapes can be computed. For less target regular shapes, stochastic methods
such the Ising model [76] in combination with Monte Carlo methods such as
the Metropolis-Hastings algorithm [11], and body similarity metrics like the
Hausdorff distance [74] can be used .
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 33

3.1.1.1 Red blood cell shapes

Here, we model the relaxation of a red blood cell from a spherical shape to
the biconcave shape characteristic of red blood cells. Evans et al. [46] put
forward an equation to describe this shape (eq. 3.9). This equation expresses
the new 𝑦 coordinate changes as a function of coordinates 𝑥, 𝑦 and 𝑧, with
parameters 𝑎0 , 𝑎1 , 𝑎2 .
⎧ √︁ (︁ )︁
⎨𝐷 1 − 4(𝑥2 +𝑧 2)
𝑥2 +𝑧 2 (𝑥2 +𝑧 2 )2
𝑎0 + 𝑎1 𝐷 2 + 𝑎2 𝐷 4 𝑖𝑓 𝑦 ≥ 0
⎪
𝐷2
𝑦 ′ (𝑥, 𝑦, 𝑧) = √︁ (︁ )︁
⎩−𝐷 1 − 4(𝑥2 +𝑧 2)
𝑥2 +𝑧 2 (𝑥2 +𝑧 2 )2
𝑎 + 𝑎 + 𝑎 𝑖𝑓 𝑦 < 0
⎪
𝐷2 0 1 𝐷2 2 𝐷4

(3.9)

Hence, we deduce the following equations for new equilibrium lengths

⎧ √︁
𝜒𝑖
⎨𝑟eq = 𝑥2𝑖 + (𝑦 ′ (𝑥𝑖 , 𝑦𝑖 , 𝑧𝑖 ))2 + 𝑧𝑖2
⎪
√︁ (3.10)
⎩𝑟𝑖𝑗 = (𝑥𝑖 − 𝑥𝑗 )2 + (𝑦 ′ (𝑥𝑖 , 𝑦𝑖 , 𝑧𝑖 ) − 𝑦 ′ (𝑥𝑗 , 𝑦𝑗 , 𝑧𝑗 ))2 + (𝑧𝑖 − 𝑧𝑗 )2
⎪
eq

Figure 3.2B demonstrates this shape change.

With this example, we have an opportunity to highlight how MG# dif-

fers from other vertex-based approaches. Generally, vertex models do not
feature intracellular particles. Hence, the forces that shape the cell emanate
exclusively from membrane particles interactions. In order to emulate this
behaviour in this context, we ran new simulations of cell shape transfor-
mation from spherical to RBC shape, this time suppressing the action of
cytoskeleton forces and allowing the cell to remodel under the sole influence
of membrane forces. We observed that particles did not rearrange in the
previous configuration that outlined the biconcave shape. Rather, though
the cell relaxes to a steady state, it maintains a convex shape (Fig. 3.2C).
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 34

We also take advantage of this example to examine the numerical sta-

bility and convergence of our model. To test these properties, we conducted
further simulations in which we varied the time step (𝑑𝑡) over multiple level
of magnitude. Additionally, we introduce the notion of the elastic energy of
a cell (𝐸) to measure how well the cell has reached the equilibrium state of
the target shape. This energy is defined as the sum over all cell springs of
the squared difference between equilibrium and actual lengths (Fig. 3.11).
In their resting states, cells springs have the same elongation as their equi-
librium lengths. Hence, cells exhibit the property that their elastic energy
equals zero at rest.

∑︁
𝑘𝑠
𝐸= (𝑟eq − 𝑟𝑘𝑠 )2 (3.11)
𝑠≤𝑁𝑘

Here 𝑁𝑘 is the number of springs in the cell (𝑘).

For each simulation, we plotted the evolution of the elastic energy per
time. Figure 3.3 shows the outcomes of these simulations. Whereas higher
time steps (𝑑𝑡 = 0.35, 𝑑𝑡 = 0.3, Fig. 3.3B, blue, yellow ) yielded unstable
simulations where the elastic energy indefinitely oscillated around non zero
values, lower time steps (𝑑𝑡 = 0.25, 𝑑𝑡 = 0.1, 𝑑𝑡 = 0.01, 𝑑𝑡 = 0.005, Fig. 3.3B,
green, red, purple, brown) not only produced simulations where the elastic
energy converged towards zero, but the curve also exhibited stability. In the
latter category, smaller time steps (𝑑𝑡 = 0.01, 𝑑𝑡 = 0.005) however needed
more simulation time to fully converge.

3.1.1.2 Cell elongation

Here, we model the elongation of a cell from a spherical shape to that of

a cylinder. Because of the spherical symmetry of the initial cell, we can
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 35

use the parametric representation of a sphere (eq. 3.12), with coordinates

(𝑥, 𝑦, 𝑧) relative to the cell centre.
⎧
⎪
⎪
⎪
⎪ 𝑥 = 𝑅 sin 𝜃 cos 𝜙
⎪
⎨
𝑦 = 𝑅 cos 𝜃 (3.12)
⎪
⎪
⎪
⎪
⎩𝑧 = 𝑅 sin 𝜃 sin 𝜙
⎪

If we assume that the new ellipsoidal cell will be characterised by radius

and height (ℎ, 𝑟), the shape change from spherical to cylindrical will require
that new positions of particles in the elongated cell satisfy equation 3.13.
⎧
𝑥′ 𝑎
⎪
⎪
⎪
⎪ = 𝑅
𝑥
⎪
⎨
𝑦′ = 𝑏
𝑦 (3.13)
⎪ 𝑅
⎪
⎪
⎩𝑧 ′ 𝑐
⎪
= 𝑧
⎪
𝑅

Combining (eq. 3.12) and (eq. 3.13), we then have the following rela-
tionships between initial and target coordinates.
⎧
𝑥′ 𝑎
⎪
⎪
⎪
⎪ = 𝑅
𝑥
⎪
⎨
𝑦′ = 𝑏
𝑦 (3.14)
⎪ 𝑅
⎪
⎪
⎩𝑧 ′ 𝑐
⎪
= 𝑧
⎪
𝑅

We can therefore compute required equilibrium lengths for springs 𝑖𝑗

and 𝜒𝑖 by computing particle distances from their new positions.
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 36

⎧ √︁(︀ )︀
𝜒𝑖 𝑎𝑥 2
(︀ 𝑏𝑦𝑖 )︀2 (︀ 𝑐𝑧 )︀2
⎨𝑟eq
⎪ = 𝑖
+ + 𝑅𝑖
√︂ 𝑅 𝑅
(3.15)
(︀ 𝑎𝑥𝑖 −𝑎𝑥𝑗 )︀2 (︁ 𝑏𝑦𝑖 −𝑏𝑦𝑗 )︁2 (︀ 𝑐𝑧𝑖 −𝑐𝑧𝑗 )︀2
⎩𝑟𝑖𝑗
⎪
= + +
eq 𝑅 𝑅 𝑅

Figure 3.2A demonstrates this shape change.

3.1.1.3 Epithelial cell shapes

Here, we model the transformation of a cell from a spherical shape to that

of a cylinder. Because of the spherical symmetry of the initial cell, we can
use the parametric representation of a sphere (Eq.3.16), with coordinates
(𝑥, 𝑦, 𝑧) relative to the cell centre.
⎧
⎪
⎪𝑥 = 𝑅 sin 𝜃 cos 𝜙
⎪
⎪
⎪
⎨
𝑦 = 𝑅 cos 𝜃 (3.16)
⎪
⎪
⎪
⎪
⎩𝑧 = 𝑅 sin 𝜃 sin 𝜙
⎪

If we assume that the new cylindrical cell will be characterised by radius

and height (ℎ, 𝑟), the shape change from spherical to cylindrical will require
that new positions of particles in the elongated cell satisfy equation (3.17).
⎧
𝑥′
⎪
⎪
⎪
⎪ = 𝑟 cos 𝜙
⎪
⎨
𝑦′ = ℎ𝑦 (3.17)
⎪
⎪
⎪
⎩𝑧 ′
⎪
= 𝑟 sin 𝜙
⎪
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 37

From equations (eq. 3.16) and (eq. 3.17), we deduce that:

⎧
𝑥′ √ 𝑟
⎪
⎪
⎪
⎪ = 𝑥2 +𝑧 2
𝑥
⎪
⎨
𝑦′ = ℎ𝑦 (3.18)
⎪
⎪
⎪
⎩𝑧 ′ √ 𝑟
⎪
= 𝑧
⎪
𝑥2 +𝑧 2

We can therefore compute required equilibrium lengths for springs 𝑖𝑗

and 𝜒𝑖 by calculating particle distances from their new positions.
⎧ √︃(︂ )︂2 (︂ )︂2
⎪
𝜒𝑖 √ 𝑟
ℎ2 𝑦𝑖2 √ 𝑟
⎪
⎨𝑟eq = 𝑥𝑖 + + 𝑧𝑖
⎪
⎪
𝑥2𝑖 +𝑧𝑖2 𝑥2𝑖 +𝑧𝑖2
√︃(︂ )︂2 (︂ )︂2
⎪
2
𝑖𝑗 √ 𝑟
𝑥𝑖 − √ 𝑟 √ 𝑟
𝑧𝑖 − √ 𝑟
⎪
⎩𝑟eq
⎪
⎪ = 𝑥𝑗 + ℎ2 (𝑦𝑖 − 𝑦𝑗 ) + 𝑧𝑗
𝑥2𝑖 +𝑧𝑖2 𝑥2𝑗 +𝑧𝑗2 𝑥2𝑖 +𝑧𝑖2 𝑥2𝑗 +𝑧𝑗2

(3.19)

Figure 3.2D demonstrates this shape change.

MG#: A modelling framework and simulation platform for cell and tissue
mechanics 38

Figure 3.2: Regular cell shapes obtained by shape metamorpho-

sis. A. Top: initial spherical cell. Bottom: Ellipsoid cell shape represent-
ing an elongated cell. Parameter values: 𝑎 = 0.707, 𝑏 = 1.25, 𝑐 = 1 B. Top:
initial spherical cell. Bottom: Biconcave cell shape characteristic of Red
Blood Cells. Parameter values: 𝑎0 = 0.0518, 𝑎1 = 0.0518, 𝑎2 = −4.491.
C. Top: initial spherical cell. Bottom: Red Blood cells biconcave shape is
not attained when radial forces are suppressed, though simulation time is
10 times longer than in C. Parameter values: 𝑎 = 0.707, 𝑏 = 1.25, 𝑐 = 1.
D. Top: initial spherical cell. Bottom: Cylindrical cell shape characteristic
of epithelial cells. Parameter values: ℎ = 1, 𝑟 = 0.5.
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 39

Figure 3.3: Convergence of Red Blood cell Shapes. A. Cell shapes

obtained for different values of the simulation time step. B. Elastic en-
ergy charts for different simulation scenarios. Higher time steps con-
verge towards non-zero values, but yield unstable simulations (blue, yel-
low ). Lower time steps converge towards zero and yield stable simulations
(green, red, purple, brown). In each of these simulations, we use parameter
values 𝑎0 = 0.0518, 𝑎1 = 0.0518, 𝑎2 = −4.491.

3.1.2 Apoptosis

Biological life is an equilibrium between life and death. In order for organisms
to live and develop, new cells need to be born, while old cells need to die.
Apoptosis, which designates the programmed death of biological cells, is one
the mechanisms that mediate the right balance between life and death in
living beings. Dysfunctional control of cell birth and death mechanisms lead
to several physiological diseases including cancers [33, 89]. Apoptosis is also
very instrumental in other development processes such as the separation of
digits [140].

Here, we propose to simulate apoptosis by gradually reducing the vol-

ume of the cell. For this, we shrink all cell spring equilibrium lengths to
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 40

zero. As a response, the cell spring lengths decrease until the cell reaches a
negligible volume (Fig.3.4).

Figure 3.4: A cell undergoing apoptosis. A cell decreases in volume

until it vanishes

3.1.3 Cell polarity

Polarity is an important property of biological cells. In epithelial tissues, cells

exhibit an apico-basal polarity with a strict distinction between their basal
and apical faces. This type of polarisation enables absorptive epithelia to
extract materials from a particular side of the sheet [55], or form multicellular
structures such as tubes [28], or rosettes [65]. Mesenchymal cells, on the
other hand, feature polarity through mechanisms such as protrusion, which
induce directed shape changes and cell movement [57]. Polarisation, along
with apoptosis, appears to play a critical in lumenogenesis [33, 155].

In MG#, we propose to model polarity by assigning to a membrane

particle the lead role in the cell. In this way, the said particle defines the
orientation and polarity of the cell. Figure 3.5 shows different scenarios of
orientation for apico-basal polarity of an epithelial cell. The different cells
have been obtained by shape transformation of the same initial spherical cell.
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 41

Figure 3.5: Cell Polarity. Different spatial orientations for the api-
cobasal polarity of an epithelial cell. In each case, the epithelial cell has
been obtained by shape metamorphosis of an initial spherical cell. A dif-
ferent polarity orientation has been specified for each cell.

3.1.4 Cell cycle and division

Cell cycle is a mechanism at the heart of morphogenesis. From the zygote

stage to the adult vertebrate, cell growth and division ensure that the or-
ganism develops and renews its organs. We implement a simple cell cycle
with a cycle period for each cell. While going through their cycle, each cell
increments its internal clock, then divides into two cells when their counter
reaches the cycle period.

The process of division takes as input the cell’s mesh and outputs two
meshes corresponding to the daughter cells (Fig. 3.6). It starts with comput-
ing the division plane, which is fully determined by a normal vector (→−
𝑛 ) and
a characteristic point (𝑂). The normal vector is either explicitly given as a
parameter, or numerically determined by computing the elongation axis of
the cell. For symmetric division, we use the centre of the cell as characteristic
point. Our model may also account for asymmetric division by choosing a
point different from the centre of the cell to compute the division plane. Hav-
ing set the division plane, each particle’s position relative to this plane (𝑃𝑖 )
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 42

−−→ −
is determined by evaluating the sign of the dot product 𝑂𝑃𝑖 · →
𝑛 , and using
the terminology ‘above’ if it is positive and ‘below’ otherwise. This operation
partitions the cell mesh into two distinct sets: the ‘upper mesh’, made of all
particles whose relative position to the plane is above, and the ‘lower mesh’,
consisting of particles which are below the plane. However, these meshes are
not equivalent in terms of number of particles and topology. In order to pre-
serve these characteristic features of our model, we extend each mesh with
new particles obtained by an orthogonal projection on the division plane of
the other mesh’s vertices.

In the cleavage simulation below (Fig. 3.6), a single cell (zygote) goes
through a series of successive divisions. At each division, the division plane
is chosen such that its normal vector maximises the dispersion of the cloud
point (formed by the cell particles) around their centre of mass. The vector
is determined by calculating the co-variance matrix of the cloud point and
then computing the eigen vector corresponding to the minimal eigen value.

Figure 3.6: A simulated cell going through a series of successive divisions

MG#: A modelling framework and simulation platform for cell and tissue
mechanics 43

3.2 MG#: The open simulation framework

To produce simulations based on our model, we initially built a modelling

platform within the game engine Unity1 . It had the advantage of offering
a unique environment for programming, running and viewing simulations.
However, combining at every time step the computation of particle config-
urations together with their rendering significantly slowed down the whole
process. In addition, once a simulation was done, all results were lost and
should the same experiment be run again, it was not possible to reuse or
play back the previous calculations. Furthermore, a unity-based simulation
platform could not allow us to run simultaneously multiple variants of a
simulation.

To overcome these limitations, and following the lead of Chaste, an open

source library for computational biology and physiology [99], we adopted a
design principle driven by Dijkstra’s separation of concerns [37]. We opted
for a separation of computing and visualisation into two distinct programs.
On the one hand, we developed in C# a custom back-end physics engine spe-
cific to our model, and independent from any graphical interface, based on
implementations of algebraic and geometric primitives described by Hardy
and Steeb in their book [66]. Figure 3.7 gives a brief overview in UML rep-
resentation of the class diagram of MG#. This new physics engine updates
at every time frame cell particles’ positions and thus the spatial configura-
tion of simulated cell populations. At the end of simulations, the physics
engine logs these configurations to log files using either the well established
VTK format, or our custom format (.MG files), which has the advantage
of producing overall more compact logs. On the other hand, we developed
a Unity-based viewer to render simulations logged in our custom (Fig. 3.8).
1
https://unity3d.com
MG#: A modelling framework and simulation platform for cell and tissue
mechanics 44

VTK log files can be rendered with open source visualisation tools such as
paraview [1].

Figure 3.7: MG# Class Architecture. An MG# user experiment

*.UserExperiment (blue) is implemented in the form of a simulation class
which inherits from the base simulator class (MGSharp.Core.Simulator ).
A simulation is performed on a population of cells (*.BiologicalEnti-
ties.CellPopulation), which may be partitioned into tissues (*.Biologi-
calEntities.Tissue), which themselves are specific cell populations. Each
cell population or tissue is made of cells (*.BiologicalEntities.MGCell ),
whose representation in the 3D space are geometric meshes (*.Geomet-
ricPrimitives.Mesh). The generic mesh class is built upon more el-
ementary geometric classes not shown here (vector, face, etc.) and
extended from Hardy and Steeb implementations [66]. The generic
MGCell can be specialised further into child classes with specific be-
haviours (*.BiologicalEntities.UserClass). Helper classes and methods
(MGSharp.Core.Helpers.* ), and the generic model class (*.MGMod-
els.MGModel ) that contains the definition of most generic model parame-
ters, are transversal to this architecture, and can be used at any level. An
instance of a user experiment class (green, underlined) implements inher-
ited methods from the Simulator class, and may define custom methods.
Modelling of epithelial morphogenesis 45

Figure 3.8: Simulation framework: GUI-less physics engine and viewer

Chapter 4

Modelling of epithelial
morphogenesis

Epithelial sheets can be defined as densely packed arrays of cells tightly

connected at their junctions to form layers of cells. Their inherent rigid
topology confers them natural functions of acting as barriers and selective
filters [55]. Epithelial sheets are intensely present in adult organisms. They
form the tissues that absorb nutrients and metabolites from the environment,
and make up the tubes, canals and cavities of glandular organs. They are
responsible for transport, filtration, the synthetic functions of endocrine and
exocrine, and for maintaining different electric potentials between parts of
the body [82]. However, of much more interest to us, epithelia play a central
role in morphogenesis: the complex geometry of embryos and organs can
often be traced back to biomechanical processes occurring within epithelial
sheets [32, 69]. Spectacular cellular arrangements during key morphogenesis
events such as Neural Tube Closure in Xenopus [25], the formation of the
ventral furrow during Drosophila gastrulation [19], or the remodelling of
mouse embryonic and extraembryonic tissues during implantation [8, 9, 26],
47
Modelling of epithelial morphogenesis 48

are all driven by epithelia. In general, epithelial tissues present a set of

canonical behaviours from which proceed the basic structure of most organs.
These include tissue elongation or shortening, folding, spreading, budding,
cavitation, and delamination of epithelial sheets [69, 82]. Here, we would
like to show how we can use the model presented in the previous chapter to
simulate a subset of these canonical behaviours.

In biological development, higher scale processes often emerge as a re-

sult of lower scale dynamics. This key feature of morphogenesis is present
in epithelial sheets, where tissue morphogenesis is inextricably linked with
questions of cell behaviour[55]. However, cell shape changes may proceed
from either active or passive response to the mechanical stress or chemical
signals they receive [32]. For instance, Apical Constriction (AC), which is
known to be one of the mechanisms driving bending in epithelial sheets, has
been shown to be a result of lower scale cellular biochemical processes [131].
On the other hand, cells transition from squamous to columnar shape might
be tributary to stresses imposed by tissues’ boundaries [26, 32]. However, in
both scenarios, understanding the processes by which shape changes in ep-
ithelial tissues are brought about require the study of how their elementary
units, epithelial cells, behave. Furthermore, even when the action of indi-
vidual cells trigger epithelial morphogenesis, the coordination of such action
remains essential to the formation of coherent tissues[55]. In the first part
of this chapter we will focus on modelling the behaviours of single epithelial
cells. From there, we will move on to address the coordinated implementation
of these behaviours in epithelial sheets to simulate emergent phenomena.
Modelling of epithelial morphogenesis 49

4.1 Morphological changes in single epithelial

cells

In this section we show how we can use MG# to model single epithelial
cell shape changes. Here, we consider modelling the mechanics of Planar Po-
larised Constriction (PPC), Apical Constriction (AC) and Expansion, Apical
Constriction with volume conservation. Finally, we examine how we can in-
tegrate chemical variables to simulate Apical Constriction.

To describe single cell shape changes in epithelial tissues, we use the

MG# mechanical framework presented in the previous chapter. Epithelial
cell shapes can be abstracted as cells with a cylindrical geometry and apical-
basal polarity. In MG#, cell shape changes emanate from alterations to
equilibrium lengths of cell springs. Hence, triggering these behaviours re-
quires setting new equilibrium lengths for each cell. Because, in addition to
their simple geometry, elementary shape changes in epithelial cells may also
exhibit straightforward geometry, we will establish analytical equations to
describe these changes. In the following equations, cell dimensions are as-
sumed to be normalised by a standard distance, making them dimensionless,
and the equations non-dimensionalised.

4.1.1 Planar Polarised Constriction

4.1.1.1 Biological motivation

Planar Polarised Constriction (PPC) is the process by which epithelial cells

constrict their lateral domain. PPC exists as a consequence of Planar Polar-
ity, which, unlike the characteristic apicobasal polarity of epithelial sheets,
Modelling of epithelial morphogenesis 50

is generated within the plane of the sheet[12]. This type of polarity co-
ordinates the asymmetric distribution of molecules within individual cells,
creating forces that break the symmetry of the cell and causes it to constrict
laterally. Planar Polarised Constriction is known to play a role in multicel-
lular rosette formation, convergent extension and the elongation of epithelial
sheets [12, 65, 87].

4.1.1.2 Computational modelling

In this section, for purposes of simplification, we consider an epithelial cell

with a square base. In analogy to biology, we simulate planar polarised
constriction by shrinking the equilibrium lengths of cell springs. The closer
the springs are to the constricting side, the greater we reduce their resting
lengths. Here, we show how we obtain new equilibrium lengths when we
target to reduce the width of the constricting face by a total length of 2𝑑.

We consider a 2D cut of the constricted cell as in figure 4.1B. In addition,

let us consider sub-cellular vertices 𝑀𝑖 (𝑅, 𝑦𝑖 , 𝑧𝑖 ) and 𝑀𝑗 (𝑅, 𝑦𝑗 , 𝑧𝑗 ) located
on the lateral side of the cell. Once the cell constricts by reducing one of
its lateral sides by 𝑑, vertices 𝑀𝑖 and 𝑀𝑗 will have moved respectively to
positions 𝑀𝑖′ (𝑥′𝑖 , 𝑦𝑖 , 𝑧𝑖 ) and 𝑀𝑗′ (𝑥′𝑗 , 𝑦𝑗 , 𝑧𝑗 ).

Equilibrium lengths of membrane spring 𝑖𝑗 and 𝜒𝑖 in the constricted

cell are given by equation (4.1).
⎧
−−→
𝜒𝑖
⎨ 𝑟eq (𝑑) = ||𝑂𝑀𝑖′ ||
⎪
(4.1)
⎩ 𝑟𝑖𝑗 (𝑑) = ||−
⎪ −−→
𝑀𝑖′ 𝑀𝑗′ ||
eq
Modelling of epithelial morphogenesis 51

Finding equilibrium lengths necessary to constraint the cell to undergo

planar polarised constriction is therefore equivalent to computing the coor-
dinates of vertices 𝑀𝑖′ and 𝑀𝑗′ . Given that the same process can be used to
calculate the coordinates of 𝑀𝑖′ and 𝑀𝑗′ , we will show the steps for 𝑀𝑖′ , and
use equivalent results for 𝑀𝑗′ .

We define point 𝑂𝑖 (0, 𝑦𝑖 , 𝑧𝑖 ) as the centre of the disc formed by the

intersection of the cell and the plane containing 𝑀𝑖 and parallel to the base
−−−→ −−−→
of cell (see figure 4.1). Using the Chasles theorem for vectors 𝑂𝑖 𝑀𝑖′ , 𝑂𝑖 𝑀𝑖
−−−→
and 𝑀𝑖′ 𝑀𝑖 , we can write:

−−−→′ −−−→ −−− →

𝑂𝑖 𝑀𝑖 = 𝑂𝑖 𝑀𝑖 − 𝑀𝑖′ 𝑀𝑖 (4.2)

The Thales theorem applied to the triangle (𝐴′ 𝐵𝐴) yields

−−−→ −−→
||𝑀𝑖′ 𝑀𝑖 || ||𝑀𝑖 𝐵||
−−→ = −→ (4.3)
||𝐴′ 𝐴|| ||𝐴𝐵||

Hence −−→
−−− → ||𝑀𝑖 𝐵|| −−→
||𝑀𝑖 𝑀𝑖 || = −→ ||𝐴′ 𝐴||
′
(4.4)
||𝐴𝐵||

−−−→ −−→
Meanwhile, 𝑂𝑖 𝑀𝑖 and 𝐴′ 𝐴 are co-linear, and

−−→ 𝑑 −−−→
𝐴′ 𝐴 = × 𝑂𝑖 𝑀𝑖 (4.5)
𝑅

Combining equations (4.2), (4.3), (4.4) and (4.5), we can deduce

−−→
−−−→′ −−−→ 𝑑 ||𝑀𝑖 𝐵|| −−−→
𝑂𝑖 𝑀𝑖 = 𝑂𝑖 𝑀𝑖 − × −→ 𝑂𝑖 𝑀𝑖 (4.6)
𝑅 ||𝐴𝐵||
Modelling of epithelial morphogenesis 52

Hence, (︃ −−→ )︃
−−−→′ 𝑑 ||𝑀𝑖 𝐵|| −−−→
𝑂𝑖 𝑀𝑖 = 1 − × −→ 𝑂𝑖 𝑀𝑖 (4.7)
𝑅 ||𝐴𝐵||

One can easily verify that

−−→
||𝑀𝑖 𝐵|| 𝑧𝑖 + 𝑅
−→ = 2𝑅 (4.8)
||𝐴𝐵||

Hence
−−−→′
(︂ )︂
𝑑 (𝑧𝑖 + 𝑅) −−−→
𝑂𝑖 𝑀𝑖 = 1− 𝑂𝑖 𝑀𝑖 (4.9)
2𝑅2

Hence, in the coordinates system, we get

⎛ (︁ )︁ ⎞
𝑑(𝑧𝑖 +𝑅)
1− 2𝑅2
𝑥𝑖
−−−→′ ⎜ ⎟
𝑂𝑖 𝑀𝑖 = ⎜ (4.10)
⎜ ⎟
0 ⎟
⎝ ⎠
0

Hence the coordinates of 𝑀𝑖′ , the new position of particle 𝑖 in the con-
stricted cell is given by equation (4.11).
(︂(︂ )︂ )︂
𝑑 (𝑧𝑖 + 𝑅)
𝑀𝑖′ 1− 𝑥𝑖 , 𝑦𝑖 , 𝑧𝑖 (4.11)
2𝑅2

Similarly, for a particle 𝑗 with position 𝑀𝑗 on the cylindrical cell, its

position on the constricted cell will be given by equation (4.12).
(︂(︂ )︂ )︂
𝑑 (𝑧𝑗 + 𝑅)
𝑀𝑗′ 1− 𝑥𝑗 , 𝑦𝑗 , 𝑧𝑗 (4.12)
2𝑅2

From this and equations (4.1) and (4.2), we deduce new equilibrium
lengths for all springs 𝑀𝑖 𝑀𝑗 and 𝑂𝑀𝑖 required to constrict apically our
Modelling of epithelial morphogenesis 53

epithelial cell.
⎧ √︂(︁ )︁2
𝜒𝑖 𝑑(𝑧𝑖 +𝑅)
1 − 2𝑅2 𝑥2𝑖 + 𝑦𝑖2 + 𝑧𝑖2
⎪
⎨ 𝑟eq (𝑑) =
⎪
√︂(︁(︁ )︁ (︁ )︁ )︁2
𝑑(𝑧𝑗 +𝑅)
𝑖𝑗 𝑑(𝑧𝑖 +𝑅)
𝑥𝑗 + (𝑦𝑖 − 𝑦𝑗 )2 + (𝑧𝑖 − 𝑧𝑗 )2
⎪
⎩ 𝑟eq
⎪ (𝑑) = 1 − 2𝑅2 𝑥𝑖 − 1 − 2𝑅2

(4.13)

Figure 4.1C showcases a simulation of planar polarised constriction.

Figure 4.1: Planar Polarised Constriction. A. Apicobasal polari-

sation of a planar polarised constricted cell B. A 2D cut of a constricted
cell. The cell underwent planar polarised constriction by reducing its
lateral width by a length of 2𝑑. C. Simulation of planar polarised con-
striction using an epithelial cell with square base. Parameter values:
𝑅 = 0.5, 𝑑 = 0.35

4.1.2 Apical Constriction & Apical Expansion

4.1.2.1 Biological motivation

Apical constriction can be defined as the shrinkage of the apical surface of

an epithelial cell. AC is the result of the contraction of actomyosin networks
Modelling of epithelial morphogenesis 54

near the apical face of the cell[93]. This seemingly simple behaviour of indi-
vidual cell may induce dramatic changes in a sheet’s morphology [92, 131].
Apical Constriction is involved in many processes driving morphogenesis in
development, including the formation of tubes [3, 28], neurulation [25, 75],
or during gastrulation in several species [129]. The roles and regulation of
AC in development has been extensively reviewed in [91–94, 131].

Apical Expansion is the increase of the apical surface area in an epithe-

lial cell. By definition, it appears to be the reverse process of Apical Con-
striction. However, several morphogenesis events involve Apical Expansion.
It has been shown that Apical Expansion, under the influence of Cadherin
99C, fosters epithelial tube elongation [29]. Apical Expasion has also been
found to be a driver of ascidian gastrulation [32]. Hence, we propose to
simulate apical expansion, in addition to apical constriction.

4.1.2.2 Computational modelling

We simulate apical constriction (resp. expansion) by shrinking the equilib-

rium lengths of membrane springs. To obtain conic shapes, the new resting
lengths among nodes at a given height are set to decrease (resp. increase)
with the distance from the basal face. Here we will establish the expres-
sions for new equilibrium lengths when we target to reduce (resp. increase)
the apical radius (𝑅) by a total amount of 𝑑. In order to account for both
Constriction and Expansion, we allow 𝑑 to be negative, however, with the
constraint that −𝑅 < 𝑑.

We will consider a 2D projection of an epithelial cell as described in

figure 4.2B. In addition, we consider sub-cellular vertices 𝑀𝑖 (𝑥𝑖 , 𝑦𝑖 , 𝑧𝑖 ) and
𝑀𝑗 (𝑥𝑗 , 𝑦𝑗 , 𝑧𝑗 ) located on the lateral side of the cell. Once the cell constricts
Modelling of epithelial morphogenesis 55

by reducing its apical radius by 𝑑, vertices 𝑀𝑖 and 𝑀𝑗 would have moved

respectively to positions 𝑀𝑖′ (𝑥′𝑖 , 𝑦𝑖′ , 𝑧𝑖′ ) and 𝑀𝑗′ (𝑥′𝑗 , 𝑦𝑗′ , 𝑧𝑗′ ).

Equilibrium lengths of membrane spring 𝑖𝑗 for the constricted cell is

given by equation (4.14).
⎧
−−→
𝜒𝑖
⎨ 𝑟eq (𝑑) = ||𝑂𝑀𝑖′ ||
⎪
(4.14)
⎩ 𝑟𝑖𝑗 (𝑑) = ||−
⎪ −−→
𝑀𝑖′ 𝑀𝑗′ ||
eq

Hence, finding equilibrium lengths necessary to constraint the cell to

undergo apical constriction is equivalent to computing the coordinates of
vertices 𝑀𝑖′ and 𝑀𝑗′ . The same process can be used to calculate the coordi-
nates of 𝑀𝑖′ and 𝑀𝑗′ . Therefore, we will show the steps for 𝑀𝑖′ , and use the
similar result for 𝑀𝑗′ .

We define point 𝑂𝑖 (0, 𝑦𝑖 , 0) as the centre of the disc formed by the in-
tersection of the cell and the plane containing 𝑀𝑖 and parallel to the base
−−−→ −−−→
of cell (see figure 4.2B). Using the Chasles theorem for vectors 𝑂𝑖 𝑀𝑖′ , 𝑂𝑖 𝑀𝑖
−−−→
and 𝑀𝑖′ 𝑀𝑖 , we can write:

−−−→′ −−−→ −−− →

𝑂𝑖 𝑀𝑖 = 𝑂𝑖 𝑀𝑖 − 𝑀𝑖′ 𝑀𝑖 (4.15)

Using the property of Thales applied to the triangle (𝐴′ 𝐵𝐴)

−−−→ −−→
||𝑀𝑖′ 𝑀𝑖 || ||𝑀𝑖 𝐵||
−−→ = −→ (4.16)
||𝐴′ 𝐴|| ||𝐴𝐵||

Hence −−→
−−− → ||𝑀𝑖 𝐵|| −−→
||𝑀𝑖 𝑀𝑖 || = −→ ||𝐴′ 𝐴||
′
(4.17)
||𝐴𝐵||
Modelling of epithelial morphogenesis 56

−−−→ −−→
Meanwhile, 𝑂𝑖 𝑀𝑖 and 𝐴′ 𝐴 are colinear, and

−−→ 𝑑 −−−→
𝐴′ 𝐴 = × 𝑂𝑖 𝑀𝑖 (4.18)
𝑅

Therefore,
−−→
−−−→′ −−−→ 𝑑 ||𝑀𝑖 𝐵|| −−−→
𝑂𝑖 𝑀𝑖 = 𝑂𝑖 𝑀𝑖 − × −→ 𝑂𝑖 𝑀𝑖 (4.19)
𝑅 ||𝐴𝐵||

Hence, (︃ −−→ )︃
−−−→′ 𝑑 ||𝑀𝑖 𝐵|| −−−→
𝑂𝑖 𝑀𝑖 = 1 − × −→ 𝑂𝑖 𝑀𝑖 (4.20)
𝑅 ||𝐴𝐵||

One can easily verify that

−−→ ℎ
||𝑀𝑖 𝐵|| 𝑦𝑖 + 2
−→ = ℎ (4.21)
||𝐴𝐵||

Hence (︃ )︀ )︃
𝑑 𝑦𝑖 + ℎ2
(︀
−−−→′ −−−→
𝑂𝑖 𝑀𝑖 = 1− 𝑂𝑖 𝑀𝑖 (4.22)
𝑅ℎ

In the Cartesian coordinates system, we get

⎛ (︂ )︂ ⎞
𝑑(𝑦𝑖 + ℎ
2)
⎜ 1− 𝑅ℎ
𝑥𝑖 ⎟
−−−→′ ⎜ ⎟
𝑂𝑖 𝑀𝑖 = ⎜ (4.23)
⎜ ⎟
0 ⎟
⎜ (︂ )︂ ⎟
⎝ 𝑑(𝑦𝑖 + ℎ
2)
⎠
1 − 𝑅ℎ 𝑧𝑖

Hence the coordinates of 𝑀𝑖′ , image of 𝑀𝑖 by the transformation that

transforms the cylindrical cell to the cone-shaped apical constricted cell is
Modelling of epithelial morphogenesis 57

given by equation (4.24).

(︃(︃ )︀ )︃ (︃ )︀ )︃ )︃
𝑑 𝑦𝑖 + ℎ2 𝑑 𝑦𝑖 + ℎ2
(︀ (︀
𝑀𝑖′ 1− 𝑥𝑖 , 𝑦𝑖 , 1 − 𝑧𝑖 ) (4.24)
𝑅ℎ 𝑅ℎ

Similarly, for a particle 𝑀𝑗 on the cylindrical cell, its position on the

constricted cell will be given by equation (4.25).
(︃(︃ )︀ )︃ (︃ )︀ )︃ )︃
𝑑 𝑦𝑗 + ℎ2 𝑑 𝑦𝑗 + ℎ2
(︀ (︀
𝑀𝑗′ 1− 𝑥𝑗 , 𝑦𝑗 , 1 − 𝑧𝑗 (4.25)
𝑅ℎ 𝑅ℎ

From this and equations (4.14) and (4.15) we deduce the new equilib-
rium lengths for all springs 𝑀𝑖 𝑀𝑗 and 𝑂𝑀𝑖 required to constrict apically our
epithelial cell.

√︁
𝜒𝑖
𝑟eq (𝑑) = (𝐴(𝑥𝑖 ))2 + 𝑦𝑖2 + (𝐴(𝑧𝑖 ))2 (4.26)
√︁
𝑟eq (𝑑) = (𝐴(𝑥𝑖 ) − 𝐴(𝑥𝑗 ))2 + (𝑦𝑖 − 𝑦𝑗 )2 + (𝐴(𝑧𝑖 ) − 𝐴(𝑧𝑗 ))2
𝑖𝑗
(4.27)

𝑑( ℎ
(︁ )︁
+𝑦𝑘 )
Here, 𝐴(𝑡𝑘 ) = 1 − 2
𝑅ℎ
𝑡𝑘 and 𝑥𝑖,𝑗 , 𝑦𝑖,𝑗 , 𝑧𝑖,𝑗 are relative to the centre of
the cell.

An interesting feature of these equations is that they are independent

of the topology of the initial mesh, which also needs to be cylindrical in
shape. Figures 4.1C and 4.1D show examples of apical constriction with two
different types of cylindrical meshes, respectively with square and hexagonal
base.
Modelling of epithelial morphogenesis 58

Figure 4.2: Apical Constriction. A. Schema of an apically constricted

epithelial cell with hexagonal base. B. A 2D section of a constricted cell.
The cell underwent apical constriction by reducing its apical radius by a
length of 𝑑. Initial cylindrical section in white, constricted cell in red.
C. Simulation of apical constriction with an epithelial cell with square
base. Parameter values: 𝐻 = 2, 𝑅 = 0.5, 𝑑 = 0.35. D. Simulation of
apical constriction with an epithelial cell with hexagonal base. Parameter
values: 𝐻 = 2, 𝑅 = 0.5, 𝑑 = 0.35.
Modelling of epithelial morphogenesis 59

4.1.3 Apical constriction with volume conversation:

basal-lateral modulation

4.1.3.1 Biological motivation

In several situations during their shape changes, cells attempt to maintain

their volume unchanged. This feature has been observed in certain cases of
Apical Constriction, where the cell, while constricting, simultaneously em-
ploys volume conservation mechanisms to compensate the loss of volume
induced by the constriction of the apical side. These mechanisms involve
basal modulation, in which the epithelial cell enlarges it basal face, and
lateral lengthening, where the cell increases its lateral height [32, 136, 151].
These variants to AC are present in multiple morphogenesis events in epithe-
lial tissues. For instance, there is evidence that coordination between apical
and basolateral contractility is necessary for ascidian endoderm invagination
[32, 136]. As in simple Apical Constriction and Planar Polarised Constric-
tion, these shape changes are dependent on the differential localisation of
activated actomyosin networks within the cell.

4.1.3.2 Computational modelling

In the scenario presented in figure 4.3B, the cell reduces its apical radius by a
given amount 𝑑, and, as a consequence of volume conservation, increases its
lateral height by 𝐻0 , and its basal radius by 𝑅0 . In order to enforce volume
conservation on the constricting cell, we need to compute appropriate values
𝐻0 and 𝑅0 .
Modelling of epithelial morphogenesis 60

Computing 𝑅0 and 𝐻0

We consider an epithelial cell with cylindrical shape. In its initial state, the
volume of the cell is given by

𝒱0 = ℬ × 𝐻 (4.28)

Here, ℬ is the surface area of the cylinder base, the specific formula
for ℬ depending on the actual shape of the base. In regular cylinders (with
regular base), the formula for ℬ will be given by

ℬ = 𝛾 × 𝑅2 (4.29)

Here, 𝛾 is a multiplicative factor. For a square base, 𝛾 = 4, for the

inscribed circular base 𝛾 = 𝜋 , for the inscribed hexagonal base (inside the
circle) 𝛾 = 3 etc.

Hence, the volume of the cell at its initial resting state is

𝒱0 = 𝛾 × 𝑅2 × 𝐻 (4.30)

On the other hand, once the cell is fully constricted having reduced its
apical radius by 𝑑, increased its height by 𝐻0 and its basal radius by 𝑅0 , the
volume is given by
∫︁ 𝐻
2
𝒱= 𝛾𝑟′2 (𝑦 ′ )𝑑𝑦 ′ (4.31)
−𝐻
2
−𝐻0

Here, 𝑟′ (𝑦 ′ ) is the radius of the constricted cell at height 𝑦 ′ . If the cell

radius decreases in an homogeneous way, 𝑟′ (𝑦 ′ ) takes the form of an affine
Modelling of epithelial morphogenesis 61

function such that

𝐻 𝐻
𝑟′ (− − 𝐻0 ) = 𝑅 + 𝑅0 and 𝑟′ ( ) = 𝑅 − 𝑑 (4.32)
2 2

Solving 𝑟′ (𝑦 ′ ), we get

𝑅0 + 𝑑 ′ (𝑑 + 𝑅0 ) × 𝐻
𝑟′ (𝑦 ′ ) = − 𝑦 +𝑅−𝑑+ (4.33)
𝐻 + 𝐻0 2(𝐻 + 𝐻0 )

Hence
𝐻 + 𝐻0 ′
𝑑𝑦 ′ = − 𝑑𝑟 (4.34)
𝑅0 + 𝑑

By a change of variable, we get the following equation for the volume

∫︁ 𝑅−𝑑 (︂ )︂
𝐻 + 𝐻0 ′
′2
𝒱= 𝛾𝑟 − 𝑑𝑟 (4.35)
𝑅+𝑅0 𝑅0 + 𝑑
𝐻 + 𝐻0 𝑅−𝑑 ′2 ′
∫︁
= −𝛾 × 𝑟 𝑑𝑟 (4.36)
𝑅0 + 𝑑 𝑅+𝑅0
𝛾 𝐻 + 𝐻0 [︀ ′3 ]︀𝑅−𝑑
=− × 𝑟 𝑅+𝑅0 (4.37)
3 𝑅0 + 𝑑
𝛾 𝐻 + 𝐻0 (︀
(𝑅 − 𝑑)3 − (𝑅 + 𝑅0 )3
)︀
=− × (4.38)
3 𝑅0 + 𝑑
𝛾 𝐻 + 𝐻0
(−𝑑 − 𝑅0 ) (𝑅 − 𝑑)2 + (𝑅 − 𝑑)(𝑅 + 𝑅0 ) + (𝑅 + 𝑅0 )2
(︀ )︀
=− ×
3 𝑅0 + 𝑑
(4.39)
𝛾
× (𝐻 + 𝐻0 ) (𝑅 − 𝑑)2 + (𝑅 − 𝑑)(𝑅 + 𝑅0 ) + (𝑅 + 𝑅0 )2
(︀ )︀
= (4.40)
3

Hence,

𝛾
× (𝐻 + 𝐻0 ) 𝑅02 + (3𝑅 − 𝑑)𝑅0 + 3𝑅2 − 3𝑅𝑑 + 𝑑2
(︀ )︀
𝒱= (4.41)
3
Modelling of epithelial morphogenesis 62

The volume conservation hypothesis dictates that 𝒱0 = 𝒱. Hence, the

right hand terms of equations (4.30) and (4.41) are equal. Therefore,

𝛾
𝛾𝑅2 𝐻 = × (𝐻 + 𝐻0 ) 𝑅02 + (3𝑅 − 𝑑)𝑅0 + 3𝑅2 − 3𝑅𝑑 + 𝑑2
(︀ )︀
(4.42)
3

Hence the following equation describing volume conservation in a con-

stricting epithelial cell. This equation is quadratic in 𝑅0 , and linear in 𝐻0 .

(𝐻 + 𝐻0 ) 𝑅02 + (3𝑅 − 𝑑)𝑅0 + 3𝑅2 − 3𝑅𝑑 + 𝑑2 − 3𝑅2 𝐻 = 0

(︀ )︀
(4.43)

Therefore, for given values of 𝑑 and 𝑅0 such that 𝑑 ≤ 𝑅, 𝑅0 >= −𝑅 and

(𝑑, 𝑅0 ) ̸= (𝑅, −𝑅), equation (4.43) gives the resulting lateral modulation.

3𝑅2 𝐻
𝐻0 = −𝐻 (4.44)
(𝑅02 + (3𝑅 − 𝑑)𝑅0 + 3𝑅2 − 3𝑅𝑑 + 𝑑2 )

Conversely, for given values of 𝑑 and 𝐻0 such that 𝑑 ≤ 𝑅 and 𝐻0 >

−𝐻, one can easily show that the expected basal modulation is given as in
equation (4.45).
√︂ (︁ )︁
3𝑅2 𝐻
(3𝑅 − 𝑑) − 4 3𝑅2 − 3𝑅𝑑 + 𝑑2 −
2
𝐻+𝐻0
− (3𝑅 − 𝑑)
𝑅0 = (4.45)
2

Special case 1: Basal modulation

Here, in order to compensate for volume loss induced by apical constric-

tion, the epithelial cell strictly increases its basal radius, keeping its height
Modelling of epithelial morphogenesis 63

unchanged. In this case, 𝐻0 = 0 and 𝑅0 is given by equation (4.46).

√︀
3(3𝑅 − 𝑑)(𝑅 + 𝑑) − (3𝑅 − 𝑑)
𝑅0 = (4.46)
2

Special case 2: Lateral modulation

Here, in order to compensate for volume loss induced by apical constric-

tion, the epithelial cell strictly increases its height, keeping its basal radius
unchanged. In this case, 𝑅0 = 0 and 𝐻0 is given by equation (4.47).

3𝑅2 𝐻
𝐻0 = −𝐻 (4.47)
(3𝑅2 − 3𝑅𝑑 + 𝑑2 )

Having computed values for 𝐻0 and 𝑅0 , we can now proceed with cal-
culating new equilibrium distances for springs connected to point 𝑀𝑖′ .

Computing the coordinates of 𝑀𝑖′

We will consider a 2D section of an epithelial cell going through apical

constriction with basal-lateral modulation (Fig.4.3B). In addition, we con-
sider sub-cellular particles 𝑖 and 𝑗 with respective positions 𝑀𝑖 (𝑥𝑖 , 𝑦𝑖 , 𝑧𝑖 ) and
𝑀𝑖 (𝑥𝑗 , 𝑦𝑗 , 𝑧𝑗 ) located on the lateral side of the cell. Once the cell constricts by
reducing its apical radius by 𝑑, particles 𝑖 and 𝑗 will have moved respectively
to positions 𝑀𝑖′ (𝑥′𝑖 , 𝑦𝑖′ , 𝑧𝑖′ ) and 𝑀𝑗′ (𝑥′𝑗 , 𝑦𝑗′ , 𝑧𝑗′ ).
Modelling of epithelial morphogenesis 64

Equilibrium lengths of membrane spring 𝑀𝑖 𝑀𝑗 for the constricted cell

is given by equation (4.48).
⎧
−−→
𝜒𝑖
⎨ 𝑟eq (𝑑) = ||𝑂𝑀𝑖′ ||
⎪
(4.48)
⎩ 𝑟𝑖𝑗 (𝑑) = ||−
⎪ −−→
𝑀𝑖′ 𝑀𝑗′ ||
eq

Hence, finding equilibrium lengths necessary to constraint the cell to

undergo apical constriction is equivalent to computing the coordinates of
vertices 𝑀𝑖′ and 𝑀𝑗′ . The same process can be used to calculate the coordi-
nates of 𝑀𝑖′ and 𝑀𝑗′ . Therefore, we will show the steps for 𝑀𝑖′ , and use the
similar result for 𝑀𝑗′ .

Here we use a different approach to the previous sections. Because ep-

ithelial cells, thanks to their shape, have a cylindrical symmetry, We consider
the coordinates of points 𝑀𝑖 and 𝑀𝑖′ in the cylindrical system of coordinates,
respectively 𝑀𝑖 (𝑟𝑖 , 𝜙𝑖 , 𝑦𝑖 ) and 𝑀𝑖′ (𝑟𝑖′ , 𝜙′𝑖 , 𝑦𝑖′ ).

To switch between cylindrical and Cartesian coordinates systems, we

have the following relationships:
⎧
⎨ 𝑥𝑖 = 𝑟𝑖 cos 𝜙 and 𝑥′𝑖 = 𝑟𝑖′ cos 𝜙′
⎪
(4.49)
⎩ 𝑧𝑖 = 𝑟𝑖 sin 𝜙 and 𝑧 ′ = 𝑟′ sin 𝜙′
⎪
𝑖 𝑖

While constricting, if the cell does not rotate, we can assume that the
angle 𝜙 is not changed by the process.

𝜙′𝑖 = 𝜙𝑖 (4.50)
Modelling of epithelial morphogenesis 65

Hence, combining equations (4.49) and (4.50), we have the following

equations for 𝑥′𝑖 and 𝑧𝑖′ :
⎧
⎨ 𝑥′𝑖 = 𝑟𝑖′ 𝑥𝑖
⎪
𝑟𝑖
(4.51)
⎩ 𝑧 ′ = 𝑟 ′ 𝑧𝑖
⎪
𝑖 𝑖 𝑟𝑖

We have already established that 𝑟𝑖′ can be expressed as an affine func-

tion of the 𝑦 coordinate of 𝑀𝑖′ as in equation (4.33). We found

𝑅0 + 𝑑 ′ (𝑑 + 𝑅0 ) × 𝐻
𝑟′ (𝑦 ′ ) = − 𝑦 +𝑅−𝑑+ (4.52)
𝐻 + 𝐻0 2(𝐻 + 𝐻0 )

Furthermore, 𝑦 ′ can be expressed as an affine function of 𝑦, such that

𝐻 𝐻 𝐻 𝐻
𝑦 ′ (− − 𝐻0 ) = − and 𝑦 ′ ( )= (4.53)
2 2 2 2

We therefore get
(︂ )︂
′ 𝐻0 𝐻0
𝑦 (𝑦) = 1 + 𝑦− (4.54)
𝐻 2

Hence, We can write 𝑟′ as a function of 𝑦

𝑑 + 𝑅0 𝑑 + 𝑅0
𝑟′ (𝑦) = − 𝑦+𝑅−𝑑+ (4.55)
𝐻 2
Modelling of epithelial morphogenesis 66

Therefore, we get the following equations for 𝑀𝑖′ coordinates:

⎧
× √ 𝑥2𝑖 2
′
(︀ 𝑑+𝑅 𝑑+𝑅0
)︀
⎪
⎪ 𝑥 𝑖 = − 𝐻
0
𝑦 𝑖 + 𝑅 − 𝑑 + 2 𝑥𝑖 +𝑧𝑖
⎪
⎪
⎪
⎨
𝑦𝑖′ = 1 + 𝐻𝐻0 𝑦𝑖 − 𝐻20 (4.56)
(︀ )︀
⎪
⎪
⎪
× √ 𝑧2𝑖 2
⎪ ′ (︀ 𝑑+𝑅
𝑦𝑖 + 𝑅 − 𝑑 + 𝑑+𝑅
)︀
⎩ 𝑧𝑖 = −
⎪ 0 0
𝐻 2 𝑥𝑖 +𝑧𝑖

For a particle 𝑗 on the cylindrical cell, its position 𝑀𝑗′ on the constricted
cell will be given by a similar equation (Eq.4.57).
⎧
′
(︀ 𝑑+𝑅 𝑑+𝑅0 𝑥
× √ 2𝑗 2
)︀
⎪
⎪ 𝑥 𝑗 = − 𝐻
0
𝑦 𝑗 + 𝑅 − 𝑑 + 2 𝑥 𝑗 +𝑧𝑗
⎪
⎪
⎪
⎨
𝑦𝑗′ = 1 + 𝐻𝐻0 𝑦𝑗 − 𝐻20 (4.57)
(︀ )︀
⎪
⎪
⎪
⎪ ′ (︀ 𝑑+𝑅0 𝑧
𝑧𝑗 = − 𝐻 𝑦𝑗 + 𝑅 − 𝑑 + 𝑑+𝑅 × √ 2𝑗 2
⎪ )︀
⎩ 0
2 𝑥𝑗 +𝑧𝑗

From these equations (4.28) and (4.29) we deduce the new equilibrium
lengths for all springs 𝑖𝑗 and 𝜒𝑖 required to enforce volume conservation
during the apical constriction of an epithelial cell.
⎧
√︀
𝜒𝑖
⎨ 𝑟eq (𝑑) = 𝑥′𝑖 2 + 𝑦𝑖′ 2 + 𝑧𝑖′ 2
⎪
√︁ (4.58)
⎩ 𝑟𝑖𝑗 (𝑑) = (𝑥′ − 𝑥′ )2 + (𝑦 ′ − 𝑦 ′ )2 + (𝑧 ′ − 𝑧 ′ )2
⎪
eq 𝑖 𝑗 𝑖 𝑗 𝑖 𝑗
Modelling of epithelial morphogenesis 67

Figure 4.3: Apical Constriction with volume conservation. A.

Schema of an apically constricted epithelial cell with hexagonal base. B.
A 2D section of a constricted cell. The cell underwent apical constriction
by reducing its apical radius by a length of 𝑑. Initial cylindrical section
in white, constricted cell in red. C. Simulation of apical constriction with
volume conservation using an epithelial cell with hexagonal base. Param-
eter values: 𝐻 = 2, 𝑅 = 0.5, 𝑑 = 0.4, 𝐻0 = 0.25. Using equation (4.45),
we find 𝑅0 = 0.2618908.

4.1.4 Integrating chemical variables: 𝐶𝑎-controlled

Apical Constriction

4.1.4.1 Biological motivation

Several studies have provided evidence for correlation of Apical Constriction

events with high concentrations of 𝐶𝑎2+ within epithelial cells. In partic-
ular, these studies have shown that periodic spikes of calcium levels pre-
ceding the enrichment in apical actin, result in pulsed contractions of cells’
actomyosin networks which drive apical constriction [25, 93, 141]. In this
Modelling of epithelial morphogenesis 68

dynamic system, cells do not constrict all at once. Rather, AC manifests

as a continuous process involving alternative phases of apical surface area
shrinkage and expansion [25]. Although these studies evidenced strong cor-
relations between the spikes and apical contractions, and established the
precedence of Ca2+ spikes, the detailed chain of causation remains to be
elucidated. Christodoulou et al. hypothesised that these could involve the
calcium-driven activation of RhoA which in turn activates ROCK, responsi-
ble for the contraction of actin filaments [25]. Moreover, it was shown that
calcium regulates the cytoskeletal dynamics of AC during mouse neural tube
closure through the secretory pathway calcium ATPase-1 [20].

Furthermore, it has been established that calcium levels in Xenopus

are regulated by inositol 1,4,5-triphosphate [5]. Building on these studies,
Kaouri et al. [78] proposed a simple mechanochemical model capturing the
dynamics of 𝐶𝑎2+ and 𝐼𝑃 3, as well as the interplay between their levels of
concentration and changes in the apical surface area of cells. In order to
demonstrate how chemical modelling can be coupled with cell mechanics in
MG#, we propose an application to the control system of calcium levels
during apical constriction.

4.1.4.2 Computational modelling

Calcium signalling within single cells involves the storage and release of cal-
cium cations in and from cellular stores such the Endoplasmic Reticulum
(ER) or the Sarcoplasmic Reticulum. Several conceptual modelling attempts
aiming at capturing these dynamics have been put forward over the years
[44]. Among those, the Atri model [5] presents the advantage of strong
agreement with experimental findings [45].
Modelling of epithelial morphogenesis 69

Kaouri et al. [78] propose an enhancement of the Atri model which

accounts for the mechanochemical feedback loop between calcium concen-
trations and the apical surface constriction. This model takes the form of
a system of three ODEs describing the evolution over time of the molecular
concentrations of 𝐶𝑎2+ (𝑐) and 𝐼𝑃 3 (ℎ), a receptor molecule in the ER, and
the dynamics of the apical surface area of the cell (𝜃).

Here, the variations of calcium concentrations are influenced by the re-

lease of calcium from ER stores into the cytoplasm through IP3 receptors
(first term in (Eq. 4.59)), the flux of calcium out of the cytoplasm (negative
term in (Eq. 4.59)), and the apical surface area (third term in (Eq. 4.59)). Re-
ciprocally, the variations of 𝐼𝑃 3 concentrations and of the apical surface area
are influenced by the concentration of calcium cations in the cell. This results
in the following non-dimensionalised equations ((Eq. 4.59) - (Eq. 4.61)).

𝑑𝑐 𝑏+𝑐 Γ𝑐
= 𝜇ℎ𝐾1 − + 𝜆𝜃 = 𝑅1 (𝑐, 𝜃, ℎ; 𝜇, 𝜆) (4.59)
𝑑𝑡 1+𝑐 𝐾 +𝑐
𝑑ℎ 𝐾22
= 2 − ℎ = 𝑅3 (𝑐, ℎ) (4.60)
𝑑𝑡 𝐾 2 + 𝑐2
𝑑𝜃
= −𝐾𝜃 𝜃 + 𝑇̂︀(𝑐) = 𝑅2 (𝑐, 𝜃) (4.61)
𝑑𝑡

𝑘𝛾
In (Eq. 4.59)-(Eq. 4.61), 𝐾1 = 𝑘𝑓 𝜏𝑘ℎ1 , Γ = 𝛾𝜏ℎ 𝑘1 , 𝐾 = 𝑘1
and 𝜆 = 𝜏 𝑆/𝑘1 .
𝜏ℎ
𝑘𝜃 = 𝜏ℎ 𝐸 ′ (1 + 𝜈 ′ )𝜉1 + 𝜉2 and 𝑇 (𝑐) = 𝑇 (𝑐)
𝜉1 +𝜉2 𝐷
and 𝐾2 = 𝑘2 /𝑘1 . Using
40
the parameter values of Atri et al. [5], they obtain 𝐾2 = 1, Γ = 7
, and
1
𝐾 = 7
. Also, taking values of 𝐸, 𝜈 and of the viscosity from Zhou et
al. [156] (𝐸 = 8.5𝑃 𝑎, 𝜈 = 0.4 and 𝜉1 + 𝜉2 = 100Pa.s) they find that 𝑘𝜃 is
𝜏ℎ 𝜏ℎ
0.4. Furthermore, 𝑇 (𝑐) = 𝑇 (𝑐)
𝜉1 +𝜉2 𝐷
= 𝑇 𝑇̂︀(𝑐),
𝜉1 +𝜉2 0𝐷
where 𝑇̂︀(𝑐) is non-
𝜏ℎ
dimensional, and they also fix 𝑇
𝜉1 +𝜉2 0𝐷
= 1.
Modelling of epithelial morphogenesis 70

Integrating the work of Kaouri et al. [78] into our model of cell me-
chanics, we propose here a mechanochemical model of apical constriction,
linking molecular regulation within epithelial cells to contractile forces caus-
ing apical constriction, i.e. making the mechanical parameters 𝑟eq functions
of certain protein concentrations ℎ (molecular concentration of Inositol 1,4,5-
triphosphate), and 𝑐 (molecular concentration of 𝐶𝑎2+ ).

In the previous section on apical constriction, we established for our

model a mathematical link between the apical radius of a constricting cell
and the mechanical forces driving the constriction (Eq.4.25). From this, we
can deduce a relationship between the surface area of a constricting cell and
those mechanical forces.

For an hexagonal epithelial cell with initial apical radius 𝑅, the apical
√
surface area is given by 𝐴0 = (3 3/2)𝑅2 , and, for a constricted cell, whose
apical radius has been reduced by 𝑑, the apical surface area is given by
√
𝐴(𝑑) = (3 3/2)(𝑅 − 𝑑)2 . One can easily establish the relationship between
𝜃, the normalised surface area, and 𝑑, given by equation (4.62) - (4.63).

1 (︁ √ )︁
𝑑= 1− 𝜃 (4.62)
2
𝐴(𝑑)
𝜃= (4.63)
𝐴0

Hence, we can express the contractile forces required to reduce cells

radius by a length 𝑑 as a function of the apical surface area via the constricted
𝑖𝑗 𝑖𝜒
cells springs’ equilibrium lengths 𝑟eq (𝑑) and 𝑟eq (𝑑), themselves functions of
𝑑 (See equation (4.25)).

With these equations, we are able to establish a direct feedback loop

between the chemical and mechanical variables that control the cell shape.
Modelling of epithelial morphogenesis 71

For a given initial state (𝑐0 , 𝜃0 , ℎ0 ), we can solve this system of ODEs and
infer the evolution of the system over the whole simulation length. Having
computed the apical surface area for each simulation frame, we can calcu-
late the corresponding quantity 𝑑 (eq. 4.62) periodically and trigger apical
constriction/expansion. Figure 4.4 shows an example simulation of apical
constriction regulated by AC levels.

Figure 4.4: 𝐶𝑎2+ regulated Apical Constriction. A. Levels of con-

centrations of 𝐶𝑎2+ (orange) and IP3 (grey) within an epithelial under-
going AC. The yellow curve shows the theoretical evolution of the apical
cell area through time. The apical faces goes through alternative phases
of growth and decrease as reported in [25]. B. Plots of theoretical and em-
pirical apical surface area as yielded by our simulation. Observed slight
differences are caused by the Euler explicit scheme used C. Snapshots of
the constricting cell at different time points showing the evolution of the
apical surface area. Parameter values: 𝐻 = 2, 𝑅 = 0.5. From (eq. 4.59 -
eq. 4.61), we fix 𝑐0 = 1, 𝜃0 = 1, ℎ0 = 1, 𝜆 = 0.5, 𝜇 = 0.289, 𝑘1 = 0.7.

4.2 Morphogenesis in epithelial tissues

The bulk of epithelial morphogenesis can be broken down into interactions

of a few canonical epithelial behaviours [69]. In this section, we describe and
Modelling of epithelial morphogenesis 72

simulate using MG# some of these canonical behaviours, notably epithelial

folding and the formation of multicellular rosettes. In order to simulate these
development episodes, we implement the single cellular behaviours described
in previous sections in the context of multicellular epithelial sheets.

4.2.1 Folding of epithelial sheets

4.2.1.1 Biological motivation

Epithelial folding is undoubtedly one of the most common regulators of shape

in development across living organisms [119]. It can be defined as the process
during which a flat epithelial sheet goes out of plane by acquiring a curved
shape. Folding is the main mechanism at the heart of important morpho-
genesis events such as gastrulation [75], Neural Tube Closure [25, 95, 141]
and the formation of epithelial tubes [3, 28]. A plethora of single epithelial
cells changes are able to drive the bending of epithelial sheets including dif-
ferent shades of apical constriction, basal expansion, and lateral modulation
[119, 139, 151]. More details on epithelial folding can be found in reviews by
Wen et al. and Pearl et al [32, 119, 151].

4.2.1.2 Computational modelling

Founded on the biological background of the previous section, we propose

to simulate the folding of a single epithelial layer. However, here we limit
ourselves to epithelial folding driven by apical constriction, the principle
being the same for other drivers of folding.

It has often been suggested in litterature that during epithelial folding,

single cell mechanisms at work are activated in discriminatory ways both in
Modelling of epithelial morphogenesis 73

space and time. Suzuki et al. [141] confirmed observations by Christodoulou

et al. [25] that 𝐶𝑎2+ pulses, the biochemical metabolism fuelling AC in
single cells during Xenopus Neural Tube Closure, occurred in neural plate
cells rather than in the non-neural epidermis, which surrounds the Neural
Plate. Hence, epidermis cells, located at the boundary of the neural plate,
did not undergo AC, nor folding morphogenesis. Furthermore, Ogura et
al.[109] evidenced the requirement of a timely wave of cellular contractility
during epithelial invagination in Drosophila tracheal placode.

In order to account for this behaviour in our modelling, we speculate the

presence of an “epicentre” within the sheet where cells actively undergo apical
constriction (Fig. 4.5A). Cells beyond the epicentre (a region of space that
we shall name the “boundary”) do not constrict apically, and only respond to
the mechanical action of cells in the epicentre. In other words, the distance
𝑑 by which we reduce the apical radius of a cell is a function of the position
of the cell within the sheet. In this case of epithelial folding, equation (4.64)
describes this function.
⎧
−−→ ⎨ 0 if cell is on boundary
⎪
𝑑(𝑋𝑐𝑒𝑙𝑙 ) = (4.64)
⎩ 𝑅 if cell is on epicentre
⎪

The coordinated movement of cells induced by these positional laws

causes the tissue to bend and fold (Fig. 4.5A,B).
Modelling of epithelial morphogenesis 74

Figure 4.5: Epithelial Folding. A. Top view of a single epithelial

layer folding under the influence of single cells at the centre of the sheet
(coloured in yellow) constricting apically. The sheet is composed of 81
epithelial cells with hexagonal base. The epicentre is represented by the
25 yellow cells at the centre of the sheet. B. Lateral view of the folding
process in A. The sheet goes out of plane. Parameter values: ℎ = 2, 𝑅 =
0.5, 𝑑 = 0.5

4.2.2 Morphogenesis of multi-cellular rosettes

4.2.2.1 Biological motivation

Multi-cellular rosettes are concentric biological structures formed by agglom-

erations of polarised cells in epithelial tissues. These epithelial structures
have observed across many species during development, including mouse
[8, 24], Drosophila, chick and zebrafish [65]. Multi-cellular rosettes play an
important role in several cellular behaviours: tissue elongation [3], lumen for-
mation and maintenance of cell pluripotency [65]. In mammalian embryos
for instance, they mediate the formation of a lumen in the early epiblast
[8, 9, 132–134], and later the pro-amniotic cavity, bridging through embry-
onic and extra-embryonic tissues [24].
Modelling of epithelial morphogenesis 75

Like many processes in epithelial sheets, mechanisms of multicellular

rosettes formation find their source in the contraction of actomyosin networks
within individual cells. However, depending on the localisation of these
contractions within the cell, two fundamental mechanisms have been found
to drive multicellular rosette formation. On the one hand, polarised apical
constriction, where actin-myosin networks localised near the apical surface
of the cell constricts leading to the narrowing of the apical domain of cells.
Here, polarity regulators (such as Par-3, Par-6, and atypical protein kinase
C) adherens junction proteins (cadherins), and tight junction proteins (ZO-1)
are responsible for the actin-myosin network localisation [8, 65]. On the other
hand, polarised apical constriction [8, 65], where the network is localised near
the apical surface of the cell, and planar polarised constriction, where the
network is localised near the constricting lateral face of the cell [12, 65]. In
both cases, cell polarity is critical to the formation of the rosette [8, 12].

4.2.2.2 Computational modelling

Founded on the biological background of the previous section, we propose to

simulate rosette formation on the basis of the distinct single cell mechanisms
that have been put forward as regulators of multicellular rosette formation.

The first scenario we simulate is the formation of planar rosettes

(Fig. 4.6A). Here, an epithelial sheet with a relatively small amount of cells
(8 cells) metamorphoses into a planar rosette by the action of single cells
triggering planar polarised constriction. In the second scenario, cells in a
single layered epithelial sheet constrict apically, prompting the sheet to turn
into a rosette (Fig. 4.6B). Finally, in a polarised epithelial tissue with two
layers, cells undergo apical constriction. This results in the emergence of a
Modelling of epithelial morphogenesis 76

near spherical epithelial rosette, reminiscent of mammalian embryonic tissues

[132–134] (Fig. 4.6C).

In a pure mathematical sense, we can extend the idea of the positional

shrinking function defined for positional folding by noticing that here also,
whether a cell constrict or not depends on the position of the cell. How-
ever, in all three cases, this function is homogeneous over the tissue, and
the presented simulations is equal to the initial apical radius of the cells.
(eq. 4.65).
−−→
𝑑(𝑋𝑐𝑒𝑙𝑙 ) = 𝑅 (4.65)
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 77

Figure 4.6: Morphogenesis of multi-cellular rosettes. A. Sim-

ulation of the morphogenesis of a multicellular rosette through planar
polarised constriction of single cells. The cell population is made of 8
epithelial cells with square base. B. Simulation of the morphogenesis of
a multicellular rosette via apical constriction of cells in a single-layered
epithelial sheet. C. Simulation of the morphogenesis of a multicellular
rosette via apical constriction of polarised cells in a double-layered epithe-
lial sheet. In B and C, the the sheets are respectively composed of 25 and
50 hexagonal epithelial cells. Parameter values: ℎ = 2, 𝑅 = 0.5, 𝑑 = 0.5
Chapter 5

Computational modelling unveils

how epiblast remodelling and
positioning rely on
trophectoderm morphogenesis
during mouse implantation

Understanding the processes by which the mammalian embryo implants in

the maternal uterus is a long-standing challenge in embryology. New insights
into this morphogenetic event could be of great importance in helping, for
example, to reduce human infertility. During implantation the blastocyst,
composed of epiblast (EPI) and trophectoderm (TE) and the primitive en-
doderm (PE), undergoes significant remodelling from an oval ball to an egg
cylinder. A main feature of this transformation is symmetry breaking and

79
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 80

reshaping of the epiblast into a “cup”. Based on previous studies, we hy-

pothesise that this event is the result of mechanical constraints originating
from the trophectoderm, which is also significantly transformed during this
process. In order to investigate this hypothesis we propose MG#, an orig-
inal computational model of biomechanics able to reproduce key cell shape
changes and tissue level behaviours in silico. With this model, we simulate
epiblast and trophectoderm morphogenesis during implantation. First, our
results uphold experimental findings that repulsion at the apical surface of
the epiblast is sufficient to drive lumenogenesis. Then, we provide new the-
oretical evidence that trophectoderm morphogenesis is sufficient to dictate
the cup shape acquisition by the epiblast and to foster its movement to-
wards the uterine tissue. We also conduct a sensitivity analysis, where we
show how different sets of model parameters influence simulation outcomes.
Together, these results offer mechanical insights into mouse implantation and
highlight the usefulness of agent-based modelling methods in the study of
embryogenesis.

5.1 Introduction

A critical milestone of mouse development is reached when the embryo im-

plants in the maternal uterine tissue [132, 149]. Prior to implantation, a
series of cell fate decisions concomitant with multiple rounds of divisions
gradually transform the initial zygote into a blastocyst featuring three dif-
ferent cell lineages: a spherical embryonic epiblast (EPI) wrapped into two
extraembryonic tissues, the trophectoderm (TE) and primitive or visceral en-
doderm (PE/VE) [4, 8]. Upon implantation, the embryo moves towards ma-
ternal sites, and undergoes significant remodelling, culminating in the case of
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 81

the mouse in an egg cylinder, a body structure essential to post-implantation

phases such as gastrulation [8, 24? ]. A key feature of this blastocyst-to-egg-
cylinder transition, still poorly understood, is the appearance of symmetry
breaking within the epiblast and its reshaping into a cup [8, 9], which occurs
roughly between stages E4.5 and E4.75 of embryonic development.

Many of the important structural changes that occur during implanta-

tion have been explained in terms of chemical signals within and between
embryonic and extraembryonic compartments [117, 149]. For instance, it
was shown that at the onset of implantation epiblast cells exit their naive
pluripotency state, self-organise into a highly polarised rosette, and initiate
lumenogenesis under the influence of 𝛽1-integrin signalling [9, 133]. Shortly
after implantation, 𝛽1-integrin enables pro-amniotic cavity formation along
the entire egg cylinder via the resolution of multiple rosettes both in extraem-
bryonic cell populations and at their interface with the embryonic tissue [24].
Moreover, differentiation of the primitive trophectoderm into polar and mu-
ral trophectoderm leading to the formation of a boundary between the two
tissues was traced back to fibroblast growth factors (FGFs) signalling [26].

As D’Arcy Thompson already noted about genetics, however, develop-

ment cannot be construed solely in terms of biochemical signals either: the
mechanical interactions between cells and tissues equally and reciprocally
contribute to embryogenesis [68, 143]. On the subject of the epiblast remod-
elling into a cup, a series of biological works have paved the way and triggered
further investigation into the mechanics involved. Because it was observed
that the EPI did not initiate specific tissue-level symmetry-breaking be-
haviours, one study stated that after the basement membrane disintegrated
between the EPI and TE, the membrane between the EPI and the PE acted
like a basket that moulded the epiblast into its cup shape [8] (Fig. 5.1A).
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 82

Although this hypothesis put the spotlight on the basement membrane, it

also suggested that the TE in direct contact with the EPI could play a
role in this shape change. Evidence supporting this hypothesis grew when
“ETS-embryoids” (ETS: embryonic and trophoblast stem-cell) assembled in
vitro from EPI and TE stem cells, surrounded by the extracellular matrix
(ECM) acting as the basement membrane, replicated embryonic transition
from blastocyst to egg cylinder [67] (Fig. 5.1B). Furthermore, a recent study
highlighted more clearly the role of the trophectoderm [155]. In this study,
ExE-embryoids (ExE: extra-embryonic ectoderm), cultured from EPI and
PE stem cells separated by an ECM basement membrane, did not break the
symmetry of their initial spherical shape (Fig. 5.1C). In contrast, both ETS-
and ETX-embryoids (ETX: embryonic, trophoblast and extra-embryonic en-
doderm) made from all three blastocyst lineages did reproduce the symmetry
breaking observed in real embryos. Together, these studies established the
necessity of the trophectoderm for the remodelling of the epiblast [67, 155].
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 83

Figure 5.1: Review of epiblast symmetry breaking theories.

A. The basement membrane separating the epiblast and the primitive en-
doderm moulds the epiblast into a cup while it disintegrates between the
epiblast and the trophectoderm in mouse embryos [8]. B. Embryoid struc-
tures featuring epiblast and trophectoderm stem cells surrounded by an
ECM acting as a basement membrane (ETS-embryoids) replicate mouse
embryogenesis by forming body structures similar to those observed in
normal embryonic development [67]. Here the presence of the trophecdo-
derm shows that this tissue might be required for symmetry breaking in
the epiblast and cup shape acquisition. C. Embryoid structures featuring
epiblast and primitive endoderm stem cells surrounded by an ECM acting
as a basement membrane (EXE-embryoids) do not break symmetry in the
epiblast, but initiate lumenogenesis [155]. This evidences the requirement
of the trophectoderm for the remodelling of the epiblast. D. Trophecto-
derm morphogenesis during mouse implantation. Trophectodermal cells
elongate, then undergo apical constriction, resulting in the tissue folding
and invaginating the epiblast [26]. This suggests that epiblast remodelling
into a cup might be a mechanical response to trophectoderm dynamics
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 84

On the other hand, how exactly trophectoderm morphogenesis influ-

ences shape change in the epiblast has not been elucidated yet because very
little is known on trophectoderm morphogenesis during implantation. In
the light of recent detailed descriptions of extra-embryonic tissue morpho-
genesis during implantation [26], it appears increasingly plausible that tro-
phectoderm morphogenesis regulated epiblast remodelling via mechanical
interactions at their common boundary. This study showed that polar tro-
phectodermal cells exhibited drastic morphological changes throughout the
implantation period. Whereas early implanting blastocysts featured squa-
mous cells in the polar trophectoderm, these cells, driven by a high mitotic
and space restrictions due to the formation of a boundary with the mural
trophectoderm, later transited to cuboidal, then elongated to acquire colum-
nar shapes. These changes were followed by apical constriction resulting in
the folding of the whole tissue, and invagination of the epiblast (Fig. 5.1D).
Moreover, this study provided experimental evidence that other structural
changes, most notably the stretching of PE (Primitive Endoderm) cells, re-
sulted from TE (Trophectoderm) morphogenesis [26]. Hence, we want to
investigate the hypothesis that trophectoderm morphogenesis drives the re-
modelling of the epiblast into a cup via mechanical interactions at their
common boundary.

Building on the increasing power of computational modelling in devel-

opmental biology [17, 35, 135, 147], we examine the influence of trophecto-
derm morphogenesis on the epiblast. The requirement of dramatic cell shape
changes in trophectodermal cells, notably apical constriction [26], orients
modelling options toward the family of deformable cell models (DCM) [146].
In this category, two classes of models have been predominant in recent
research: vertex models (VM) and sub-cellular element models (SEM).
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 85

Although vertex models were used extensively to study epithelial dynam-

ics [2, 52], accounting for various mechanical behaviours of individual cells
remains challenging in a global energy-based approach. Hence, we set our
choice on SEM, where cells are represented by an agglomeration of com-
putational particles interacting with one another via short-range potentials
emulating the viscoelastic properties of their cytoskeleton [98, 102, 130].
However, in order to exhibit realistic cell shapes, SEM generally involve an
important number of particles, many of which reside within the cell, thus do
not have a direct influence on cell shape. This leads to increased computa-
tional complexity, limiting the size of cell populations that can be simulated.

Here, using MG#, we first reproduce the experimental observation that

repulsion at the apical surface is sufficient for lumenogenesis in the epiblast.
Then, we reproduce trophectoderm morphogenesis during implantation and
we provide theoretical support that epiblast remodelling into a cup shape and
its movement towards the maternal uterine tissue can be explained by tro-
phectoderm morphogenesis. We also conduct a sensitivity analysis, where we
show how different sets of model parameters influence simulation outcomes.

5.2 Results

In this section, we applied our model to the study of mouse embryo morpho-
genesis during implantation. Here we focused on epiblast and trophectoderm
tissues. First, we tested the hypothesis of whether repulsion at the apical
surface of the epiblast was sufficient to account for lumenogenesis. Then, we
simulated both tissues’ morphogenesis and showed that the epiblast remod-
elling into a cup shape and its movement towards the maternal uterine tissue
could be explained by trophectoderm morphogenesis. Next, we conducted
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 86

a sensitivity analysis, to show how different sets of parameters influenced

simulation outcomes.

5.2.1 Repulsion at the apical surface of the epiblast is

sufficient for lumenogenesis

The study of how lumens arise in epithelial tissues has revealed two predomi-
nant mechanisms: cavitation mediated by apoptosis, and hollowing, in which
the lumen is formed by exocytosis and membrane separation [31, 152]. In
the case of highly polarised epithelia, it was shown that cavitation was not
necessary for lumenogenesis [94]. Hence, the hollowing mechanism was privi-
leged in epiblast lumenogenesis, which features highly polarised cells spatially
organised in the shape of a rosette. Moreover, it was hypothesised that repul-
sion mediated by anti-adhesive molecules such as podocalyxin (PCX) drove
lumen formation in the epiblast [8, 9, 133, 155]. Furthermore, evidence for
hollowing in the epiblast was observed in a recent study [155], where apop-
tosis was found not to regulate lumenogenesis, but PCX was discovered to
be predominant at the apical surface of cells facing the lumen.

Using our model, we sought to determine theoretically whether hollow-

ing via repulsion at the apical surface of the epiblast rosette was a viable
mechanism for lumenogenesis in this tissue. First, we built a 3D rosette-
shaped epiblast by submitting polarised epithelial cells to apical constric-
tion [9] (Fig. 5.2A,B, Supplementary Fig. 5.6A). Then, inspired by the anti-
adhesive role of PCX, we broke adhesive links between cell membranes in
contact at the apical surface of the rosette, meaning that certain neighbour-
ing pairs of particles were not more submitted to the exact same forces, but
rather could be repelled in different directions.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 87

Figure 5.2: Lumenogenesis in the epiblast. A. A 3D model of a

rosette-shaped epiblast. B. A 2D slice of the epiblast in A showing apically
constricted cells of the building block of the epiblast rosette. C. Creation
of the lumen cavity by repulsion at the apical surface of the epiblast. Green
arrows represent the direction of repulsive forces. The snapshots (from
left to right) were taken respectively at 𝑡 = 0, 500 and 2000. D. Lateral
view of the sliced epiblast showing the lumen volume. E. Dynamics in
time of the volume of the lumen. Values of the equation parameters:
𝐽EPI = 2.5, 𝜆med = 𝜆𝜒 = 2, 𝜌 = 1, 𝑅lum = 0.25.

We then created a virtual source (𝑂) at the centre of the lumen to exert
repulsive forces on apical particles. Bedzhov et al. argue that these repulsion
forces are driven by electrical charges [9], a possibility also explored in [43].
To model these effects, we used conservative forces from a Morse potential
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 88

(Eq. 5.1).
→
− rep
𝐹 𝑖 = 2𝐽EPI 𝜌 (𝑒2𝜌(𝑟 −𝑅lum ) − 𝑒𝜌(𝑟 −𝑅lum ) ) →
−
𝑂𝑖 𝑂𝑖
𝑢 𝑂𝑖 (5.1)

Here, 𝑅lum is the radius of the lumen. These forces prompted neighbour-
ing apical particles and surfaces to drift apart from each other, initiate the
creation of a lumen at the centre of the rosette (Fig. 5.2C-E). This result,
upholding experimental data, suggests that hollowing via apical repulsion is
sufficient to drive the onset of lumenogenesis in the mouse epiblast.

5.2.2 Mechanical constraints imposed by TE morpho-

genesis on the epiblast drive cup shape acquisition

A key feature of the blastocyst-to-egg-cylinder transition is the symmetry

breaking within the epiblast and its shaping into a cup [8, 9]. During this
transformation, the epiblast remodels from an oval ball to a tissue with a flat
surface at its boundary with the trophectoderm. Previous studies have estab-
lished the requirement of the trophectoderm in this shape change [67, 155].
Using the presented model, we investigated how trophectoderm morphogen-
esis influenced the cup shape acquisition by the epiblast. Our simulation
protocol consisted of reproducing the sequence of morphological events ob-
served in the trophectoderm as described in [26] (elongation followed folding
via apical constriction), and keeping track of the consequent changes in the
epiblast. For simplicity and to keep the model computationally efficient, we
assumed that there were no cell divisions in the tissue.

We built a virtual embryo consisting of a TE sheet with initial cuboidal

cells laying on top of an oval rosette-shaped epiblast (Supplementary
Fig. 5.6B). At the initial stage (Fig. 5.3A,E), new equilibrium lengths were
computed for all TE cells, with the goal of triggering a transition from
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 89

cuboidal cells to more elongated columnar shapes with smaller apical surface.
These cells lost their resting state and regained it by gradually aligning their
actual springs lengths with the calculated equilibrium lengths (Fig. 5.3B,F).
After that, we initiated invagination in the TE. Single cell mechanisms at
work are often activated in discriminatory ways both in space and time
[25, 109, 141]. In our simulations, the distribution over the entire sheet of
the length 𝑑 by which the apical radius of cells 𝑅 was shrunk depended on the
position of the cell in relation to the centre of the sheet via a step function:
cells in the middle of the sheet were set to constrict completely (𝑑 = 𝑅), while
cells on the boundary did not constrict (𝑑 = 0, Supplementary Fig. 5.7). The
coordinated movement of cells induced by these positional laws caused the
tissue to fold and invaginate the epiblast. Short after TE invagination begins,
we initiated lumenogenesis in the epiblast (Fig. 5.3G). In order to highlight
the requirement of the TE, following TE folding (Fig. 5.3C,G), we broke
the contacts between the TE and the epiblast for the remaining time of the
simulation, inhibiting any mechanical interactions between the two tissues,
but maintaining both tissues’ own mechanics (Fig. 5.3D,H). We noted that
throughout the experiment, with the exception of lumenogenesis, epiblast
cells did not initiate any behaviours, the epiblast as a whole simply reacted
to the mechanics induced by either the presence or the absence of the TE.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 90

Figure 5.3: Trophectoderm morphogenesis regulates epiblast

shape. A-D. 3D snapshots of the simulation of TE and EPI morphogen-
esis during mouse implantation, and the regulation of EPI shape, taken
respectively at 𝑡 = 0, 3000, 6000 and 9000. E-H. Corresponding 2D slices
of the cell population at the same stages. (A,E). The initial stage features
a single layered TE with cuboidal cells resting upon the rosette-shaped epi-
blast. (B,F). TE cells have transited to a columnar shape. (C,G). The
TE has folded by apical constriction of single cells. Concomitantly, lu-
menogenesis was initiated in the epiblast (the process starts at 𝑡 = 4000).
(D,H). After adhesive links were broken between TE and EPI, the EPI
bounces back to its near spherical shape. I. Definitions of the metrics used
to evaluate the model, involving the curvature 𝜃, TE/EPI interface diame-
ter 𝐷, TE/EPI interface length 𝐿, and interface ratio 𝐿/𝐷. J. Plot of the
population’s elastic energy 𝐸. Discontinuities mark the start of new mor-
phological events at 𝑡 = 0, 3000, 4000, and 6000). After removal of the TE,
𝐸 falls closer to zero than ever before, meaning that cells are closer to their
resting stage, hence less externally constrained. K. Plot of the interface
curvature 𝜃. During TE morphogenesis, 𝜃 rises towards a flat angle, then
sharply drops when the TE is removed. L. Plot of the interface ratio 𝐿/𝐷.
During TE morphogenesis, the interface curvature decreases towards 1,
then sharply increases when the TE is removed. Values of the equation
parameters: 𝐽EPI = 𝐽TE = 2.5, 𝜆med = 𝜆 = 2, 𝜌 = 1, 𝑅lum = 0.25.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 91

To appreciate the impact of the TE on the epiblast, we used the elastic

energy 𝐸𝑖 of a cell 𝑖 as the sum over all cell springs of the squared difference
between equilibrium and actual lengths. We extended this notion by defining
the total elastic energy of a tissue or an entire population of cells as the sum
of 𝐸𝑖 ’s in the population (eq. 5.2).
(︃ )︃
∑︁ ∑︁
𝑘𝑠
𝐸= (𝑟eq − 𝑟𝑘𝑠 )2 (5.2)
𝑘≤𝑁 𝑠≤𝑁𝑘

Here, 𝑁 is the number of cells in the population and 𝑁𝑘 the number of

springs in cell 𝑘.

Cells always tended to minimise this energy, which can also be viewed
as the degree of relaxation of cell: the closer it is to zero, the closer the
cell is in its resting state, the more relaxed it is, hence the less constrained.
In addition, we monitored the curvature of the epiblast, i.e. the inclination
angle 𝜃 of the epiblast surface covered by the trophectoderm (Fig. 5.3I). An
increasing curvature, trending towards a flat surface, was characteristic of the
epiblast’s transition from an oval rosette to a cup. Moreover, we measured
the length (𝐿) and diameter (𝐷) of the interface between EPI and TE, and
considered their interface ratio (𝐼𝑟 = 𝐿/𝐷) as our third evaluation metric
(Fig. 5.3I). It was expected that this ratio would decrease towards 1 as the
epiblast flattened. We plotted the profiles of the curvature, the interface
ratio and the elastic energy throughout our simulation.

Our model matched biological expectations by replicating, on the

one hand, an increasing curvature and a decreasing interface ratio, with
ultimately a flat TE/EPI interface just before we removed the TE
(Fig. 5.3C,G,K,L). On the other hand, as soon as the TE was removed, the
epiblast bounced back to its original shape (Fig. 5.3D,H,K,L). This result
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 92

agrees with the experimental observation that without the TE, the epiblast
does not break symmetry [155]. The elastic energy profiles tie these be-
haviours to the mechanical influence of the TE over the epiblast. Actually,
breaking mechanical interactions between the TE and the EPI not only re-
sulted in a sharp drop in elastic energy, but this energy also plateaued at a
value significantly lower than in other stages (Fig. 5.3J), demonstrating that
cells were more mechanically constrained when both tissues were in contact.

These observations suggest that the presence of the TE imposes me-

chanical stress on epiblast cells, hinting to the necessity of this tissue’s mor-
phogenesis in the remodelling of the epiblast.

5.2.3 Trophectoderm morphogenesis fosters epiblast

movement towards the uterine tissue

An important requirement of implantation is close contact between the em-

bryo and the uterine tissue. As soon as the three pre-implantation lineages
are specified, the blastocyst hatches out of the zona pellucida and initiates
the process of implantation [8]. However, there exists a gap between the
hatched blastocyst and attachment sites in the uterus. In order to close
this gap, the embryo needs to move towards the uterus. It was recently
established that this movement of the embryo towards maternal sites occur
concomitantly to the drastic morphological changes observed in the TE [26].
Furthermore, it was observed in that same study that primitive endoderm
expansion over the whole embryo is driven by TE morphogenesis. Given that
the trophectoderm keeps close contact with the epiblast during these events,
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 93

we hypothesised that epiblast positioning could also be affected by TE mor-

phogenesis. We employed computational modelling to examine whether TE
morphological changes could influence the trajectory of the epiblast.

Here, as previously, we reproduced the sequence of TE morphogenesis

(elongation followed by folding via apical constriction), and observed how it
affected the position of the epiblast (which also undergoes lumenogenesis).
To highlight how the TE influences the trajectory of the epiblast, we defined
what we designated as the “pushing distance”. We computed this distance at
any given time point of the simulation by calculating the difference in height
between the lowest point of the epiblast at that time point and the lowest
point at the initial stage (Fig. 5.4A). We plotted the profiles of this metric and
observed an increasing pushing distance as the TE transited from cuboidal
to columnar, then as the TE folded (Fig. 5.4B). The sudden soar observed at
𝑡 = 4000 reflects the slight elongation of the tissue due to hollowing-driven
lumenogenesis in the epiblast.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 94

Figure 5.4: Trophectoderm fosters epiblast movement towards

maternal sites. A. Snapshots of the simulation of TE and EPI morpho-
genesis during mouse implantation, and their influence on EPI positioning,
taken respectively at 𝑡 = 0 and 6000. B. Plot of the pushing distance,
which increases with time. C. Plot of the elastic energy 𝐸. Discontinuities
mark the start of new morphological events (𝑡 = 0 and 3000). The sudden
soar observed at 𝑡 = 4000 reflects the slight elongation of the tissue due
to hollowing-driven lumenogenesis in the epiblast. D. Plot of the pushing
distance on the epiblast Centre of Mass (CoM), which also increases with
time. E. Plot of the pushing distance on the cell population Centre of
Mass (CoM), which also increases with time. Values of the equation pa-
rameters: 𝐽EPI = 𝐽TE = 2.5, 𝜆med = 𝜆𝜒 = 2, 𝜌 = 1, ,d=0.5, 𝑅lum = 0.25.

We chose to monitor the lower end of the epiblast because it is via this
pole that the epiblast attaches to maternal sties. However, To ensure that
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 95

the observed changes did not merely represent an elongation of the epiblast,
we also tracked the trajectory of the Centre of Mass (CoM) of both the
epiblast (Fig. 5.4D) and the entire cell population (Fig. 5.4E). Similarly,
these metrics reaffirmed that the epiblast indeed engages in a downwards
movement. Furthermore, we checked that lumenogenesis in the epiblast was
not necessary to foster this motion (Fig. 5.8). These results suggest that
TE morphogenesis, while reshaping the epiblast, also fosters the embryo’s
movement towards maternal sites.

5.2.4 Sensitivity analysis

Physical properties are generally a segregating factor between differentiated

cells in development [14, 120]. Although the mouse trophectoderm and epi-
blast form distinct cell lineages, we have so far assumed similar characteris-
tics for both types of cells. The nature of our cell model allows for global
physical properties such as mechanical stiffness to emerge from lower scale
interactions between subcellular elements. In order to characterise cells by
their stiffness and thus differentiate trophectoderm and epiblast cells, we first
needed to establish how this property depended on intrinsic model parame-
ters. In the following case study, we set out to determine how cell stiffness
relates to parameter 𝐽, the scaling factor of the elastic force between neigh-
bouring particles.

We used an “in Silico” adaptation of the experimental protocol described

in [97] to estimate cell stiffness based on the computation of a measure
of their elasticity modulus (also known as Young modulus). For a given
value of 𝐽, we perform a series of simulations consisting of applying forces
of increasing magnitudes (𝐹 ) on the apical and basal faces of an epithelial
𝐹
cell (Fig. 5.5A). For each force, we calculate the associated stress (𝜎 = 𝑆
,
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 96

where 𝑆 is the surface area of each face) and note the resulting deformation
Δ𝐿
(strain, 𝜖 = 𝐿0
). We then plotted the stress-strain curve and estimated the
Young modulus (𝑌 ) as the slope of the curve using a linear regression model
(fig. 5.5B). Using this protocol, we ran simulations with 50 different values
of 𝐽 uniformly distributed between 0 and 5, recording estimated values of
𝑌 after every simulation. The plot in (fig. 5.5C) suggests that 𝑌 relates to
𝐽 in a measurable way. More broadly, 𝑌 increases with 𝐽. In other words,
the interaction strength between subcellular particles regulates cells global
stiffness: the stronger this interaction is, the stiffer the cell.

We conducted the same analysis on how friction forces coefficients 𝜆𝜒

and 𝜆med affect cell mechanical properties. We fixed 𝜆med (𝜆med = 2), var-
ied 𝜆𝜒 , and observed the evolution of cells elasticity modulus as a function
𝜆𝜒
of 𝜆med
. Simulated show that above a certain threshold ( 𝜆𝜆med
𝜒
≥ 0.161),
cells elasticity modulus was constant (fig. 5.9A,B). Below this threshold,
the structure of the cell was compromised (fig. 5.9C). Overall, these observa-
tions suggested that cell mechanical properties did not depend on differences
between friction parameters within and without the cell. Furthermore, we
refined the cell mesh, taking the number of vertices to 42 (fig. 5.9D), and
repeated the experiments, varying values of parameter 𝐽 (fig. 5.9E,F). Re-
sults show that mechanical properties changed with mesh refinement (For
𝐽42 = 2.5 → 𝑌42 = 2.75). However, while refining the mesh, parameters can
be tuned in order maintain cell stiffness (𝐽42 = 2.6 → 𝑌42 = 2.90) to allow
similar responses to external stress.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 97

Figure 5.5: Mechanical properties of EPI and TE determine

mouse implantation. A. In“Silico” experimental protocol used to de-
termine cells elastic modulus. B. Stress-Strain curve (black ) for a single
epithelial cell (34 vertices) with 𝐽 = 2.5. (blue) Linear approximation of
the Stress-Strain curve. The elastic modulus of the cell is determined by
the slope of this line (𝑌 = 2.92). C. Plot of the Elastic (Young) modulus
of cells as a function of parameter 𝐽, the interaction strength between sub-
cellular particles. D,E,F. Respective Plots of the Interface curvature, the
Interface ratio and the Pushing Distance as functions of the mechanical
stiffness of TE cells (determined by 𝐽TE as in C). G. Plot of the fitness
metric as functions of the mechanical stiffness of TE cells (determined by
𝐽TE as in C). H. Snapshots of the epiblast shape at the end of simula-
tions for different values of 𝐽TE . With equal stiffness (middle, 𝐽TE = 2.5,
𝐽EPI = 2.5), trophectoderm morphogenesis flatten the epiblast, which
acquires a cup shape. However, with significantly lower stiffness (left,
𝐽TE = 0.3, 𝐽EPI = 2.5), trophectoderm morphogenesis barely reshape the
epiblast; meanwhile, with considerably higher stiffness (right, 𝐽TE = 4.9,
𝐽EPI = 2.5), the trophectoderm invaginates the epiblast, forcing a concave
interface with the epiblast. Other parameter values, 𝜆med = 𝜆𝜒 = 2, 𝜌 = 1,
𝑑 = 0.5.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 98

Having established how model parameters regulate cell stiffness, we were

able to discriminate between cell types based on parameter values we set for
each. We then sought to investigate how differences between physical proper-
ties of trophectoderm and epiblast cells would influence mouse implantation.
For this, we conducted a parameter space exploration in the one dimensional
space of values of parameter 𝐽TE , maintaining the value of 𝐽EPI constant to
a value of 2.5. This series of experiments consisted of running 50 different
simulations of mouse implantation, with values of 𝐽TE ranging from 0 to 5
with a step of 0.1. To better appreciate the impact of the trophectoderm
on the epiblast, we do not trigger lumenogenesis in the epiblast. For every
simulation, we recorded the curvature, interface ratio and pushing distance
as defined in previous section, and plotted their values against values of 𝐽TE
(fig. 5.5D, E, F). In order to determine which values of 𝐽TE perform best
overall for these metrics, we defined a normalised fitness measure consisting
of a combination of these metrics as previously done in [35]. If we denote by
𝜃(𝐽TE ), 𝐼𝑟(𝐽TE ) and 𝐻(𝐽TE ) the respective values of the curvature, interface
ratio and pushing distance for a given value of 𝐽TE , and 𝜃min,max , 𝐼𝑟min,max
𝐻min,max their optimal values in the simulated data, the fitness metric (𝑀 )
is defined by equation (5.3).

(︃(︂ )︂2 (︂ )︂2 (︂ )︂2 )︃

1 𝜃(𝐽TE ) − 𝜃max 𝐼𝑟(𝐽TE ) − 𝐼𝑟min 𝐻(𝐽TE ) − 𝐻max
𝑀 (𝐽TE ) = + +
3 𝜃max − 𝜃min 𝐼𝑟max − 𝐼𝑟min 𝐻max − 𝐻min
(5.3)

It can be observed that function 𝑀 admits a minimum and its values

are constrained in [0, 1]. We plotted this metric against values of 𝐽TE and
considered that areas where the fitness fell below 0.1 represented simulations
featuring a good compromise between curvature, interface ratio and pushing
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 99

distance (fig. 5.5G, green points). The plotted data hint the existence of
a preferential range of values that yield optimal fitness with respect to the
three metrics involved (fig. 5.5G green points, fig. 5.5H middle). Within this
range, the strength of subcellular interactions is always always higher for
trophectoderm cells (𝐽TE ∈ [2.5, 3.5]) than for epiblast cells (𝐽TE = 2.5). As-
suming that cell stiffness remain constant through implantation, this result
suggest that mouse implantation requires trophectoderm cells to be generally
stiffer than epiblast cells. However, outside of this range, simulations appear
to perform poorly. For instance, below this range i.e. with TE cells more
ductile than EPI cells, the epiblast is not sufficiently remodelled into a cup
(fig. 5.5H, left), as attested by moderate performances of the interface cur-
vature and ratio (fig. 5.5D,E)). Above this range i.e. simulations featuring
TE cells significantly more rigid than EPI cells, the trophectoderm consider-
ably invaginates the epiblast, creating a concave interface ((fig. 5.5H, right)).
This reflects poorly on the pushing distance as highlighted by the negative
slope of its curve (fig. 5.5F)).

5.3 Discussion and Conclusion

Understanding the processes by which the mammalian embryo implants in

the maternal uterus is crucial to many breakthroughs in embryology [149].
New insights into these morphogenesis events could be of great importance in
helping for example to reduce human infertility [56]. Although advances have
been made by studying biochemical cues involved in these events, we focused
here on the mechanical basis at the cellular level of epiblast morphogenesis.
In order to study the physical dynamics of mouse implantation, we have
designed a novel, computationally efficient model of biological cells and tissue
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 100

mechanics able to simulate key episodes of vertebrate morphogenesis. With

this model, we were able to schematically reproduce lumenogenesis in the
epiblast, trophectoderm morphogenesis driven by single cells elongation and
apical constriction, as well as provide theoretical support to the fact that
this morphogenesis regulates the remodelling and positioning of the epiblast
during implantation.

A well-known shortcoming of agent-based modelling is the risk to intro-

duce disputable artefacts in the simulations. Within the scope of this work,
we have shown that our model adhered well to biology by successfully simu-
lating tissue-level morphological changes based solely on changes triggered at
the cellular level, in a bottom-up, emergent fashion. We did this in particular
for epithelial bending through apical constriction [131], rosette formation via
polarised apical constriction [65], and repulsion-driven lumenogenesis [8, 9].
Nonetheless, some nuance should be added to certain quantitative features of
the simulations. For instance, although it is a biological fact that the epiblast
lumen’s volume increases as a result of cells drifting apart, the rate of this
growth as exhibited in the graph of Fig. 5.2E may not reflect the actual rate
curve in mouse embryos. The same could be said of the rate at which the epi-
blast reshapes (Fig. 5.3K,L), or the trophectoderm-induced epiblast velocity
in its motion towards maternal sites (Fig. 5.4B). While not invalidating our
main conclusions, these quantitative outputs are essentially contingent upon
the choice of the potential function (here the Morse potential) and parameter
values. This limitation could be overcome by experimenting with other po-
tential functions, searching parameter space, and comparing results against
real biological data.

Another weakness of computational modelling is its inability to inte-

grate all possible details of a real-world problem, as this would inevitably
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 101

increase complexity and demand unavailable computing power. Clearly, effi-

ciency in our simulations was achieved by stripping the model of noticeable
features of biological development. One important approximation is that
we ignored the hypothetical impact of proliferation, although it is a per-
vasive phenomenon in both tissues. However, while it may be argued that
proliferation plays a non-trivial role in the elongation of trophectodermal
cells [26], it is difficult to make a case that proliferation would be central
in reshaping the epiblast, or the folding of the trophectoderm. In fact, this
particular lack in our approach could even be considered an advantage, since
neglecting proliferation also allowed isolating, hence highlighting the effects
of pure mechanical interactions within and between the trophectoderm and
the epiblast. Another simplification is that we neglected stochastic effects
related for example to cell movements during these embryogenesis episodes.
Although it was shown that cell membrane blebbing influenced cell differ-
entiation [153], in many epithelial settings, stochastic effects are often com-
pensated by strong interactions between cells [54]. Furthermore, in general,
deterministic models, still exhibit good predictive power while remaining
computationally practical [78].

In summary, although relatively abstract and schematic, our computa-

tional model and simulations offer new insights into mouse embryo implanta-
tion. Looking forward, refinements could combine the effects of mechanical
interactions with proliferation and the stochasticity of biological processes to
further investigate tissue shape changes. Then, the variables and parameters
in these simulations could be tuned to fit quantitative metrics based on real
measurements gathered from implanting embryos.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 102

Figure 5.6: Epiblast and trophectoderm population reconstruc-

tion. A. The rosette-shaped EPI tissue is built by submitting polarised
cells in a double epithelial layer to apical constriction. Green arrows in-
dicate the apical surface of the cells, where constriction occurs. B. The
initial cell population (TE and EPI) is built by adding an epithelial layer
to the forming the EPI.

Figure 5.7: Top view of trophectoderm morphogenesis. A. Initial

stage with cuboidal cells. B. Columnar TE initiating apical constriction.
Red arrows highlight cells undergoing apical constriction. In this case,
only cells in the middle constrict (light blue) to enable folding. C. Folded
TE. D. Folded TE after separation from the EPI.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 103

Figure 5.8: Trophectoderm fosters epiblast movement towards

maternal sites (Without lumenogenesis in epiblast). A. Snapshots
of the simulation of TE and EPI morphogenesis during mouse implanta-
tion, and their influence on EPI positioning, taken respectively at 𝑡 = 0
and 6000. B. Plot of the pushing distance, which increases with time.
C. Plot of the elastic energy 𝐸. Discontinuities mark the start of new
morphological events (𝑡 = 0 and 3000). D. Plot of the pushing distance
on the epiblast Centre of Mass (CoM), which also increases with time.
E. Plot of the pushing distance on the cell population Centre of Mass
(CoM), which also increases with time. Values of the equation parame-
ters: 𝐽EPI = 𝐽TE = 2.5, 𝜆0 = 𝜆𝜒 = 2, 𝜌 = 1.
Computational modelling unveils how epiblast remodelling and positioning
rely on trophectoderm morphogenesis during mouse implantation 104

Figure 5.9: Sensitivity analysis (Supplementary). A. Stress-Strain

curve (black ) for a single epithelial cell (34 vertices) with 𝐽 = 2.5 and
𝜆med = 𝜆𝜒 = 2. (blue) Linear approximation of the Stress-Strain curve.
The elastic modulus of the cell is determined by the slope of this line
(𝑌 = 2.92). B. Plot of the Elastic (Young) modulus of cells as a function
𝜆𝜒
of the parameter ratio ( 𝜆med ). Young’s modulus is defined and constant for
𝜆 𝜒
values of 𝜆med greater or equal to approximately 0.161. Below this value,
simulated cells do not behave as physical materials, and the elasticity
modulus cannot be defined as illustrated in the next plot. C. Stress-
Strain curve (black ) for a single epithelial cell (34 vertices) with 𝐽 = 2.5,
𝜆med = 2 and 𝜆𝜒 = 0.25. The discontinuity in the curve shows that the set
of parameters is not suitable for a cell. D. 3D rendering of an epithelial
cell with square basis and 42 vertices. E. Stress-Strain curve (black ) for a
single epithelial cell (42 vertices) with 𝐽 = 2.5 and 𝜆med = 𝜆𝜒 = 2. (blue)
Linear approximation of the Stress-Strain curve. The elastic modulus of
the cell is determined by the slope of this line (𝑌 = 2.75). F. Plot of
the Elastic (Young) modulus of a cell (42 vertices) as a function of the
parameter 𝐽, the interaction strength between subcellular particles. In
order for such a cell (42 vertices) to have equivalent stiffness with the
previous type of cell (34 vertices, 𝐽34 = 2.5, 𝑌34 = 2.92), the parameter
𝐽42 needs to be set to approximately 2.6 (𝑌42 = 2.90).
Chapter 6

Quantification of cell behaviors

and computational modelling
show that cell directional
behaviors drive zebrafish pectoral
fin morphogenesis

Understanding the mechanisms by which the zebrafish pectoral fin develops

is expected to produce insights on how vertebrate limbs grow from a 2D cell
layer to a 3D structure. Two mechanisms have been proposed to drive limb
morphogenesis in tetrapods: a growth-based morphogenesis with a higher
proliferation rate at the distal tip of the limb bud than at the proximal side,
and directed cell behaviors that include elongation, division and migration in
a nonrandom manner. Based on quantitative experimental biological data
at the level of individual cells in the whole developing organ, we test the

105
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 106

conditions for the dynamics of pectoral fin early morphogenesis. We found

that during the development of the zebrafish pectoral fin, cells have a pref-
erential elongation axis that gradually aligns along the proximodistal axis
(PD) of the organ. Based on these quantitative observations, we build a
center-based cell model enhanced with a polarity term and cell proliferation
to simulate fin growth. Our simulations resulted in 3D fins similar in shape to
the observed ones, suggesting that the existence of a preferential axis of cell
polarization is essential to drive fin morphogenesis in zebrafish, as observed
in the development of limbs in the mouse, but distal tip-based expansion is
not.

6.1 Introduction

Vertebrate limb development is a classical model system for understanding

pattern formation: the process in which spatial organization of differentiated
cells and tissues is generated in the embryo. Various tissue types contribut-
ing to the mature limb are derived from several embryonic tissues including
the lateral plate mesoderm, the somites and the ectoderm [16, 21, 23, 62, 73].
How the pectoral fin lateral plate mesoderm (LPM), which gives rise to skele-
tal elements and tendons, grows outward from the body trunk and acquires
its particular shape remains unclear. The in vivo observation of the whole
process and the quantification of cell behaviors underlying morphogenesis are
still open challenges. The formation of the zebrafish pectoral fin is a model
for limb development, as it is especially suited for long-term imaging owing
to its external embryonic development and translucent body. The formation
of the pectoral fin initiates at 18 hours post fertilization (hpf), when LPM
cells condense at the prospective fin location as a flat 2D cell layer under a
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 107

single layer of ectodermal cells. Over the course of the next 30 hours, the fin
bud grows and forms a 3D structure. LPM cells proliferate and myoblasts
coming from neighbouring somites enter the fin, where they will give rise to
muscle bundles. At the distal tip of the fin, ectodermal cells align to form
the so-called apical ectodermal ridge (AER) known to act as a source of
molecular signaling required for the fin growth.

Over the past 40 years, the “proliferation gradient” model has been the
dominant hypothesis to explain the limb bud elongation. This model sug-
gests that a diffusible signal from the AER sets up a spatial concentration
gradient. This molecule “signals the mesenchyme immediately underlying it,
termed the ‘progress’ or ‘proliferative’ zone, to proliferate, resulting in di-
rected proximodistal outgrowth” [105]. The AER indeed was shown to have
mitogenic properties [126]. On the other hand, a few studies hypothesized
that directionally oriented cell behaviors drive limb elongation. Li et al. [86]
demonstrated that some mesenchymal cells in the chick wing bud are capable
of migrating toward an ectopic source of Fgf4 (one of the molecules produced
by the AER) implanted in the center of the bud. This work directly sup-
ported the idea that mesenchymal cells could consider the Fgf gradient as a
chemoattractant rather than as a mitogen.

A number of computer simulations of limb bud elongation have been

produced over the last decade, most of which are in two dimensions and
based on the proliferation gradient hypothesis but do not incorporate real
quantitative data into their model [38, 100, 121]. Recently, Boehm et al. [13]
proposed a new three-dimensional computer model based on the actual shape
of the mouse limb bud captured by optical projection tomography imaging.
They found that the proliferation gradient would have to be extreme from the
distal end to the proximal end and subsequently too unrealistic to account
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 108

for the bud shape. Similarly, Gros et al. [63] concluded that uniform cell
division distribution and focal regions of cell death in the chick limb bud are
unlikely to be sufficient to drive the anisotropic nature of its growth. Both
studies observed that cell shapes were oriented toward the nearest ectoderm,
rather than distally toward the apical ectodermal ridge. They have also
demonstrated the presence of filopodia suggesting active cell movement.

With improvements in in vivo imaging setups and data processing, a re-

cently available complete 3D tracking of the different cell types in the early
fin bud reveals their heterogeneous and complex cellular rearrangement dur-
ing the transition from 2D to 3D. The quantitative analysis of cell behaviors
during the first 20 hours of the zebrafish pectoral fin formation are used to
build a model of fin growth. The comparison between simulated and bio-
logical data highlights the pattern of cellular rearrangement underlying the
pectoral fin morphogenesis.

6.2 Methods

6.2.1 Data acquisition workflow

For this study, 4D imaging data of developing zebrafish pectoral fins were
acquired and pre-processed by our partners at the CNRS/Bioemergences lab
1
. Here we give a brief overview of the data acquisition workflow.
1
http://bioemergences.iscpif.fr/bioemergences/index.php
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 109

Figure 6.1: Geometry of the pectoral fin based on live imaging

and image processing data. (a) 3D rendering of raw data nuclear
staining at 𝑡 = 47.7 hpf: dorsal view of the zebrafish body with detection of
approximate nuclear centers of the pectoral fin cells highlighted by colored
dots, where the color code depends on the cell type; scale bar: 20 𝜇m.
Imaging data and individual cell tracking have been generated by the
CNRS/Bioemerges lab. (b-d) After applying cell detection methods: 3D
rendering of the approximate nucleus centers of LPM cells in the pectoral
fin at different stages of development, respectively 𝑡 = 28 hpf, 𝑡 = 37.9 hpf
and 𝑡 = 47.7 hpf (AP: anteroposterior axis; DV: dorsoventral axis). (e-
g) 3D rendering of the pectoral fin at the same times along the AP axis
and PD (proximodistal) axis. (h-j) Evolution over time of the fin size
in 𝜇m along the PD, AP and DV axes respectively. Fin expansion occurs
mainly along PD. It undergoes a slight compaction along the other two
axes, more pronounced along the DV axis.
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 110

6.2.1.1 Zebrafish husbandry

Adult zebrafish (Danio rerio) were maintained at 28∘ C according to stan-

dard procedures as described in [80]. Embryos were kept at 0.3% Danieau’s
medium at 28.5∘ C. The transgenic line Tg(X1a.Eef11:H2B-mCherry) was
used to visualize nuclei [124].

6.2.1.2 Imaging pectoral fin growth

Fin growth was monitored using the protocol described in [103]. Briefly, this
was achieved on the upright Zeiss LSM780 confocal microscope equipped
with 20x water dipping lens objective (Zeiss Objective W “Plan-Apochromat”
20x/1.0 DIC). The embryos were kept at 28.5∘ C and the chorion was re-
moved using forceps prior imaging. Embryos were anaesthetized with 0.04%
of tricaine methanesulfonate (Sigma) in embryo medium and mounted us-
ing agarose molds. The mounting strategy immobilized the embryo while
allowing pectoral fin to develop unperturbed. The imaging plane was paral-
lel to the plane formed by the anteroposterior (AP) and dorsoventral (DV)
axes (Fig. 6.1a). Images of the developmental process were taken at a time
invertal of 2 minutes and 18 seconds.

6.2.1.3 Image processing and reconstruction

The first analysis steps were performed using Fiji, an open source Java-based
image processing (ImageJ ) package. Raw datasets needed to undergo regis-
tration to keep the region of interest at about the same XYZ location, despite
the embryos undergoing significant morphological changes during overnight
imaging. To compensate for possible photo-bleaching, bleach correction by
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 111

histogram matching was also performed. Finally, images were uploaded into
the Bioemergences workflow [48], an online software platform integrating
original mathematical methods and algorithms to perform image filtering,
nucleus center detection, and cell tracking (Fig. 6.1b-g). The outcome of cell
detection was then manually validated using Mov-IT, an interactive visual-
ization and editing tool complementing Bioemergences.

6.2.2 Tracking zebrafish pectoral fin growth along PD,

AP and DV axes

The quantitative analysis required for this work necessitated that we track
the fin’s dynamics along its main axes: proximal-distal (PD), anterior-
posterior (AP), dorsal-ventral (DV). Precise tracking of the fin’s size along
these axes is a challenge due to the fact that the embryo body moves during
imaging, most likely due to growth. Embryo movements mean that its main
axes do not always maintain a static orientation. This in turn poses the
problem of precise identification of the fin’s main axis for every time point.
To tackle this issue, at each time point, we computed new orientations for
the PD, AP, DV axes as functions of both the cloud of points (cell centers)
at the current time and their orientations at the previous time step. First,
we applied the Principal Component Analysis (PCA) algorithm to the cell
centers in order to determine the three main directions of the point cloud.
Then, we compared each of PCA output axes with the PD axis at the pre-
vious time step and kept the most parallel one as our new PD axis. Given
that PCA does not keep track of the orientation at previous time steps, the
basis formed by the two remaining PCA axes could be significantly rotated
compared to the previous AP-DV basis. To correct this issue, we projected
onto the plane formed by these two vectors the AP and DV axes computed
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 112

at the previous time steps, and considered these projections to be our new
AP and DV axes. When the movement of the fin as a whole between two
time steps is not significant enough to change the orientation of the axes,
we keep the orientations computed in the previous time step. Here, notic-
ing that the fin as whole, although still growing, is relatively stable as from
about 43.3hpf, we stopped computing new axes orientations as from about
that time point.

6.2.3 Computational model

In order to test our data-derived hypothesis, we turned to computational

modelling. Although MG# was a good candidate model to perform this
study, other factors made it more appealing to look for different models.
As reviewed in chapter 2, when it comes to spatially explicit simulations
of biological development, several computational approaches stand out. De-
pending on the biological realism they include, the spatiotemporal scale they
capture, or the nature of variables they manipulate, models exhibit different
properties and suit different purposes. In this case, the interest of our study
did not lie in lower scale cell changes, but rather in the tissue morphogene-
sis emerging from their collective dynamics. Moreover, our datasets featured
cells represented solely be their centre. Hence, we found it suitable to set our
choice on the family of centre-based models (CBM). In this way, not only did
we gain in computational efficient in comparison to deformable cell models
(MG#), but this choice also meant that we could simulate cells using the
initial states provided from the dataset, and use same metrics to compare
our results against the biological data.
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 113

Figure 6.2: Center-based computational model of multicellular

dynamics. (a) Schema of a local cell neighborhood and the abstract forces
→
−
on cell centers. 𝐹 AR𝑗𝑖 is the passive attraction-repulsion force exerted on
→
−
a cell 𝑖 by a cell 𝑗. 𝐹 Pol
𝑖 is the active migration force driven by the cell’s
polarity (specified in Section 6.3.4). (b) Plot of the Morse force profile
→
−
(derivative of the Morse potential) defining 𝐹 AR , for different parameter
values. This curve presents two regimes: a positive regime (attraction)
below an equilibrium distance 𝑟eq and a negative regime (repulsion) above.

6.2.3.1 Model description

In CBM, cells are described by simple geometrical shapes whose represen-

tation can be reduced to their centers [146]. These models assume that cell
trajectories in space can be assimilated to the motion of particles, which are
governed by an equation of motion. CBMs have been used extensively to
study the development of multiple organisms [35, 57, 146, 147]. We used a
simplified mechanistic formulation of a CBM approach to simulate pectoral
fin growth (Fig. 6.2a).
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 114

Cells are subjected in their center to forces governing their behavior.

Similar to Julien Delile’s MecaGen model [35], here we distinguished two
main types of forces acting on a cell. On the one hand, passive attraction-
→
−
repulsion (AR) forces 𝐹 AR regulate interactions between cells and their local
neighborhoods. AR forces translate the biophysical property that individ-
ual cells occupy a certain volume in space, hence their centres cannot be
indefinitely close or indefinitely apart from each other. In the literature,
these forces often derive from elastic potential mimicking linear or non-linear
springs [7, 42]. Here, we followed this rule and derived AR forces from a
Morse potential, a curve exhibiting a quadratic minimum framed by vertical
and horizontal asymptotes (see its derivative in Fig. 6.2b). Moreover, we
also defined a neighborhood for each cell by computing the 3D Delaunay
tetrahedralization of the system. Two cells are deemed neighbors if their
centers belong to the same tetrahedron.

On the other hand, cells also exhibit an active migration force which
shapes their motion. Informed by empirical evidence, we constructed a
→
−
polarity-driven migration force 𝐹 Pol governing cells’ intrinsic mechanics (ex-
plained in Results, Section 6.3.4). Finally, we neglected the effects of inertia
due to a low Reynolds number [35, 106], and only considered viscosity-driven
friction via a constant coefficient 𝜆. Altogether, the equation of motion for
a cell 𝑖 with neighborhood 𝒩𝑖 reads:
(︃ )︃
∑︁ →
− AR →
−
𝜆→
−
𝑣𝑖= 𝐹 𝑗𝑖 + 𝐹 Pol
𝑖 (6.1)
𝑗∈𝒩𝑖

Here, →
−
𝑣 𝑖 is the velocity of cell 𝑖.

In CBM, cell divisions are often simulated by adding daughter cells

to the system at appropriate positions, notably in the neighborhood of the
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 115

mother cell. Here, we assumed that cells divide along their long axis and
added new cells such that mother and daughter cells were aligned along this
axis. Furthermore, we dealt with cell cycles by setting a global cycle period
for all cells, and assigning random initial phases to individual cells. Using
this model, we simulated zebrafish pectoral fin morphogenesis starting from
the initial cell arrangement provided by the imaging data.

6.2.3.2 Computational implementation

The computational implementation of the described model requires a time

→
− →
− (︁→
− )︁𝑡
discretisation of the equation of motion. Given 𝑋 𝑡𝑖 , 𝜆 and 𝐹 𝑡𝑖 = 𝐹 AR 𝑖 +
(︁→
− Pol )︁𝑡
𝐹𝑖 , respectively the position, friction coefficient and total force applied
on the centre of cell 𝑖 at time point 𝑡, an explicit Euler scheme is used to
→
−
calculate the position 𝑋 𝑡+1
𝑖 of the cell at the next time point. If 𝑑𝑡 is the
time laps between instants 𝑡 and (𝑡 + 1), cells update their configuration
based on the following rule (Eq. 6.2).

→
− 𝑡+1 →
− →
−
𝑋 𝑖 = 𝑋 𝑡𝑖 + 𝑉 𝑡𝑖 × 𝑑𝑡 (6.2)

→
−
In Equation (6.2), 𝑉 𝑡𝑖 is nor given, neither determined. An expression
→
−
of 𝑉 𝑡𝑖 can however be drawn from the equation of motion (Eq. 6.3).

→
− →
−
𝜆 𝑉 𝑡𝑖 = 𝐹 𝑡𝑖 (6.3)

Hence we update cell positions in simulations according to the following

rule (Eq. 6.4).
→
−
→ − 𝑡 𝐹 𝑡𝑖
− 𝑡+1 →
𝑋𝑖 = 𝑋𝑖 + × 𝑑𝑡 (6.4)
𝜆
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 116

This model has been implemented from scratch using the C# program-
ming language, and the code source has been published on the code sharing
platform Github2 .

6.3 Results

6.3.1 Zebrafish pectoral fin morphogenesis is proximal

distal oriented

Using the approach described in the methods, we computed the pectoral

fin’s main axes for each time point. Then, we proceeded with calculating
the size of the fin along each direction (Fig. 6.1h-j). Our data suggests that
the fin expands principally along the PD axis with a quasi-linear slope, after
an initial oscillatory behavior (Fig. 6.1h). This result aligns with previous
observations which has consistently shown that a common property of limb
development in vertebrates is the distal orientation of their growth [13, 72].
Furthermore, in this dataset, while fin’s length along the AP axis oscillates
somewhat, no significant overall change is recorded (Fig. 6.1i). Along the
DV axis, however, the fin seems to contract slightly over the length of devel-
opment (Fig. 6.1j), but seems to recover in the latest time steps.

6.3.2 Distal tip-based growth does not account for ze-

brafish pectoral fin morphogenesis

We sought to determine the role of proliferation in zebrafish pectoral fin

growth. For this, we computed for every time point of development the
2
https://github.com/guijoe/MaSoFin
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 117

bounding box encapsulating the pectoral fin. Then, we discretized this

bounding box using the same volume unit everywhere. Next, we calculated
the cumulative number of cell divisions in each volume unit of that space
over the duration of fin growth. In order to understand the distribution of
proliferation in 3D space, we proceeded with plotting the marginal distribu-
tions of cumulative divisions along the three main axes of the pectoral fin
(Fig. 6.3d-e). To avoid the possibility of a frequency accumulation bias due
to fact that the fin grows over time, we also analysed proliferation over a rel-
ative bounding box fitting the fin size at all time points. With this method,
the last few layers of the PD axis always represented the actual distal end of
the fin. We then also evaluated the cumulative number of division in each
of these layers along the fin’s axis (Fig. 6.3g-i).

Marginal distributions of proliferation along the AP and DV axes show

that during pectoral fin morphogenesis, the bulk of proliferation is concen-
trated at the center of the fin, while only a few divisions are observed near
the lateral surfaces (Fig. 6.3a-i, see supplementary figure S1 a,c). Differen-
tial behaviors of cells based on their location is a well-established biological
mechanism, reminiscent of the so-called “French flag” model of Wolpert’s
positional information [? ]. Furthermore, histogram plots of these marginal
distributions may suggest that proliferation along the AP and DV axes can
be assimilated to Gaussian processes (Fig. 6.3d-f, see supplementary figure
S1 d-f). Although it could be expected that such behavior facilitates the
development of the fin toward its known shape, it is not clear whether it is
sufficient to drive this growth. However, it is also likely that the accumula-
tion of cell division in the inner volume of the fin (Fig. 6.3g-i, see supplemen-
tary figure S1 g-i) is merely a consequence of the fin’s geometry, namely its
overall conic shape, favoring higher number of cells in the middle than near
the lateral surfaces. This interpretation seems to be favoured by the data in
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 118

Fig. 6.3j,l (also see supplementary figure S1 j,l), which present quasi-uniform
histograms of number of divisions per time step over the length of AP and
DV axes.

Figure 6.3: Analysis of proliferation in the zebrafish pectoral fin. (a-

c) Frequencies of divisions along the AP, PD and DV axes respectively,
highlighted by a yellow-red color gradient coding for differences in prolif-
eration rates across the fin. (a,c) The preponderance of red at the center
of the fin shows where the bulk of cell divisions takes place, with only a
few of them occurring near the lateral surfaces (yellow). (b) A decreas-
ing gradient of proliferation rates from the proximal pole to the distal tip
characterizes the PD axis. (d-f) Marginal distributions of proliferation
along the AP, PD and DV axes respectively, expressed in numbers of cells
with respect to the absolute distance in 𝜇 m along the axis. (g-i) Same
distributions with respect to the relative distance on the axis. (j-l) Same
distributions expressed in proportions of cells with respect to the absolute
distance.
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 119

Along the PD axis, the marginal distribution of proliferation shows a

decreasing gradient of cell division from the proximal pole to the distal tip
of the fin (Fig. 6.3b, see supplementary figure S1 b). This simply translates
into the fact that proximal layers of the fin, which form early, host more
divisions than distal regions, which develop later. This observation stands
in contradiction with the growth-based morphogenesis hypothesis that stip-
ulates higher proliferation rates at the distal tip of the fin. Hence, this result
suggests that growth-based morphogenesis might not be the main drive for
zebrafish fin pectoral morphogenesis.

6.3.3 Zebrafish pectoral fin cells exhibit preferential di-

rectional behaviors

Next, we looked whether cells exhibited peculiar behaviors along preferential

directions that could influence the shaping of the zebrafish fin. To this
goal, we decided to analyse the dynamics over time of the elongation axis of
each cell. We determined the elongation axis of a cell 𝑖 by computing the
direction of maximum variance of the cloud of points consisting of cell 𝑖 and
its Delaunay neighborhood 𝒩𝑖 . This direction was given by the eigenvector
corresponding to the maximum eigenvalue of the covariance matrix of 𝒩𝑖 .
We denote this vector by →
−
𝑒 Max
𝑖 , which we also consider to be the polarity
vector of the cell (Fig. 6.4a).
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 120

Figure 6.4: Analysis of directional cell behaviors in the zebrafish

pectoral fin. (a) Schematics in 2D of the method used to analyse direc-
tional cell behaviors: for each cell 𝑖, 𝜃𝑖 denotes the polarity angle that this
cell forms between its elongation axis → −𝑒 Max
𝑖 (extracted from the maxi-
mum eigenvalue of the covariance matrix of its neighborhood 𝒩𝑖 ) and the
PD axis → −𝑢 . (b-d) Lateral view of the pectoral fin at different stages of
development, respectively 𝑡 = 28 hpf, 𝑡 = 37.9 hpf and 𝑡 = 47.7 hpf. (e-
g) Vector field of the cells’ elongation axes → − 𝑒 Max
𝑖 in the pectoral fin at
the same stages. (h-j) Distribution of the polarity angles 𝜃𝑖 of the cells in
the pectoral fin at the same stages, compared with the standard distribu-
tion of random angles formed by two arbitrary vectors in 3D (red curve).
(k) Evolution over time of the average polarity angle 𝜃 of the fin cells ±
its standard deviation Δ𝜃 shown in red.

Having computed the elongation axis of each cell through every time
point of development, we observed that cells at initial stages are elongated
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 121

perpendicularly to the PD axis. We further noticed that, during develop-

ment, cells gradually bring their long axis in closer alignment to the PD axis
(Fig. 6.4e-g). This result is consistent with previous studies reporting high
polarisation during limb development [72].In order to confirm this qualita-
tive observation, we measured the angle that cells form via their long axis
with the PD axis of the pectoral fin, and called it the “polarity angle” with
notation 𝜃𝑖 (Fig. 6.4a). At the initial time point, the distribution of polarity
angles was clustered around 90∘ , confirming the previous observation that
cells were elongated perpendicular to the PD axis (Fig. 6.4h). During de-
velopment, this distribution spread in a nonrandom way between 0 and 90∘ ,
where the average polarity angle decreased toward a value of 60∘ , meaning
that cells exhibited preferential directionality by orienting their long axis
toward the PD axis (Fig. 6.4i,j). To ensure that our observations were not
mere features of a single developing fin, we applied the same analysis to a
different dataset which lead to a similar dynamics (see supplementary figure
S2).

6.3.4 Directional cell behaviors are essential to drive

zebrafish pectoral fin morphogenesis

We wanted to find out whether directional behaviors of cells were sufficient

to drive fin morphogenesis. In the previous analysis, we observed that cells
tended to align their long axis in the direction of the PD axis, eventually
forming an average polarity angle 𝜃 = 60∘ . We noticed that such behavior
could explain the overall conic shape of the fin. Based on this observation,
→
−
we designed the polarity force term 𝐹 Pol of the 3D model as follows. First,
we defined a global polarity energy of the cell population, denoted by 𝐻 Pol .
Models in which forces derive from problem-specific global energy have been
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 122

used in different contexts such as cell sorting, molecular signaling, or epithe-

lial morphogenesis in the developing Drosophila [52, 116]. Germann et al.
[57] also use an energy term conjointly with CBM by defining a tissue po-
larity potential and an apicobasal polarity to distinguish between epithelial
and mesenchymal tissues.

Here 𝐻 Pol is expressed over all 𝑁 cells by:

(︃ )︃2
1 ∑︁ )︀ 𝜋
arccos →
(︀− →
𝐻 Pol = 𝑢 ·−
𝑒 Max
𝑖 − (6.5)
𝑁 1≤𝑖≤𝑁 3

Here, →
−
𝑢 denotes the unit vector of the PD axis, which constitutes the direc-
tion of fin growth. It has been suggested that this axis is defined by the body
plan at the onset of fin growth (AP, DV axis). Moreover, cells can sense this
axis through molecular cues originating from determined mesoderm cells,
involving for instance gradients of Wnt or FGF [144].

Although 𝐻 Pol is defined globally, individual cells contribute to this

potential only to the extent of their local neighborhood. This energy was
designed such that its minimum corresponds to 𝜃 = 60∘ . Then, we set the
polarity force to be proportional to the opposite of the gradient of 𝐻 Pol with
respect to each cell:
→
− Pol →
−
𝐹 𝑖 = −𝜈 ∇ 𝑖 𝐻 Pol (6.6)

Having defined these laws governing cell interactions, we proceeded to sim-

ulating limb morphogenesis. In order to highlight the influence of the po-
larity force, we used a uniform cell cycle period with a random initial phase
for each cell. Our virtual limb featured similar properties to the imaged
→
−
limb (Fig. 6.5). Driven by the polarity force 𝐹 Pol and constrained by the
→
−
elastic force 𝐹 AR defined in Section 6.2.3, cells moved and reshaped their
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 123

neighborhoods to minimize the polarity energy 𝐻 Pol , resulting in 𝜃 effec-

tively decreasing throughout development toward 60∘ (Fig. 6.5n), in a way
similar to the real fin. The polarity angle distribution 𝜃𝑖 , which clustered
around 90∘ at the initial time point, progressively spread between 0 and
90∘ (Fig. 6.5h-j). Although the decrease of 𝜃 towards 60∘ was predictable
from the definition of the polarity force, a striking result is that this force
restricted the fin growth toward the distal pole, as observed in development.
Furthermore, over the same period of time as in our dataset, the virtual fin
grew to a size comparable to that of the real fin, from about 12.71 𝜇m to
about 52.94 𝜇m (Fig. 6.5k), presenting over imaged fin an increase of just
below 7%. Finally, our simulated fin acquired a global conic shape similar to
that of the real fin (Fig. 6.5b-d,e-g). In order to get this quantitative agree-
→
− →
−
ment, a right balance between the forces in presence ( 𝐹 AR , 𝐹 Pol ) needed to
be achieved. Proper scaling of these forces was done through model param-
eters (𝐽 = 0.001, 𝜈 = 1, 𝜆 = 0.2). Taken together, these results suggest that
directional cell behaviors, in particular alignment toward the PD axis of the
zebrafish pectoral fin, could be an essential drive for its morphogenesis.
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 124

Figure 6.5: Simulation of pectoral fin morphogenesis based on

directional cell behaviors. Values of the equation parameters: 𝐽 =
0.001, 𝜆 = 0.2, 𝜈 = 1. (a) 3D view of the simulated fin at the final stage
𝑡 = 47.8 hpf. (b-d) Lateral view of the simulated fin at different stages of
development, respectively 𝑡 = 28 hpf, 𝑡 = 37.9 hpf and 𝑡 = 47.7 hpf. (e-
g) Vector field of the cells’ elongation axes →
−
𝑒 Max
𝑖 in the simulated fin at the
same stages. (h-j) Distribution of the polarity angles 𝜃𝑖 of the cells in the
simulated fin at the same stages, compared with the standard distribution
of random angles formed by two arbitrary vectors in 3D (red curve). (k-
m) Evolution over time of the simulated fin size in 𝜇m along the PD, AP
and DV axes respectively. We observe roughly the same behavior as the
real fin in Fig. 6.1h-j. (n) Evolution over time of the average polarity angle
𝜃 of the simulated fin cells ± its standard deviation Δ𝜃 shown in red. This
curve is more scattered than Fig. 6.4k.
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 125

6.4 Discussion

The question of how vertebrates make their limbs is a fascinating problem in

embryology that has been widely investigated across multiple species [144].
Although the study of molecular patterns underlying this morphogenesis
has provided rich insights, the cellular basis of limb formation has not been
completely elucidated. The availability of in toto imaging with resolution at
the single cell level provides quantitative data for cells’ behavior along their
trajectory. New methods need to be developed to analyze such data, to use
it to feed realistic models and evaluate hypotheses through a quantitative
comparison between in vivo and in silico data. This work brings insights
into the development of the zebrafish pectoral fin using quantitative analysis
of imaging data and computational modelling.

Here, we investigated zebrafish pectoral fin development under the prism

of the two dominant hypotheses of cellular behavior during limb growth. On
the one hand, we analyzed proliferation behaviors in different regions of the
fin, and found that proliferation gradients could not account for the observed
growth. On the other hand, analysis of cellular elongation directions showed
that cells tended on average to lower the angle they formed with the PD
axis via their long axis, an indication of preferential polarity. To test this
hypothesis, we formulated a simple mechanistic model of cells and simulated
pectoral fin morphogenesis.

Our model of fin development based on quantitative biological data

derived from live imaging and image processing accounts for the directional
growth and the shape of the fin. However, due to some differences in compar-
ison with imaging data, it also helps refine hypotheses concerning additional
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 126

constraints that were not integrated here. The simulated fin does not ex-
hibit the same amplitude of slight compaction along the DV axis as the one
measured in the biological data. In addition, the simulated fin shows a first
phase of fast growth not quite observed in the zebrafish where the fin growth
rate is closer to linear. We hypothesize that constraints imposed by the out-
side cell layers (i.e. ectodermal layer and enveloping layer) also contribute to
the regulation of the fin growth and shaping, as observed in chick limb mor-
phogenesis [121]. Chemical signalling from the outside cell layers, or other
sources including for example electrical fields [72], could precisely play a role
in the emergence of the cell polarization axes implemented in our model.
Additionally, we considered the contribution of somitic cells that invade the
fin bud during the time course of our observation as neutral regarding the
overall growth and shaping. Both an in vivo and in silico experimentation
are needed to support this view.

The biomechanical aspects privileged in our study are part of a com-

plex interplay of genetic and mechanical cues characteristic of biological de-
velopment. On the mechanical side, the requirement for cellular directional
behaviours has been highlighted. However, local growth, evidenced by sig-
nificant rise in cell population over the observed time span, is a key factor,
serving to provide enough cells for the development of the organ, a con-
clusion also reached in mouse limb morphogenesis [13]. Moreover, other
mechanisms such as cells coalignment or cell competition, which was shown
to be preponderant in polarised tissues and organ size control, could equally
be at play in pectoral fin morphogenesis [148]. To test the latter, a cor-
relation analysis between cell polarisation and cell death would need to be
performed. If the response of LPM cells during early fin development may
mainly involve mechanical interactions, it is certainly required to integrate
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 127

genetic and molecular interactions as well in the transformation of the ecto-

derm that leads to shape the AER. Further studies may be integrate these
mechanisms for more insights into zebrafish pectoral fin morphogenesis.
Quantification of cell behaviors and computational modelling show that cell
directional behaviors drive zebrafish pectoral fin morphogenesis 128

Supplementary figure 1

Figure S1: Analysis of proliferation in the zebrafish pectoral fin

(supplementary dataset). (a-c) Frequencies of divisions along the AP,
PD and DV axes respectively, highlighted by a yellow-red color gradient
coding for differences in proliferation rates across the fin. (a,c) The pre-
ponderance of red at the center of the fin shows where the bulk of cell
divisions takes place, with only a few of them occurring near the lateral
surfaces (yellow). (b) A decreasing gradient of proliferation rates from the
proximal pole to the distal tip characterizes the PD axis. (d-f) Marginal
distributions of proliferation along the AP, PD and DV axes respectively,
expressed in numbers of cells with respect to the absolute distance in 𝜇𝑚
along the axis. (g-i) Same distributions with respect to the relative dis-
tance on the axis. (j-l) Same distributions expressed in proportions of cells
with respect to the absolute distance.
Conclusion 129

Supplementary figure 2

Figure S2: Analysis of directional cell behaviors in the zebrafish

pectoral fin (supplementary dataset). (a) Schematics in 2D of the
method used to analyse directional cell behaviors: for each cell 𝑖, 𝜃𝑖 denotes
the polarity angle that this cell forms between its elongation axis → −𝑒 Max
𝑖
(extracted from the maximum eigenvalue of the covariance matrix of its
neighborhood 𝒩𝑖 ) and the PD axis → −𝑢 . (b-d) Lateral view of the pectoral
fin at different stages of development, respectively 𝑡 = 28.0 hpf, 𝑡 = 36.1 hpf
and 𝑡 = 44.2 hpf. (e-g) Vector field of the cells’ elongation axes →
−
𝑒 Max
𝑖 in the
pectoral fin at the same stages. (h-j) Distribution of the polarity angles
𝜃𝑖 of the cells in the pectoral fin at the same stages, compared with the
standard distribution of random angles formed by two arbitrary vectors
in 3D (red curve). (k) Evolution over time of the average polarity angle 𝜃
of the fin cells ± its standard deviation Δ𝜃 shown in red.
Chapter 7

Conclusion

Our goal in this research journey was to address the question of how com-
putational models can contribute in unravelling the mysteries of biological
development. Practically, the bulk of our work consists in formulating the-
oretical models that capture the dynamics of living tissues and simulating
these models to investigate, in close collaboration with biologists, morpho-
genesis events in living systems. During our doctoral studies, this programme
has been implemented through the development of agent-based models of cell
and tissue mechanics, which were then used to gain valuable insights into
the regulation of the epiblast shape and motion during mouse implantation,
and the morphogenesis of the zebrafish pectoral fin.

In our initial project, we built upon the fundamental principles

of Sub-cellular Element Models (SEM) to develop MG#, a multi-purpose
novel model of cells and tissues biomechanics. In this model, a biologi-
cal cell is abstracted by a cohort of particles whose triangulation form the
cell membrane, and a single intracellular particle represent the biological
cell. Non-linear elastic potentials between particles mimic the plasticity of

131
Conclusion 132

cell membranes and the activity of the cytoskeleton within the cell. These
characteristics enable simulated cells to exhibit key behavioural properties
of biological cells, such as division, apoptosis, bio-realistic shapes, and the
feedback loop between biomechanical and biochemical effects, making MG#
suitable for a broad range of morphological phenomena. In particular, we
have shown that cell populations simulated with MG# can reproduce living
tissue behaviours such as epithelial folding or the formation of multicellular
rosettes.

In our next endeavour, we leveraged MG# to investigate the driv-

ing factor behind the symmetry-breaking in the mouse embryo epiblast dur-
ing mouse implantation. At the onset of implantation, the mouse embryo
consists of three cell lines: the embryonic tissue (epiblast), and two extra-
embryonic tissues (trophectoderm and primitive endoderm). Building on
previous studies, we hypothesised that epiblast remodelling into a cup re-
sults from mechanical constraints imposed by the polar trophectoderm, also
undergoing significant morphogenesis in this time. We investigated this hy-
pothesis by simulating both tissues with MG#.

In our latest project, we used computational modelling to study the

morphogenesis of the zebrafish pectoral fin. More specifically, in collabora-
tion with experimental biologists at the French CNRS BioEmergences lab,
we sought to understand how this 3D organ develops from an initial layer of
the cells. By analysing quantitative data resulting from the automatic track-
ing of single fin cells, we were able to rule out growth-based morphogenesis
as the predominant factor for fin expansion along the PD axis. This result
aligned with previous observations in mouse limbs, which also pointed us in
the direction of examining cell elongation patterns in the fin. We found out
that, during fin growth, cells aligned their long axis to the proximal-distal
Conclusion 133

axis (PD) of the zebrafish fin overtime. Building on this quantitative result,
we enhanced the basic centre-based model by adding a polarity force term
and specified proliferation rules based on cell cycle time. With this enhanced
CBM model simulated fin growth.

In summary, the studies presented in this document highlight the effec-

tiveness of mathematical and computational models in developmental biology
and provide new insights in this field. Furthermore, the formulated models
have been implemented as open source tools that were made available to
the wider community by publishing their code on the open source sharing
platform Github.

7.1 Future perspectives

The work carried out in this thesis can be further developed or adapted to
accommodate more investigations of biological phenomena.

First, the models developed could be adapted for the purpose of studying
other biological phenomena in development including differential-adhesion or
differential-tension induced cell sorting, or pathological behaviours such as
the growth of tumors. Although MG# simulations presented in this the-
sis were mostly limited to the scope of epithelial tissues, MG#, because of
its deformable nature, can be enhanced to capture the dynamics of more
complex cell arrangements and tissue topologies, making it suitable for both
epithelia and mesenchyme. A step in this direction could be to graft onto
MG# a theory of cell external contacts that accounts for cases where there
may not exist an ideal mapping between external neighbouring particles.
In order to achieve this, established contact mechanics methods such as
Hertzian contact, Johnson-Kendall-Roberts or Maugis-Dugtale can be used
Conclusion 134

[27, 146]. To simulate differential-adhesion cell sorting for instance, single

cells entourage would need to be updated periodically through their particles
external neighbourhoods. In addition, contact surfaces, instead of punctual
particle-particle contacts, would need to be defined around interacting parti-
cles of neighbouring cells. Moreover, different physical properties such as the
modulus of elasticity of cells or the Poisson ratio would need to be defined
in other to resolve adhesions on these contact surfaces.

In Chapter 6, we used a data-driven approach to examine the morpho-

genesis of the zebrafish pectoral fin. In this approach, we derived biological
hypothesis directly from the analysis of the cell tracking data of fin cells and
tested these hypothesis through simulations when necessary. However, more
information, going from the subcellular activities of proteins, or genetic regu-
lation, to cellular and tissue level like mechanical stresses, can be drawn from
biological data. This data can then be integrated in the modelling workflow
to infer model parameters or test model predictions [51].
Bibliography

[1] Ahrens, J., Geveci, B., and Law, C. Paraview: An end-user tool
for large data visualization. The visualization handbook 717 (2005).

[2] Alt, S., Ganguly, P., and Salbreux, G. Vertex models: From
cell mechanics to tissue morphogenesis. Philosophical transactions of
the Royal Society of London. Series B, Biological sciences 372, 1720 (5
2017), 20150520.

[3] Andrew, D. J., and Ewald, A. J. Morphogenesis of epithelial

tubes: Insights into tube formation, elongation, and elaboration. De-
velopmental biology 341, 1 (2010), 34–55.

[4] Arnold, S. J., and Robertson, E. J. Making a commitment:

Cell lineage allocation and axis patterning in the early mouse embryo.
Nature reviews Molecular cell biology 10, 2 (2009), 91–103.

[5] Atri, A., Amundson, J., Clapham, D., and Sneyd, J. A single-
pool model for intracellular calcium oscillations and waves in the xeno-
pus laevis oocyte. Biophysical Journal 65, 4 (1993), 1727–1739.

[6] Balbi, V., Destrade, M., and Goriely, A. The mechanics of

human brain organoids. Physical Review E 101 (2020).

135
Bibliography 136

[7] Basan, M., Prost, J., Joanny, J.-F., and Elgeti, J. Dissipa-
tive particle dynamics simulations for biological tissues: rheology and
competition. Physical biology 8, 2 (2011), 026014.

[8] Bedzhov, I., Graham, S. J., Leung, C. Y., and Zernicka-

Goetz, M. Developmental plasticity, cell fate specification and mor-
phogenesis in the early mouse embryo. Philosophical Transactions of
the Royal Society B: Biological Sciences 369, 1657 (2014), 20130538.

[9] Bedzhov, I., and Zernicka-Goetz, M. Self-organizing properties

of mouse pluripotent cells initiate morphogenesis upon implantation.
Cell 156, 5 (2014), 1032–1044.

[10] Belmonte, J. M., Swat, M. H., and Glazier, J. A. Filopodial-

tension model of convergent-extension of tissues. PLoS computational
biology 12, 6 (2016), e1004952.

[11] Binder, K., Ceperley, D. M., Hansen, J.-P., Kalos, M., Lan-
dau, D., Levesque, D., Mueller-Krumbhaar, H., Stauffer,
D., and Weis, J.-J. Monte Carlo methods in statistical physics, vol. 7.
Springer Science & Business Media, 2012.

[12] Blankenship, J. T., Backovic, S. T., Sanny, J. S., Weitz, O.,

and Zallen, J. A. Multicellular rosette formation links planar cell
polarity to tissue morphogenesis. Developmental cell 11, 4 (2006), 459–
470.

[13] Boehm, B., Westerberg, H., Lesnicar-Pucko, G., Raja, S.,

Rautschka, M., Cotterell, J., Swoger, J., and Sharpe, J.
The role of spatially controlled cell proliferation in limb bud morpho-
genesis. PLoS biology 8, 7 (2010).
Bibliography 137

[14] Bongiorno, T., Kazlow, J., Mezencev, R., Griffiths, S.,

Olivares-Navarrete, R., McDonald, J. F., Schwartz, Z.,
Boyan, B. D., McDevitt, T. C., and Sulchek, T. Mechani-
cal stiffness as an improved single-cell indicator of osteoblastic human
mesenchymal stem cell differentiation. Journal of biomechanics 47, 9
(2014), 2197–2204.

[15] Boocock, D., Hino, N., Ruzickova, N., Hirashima, T., and
Hannezo, E. Theory of mechanochemical patterning and optimal
migration in cell monolayers. Nature Physics 17, 2 (2021), 267–274.

[16] Brand-Saberi, B., Seifert, R., Grim, M., Wilting, J.,

Köhlewein, M., and Christ, B. Blood vessel formation in the
avian limb bud involves angioblastic and angiotrophic growth. Devel-
opmental Dynamics 202, 2 (1995), 181–194.

[17] Brodland, G. W. How computational models can help unlock bio-

logical systems, 2015.

[18] Brodland, G. W., and Chen, H. H. The mechanics of heterotypic

cell aggregates: insights from computer simulations. Journal of Biome-
chanical Engineering 122, 4 (2000), 402.

[19] Brodland, G. W., Conte, V., Cranston, P. G., Veldhuis,

J., Narasimhan, S., Hutson, M. S., Jacinto, A., Ulrich, F.,
Baum, B., and Miodownik, M. Video force microscopy reveals the
mechanics of ventral furrow invagination in drosophila. Proceedings of
the National Academy of Sciences 107, 51 (2010), 22111–22116.

[20] Brown, J. M., and Garcı́a-Garcı́a, M. J. Secretory pathway cal-

cium atpase 1 (spca1) controls mouse neural tube closure by regulating
cytoskeletal dynamics. Development 145, 19 (2018).
Bibliography 138

[21] Cameron, J., and McCredie, J. Innervation of the undifferentiated

limb bud in rabbit embryo. Journal of anatomy 134, Pt 4 (1982), 795.

[22] Chan, C. J., Costanzo, M., Ruiz-Herrero, T., Mönke, G.,

Petrie, R. J., Bergert, M., Diz-Munoz, A., Mahadevan, L.,
and Hiiragi, T. Hydraulic control of mammalian embryo size and
cell fate. Nature 571, 7763 (2019), 112–116.

[23] Christ, B., and Brand-Saberi, B. Limb muscle development.

International Journal of Developmental Biology 46, 7 (2002), 905–914.

[24] Christodoulou, N., Kyprianou, C., Weberling, A., Wang,

R., Cui, G., Peng, G., Jing, N., and Zernicka-Goetz, M. Se-
quential formation and resolution of multiple rosettes drive embryo
remodelling after implantation. Nature Cell Biology 20, 11 (2018),
1278–1289.

[25] Christodoulou, N., and Skourides, P. A. Cell-autonomous

Ca2+ flashes elicit pulsed contractions of an apical actin network to
drive apical constriction during neural tube closure. Cell reports 13,
10 (2015), 2189–2202.

[26] Christodoulou, N., Weberling, A., Strathdee, D., Ander-

son, K. I., Timpson, P., and Zernicka-Goetz, M. Morphogenesis
of extra-embryonic tissues directs the remodelling of the mouse embryo
at implantation. Nature Communications 10, 1 (2019), 1–12.

[27] Chu, Y.-S., Dufour, S., Thiery, J. P., Perez, E., and Pincet,
F. Johnson-kendall-roberts theory applied to living cells. Physical
review letters 94, 2 (2005), 028102.

[28] Chung, S., and Andrew, D. J. The formation of epithelial tubes.

Journal of cell science 121, 21 (2008), 3501–3504.
Bibliography 139

[29] Chung, S., and Andrew, D. J. Cadherin 99C regulates apical

expansion and cell rearrangement during epithelial tube elongation.
Development 141, 9 (2014), 1950–1960.

[30] Coen, E. The art of genes: how organisms make themselves. Oxford
University Press, 2000.

[31] Coucouvanis, E., and Martin, G. R. Signals for death and sur-
vival: A two-step mechanism for cavitation in the vertebrate embryo.
Cell 83, 2 (1995), 279–287.

[32] Davidson, L. A. Epithelial machines that shape the embryo. Trends

in cell biology 22, 2 (2012), 82–87.

[33] Debnath, J., Mills, K. R., Collins, N. L., Reginato, M. J.,

Muthuswamy, S. K., and Brugge, J. S. The role of apoptosis in
creating and maintaining luminal space within normal and oncogene-
expressing mammary acini. Cell 111, 1 (2002), 29–40.

[34] Delile, J., Doursat, R., and Peyriéras, N. Computational mod-

eling and simulation of animal early embryogenesis with the MecaGen
platform. In Computational Systems Biology. Elsevier, 2014, pp. 359–
405.

[35] Delile, J., Herrmann, M., Peyriéras, N., and Doursat, R. A

cell-based computational model of early embryogenesis coupling me-
chanical behaviour and gene regulation. Nature Communications 8, 1
(2017), 1–10.

[36] Dequidt, A., Devemy, J., and Padua, A. A. Thermalized drude

oscillators with the lammps molecular dynamics simulator. Journal of
chemical information and modeling 56, 1 (2016), 260–268.
Bibliography 140

[37] Dijkstra, E. W. On the role of scientific thought. In Selected writings

on computing: a personal perspective. Springer, 1982, pp. 60–66.

[38] Dillon, R., and Othmer, H. G. A mathematical model for out-

growth and spatial patterning of the vertebrate limb bud. Journal of
theoretical biology 197, 3 (1999), 295–330.

[39] Disset, J. Simulation de Cellules et Réseaux de Regulation Génétique

Artificiels pour le Développement et l’Auto-organisation de Créatures
Multicellulaires. PhD dissertation, Université de Toulouse, 2017.

[40] Disset, J., Cussat-Blanc, S., and Duthen, Y. MecaCell: an

Open-source Efficient Cellular Physics Engine. In ECAL (2015).

[41] Drasdo, D., and Höhme, S. A single-cell-based model of tumor

growth in vitro: monolayers and spheroids. Physical biology 2, 3 (2005),
133.

[42] Drasdo, D., and Loeffler, M. Individual-based models to growth

and folding in one-layered tissues: intestinal crypts and early devel-
opment. Nonlinear Analysis: Theory, Methods & Applications 47, 1
(2001), 245–256.

[43] Duclut, C., Sarkar, N., Prost, J., and Jülicher, F. Fluid
pumping and active flexoelectricity can promote lumen nucleation in
cell assemblies. Proceedings of the National Academy of Sciences 116,
39 (2019), 19264–19273.

[44] Dupont, G., Falcke, M., Kirk, V., and Sneyd, J. Models of
calcium signalling, vol. 43. Springer, 2016.
Bibliography 141

[45] Estrada, J., Andrew, N., Gibson, D., Chang, F., Gnad, F.,
and Gunawardena, J. Cellular interrogation: exploiting cell-to-
cell variability to discriminate regulatory mechanisms in oscillatory
signalling. PLoS computational biology 12, 7 (2016), e1004995.

[46] Evans, E., and Fung, Y.-C. Improved measurements of the ery-
throcyte geometry. Microvascular research 4, 4 (1972), 335–347.

[47] Farhadifar, R., Röper, J.-C., Aigouy, B., Eaton, S., and
Jülicher, F. The influence of cell mechanics, cell-cell interactions,
and proliferation on epithelial packing. Current Biology 17, 24 (12
2007), 2095–2104.

[48] Faure, E., Savy, T., Rizzi, B., Melani, C., Stašová, O.,
Fabrèges, D., Špir, R., Hammons, M., Čúnderlı́k, R., Recher,
G., et al. A workflow to process 3d+ time microscopy images of
developing organisms and reconstruct their cell lineage. Nature Com-
munications 7, 1 (2016), 1–10.

[49] Fedosov, D. A., Caswell, B., and Karniadakis, G. E. System-

atic coarse-graining of spectrin-level red blood cell models. Computer
Methods in Applied Mechanics and Engineering 199, 29-32 (2010),
1937–1948.

[50] Fedosov, D. A., Lei, H., Caswell, B., Suresh, S., and Karni-
adakis, G. E. Multiscale modeling of red blood cell mechanics and
blood flow in malaria. PLoS Computational Biology 7, 12 (12 2011),
e1002270.

[51] Fletcher, A., and Osborne, J. Seven challenges in the multiscale

modelling of multicellular tissues. Preprints (2020).
Bibliography 142

[52] Fletcher, A., Osterfield, M., Baker, R., and Shvartsman,

S. Vertex Models of Epithelial Morphogenesis. Biophysical Journal
106, 11 (6 2014), 2291–2304.

[53] Fletcher, A. G., Cooper, F., and Baker, R. E. Mechanocellular

models of epithelial morphogenesis. Philosophical Transactions of the
Royal Society B: Biological Sciences 372, 1720 (2017), 20150519.

[54] Fletcher, A. G., Osterfield, M., Baker, R. E., and Shvarts-

man, S. Y. Vertex models of epithelial morphogenesis. Biophysical
Journal 106, 11 (6 2014), 2291–2304.

[55] Fristrom, D. The cellular basis of epithelial morphogenesis. a review.

Tissue and Cell 20, 5 (1988), 645–690.

[56] Fukui, Y., Hirota, Y., Matsuo, M., Gebril, M., Akaeda, S.,
Hiraoka, T., and Osuga, Y. Uterine receptivity, embryo attach-
ment, and embryo invasion: Multistep processes in embryo implanta-
tion. Reproductive medicine and biology 18, 3 (2019), 234–240.

[57] Germann, P., Marin-Riera, M., and Sharpe, J. ya|| a: Gpu-

powered spheroid models for mesenchyme and epithelium. Cell systems
8, 3 (2019), 261–266.

[58] Glazier, J. A., and Graner, F. Simulation of the differential

adhesion driven rearrangement of biological cells. Physical Review E
47, 3 (3 1993), 2128–2154.

[59] Goriely, A., Erlich, A., Jones, G., Tisseur, F., and Moul-
ton, D. The role of topology and mechanics in uniaxially growing cell
networks. Proceedings of the Royal Society A: Mathematical, Physical
and Engineering Sciences 476, 2233 (2020).
Bibliography 143

[60] Graner, F., and Glazier, J. A. Simulation of biological cell sorting

using a two-dimensional extended Potts model. Physical Review Letters
69, 13 (9 1992), 2013–2016.

[61] Green, J. B. A., and Sharpe, J. Positional information and

reaction-diffusion: two big ideas in developmental biology combine.
Development (Cambridge, England) 142, 7 (4 2015), 1203–11.

[62] Grim, M., and Christ, B. Neural crest cell migration into the limb
bud of avian embryos. Progress in clinical and biological research 383
(1993), 391–402.

[63] Gros, J., Hu, J. K.-H., Vinegoni, C., Feruglio, P. F.,

Weissleder, R., and Tabin, C. J. WNT5A/JNK and FGF/MAPK
pathways regulate the cellular events shaping the vertebrate limb bud.
Current Biology 20, 22 (2010), 1993–2002.

[64] Hannezo, E., and Heisenberg, C.-P. Mechanochemical feedback

loops in development and disease. Cell 178, 1 (2019), 12–25.

[65] Harding, M. J., McGraw, H. F., and Nechiporuk, A. The roles

and regulation of multicellular rosette structures during morphogene-
sis. Development 141, 13 (2014), 2549–2558.

[66] Hardy, A., and Steeb, W.-H. Mathematical Tools in Computer

Graphics with C# Implementations. WORLD SCIENTIFIC, 1 2008.

[67] Harrison, S. E., Sozen, B., Christodoulou, N., Kyprianou,

C., and Zernicka-Goetz, M. Assembly of embryonic and extraem-
bryonic stem cells to mimic embryogenesis in vitro. Science 356, 6334
(2017), eaal1810.
Bibliography 144

[68] Heisenberg, C.-P. D’arcy thompson’s ‘on growth and form’: From
soap bubbles to tissue self-organization. Mechanisms of Development
145 (6 2017), 32–37.

[69] Honda, H. The world of epithelial sheets. Development, Growth &

Differentiation 59, 5 (2017), 306–316.

[70] Honda, H., and Nagai, T. Cell models lead to understanding of

multi-cellular morphogenesis consisting of successive self-construction
of cells. Journal of Biochemistry 157, 3 (3 2015), 129–136.

[71] Honda, H., Tanemura, M., and Nagai, T. A three-dimensional

vertex dynamics cell model of space-filling polyhedra simulating cell
behavior in a cell aggregate. Journal of theoretical biology 226, 4 (2004),
439–453.

[72] Hopyan, S. Biophysical regulation of early limb bud morphogenesis.

Developmental biology 429, 2 (2017), 429–433.

[73] Huang, R., Zhi, Q., and Christ, B. The relationship between limb
muscle and endothelial cells migrating from single somite. Anatomy
and embryology 206, 4 (2003), 283–289.

[74] Huttenlocher, D. P., Klanderman, G. A., and Rucklidge,

W. J. Comparing images using the Hausdorff distance. IEEE Trans-
actions on pattern analysis and machine intelligence 15, 9 (1993), 850–
863.

[75] Inoue, Y., Suzuki, M., Watanabe, T., Yasue, N., Tateo, I.,
Adachi, T., and Ueno, N. Mechanical roles of apical constriction,
cell elongation, and cell migration during neural tube formation in
xenopus. Biomechanics and modeling in mechanobiology 15, 6 (2016),
1733–1746.
Bibliography 145

[76] Ising, E. Beitrag zur theorie des ferromagnetismus. Zeitschrift für

Physik 31, 1 (1925), 253–258.

[77] Jacobson, A. G., and Gordon, R. Changes in the shape of the

developing vertebrate nervous system analyzed experimentally, math-
ematically and by computer simulation. Journal of Experimental Zo-
ology 197, 2 (1976), 191–246.

[78] Kaouri, K., Maini, P. K., Skourides, P., Christodoulou, N.,

and Chapman, S. J. A simple mechanochemical model for calcium
signalling in embryonic epithelial cells. Journal of mathematical biology
78, 7 (2019), 2059–2092.

[79] Keller, R. Physical biology returns to morphogenesis. Science 338,

6104 (10 2012), 201–203.

[80] Kimmel, C. B., Ballard, W. W., Kimmel, S. R., Ullmann,

B., and Schilling, T. F. Stages of embryonic development of the
zebrafish. Developmental dynamics 203, 3 (1995), 253–310.

[81] Kojima, Y., Tam, O. H., and Tam, P. P. Timing of developmental

events in the early mouse embryo. In Seminars in cell & developmental
biology (2014), vol. 34, Elsevier, pp. 65–75.

[82] Kolega, J. The cellular basis of epithelial morphogenesis. In The

Cellular Basis of Morphogenesis. Springer, 1986, pp. 103–143.

[83] Kursawe, J., Baker, R. E., and Fletcher, A. G. Approximate

Bayesian computation reveals the importance of repeated measure-
ments for parameterising cell-based models of growing tissues. Journal
of Theoretical Biology 443 (4 2018), 66–81.
Bibliography 146

[84] Lecuit, T., and Lenne, P.-F. Cell surface mechanics and the con-
trol of cell shape, tissue patterns and morphogenesis. Nature Reviews
Molecular Cell Biology 8, 8 (8 2007), 633–644.

[85] Lenne, P. F., Munro, E., Heemskerk, I., Warmflash, A.,

Bocanegra-Moreno, L., Kishi, K., Kicheva, A., Long, Y.,
Fruleux, A., Boudaoud, A., et al. Roadmap on multiscale cou-
pling of biochemical and mechanical signals during development. Phys-
ical Biology (2020).

[86] Li, S., and Muneoka, K. Cell migration and chick limb devel-
opment: chemotactic action of FGF-4 and the AER. Developmental
biology 211, 2 (1999), 335–347.

[87] Lienkamp, S. S., Liu, K., Karner, C. M., Carroll, T. J.,

Ronneberger, O., Wallingford, J. B., and Walz, G. Verte-
brate kidney tubules elongate using a planar cell polarity–dependent,
rosette-based mechanism of convergent extension. Nature genetics 44,
12 (2012), 1382–1387.

[88] Malmi Kakkada, A., Li, X., Samanta, H. S., Sinha, S., and
Thirumalai, D. Cell growth rate dictates the onset of glass to fluid-
like transition and long time super-diffusion in an evolving cell colony.
Biophysical Journal 114, 3 (2 2018), 323a.

[89] Malmi-Kakkada, A. N., Li, X., Samanta, H. S., Sinha, S., and
Thirumalai, D. Cell growth rate dictates the onset of glass to fluid-
like transition and long time superdiffusion in an evolving cell colony.
Physical Review X 8, 2 (2018), 021025.

[90] Marin-Riera, M., Brun-Usan, M., Zimm, R., Välikangas, T.,

and Salazar-Ciudad, I. Computational modeling of development
Bibliography 147

by epithelia, mesenchyme and their interactions: a unified model.

Bioinformatics 32, 2 (9 2015), btv527.

[91] Martin, A. C., Gelbart, M., Fernandez-Gonzalez, R.,

Kaschube, M., and Wieschaus, E. F. Integration of contractile
forces during tissue invagination. Journal of Cell Biology 188, 5 (2010),
735–749.

[92] Martin, A. C., and Goldstein, B. Apical constriction: themes

and variations on a cellular mechanism driving morphogenesis. Devel-
opment 141, 10 (2014), 1987–1998.

[93] Martin, A. C., Kaschube, M., and Wieschaus, E. F. Pulsed

contractions of an actin–myosin network drive apical constriction. Na-
ture 457, 7228 (2009), 495–499.

[94] Martı́n-Belmonte, F., Yu, W., Rodrı́guez-Fraticelli, A. E.,

Ewald, A., Werb, Z., Alonso, M. A., and Mostov, K. Cell-
polarity dynamics controls the mechanism of lumen formation in ep-
ithelial morphogenesis. Current Biology 18, 7 (2008), 507–513.

[95] McShane, S. G., Molè, M. A., Savery, D., Greene, N. D., Tam,
P. P., and Copp, A. J. Cellular basis of neuroepithelial bending
during mouse spinal neural tube closure. Developmental biology 404, 2
(2015), 113–124.

[96] Merks, R. M., Guravage, M., Inzé, D., and Beemster, G. T.

Virtualleaf: an open-source framework for cell-based modeling of plant
tissue growth and development. Plant physiology 155, 2 (2011), 656–
666.
Bibliography 148

[97] Micoulet, A., Spatz, J. P., and Ott, A. Mechanical

response analysis and power generation by single-cell stretching.
ChemPhysChem 6, 4 (2005), 663–670.

[98] Milde, F., Tauriello, G., Haberkern, H., and Koumout-

sakos, P. SEM++: A particle model of cellular growth, signaling
and migration. Computational Particle Mechanics 1, 2 (6 2014), 211–
227.

[99] Mirams, G. R., Arthurs, C. J., Bernabeu, M. O., Bordas, R.,

Cooper, J., Corrias, A., Davit, Y., Dunn, S.-J., Fletcher,
A. G., Harvey, D. G., et al. Chaste: an open source c++ library
for computational physiology and biology. PLoS Comput Biol 9, 3
(2013), e1002970.

[100] Morishita, Y., and Iwasa, Y. Growth based morphogenesis of

vertebrate limb bud. Bulletin of mathematical biology 70, 7 (2008),
1957–1978.

[101] Nagai, T., and Honda, H. A dynamic cell model for the formation
of epithelial tissues. Philosophical Magazine B 81, 7 (2001), 699–719.

[102] Newman, T. J. Modeling multi-cellular systems using sub-cellular

elements. arXiv preprint q-bio/0504028 (2005).

[103] Nguyen, H., Boix-Fabrés, J., Peyriéras, N., and Kardash,

E. 3d+ time imaging and image reconstruction of pectoral fin during
zebrafish embryogenesis. In Computer Optimized Microscopy. Springer,
2019, pp. 135–153.

[104] Niculescu, I., Textor, J., and de Boer, R. J. Crawling and

gliding: a computational model for shape-driven cell migration. PLOS
Computational Biology 11, 10 (10 2015), e1004280.
Bibliography 149

[105] Niswander, L., Tickle, C., Vogel, A., Booth, I., and Mar-
tin, G. R. FGF-4 replaces the apical ectodermal ridge and directs
outgrowth and patterning of the limb. Cell 75, 3 (1993), 579–587.

[106] Odell, G. M., Oster, G., Alberch, P., and Burnside, B. The
mechanical basis of morphogenesis: I. epithelial folding and invagina-
tion. Developmental biology 85, 2 (1981), 446–462.

[107] Odenthal, T., Smeets, B., Van Liedekerke, P., Tijskens, E.,
Van Oosterwyck, H., and Ramon, H. Analysis of initial cell
spreading using mechanistic contact formulations for a deformable cell
model. PLoS Computational Biology 9, 10 (10 2013), e1003267.

[108] Odenthal, T., Smeets, B., Van Liedekerke, P., Tijskens, E.,
Van Oosterwyck, H., and Ramon, H. Analysis of initial cell
spreading using mechanistic contact formulations for a deformable cell
model. PLoS computational biology 9, 10 (2013).

[109] Ogura, Y., Wen, F.-L., Sami, M. M., Shibata, T., and Hayashi,
S. A switch-like activation relay of EGFR-ERK signaling regulates a
wave of cellular contractility for epithelial invagination. Developmental
cell 46, 2 (2018), 162–172.

[110] Okuda, S., Inoue, Y., Eiraku, M., Adachi, T., and Sasai, Y.
Vertex dynamics simulations of viscosity-dependent deformation dur-
ing tissue morphogenesis. Biomechanics and Modeling in Mechanobi-
ology 14, 2 (4 2015), 413–425.

[111] Okuda, S., Inoue, Y., Eiraku, M., Sasai, Y., and Adachi,
T. Apical contractility in growing epithelium supports robust main-
tenance of smooth curvatures against cell-division-induced mechanical
disturbance. Journal of Biomechanics 46, 10 (6 2013), 1705–1713.
Bibliography 150

[112] Okuda, S., Inoue, Y., Eiraku, M., Sasai, Y., and Adachi,
T. Modeling cell proliferation for simulating three-dimensional tissue
morphogenesis based on a reversible network reconnection framework.
Biomechanics and Modeling in Mechanobiology 12, 5 (10 2013), 987–
996.

[113] Okuda, S., Inoue, Y., Eiraku, M., Sasai, Y., and Adachi, T.
Reversible network reconnection model for simulating large deforma-
tion in dynamic tissue morphogenesis. Biomechanics and Modeling in
Mechanobiology 12, 4 (8 2013), 627–644.

[114] Okuda, S., Inoue, Y., Watanabe, T., and Adachi, T. Cou-
pling intercellular molecular signalling with multicellular deformation
for simulating three-dimensional tissue morphogenesis. Interface Focus
(2015).

[115] Okuda, S., Miura, T., Inoue, Y., Adachi, T., and Eiraku, M.
Combining Turing and 3D vertex models reproduces autonomous mul-
ticellular morphogenesis with undulation, tubulation, and branching.
Scientific Reports 8, 1 (12 2018), 2386.

[116] Osborne, J. M., Fletcher, A. G., Pitt-Francis, J. M., Maini,

P. K., and Gavaghan, D. J. Comparing individual-based ap-
proaches to modelling the self-organization of multicellular tissues.
PLoS computational biology 13, 2 (2017), e1005387.

[117] Parfitt, D.-E., and Shen, M. M. From blastocyst to gastrula:

Gene regulatory networks of embryonic stem cells and early mouse
embryogenesis. Philosophical Transactions of the Royal Society B: Bi-
ological Sciences 369, 1657 (2014), 20130542.
Bibliography 151

[118] Pathmanathan, P., Cooper, J., Fletcher, A., Mirams, G.,

Murray, P., Osborne, J., Pitt-Francis, J., Walter, A., and
Chapman, S. A computational study of discrete mechanical tissue
models. Physical biology 6, 3 (2009), 036001.

[119] Pearl, E. J., Li, J., and Green, J. B. Cellular systems for ep-
ithelial invagination. Philosophical Transactions of the Royal Society
B: Biological Sciences 372, 1720 (2017), 20150526.

[120] Pillarisetti, A., Desai, J. P., Ladjal, H., Schiffmacher, A.,

Ferreira, A., and Keefer, C. L. Mechanical phenotyping of
mouse embryonic stem cells: increase in stiffness with differentiation.
Cellular Reprogramming (Formerly" Cloning and Stem Cells") 13, 4
(2011), 371–380.

[121] Poplawski, N. J., Swat, M., Gens, J. S., and Glazier, J. A.

Adhesion between cells, diffusion of growth factors, and elasticity of the
aer produce the paddle shape of the chick limb. Physica A: Statistical
Mechanics and its Applications 373 (2007), 521–532.

[122] Rashevsky, N. An approach to the mathematical biophysics of bio-

logical self-regulation and of cell polarity. The bulletin of mathematical
biophysics 2, 1 (1940), 15–25.

[123] Rashevsky, N. Topology and life: in search of general mathemati-

cal principles in biology and sociology. The bulletin of mathematical
biophysics 16, 4 (1954), 317–348.

[124] Recher, G., Jouralet, J., Brombin, A., Heuzé, A., Mugniery,
E., Hermel, J.-M., Desnoulez, S., Savy, T., Herbomel, P.,
Bourrat, F., et al. Zebrafish midbrain slow-amplifying progenitors
Bibliography 152

exhibit high levels of transcripts for nucleotide and ribosome biogenesis.

Development 140, 24 (2013), 4860–4869.

[125] Recho, P., Hallou, A., and Hannezo, E. Theory of

mechanochemical patterning in biphasic biological tissues. Proceedings
of the National Academy of Sciences 116, 12 (2019), 5344–5349.

[126] Reiter, R. S., and Solursh, M. Mitogenic property of the apical

ectodermal ridge. Developmental biology 93, 1 (1982), 28–35.

[127] Revell, C., Blumenfeld, R., and Chalut, K. J. Force-based

three-dimensional model predicts mechanical drivers of cell sorting.
Proceedings of the Royal Society B 286, 1895 (2019), 20182495.

[128] Rivera-Pérez, J. A., and Hadjantonakis, A.-K. The dynam-

ics of morphogenesis in the early mouse embryo. Cold Spring Harbor
perspectives in biology 7, 11 (6 2014), a015867.

[129] Rohrschneider, M. R., and Nance, J. Polarity and cell fate

specification in the control of caenorhabditis elegans gastrulation. De-
velopmental Dynamics 238, 4 (2009), 789–796.

[130] Sandersius, S. A., and Newman, T. J. Modeling cell rheology with

the subcellular element model. Physical biology 5, 1 (2008), 015002.

[131] Sawyer, J. M., Harrell, J. R., Shemer, G., Sullivan-Brown,

J., Roh-Johnson, M., and Goldstein, B. Apical constriction: a
cell shape change that can drive morphogenesis. Developmental biology
341, 1 (2010), 5–19.

[132] Shahbazi, M. N., Jedrusik, A., Vuoristo, S., Recher, G., Hu-
palowska, A., Bolton, V., Fogarty, N. M., Campbell, A.,
Devito, L. G., Ilic, D., et al. Self-organization of the human
Bibliography 153

embryo in the absence of maternal tissues. Nature Cell Biology 18, 6

(2016), 700–708.

[133] Shahbazi, M. N., Scialdone, A., Skorupska, N., Weberling,

A., Recher, G., Zhu, M., Jedrusik, A., Devito, L. G., Noli,
L., Macaulay, I. C., et al. Pluripotent state transitions coordinate
morphogenesis in mouse and human embryos. Nature 552, 7684 (2017),
239–243.

[134] Shahbazi, M. N., and Zernicka-Goetz, M. Deconstructing and

reconstructing the mouse and human early embryo. Nature Cell Biology
20, 8 (2018), 878–887.

[135] Sharpe, J. Computer modeling in developmental biology: Growing

today, essential tomorrow. Development 144, 23 (2017), 4214–4225.

[136] Sherrard, K., Robin, F., Lemaire, P., and Munro, E. Sequen-
tial activation of apical and basolateral contractility drives ascidian
endoderm invagination. Current Biology 20, 17 (2010), 1499–1510.

[137] Shyer, A. E., Tallinen, T., Nerurkar, N. L., Wei, Z., Gil,
E. S., Kaplan, D. L., Tabin, C. J., and Mahadevan, L. Villifi-
cation: how the gut gets its villi. Science 342, 6155 (2013), 212–218.

[138] Steinberg, M. S. Adhesion-guided multicellular assembly: a com-

mentary upon the postulates, real and imagined, of the differential
adhesion hypothesis, with special attention to computer simulations of
cell sorting. Journal of theoretical biology 55, 2 (1975), 431–443.

[139] Sui, L., Alt, S., Weigert, M., Dye, N., Eaton, S., Jug, F.,
Myers, E. W., Jülicher, F., Salbreux, G., and Dahmann, C.
Differential lateral and basal tension drive folding of drosophila wing
Bibliography 154

discs through two distinct mechanisms. Nature Communications 9, 1

(2018), 1–13.

[140] Suzanne, M., and Steller, H. Shaping organisms with apoptosis.

Cell Death & Differentiation 20, 5 (2013), 669–675.

[141] Suzuki, M., Sato, M., Koyama, H., Hara, Y., Hayashi, K.,
Yasue, N., Imamura, H., Fujimori, T., Nagai, T., Campbell,
R. E., et al. Distinct intracellular Ca2+ dynamics regulate apical
constriction and differentially contribute to neural tube closure. De-
velopment 144, 7 (2017), 1307–1316.

[142] Swat, M. H., Thomas, G. L., Belmonte, J. M., Shirinifard, A.,

Hmeljak, D., and Glazier, J. A. Multi-scale modeling of tissues
using compucell3D. Methods in cell biology 110 (2012), 325–366.

[143] Thompson, D. W., et al. On growth and form. On growth and

form. (1942).

[144] Tickle, C. How the embryo makes a limb: determination, polarity

and identity. Journal of anatomy 227, 4 (2015), 418–430.

[145] Turing, A. M. The Chemical Basis of Morphogenesis, vol. 237. Royal

Society.

[146] Van Liedekerke, P., Palm, M., Jagiella, N., and Drasdo,
D. Simulating tissue mechanics with agent-based models: concepts,
perspectives and some novel results. Computational particle mechanics
2, 4 (2015), 401–444.

[147] Villoutreix, P., Delile, J., Rizzi, B., Duloquin, L., Savy, T.,
Bourgine, P., Doursat, R., and Peyriéras, N. An integrated
Bibliography 155

modelling framework from cells to organism based on a cohort of digital

embryos. Scientific reports 6 (2016), 37438.

[148] Vincent, J.-P., Fletcher, A. G., and Baena-Lopez, L. A.

Mechanisms and mechanics of cell competition in epithelia. Nature
reviews Molecular cell biology 14, 9 (2013), 581–591.

[149] Wang, H., and Dey, S. K. Roadmap to embryo implantation: Clues

from mouse models. Nature Reviews Genetics 7, 3 (2006), 185–199.

[150] Wang, Q., Holmes, W. R., Sosnik, J., Schilling, T., and
Nie, Q. Cell Sorting and Noise-Induced Cell Plasticity Coordinate to
Sharpen Boundaries between Gene Expression Domains. PLOS Com-
putational Biology 13, 1 (1 2017), e1005307.

[151] Wen, F.-L., Wang, Y.-C., and Shibata, T. Epithelial folding

driven by apical or basal-lateral modulation: geometric features, me-
chanical inference, and boundary effects. Biophysical Journal 112, 12
(2017), 2683–2695.

[152] Wolpert, L., Tickle, C., and Arias, A. M. Principles of devel-

opment. Oxford university press, 2015.

[153] Yanagida, A., Revell, C., Stirparo, G. G., Corujo-Simon,

E., Aspalter, I. M., Peters, R., De Belly, H., Cassani, D. A.,
Achouri, S., Blumenfeld, R., et al. Cell surface fluctuations
regulate early embryonic lineage sorting. BioRxiv (2020).

[154] Yu, J. C., and Fernandez-Gonzalez, R. Quantitative modelling

of epithelial morphogenesis: integrating cell mechanics and molecular
dynamics. Seminars in Cell & Developmental Biology 67 (7 2017),
153–160.
Bibliography 156

[155] Zhang, S., Chen, T., Chen, N., Gao, D., Shi, B., Kong, S.,
West, R. C., Yuan, Y., Zhi, M., Wei, Q., et al. Implanta-
tion initiation of self-assembled embryo-like structures generated using
three types of mouse blastocyst-derived stem cells. Nature Communi-
cations 10, 1 (2019), 1–17.

[156] Zhou, J., Kim, H. Y., and Davidson, L. A. Actomyosin stiffens

the vertebrate embryo during crucial stages of elongation and neural
tube closure. Development 136, 4 (2009), 677–688.
List of Figures

2.1 A population of cells, initially randomly scattered, gradually

sorts itself into two regions due to differences in adhesion
strength between cell types. . . . . . . . . . . . . . . . . . . 16

3.1 Computational model (MG#). A,B. 3D representation

of cells: The cell is abstracted by an agglomeration of particles
(small white spheres), whose triangulation (white edges) forms
the membrane, and by an intracellular particle (big white
sphere). Interactions between the intracellular and membrane
particles (blue lines) mimic the cytoskeleton. Membrane par-
ticle 𝑖 is under forces from neighbouring particles (𝑗, 𝑘, ...),
and from the intracellular particle 𝜒. A. Cell with 42 vertices.
B. Columnar cell with 34 vertices. C. A view of particle neigh-
bourhoods in a square arrangement of four cells (nuclei not
displayed). white: internal neighbourhood links. Yellow: ex-
ternal neighbourhood links. In this case, diagonal cells are not
touching. D,E. Forces acting on a cell. Membrane particle 𝑖
is under forces from neighbouring particles (𝑗, 𝑘, ...), and from
the intracellular particle 𝜒. F. External forces acting on a cell
→
− →
− ext →
− int →
− int →
−𝜒
via its particles. Here, 𝐹 ext
𝑖2 = 𝐹 𝑗2 𝑖2 = ( 𝐹 𝑗1 𝑗2 + 𝐹 𝑗3 𝑗2 ) + 𝐹 𝑗2 .
G. Plots of the magnitude of Morse forces under different val-
ues of 𝐽, with 𝜌 = 1 and 𝑟eq = 0.5. . . . . . . . . . . . . . . 27

157
Bibliography 158

3.2 Regular cell shapes obtained by shape metamorpho-

sis. A. Top: initial spherical cell. Bottom: Ellipsoid cell
shape representing an elongated cell. Parameter values: 𝑎 =
0.707, 𝑏 = 1.25, 𝑐 = 1 B. Top: initial spherical cell. Bottom:
Biconcave cell shape characteristic of Red Blood Cells. Pa-
rameter values: 𝑎0 = 0.0518, 𝑎1 = 0.0518, 𝑎2 = −4.491. C.
Top: initial spherical cell. Bottom: Red Blood cells bicon-
cave shape is not attained when radial forces are suppressed,
though simulation time is 10 times longer than in C. Parame-
ter values: 𝑎 = 0.707, 𝑏 = 1.25, 𝑐 = 1. D. Top: initial spherical
cell. Bottom: Cylindrical cell shape characteristic of epithelial
cells. Parameter values: ℎ = 1, 𝑟 = 0.5. . . . . . . . . . . . . 38
3.3 Convergence of Red Blood cell Shapes. A. Cell shapes
obtained for different values of the simulation time step. B.
Elastic energy charts for different simulation scenarios. Higher
time steps converge towards non-zero values, but yield un-
stable simulations (blue, yellow ). Lower time steps converge
towards zero and yield stable simulations (green, red, purple,
brown). In each of these simulations, we use parameter values
𝑎0 = 0.0518, 𝑎1 = 0.0518, 𝑎2 = −4.491. . . . . . . . . . . . . . 39
3.4 A cell undergoing apoptosis. A cell decreases in volume
until it vanishes . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.5 Cell Polarity. Different spatial orientations for the api-
cobasal polarity of an epithelial cell. In each case, the ep-
ithelial cell has been obtained by shape metamorphosis of an
initial spherical cell. A different polarity orientation has been
specified for each cell. . . . . . . . . . . . . . . . . . . . . . . 41
3.6 A simulated cell going through a series of successive divisions 42
Bibliography 159

3.7 MG# Class Architecture. An MG# user experiment

*.UserExperiment (blue) is implemented in the form of a
simulation class which inherits from the base simulator class
(MGSharp.Core.Simulator ). A simulation is performed on
a population of cells (*.BiologicalEntities.CellPopulation),
which may be partitioned into tissues (*.BiologicalEnti-
ties.Tissue), which themselves are specific cell populations.
Each cell population or tissue is made of cells (*.BiologicalEn-
tities.MGCell ), whose representation in the 3D space are ge-
ometric meshes (*.GeometricPrimitives.Mesh). The generic
mesh class is built upon more elementary geometric classes
not shown here (vector, face, etc.) and extended from Hardy
and Steeb implementations [66]. The generic MGCell can be
specialised further into child classes with specific behaviours
(*.BiologicalEntities.UserClass). Helper classes and meth-
ods (MGSharp.Core.Helpers.* ), and the generic model class
(*.MGModels.MGModel ) that contains the definition of most
generic model parameters, are transversal to this architecture,
and can be used at any level. An instance of a user experi-
ment class (green, underlined) implements inherited methods
from the Simulator class, and may define custom methods. . 44
3.8 Simulation framework: GUI-less physics engine and viewer . 45

4.1 Planar Polarised Constriction. A. Apicobasal polarisa-

tion of a planar polarised constricted cell B. A 2D cut of
a constricted cell. The cell underwent planar polarised con-
striction by reducing its lateral width by a length of 2𝑑. C.
Simulation of planar polarised constriction using an epithelial
cell with square base. Parameter values: 𝑅 = 0.5, 𝑑 = 0.35 . 53
4.2 Apical Constriction. A. Schema of an apically constricted
epithelial cell with hexagonal base. B. A 2D section of a
constricted cell. The cell underwent apical constriction by
reducing its apical radius by a length of 𝑑. Initial cylindrical
section in white, constricted cell in red. C. Simulation of
apical constriction with an epithelial cell with square base.
Parameter values: 𝐻 = 2, 𝑅 = 0.5, 𝑑 = 0.35. D. Simulation
of apical constriction with an epithelial cell with hexagonal
base. Parameter values: 𝐻 = 2, 𝑅 = 0.5, 𝑑 = 0.35. . . . . . . 58
Bibliography 160

4.3 Apical Constriction with volume conservation. A.

Schema of an apically constricted epithelial cell with hexag-
onal base. B. A 2D section of a constricted cell. The cell
underwent apical constriction by reducing its apical radius by
a length of 𝑑. Initial cylindrical section in white, constricted
cell in red. C. Simulation of apical constriction with volume
conservation using an epithelial cell with hexagonal base. Pa-
rameter values: 𝐻 = 2, 𝑅 = 0.5, 𝑑 = 0.4, 𝐻0 = 0.25. Using
equation (4.45), we find 𝑅0 = 0.2618908. . . . . . . . . . . . 67
4.4 𝐶𝑎2+ regulated Apical Constriction. A. Levels of con-
centrations of 𝐶𝑎2+ (orange) and IP3 (grey) within an epithe-
lial undergoing AC. The yellow curve shows the theoretical
evolution of the apical cell area through time. The apical
faces goes through alternative phases of growth and decrease
as reported in [25]. B. Plots of theoretical and empirical
apical surface area as yielded by our simulation. Observed
slight differences are caused by the Euler explicit scheme used
C. Snapshots of the constricting cell at different time points
showing the evolution of the apical surface area. Parameter
values: 𝐻 = 2, 𝑅 = 0.5. From (eq. 4.59 - eq. 4.61), we fix
𝑐0 = 1, 𝜃0 = 1, ℎ0 = 1, 𝜆 = 0.5, 𝜇 = 0.289, 𝑘1 = 0.7. . . . . . . 71
4.5 Epithelial Folding. A. Top view of a single epithelial layer
folding under the influence of single cells at the centre of the
sheet (coloured in yellow) constricting apically. The sheet
is composed of 81 epithelial cells with hexagonal base. The
epicentre is represented by the 25 yellow cells at the centre of
the sheet. B. Lateral view of the folding process in A. The
sheet goes out of plane. Parameter values: ℎ = 2, 𝑅 = 0.5, 𝑑 =
0.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.6 Morphogenesis of multi-cellular rosettes. A. Simulation
of the morphogenesis of a multicellular rosette through pla-
nar polarised constriction of single cells. The cell population
is made of 8 epithelial cells with square base. B. Simula-
tion of the morphogenesis of a multicellular rosette via apical
constriction of cells in a single-layered epithelial sheet. C.
Simulation of the morphogenesis of a multicellular rosette via
apical constriction of polarised cells in a double-layered ep-
ithelial sheet. In B and C, the the sheets are respectively
composed of 25 and 50 hexagonal epithelial cells. Parameter
values: ℎ = 2, 𝑅 = 0.5, 𝑑 = 0.5 . . . . . . . . . . . . . . . . . 77
Bibliography 161

5.1 Review of epiblast symmetry breaking theories.

A. The basement membrane separating the epiblast and the
primitive endoderm moulds the epiblast into a cup while it
disintegrates between the epiblast and the trophectoderm in
mouse embryos [8]. B. Embryoid structures featuring epiblast
and trophectoderm stem cells surrounded by an ECM acting
as a basement membrane (ETS-embryoids) replicate mouse
embryogenesis by forming body structures similar to those ob-
served in normal embryonic development [67]. Here the pres-
ence of the trophecdoderm shows that this tissue might be
required for symmetry breaking in the epiblast and cup shape
acquisition. C. Embryoid structures featuring epiblast and
primitive endoderm stem cells surrounded by an ECM acting
as a basement membrane (EXE-embryoids) do not break sym-
metry in the epiblast, but initiate lumenogenesis [155]. This
evidences the requirement of the trophectoderm for the re-
modelling of the epiblast. D. Trophectoderm morphogenesis
during mouse implantation. Trophectodermal cells elongate,
then undergo apical constriction, resulting in the tissue folding
and invaginating the epiblast [26]. This suggests that epiblast
remodelling into a cup might be a mechanical response to tro-
phectoderm dynamics . . . . . . . . . . . . . . . . . . . . . . 83
5.2 Lumenogenesis in the epiblast. A. A 3D model of a
rosette-shaped epiblast. B. A 2D slice of the epiblast in A
showing apically constricted cells of the building block of the
epiblast rosette. C. Creation of the lumen cavity by repulsion
at the apical surface of the epiblast. Green arrows represent
the direction of repulsive forces. The snapshots (from left to
right) were taken respectively at 𝑡 = 0, 500 and 2000. D. Lat-
eral view of the sliced epiblast showing the lumen volume.
E. Dynamics in time of the volume of the lumen. Values of
the equation parameters: 𝐽EPI = 2.5, 𝜆med = 𝜆𝜒 = 2, 𝜌 =
1, 𝑅lum = 0.25. . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Bibliography 162

5.3 Trophectoderm morphogenesis regulates epiblast

shape. A-D. 3D snapshots of the simulation of TE and
EPI morphogenesis during mouse implantation, and the reg-
ulation of EPI shape, taken respectively at 𝑡 = 0, 3000, 6000
and 9000. E-H. Corresponding 2D slices of the cell popula-
tion at the same stages. (A,E). The initial stage features a
single layered TE with cuboidal cells resting upon the rosette-
shaped epiblast. (B,F). TE cells have transited to a columnar
shape. (C,G). The TE has folded by apical constriction of
single cells. Concomitantly, lumenogenesis was initiated in
the epiblast (the process starts at 𝑡 = 4000). (D,H). Af-
ter adhesive links were broken between TE and EPI, the EPI
bounces back to its near spherical shape. I. Definitions of
the metrics used to evaluate the model, involving the cur-
vature 𝜃, TE/EPI interface diameter 𝐷, TE/EPI interface
length 𝐿, and interface ratio 𝐿/𝐷. J. Plot of the popula-
tion’s elastic energy 𝐸. Discontinuities mark the start of new
morphological events at 𝑡 = 0, 3000, 4000, and 6000). Af-
ter removal of the TE, 𝐸 falls closer to zero than ever be-
fore, meaning that cells are closer to their resting stage, hence
less externally constrained. K. Plot of the interface curva-
ture 𝜃. During TE morphogenesis, 𝜃 rises towards a flat an-
gle, then sharply drops when the TE is removed. L. Plot of
the interface ratio 𝐿/𝐷. During TE morphogenesis, the in-
terface curvature decreases towards 1, then sharply increases
when the TE is removed. Values of the equation parameters:
𝐽EPI = 𝐽TE = 2.5, 𝜆med = 𝜆 = 2, 𝜌 = 1, 𝑅lum = 0.25. . . . . . 90
5.4 Trophectoderm fosters epiblast movement towards
maternal sites. A. Snapshots of the simulation of TE and
EPI morphogenesis during mouse implantation, and their in-
fluence on EPI positioning, taken respectively at 𝑡 = 0 and
6000. B. Plot of the pushing distance, which increases with
time. C. Plot of the elastic energy 𝐸. Discontinuities mark
the start of new morphological events (𝑡 = 0 and 3000). The
sudden soar observed at 𝑡 = 4000 reflects the slight elongation
of the tissue due to hollowing-driven lumenogenesis in the epi-
blast. D. Plot of the pushing distance on the epiblast Centre
of Mass (CoM), which also increases with time. E. Plot of the
pushing distance on the cell population Centre of Mass (CoM),
which also increases with time. Values of the equation param-
eters: 𝐽EPI = 𝐽TE = 2.5, 𝜆med = 𝜆𝜒 = 2, 𝜌 = 1, ,d=0.5, 𝑅lum =
0.25. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Bibliography 163

5.5 Mechanical properties of EPI and TE determine

mouse implantation. A. In“Silico” experimental protocol
used to determine cells elastic modulus. B. Stress-Strain curve
(black ) for a single epithelial cell (34 vertices) with 𝐽 = 2.5.
(blue) Linear approximation of the Stress-Strain curve. The
elastic modulus of the cell is determined by the slope of this
line (𝑌 = 2.92). C. Plot of the Elastic (Young) modulus of
cells as a function of parameter 𝐽, the interaction strength
between subcellular particles. D,E,F. Respective Plots of the
Interface curvature, the Interface ratio and the Pushing Dis-
tance as functions of the mechanical stiffness of TE cells (de-
termined by 𝐽TE as in C). G. Plot of the fitness metric as
functions of the mechanical stiffness of TE cells (determined
by 𝐽TE as in C). H. Snapshots of the epiblast shape at the
end of simulations for different values of 𝐽TE . With equal
stiffness (middle, 𝐽TE = 2.5, 𝐽EPI = 2.5), trophectoderm mor-
phogenesis flatten the epiblast, which acquires a cup shape.
However, with significantly lower stiffness (left, 𝐽TE = 0.3,
𝐽EPI = 2.5), trophectoderm morphogenesis barely reshape the
epiblast; meanwhile, with considerably higher stiffness (right,
𝐽TE = 4.9, 𝐽EPI = 2.5), the trophectoderm invaginates the
epiblast, forcing a concave interface with the epiblast. Other
parameter values, 𝜆med = 𝜆𝜒 = 2, 𝜌 = 1, 𝑑 = 0.5. . . . . . . . 97
5.6 Epiblast and trophectoderm population reconstruc-
tion. A. The rosette-shaped EPI tissue is built by submitting
polarised cells in a double epithelial layer to apical constric-
tion. Green arrows indicate the apical surface of the cells,
where constriction occurs. B. The initial cell population (TE
and EPI) is built by adding an epithelial layer to the forming
the EPI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.7 Top view of trophectoderm morphogenesis. A. Initial
stage with cuboidal cells. B. Columnar TE initiating api-
cal constriction. Red arrows highlight cells undergoing apical
constriction. In this case, only cells in the middle constrict
(light blue) to enable folding. C. Folded TE. D. Folded TE
after separation from the EPI. . . . . . . . . . . . . . . . . . 102
Bibliography 164

5.8 Trophectoderm fosters epiblast movement towards

maternal sites (Without lumenogenesis in epiblast).
A. Snapshots of the simulation of TE and EPI morphogen-
esis during mouse implantation, and their influence on EPI
positioning, taken respectively at 𝑡 = 0 and 6000. B. Plot
of the pushing distance, which increases with time. C. Plot
of the elastic energy 𝐸. Discontinuities mark the start of
new morphological events (𝑡 = 0 and 3000). D. Plot of
the pushing distance on the epiblast Centre of Mass (CoM),
which also increases with time. E. Plot of the pushing dis-
tance on the cell population Centre of Mass (CoM), which
also increases with time. Values of the equation parameters:
𝐽EPI = 𝐽TE = 2.5, 𝜆0 = 𝜆𝜒 = 2, 𝜌 = 1. . . . . . . . . . . . . . 103
5.9 Sensitivity analysis (Supplementary). A. Stress-Strain
curve (black ) for a single epithelial cell (34 vertices) with
𝐽 = 2.5 and 𝜆med = 𝜆𝜒 = 2. (blue) Linear approximation
of the Stress-Strain curve. The elastic modulus of the cell is
determined by the slope of this line (𝑌 = 2.92). B. Plot of the
Elastic (Young) modulus of cells as a function of the param-
eter ratio ( 𝜆𝜆med
𝜒
). Young’s modulus is defined and constant
𝜆𝜒
for values of 𝜆med greater or equal to approximately 0.161.
Below this value, simulated cells do not behave as physical
materials, and the elasticity modulus cannot be defined as il-
lustrated in the next plot. C. Stress-Strain curve (black ) for a
single epithelial cell (34 vertices) with 𝐽 = 2.5, 𝜆med = 2 and
𝜆𝜒 = 0.25. The discontinuity in the curve shows that the set
of parameters is not suitable for a cell. D. 3D rendering of
an epithelial cell with square basis and 42 vertices. E. Stress-
Strain curve (black ) for a single epithelial cell (42 vertices)
with 𝐽 = 2.5 and 𝜆med = 𝜆𝜒 = 2. (blue) Linear approxima-
tion of the Stress-Strain curve. The elastic modulus of the cell
is determined by the slope of this line (𝑌 = 2.75). F. Plot
of the Elastic (Young) modulus of a cell (42 vertices) as a
function of the parameter 𝐽, the interaction strength between
subcellular particles. In order for such a cell (42 vertices) to
have equivalent stiffness with the previous type of cell (34 ver-
tices, 𝐽34 = 2.5, 𝑌34 = 2.92), the parameter 𝐽42 needs to be
set to approximately 2.6 (𝑌42 = 2.90). . . . . . . . . . . . . . 104
Bibliography 165

6.1 Geometry of the pectoral fin based on live imaging

and image processing data. (a) 3D rendering of raw data
nuclear staining at 𝑡 = 47.7 hpf: dorsal view of the zebrafish
body with detection of approximate nuclear centers of the
pectoral fin cells highlighted by colored dots, where the color
code depends on the cell type; scale bar: 20 𝜇m. Imaging
data and individual cell tracking have been generated by the
CNRS/Bioemerges lab. (b-d) After applying cell detection
methods: 3D rendering of the approximate nucleus centers of
LPM cells in the pectoral fin at different stages of develop-
ment, respectively 𝑡 = 28 hpf, 𝑡 = 37.9 hpf and 𝑡 = 47.7 hpf
(AP: anteroposterior axis; DV: dorsoventral axis). (e-g) 3D
rendering of the pectoral fin at the same times along the AP
axis and PD (proximodistal) axis. (h-j) Evolution over time of
the fin size in 𝜇m along the PD, AP and DV axes respectively.
Fin expansion occurs mainly along PD. It undergoes a slight
compaction along the other two axes, more pronounced along
the DV axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2 Center-based computational model of multicellular
dynamics. (a) Schema of a local cell neighborhood and the
→
−
abstract forces on cell centers. 𝐹 AR
𝑗𝑖 is the passive attraction-
→
−
repulsion force exerted on a cell 𝑖 by a cell 𝑗. 𝐹 Pol 𝑖 is the
active migration force driven by the cell’s polarity (specified
in Section 6.3.4). (b) Plot of the Morse force profile (derivative
→
−
of the Morse potential) defining 𝐹 AR , for different parameter
values. This curve presents two regimes: a positive regime
(attraction) below an equilibrium distance 𝑟eq and a negative
regime (repulsion) above. . . . . . . . . . . . . . . . . . . . . 113
Bibliography 166

6.3 Analysis of proliferation in the zebrafish pectoral fin. (a-

c) Frequencies of divisions along the AP, PD and DV axes
respectively, highlighted by a yellow-red color gradient coding
for differences in proliferation rates across the fin. (a,c) The
preponderance of red at the center of the fin shows where the
bulk of cell divisions takes place, with only a few of them oc-
curring near the lateral surfaces (yellow). (b) A decreasing
gradient of proliferation rates from the proximal pole to the
distal tip characterizes the PD axis. (d-f) Marginal distribu-
tions of proliferation along the AP, PD and DV axes respec-
tively, expressed in numbers of cells with respect to the abso-
lute distance in 𝜇 m along the axis. (g-i) Same distributions
with respect to the relative distance on the axis. (j-l) Same
distributions expressed in proportions of cells with respect to
the absolute distance. . . . . . . . . . . . . . . . . . . . . . . 118
6.4 Analysis of directional cell behaviors in the zebrafish
pectoral fin. (a) Schematics in 2D of the method used to
analyse directional cell behaviors: for each cell 𝑖, 𝜃𝑖 denotes
the polarity angle that this cell forms between its elongation
axis →−𝑒 Max
𝑖 (extracted from the maximum eigenvalue of the
covariance matrix of its neighborhood 𝒩𝑖 ) and the PD axis
→
−𝑢 . (b-d) Lateral view of the pectoral fin at different stages
of development, respectively 𝑡 = 28 hpf, 𝑡 = 37.9 hpf and
𝑡 = 47.7 hpf. (e-g) Vector field of the cells’ elongation axes
→
−𝑒 Max in the pectoral fin at the same stages. (h-j) Distribu-
𝑖
tion of the polarity angles 𝜃𝑖 of the cells in the pectoral fin
at the same stages, compared with the standard distribution
of random angles formed by two arbitrary vectors in 3D (red
curve). (k) Evolution over time of the average polarity angle
𝜃 of the fin cells ± its standard deviation ∆𝜃 shown in red. . 120
Bibliography 167

6.5 Simulation of pectoral fin morphogenesis based on di-

rectional cell behaviors. Values of the equation parame-
ters: 𝐽 = 0.001, 𝜆 = 0.2, 𝜈 = 1. (a) 3D view of the simulated
fin at the final stage 𝑡 = 47.8 hpf. (b-d) Lateral view of the
simulated fin at different stages of development, respectively
𝑡 = 28 hpf, 𝑡 = 37.9 hpf and 𝑡 = 47.7 hpf. (e-g) Vector field
of the cells’ elongation axes →−𝑒 Max
𝑖 in the simulated fin at the
same stages. (h-j) Distribution of the polarity angles 𝜃𝑖 of the
cells in the simulated fin at the same stages, compared with
the standard distribution of random angles formed by two ar-
bitrary vectors in 3D (red curve). (k-m) Evolution over time
of the simulated fin size in 𝜇m along the PD, AP and DV
axes respectively. We observe roughly the same behavior as
the real fin in Fig. 6.1h-j. (n) Evolution over time of the aver-
age polarity angle 𝜃 of the simulated fin cells ± its standard
deviation ∆𝜃 shown in red. This curve is more scattered than
Fig. 6.4k. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
S1 Analysis of proliferation in the zebrafish pectoral fin
(supplementary dataset). (a-c) Frequencies of divisions
along the AP, PD and DV axes respectively, highlighted by a
yellow-red color gradient coding for differences in proliferation
rates across the fin. (a,c) The preponderance of red at the cen-
ter of the fin shows where the bulk of cell divisions takes place,
with only a few of them occurring near the lateral surfaces
(yellow). (b) A decreasing gradient of proliferation rates from
the proximal pole to the distal tip characterizes the PD axis.
(d-f) Marginal distributions of proliferation along the AP, PD
and DV axes respectively, expressed in numbers of cells with
respect to the absolute distance in 𝜇𝑚 along the axis. (g-
i) Same distributions with respect to the relative distance on
the axis. (j-l) Same distributions expressed in proportions of
cells with respect to the absolute distance. . . . . . . . . . . 128
Bibliography 168

S2 Analysis of directional cell behaviors in the zebrafish

pectoral fin (supplementary dataset). (a) Schematics in
2D of the method used to analyse directional cell behaviors:
for each cell 𝑖, 𝜃𝑖 denotes the polarity angle that this cell forms
between its elongation axis → −
𝑒 Max
𝑖 (extracted from the maxi-
mum eigenvalue of the covariance matrix of its neighborhood
𝒩𝑖 ) and the PD axis → −
𝑢 . (b-d) Lateral view of the pectoral fin
at different stages of development, respectively 𝑡 = 28.0 hpf,
𝑡 = 36.1 hpf and 𝑡 = 44.2 hpf. (e-g) Vector field of the cells’
elongation axes → −
𝑒 Max
𝑖 in the pectoral fin at the same stages.
(h-j) Distribution of the polarity angles 𝜃𝑖 of the cells in the
pectoral fin at the same stages, compared with the standard
distribution of random angles formed by two arbitrary vec-
tors in 3D (red curve). (k) Evolution over time of the average
polarity angle 𝜃 of the fin cells ± its standard deviation ∆𝜃
shown in red. . . . . . . . . . . . . . . . . . . . . . . . . . . 129